Print this page
cstyle fixes
dsl_dataset_set_fsid_guid should use ZFS_SPACE_CHECK_RESERVED
dsl_dataset_set_fsid_guid _check and _sync func declared static,
removed from dsl_dataset.h
rewrite unique_valid
6333 ZFS should let the user specify or modify the fsid_guid of a dataset
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dsl_dataset.c
+++ new/usr/src/uts/common/fs/zfs/dsl_dataset.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25 25 * Copyright (c) 2014 RackTop Systems.
26 26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 27 */
28 28
29 29 #include <sys/dmu_objset.h>
30 30 #include <sys/dsl_dataset.h>
31 31 #include <sys/dsl_dir.h>
32 32 #include <sys/dsl_prop.h>
33 33 #include <sys/dsl_synctask.h>
34 34 #include <sys/dmu_traverse.h>
35 35 #include <sys/dmu_impl.h>
36 36 #include <sys/dmu_tx.h>
37 37 #include <sys/arc.h>
38 38 #include <sys/zio.h>
39 39 #include <sys/zap.h>
40 40 #include <sys/zfeature.h>
41 41 #include <sys/unique.h>
42 42 #include <sys/zfs_context.h>
43 43 #include <sys/zfs_ioctl.h>
44 44 #include <sys/spa.h>
45 45 #include <sys/zfs_znode.h>
46 46 #include <sys/zfs_onexit.h>
47 47 #include <sys/zvol.h>
48 48 #include <sys/dsl_scan.h>
49 49 #include <sys/dsl_deadlist.h>
50 50 #include <sys/dsl_destroy.h>
51 51 #include <sys/dsl_userhold.h>
52 52 #include <sys/dsl_bookmark.h>
53 53 #include <sys/dmu_send.h>
54 54 #include <sys/zio_checksum.h>
55 55 #include <sys/zio_compress.h>
56 56 #include <zfs_fletcher.h>
57 57
58 58 /*
59 59 * The SPA supports block sizes up to 16MB. However, very large blocks
60 60 * can have an impact on i/o latency (e.g. tying up a spinning disk for
61 61 * ~300ms), and also potentially on the memory allocator. Therefore,
62 62 * we do not allow the recordsize to be set larger than zfs_max_recordsize
63 63 * (default 1MB). Larger blocks can be created by changing this tunable,
64 64 * and pools with larger blocks can always be imported and used, regardless
65 65 * of this setting.
66 66 */
67 67 int zfs_max_recordsize = 1 * 1024 * 1024;
68 68
69 69 #define SWITCH64(x, y) \
70 70 { \
71 71 uint64_t __tmp = (x); \
72 72 (x) = (y); \
73 73 (y) = __tmp; \
74 74 }
75 75
76 76 #define DS_REF_MAX (1ULL << 62)
77 77
78 78 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
79 79
80 80 /*
81 81 * Figure out how much of this delta should be propogated to the dsl_dir
82 82 * layer. If there's a refreservation, that space has already been
83 83 * partially accounted for in our ancestors.
84 84 */
85 85 static int64_t
86 86 parent_delta(dsl_dataset_t *ds, int64_t delta)
87 87 {
88 88 dsl_dataset_phys_t *ds_phys;
89 89 uint64_t old_bytes, new_bytes;
90 90
91 91 if (ds->ds_reserved == 0)
92 92 return (delta);
93 93
94 94 ds_phys = dsl_dataset_phys(ds);
95 95 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
96 96 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
97 97
98 98 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
99 99 return (new_bytes - old_bytes);
100 100 }
101 101
102 102 void
103 103 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
104 104 {
105 105 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
106 106 int compressed = BP_GET_PSIZE(bp);
107 107 int uncompressed = BP_GET_UCSIZE(bp);
108 108 int64_t delta;
109 109
110 110 dprintf_bp(bp, "ds=%p", ds);
111 111
112 112 ASSERT(dmu_tx_is_syncing(tx));
113 113 /* It could have been compressed away to nothing */
114 114 if (BP_IS_HOLE(bp))
115 115 return;
116 116 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
117 117 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
118 118 if (ds == NULL) {
119 119 dsl_pool_mos_diduse_space(tx->tx_pool,
120 120 used, compressed, uncompressed);
121 121 return;
122 122 }
123 123
124 124 dmu_buf_will_dirty(ds->ds_dbuf, tx);
125 125 mutex_enter(&ds->ds_lock);
126 126 delta = parent_delta(ds, used);
127 127 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
128 128 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
129 129 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
130 130 dsl_dataset_phys(ds)->ds_unique_bytes += used;
131 131
132 132 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
133 133 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
134 134 B_TRUE;
135 135 }
136 136
137 137 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
138 138 if (f != SPA_FEATURE_NONE)
139 139 ds->ds_feature_activation_needed[f] = B_TRUE;
140 140
141 141 mutex_exit(&ds->ds_lock);
142 142 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
143 143 compressed, uncompressed, tx);
144 144 dsl_dir_transfer_space(ds->ds_dir, used - delta,
145 145 DD_USED_REFRSRV, DD_USED_HEAD, tx);
146 146 }
147 147
148 148 int
149 149 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
150 150 boolean_t async)
151 151 {
152 152 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
153 153 int compressed = BP_GET_PSIZE(bp);
154 154 int uncompressed = BP_GET_UCSIZE(bp);
155 155
156 156 if (BP_IS_HOLE(bp))
157 157 return (0);
158 158
159 159 ASSERT(dmu_tx_is_syncing(tx));
160 160 ASSERT(bp->blk_birth <= tx->tx_txg);
161 161
162 162 if (ds == NULL) {
163 163 dsl_free(tx->tx_pool, tx->tx_txg, bp);
164 164 dsl_pool_mos_diduse_space(tx->tx_pool,
165 165 -used, -compressed, -uncompressed);
166 166 return (used);
167 167 }
168 168 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
169 169
170 170 ASSERT(!ds->ds_is_snapshot);
171 171 dmu_buf_will_dirty(ds->ds_dbuf, tx);
172 172
173 173 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
174 174 int64_t delta;
175 175
176 176 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
177 177 dsl_free(tx->tx_pool, tx->tx_txg, bp);
178 178
179 179 mutex_enter(&ds->ds_lock);
180 180 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
181 181 !DS_UNIQUE_IS_ACCURATE(ds));
182 182 delta = parent_delta(ds, -used);
183 183 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
184 184 mutex_exit(&ds->ds_lock);
185 185 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
186 186 delta, -compressed, -uncompressed, tx);
187 187 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
188 188 DD_USED_REFRSRV, DD_USED_HEAD, tx);
189 189 } else {
190 190 dprintf_bp(bp, "putting on dead list: %s", "");
191 191 if (async) {
192 192 /*
193 193 * We are here as part of zio's write done callback,
194 194 * which means we're a zio interrupt thread. We can't
195 195 * call dsl_deadlist_insert() now because it may block
196 196 * waiting for I/O. Instead, put bp on the deferred
197 197 * queue and let dsl_pool_sync() finish the job.
198 198 */
199 199 bplist_append(&ds->ds_pending_deadlist, bp);
200 200 } else {
201 201 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
202 202 }
203 203 ASSERT3U(ds->ds_prev->ds_object, ==,
204 204 dsl_dataset_phys(ds)->ds_prev_snap_obj);
205 205 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
206 206 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
207 207 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
208 208 ds->ds_object && bp->blk_birth >
209 209 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
210 210 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
211 211 mutex_enter(&ds->ds_prev->ds_lock);
212 212 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
213 213 mutex_exit(&ds->ds_prev->ds_lock);
214 214 }
215 215 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
216 216 dsl_dir_transfer_space(ds->ds_dir, used,
217 217 DD_USED_HEAD, DD_USED_SNAP, tx);
218 218 }
219 219 }
220 220 mutex_enter(&ds->ds_lock);
221 221 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
222 222 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
223 223 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
224 224 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
225 225 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
226 226 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
227 227 mutex_exit(&ds->ds_lock);
228 228
229 229 return (used);
230 230 }
231 231
232 232 uint64_t
233 233 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
234 234 {
235 235 uint64_t trysnap = 0;
236 236
237 237 if (ds == NULL)
238 238 return (0);
239 239 /*
240 240 * The snapshot creation could fail, but that would cause an
241 241 * incorrect FALSE return, which would only result in an
242 242 * overestimation of the amount of space that an operation would
243 243 * consume, which is OK.
244 244 *
245 245 * There's also a small window where we could miss a pending
246 246 * snapshot, because we could set the sync task in the quiescing
247 247 * phase. So this should only be used as a guess.
248 248 */
249 249 if (ds->ds_trysnap_txg >
250 250 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
251 251 trysnap = ds->ds_trysnap_txg;
252 252 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
253 253 }
254 254
255 255 boolean_t
256 256 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
257 257 uint64_t blk_birth)
258 258 {
259 259 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
260 260 (bp != NULL && BP_IS_HOLE(bp)))
261 261 return (B_FALSE);
262 262
263 263 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
264 264
265 265 return (B_TRUE);
266 266 }
267 267
268 268 static void
269 269 dsl_dataset_evict(void *dbu)
270 270 {
271 271 dsl_dataset_t *ds = dbu;
272 272
273 273 ASSERT(ds->ds_owner == NULL);
274 274
275 275 ds->ds_dbuf = NULL;
276 276
277 277 unique_remove(ds->ds_fsid_guid);
278 278
279 279 if (ds->ds_objset != NULL)
280 280 dmu_objset_evict(ds->ds_objset);
281 281
282 282 if (ds->ds_prev) {
283 283 dsl_dataset_rele(ds->ds_prev, ds);
284 284 ds->ds_prev = NULL;
285 285 }
286 286
287 287 bplist_destroy(&ds->ds_pending_deadlist);
288 288 if (ds->ds_deadlist.dl_os != NULL)
289 289 dsl_deadlist_close(&ds->ds_deadlist);
290 290 if (ds->ds_dir)
291 291 dsl_dir_async_rele(ds->ds_dir, ds);
292 292
293 293 ASSERT(!list_link_active(&ds->ds_synced_link));
294 294
295 295 list_destroy(&ds->ds_prop_cbs);
296 296 mutex_destroy(&ds->ds_lock);
297 297 mutex_destroy(&ds->ds_opening_lock);
298 298 mutex_destroy(&ds->ds_sendstream_lock);
299 299 refcount_destroy(&ds->ds_longholds);
300 300
301 301 kmem_free(ds, sizeof (dsl_dataset_t));
302 302 }
303 303
304 304 int
305 305 dsl_dataset_get_snapname(dsl_dataset_t *ds)
306 306 {
307 307 dsl_dataset_phys_t *headphys;
308 308 int err;
309 309 dmu_buf_t *headdbuf;
310 310 dsl_pool_t *dp = ds->ds_dir->dd_pool;
311 311 objset_t *mos = dp->dp_meta_objset;
312 312
313 313 if (ds->ds_snapname[0])
314 314 return (0);
315 315 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
316 316 return (0);
317 317
318 318 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
319 319 FTAG, &headdbuf);
320 320 if (err != 0)
321 321 return (err);
322 322 headphys = headdbuf->db_data;
323 323 err = zap_value_search(dp->dp_meta_objset,
324 324 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
325 325 dmu_buf_rele(headdbuf, FTAG);
326 326 return (err);
327 327 }
328 328
329 329 int
330 330 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
331 331 {
332 332 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
333 333 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
334 334 matchtype_t mt;
335 335 int err;
336 336
337 337 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
338 338 mt = MT_FIRST;
339 339 else
340 340 mt = MT_EXACT;
341 341
342 342 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
343 343 value, mt, NULL, 0, NULL);
344 344 if (err == ENOTSUP && mt == MT_FIRST)
345 345 err = zap_lookup(mos, snapobj, name, 8, 1, value);
346 346 return (err);
347 347 }
348 348
349 349 int
350 350 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
351 351 boolean_t adj_cnt)
352 352 {
353 353 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
354 354 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
355 355 matchtype_t mt;
356 356 int err;
357 357
358 358 dsl_dir_snap_cmtime_update(ds->ds_dir);
359 359
360 360 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
361 361 mt = MT_FIRST;
362 362 else
363 363 mt = MT_EXACT;
364 364
365 365 err = zap_remove_norm(mos, snapobj, name, mt, tx);
366 366 if (err == ENOTSUP && mt == MT_FIRST)
367 367 err = zap_remove(mos, snapobj, name, tx);
368 368
369 369 if (err == 0 && adj_cnt)
370 370 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
371 371 DD_FIELD_SNAPSHOT_COUNT, tx);
372 372
373 373 return (err);
374 374 }
375 375
376 376 boolean_t
377 377 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
378 378 {
379 379 dmu_buf_t *dbuf = ds->ds_dbuf;
380 380 boolean_t result = B_FALSE;
381 381
382 382 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
383 383 ds->ds_object, DMU_BONUS_BLKID, tag)) {
384 384
385 385 if (ds == dmu_buf_get_user(dbuf))
386 386 result = B_TRUE;
387 387 else
388 388 dmu_buf_rele(dbuf, tag);
389 389 }
390 390
391 391 return (result);
392 392 }
393 393
394 394 int
395 395 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
396 396 dsl_dataset_t **dsp)
397 397 {
398 398 objset_t *mos = dp->dp_meta_objset;
399 399 dmu_buf_t *dbuf;
400 400 dsl_dataset_t *ds;
401 401 int err;
402 402 dmu_object_info_t doi;
403 403
404 404 ASSERT(dsl_pool_config_held(dp));
405 405
406 406 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
407 407 if (err != 0)
408 408 return (err);
409 409
410 410 /* Make sure dsobj has the correct object type. */
411 411 dmu_object_info_from_db(dbuf, &doi);
412 412 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
413 413 dmu_buf_rele(dbuf, tag);
414 414 return (SET_ERROR(EINVAL));
415 415 }
416 416
417 417 ds = dmu_buf_get_user(dbuf);
418 418 if (ds == NULL) {
419 419 dsl_dataset_t *winner = NULL;
420 420
421 421 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
422 422 ds->ds_dbuf = dbuf;
423 423 ds->ds_object = dsobj;
424 424 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
425 425
426 426 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
427 427 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
428 428 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
429 429 refcount_create(&ds->ds_longholds);
430 430
431 431 bplist_create(&ds->ds_pending_deadlist);
432 432 dsl_deadlist_open(&ds->ds_deadlist,
433 433 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
434 434
435 435 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
436 436 offsetof(dmu_sendarg_t, dsa_link));
437 437
438 438 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
439 439 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
440 440
441 441 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
442 442 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
443 443 if (!(spa_feature_table[f].fi_flags &
444 444 ZFEATURE_FLAG_PER_DATASET))
445 445 continue;
446 446 err = zap_contains(mos, dsobj,
447 447 spa_feature_table[f].fi_guid);
448 448 if (err == 0) {
449 449 ds->ds_feature_inuse[f] = B_TRUE;
450 450 } else {
451 451 ASSERT3U(err, ==, ENOENT);
452 452 err = 0;
453 453 }
454 454 }
455 455 }
456 456
457 457 err = dsl_dir_hold_obj(dp,
458 458 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
459 459 if (err != 0) {
460 460 mutex_destroy(&ds->ds_lock);
461 461 mutex_destroy(&ds->ds_opening_lock);
462 462 mutex_destroy(&ds->ds_sendstream_lock);
463 463 refcount_destroy(&ds->ds_longholds);
464 464 bplist_destroy(&ds->ds_pending_deadlist);
465 465 dsl_deadlist_close(&ds->ds_deadlist);
466 466 kmem_free(ds, sizeof (dsl_dataset_t));
467 467 dmu_buf_rele(dbuf, tag);
468 468 return (err);
469 469 }
470 470
471 471 if (!ds->ds_is_snapshot) {
472 472 ds->ds_snapname[0] = '\0';
473 473 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
474 474 err = dsl_dataset_hold_obj(dp,
475 475 dsl_dataset_phys(ds)->ds_prev_snap_obj,
476 476 ds, &ds->ds_prev);
477 477 }
478 478 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
479 479 int zaperr = zap_lookup(mos, ds->ds_object,
480 480 DS_FIELD_BOOKMARK_NAMES,
481 481 sizeof (ds->ds_bookmarks), 1,
482 482 &ds->ds_bookmarks);
483 483 if (zaperr != ENOENT)
484 484 VERIFY0(zaperr);
485 485 }
486 486 } else {
487 487 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
488 488 err = dsl_dataset_get_snapname(ds);
489 489 if (err == 0 &&
490 490 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
491 491 err = zap_count(
492 492 ds->ds_dir->dd_pool->dp_meta_objset,
493 493 dsl_dataset_phys(ds)->ds_userrefs_obj,
494 494 &ds->ds_userrefs);
495 495 }
496 496 }
497 497
498 498 if (err == 0 && !ds->ds_is_snapshot) {
499 499 err = dsl_prop_get_int_ds(ds,
500 500 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
501 501 &ds->ds_reserved);
502 502 if (err == 0) {
503 503 err = dsl_prop_get_int_ds(ds,
504 504 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
505 505 &ds->ds_quota);
506 506 }
507 507 } else {
508 508 ds->ds_reserved = ds->ds_quota = 0;
509 509 }
510 510
511 511 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
512 512 if (err == 0)
513 513 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
514 514
515 515 if (err != 0 || winner != NULL) {
516 516 bplist_destroy(&ds->ds_pending_deadlist);
517 517 dsl_deadlist_close(&ds->ds_deadlist);
518 518 if (ds->ds_prev)
519 519 dsl_dataset_rele(ds->ds_prev, ds);
520 520 dsl_dir_rele(ds->ds_dir, ds);
521 521 mutex_destroy(&ds->ds_lock);
522 522 mutex_destroy(&ds->ds_opening_lock);
523 523 mutex_destroy(&ds->ds_sendstream_lock);
524 524 refcount_destroy(&ds->ds_longholds);
525 525 kmem_free(ds, sizeof (dsl_dataset_t));
526 526 if (err != 0) {
527 527 dmu_buf_rele(dbuf, tag);
528 528 return (err);
529 529 }
530 530 ds = winner;
531 531 } else {
532 532 ds->ds_fsid_guid =
533 533 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
534 534 }
535 535 }
536 536 ASSERT3P(ds->ds_dbuf, ==, dbuf);
537 537 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
538 538 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
539 539 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
540 540 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
541 541 *dsp = ds;
542 542 return (0);
543 543 }
544 544
545 545 int
546 546 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
547 547 void *tag, dsl_dataset_t **dsp)
548 548 {
549 549 dsl_dir_t *dd;
550 550 const char *snapname;
551 551 uint64_t obj;
552 552 int err = 0;
553 553 dsl_dataset_t *ds;
554 554
555 555 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
556 556 if (err != 0)
557 557 return (err);
558 558
559 559 ASSERT(dsl_pool_config_held(dp));
560 560 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
561 561 if (obj != 0)
562 562 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
563 563 else
564 564 err = SET_ERROR(ENOENT);
565 565
566 566 /* we may be looking for a snapshot */
567 567 if (err == 0 && snapname != NULL) {
568 568 dsl_dataset_t *snap_ds;
569 569
570 570 if (*snapname++ != '@') {
571 571 dsl_dataset_rele(ds, tag);
572 572 dsl_dir_rele(dd, FTAG);
573 573 return (SET_ERROR(ENOENT));
574 574 }
575 575
576 576 dprintf("looking for snapshot '%s'\n", snapname);
577 577 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
578 578 if (err == 0)
579 579 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
580 580 dsl_dataset_rele(ds, tag);
581 581
582 582 if (err == 0) {
583 583 mutex_enter(&snap_ds->ds_lock);
584 584 if (snap_ds->ds_snapname[0] == 0)
585 585 (void) strlcpy(snap_ds->ds_snapname, snapname,
586 586 sizeof (snap_ds->ds_snapname));
587 587 mutex_exit(&snap_ds->ds_lock);
588 588 ds = snap_ds;
589 589 }
590 590 }
591 591 if (err == 0)
592 592 *dsp = ds;
593 593 dsl_dir_rele(dd, FTAG);
594 594 return (err);
595 595 }
596 596
597 597 int
598 598 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
599 599 void *tag, dsl_dataset_t **dsp)
600 600 {
601 601 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
602 602 if (err != 0)
603 603 return (err);
604 604 if (!dsl_dataset_tryown(*dsp, tag)) {
605 605 dsl_dataset_rele(*dsp, tag);
606 606 *dsp = NULL;
607 607 return (SET_ERROR(EBUSY));
608 608 }
609 609 return (0);
610 610 }
611 611
612 612 int
613 613 dsl_dataset_own(dsl_pool_t *dp, const char *name,
614 614 void *tag, dsl_dataset_t **dsp)
615 615 {
616 616 int err = dsl_dataset_hold(dp, name, tag, dsp);
617 617 if (err != 0)
618 618 return (err);
619 619 if (!dsl_dataset_tryown(*dsp, tag)) {
620 620 dsl_dataset_rele(*dsp, tag);
621 621 return (SET_ERROR(EBUSY));
622 622 }
623 623 return (0);
624 624 }
625 625
626 626 /*
627 627 * See the comment above dsl_pool_hold() for details. In summary, a long
628 628 * hold is used to prevent destruction of a dataset while the pool hold
629 629 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
630 630 *
631 631 * The dataset and pool must be held when this function is called. After it
632 632 * is called, the pool hold may be released while the dataset is still held
633 633 * and accessed.
634 634 */
635 635 void
636 636 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
637 637 {
638 638 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
639 639 (void) refcount_add(&ds->ds_longholds, tag);
640 640 }
641 641
642 642 void
643 643 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
644 644 {
645 645 (void) refcount_remove(&ds->ds_longholds, tag);
646 646 }
647 647
648 648 /* Return B_TRUE if there are any long holds on this dataset. */
649 649 boolean_t
650 650 dsl_dataset_long_held(dsl_dataset_t *ds)
651 651 {
652 652 return (!refcount_is_zero(&ds->ds_longholds));
653 653 }
654 654
655 655 void
656 656 dsl_dataset_name(dsl_dataset_t *ds, char *name)
657 657 {
658 658 if (ds == NULL) {
659 659 (void) strcpy(name, "mos");
660 660 } else {
661 661 dsl_dir_name(ds->ds_dir, name);
662 662 VERIFY0(dsl_dataset_get_snapname(ds));
663 663 if (ds->ds_snapname[0]) {
664 664 (void) strcat(name, "@");
665 665 /*
666 666 * We use a "recursive" mutex so that we
667 667 * can call dprintf_ds() with ds_lock held.
668 668 */
669 669 if (!MUTEX_HELD(&ds->ds_lock)) {
670 670 mutex_enter(&ds->ds_lock);
671 671 (void) strcat(name, ds->ds_snapname);
672 672 mutex_exit(&ds->ds_lock);
673 673 } else {
674 674 (void) strcat(name, ds->ds_snapname);
675 675 }
676 676 }
677 677 }
678 678 }
679 679
680 680 void
681 681 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
682 682 {
683 683 dmu_buf_rele(ds->ds_dbuf, tag);
684 684 }
685 685
686 686 void
687 687 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
688 688 {
689 689 ASSERT3P(ds->ds_owner, ==, tag);
690 690 ASSERT(ds->ds_dbuf != NULL);
691 691
692 692 mutex_enter(&ds->ds_lock);
693 693 ds->ds_owner = NULL;
694 694 mutex_exit(&ds->ds_lock);
695 695 dsl_dataset_long_rele(ds, tag);
696 696 dsl_dataset_rele(ds, tag);
697 697 }
698 698
699 699 boolean_t
700 700 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
701 701 {
702 702 boolean_t gotit = FALSE;
703 703
704 704 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
705 705 mutex_enter(&ds->ds_lock);
706 706 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
707 707 ds->ds_owner = tag;
708 708 dsl_dataset_long_hold(ds, tag);
709 709 gotit = TRUE;
710 710 }
711 711 mutex_exit(&ds->ds_lock);
712 712 return (gotit);
713 713 }
714 714
715 715 boolean_t
716 716 dsl_dataset_has_owner(dsl_dataset_t *ds)
717 717 {
718 718 boolean_t rv;
719 719 mutex_enter(&ds->ds_lock);
720 720 rv = (ds->ds_owner != NULL);
721 721 mutex_exit(&ds->ds_lock);
722 722 return (rv);
723 723 }
724 724
725 725 static void
726 726 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
727 727 {
728 728 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
729 729 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
730 730 uint64_t zero = 0;
731 731
732 732 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
733 733
734 734 spa_feature_incr(spa, f, tx);
735 735 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
736 736
737 737 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
738 738 sizeof (zero), 1, &zero, tx));
739 739 }
740 740
741 741 void
742 742 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
743 743 {
744 744 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
745 745 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
746 746
747 747 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
748 748
749 749 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
750 750 spa_feature_decr(spa, f, tx);
751 751 }
752 752
753 753 uint64_t
754 754 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
755 755 uint64_t flags, dmu_tx_t *tx)
756 756 {
757 757 dsl_pool_t *dp = dd->dd_pool;
758 758 dmu_buf_t *dbuf;
759 759 dsl_dataset_phys_t *dsphys;
760 760 uint64_t dsobj;
761 761 objset_t *mos = dp->dp_meta_objset;
762 762
763 763 if (origin == NULL)
764 764 origin = dp->dp_origin_snap;
765 765
766 766 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
767 767 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
768 768 ASSERT(dmu_tx_is_syncing(tx));
769 769 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
770 770
771 771 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
772 772 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
773 773 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
774 774 dmu_buf_will_dirty(dbuf, tx);
775 775 dsphys = dbuf->db_data;
776 776 bzero(dsphys, sizeof (dsl_dataset_phys_t));
777 777 dsphys->ds_dir_obj = dd->dd_object;
778 778 dsphys->ds_flags = flags;
779 779 dsphys->ds_fsid_guid = unique_create();
780 780 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
781 781 sizeof (dsphys->ds_guid));
782 782 dsphys->ds_snapnames_zapobj =
783 783 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
784 784 DMU_OT_NONE, 0, tx);
785 785 dsphys->ds_creation_time = gethrestime_sec();
786 786 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
787 787
788 788 if (origin == NULL) {
789 789 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
790 790 } else {
791 791 dsl_dataset_t *ohds; /* head of the origin snapshot */
792 792
793 793 dsphys->ds_prev_snap_obj = origin->ds_object;
794 794 dsphys->ds_prev_snap_txg =
795 795 dsl_dataset_phys(origin)->ds_creation_txg;
796 796 dsphys->ds_referenced_bytes =
797 797 dsl_dataset_phys(origin)->ds_referenced_bytes;
798 798 dsphys->ds_compressed_bytes =
799 799 dsl_dataset_phys(origin)->ds_compressed_bytes;
800 800 dsphys->ds_uncompressed_bytes =
801 801 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
802 802 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
803 803
804 804 /*
805 805 * Inherit flags that describe the dataset's contents
806 806 * (INCONSISTENT) or properties (Case Insensitive).
807 807 */
808 808 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
809 809 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
810 810
811 811 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
812 812 if (origin->ds_feature_inuse[f])
813 813 dsl_dataset_activate_feature(dsobj, f, tx);
814 814 }
815 815
816 816 dmu_buf_will_dirty(origin->ds_dbuf, tx);
817 817 dsl_dataset_phys(origin)->ds_num_children++;
818 818
819 819 VERIFY0(dsl_dataset_hold_obj(dp,
820 820 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
821 821 FTAG, &ohds));
822 822 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
823 823 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
824 824 dsl_dataset_rele(ohds, FTAG);
825 825
826 826 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
827 827 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
828 828 dsl_dataset_phys(origin)->ds_next_clones_obj =
829 829 zap_create(mos,
830 830 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
831 831 }
832 832 VERIFY0(zap_add_int(mos,
833 833 dsl_dataset_phys(origin)->ds_next_clones_obj,
834 834 dsobj, tx));
835 835 }
836 836
837 837 dmu_buf_will_dirty(dd->dd_dbuf, tx);
838 838 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
839 839 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
840 840 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
841 841 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
842 842 dsl_dir_phys(origin->ds_dir)->dd_clones =
843 843 zap_create(mos,
844 844 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
845 845 }
846 846 VERIFY0(zap_add_int(mos,
847 847 dsl_dir_phys(origin->ds_dir)->dd_clones,
848 848 dsobj, tx));
849 849 }
850 850 }
851 851
852 852 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
853 853 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
854 854
855 855 dmu_buf_rele(dbuf, FTAG);
856 856
857 857 dmu_buf_will_dirty(dd->dd_dbuf, tx);
858 858 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
859 859
860 860 return (dsobj);
861 861 }
862 862
863 863 static void
864 864 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
865 865 {
866 866 objset_t *os;
867 867
868 868 VERIFY0(dmu_objset_from_ds(ds, &os));
869 869 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
870 870 dsl_dataset_dirty(ds, tx);
871 871 }
872 872
873 873 uint64_t
874 874 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
875 875 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
876 876 {
877 877 dsl_pool_t *dp = pdd->dd_pool;
878 878 uint64_t dsobj, ddobj;
879 879 dsl_dir_t *dd;
880 880
881 881 ASSERT(dmu_tx_is_syncing(tx));
882 882 ASSERT(lastname[0] != '@');
883 883
884 884 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
885 885 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
886 886
887 887 dsobj = dsl_dataset_create_sync_dd(dd, origin,
888 888 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
889 889
890 890 dsl_deleg_set_create_perms(dd, tx, cr);
891 891
892 892 /*
893 893 * Since we're creating a new node we know it's a leaf, so we can
894 894 * initialize the counts if the limit feature is active.
895 895 */
896 896 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
897 897 uint64_t cnt = 0;
898 898 objset_t *os = dd->dd_pool->dp_meta_objset;
899 899
900 900 dsl_dir_zapify(dd, tx);
901 901 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
902 902 sizeof (cnt), 1, &cnt, tx));
903 903 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
904 904 sizeof (cnt), 1, &cnt, tx));
905 905 }
906 906
907 907 dsl_dir_rele(dd, FTAG);
908 908
909 909 /*
910 910 * If we are creating a clone, make sure we zero out any stale
911 911 * data from the origin snapshots zil header.
912 912 */
913 913 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
914 914 dsl_dataset_t *ds;
915 915
916 916 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
917 917 dsl_dataset_zero_zil(ds, tx);
918 918 dsl_dataset_rele(ds, FTAG);
919 919 }
920 920
921 921 return (dsobj);
922 922 }
923 923
924 924 /*
925 925 * The unique space in the head dataset can be calculated by subtracting
926 926 * the space used in the most recent snapshot, that is still being used
927 927 * in this file system, from the space currently in use. To figure out
928 928 * the space in the most recent snapshot still in use, we need to take
929 929 * the total space used in the snapshot and subtract out the space that
930 930 * has been freed up since the snapshot was taken.
931 931 */
932 932 void
933 933 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
934 934 {
935 935 uint64_t mrs_used;
936 936 uint64_t dlused, dlcomp, dluncomp;
937 937
938 938 ASSERT(!ds->ds_is_snapshot);
939 939
940 940 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
941 941 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
942 942 else
943 943 mrs_used = 0;
944 944
945 945 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
946 946
947 947 ASSERT3U(dlused, <=, mrs_used);
948 948 dsl_dataset_phys(ds)->ds_unique_bytes =
949 949 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
950 950
951 951 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
952 952 SPA_VERSION_UNIQUE_ACCURATE)
953 953 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
954 954 }
955 955
956 956 void
957 957 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
958 958 dmu_tx_t *tx)
959 959 {
960 960 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
961 961 uint64_t count;
962 962 int err;
963 963
964 964 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
965 965 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
966 966 obj, tx);
967 967 /*
968 968 * The err should not be ENOENT, but a bug in a previous version
969 969 * of the code could cause upgrade_clones_cb() to not set
970 970 * ds_next_snap_obj when it should, leading to a missing entry.
971 971 * If we knew that the pool was created after
972 972 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
973 973 * ENOENT. However, at least we can check that we don't have
974 974 * too many entries in the next_clones_obj even after failing to
975 975 * remove this one.
976 976 */
977 977 if (err != ENOENT)
978 978 VERIFY0(err);
979 979 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
980 980 &count));
981 981 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
982 982 }
983 983
984 984
985 985 blkptr_t *
986 986 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
987 987 {
988 988 return (&dsl_dataset_phys(ds)->ds_bp);
989 989 }
990 990
991 991 void
992 992 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
993 993 {
994 994 ASSERT(dmu_tx_is_syncing(tx));
995 995 /* If it's the meta-objset, set dp_meta_rootbp */
996 996 if (ds == NULL) {
997 997 tx->tx_pool->dp_meta_rootbp = *bp;
998 998 } else {
999 999 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1000 1000 dsl_dataset_phys(ds)->ds_bp = *bp;
1001 1001 }
1002 1002 }
1003 1003
1004 1004 spa_t *
1005 1005 dsl_dataset_get_spa(dsl_dataset_t *ds)
1006 1006 {
1007 1007 return (ds->ds_dir->dd_pool->dp_spa);
1008 1008 }
1009 1009
1010 1010 void
1011 1011 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1012 1012 {
1013 1013 dsl_pool_t *dp;
1014 1014
1015 1015 if (ds == NULL) /* this is the meta-objset */
1016 1016 return;
1017 1017
1018 1018 ASSERT(ds->ds_objset != NULL);
1019 1019
1020 1020 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1021 1021 panic("dirtying snapshot!");
1022 1022
1023 1023 dp = ds->ds_dir->dd_pool;
1024 1024
1025 1025 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1026 1026 /* up the hold count until we can be written out */
1027 1027 dmu_buf_add_ref(ds->ds_dbuf, ds);
1028 1028 }
1029 1029 }
1030 1030
1031 1031 boolean_t
1032 1032 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1033 1033 {
1034 1034 for (int t = 0; t < TXG_SIZE; t++) {
1035 1035 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1036 1036 ds, t))
1037 1037 return (B_TRUE);
1038 1038 }
1039 1039 return (B_FALSE);
1040 1040 }
1041 1041
1042 1042 static int
1043 1043 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1044 1044 {
1045 1045 uint64_t asize;
1046 1046
1047 1047 if (!dmu_tx_is_syncing(tx))
1048 1048 return (0);
1049 1049
1050 1050 /*
1051 1051 * If there's an fs-only reservation, any blocks that might become
1052 1052 * owned by the snapshot dataset must be accommodated by space
1053 1053 * outside of the reservation.
1054 1054 */
1055 1055 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1056 1056 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1057 1057 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1058 1058 return (SET_ERROR(ENOSPC));
1059 1059
1060 1060 /*
1061 1061 * Propagate any reserved space for this snapshot to other
1062 1062 * snapshot checks in this sync group.
1063 1063 */
1064 1064 if (asize > 0)
1065 1065 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1066 1066
1067 1067 return (0);
1068 1068 }
1069 1069
1070 1070 typedef struct dsl_dataset_snapshot_arg {
1071 1071 nvlist_t *ddsa_snaps;
1072 1072 nvlist_t *ddsa_props;
1073 1073 nvlist_t *ddsa_errors;
1074 1074 cred_t *ddsa_cr;
1075 1075 } dsl_dataset_snapshot_arg_t;
1076 1076
1077 1077 int
1078 1078 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1079 1079 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1080 1080 {
1081 1081 int error;
1082 1082 uint64_t value;
1083 1083
1084 1084 ds->ds_trysnap_txg = tx->tx_txg;
1085 1085
1086 1086 if (!dmu_tx_is_syncing(tx))
1087 1087 return (0);
1088 1088
1089 1089 /*
1090 1090 * We don't allow multiple snapshots of the same txg. If there
1091 1091 * is already one, try again.
1092 1092 */
1093 1093 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1094 1094 return (SET_ERROR(EAGAIN));
1095 1095
1096 1096 /*
1097 1097 * Check for conflicting snapshot name.
1098 1098 */
1099 1099 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1100 1100 if (error == 0)
1101 1101 return (SET_ERROR(EEXIST));
1102 1102 if (error != ENOENT)
1103 1103 return (error);
1104 1104
1105 1105 /*
1106 1106 * We don't allow taking snapshots of inconsistent datasets, such as
1107 1107 * those into which we are currently receiving. However, if we are
1108 1108 * creating this snapshot as part of a receive, this check will be
1109 1109 * executed atomically with respect to the completion of the receive
1110 1110 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1111 1111 * case we ignore this, knowing it will be fixed up for us shortly in
1112 1112 * dmu_recv_end_sync().
1113 1113 */
1114 1114 if (!recv && DS_IS_INCONSISTENT(ds))
1115 1115 return (SET_ERROR(EBUSY));
1116 1116
1117 1117 /*
1118 1118 * Skip the check for temporary snapshots or if we have already checked
1119 1119 * the counts in dsl_dataset_snapshot_check. This means we really only
1120 1120 * check the count here when we're receiving a stream.
1121 1121 */
1122 1122 if (cnt != 0 && cr != NULL) {
1123 1123 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1124 1124 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1125 1125 if (error != 0)
1126 1126 return (error);
1127 1127 }
1128 1128
1129 1129 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1130 1130 if (error != 0)
1131 1131 return (error);
1132 1132
1133 1133 return (0);
1134 1134 }
1135 1135
1136 1136 static int
1137 1137 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1138 1138 {
1139 1139 dsl_dataset_snapshot_arg_t *ddsa = arg;
1140 1140 dsl_pool_t *dp = dmu_tx_pool(tx);
1141 1141 nvpair_t *pair;
1142 1142 int rv = 0;
1143 1143
1144 1144 /*
1145 1145 * Pre-compute how many total new snapshots will be created for each
1146 1146 * level in the tree and below. This is needed for validating the
1147 1147 * snapshot limit when either taking a recursive snapshot or when
1148 1148 * taking multiple snapshots.
1149 1149 *
1150 1150 * The problem is that the counts are not actually adjusted when
1151 1151 * we are checking, only when we finally sync. For a single snapshot,
1152 1152 * this is easy, the count will increase by 1 at each node up the tree,
1153 1153 * but its more complicated for the recursive/multiple snapshot case.
1154 1154 *
1155 1155 * The dsl_fs_ss_limit_check function does recursively check the count
1156 1156 * at each level up the tree but since it is validating each snapshot
1157 1157 * independently we need to be sure that we are validating the complete
1158 1158 * count for the entire set of snapshots. We do this by rolling up the
1159 1159 * counts for each component of the name into an nvlist and then
1160 1160 * checking each of those cases with the aggregated count.
1161 1161 *
1162 1162 * This approach properly handles not only the recursive snapshot
1163 1163 * case (where we get all of those on the ddsa_snaps list) but also
1164 1164 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1165 1165 * validate the limit on 'a' using a count of 2).
1166 1166 *
1167 1167 * We validate the snapshot names in the third loop and only report
1168 1168 * name errors once.
1169 1169 */
1170 1170 if (dmu_tx_is_syncing(tx)) {
1171 1171 nvlist_t *cnt_track = NULL;
1172 1172 cnt_track = fnvlist_alloc();
1173 1173
1174 1174 /* Rollup aggregated counts into the cnt_track list */
1175 1175 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1176 1176 pair != NULL;
1177 1177 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1178 1178 char *pdelim;
1179 1179 uint64_t val;
1180 1180 char nm[MAXPATHLEN];
1181 1181
1182 1182 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1183 1183 pdelim = strchr(nm, '@');
1184 1184 if (pdelim == NULL)
1185 1185 continue;
1186 1186 *pdelim = '\0';
1187 1187
1188 1188 do {
1189 1189 if (nvlist_lookup_uint64(cnt_track, nm,
1190 1190 &val) == 0) {
1191 1191 /* update existing entry */
1192 1192 fnvlist_add_uint64(cnt_track, nm,
1193 1193 val + 1);
1194 1194 } else {
1195 1195 /* add to list */
1196 1196 fnvlist_add_uint64(cnt_track, nm, 1);
1197 1197 }
1198 1198
1199 1199 pdelim = strrchr(nm, '/');
1200 1200 if (pdelim != NULL)
1201 1201 *pdelim = '\0';
1202 1202 } while (pdelim != NULL);
1203 1203 }
1204 1204
1205 1205 /* Check aggregated counts at each level */
1206 1206 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1207 1207 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1208 1208 int error = 0;
1209 1209 char *name;
1210 1210 uint64_t cnt = 0;
1211 1211 dsl_dataset_t *ds;
1212 1212
1213 1213 name = nvpair_name(pair);
1214 1214 cnt = fnvpair_value_uint64(pair);
1215 1215 ASSERT(cnt > 0);
1216 1216
1217 1217 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1218 1218 if (error == 0) {
1219 1219 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1220 1220 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1221 1221 ddsa->ddsa_cr);
1222 1222 dsl_dataset_rele(ds, FTAG);
1223 1223 }
1224 1224
1225 1225 if (error != 0) {
1226 1226 if (ddsa->ddsa_errors != NULL)
1227 1227 fnvlist_add_int32(ddsa->ddsa_errors,
1228 1228 name, error);
1229 1229 rv = error;
1230 1230 /* only report one error for this check */
1231 1231 break;
1232 1232 }
1233 1233 }
1234 1234 nvlist_free(cnt_track);
1235 1235 }
1236 1236
1237 1237 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1238 1238 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1239 1239 int error = 0;
1240 1240 dsl_dataset_t *ds;
1241 1241 char *name, *atp;
1242 1242 char dsname[MAXNAMELEN];
1243 1243
1244 1244 name = nvpair_name(pair);
1245 1245 if (strlen(name) >= MAXNAMELEN)
1246 1246 error = SET_ERROR(ENAMETOOLONG);
1247 1247 if (error == 0) {
1248 1248 atp = strchr(name, '@');
1249 1249 if (atp == NULL)
1250 1250 error = SET_ERROR(EINVAL);
1251 1251 if (error == 0)
1252 1252 (void) strlcpy(dsname, name, atp - name + 1);
1253 1253 }
1254 1254 if (error == 0)
1255 1255 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1256 1256 if (error == 0) {
1257 1257 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1258 1258 error = dsl_dataset_snapshot_check_impl(ds,
1259 1259 atp + 1, tx, B_FALSE, 0, NULL);
1260 1260 dsl_dataset_rele(ds, FTAG);
1261 1261 }
1262 1262
1263 1263 if (error != 0) {
1264 1264 if (ddsa->ddsa_errors != NULL) {
1265 1265 fnvlist_add_int32(ddsa->ddsa_errors,
1266 1266 name, error);
1267 1267 }
1268 1268 rv = error;
1269 1269 }
1270 1270 }
1271 1271
1272 1272 return (rv);
1273 1273 }
1274 1274
1275 1275 void
1276 1276 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1277 1277 dmu_tx_t *tx)
1278 1278 {
1279 1279 static zil_header_t zero_zil;
1280 1280
1281 1281 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1282 1282 dmu_buf_t *dbuf;
1283 1283 dsl_dataset_phys_t *dsphys;
1284 1284 uint64_t dsobj, crtxg;
1285 1285 objset_t *mos = dp->dp_meta_objset;
1286 1286 objset_t *os;
1287 1287
1288 1288 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1289 1289
1290 1290 /*
1291 1291 * If we are on an old pool, the zil must not be active, in which
1292 1292 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1293 1293 */
1294 1294 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1295 1295 dmu_objset_from_ds(ds, &os) != 0 ||
1296 1296 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1297 1297 sizeof (zero_zil)) == 0);
1298 1298
1299 1299 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1300 1300
1301 1301 /*
1302 1302 * The origin's ds_creation_txg has to be < TXG_INITIAL
1303 1303 */
1304 1304 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1305 1305 crtxg = 1;
1306 1306 else
1307 1307 crtxg = tx->tx_txg;
1308 1308
1309 1309 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1310 1310 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1311 1311 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1312 1312 dmu_buf_will_dirty(dbuf, tx);
1313 1313 dsphys = dbuf->db_data;
1314 1314 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1315 1315 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1316 1316 dsphys->ds_fsid_guid = unique_create();
1317 1317 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1318 1318 sizeof (dsphys->ds_guid));
1319 1319 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1320 1320 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1321 1321 dsphys->ds_next_snap_obj = ds->ds_object;
1322 1322 dsphys->ds_num_children = 1;
1323 1323 dsphys->ds_creation_time = gethrestime_sec();
1324 1324 dsphys->ds_creation_txg = crtxg;
1325 1325 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1326 1326 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1327 1327 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1328 1328 dsphys->ds_uncompressed_bytes =
1329 1329 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1330 1330 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1331 1331 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1332 1332 dmu_buf_rele(dbuf, FTAG);
1333 1333
1334 1334 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1335 1335 if (ds->ds_feature_inuse[f])
1336 1336 dsl_dataset_activate_feature(dsobj, f, tx);
1337 1337 }
1338 1338
1339 1339 ASSERT3U(ds->ds_prev != 0, ==,
1340 1340 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1341 1341 if (ds->ds_prev) {
1342 1342 uint64_t next_clones_obj =
1343 1343 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1344 1344 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1345 1345 ds->ds_object ||
1346 1346 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1347 1347 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1348 1348 ds->ds_object) {
1349 1349 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1350 1350 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1351 1351 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1352 1352 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1353 1353 } else if (next_clones_obj != 0) {
1354 1354 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1355 1355 dsphys->ds_next_snap_obj, tx);
1356 1356 VERIFY0(zap_add_int(mos,
1357 1357 next_clones_obj, dsobj, tx));
1358 1358 }
1359 1359 }
1360 1360
1361 1361 /*
1362 1362 * If we have a reference-reservation on this dataset, we will
1363 1363 * need to increase the amount of refreservation being charged
1364 1364 * since our unique space is going to zero.
1365 1365 */
1366 1366 if (ds->ds_reserved) {
1367 1367 int64_t delta;
1368 1368 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1369 1369 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1370 1370 ds->ds_reserved);
1371 1371 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1372 1372 delta, 0, 0, tx);
1373 1373 }
1374 1374
1375 1375 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1376 1376 dsl_dataset_phys(ds)->ds_deadlist_obj =
1377 1377 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1378 1378 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1379 1379 dsl_deadlist_close(&ds->ds_deadlist);
1380 1380 dsl_deadlist_open(&ds->ds_deadlist, mos,
1381 1381 dsl_dataset_phys(ds)->ds_deadlist_obj);
1382 1382 dsl_deadlist_add_key(&ds->ds_deadlist,
1383 1383 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1384 1384
1385 1385 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1386 1386 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1387 1387 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1388 1388 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1389 1389 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1390 1390 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1391 1391
1392 1392 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1393 1393 snapname, 8, 1, &dsobj, tx));
1394 1394
1395 1395 if (ds->ds_prev)
1396 1396 dsl_dataset_rele(ds->ds_prev, ds);
1397 1397 VERIFY0(dsl_dataset_hold_obj(dp,
1398 1398 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1399 1399
1400 1400 dsl_scan_ds_snapshotted(ds, tx);
1401 1401
1402 1402 dsl_dir_snap_cmtime_update(ds->ds_dir);
1403 1403
1404 1404 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1405 1405 }
1406 1406
1407 1407 static void
1408 1408 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1409 1409 {
1410 1410 dsl_dataset_snapshot_arg_t *ddsa = arg;
1411 1411 dsl_pool_t *dp = dmu_tx_pool(tx);
1412 1412 nvpair_t *pair;
1413 1413
1414 1414 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1415 1415 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1416 1416 dsl_dataset_t *ds;
1417 1417 char *name, *atp;
1418 1418 char dsname[MAXNAMELEN];
1419 1419
1420 1420 name = nvpair_name(pair);
1421 1421 atp = strchr(name, '@');
1422 1422 (void) strlcpy(dsname, name, atp - name + 1);
1423 1423 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1424 1424
1425 1425 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1426 1426 if (ddsa->ddsa_props != NULL) {
1427 1427 dsl_props_set_sync_impl(ds->ds_prev,
1428 1428 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1429 1429 }
1430 1430 dsl_dataset_rele(ds, FTAG);
1431 1431 }
1432 1432 }
1433 1433
1434 1434 /*
1435 1435 * The snapshots must all be in the same pool.
1436 1436 * All-or-nothing: if there are any failures, nothing will be modified.
1437 1437 */
1438 1438 int
1439 1439 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1440 1440 {
1441 1441 dsl_dataset_snapshot_arg_t ddsa;
1442 1442 nvpair_t *pair;
1443 1443 boolean_t needsuspend;
1444 1444 int error;
1445 1445 spa_t *spa;
1446 1446 char *firstname;
1447 1447 nvlist_t *suspended = NULL;
1448 1448
1449 1449 pair = nvlist_next_nvpair(snaps, NULL);
1450 1450 if (pair == NULL)
1451 1451 return (0);
1452 1452 firstname = nvpair_name(pair);
1453 1453
1454 1454 error = spa_open(firstname, &spa, FTAG);
1455 1455 if (error != 0)
1456 1456 return (error);
1457 1457 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1458 1458 spa_close(spa, FTAG);
1459 1459
1460 1460 if (needsuspend) {
1461 1461 suspended = fnvlist_alloc();
1462 1462 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1463 1463 pair = nvlist_next_nvpair(snaps, pair)) {
1464 1464 char fsname[MAXNAMELEN];
1465 1465 char *snapname = nvpair_name(pair);
1466 1466 char *atp;
1467 1467 void *cookie;
1468 1468
1469 1469 atp = strchr(snapname, '@');
1470 1470 if (atp == NULL) {
1471 1471 error = SET_ERROR(EINVAL);
1472 1472 break;
1473 1473 }
1474 1474 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1475 1475
1476 1476 error = zil_suspend(fsname, &cookie);
1477 1477 if (error != 0)
1478 1478 break;
1479 1479 fnvlist_add_uint64(suspended, fsname,
1480 1480 (uintptr_t)cookie);
1481 1481 }
1482 1482 }
1483 1483
1484 1484 ddsa.ddsa_snaps = snaps;
1485 1485 ddsa.ddsa_props = props;
1486 1486 ddsa.ddsa_errors = errors;
1487 1487 ddsa.ddsa_cr = CRED();
1488 1488
1489 1489 if (error == 0) {
1490 1490 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1491 1491 dsl_dataset_snapshot_sync, &ddsa,
1492 1492 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1493 1493 }
1494 1494
1495 1495 if (suspended != NULL) {
1496 1496 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1497 1497 pair = nvlist_next_nvpair(suspended, pair)) {
1498 1498 zil_resume((void *)(uintptr_t)
1499 1499 fnvpair_value_uint64(pair));
1500 1500 }
1501 1501 fnvlist_free(suspended);
1502 1502 }
1503 1503
1504 1504 return (error);
1505 1505 }
1506 1506
1507 1507 typedef struct dsl_dataset_snapshot_tmp_arg {
1508 1508 const char *ddsta_fsname;
1509 1509 const char *ddsta_snapname;
1510 1510 minor_t ddsta_cleanup_minor;
1511 1511 const char *ddsta_htag;
1512 1512 } dsl_dataset_snapshot_tmp_arg_t;
1513 1513
1514 1514 static int
1515 1515 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1516 1516 {
1517 1517 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1518 1518 dsl_pool_t *dp = dmu_tx_pool(tx);
1519 1519 dsl_dataset_t *ds;
1520 1520 int error;
1521 1521
1522 1522 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1523 1523 if (error != 0)
1524 1524 return (error);
1525 1525
1526 1526 /* NULL cred means no limit check for tmp snapshot */
1527 1527 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1528 1528 tx, B_FALSE, 0, NULL);
1529 1529 if (error != 0) {
1530 1530 dsl_dataset_rele(ds, FTAG);
1531 1531 return (error);
1532 1532 }
1533 1533
1534 1534 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1535 1535 dsl_dataset_rele(ds, FTAG);
1536 1536 return (SET_ERROR(ENOTSUP));
1537 1537 }
1538 1538 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1539 1539 B_TRUE, tx);
1540 1540 if (error != 0) {
1541 1541 dsl_dataset_rele(ds, FTAG);
1542 1542 return (error);
1543 1543 }
1544 1544
1545 1545 dsl_dataset_rele(ds, FTAG);
1546 1546 return (0);
1547 1547 }
1548 1548
1549 1549 static void
1550 1550 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1551 1551 {
1552 1552 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1553 1553 dsl_pool_t *dp = dmu_tx_pool(tx);
1554 1554 dsl_dataset_t *ds;
1555 1555
1556 1556 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1557 1557
1558 1558 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1559 1559 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1560 1560 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1561 1561 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1562 1562
1563 1563 dsl_dataset_rele(ds, FTAG);
1564 1564 }
1565 1565
1566 1566 int
1567 1567 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1568 1568 minor_t cleanup_minor, const char *htag)
1569 1569 {
1570 1570 dsl_dataset_snapshot_tmp_arg_t ddsta;
1571 1571 int error;
1572 1572 spa_t *spa;
1573 1573 boolean_t needsuspend;
1574 1574 void *cookie;
1575 1575
1576 1576 ddsta.ddsta_fsname = fsname;
1577 1577 ddsta.ddsta_snapname = snapname;
1578 1578 ddsta.ddsta_cleanup_minor = cleanup_minor;
1579 1579 ddsta.ddsta_htag = htag;
1580 1580
1581 1581 error = spa_open(fsname, &spa, FTAG);
1582 1582 if (error != 0)
1583 1583 return (error);
1584 1584 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1585 1585 spa_close(spa, FTAG);
1586 1586
1587 1587 if (needsuspend) {
1588 1588 error = zil_suspend(fsname, &cookie);
1589 1589 if (error != 0)
1590 1590 return (error);
1591 1591 }
1592 1592
1593 1593 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1594 1594 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1595 1595
1596 1596 if (needsuspend)
1597 1597 zil_resume(cookie);
1598 1598 return (error);
1599 1599 }
1600 1600
1601 1601
1602 1602 void
1603 1603 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1604 1604 {
1605 1605 ASSERT(dmu_tx_is_syncing(tx));
1606 1606 ASSERT(ds->ds_objset != NULL);
1607 1607 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1608 1608
1609 1609 /*
1610 1610 * in case we had to change ds_fsid_guid when we opened it,
1611 1611 * sync it out now.
1612 1612 */
1613 1613 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1614 1614 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1615 1615
1616 1616 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1617 1617 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1618 1618 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1619 1619 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1620 1620 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1621 1621 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1622 1622 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1623 1623 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1624 1624 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1625 1625 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1626 1626 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1627 1627 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1628 1628 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1629 1629 }
1630 1630
1631 1631 dmu_objset_sync(ds->ds_objset, zio, tx);
1632 1632
1633 1633 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1634 1634 if (ds->ds_feature_activation_needed[f]) {
1635 1635 if (ds->ds_feature_inuse[f])
1636 1636 continue;
1637 1637 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1638 1638 ds->ds_feature_inuse[f] = B_TRUE;
1639 1639 }
1640 1640 }
1641 1641 }
1642 1642
1643 1643 static void
1644 1644 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1645 1645 {
1646 1646 uint64_t count = 0;
1647 1647 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1648 1648 zap_cursor_t zc;
1649 1649 zap_attribute_t za;
1650 1650 nvlist_t *propval = fnvlist_alloc();
1651 1651 nvlist_t *val = fnvlist_alloc();
1652 1652
1653 1653 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1654 1654
1655 1655 /*
1656 1656 * There may be missing entries in ds_next_clones_obj
1657 1657 * due to a bug in a previous version of the code.
1658 1658 * Only trust it if it has the right number of entries.
1659 1659 */
1660 1660 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1661 1661 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1662 1662 &count));
1663 1663 }
1664 1664 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1665 1665 goto fail;
1666 1666 for (zap_cursor_init(&zc, mos,
1667 1667 dsl_dataset_phys(ds)->ds_next_clones_obj);
1668 1668 zap_cursor_retrieve(&zc, &za) == 0;
1669 1669 zap_cursor_advance(&zc)) {
1670 1670 dsl_dataset_t *clone;
1671 1671 char buf[ZFS_MAXNAMELEN];
1672 1672 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1673 1673 za.za_first_integer, FTAG, &clone));
1674 1674 dsl_dir_name(clone->ds_dir, buf);
1675 1675 fnvlist_add_boolean(val, buf);
1676 1676 dsl_dataset_rele(clone, FTAG);
1677 1677 }
1678 1678 zap_cursor_fini(&zc);
1679 1679 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1680 1680 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1681 1681 fail:
1682 1682 nvlist_free(val);
1683 1683 nvlist_free(propval);
1684 1684 }
1685 1685
1686 1686 static void
1687 1687 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1688 1688 {
1689 1689 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1690 1690
1691 1691 if (dsl_dataset_has_resume_receive_state(ds)) {
1692 1692 char *str;
1693 1693 void *packed;
1694 1694 uint8_t *compressed;
1695 1695 uint64_t val;
1696 1696 nvlist_t *token_nv = fnvlist_alloc();
1697 1697 size_t packed_size, compressed_size;
1698 1698
1699 1699 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1700 1700 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1701 1701 fnvlist_add_uint64(token_nv, "fromguid", val);
1702 1702 }
1703 1703 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1704 1704 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1705 1705 fnvlist_add_uint64(token_nv, "object", val);
1706 1706 }
1707 1707 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1708 1708 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1709 1709 fnvlist_add_uint64(token_nv, "offset", val);
1710 1710 }
1711 1711 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1712 1712 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1713 1713 fnvlist_add_uint64(token_nv, "bytes", val);
1714 1714 }
1715 1715 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1716 1716 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1717 1717 fnvlist_add_uint64(token_nv, "toguid", val);
1718 1718 }
1719 1719 char buf[256];
1720 1720 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1721 1721 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1722 1722 fnvlist_add_string(token_nv, "toname", buf);
1723 1723 }
1724 1724 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1725 1725 DS_FIELD_RESUME_EMBEDOK) == 0) {
1726 1726 fnvlist_add_boolean(token_nv, "embedok");
1727 1727 }
1728 1728 packed = fnvlist_pack(token_nv, &packed_size);
1729 1729 fnvlist_free(token_nv);
1730 1730 compressed = kmem_alloc(packed_size, KM_SLEEP);
1731 1731
1732 1732 compressed_size = gzip_compress(packed, compressed,
1733 1733 packed_size, packed_size, 6);
1734 1734
1735 1735 zio_cksum_t cksum;
1736 1736 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1737 1737
1738 1738 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1739 1739 for (int i = 0; i < compressed_size; i++) {
1740 1740 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1741 1741 }
1742 1742 str[compressed_size * 2] = '\0';
1743 1743 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1744 1744 ZFS_SEND_RESUME_TOKEN_VERSION,
1745 1745 (longlong_t)cksum.zc_word[0],
1746 1746 (longlong_t)packed_size, str);
1747 1747 dsl_prop_nvlist_add_string(nv,
1748 1748 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1749 1749 kmem_free(packed, packed_size);
1750 1750 kmem_free(str, compressed_size * 2 + 1);
1751 1751 kmem_free(compressed, packed_size);
1752 1752 strfree(propval);
1753 1753 }
1754 1754 }
1755 1755
1756 1756 void
1757 1757 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1758 1758 {
1759 1759 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1760 1760 uint64_t refd, avail, uobjs, aobjs, ratio;
1761 1761
1762 1762 ASSERT(dsl_pool_config_held(dp));
1763 1763
1764 1764 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1765 1765 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1766 1766 dsl_dataset_phys(ds)->ds_compressed_bytes);
1767 1767
1768 1768 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1769 1769 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1770 1770 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1771 1771
1772 1772 if (ds->ds_is_snapshot) {
1773 1773 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1774 1774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1775 1775 dsl_dataset_phys(ds)->ds_unique_bytes);
1776 1776 get_clones_stat(ds, nv);
1777 1777 } else {
1778 1778 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1779 1779 char buf[MAXNAMELEN];
1780 1780 dsl_dataset_name(ds->ds_prev, buf);
1781 1781 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1782 1782 }
1783 1783
1784 1784 dsl_dir_stats(ds->ds_dir, nv);
1785 1785 }
1786 1786
1787 1787 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1788 1788 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1789 1789 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1790 1790
|
↓ open down ↓ |
1790 lines elided |
↑ open up ↑ |
1791 1791 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1792 1792 dsl_dataset_phys(ds)->ds_creation_time);
1793 1793 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1794 1794 dsl_dataset_phys(ds)->ds_creation_txg);
1795 1795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1796 1796 ds->ds_quota);
1797 1797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1798 1798 ds->ds_reserved);
1799 1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1800 1800 dsl_dataset_phys(ds)->ds_guid);
1801 + dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_FSID_GUID,
1802 + dsl_dataset_phys(ds)->ds_fsid_guid);
1801 1803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1802 1804 dsl_dataset_phys(ds)->ds_unique_bytes);
1803 1805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1804 1806 ds->ds_object);
1805 1807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1806 1808 ds->ds_userrefs);
1807 1809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1808 1810 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1809 1811
1810 1812 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1811 1813 uint64_t written, comp, uncomp;
1812 1814 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1813 1815 dsl_dataset_t *prev;
1814 1816
1815 1817 int err = dsl_dataset_hold_obj(dp,
1816 1818 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1817 1819 if (err == 0) {
1818 1820 err = dsl_dataset_space_written(prev, ds, &written,
1819 1821 &comp, &uncomp);
1820 1822 dsl_dataset_rele(prev, FTAG);
1821 1823 if (err == 0) {
1822 1824 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1823 1825 written);
1824 1826 }
1825 1827 }
1826 1828 }
1827 1829
1828 1830 if (!dsl_dataset_is_snapshot(ds)) {
1829 1831 /*
1830 1832 * A failed "newfs" (e.g. full) resumable receive leaves
1831 1833 * the stats set on this dataset. Check here for the prop.
1832 1834 */
1833 1835 get_receive_resume_stats(ds, nv);
1834 1836
1835 1837 /*
1836 1838 * A failed incremental resumable receive leaves the
1837 1839 * stats set on our child named "%recv". Check the child
1838 1840 * for the prop.
1839 1841 */
1840 1842 char recvname[ZFS_MAXNAMELEN];
1841 1843 dsl_dataset_t *recv_ds;
1842 1844 dsl_dataset_name(ds, recvname);
1843 1845 (void) strcat(recvname, "/");
1844 1846 (void) strcat(recvname, recv_clone_name);
1845 1847 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1846 1848 get_receive_resume_stats(recv_ds, nv);
1847 1849 dsl_dataset_rele(recv_ds, FTAG);
1848 1850 }
1849 1851 }
1850 1852 }
1851 1853
1852 1854 void
1853 1855 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1854 1856 {
1855 1857 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1856 1858 ASSERT(dsl_pool_config_held(dp));
1857 1859
1858 1860 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1859 1861 stat->dds_inconsistent =
1860 1862 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1861 1863 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1862 1864 stat->dds_origin[0] = '\0';
1863 1865 if (ds->ds_is_snapshot) {
1864 1866 stat->dds_is_snapshot = B_TRUE;
1865 1867 stat->dds_num_clones =
1866 1868 dsl_dataset_phys(ds)->ds_num_children - 1;
1867 1869 } else {
1868 1870 stat->dds_is_snapshot = B_FALSE;
1869 1871 stat->dds_num_clones = 0;
1870 1872
1871 1873 if (dsl_dir_is_clone(ds->ds_dir)) {
1872 1874 dsl_dataset_t *ods;
1873 1875
1874 1876 VERIFY0(dsl_dataset_hold_obj(dp,
1875 1877 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1876 1878 FTAG, &ods));
1877 1879 dsl_dataset_name(ods, stat->dds_origin);
1878 1880 dsl_dataset_rele(ods, FTAG);
1879 1881 }
1880 1882 }
1881 1883 }
1882 1884
1883 1885 uint64_t
1884 1886 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1885 1887 {
1886 1888 return (ds->ds_fsid_guid);
1887 1889 }
1888 1890
1889 1891 void
1890 1892 dsl_dataset_space(dsl_dataset_t *ds,
1891 1893 uint64_t *refdbytesp, uint64_t *availbytesp,
1892 1894 uint64_t *usedobjsp, uint64_t *availobjsp)
1893 1895 {
1894 1896 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1895 1897 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1896 1898 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1897 1899 *availbytesp +=
1898 1900 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1899 1901 if (ds->ds_quota != 0) {
1900 1902 /*
1901 1903 * Adjust available bytes according to refquota
1902 1904 */
1903 1905 if (*refdbytesp < ds->ds_quota)
1904 1906 *availbytesp = MIN(*availbytesp,
1905 1907 ds->ds_quota - *refdbytesp);
1906 1908 else
1907 1909 *availbytesp = 0;
1908 1910 }
1909 1911 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1910 1912 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1911 1913 }
1912 1914
1913 1915 boolean_t
1914 1916 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1915 1917 {
1916 1918 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1917 1919
1918 1920 ASSERT(dsl_pool_config_held(dp));
1919 1921 if (snap == NULL)
1920 1922 return (B_FALSE);
1921 1923 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1922 1924 dsl_dataset_phys(snap)->ds_creation_txg) {
1923 1925 objset_t *os, *os_snap;
1924 1926 /*
1925 1927 * It may be that only the ZIL differs, because it was
1926 1928 * reset in the head. Don't count that as being
1927 1929 * modified.
1928 1930 */
1929 1931 if (dmu_objset_from_ds(ds, &os) != 0)
1930 1932 return (B_TRUE);
1931 1933 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1932 1934 return (B_TRUE);
1933 1935 return (bcmp(&os->os_phys->os_meta_dnode,
1934 1936 &os_snap->os_phys->os_meta_dnode,
1935 1937 sizeof (os->os_phys->os_meta_dnode)) != 0);
1936 1938 }
1937 1939 return (B_FALSE);
1938 1940 }
1939 1941
1940 1942 typedef struct dsl_dataset_rename_snapshot_arg {
1941 1943 const char *ddrsa_fsname;
1942 1944 const char *ddrsa_oldsnapname;
1943 1945 const char *ddrsa_newsnapname;
1944 1946 boolean_t ddrsa_recursive;
1945 1947 dmu_tx_t *ddrsa_tx;
1946 1948 } dsl_dataset_rename_snapshot_arg_t;
1947 1949
1948 1950 /* ARGSUSED */
1949 1951 static int
1950 1952 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1951 1953 dsl_dataset_t *hds, void *arg)
1952 1954 {
1953 1955 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1954 1956 int error;
1955 1957 uint64_t val;
1956 1958
1957 1959 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1958 1960 if (error != 0) {
1959 1961 /* ignore nonexistent snapshots */
1960 1962 return (error == ENOENT ? 0 : error);
1961 1963 }
1962 1964
1963 1965 /* new name should not exist */
1964 1966 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1965 1967 if (error == 0)
1966 1968 error = SET_ERROR(EEXIST);
1967 1969 else if (error == ENOENT)
1968 1970 error = 0;
1969 1971
1970 1972 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1971 1973 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1972 1974 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1973 1975 error = SET_ERROR(ENAMETOOLONG);
1974 1976
1975 1977 return (error);
1976 1978 }
1977 1979
1978 1980 static int
1979 1981 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1980 1982 {
1981 1983 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1982 1984 dsl_pool_t *dp = dmu_tx_pool(tx);
1983 1985 dsl_dataset_t *hds;
1984 1986 int error;
1985 1987
1986 1988 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1987 1989 if (error != 0)
1988 1990 return (error);
1989 1991
1990 1992 if (ddrsa->ddrsa_recursive) {
1991 1993 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1992 1994 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1993 1995 DS_FIND_CHILDREN);
1994 1996 } else {
1995 1997 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1996 1998 }
1997 1999 dsl_dataset_rele(hds, FTAG);
1998 2000 return (error);
1999 2001 }
2000 2002
2001 2003 static int
2002 2004 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2003 2005 dsl_dataset_t *hds, void *arg)
2004 2006 {
2005 2007 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2006 2008 dsl_dataset_t *ds;
2007 2009 uint64_t val;
2008 2010 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2009 2011 int error;
2010 2012
2011 2013 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2012 2014 ASSERT(error == 0 || error == ENOENT);
2013 2015 if (error == ENOENT) {
2014 2016 /* ignore nonexistent snapshots */
2015 2017 return (0);
2016 2018 }
2017 2019
2018 2020 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2019 2021
2020 2022 /* log before we change the name */
2021 2023 spa_history_log_internal_ds(ds, "rename", tx,
2022 2024 "-> @%s", ddrsa->ddrsa_newsnapname);
2023 2025
2024 2026 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2025 2027 B_FALSE));
2026 2028 mutex_enter(&ds->ds_lock);
2027 2029 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2028 2030 mutex_exit(&ds->ds_lock);
2029 2031 VERIFY0(zap_add(dp->dp_meta_objset,
2030 2032 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2031 2033 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2032 2034
2033 2035 dsl_dataset_rele(ds, FTAG);
2034 2036 return (0);
2035 2037 }
2036 2038
2037 2039 static void
2038 2040 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2039 2041 {
2040 2042 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2041 2043 dsl_pool_t *dp = dmu_tx_pool(tx);
2042 2044 dsl_dataset_t *hds;
2043 2045
2044 2046 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2045 2047 ddrsa->ddrsa_tx = tx;
2046 2048 if (ddrsa->ddrsa_recursive) {
2047 2049 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2048 2050 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2049 2051 DS_FIND_CHILDREN));
2050 2052 } else {
2051 2053 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2052 2054 }
2053 2055 dsl_dataset_rele(hds, FTAG);
2054 2056 }
2055 2057
2056 2058 int
2057 2059 dsl_dataset_rename_snapshot(const char *fsname,
2058 2060 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2059 2061 {
2060 2062 dsl_dataset_rename_snapshot_arg_t ddrsa;
2061 2063
2062 2064 ddrsa.ddrsa_fsname = fsname;
2063 2065 ddrsa.ddrsa_oldsnapname = oldsnapname;
2064 2066 ddrsa.ddrsa_newsnapname = newsnapname;
2065 2067 ddrsa.ddrsa_recursive = recursive;
2066 2068
2067 2069 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2068 2070 dsl_dataset_rename_snapshot_sync, &ddrsa,
2069 2071 1, ZFS_SPACE_CHECK_RESERVED));
2070 2072 }
2071 2073
2072 2074 /*
2073 2075 * If we're doing an ownership handoff, we need to make sure that there is
2074 2076 * only one long hold on the dataset. We're not allowed to change anything here
2075 2077 * so we don't permanently release the long hold or regular hold here. We want
2076 2078 * to do this only when syncing to avoid the dataset unexpectedly going away
2077 2079 * when we release the long hold.
2078 2080 */
2079 2081 static int
2080 2082 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2081 2083 {
2082 2084 boolean_t held;
2083 2085
2084 2086 if (!dmu_tx_is_syncing(tx))
2085 2087 return (0);
2086 2088
2087 2089 if (owner != NULL) {
2088 2090 VERIFY3P(ds->ds_owner, ==, owner);
2089 2091 dsl_dataset_long_rele(ds, owner);
2090 2092 }
2091 2093
2092 2094 held = dsl_dataset_long_held(ds);
2093 2095
2094 2096 if (owner != NULL)
2095 2097 dsl_dataset_long_hold(ds, owner);
2096 2098
2097 2099 if (held)
2098 2100 return (SET_ERROR(EBUSY));
2099 2101
2100 2102 return (0);
2101 2103 }
2102 2104
2103 2105 typedef struct dsl_dataset_rollback_arg {
2104 2106 const char *ddra_fsname;
2105 2107 void *ddra_owner;
2106 2108 nvlist_t *ddra_result;
2107 2109 } dsl_dataset_rollback_arg_t;
2108 2110
2109 2111 static int
2110 2112 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2111 2113 {
2112 2114 dsl_dataset_rollback_arg_t *ddra = arg;
2113 2115 dsl_pool_t *dp = dmu_tx_pool(tx);
2114 2116 dsl_dataset_t *ds;
2115 2117 int64_t unused_refres_delta;
2116 2118 int error;
2117 2119
2118 2120 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2119 2121 if (error != 0)
2120 2122 return (error);
2121 2123
2122 2124 /* must not be a snapshot */
2123 2125 if (ds->ds_is_snapshot) {
2124 2126 dsl_dataset_rele(ds, FTAG);
2125 2127 return (SET_ERROR(EINVAL));
2126 2128 }
2127 2129
2128 2130 /* must have a most recent snapshot */
2129 2131 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2130 2132 dsl_dataset_rele(ds, FTAG);
2131 2133 return (SET_ERROR(EINVAL));
2132 2134 }
2133 2135
2134 2136 /* must not have any bookmarks after the most recent snapshot */
2135 2137 nvlist_t *proprequest = fnvlist_alloc();
2136 2138 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2137 2139 nvlist_t *bookmarks = fnvlist_alloc();
2138 2140 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2139 2141 fnvlist_free(proprequest);
2140 2142 if (error != 0)
2141 2143 return (error);
2142 2144 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2143 2145 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2144 2146 nvlist_t *valuenv =
2145 2147 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2146 2148 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2147 2149 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2148 2150 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2149 2151 fnvlist_free(bookmarks);
2150 2152 dsl_dataset_rele(ds, FTAG);
2151 2153 return (SET_ERROR(EEXIST));
2152 2154 }
2153 2155 }
2154 2156 fnvlist_free(bookmarks);
2155 2157
2156 2158 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2157 2159 if (error != 0) {
2158 2160 dsl_dataset_rele(ds, FTAG);
2159 2161 return (error);
2160 2162 }
2161 2163
2162 2164 /*
2163 2165 * Check if the snap we are rolling back to uses more than
2164 2166 * the refquota.
2165 2167 */
2166 2168 if (ds->ds_quota != 0 &&
2167 2169 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2168 2170 dsl_dataset_rele(ds, FTAG);
2169 2171 return (SET_ERROR(EDQUOT));
2170 2172 }
2171 2173
2172 2174 /*
2173 2175 * When we do the clone swap, we will temporarily use more space
2174 2176 * due to the refreservation (the head will no longer have any
2175 2177 * unique space, so the entire amount of the refreservation will need
2176 2178 * to be free). We will immediately destroy the clone, freeing
2177 2179 * this space, but the freeing happens over many txg's.
2178 2180 */
2179 2181 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2180 2182 dsl_dataset_phys(ds)->ds_unique_bytes);
2181 2183
2182 2184 if (unused_refres_delta > 0 &&
2183 2185 unused_refres_delta >
2184 2186 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2185 2187 dsl_dataset_rele(ds, FTAG);
2186 2188 return (SET_ERROR(ENOSPC));
2187 2189 }
2188 2190
2189 2191 dsl_dataset_rele(ds, FTAG);
2190 2192 return (0);
2191 2193 }
2192 2194
2193 2195 static void
2194 2196 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2195 2197 {
2196 2198 dsl_dataset_rollback_arg_t *ddra = arg;
2197 2199 dsl_pool_t *dp = dmu_tx_pool(tx);
2198 2200 dsl_dataset_t *ds, *clone;
2199 2201 uint64_t cloneobj;
2200 2202 char namebuf[ZFS_MAXNAMELEN];
2201 2203
2202 2204 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2203 2205
2204 2206 dsl_dataset_name(ds->ds_prev, namebuf);
2205 2207 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2206 2208
2207 2209 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2208 2210 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2209 2211
2210 2212 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2211 2213
2212 2214 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2213 2215 dsl_dataset_zero_zil(ds, tx);
2214 2216
2215 2217 dsl_destroy_head_sync_impl(clone, tx);
2216 2218
2217 2219 dsl_dataset_rele(clone, FTAG);
2218 2220 dsl_dataset_rele(ds, FTAG);
2219 2221 }
2220 2222
2221 2223 /*
2222 2224 * Rolls back the given filesystem or volume to the most recent snapshot.
2223 2225 * The name of the most recent snapshot will be returned under key "target"
2224 2226 * in the result nvlist.
2225 2227 *
2226 2228 * If owner != NULL:
2227 2229 * - The existing dataset MUST be owned by the specified owner at entry
2228 2230 * - Upon return, dataset will still be held by the same owner, whether we
2229 2231 * succeed or not.
2230 2232 *
2231 2233 * This mode is required any time the existing filesystem is mounted. See
2232 2234 * notes above zfs_suspend_fs() for further details.
2233 2235 */
2234 2236 int
2235 2237 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2236 2238 {
2237 2239 dsl_dataset_rollback_arg_t ddra;
2238 2240
2239 2241 ddra.ddra_fsname = fsname;
2240 2242 ddra.ddra_owner = owner;
2241 2243 ddra.ddra_result = result;
2242 2244
2243 2245 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2244 2246 dsl_dataset_rollback_sync, &ddra,
2245 2247 1, ZFS_SPACE_CHECK_RESERVED));
2246 2248 }
2247 2249
2248 2250 struct promotenode {
2249 2251 list_node_t link;
2250 2252 dsl_dataset_t *ds;
2251 2253 };
2252 2254
2253 2255 typedef struct dsl_dataset_promote_arg {
2254 2256 const char *ddpa_clonename;
2255 2257 dsl_dataset_t *ddpa_clone;
2256 2258 list_t shared_snaps, origin_snaps, clone_snaps;
2257 2259 dsl_dataset_t *origin_origin; /* origin of the origin */
2258 2260 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2259 2261 char *err_ds;
2260 2262 cred_t *cr;
2261 2263 } dsl_dataset_promote_arg_t;
2262 2264
2263 2265 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2264 2266 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2265 2267 void *tag);
2266 2268 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2267 2269
2268 2270 static int
2269 2271 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2270 2272 {
2271 2273 dsl_dataset_promote_arg_t *ddpa = arg;
2272 2274 dsl_pool_t *dp = dmu_tx_pool(tx);
2273 2275 dsl_dataset_t *hds;
2274 2276 struct promotenode *snap;
2275 2277 dsl_dataset_t *origin_ds;
2276 2278 int err;
2277 2279 uint64_t unused;
2278 2280 uint64_t ss_mv_cnt;
2279 2281 size_t max_snap_len;
2280 2282
2281 2283 err = promote_hold(ddpa, dp, FTAG);
2282 2284 if (err != 0)
2283 2285 return (err);
2284 2286
2285 2287 hds = ddpa->ddpa_clone;
2286 2288 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2287 2289
2288 2290 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2289 2291 promote_rele(ddpa, FTAG);
2290 2292 return (SET_ERROR(EXDEV));
2291 2293 }
2292 2294
2293 2295 /*
2294 2296 * Compute and check the amount of space to transfer. Since this is
2295 2297 * so expensive, don't do the preliminary check.
2296 2298 */
2297 2299 if (!dmu_tx_is_syncing(tx)) {
2298 2300 promote_rele(ddpa, FTAG);
2299 2301 return (0);
2300 2302 }
2301 2303
2302 2304 snap = list_head(&ddpa->shared_snaps);
2303 2305 origin_ds = snap->ds;
2304 2306
2305 2307 /* compute origin's new unique space */
2306 2308 snap = list_tail(&ddpa->clone_snaps);
2307 2309 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2308 2310 origin_ds->ds_object);
2309 2311 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2310 2312 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2311 2313 &ddpa->unique, &unused, &unused);
2312 2314
2313 2315 /*
2314 2316 * Walk the snapshots that we are moving
2315 2317 *
2316 2318 * Compute space to transfer. Consider the incremental changes
2317 2319 * to used by each snapshot:
2318 2320 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2319 2321 * So each snapshot gave birth to:
2320 2322 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2321 2323 * So a sequence would look like:
2322 2324 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2323 2325 * Which simplifies to:
2324 2326 * uN + kN + kN-1 + ... + k1 + k0
2325 2327 * Note however, if we stop before we reach the ORIGIN we get:
2326 2328 * uN + kN + kN-1 + ... + kM - uM-1
2327 2329 */
2328 2330 ss_mv_cnt = 0;
2329 2331 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2330 2332 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2331 2333 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2332 2334 for (snap = list_head(&ddpa->shared_snaps); snap;
2333 2335 snap = list_next(&ddpa->shared_snaps, snap)) {
2334 2336 uint64_t val, dlused, dlcomp, dluncomp;
2335 2337 dsl_dataset_t *ds = snap->ds;
2336 2338
2337 2339 ss_mv_cnt++;
2338 2340
2339 2341 /*
2340 2342 * If there are long holds, we won't be able to evict
2341 2343 * the objset.
2342 2344 */
2343 2345 if (dsl_dataset_long_held(ds)) {
2344 2346 err = SET_ERROR(EBUSY);
2345 2347 goto out;
2346 2348 }
2347 2349
2348 2350 /* Check that the snapshot name does not conflict */
2349 2351 VERIFY0(dsl_dataset_get_snapname(ds));
2350 2352 if (strlen(ds->ds_snapname) >= max_snap_len) {
2351 2353 err = SET_ERROR(ENAMETOOLONG);
2352 2354 goto out;
2353 2355 }
2354 2356 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2355 2357 if (err == 0) {
2356 2358 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2357 2359 err = SET_ERROR(EEXIST);
2358 2360 goto out;
2359 2361 }
2360 2362 if (err != ENOENT)
2361 2363 goto out;
2362 2364
2363 2365 /* The very first snapshot does not have a deadlist */
2364 2366 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2365 2367 continue;
2366 2368
2367 2369 dsl_deadlist_space(&ds->ds_deadlist,
2368 2370 &dlused, &dlcomp, &dluncomp);
2369 2371 ddpa->used += dlused;
2370 2372 ddpa->comp += dlcomp;
2371 2373 ddpa->uncomp += dluncomp;
2372 2374 }
2373 2375
2374 2376 /*
2375 2377 * If we are a clone of a clone then we never reached ORIGIN,
2376 2378 * so we need to subtract out the clone origin's used space.
2377 2379 */
2378 2380 if (ddpa->origin_origin) {
2379 2381 ddpa->used -=
2380 2382 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2381 2383 ddpa->comp -=
2382 2384 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2383 2385 ddpa->uncomp -=
2384 2386 dsl_dataset_phys(ddpa->origin_origin)->
2385 2387 ds_uncompressed_bytes;
2386 2388 }
2387 2389
2388 2390 /* Check that there is enough space and limit headroom here */
2389 2391 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2390 2392 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2391 2393 if (err != 0)
2392 2394 goto out;
2393 2395
2394 2396 /*
2395 2397 * Compute the amounts of space that will be used by snapshots
2396 2398 * after the promotion (for both origin and clone). For each,
2397 2399 * it is the amount of space that will be on all of their
2398 2400 * deadlists (that was not born before their new origin).
2399 2401 */
2400 2402 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2401 2403 uint64_t space;
2402 2404
2403 2405 /*
2404 2406 * Note, typically this will not be a clone of a clone,
2405 2407 * so dd_origin_txg will be < TXG_INITIAL, so
2406 2408 * these snaplist_space() -> dsl_deadlist_space_range()
2407 2409 * calls will be fast because they do not have to
2408 2410 * iterate over all bps.
2409 2411 */
2410 2412 snap = list_head(&ddpa->origin_snaps);
2411 2413 err = snaplist_space(&ddpa->shared_snaps,
2412 2414 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2413 2415 if (err != 0)
2414 2416 goto out;
2415 2417
2416 2418 err = snaplist_space(&ddpa->clone_snaps,
2417 2419 snap->ds->ds_dir->dd_origin_txg, &space);
2418 2420 if (err != 0)
2419 2421 goto out;
2420 2422 ddpa->cloneusedsnap += space;
2421 2423 }
2422 2424 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2423 2425 DD_FLAG_USED_BREAKDOWN) {
2424 2426 err = snaplist_space(&ddpa->origin_snaps,
2425 2427 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2426 2428 &ddpa->originusedsnap);
2427 2429 if (err != 0)
2428 2430 goto out;
2429 2431 }
2430 2432
2431 2433 out:
2432 2434 promote_rele(ddpa, FTAG);
2433 2435 return (err);
2434 2436 }
2435 2437
2436 2438 static void
2437 2439 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2438 2440 {
2439 2441 dsl_dataset_promote_arg_t *ddpa = arg;
2440 2442 dsl_pool_t *dp = dmu_tx_pool(tx);
2441 2443 dsl_dataset_t *hds;
2442 2444 struct promotenode *snap;
2443 2445 dsl_dataset_t *origin_ds;
2444 2446 dsl_dataset_t *origin_head;
2445 2447 dsl_dir_t *dd;
2446 2448 dsl_dir_t *odd = NULL;
2447 2449 uint64_t oldnext_obj;
2448 2450 int64_t delta;
2449 2451
2450 2452 VERIFY0(promote_hold(ddpa, dp, FTAG));
2451 2453 hds = ddpa->ddpa_clone;
2452 2454
2453 2455 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2454 2456
2455 2457 snap = list_head(&ddpa->shared_snaps);
2456 2458 origin_ds = snap->ds;
2457 2459 dd = hds->ds_dir;
2458 2460
2459 2461 snap = list_head(&ddpa->origin_snaps);
2460 2462 origin_head = snap->ds;
2461 2463
2462 2464 /*
2463 2465 * We need to explicitly open odd, since origin_ds's dd will be
2464 2466 * changing.
2465 2467 */
2466 2468 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2467 2469 NULL, FTAG, &odd));
2468 2470
2469 2471 /* change origin's next snap */
2470 2472 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2471 2473 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2472 2474 snap = list_tail(&ddpa->clone_snaps);
2473 2475 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2474 2476 origin_ds->ds_object);
2475 2477 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2476 2478
2477 2479 /* change the origin's next clone */
2478 2480 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2479 2481 dsl_dataset_remove_from_next_clones(origin_ds,
2480 2482 snap->ds->ds_object, tx);
2481 2483 VERIFY0(zap_add_int(dp->dp_meta_objset,
2482 2484 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2483 2485 oldnext_obj, tx));
2484 2486 }
2485 2487
2486 2488 /* change origin */
2487 2489 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2488 2490 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2489 2491 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2490 2492 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2491 2493 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2492 2494 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2493 2495 origin_head->ds_dir->dd_origin_txg =
2494 2496 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2495 2497
2496 2498 /* change dd_clone entries */
2497 2499 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2498 2500 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2499 2501 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2500 2502 VERIFY0(zap_add_int(dp->dp_meta_objset,
2501 2503 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2502 2504 hds->ds_object, tx));
2503 2505
2504 2506 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2505 2507 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2506 2508 origin_head->ds_object, tx));
2507 2509 if (dsl_dir_phys(dd)->dd_clones == 0) {
2508 2510 dsl_dir_phys(dd)->dd_clones =
2509 2511 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2510 2512 DMU_OT_NONE, 0, tx);
2511 2513 }
2512 2514 VERIFY0(zap_add_int(dp->dp_meta_objset,
2513 2515 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2514 2516 }
2515 2517
2516 2518 /* move snapshots to this dir */
2517 2519 for (snap = list_head(&ddpa->shared_snaps); snap;
2518 2520 snap = list_next(&ddpa->shared_snaps, snap)) {
2519 2521 dsl_dataset_t *ds = snap->ds;
2520 2522
2521 2523 /*
2522 2524 * Property callbacks are registered to a particular
2523 2525 * dsl_dir. Since ours is changing, evict the objset
2524 2526 * so that they will be unregistered from the old dsl_dir.
2525 2527 */
2526 2528 if (ds->ds_objset) {
2527 2529 dmu_objset_evict(ds->ds_objset);
2528 2530 ds->ds_objset = NULL;
2529 2531 }
2530 2532
2531 2533 /* move snap name entry */
2532 2534 VERIFY0(dsl_dataset_get_snapname(ds));
2533 2535 VERIFY0(dsl_dataset_snap_remove(origin_head,
2534 2536 ds->ds_snapname, tx, B_TRUE));
2535 2537 VERIFY0(zap_add(dp->dp_meta_objset,
2536 2538 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2537 2539 8, 1, &ds->ds_object, tx));
2538 2540 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2539 2541 DD_FIELD_SNAPSHOT_COUNT, tx);
2540 2542
2541 2543 /* change containing dsl_dir */
2542 2544 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2543 2545 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2544 2546 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2545 2547 ASSERT3P(ds->ds_dir, ==, odd);
2546 2548 dsl_dir_rele(ds->ds_dir, ds);
2547 2549 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2548 2550 NULL, ds, &ds->ds_dir));
2549 2551
2550 2552 /* move any clone references */
2551 2553 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2552 2554 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2553 2555 zap_cursor_t zc;
2554 2556 zap_attribute_t za;
2555 2557
2556 2558 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2557 2559 dsl_dataset_phys(ds)->ds_next_clones_obj);
2558 2560 zap_cursor_retrieve(&zc, &za) == 0;
2559 2561 zap_cursor_advance(&zc)) {
2560 2562 dsl_dataset_t *cnds;
2561 2563 uint64_t o;
2562 2564
2563 2565 if (za.za_first_integer == oldnext_obj) {
2564 2566 /*
2565 2567 * We've already moved the
2566 2568 * origin's reference.
2567 2569 */
2568 2570 continue;
2569 2571 }
2570 2572
2571 2573 VERIFY0(dsl_dataset_hold_obj(dp,
2572 2574 za.za_first_integer, FTAG, &cnds));
2573 2575 o = dsl_dir_phys(cnds->ds_dir)->
2574 2576 dd_head_dataset_obj;
2575 2577
2576 2578 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2577 2579 dsl_dir_phys(odd)->dd_clones, o, tx));
2578 2580 VERIFY0(zap_add_int(dp->dp_meta_objset,
2579 2581 dsl_dir_phys(dd)->dd_clones, o, tx));
2580 2582 dsl_dataset_rele(cnds, FTAG);
2581 2583 }
2582 2584 zap_cursor_fini(&zc);
2583 2585 }
2584 2586
2585 2587 ASSERT(!dsl_prop_hascb(ds));
2586 2588 }
2587 2589
2588 2590 /*
2589 2591 * Change space accounting.
2590 2592 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2591 2593 * both be valid, or both be 0 (resulting in delta == 0). This
2592 2594 * is true for each of {clone,origin} independently.
2593 2595 */
2594 2596
2595 2597 delta = ddpa->cloneusedsnap -
2596 2598 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2597 2599 ASSERT3S(delta, >=, 0);
2598 2600 ASSERT3U(ddpa->used, >=, delta);
2599 2601 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2600 2602 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2601 2603 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2602 2604
2603 2605 delta = ddpa->originusedsnap -
2604 2606 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2605 2607 ASSERT3S(delta, <=, 0);
2606 2608 ASSERT3U(ddpa->used, >=, -delta);
2607 2609 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2608 2610 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2609 2611 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2610 2612
2611 2613 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2612 2614
2613 2615 /* log history record */
2614 2616 spa_history_log_internal_ds(hds, "promote", tx, "");
2615 2617
2616 2618 dsl_dir_rele(odd, FTAG);
2617 2619 promote_rele(ddpa, FTAG);
2618 2620 }
2619 2621
2620 2622 /*
2621 2623 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2622 2624 * (exclusive) and last_obj (inclusive). The list will be in reverse
2623 2625 * order (last_obj will be the list_head()). If first_obj == 0, do all
2624 2626 * snapshots back to this dataset's origin.
2625 2627 */
2626 2628 static int
2627 2629 snaplist_make(dsl_pool_t *dp,
2628 2630 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2629 2631 {
2630 2632 uint64_t obj = last_obj;
2631 2633
2632 2634 list_create(l, sizeof (struct promotenode),
2633 2635 offsetof(struct promotenode, link));
2634 2636
2635 2637 while (obj != first_obj) {
2636 2638 dsl_dataset_t *ds;
2637 2639 struct promotenode *snap;
2638 2640 int err;
2639 2641
2640 2642 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2641 2643 ASSERT(err != ENOENT);
2642 2644 if (err != 0)
2643 2645 return (err);
2644 2646
2645 2647 if (first_obj == 0)
2646 2648 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2647 2649
2648 2650 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2649 2651 snap->ds = ds;
2650 2652 list_insert_tail(l, snap);
2651 2653 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2652 2654 }
2653 2655
2654 2656 return (0);
2655 2657 }
2656 2658
2657 2659 static int
2658 2660 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2659 2661 {
2660 2662 struct promotenode *snap;
2661 2663
2662 2664 *spacep = 0;
2663 2665 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2664 2666 uint64_t used, comp, uncomp;
2665 2667 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2666 2668 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2667 2669 *spacep += used;
2668 2670 }
2669 2671 return (0);
2670 2672 }
2671 2673
2672 2674 static void
2673 2675 snaplist_destroy(list_t *l, void *tag)
2674 2676 {
2675 2677 struct promotenode *snap;
2676 2678
2677 2679 if (l == NULL || !list_link_active(&l->list_head))
2678 2680 return;
2679 2681
2680 2682 while ((snap = list_tail(l)) != NULL) {
2681 2683 list_remove(l, snap);
2682 2684 dsl_dataset_rele(snap->ds, tag);
2683 2685 kmem_free(snap, sizeof (*snap));
2684 2686 }
2685 2687 list_destroy(l);
2686 2688 }
2687 2689
2688 2690 static int
2689 2691 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2690 2692 {
2691 2693 int error;
2692 2694 dsl_dir_t *dd;
2693 2695 struct promotenode *snap;
2694 2696
2695 2697 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2696 2698 &ddpa->ddpa_clone);
2697 2699 if (error != 0)
2698 2700 return (error);
2699 2701 dd = ddpa->ddpa_clone->ds_dir;
2700 2702
2701 2703 if (ddpa->ddpa_clone->ds_is_snapshot ||
2702 2704 !dsl_dir_is_clone(dd)) {
2703 2705 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2704 2706 return (SET_ERROR(EINVAL));
2705 2707 }
2706 2708
2707 2709 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2708 2710 &ddpa->shared_snaps, tag);
2709 2711 if (error != 0)
2710 2712 goto out;
2711 2713
2712 2714 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2713 2715 &ddpa->clone_snaps, tag);
2714 2716 if (error != 0)
2715 2717 goto out;
2716 2718
2717 2719 snap = list_head(&ddpa->shared_snaps);
2718 2720 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2719 2721 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2720 2722 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2721 2723 &ddpa->origin_snaps, tag);
2722 2724 if (error != 0)
2723 2725 goto out;
2724 2726
2725 2727 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2726 2728 error = dsl_dataset_hold_obj(dp,
2727 2729 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2728 2730 tag, &ddpa->origin_origin);
2729 2731 if (error != 0)
2730 2732 goto out;
2731 2733 }
2732 2734 out:
2733 2735 if (error != 0)
2734 2736 promote_rele(ddpa, tag);
2735 2737 return (error);
2736 2738 }
2737 2739
2738 2740 static void
2739 2741 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2740 2742 {
2741 2743 snaplist_destroy(&ddpa->shared_snaps, tag);
2742 2744 snaplist_destroy(&ddpa->clone_snaps, tag);
2743 2745 snaplist_destroy(&ddpa->origin_snaps, tag);
2744 2746 if (ddpa->origin_origin != NULL)
2745 2747 dsl_dataset_rele(ddpa->origin_origin, tag);
2746 2748 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2747 2749 }
2748 2750
2749 2751 /*
2750 2752 * Promote a clone.
2751 2753 *
2752 2754 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2753 2755 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2754 2756 */
2755 2757 int
2756 2758 dsl_dataset_promote(const char *name, char *conflsnap)
2757 2759 {
2758 2760 dsl_dataset_promote_arg_t ddpa = { 0 };
2759 2761 uint64_t numsnaps;
2760 2762 int error;
2761 2763 objset_t *os;
2762 2764
2763 2765 /*
2764 2766 * We will modify space proportional to the number of
2765 2767 * snapshots. Compute numsnaps.
2766 2768 */
2767 2769 error = dmu_objset_hold(name, FTAG, &os);
2768 2770 if (error != 0)
2769 2771 return (error);
2770 2772 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2771 2773 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2772 2774 &numsnaps);
2773 2775 dmu_objset_rele(os, FTAG);
2774 2776 if (error != 0)
2775 2777 return (error);
2776 2778
2777 2779 ddpa.ddpa_clonename = name;
2778 2780 ddpa.err_ds = conflsnap;
2779 2781 ddpa.cr = CRED();
2780 2782
2781 2783 return (dsl_sync_task(name, dsl_dataset_promote_check,
2782 2784 dsl_dataset_promote_sync, &ddpa,
2783 2785 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2784 2786 }
2785 2787
2786 2788 int
2787 2789 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2788 2790 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2789 2791 {
2790 2792 int64_t unused_refres_delta;
2791 2793
2792 2794 /* they should both be heads */
2793 2795 if (clone->ds_is_snapshot ||
2794 2796 origin_head->ds_is_snapshot)
2795 2797 return (SET_ERROR(EINVAL));
2796 2798
2797 2799 /* if we are not forcing, the branch point should be just before them */
2798 2800 if (!force && clone->ds_prev != origin_head->ds_prev)
2799 2801 return (SET_ERROR(EINVAL));
2800 2802
2801 2803 /* clone should be the clone (unless they are unrelated) */
2802 2804 if (clone->ds_prev != NULL &&
2803 2805 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2804 2806 origin_head->ds_dir != clone->ds_prev->ds_dir)
2805 2807 return (SET_ERROR(EINVAL));
2806 2808
2807 2809 /* the clone should be a child of the origin */
2808 2810 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2809 2811 return (SET_ERROR(EINVAL));
2810 2812
2811 2813 /* origin_head shouldn't be modified unless 'force' */
2812 2814 if (!force &&
2813 2815 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2814 2816 return (SET_ERROR(ETXTBSY));
2815 2817
2816 2818 /* origin_head should have no long holds (e.g. is not mounted) */
2817 2819 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2818 2820 return (SET_ERROR(EBUSY));
2819 2821
2820 2822 /* check amount of any unconsumed refreservation */
2821 2823 unused_refres_delta =
2822 2824 (int64_t)MIN(origin_head->ds_reserved,
2823 2825 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2824 2826 (int64_t)MIN(origin_head->ds_reserved,
2825 2827 dsl_dataset_phys(clone)->ds_unique_bytes);
2826 2828
2827 2829 if (unused_refres_delta > 0 &&
2828 2830 unused_refres_delta >
2829 2831 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2830 2832 return (SET_ERROR(ENOSPC));
2831 2833
2832 2834 /* clone can't be over the head's refquota */
2833 2835 if (origin_head->ds_quota != 0 &&
2834 2836 dsl_dataset_phys(clone)->ds_referenced_bytes >
2835 2837 origin_head->ds_quota)
2836 2838 return (SET_ERROR(EDQUOT));
2837 2839
2838 2840 return (0);
2839 2841 }
2840 2842
2841 2843 void
2842 2844 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2843 2845 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2844 2846 {
2845 2847 dsl_pool_t *dp = dmu_tx_pool(tx);
2846 2848 int64_t unused_refres_delta;
2847 2849
2848 2850 ASSERT(clone->ds_reserved == 0);
2849 2851 ASSERT(origin_head->ds_quota == 0 ||
2850 2852 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2851 2853 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2852 2854
2853 2855 /*
2854 2856 * Swap per-dataset feature flags.
2855 2857 */
2856 2858 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2857 2859 if (!(spa_feature_table[f].fi_flags &
2858 2860 ZFEATURE_FLAG_PER_DATASET)) {
2859 2861 ASSERT(!clone->ds_feature_inuse[f]);
2860 2862 ASSERT(!origin_head->ds_feature_inuse[f]);
2861 2863 continue;
2862 2864 }
2863 2865
2864 2866 boolean_t clone_inuse = clone->ds_feature_inuse[f];
2865 2867 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2866 2868
2867 2869 if (clone_inuse) {
2868 2870 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2869 2871 clone->ds_feature_inuse[f] = B_FALSE;
2870 2872 }
2871 2873 if (origin_head_inuse) {
2872 2874 dsl_dataset_deactivate_feature(origin_head->ds_object,
2873 2875 f, tx);
2874 2876 origin_head->ds_feature_inuse[f] = B_FALSE;
2875 2877 }
2876 2878 if (clone_inuse) {
2877 2879 dsl_dataset_activate_feature(origin_head->ds_object,
2878 2880 f, tx);
2879 2881 origin_head->ds_feature_inuse[f] = B_TRUE;
2880 2882 }
2881 2883 if (origin_head_inuse) {
2882 2884 dsl_dataset_activate_feature(clone->ds_object, f, tx);
2883 2885 clone->ds_feature_inuse[f] = B_TRUE;
2884 2886 }
2885 2887 }
2886 2888
2887 2889 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2888 2890 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2889 2891
2890 2892 if (clone->ds_objset != NULL) {
2891 2893 dmu_objset_evict(clone->ds_objset);
2892 2894 clone->ds_objset = NULL;
2893 2895 }
2894 2896
2895 2897 if (origin_head->ds_objset != NULL) {
2896 2898 dmu_objset_evict(origin_head->ds_objset);
2897 2899 origin_head->ds_objset = NULL;
2898 2900 }
2899 2901
2900 2902 unused_refres_delta =
2901 2903 (int64_t)MIN(origin_head->ds_reserved,
2902 2904 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2903 2905 (int64_t)MIN(origin_head->ds_reserved,
2904 2906 dsl_dataset_phys(clone)->ds_unique_bytes);
2905 2907
2906 2908 /*
2907 2909 * Reset origin's unique bytes, if it exists.
2908 2910 */
2909 2911 if (clone->ds_prev) {
2910 2912 dsl_dataset_t *origin = clone->ds_prev;
2911 2913 uint64_t comp, uncomp;
2912 2914
2913 2915 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2914 2916 dsl_deadlist_space_range(&clone->ds_deadlist,
2915 2917 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2916 2918 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2917 2919 }
2918 2920
2919 2921 /* swap blkptrs */
2920 2922 {
2921 2923 blkptr_t tmp;
2922 2924 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2923 2925 dsl_dataset_phys(origin_head)->ds_bp =
2924 2926 dsl_dataset_phys(clone)->ds_bp;
2925 2927 dsl_dataset_phys(clone)->ds_bp = tmp;
2926 2928 }
2927 2929
2928 2930 /* set dd_*_bytes */
2929 2931 {
2930 2932 int64_t dused, dcomp, duncomp;
2931 2933 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2932 2934 uint64_t odl_used, odl_comp, odl_uncomp;
2933 2935
2934 2936 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2935 2937 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2936 2938
2937 2939 dsl_deadlist_space(&clone->ds_deadlist,
2938 2940 &cdl_used, &cdl_comp, &cdl_uncomp);
2939 2941 dsl_deadlist_space(&origin_head->ds_deadlist,
2940 2942 &odl_used, &odl_comp, &odl_uncomp);
2941 2943
2942 2944 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2943 2945 cdl_used -
2944 2946 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2945 2947 odl_used);
2946 2948 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2947 2949 cdl_comp -
2948 2950 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2949 2951 odl_comp);
2950 2952 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2951 2953 cdl_uncomp -
2952 2954 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2953 2955 odl_uncomp);
2954 2956
2955 2957 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2956 2958 dused, dcomp, duncomp, tx);
2957 2959 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2958 2960 -dused, -dcomp, -duncomp, tx);
2959 2961
2960 2962 /*
2961 2963 * The difference in the space used by snapshots is the
2962 2964 * difference in snapshot space due to the head's
2963 2965 * deadlist (since that's the only thing that's
2964 2966 * changing that affects the snapused).
2965 2967 */
2966 2968 dsl_deadlist_space_range(&clone->ds_deadlist,
2967 2969 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2968 2970 &cdl_used, &cdl_comp, &cdl_uncomp);
2969 2971 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2970 2972 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2971 2973 &odl_used, &odl_comp, &odl_uncomp);
2972 2974 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2973 2975 DD_USED_HEAD, DD_USED_SNAP, tx);
2974 2976 }
2975 2977
2976 2978 /* swap ds_*_bytes */
2977 2979 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2978 2980 dsl_dataset_phys(clone)->ds_referenced_bytes);
2979 2981 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2980 2982 dsl_dataset_phys(clone)->ds_compressed_bytes);
2981 2983 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2982 2984 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2983 2985 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2984 2986 dsl_dataset_phys(clone)->ds_unique_bytes);
2985 2987
2986 2988 /* apply any parent delta for change in unconsumed refreservation */
2987 2989 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2988 2990 unused_refres_delta, 0, 0, tx);
2989 2991
2990 2992 /*
2991 2993 * Swap deadlists.
2992 2994 */
2993 2995 dsl_deadlist_close(&clone->ds_deadlist);
2994 2996 dsl_deadlist_close(&origin_head->ds_deadlist);
2995 2997 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2996 2998 dsl_dataset_phys(clone)->ds_deadlist_obj);
2997 2999 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2998 3000 dsl_dataset_phys(clone)->ds_deadlist_obj);
2999 3001 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3000 3002 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3001 3003
3002 3004 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3003 3005
3004 3006 spa_history_log_internal_ds(clone, "clone swap", tx,
3005 3007 "parent=%s", origin_head->ds_dir->dd_myname);
3006 3008 }
3007 3009
3008 3010 /*
3009 3011 * Given a pool name and a dataset object number in that pool,
3010 3012 * return the name of that dataset.
3011 3013 */
3012 3014 int
3013 3015 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3014 3016 {
3015 3017 dsl_pool_t *dp;
3016 3018 dsl_dataset_t *ds;
3017 3019 int error;
3018 3020
3019 3021 error = dsl_pool_hold(pname, FTAG, &dp);
3020 3022 if (error != 0)
3021 3023 return (error);
3022 3024
3023 3025 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3024 3026 if (error == 0) {
3025 3027 dsl_dataset_name(ds, buf);
3026 3028 dsl_dataset_rele(ds, FTAG);
3027 3029 }
3028 3030 dsl_pool_rele(dp, FTAG);
3029 3031
3030 3032 return (error);
3031 3033 }
3032 3034
3033 3035 int
3034 3036 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3035 3037 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3036 3038 {
3037 3039 int error = 0;
3038 3040
3039 3041 ASSERT3S(asize, >, 0);
3040 3042
3041 3043 /*
3042 3044 * *ref_rsrv is the portion of asize that will come from any
3043 3045 * unconsumed refreservation space.
3044 3046 */
3045 3047 *ref_rsrv = 0;
3046 3048
3047 3049 mutex_enter(&ds->ds_lock);
3048 3050 /*
3049 3051 * Make a space adjustment for reserved bytes.
3050 3052 */
3051 3053 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3052 3054 ASSERT3U(*used, >=,
3053 3055 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3054 3056 *used -=
3055 3057 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3056 3058 *ref_rsrv =
3057 3059 asize - MIN(asize, parent_delta(ds, asize + inflight));
3058 3060 }
3059 3061
3060 3062 if (!check_quota || ds->ds_quota == 0) {
3061 3063 mutex_exit(&ds->ds_lock);
3062 3064 return (0);
3063 3065 }
3064 3066 /*
3065 3067 * If they are requesting more space, and our current estimate
3066 3068 * is over quota, they get to try again unless the actual
3067 3069 * on-disk is over quota and there are no pending changes (which
3068 3070 * may free up space for us).
3069 3071 */
3070 3072 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3071 3073 ds->ds_quota) {
3072 3074 if (inflight > 0 ||
3073 3075 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3074 3076 error = SET_ERROR(ERESTART);
3075 3077 else
3076 3078 error = SET_ERROR(EDQUOT);
3077 3079 }
3078 3080 mutex_exit(&ds->ds_lock);
3079 3081
3080 3082 return (error);
3081 3083 }
3082 3084
3083 3085 typedef struct dsl_dataset_set_qr_arg {
3084 3086 const char *ddsqra_name;
3085 3087 zprop_source_t ddsqra_source;
3086 3088 uint64_t ddsqra_value;
3087 3089 } dsl_dataset_set_qr_arg_t;
3088 3090
3089 3091
3090 3092 /* ARGSUSED */
3091 3093 static int
3092 3094 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3093 3095 {
3094 3096 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3095 3097 dsl_pool_t *dp = dmu_tx_pool(tx);
3096 3098 dsl_dataset_t *ds;
3097 3099 int error;
3098 3100 uint64_t newval;
3099 3101
3100 3102 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3101 3103 return (SET_ERROR(ENOTSUP));
3102 3104
3103 3105 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3104 3106 if (error != 0)
3105 3107 return (error);
3106 3108
3107 3109 if (ds->ds_is_snapshot) {
3108 3110 dsl_dataset_rele(ds, FTAG);
3109 3111 return (SET_ERROR(EINVAL));
3110 3112 }
3111 3113
3112 3114 error = dsl_prop_predict(ds->ds_dir,
3113 3115 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3114 3116 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3115 3117 if (error != 0) {
3116 3118 dsl_dataset_rele(ds, FTAG);
3117 3119 return (error);
3118 3120 }
3119 3121
3120 3122 if (newval == 0) {
3121 3123 dsl_dataset_rele(ds, FTAG);
3122 3124 return (0);
3123 3125 }
3124 3126
3125 3127 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3126 3128 newval < ds->ds_reserved) {
3127 3129 dsl_dataset_rele(ds, FTAG);
3128 3130 return (SET_ERROR(ENOSPC));
3129 3131 }
3130 3132
3131 3133 dsl_dataset_rele(ds, FTAG);
3132 3134 return (0);
3133 3135 }
3134 3136
3135 3137 static void
3136 3138 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3137 3139 {
3138 3140 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3139 3141 dsl_pool_t *dp = dmu_tx_pool(tx);
3140 3142 dsl_dataset_t *ds;
3141 3143 uint64_t newval;
3142 3144
3143 3145 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3144 3146
3145 3147 dsl_prop_set_sync_impl(ds,
3146 3148 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3147 3149 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3148 3150 &ddsqra->ddsqra_value, tx);
3149 3151
3150 3152 VERIFY0(dsl_prop_get_int_ds(ds,
3151 3153 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3152 3154
3153 3155 if (ds->ds_quota != newval) {
3154 3156 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3155 3157 ds->ds_quota = newval;
3156 3158 }
3157 3159 dsl_dataset_rele(ds, FTAG);
3158 3160 }
3159 3161
3160 3162 int
3161 3163 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3162 3164 uint64_t refquota)
3163 3165 {
3164 3166 dsl_dataset_set_qr_arg_t ddsqra;
3165 3167
3166 3168 ddsqra.ddsqra_name = dsname;
3167 3169 ddsqra.ddsqra_source = source;
3168 3170 ddsqra.ddsqra_value = refquota;
3169 3171
3170 3172 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3171 3173 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3172 3174 }
3173 3175
3174 3176 static int
3175 3177 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3176 3178 {
3177 3179 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3178 3180 dsl_pool_t *dp = dmu_tx_pool(tx);
3179 3181 dsl_dataset_t *ds;
3180 3182 int error;
3181 3183 uint64_t newval, unique;
3182 3184
3183 3185 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3184 3186 return (SET_ERROR(ENOTSUP));
3185 3187
3186 3188 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3187 3189 if (error != 0)
3188 3190 return (error);
3189 3191
3190 3192 if (ds->ds_is_snapshot) {
3191 3193 dsl_dataset_rele(ds, FTAG);
3192 3194 return (SET_ERROR(EINVAL));
3193 3195 }
3194 3196
3195 3197 error = dsl_prop_predict(ds->ds_dir,
3196 3198 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3197 3199 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3198 3200 if (error != 0) {
3199 3201 dsl_dataset_rele(ds, FTAG);
3200 3202 return (error);
3201 3203 }
3202 3204
3203 3205 /*
3204 3206 * If we are doing the preliminary check in open context, the
3205 3207 * space estimates may be inaccurate.
3206 3208 */
3207 3209 if (!dmu_tx_is_syncing(tx)) {
3208 3210 dsl_dataset_rele(ds, FTAG);
3209 3211 return (0);
3210 3212 }
3211 3213
3212 3214 mutex_enter(&ds->ds_lock);
3213 3215 if (!DS_UNIQUE_IS_ACCURATE(ds))
3214 3216 dsl_dataset_recalc_head_uniq(ds);
3215 3217 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3216 3218 mutex_exit(&ds->ds_lock);
3217 3219
3218 3220 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3219 3221 uint64_t delta = MAX(unique, newval) -
3220 3222 MAX(unique, ds->ds_reserved);
3221 3223
3222 3224 if (delta >
3223 3225 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3224 3226 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3225 3227 dsl_dataset_rele(ds, FTAG);
3226 3228 return (SET_ERROR(ENOSPC));
3227 3229 }
3228 3230 }
3229 3231
3230 3232 dsl_dataset_rele(ds, FTAG);
3231 3233 return (0);
3232 3234 }
3233 3235
3234 3236 void
3235 3237 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3236 3238 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3237 3239 {
3238 3240 uint64_t newval;
3239 3241 uint64_t unique;
3240 3242 int64_t delta;
3241 3243
3242 3244 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3243 3245 source, sizeof (value), 1, &value, tx);
3244 3246
3245 3247 VERIFY0(dsl_prop_get_int_ds(ds,
3246 3248 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3247 3249
3248 3250 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3249 3251 mutex_enter(&ds->ds_dir->dd_lock);
3250 3252 mutex_enter(&ds->ds_lock);
3251 3253 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3252 3254 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3253 3255 delta = MAX(0, (int64_t)(newval - unique)) -
3254 3256 MAX(0, (int64_t)(ds->ds_reserved - unique));
3255 3257 ds->ds_reserved = newval;
3256 3258 mutex_exit(&ds->ds_lock);
3257 3259
3258 3260 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3259 3261 mutex_exit(&ds->ds_dir->dd_lock);
3260 3262 }
3261 3263
3262 3264 static void
3263 3265 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3264 3266 {
3265 3267 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3266 3268 dsl_pool_t *dp = dmu_tx_pool(tx);
3267 3269 dsl_dataset_t *ds;
3268 3270
3269 3271 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3270 3272 dsl_dataset_set_refreservation_sync_impl(ds,
3271 3273 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3272 3274 dsl_dataset_rele(ds, FTAG);
3273 3275 }
3274 3276
3275 3277 int
3276 3278 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3277 3279 uint64_t refreservation)
3278 3280 {
3279 3281 dsl_dataset_set_qr_arg_t ddsqra;
3280 3282
3281 3283 ddsqra.ddsqra_name = dsname;
3282 3284 ddsqra.ddsqra_source = source;
3283 3285 ddsqra.ddsqra_value = refreservation;
3284 3286
3285 3287 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3286 3288 dsl_dataset_set_refreservation_sync, &ddsqra,
3287 3289 0, ZFS_SPACE_CHECK_NONE));
3288 3290 }
3289 3291
3290 3292 /*
3291 3293 * Return (in *usedp) the amount of space written in new that is not
3292 3294 * present in oldsnap. New may be a snapshot or the head. Old must be
3293 3295 * a snapshot before new, in new's filesystem (or its origin). If not then
3294 3296 * fail and return EINVAL.
3295 3297 *
3296 3298 * The written space is calculated by considering two components: First, we
3297 3299 * ignore any freed space, and calculate the written as new's used space
3298 3300 * minus old's used space. Next, we add in the amount of space that was freed
3299 3301 * between the two snapshots, thus reducing new's used space relative to old's.
3300 3302 * Specifically, this is the space that was born before old->ds_creation_txg,
3301 3303 * and freed before new (ie. on new's deadlist or a previous deadlist).
3302 3304 *
3303 3305 * space freed [---------------------]
3304 3306 * snapshots ---O-------O--------O-------O------
3305 3307 * oldsnap new
3306 3308 */
3307 3309 int
3308 3310 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3309 3311 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3310 3312 {
3311 3313 int err = 0;
3312 3314 uint64_t snapobj;
3313 3315 dsl_pool_t *dp = new->ds_dir->dd_pool;
3314 3316
3315 3317 ASSERT(dsl_pool_config_held(dp));
3316 3318
3317 3319 *usedp = 0;
3318 3320 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3319 3321 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3320 3322
3321 3323 *compp = 0;
3322 3324 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3323 3325 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3324 3326
3325 3327 *uncompp = 0;
3326 3328 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3327 3329 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3328 3330
3329 3331 snapobj = new->ds_object;
3330 3332 while (snapobj != oldsnap->ds_object) {
3331 3333 dsl_dataset_t *snap;
3332 3334 uint64_t used, comp, uncomp;
3333 3335
3334 3336 if (snapobj == new->ds_object) {
3335 3337 snap = new;
3336 3338 } else {
3337 3339 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3338 3340 if (err != 0)
3339 3341 break;
3340 3342 }
3341 3343
3342 3344 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3343 3345 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3344 3346 /*
3345 3347 * The blocks in the deadlist can not be born after
3346 3348 * ds_prev_snap_txg, so get the whole deadlist space,
3347 3349 * which is more efficient (especially for old-format
3348 3350 * deadlists). Unfortunately the deadlist code
3349 3351 * doesn't have enough information to make this
3350 3352 * optimization itself.
3351 3353 */
3352 3354 dsl_deadlist_space(&snap->ds_deadlist,
3353 3355 &used, &comp, &uncomp);
3354 3356 } else {
3355 3357 dsl_deadlist_space_range(&snap->ds_deadlist,
3356 3358 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3357 3359 &used, &comp, &uncomp);
3358 3360 }
3359 3361 *usedp += used;
3360 3362 *compp += comp;
3361 3363 *uncompp += uncomp;
3362 3364
3363 3365 /*
3364 3366 * If we get to the beginning of the chain of snapshots
3365 3367 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3366 3368 * was not a snapshot of/before new.
3367 3369 */
3368 3370 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3369 3371 if (snap != new)
3370 3372 dsl_dataset_rele(snap, FTAG);
3371 3373 if (snapobj == 0) {
3372 3374 err = SET_ERROR(EINVAL);
3373 3375 break;
3374 3376 }
3375 3377
3376 3378 }
3377 3379 return (err);
3378 3380 }
3379 3381
3380 3382 /*
3381 3383 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3382 3384 * lastsnap, and all snapshots in between are deleted.
3383 3385 *
3384 3386 * blocks that would be freed [---------------------------]
3385 3387 * snapshots ---O-------O--------O-------O--------O
3386 3388 * firstsnap lastsnap
3387 3389 *
3388 3390 * This is the set of blocks that were born after the snap before firstsnap,
3389 3391 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3390 3392 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3391 3393 * We calculate this by iterating over the relevant deadlists (from the snap
3392 3394 * after lastsnap, backward to the snap after firstsnap), summing up the
3393 3395 * space on the deadlist that was born after the snap before firstsnap.
3394 3396 */
3395 3397 int
3396 3398 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3397 3399 dsl_dataset_t *lastsnap,
3398 3400 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3399 3401 {
3400 3402 int err = 0;
3401 3403 uint64_t snapobj;
3402 3404 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3403 3405
3404 3406 ASSERT(firstsnap->ds_is_snapshot);
3405 3407 ASSERT(lastsnap->ds_is_snapshot);
3406 3408
3407 3409 /*
3408 3410 * Check that the snapshots are in the same dsl_dir, and firstsnap
3409 3411 * is before lastsnap.
3410 3412 */
3411 3413 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3412 3414 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3413 3415 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3414 3416 return (SET_ERROR(EINVAL));
3415 3417
3416 3418 *usedp = *compp = *uncompp = 0;
3417 3419
3418 3420 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3419 3421 while (snapobj != firstsnap->ds_object) {
3420 3422 dsl_dataset_t *ds;
3421 3423 uint64_t used, comp, uncomp;
3422 3424
3423 3425 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3424 3426 if (err != 0)
3425 3427 break;
3426 3428
3427 3429 dsl_deadlist_space_range(&ds->ds_deadlist,
3428 3430 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3429 3431 &used, &comp, &uncomp);
3430 3432 *usedp += used;
3431 3433 *compp += comp;
3432 3434 *uncompp += uncomp;
3433 3435
3434 3436 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3435 3437 ASSERT3U(snapobj, !=, 0);
3436 3438 dsl_dataset_rele(ds, FTAG);
3437 3439 }
3438 3440 return (err);
3439 3441 }
3440 3442
3441 3443 /*
3442 3444 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3443 3445 * For example, they could both be snapshots of the same filesystem, and
3444 3446 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3445 3447 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3446 3448 * filesystem. Or 'earlier' could be the origin's origin.
3447 3449 *
3448 3450 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3449 3451 */
3450 3452 boolean_t
3451 3453 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3452 3454 uint64_t earlier_txg)
3453 3455 {
3454 3456 dsl_pool_t *dp = later->ds_dir->dd_pool;
3455 3457 int error;
3456 3458 boolean_t ret;
3457 3459
3458 3460 ASSERT(dsl_pool_config_held(dp));
3459 3461 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3460 3462
3461 3463 if (earlier_txg == 0)
3462 3464 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3463 3465
3464 3466 if (later->ds_is_snapshot &&
3465 3467 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3466 3468 return (B_FALSE);
3467 3469
3468 3470 if (later->ds_dir == earlier->ds_dir)
3469 3471 return (B_TRUE);
3470 3472 if (!dsl_dir_is_clone(later->ds_dir))
3471 3473 return (B_FALSE);
3472 3474
3473 3475 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3474 3476 return (B_TRUE);
3475 3477 dsl_dataset_t *origin;
3476 3478 error = dsl_dataset_hold_obj(dp,
3477 3479 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3478 3480 if (error != 0)
3479 3481 return (B_FALSE);
3480 3482 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3481 3483 dsl_dataset_rele(origin, FTAG);
3482 3484 return (ret);
3483 3485 }
3484 3486
3485 3487 void
3486 3488 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3487 3489 {
3488 3490 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3489 3491 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3490 3492 }
3491 3493
3492 3494 boolean_t
3493 3495 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3494 3496 {
3495 3497 dmu_object_info_t doi;
3496 3498
|
↓ open down ↓ |
1686 lines elided |
↑ open up ↑ |
3497 3499 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3498 3500 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3499 3501 }
3500 3502
3501 3503 boolean_t
3502 3504 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3503 3505 {
3504 3506 return (dsl_dataset_is_zapified(ds) &&
3505 3507 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3506 3508 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3509 +}
3510 +
3511 +typedef struct dsl_dataset_set_fsid_guid_arg {
3512 + const char *ddsfg_name;
3513 + uint64_t ddsfg_value;
3514 +} dsl_dataset_set_fsid_guid_arg_t;
3515 +
3516 +static int
3517 +dsl_dataset_set_fsid_guid_check(void *arg, dmu_tx_t *tx)
3518 +{
3519 + int err = 0;
3520 + dsl_dataset_set_fsid_guid_arg_t *ddsfg = arg;
3521 + dsl_pool_t *dp = dmu_tx_pool(tx);
3522 + dsl_dataset_t *ds;
3523 + uint64_t newval;
3524 +
3525 + newval = ddsfg->ddsfg_value;
3526 +
3527 + err = dsl_dataset_hold(dp, ddsfg->ddsfg_name, FTAG, &ds);
3528 + if (err != 0)
3529 + return (err);
3530 +
3531 + /* Check if the FS is mounted */
3532 + if (ds->ds_owner != NULL) {
3533 + dsl_dataset_rele(ds, FTAG);
3534 + return (SET_ERROR(EBUSY));
3535 + }
3536 + dsl_dataset_rele(ds, FTAG);
3537 +
3538 + if (!unique_valid(newval))
3539 + return (SET_ERROR(EEXIST));
3540 +
3541 + return (err);
3542 +}
3543 +
3544 +static void
3545 +dsl_dataset_set_fsid_guid_sync(void *arg, dmu_tx_t *tx)
3546 +{
3547 + dsl_dataset_set_fsid_guid_arg_t *ddsfg = arg;
3548 + dsl_pool_t *dp = dmu_tx_pool(tx);
3549 + dsl_dataset_t *ds;
3550 + uint64_t newval;
3551 +
3552 + VERIFY0(dsl_dataset_hold(dp, ddsfg->ddsfg_name, FTAG, &ds));
3553 +
3554 + newval = ddsfg->ddsfg_value;
3555 +
3556 + dmu_buf_will_dirty(ds->ds_dbuf, tx);
3557 + mutex_enter(&ds->ds_lock);
3558 + ds->ds_fsid_guid = newval;
3559 + dsl_dataset_phys(ds)->ds_fsid_guid = newval;
3560 + mutex_exit(&ds->ds_lock);
3561 +
3562 + dsl_dataset_rele(ds, FTAG);
3563 +}
3564 +
3565 +int
3566 +dsl_dataset_set_fsid_guid(const char *ddname, zprop_source_t source,
3567 + uint64_t fsid_guid)
3568 +{
3569 + dsl_dataset_set_fsid_guid_arg_t ddsfg;
3570 +
3571 + ddsfg.ddsfg_name = ddname;
3572 + ddsfg.ddsfg_value = fsid_guid;
3573 +
3574 + return (dsl_sync_task(ddname, dsl_dataset_set_fsid_guid_check,
3575 + dsl_dataset_set_fsid_guid_sync, &ddsfg, 0,
3576 + ZFS_SPACE_CHECK_RESERVED));
3507 3577 }
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX