Print this page
6314 buffer overflow in dsl_dataset_name
| 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 {
|
↓ open down ↓ |
660 lines elided |
↑ open up ↑ |
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 + VERIFY3U(strlen(name) +
672 + strlen(ds->ds_snapname) + 1, <=,
673 + ZFS_MAXNAMELEN);
671 674 (void) strcat(name, ds->ds_snapname);
672 675 mutex_exit(&ds->ds_lock);
673 676 } else {
677 + VERIFY3U(strlen(name) +
678 + strlen(ds->ds_snapname) + 1, <=,
679 + ZFS_MAXNAMELEN);
674 680 (void) strcat(name, ds->ds_snapname);
675 681 }
676 682 }
677 683 }
678 684 }
679 685
680 686 void
681 687 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
682 688 {
683 689 dmu_buf_rele(ds->ds_dbuf, tag);
684 690 }
685 691
686 692 void
687 693 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
688 694 {
689 695 ASSERT3P(ds->ds_owner, ==, tag);
690 696 ASSERT(ds->ds_dbuf != NULL);
691 697
692 698 mutex_enter(&ds->ds_lock);
693 699 ds->ds_owner = NULL;
694 700 mutex_exit(&ds->ds_lock);
695 701 dsl_dataset_long_rele(ds, tag);
696 702 dsl_dataset_rele(ds, tag);
697 703 }
698 704
699 705 boolean_t
700 706 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
701 707 {
702 708 boolean_t gotit = FALSE;
703 709
704 710 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
705 711 mutex_enter(&ds->ds_lock);
706 712 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
707 713 ds->ds_owner = tag;
708 714 dsl_dataset_long_hold(ds, tag);
709 715 gotit = TRUE;
710 716 }
711 717 mutex_exit(&ds->ds_lock);
712 718 return (gotit);
713 719 }
714 720
715 721 boolean_t
716 722 dsl_dataset_has_owner(dsl_dataset_t *ds)
717 723 {
718 724 boolean_t rv;
719 725 mutex_enter(&ds->ds_lock);
720 726 rv = (ds->ds_owner != NULL);
721 727 mutex_exit(&ds->ds_lock);
722 728 return (rv);
723 729 }
724 730
725 731 static void
726 732 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
727 733 {
728 734 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
729 735 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
730 736 uint64_t zero = 0;
731 737
732 738 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
733 739
734 740 spa_feature_incr(spa, f, tx);
735 741 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
736 742
737 743 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
738 744 sizeof (zero), 1, &zero, tx));
739 745 }
740 746
741 747 void
742 748 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
743 749 {
744 750 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
745 751 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
746 752
747 753 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
748 754
749 755 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
750 756 spa_feature_decr(spa, f, tx);
751 757 }
752 758
753 759 uint64_t
754 760 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
755 761 uint64_t flags, dmu_tx_t *tx)
756 762 {
757 763 dsl_pool_t *dp = dd->dd_pool;
758 764 dmu_buf_t *dbuf;
759 765 dsl_dataset_phys_t *dsphys;
760 766 uint64_t dsobj;
761 767 objset_t *mos = dp->dp_meta_objset;
762 768
763 769 if (origin == NULL)
764 770 origin = dp->dp_origin_snap;
765 771
766 772 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
767 773 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
768 774 ASSERT(dmu_tx_is_syncing(tx));
769 775 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
770 776
771 777 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
772 778 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
773 779 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
774 780 dmu_buf_will_dirty(dbuf, tx);
775 781 dsphys = dbuf->db_data;
776 782 bzero(dsphys, sizeof (dsl_dataset_phys_t));
777 783 dsphys->ds_dir_obj = dd->dd_object;
778 784 dsphys->ds_flags = flags;
779 785 dsphys->ds_fsid_guid = unique_create();
780 786 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
781 787 sizeof (dsphys->ds_guid));
782 788 dsphys->ds_snapnames_zapobj =
783 789 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
784 790 DMU_OT_NONE, 0, tx);
785 791 dsphys->ds_creation_time = gethrestime_sec();
786 792 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
787 793
788 794 if (origin == NULL) {
789 795 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
790 796 } else {
791 797 dsl_dataset_t *ohds; /* head of the origin snapshot */
792 798
793 799 dsphys->ds_prev_snap_obj = origin->ds_object;
794 800 dsphys->ds_prev_snap_txg =
795 801 dsl_dataset_phys(origin)->ds_creation_txg;
796 802 dsphys->ds_referenced_bytes =
797 803 dsl_dataset_phys(origin)->ds_referenced_bytes;
798 804 dsphys->ds_compressed_bytes =
799 805 dsl_dataset_phys(origin)->ds_compressed_bytes;
800 806 dsphys->ds_uncompressed_bytes =
801 807 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
802 808 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
803 809
804 810 /*
805 811 * Inherit flags that describe the dataset's contents
806 812 * (INCONSISTENT) or properties (Case Insensitive).
807 813 */
808 814 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
809 815 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
810 816
811 817 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
812 818 if (origin->ds_feature_inuse[f])
813 819 dsl_dataset_activate_feature(dsobj, f, tx);
814 820 }
815 821
816 822 dmu_buf_will_dirty(origin->ds_dbuf, tx);
817 823 dsl_dataset_phys(origin)->ds_num_children++;
818 824
819 825 VERIFY0(dsl_dataset_hold_obj(dp,
820 826 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
821 827 FTAG, &ohds));
822 828 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
823 829 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
824 830 dsl_dataset_rele(ohds, FTAG);
825 831
826 832 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
827 833 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
828 834 dsl_dataset_phys(origin)->ds_next_clones_obj =
829 835 zap_create(mos,
830 836 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
831 837 }
832 838 VERIFY0(zap_add_int(mos,
833 839 dsl_dataset_phys(origin)->ds_next_clones_obj,
834 840 dsobj, tx));
835 841 }
836 842
837 843 dmu_buf_will_dirty(dd->dd_dbuf, tx);
838 844 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
839 845 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
840 846 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
841 847 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
842 848 dsl_dir_phys(origin->ds_dir)->dd_clones =
843 849 zap_create(mos,
844 850 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
845 851 }
846 852 VERIFY0(zap_add_int(mos,
847 853 dsl_dir_phys(origin->ds_dir)->dd_clones,
848 854 dsobj, tx));
849 855 }
850 856 }
851 857
852 858 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
853 859 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
854 860
855 861 dmu_buf_rele(dbuf, FTAG);
856 862
857 863 dmu_buf_will_dirty(dd->dd_dbuf, tx);
858 864 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
859 865
860 866 return (dsobj);
861 867 }
862 868
863 869 static void
864 870 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
865 871 {
866 872 objset_t *os;
867 873
868 874 VERIFY0(dmu_objset_from_ds(ds, &os));
869 875 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
870 876 dsl_dataset_dirty(ds, tx);
871 877 }
872 878
873 879 uint64_t
874 880 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
875 881 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
876 882 {
877 883 dsl_pool_t *dp = pdd->dd_pool;
878 884 uint64_t dsobj, ddobj;
879 885 dsl_dir_t *dd;
880 886
881 887 ASSERT(dmu_tx_is_syncing(tx));
882 888 ASSERT(lastname[0] != '@');
883 889
884 890 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
885 891 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
886 892
887 893 dsobj = dsl_dataset_create_sync_dd(dd, origin,
888 894 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
889 895
890 896 dsl_deleg_set_create_perms(dd, tx, cr);
891 897
892 898 /*
893 899 * Since we're creating a new node we know it's a leaf, so we can
894 900 * initialize the counts if the limit feature is active.
895 901 */
896 902 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
897 903 uint64_t cnt = 0;
898 904 objset_t *os = dd->dd_pool->dp_meta_objset;
899 905
900 906 dsl_dir_zapify(dd, tx);
901 907 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
902 908 sizeof (cnt), 1, &cnt, tx));
903 909 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
904 910 sizeof (cnt), 1, &cnt, tx));
905 911 }
906 912
907 913 dsl_dir_rele(dd, FTAG);
908 914
909 915 /*
910 916 * If we are creating a clone, make sure we zero out any stale
911 917 * data from the origin snapshots zil header.
912 918 */
913 919 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
914 920 dsl_dataset_t *ds;
915 921
916 922 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
917 923 dsl_dataset_zero_zil(ds, tx);
918 924 dsl_dataset_rele(ds, FTAG);
919 925 }
920 926
921 927 return (dsobj);
922 928 }
923 929
924 930 /*
925 931 * The unique space in the head dataset can be calculated by subtracting
926 932 * the space used in the most recent snapshot, that is still being used
927 933 * in this file system, from the space currently in use. To figure out
928 934 * the space in the most recent snapshot still in use, we need to take
929 935 * the total space used in the snapshot and subtract out the space that
930 936 * has been freed up since the snapshot was taken.
931 937 */
932 938 void
933 939 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
934 940 {
935 941 uint64_t mrs_used;
936 942 uint64_t dlused, dlcomp, dluncomp;
937 943
938 944 ASSERT(!ds->ds_is_snapshot);
939 945
940 946 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
941 947 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
942 948 else
943 949 mrs_used = 0;
944 950
945 951 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
946 952
947 953 ASSERT3U(dlused, <=, mrs_used);
948 954 dsl_dataset_phys(ds)->ds_unique_bytes =
949 955 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
950 956
951 957 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
952 958 SPA_VERSION_UNIQUE_ACCURATE)
953 959 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
954 960 }
955 961
956 962 void
957 963 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
958 964 dmu_tx_t *tx)
959 965 {
960 966 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
961 967 uint64_t count;
962 968 int err;
963 969
964 970 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
965 971 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
966 972 obj, tx);
967 973 /*
968 974 * The err should not be ENOENT, but a bug in a previous version
969 975 * of the code could cause upgrade_clones_cb() to not set
970 976 * ds_next_snap_obj when it should, leading to a missing entry.
971 977 * If we knew that the pool was created after
972 978 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
973 979 * ENOENT. However, at least we can check that we don't have
974 980 * too many entries in the next_clones_obj even after failing to
975 981 * remove this one.
976 982 */
977 983 if (err != ENOENT)
978 984 VERIFY0(err);
979 985 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
980 986 &count));
981 987 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
982 988 }
983 989
984 990
985 991 blkptr_t *
986 992 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
987 993 {
988 994 return (&dsl_dataset_phys(ds)->ds_bp);
989 995 }
990 996
991 997 void
992 998 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
993 999 {
994 1000 ASSERT(dmu_tx_is_syncing(tx));
995 1001 /* If it's the meta-objset, set dp_meta_rootbp */
996 1002 if (ds == NULL) {
997 1003 tx->tx_pool->dp_meta_rootbp = *bp;
998 1004 } else {
999 1005 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1000 1006 dsl_dataset_phys(ds)->ds_bp = *bp;
1001 1007 }
1002 1008 }
1003 1009
1004 1010 spa_t *
1005 1011 dsl_dataset_get_spa(dsl_dataset_t *ds)
1006 1012 {
1007 1013 return (ds->ds_dir->dd_pool->dp_spa);
1008 1014 }
1009 1015
1010 1016 void
1011 1017 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1012 1018 {
1013 1019 dsl_pool_t *dp;
1014 1020
1015 1021 if (ds == NULL) /* this is the meta-objset */
1016 1022 return;
1017 1023
1018 1024 ASSERT(ds->ds_objset != NULL);
1019 1025
1020 1026 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1021 1027 panic("dirtying snapshot!");
1022 1028
1023 1029 dp = ds->ds_dir->dd_pool;
1024 1030
1025 1031 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1026 1032 /* up the hold count until we can be written out */
1027 1033 dmu_buf_add_ref(ds->ds_dbuf, ds);
1028 1034 }
1029 1035 }
1030 1036
1031 1037 boolean_t
1032 1038 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1033 1039 {
1034 1040 for (int t = 0; t < TXG_SIZE; t++) {
1035 1041 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1036 1042 ds, t))
1037 1043 return (B_TRUE);
1038 1044 }
1039 1045 return (B_FALSE);
1040 1046 }
1041 1047
1042 1048 static int
1043 1049 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1044 1050 {
1045 1051 uint64_t asize;
1046 1052
1047 1053 if (!dmu_tx_is_syncing(tx))
1048 1054 return (0);
1049 1055
1050 1056 /*
1051 1057 * If there's an fs-only reservation, any blocks that might become
1052 1058 * owned by the snapshot dataset must be accommodated by space
1053 1059 * outside of the reservation.
1054 1060 */
1055 1061 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1056 1062 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1057 1063 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1058 1064 return (SET_ERROR(ENOSPC));
1059 1065
1060 1066 /*
1061 1067 * Propagate any reserved space for this snapshot to other
1062 1068 * snapshot checks in this sync group.
1063 1069 */
1064 1070 if (asize > 0)
1065 1071 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1066 1072
1067 1073 return (0);
1068 1074 }
1069 1075
1070 1076 typedef struct dsl_dataset_snapshot_arg {
1071 1077 nvlist_t *ddsa_snaps;
1072 1078 nvlist_t *ddsa_props;
1073 1079 nvlist_t *ddsa_errors;
1074 1080 cred_t *ddsa_cr;
1075 1081 } dsl_dataset_snapshot_arg_t;
1076 1082
1077 1083 int
1078 1084 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1079 1085 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1080 1086 {
1081 1087 int error;
1082 1088 uint64_t value;
1083 1089
1084 1090 ds->ds_trysnap_txg = tx->tx_txg;
1085 1091
1086 1092 if (!dmu_tx_is_syncing(tx))
1087 1093 return (0);
1088 1094
1089 1095 /*
1090 1096 * We don't allow multiple snapshots of the same txg. If there
1091 1097 * is already one, try again.
1092 1098 */
1093 1099 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1094 1100 return (SET_ERROR(EAGAIN));
1095 1101
1096 1102 /*
1097 1103 * Check for conflicting snapshot name.
1098 1104 */
1099 1105 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1100 1106 if (error == 0)
1101 1107 return (SET_ERROR(EEXIST));
1102 1108 if (error != ENOENT)
1103 1109 return (error);
1104 1110
1105 1111 /*
1106 1112 * We don't allow taking snapshots of inconsistent datasets, such as
1107 1113 * those into which we are currently receiving. However, if we are
1108 1114 * creating this snapshot as part of a receive, this check will be
1109 1115 * executed atomically with respect to the completion of the receive
1110 1116 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1111 1117 * case we ignore this, knowing it will be fixed up for us shortly in
1112 1118 * dmu_recv_end_sync().
1113 1119 */
1114 1120 if (!recv && DS_IS_INCONSISTENT(ds))
1115 1121 return (SET_ERROR(EBUSY));
1116 1122
1117 1123 /*
1118 1124 * Skip the check for temporary snapshots or if we have already checked
1119 1125 * the counts in dsl_dataset_snapshot_check. This means we really only
1120 1126 * check the count here when we're receiving a stream.
1121 1127 */
1122 1128 if (cnt != 0 && cr != NULL) {
1123 1129 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1124 1130 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1125 1131 if (error != 0)
1126 1132 return (error);
1127 1133 }
1128 1134
1129 1135 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1130 1136 if (error != 0)
1131 1137 return (error);
1132 1138
1133 1139 return (0);
1134 1140 }
1135 1141
1136 1142 static int
1137 1143 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1138 1144 {
1139 1145 dsl_dataset_snapshot_arg_t *ddsa = arg;
1140 1146 dsl_pool_t *dp = dmu_tx_pool(tx);
1141 1147 nvpair_t *pair;
1142 1148 int rv = 0;
1143 1149
1144 1150 /*
1145 1151 * Pre-compute how many total new snapshots will be created for each
1146 1152 * level in the tree and below. This is needed for validating the
1147 1153 * snapshot limit when either taking a recursive snapshot or when
1148 1154 * taking multiple snapshots.
1149 1155 *
1150 1156 * The problem is that the counts are not actually adjusted when
1151 1157 * we are checking, only when we finally sync. For a single snapshot,
1152 1158 * this is easy, the count will increase by 1 at each node up the tree,
1153 1159 * but its more complicated for the recursive/multiple snapshot case.
1154 1160 *
1155 1161 * The dsl_fs_ss_limit_check function does recursively check the count
1156 1162 * at each level up the tree but since it is validating each snapshot
1157 1163 * independently we need to be sure that we are validating the complete
1158 1164 * count for the entire set of snapshots. We do this by rolling up the
1159 1165 * counts for each component of the name into an nvlist and then
1160 1166 * checking each of those cases with the aggregated count.
1161 1167 *
1162 1168 * This approach properly handles not only the recursive snapshot
1163 1169 * case (where we get all of those on the ddsa_snaps list) but also
1164 1170 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1165 1171 * validate the limit on 'a' using a count of 2).
1166 1172 *
1167 1173 * We validate the snapshot names in the third loop and only report
1168 1174 * name errors once.
1169 1175 */
1170 1176 if (dmu_tx_is_syncing(tx)) {
1171 1177 nvlist_t *cnt_track = NULL;
1172 1178 cnt_track = fnvlist_alloc();
1173 1179
1174 1180 /* Rollup aggregated counts into the cnt_track list */
1175 1181 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1176 1182 pair != NULL;
1177 1183 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1178 1184 char *pdelim;
1179 1185 uint64_t val;
1180 1186 char nm[MAXPATHLEN];
1181 1187
1182 1188 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1183 1189 pdelim = strchr(nm, '@');
1184 1190 if (pdelim == NULL)
1185 1191 continue;
1186 1192 *pdelim = '\0';
1187 1193
1188 1194 do {
1189 1195 if (nvlist_lookup_uint64(cnt_track, nm,
1190 1196 &val) == 0) {
1191 1197 /* update existing entry */
1192 1198 fnvlist_add_uint64(cnt_track, nm,
1193 1199 val + 1);
1194 1200 } else {
1195 1201 /* add to list */
1196 1202 fnvlist_add_uint64(cnt_track, nm, 1);
1197 1203 }
1198 1204
1199 1205 pdelim = strrchr(nm, '/');
1200 1206 if (pdelim != NULL)
1201 1207 *pdelim = '\0';
1202 1208 } while (pdelim != NULL);
1203 1209 }
1204 1210
1205 1211 /* Check aggregated counts at each level */
1206 1212 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1207 1213 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1208 1214 int error = 0;
1209 1215 char *name;
1210 1216 uint64_t cnt = 0;
1211 1217 dsl_dataset_t *ds;
1212 1218
1213 1219 name = nvpair_name(pair);
1214 1220 cnt = fnvpair_value_uint64(pair);
1215 1221 ASSERT(cnt > 0);
1216 1222
1217 1223 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1218 1224 if (error == 0) {
1219 1225 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1220 1226 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1221 1227 ddsa->ddsa_cr);
1222 1228 dsl_dataset_rele(ds, FTAG);
1223 1229 }
1224 1230
1225 1231 if (error != 0) {
1226 1232 if (ddsa->ddsa_errors != NULL)
1227 1233 fnvlist_add_int32(ddsa->ddsa_errors,
1228 1234 name, error);
1229 1235 rv = error;
1230 1236 /* only report one error for this check */
1231 1237 break;
1232 1238 }
1233 1239 }
1234 1240 nvlist_free(cnt_track);
1235 1241 }
1236 1242
1237 1243 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1238 1244 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1239 1245 int error = 0;
1240 1246 dsl_dataset_t *ds;
1241 1247 char *name, *atp;
1242 1248 char dsname[MAXNAMELEN];
1243 1249
1244 1250 name = nvpair_name(pair);
1245 1251 if (strlen(name) >= MAXNAMELEN)
1246 1252 error = SET_ERROR(ENAMETOOLONG);
1247 1253 if (error == 0) {
1248 1254 atp = strchr(name, '@');
1249 1255 if (atp == NULL)
1250 1256 error = SET_ERROR(EINVAL);
1251 1257 if (error == 0)
1252 1258 (void) strlcpy(dsname, name, atp - name + 1);
1253 1259 }
1254 1260 if (error == 0)
1255 1261 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1256 1262 if (error == 0) {
1257 1263 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1258 1264 error = dsl_dataset_snapshot_check_impl(ds,
1259 1265 atp + 1, tx, B_FALSE, 0, NULL);
1260 1266 dsl_dataset_rele(ds, FTAG);
1261 1267 }
1262 1268
1263 1269 if (error != 0) {
1264 1270 if (ddsa->ddsa_errors != NULL) {
1265 1271 fnvlist_add_int32(ddsa->ddsa_errors,
1266 1272 name, error);
1267 1273 }
1268 1274 rv = error;
1269 1275 }
1270 1276 }
1271 1277
1272 1278 return (rv);
1273 1279 }
1274 1280
1275 1281 void
1276 1282 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1277 1283 dmu_tx_t *tx)
1278 1284 {
1279 1285 static zil_header_t zero_zil;
1280 1286
1281 1287 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1282 1288 dmu_buf_t *dbuf;
1283 1289 dsl_dataset_phys_t *dsphys;
1284 1290 uint64_t dsobj, crtxg;
1285 1291 objset_t *mos = dp->dp_meta_objset;
1286 1292 objset_t *os;
1287 1293
1288 1294 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1289 1295
1290 1296 /*
1291 1297 * If we are on an old pool, the zil must not be active, in which
1292 1298 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1293 1299 */
1294 1300 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1295 1301 dmu_objset_from_ds(ds, &os) != 0 ||
1296 1302 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1297 1303 sizeof (zero_zil)) == 0);
1298 1304
1299 1305 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1300 1306
1301 1307 /*
1302 1308 * The origin's ds_creation_txg has to be < TXG_INITIAL
1303 1309 */
1304 1310 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1305 1311 crtxg = 1;
1306 1312 else
1307 1313 crtxg = tx->tx_txg;
1308 1314
1309 1315 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1310 1316 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1311 1317 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1312 1318 dmu_buf_will_dirty(dbuf, tx);
1313 1319 dsphys = dbuf->db_data;
1314 1320 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1315 1321 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1316 1322 dsphys->ds_fsid_guid = unique_create();
1317 1323 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1318 1324 sizeof (dsphys->ds_guid));
1319 1325 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1320 1326 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1321 1327 dsphys->ds_next_snap_obj = ds->ds_object;
1322 1328 dsphys->ds_num_children = 1;
1323 1329 dsphys->ds_creation_time = gethrestime_sec();
1324 1330 dsphys->ds_creation_txg = crtxg;
1325 1331 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1326 1332 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1327 1333 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1328 1334 dsphys->ds_uncompressed_bytes =
1329 1335 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1330 1336 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1331 1337 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1332 1338 dmu_buf_rele(dbuf, FTAG);
1333 1339
1334 1340 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1335 1341 if (ds->ds_feature_inuse[f])
1336 1342 dsl_dataset_activate_feature(dsobj, f, tx);
1337 1343 }
1338 1344
1339 1345 ASSERT3U(ds->ds_prev != 0, ==,
1340 1346 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1341 1347 if (ds->ds_prev) {
1342 1348 uint64_t next_clones_obj =
1343 1349 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1344 1350 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1345 1351 ds->ds_object ||
1346 1352 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1347 1353 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1348 1354 ds->ds_object) {
1349 1355 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1350 1356 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1351 1357 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1352 1358 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1353 1359 } else if (next_clones_obj != 0) {
1354 1360 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1355 1361 dsphys->ds_next_snap_obj, tx);
1356 1362 VERIFY0(zap_add_int(mos,
1357 1363 next_clones_obj, dsobj, tx));
1358 1364 }
1359 1365 }
1360 1366
1361 1367 /*
1362 1368 * If we have a reference-reservation on this dataset, we will
1363 1369 * need to increase the amount of refreservation being charged
1364 1370 * since our unique space is going to zero.
1365 1371 */
1366 1372 if (ds->ds_reserved) {
1367 1373 int64_t delta;
1368 1374 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1369 1375 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1370 1376 ds->ds_reserved);
1371 1377 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1372 1378 delta, 0, 0, tx);
1373 1379 }
1374 1380
1375 1381 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1376 1382 dsl_dataset_phys(ds)->ds_deadlist_obj =
1377 1383 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1378 1384 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1379 1385 dsl_deadlist_close(&ds->ds_deadlist);
1380 1386 dsl_deadlist_open(&ds->ds_deadlist, mos,
1381 1387 dsl_dataset_phys(ds)->ds_deadlist_obj);
1382 1388 dsl_deadlist_add_key(&ds->ds_deadlist,
1383 1389 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1384 1390
1385 1391 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1386 1392 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1387 1393 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1388 1394 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1389 1395 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1390 1396 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1391 1397
1392 1398 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1393 1399 snapname, 8, 1, &dsobj, tx));
1394 1400
1395 1401 if (ds->ds_prev)
1396 1402 dsl_dataset_rele(ds->ds_prev, ds);
1397 1403 VERIFY0(dsl_dataset_hold_obj(dp,
1398 1404 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1399 1405
1400 1406 dsl_scan_ds_snapshotted(ds, tx);
1401 1407
1402 1408 dsl_dir_snap_cmtime_update(ds->ds_dir);
1403 1409
1404 1410 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1405 1411 }
1406 1412
1407 1413 static void
1408 1414 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1409 1415 {
1410 1416 dsl_dataset_snapshot_arg_t *ddsa = arg;
1411 1417 dsl_pool_t *dp = dmu_tx_pool(tx);
1412 1418 nvpair_t *pair;
1413 1419
1414 1420 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1415 1421 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1416 1422 dsl_dataset_t *ds;
1417 1423 char *name, *atp;
1418 1424 char dsname[MAXNAMELEN];
1419 1425
1420 1426 name = nvpair_name(pair);
1421 1427 atp = strchr(name, '@');
1422 1428 (void) strlcpy(dsname, name, atp - name + 1);
1423 1429 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1424 1430
1425 1431 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1426 1432 if (ddsa->ddsa_props != NULL) {
1427 1433 dsl_props_set_sync_impl(ds->ds_prev,
1428 1434 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1429 1435 }
1430 1436 dsl_dataset_rele(ds, FTAG);
1431 1437 }
1432 1438 }
1433 1439
1434 1440 /*
1435 1441 * The snapshots must all be in the same pool.
1436 1442 * All-or-nothing: if there are any failures, nothing will be modified.
1437 1443 */
1438 1444 int
1439 1445 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1440 1446 {
1441 1447 dsl_dataset_snapshot_arg_t ddsa;
1442 1448 nvpair_t *pair;
1443 1449 boolean_t needsuspend;
1444 1450 int error;
1445 1451 spa_t *spa;
1446 1452 char *firstname;
1447 1453 nvlist_t *suspended = NULL;
1448 1454
1449 1455 pair = nvlist_next_nvpair(snaps, NULL);
1450 1456 if (pair == NULL)
1451 1457 return (0);
1452 1458 firstname = nvpair_name(pair);
1453 1459
1454 1460 error = spa_open(firstname, &spa, FTAG);
1455 1461 if (error != 0)
1456 1462 return (error);
1457 1463 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1458 1464 spa_close(spa, FTAG);
1459 1465
1460 1466 if (needsuspend) {
1461 1467 suspended = fnvlist_alloc();
1462 1468 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1463 1469 pair = nvlist_next_nvpair(snaps, pair)) {
1464 1470 char fsname[MAXNAMELEN];
1465 1471 char *snapname = nvpair_name(pair);
1466 1472 char *atp;
1467 1473 void *cookie;
1468 1474
1469 1475 atp = strchr(snapname, '@');
1470 1476 if (atp == NULL) {
1471 1477 error = SET_ERROR(EINVAL);
1472 1478 break;
1473 1479 }
1474 1480 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1475 1481
1476 1482 error = zil_suspend(fsname, &cookie);
1477 1483 if (error != 0)
1478 1484 break;
1479 1485 fnvlist_add_uint64(suspended, fsname,
1480 1486 (uintptr_t)cookie);
1481 1487 }
1482 1488 }
1483 1489
1484 1490 ddsa.ddsa_snaps = snaps;
1485 1491 ddsa.ddsa_props = props;
1486 1492 ddsa.ddsa_errors = errors;
1487 1493 ddsa.ddsa_cr = CRED();
1488 1494
1489 1495 if (error == 0) {
1490 1496 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1491 1497 dsl_dataset_snapshot_sync, &ddsa,
1492 1498 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1493 1499 }
1494 1500
1495 1501 if (suspended != NULL) {
1496 1502 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1497 1503 pair = nvlist_next_nvpair(suspended, pair)) {
1498 1504 zil_resume((void *)(uintptr_t)
1499 1505 fnvpair_value_uint64(pair));
1500 1506 }
1501 1507 fnvlist_free(suspended);
1502 1508 }
1503 1509
1504 1510 return (error);
1505 1511 }
1506 1512
1507 1513 typedef struct dsl_dataset_snapshot_tmp_arg {
1508 1514 const char *ddsta_fsname;
1509 1515 const char *ddsta_snapname;
1510 1516 minor_t ddsta_cleanup_minor;
1511 1517 const char *ddsta_htag;
1512 1518 } dsl_dataset_snapshot_tmp_arg_t;
1513 1519
1514 1520 static int
1515 1521 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1516 1522 {
1517 1523 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1518 1524 dsl_pool_t *dp = dmu_tx_pool(tx);
1519 1525 dsl_dataset_t *ds;
1520 1526 int error;
1521 1527
1522 1528 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1523 1529 if (error != 0)
1524 1530 return (error);
1525 1531
1526 1532 /* NULL cred means no limit check for tmp snapshot */
1527 1533 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1528 1534 tx, B_FALSE, 0, NULL);
1529 1535 if (error != 0) {
1530 1536 dsl_dataset_rele(ds, FTAG);
1531 1537 return (error);
1532 1538 }
1533 1539
1534 1540 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1535 1541 dsl_dataset_rele(ds, FTAG);
1536 1542 return (SET_ERROR(ENOTSUP));
1537 1543 }
1538 1544 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1539 1545 B_TRUE, tx);
1540 1546 if (error != 0) {
1541 1547 dsl_dataset_rele(ds, FTAG);
1542 1548 return (error);
1543 1549 }
1544 1550
1545 1551 dsl_dataset_rele(ds, FTAG);
1546 1552 return (0);
1547 1553 }
1548 1554
1549 1555 static void
1550 1556 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1551 1557 {
1552 1558 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1553 1559 dsl_pool_t *dp = dmu_tx_pool(tx);
1554 1560 dsl_dataset_t *ds;
1555 1561
1556 1562 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1557 1563
1558 1564 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1559 1565 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1560 1566 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1561 1567 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1562 1568
1563 1569 dsl_dataset_rele(ds, FTAG);
1564 1570 }
1565 1571
1566 1572 int
1567 1573 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1568 1574 minor_t cleanup_minor, const char *htag)
1569 1575 {
1570 1576 dsl_dataset_snapshot_tmp_arg_t ddsta;
1571 1577 int error;
1572 1578 spa_t *spa;
1573 1579 boolean_t needsuspend;
1574 1580 void *cookie;
1575 1581
1576 1582 ddsta.ddsta_fsname = fsname;
1577 1583 ddsta.ddsta_snapname = snapname;
1578 1584 ddsta.ddsta_cleanup_minor = cleanup_minor;
1579 1585 ddsta.ddsta_htag = htag;
1580 1586
1581 1587 error = spa_open(fsname, &spa, FTAG);
1582 1588 if (error != 0)
1583 1589 return (error);
1584 1590 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1585 1591 spa_close(spa, FTAG);
1586 1592
1587 1593 if (needsuspend) {
1588 1594 error = zil_suspend(fsname, &cookie);
1589 1595 if (error != 0)
1590 1596 return (error);
1591 1597 }
1592 1598
1593 1599 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1594 1600 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1595 1601
1596 1602 if (needsuspend)
1597 1603 zil_resume(cookie);
1598 1604 return (error);
1599 1605 }
1600 1606
1601 1607
1602 1608 void
1603 1609 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1604 1610 {
1605 1611 ASSERT(dmu_tx_is_syncing(tx));
1606 1612 ASSERT(ds->ds_objset != NULL);
1607 1613 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1608 1614
1609 1615 /*
1610 1616 * in case we had to change ds_fsid_guid when we opened it,
1611 1617 * sync it out now.
1612 1618 */
1613 1619 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1614 1620 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1615 1621
1616 1622 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1617 1623 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1618 1624 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1619 1625 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1620 1626 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1621 1627 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1622 1628 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1623 1629 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1624 1630 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1625 1631 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1626 1632 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1627 1633 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1628 1634 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1629 1635 }
1630 1636
1631 1637 dmu_objset_sync(ds->ds_objset, zio, tx);
1632 1638
1633 1639 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1634 1640 if (ds->ds_feature_activation_needed[f]) {
1635 1641 if (ds->ds_feature_inuse[f])
1636 1642 continue;
1637 1643 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1638 1644 ds->ds_feature_inuse[f] = B_TRUE;
1639 1645 }
1640 1646 }
1641 1647 }
1642 1648
1643 1649 static void
1644 1650 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1645 1651 {
1646 1652 uint64_t count = 0;
1647 1653 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1648 1654 zap_cursor_t zc;
1649 1655 zap_attribute_t za;
1650 1656 nvlist_t *propval = fnvlist_alloc();
1651 1657 nvlist_t *val = fnvlist_alloc();
1652 1658
1653 1659 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1654 1660
1655 1661 /*
1656 1662 * There may be missing entries in ds_next_clones_obj
1657 1663 * due to a bug in a previous version of the code.
1658 1664 * Only trust it if it has the right number of entries.
1659 1665 */
1660 1666 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1661 1667 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1662 1668 &count));
1663 1669 }
1664 1670 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1665 1671 goto fail;
1666 1672 for (zap_cursor_init(&zc, mos,
1667 1673 dsl_dataset_phys(ds)->ds_next_clones_obj);
1668 1674 zap_cursor_retrieve(&zc, &za) == 0;
1669 1675 zap_cursor_advance(&zc)) {
1670 1676 dsl_dataset_t *clone;
1671 1677 char buf[ZFS_MAXNAMELEN];
1672 1678 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1673 1679 za.za_first_integer, FTAG, &clone));
1674 1680 dsl_dir_name(clone->ds_dir, buf);
1675 1681 fnvlist_add_boolean(val, buf);
1676 1682 dsl_dataset_rele(clone, FTAG);
1677 1683 }
1678 1684 zap_cursor_fini(&zc);
1679 1685 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1680 1686 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1681 1687 fail:
1682 1688 nvlist_free(val);
1683 1689 nvlist_free(propval);
1684 1690 }
1685 1691
1686 1692 static void
1687 1693 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1688 1694 {
1689 1695 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1690 1696
1691 1697 if (dsl_dataset_has_resume_receive_state(ds)) {
1692 1698 char *str;
1693 1699 void *packed;
1694 1700 uint8_t *compressed;
1695 1701 uint64_t val;
1696 1702 nvlist_t *token_nv = fnvlist_alloc();
1697 1703 size_t packed_size, compressed_size;
1698 1704
1699 1705 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1700 1706 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1701 1707 fnvlist_add_uint64(token_nv, "fromguid", val);
1702 1708 }
1703 1709 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1704 1710 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1705 1711 fnvlist_add_uint64(token_nv, "object", val);
1706 1712 }
1707 1713 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1708 1714 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1709 1715 fnvlist_add_uint64(token_nv, "offset", val);
1710 1716 }
1711 1717 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1712 1718 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1713 1719 fnvlist_add_uint64(token_nv, "bytes", val);
1714 1720 }
1715 1721 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1716 1722 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1717 1723 fnvlist_add_uint64(token_nv, "toguid", val);
1718 1724 }
1719 1725 char buf[256];
1720 1726 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1721 1727 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1722 1728 fnvlist_add_string(token_nv, "toname", buf);
1723 1729 }
1724 1730 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1725 1731 DS_FIELD_RESUME_EMBEDOK) == 0) {
1726 1732 fnvlist_add_boolean(token_nv, "embedok");
1727 1733 }
1728 1734 packed = fnvlist_pack(token_nv, &packed_size);
1729 1735 fnvlist_free(token_nv);
1730 1736 compressed = kmem_alloc(packed_size, KM_SLEEP);
1731 1737
1732 1738 compressed_size = gzip_compress(packed, compressed,
1733 1739 packed_size, packed_size, 6);
1734 1740
1735 1741 zio_cksum_t cksum;
1736 1742 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1737 1743
1738 1744 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1739 1745 for (int i = 0; i < compressed_size; i++) {
1740 1746 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1741 1747 }
1742 1748 str[compressed_size * 2] = '\0';
1743 1749 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1744 1750 ZFS_SEND_RESUME_TOKEN_VERSION,
1745 1751 (longlong_t)cksum.zc_word[0],
1746 1752 (longlong_t)packed_size, str);
1747 1753 dsl_prop_nvlist_add_string(nv,
1748 1754 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1749 1755 kmem_free(packed, packed_size);
1750 1756 kmem_free(str, compressed_size * 2 + 1);
1751 1757 kmem_free(compressed, packed_size);
1752 1758 strfree(propval);
1753 1759 }
1754 1760 }
1755 1761
1756 1762 void
1757 1763 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1758 1764 {
1759 1765 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1760 1766 uint64_t refd, avail, uobjs, aobjs, ratio;
1761 1767
1762 1768 ASSERT(dsl_pool_config_held(dp));
1763 1769
1764 1770 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1765 1771 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1766 1772 dsl_dataset_phys(ds)->ds_compressed_bytes);
1767 1773
1768 1774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1769 1775 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1770 1776 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1771 1777
1772 1778 if (ds->ds_is_snapshot) {
1773 1779 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1774 1780 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1775 1781 dsl_dataset_phys(ds)->ds_unique_bytes);
1776 1782 get_clones_stat(ds, nv);
1777 1783 } else {
1778 1784 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1779 1785 char buf[MAXNAMELEN];
1780 1786 dsl_dataset_name(ds->ds_prev, buf);
1781 1787 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1782 1788 }
1783 1789
1784 1790 dsl_dir_stats(ds->ds_dir, nv);
1785 1791 }
1786 1792
1787 1793 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1788 1794 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1789 1795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1790 1796
1791 1797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1792 1798 dsl_dataset_phys(ds)->ds_creation_time);
1793 1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1794 1800 dsl_dataset_phys(ds)->ds_creation_txg);
1795 1801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1796 1802 ds->ds_quota);
1797 1803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1798 1804 ds->ds_reserved);
1799 1805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1800 1806 dsl_dataset_phys(ds)->ds_guid);
1801 1807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1802 1808 dsl_dataset_phys(ds)->ds_unique_bytes);
1803 1809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1804 1810 ds->ds_object);
1805 1811 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1806 1812 ds->ds_userrefs);
1807 1813 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1808 1814 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1809 1815
1810 1816 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1811 1817 uint64_t written, comp, uncomp;
1812 1818 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1813 1819 dsl_dataset_t *prev;
1814 1820
1815 1821 int err = dsl_dataset_hold_obj(dp,
1816 1822 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1817 1823 if (err == 0) {
1818 1824 err = dsl_dataset_space_written(prev, ds, &written,
1819 1825 &comp, &uncomp);
1820 1826 dsl_dataset_rele(prev, FTAG);
1821 1827 if (err == 0) {
1822 1828 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1823 1829 written);
1824 1830 }
1825 1831 }
1826 1832 }
1827 1833
1828 1834 if (!dsl_dataset_is_snapshot(ds)) {
1829 1835 /*
1830 1836 * A failed "newfs" (e.g. full) resumable receive leaves
1831 1837 * the stats set on this dataset. Check here for the prop.
1832 1838 */
1833 1839 get_receive_resume_stats(ds, nv);
1834 1840
1835 1841 /*
1836 1842 * A failed incremental resumable receive leaves the
1837 1843 * stats set on our child named "%recv". Check the child
1838 1844 * for the prop.
1839 1845 */
1840 1846 char recvname[ZFS_MAXNAMELEN];
1841 1847 dsl_dataset_t *recv_ds;
1842 1848 dsl_dataset_name(ds, recvname);
1843 1849 (void) strcat(recvname, "/");
1844 1850 (void) strcat(recvname, recv_clone_name);
1845 1851 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1846 1852 get_receive_resume_stats(recv_ds, nv);
1847 1853 dsl_dataset_rele(recv_ds, FTAG);
1848 1854 }
1849 1855 }
1850 1856 }
1851 1857
1852 1858 void
1853 1859 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1854 1860 {
1855 1861 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1856 1862 ASSERT(dsl_pool_config_held(dp));
1857 1863
1858 1864 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1859 1865 stat->dds_inconsistent =
1860 1866 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1861 1867 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1862 1868 stat->dds_origin[0] = '\0';
1863 1869 if (ds->ds_is_snapshot) {
1864 1870 stat->dds_is_snapshot = B_TRUE;
1865 1871 stat->dds_num_clones =
1866 1872 dsl_dataset_phys(ds)->ds_num_children - 1;
1867 1873 } else {
1868 1874 stat->dds_is_snapshot = B_FALSE;
1869 1875 stat->dds_num_clones = 0;
1870 1876
1871 1877 if (dsl_dir_is_clone(ds->ds_dir)) {
1872 1878 dsl_dataset_t *ods;
1873 1879
1874 1880 VERIFY0(dsl_dataset_hold_obj(dp,
1875 1881 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1876 1882 FTAG, &ods));
1877 1883 dsl_dataset_name(ods, stat->dds_origin);
1878 1884 dsl_dataset_rele(ods, FTAG);
1879 1885 }
1880 1886 }
1881 1887 }
1882 1888
1883 1889 uint64_t
1884 1890 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1885 1891 {
1886 1892 return (ds->ds_fsid_guid);
1887 1893 }
1888 1894
1889 1895 void
1890 1896 dsl_dataset_space(dsl_dataset_t *ds,
1891 1897 uint64_t *refdbytesp, uint64_t *availbytesp,
1892 1898 uint64_t *usedobjsp, uint64_t *availobjsp)
1893 1899 {
1894 1900 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1895 1901 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1896 1902 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1897 1903 *availbytesp +=
1898 1904 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1899 1905 if (ds->ds_quota != 0) {
1900 1906 /*
1901 1907 * Adjust available bytes according to refquota
1902 1908 */
1903 1909 if (*refdbytesp < ds->ds_quota)
1904 1910 *availbytesp = MIN(*availbytesp,
1905 1911 ds->ds_quota - *refdbytesp);
1906 1912 else
1907 1913 *availbytesp = 0;
1908 1914 }
1909 1915 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1910 1916 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1911 1917 }
1912 1918
1913 1919 boolean_t
1914 1920 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1915 1921 {
1916 1922 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1917 1923
1918 1924 ASSERT(dsl_pool_config_held(dp));
1919 1925 if (snap == NULL)
1920 1926 return (B_FALSE);
1921 1927 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1922 1928 dsl_dataset_phys(snap)->ds_creation_txg) {
1923 1929 objset_t *os, *os_snap;
1924 1930 /*
1925 1931 * It may be that only the ZIL differs, because it was
1926 1932 * reset in the head. Don't count that as being
1927 1933 * modified.
1928 1934 */
1929 1935 if (dmu_objset_from_ds(ds, &os) != 0)
1930 1936 return (B_TRUE);
1931 1937 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1932 1938 return (B_TRUE);
1933 1939 return (bcmp(&os->os_phys->os_meta_dnode,
1934 1940 &os_snap->os_phys->os_meta_dnode,
1935 1941 sizeof (os->os_phys->os_meta_dnode)) != 0);
1936 1942 }
1937 1943 return (B_FALSE);
1938 1944 }
1939 1945
1940 1946 typedef struct dsl_dataset_rename_snapshot_arg {
1941 1947 const char *ddrsa_fsname;
1942 1948 const char *ddrsa_oldsnapname;
1943 1949 const char *ddrsa_newsnapname;
1944 1950 boolean_t ddrsa_recursive;
1945 1951 dmu_tx_t *ddrsa_tx;
1946 1952 } dsl_dataset_rename_snapshot_arg_t;
1947 1953
1948 1954 /* ARGSUSED */
1949 1955 static int
1950 1956 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1951 1957 dsl_dataset_t *hds, void *arg)
1952 1958 {
1953 1959 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1954 1960 int error;
1955 1961 uint64_t val;
1956 1962
1957 1963 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1958 1964 if (error != 0) {
1959 1965 /* ignore nonexistent snapshots */
1960 1966 return (error == ENOENT ? 0 : error);
1961 1967 }
1962 1968
1963 1969 /* new name should not exist */
1964 1970 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1965 1971 if (error == 0)
1966 1972 error = SET_ERROR(EEXIST);
1967 1973 else if (error == ENOENT)
1968 1974 error = 0;
1969 1975
1970 1976 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1971 1977 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1972 1978 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1973 1979 error = SET_ERROR(ENAMETOOLONG);
1974 1980
1975 1981 return (error);
1976 1982 }
1977 1983
1978 1984 static int
1979 1985 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1980 1986 {
1981 1987 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1982 1988 dsl_pool_t *dp = dmu_tx_pool(tx);
1983 1989 dsl_dataset_t *hds;
1984 1990 int error;
1985 1991
1986 1992 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1987 1993 if (error != 0)
1988 1994 return (error);
1989 1995
1990 1996 if (ddrsa->ddrsa_recursive) {
1991 1997 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1992 1998 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1993 1999 DS_FIND_CHILDREN);
1994 2000 } else {
1995 2001 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1996 2002 }
1997 2003 dsl_dataset_rele(hds, FTAG);
1998 2004 return (error);
1999 2005 }
2000 2006
2001 2007 static int
2002 2008 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2003 2009 dsl_dataset_t *hds, void *arg)
2004 2010 {
2005 2011 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2006 2012 dsl_dataset_t *ds;
2007 2013 uint64_t val;
2008 2014 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2009 2015 int error;
2010 2016
2011 2017 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2012 2018 ASSERT(error == 0 || error == ENOENT);
2013 2019 if (error == ENOENT) {
2014 2020 /* ignore nonexistent snapshots */
2015 2021 return (0);
2016 2022 }
2017 2023
2018 2024 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2019 2025
2020 2026 /* log before we change the name */
2021 2027 spa_history_log_internal_ds(ds, "rename", tx,
2022 2028 "-> @%s", ddrsa->ddrsa_newsnapname);
2023 2029
2024 2030 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2025 2031 B_FALSE));
2026 2032 mutex_enter(&ds->ds_lock);
2027 2033 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2028 2034 mutex_exit(&ds->ds_lock);
2029 2035 VERIFY0(zap_add(dp->dp_meta_objset,
2030 2036 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2031 2037 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2032 2038
2033 2039 dsl_dataset_rele(ds, FTAG);
2034 2040 return (0);
2035 2041 }
2036 2042
2037 2043 static void
2038 2044 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2039 2045 {
2040 2046 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2041 2047 dsl_pool_t *dp = dmu_tx_pool(tx);
2042 2048 dsl_dataset_t *hds;
2043 2049
2044 2050 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2045 2051 ddrsa->ddrsa_tx = tx;
2046 2052 if (ddrsa->ddrsa_recursive) {
2047 2053 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2048 2054 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2049 2055 DS_FIND_CHILDREN));
2050 2056 } else {
2051 2057 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2052 2058 }
2053 2059 dsl_dataset_rele(hds, FTAG);
2054 2060 }
2055 2061
2056 2062 int
2057 2063 dsl_dataset_rename_snapshot(const char *fsname,
2058 2064 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2059 2065 {
2060 2066 dsl_dataset_rename_snapshot_arg_t ddrsa;
2061 2067
2062 2068 ddrsa.ddrsa_fsname = fsname;
2063 2069 ddrsa.ddrsa_oldsnapname = oldsnapname;
2064 2070 ddrsa.ddrsa_newsnapname = newsnapname;
2065 2071 ddrsa.ddrsa_recursive = recursive;
2066 2072
2067 2073 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2068 2074 dsl_dataset_rename_snapshot_sync, &ddrsa,
2069 2075 1, ZFS_SPACE_CHECK_RESERVED));
2070 2076 }
2071 2077
2072 2078 /*
2073 2079 * If we're doing an ownership handoff, we need to make sure that there is
2074 2080 * only one long hold on the dataset. We're not allowed to change anything here
2075 2081 * so we don't permanently release the long hold or regular hold here. We want
2076 2082 * to do this only when syncing to avoid the dataset unexpectedly going away
2077 2083 * when we release the long hold.
2078 2084 */
2079 2085 static int
2080 2086 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2081 2087 {
2082 2088 boolean_t held;
2083 2089
2084 2090 if (!dmu_tx_is_syncing(tx))
2085 2091 return (0);
2086 2092
2087 2093 if (owner != NULL) {
2088 2094 VERIFY3P(ds->ds_owner, ==, owner);
2089 2095 dsl_dataset_long_rele(ds, owner);
2090 2096 }
2091 2097
2092 2098 held = dsl_dataset_long_held(ds);
2093 2099
2094 2100 if (owner != NULL)
2095 2101 dsl_dataset_long_hold(ds, owner);
2096 2102
2097 2103 if (held)
2098 2104 return (SET_ERROR(EBUSY));
2099 2105
2100 2106 return (0);
2101 2107 }
2102 2108
2103 2109 typedef struct dsl_dataset_rollback_arg {
2104 2110 const char *ddra_fsname;
2105 2111 void *ddra_owner;
2106 2112 nvlist_t *ddra_result;
2107 2113 } dsl_dataset_rollback_arg_t;
2108 2114
2109 2115 static int
2110 2116 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2111 2117 {
2112 2118 dsl_dataset_rollback_arg_t *ddra = arg;
2113 2119 dsl_pool_t *dp = dmu_tx_pool(tx);
2114 2120 dsl_dataset_t *ds;
2115 2121 int64_t unused_refres_delta;
2116 2122 int error;
2117 2123
2118 2124 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2119 2125 if (error != 0)
2120 2126 return (error);
2121 2127
2122 2128 /* must not be a snapshot */
2123 2129 if (ds->ds_is_snapshot) {
2124 2130 dsl_dataset_rele(ds, FTAG);
2125 2131 return (SET_ERROR(EINVAL));
2126 2132 }
2127 2133
2128 2134 /* must have a most recent snapshot */
2129 2135 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2130 2136 dsl_dataset_rele(ds, FTAG);
2131 2137 return (SET_ERROR(EINVAL));
2132 2138 }
2133 2139
2134 2140 /* must not have any bookmarks after the most recent snapshot */
2135 2141 nvlist_t *proprequest = fnvlist_alloc();
2136 2142 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2137 2143 nvlist_t *bookmarks = fnvlist_alloc();
2138 2144 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2139 2145 fnvlist_free(proprequest);
2140 2146 if (error != 0)
2141 2147 return (error);
2142 2148 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2143 2149 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2144 2150 nvlist_t *valuenv =
2145 2151 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2146 2152 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2147 2153 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2148 2154 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2149 2155 fnvlist_free(bookmarks);
2150 2156 dsl_dataset_rele(ds, FTAG);
2151 2157 return (SET_ERROR(EEXIST));
2152 2158 }
2153 2159 }
2154 2160 fnvlist_free(bookmarks);
2155 2161
2156 2162 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2157 2163 if (error != 0) {
2158 2164 dsl_dataset_rele(ds, FTAG);
2159 2165 return (error);
2160 2166 }
2161 2167
2162 2168 /*
2163 2169 * Check if the snap we are rolling back to uses more than
2164 2170 * the refquota.
2165 2171 */
2166 2172 if (ds->ds_quota != 0 &&
2167 2173 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2168 2174 dsl_dataset_rele(ds, FTAG);
2169 2175 return (SET_ERROR(EDQUOT));
2170 2176 }
2171 2177
2172 2178 /*
2173 2179 * When we do the clone swap, we will temporarily use more space
2174 2180 * due to the refreservation (the head will no longer have any
2175 2181 * unique space, so the entire amount of the refreservation will need
2176 2182 * to be free). We will immediately destroy the clone, freeing
2177 2183 * this space, but the freeing happens over many txg's.
2178 2184 */
2179 2185 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2180 2186 dsl_dataset_phys(ds)->ds_unique_bytes);
2181 2187
2182 2188 if (unused_refres_delta > 0 &&
2183 2189 unused_refres_delta >
2184 2190 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2185 2191 dsl_dataset_rele(ds, FTAG);
2186 2192 return (SET_ERROR(ENOSPC));
2187 2193 }
2188 2194
2189 2195 dsl_dataset_rele(ds, FTAG);
2190 2196 return (0);
2191 2197 }
2192 2198
2193 2199 static void
2194 2200 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2195 2201 {
2196 2202 dsl_dataset_rollback_arg_t *ddra = arg;
2197 2203 dsl_pool_t *dp = dmu_tx_pool(tx);
2198 2204 dsl_dataset_t *ds, *clone;
2199 2205 uint64_t cloneobj;
2200 2206 char namebuf[ZFS_MAXNAMELEN];
2201 2207
2202 2208 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2203 2209
2204 2210 dsl_dataset_name(ds->ds_prev, namebuf);
2205 2211 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2206 2212
2207 2213 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2208 2214 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2209 2215
2210 2216 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2211 2217
2212 2218 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2213 2219 dsl_dataset_zero_zil(ds, tx);
2214 2220
2215 2221 dsl_destroy_head_sync_impl(clone, tx);
2216 2222
2217 2223 dsl_dataset_rele(clone, FTAG);
2218 2224 dsl_dataset_rele(ds, FTAG);
2219 2225 }
2220 2226
2221 2227 /*
2222 2228 * Rolls back the given filesystem or volume to the most recent snapshot.
2223 2229 * The name of the most recent snapshot will be returned under key "target"
2224 2230 * in the result nvlist.
2225 2231 *
2226 2232 * If owner != NULL:
2227 2233 * - The existing dataset MUST be owned by the specified owner at entry
2228 2234 * - Upon return, dataset will still be held by the same owner, whether we
2229 2235 * succeed or not.
2230 2236 *
2231 2237 * This mode is required any time the existing filesystem is mounted. See
2232 2238 * notes above zfs_suspend_fs() for further details.
2233 2239 */
2234 2240 int
2235 2241 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2236 2242 {
2237 2243 dsl_dataset_rollback_arg_t ddra;
2238 2244
2239 2245 ddra.ddra_fsname = fsname;
2240 2246 ddra.ddra_owner = owner;
2241 2247 ddra.ddra_result = result;
2242 2248
2243 2249 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2244 2250 dsl_dataset_rollback_sync, &ddra,
2245 2251 1, ZFS_SPACE_CHECK_RESERVED));
2246 2252 }
2247 2253
2248 2254 struct promotenode {
2249 2255 list_node_t link;
2250 2256 dsl_dataset_t *ds;
2251 2257 };
2252 2258
2253 2259 typedef struct dsl_dataset_promote_arg {
2254 2260 const char *ddpa_clonename;
2255 2261 dsl_dataset_t *ddpa_clone;
2256 2262 list_t shared_snaps, origin_snaps, clone_snaps;
2257 2263 dsl_dataset_t *origin_origin; /* origin of the origin */
2258 2264 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2259 2265 char *err_ds;
2260 2266 cred_t *cr;
2261 2267 } dsl_dataset_promote_arg_t;
2262 2268
2263 2269 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2264 2270 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2265 2271 void *tag);
2266 2272 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2267 2273
2268 2274 static int
2269 2275 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2270 2276 {
2271 2277 dsl_dataset_promote_arg_t *ddpa = arg;
2272 2278 dsl_pool_t *dp = dmu_tx_pool(tx);
2273 2279 dsl_dataset_t *hds;
2274 2280 struct promotenode *snap;
2275 2281 dsl_dataset_t *origin_ds;
2276 2282 int err;
2277 2283 uint64_t unused;
2278 2284 uint64_t ss_mv_cnt;
2279 2285 size_t max_snap_len;
2280 2286
2281 2287 err = promote_hold(ddpa, dp, FTAG);
2282 2288 if (err != 0)
2283 2289 return (err);
2284 2290
2285 2291 hds = ddpa->ddpa_clone;
2286 2292 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2287 2293
2288 2294 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2289 2295 promote_rele(ddpa, FTAG);
2290 2296 return (SET_ERROR(EXDEV));
2291 2297 }
2292 2298
2293 2299 /*
2294 2300 * Compute and check the amount of space to transfer. Since this is
2295 2301 * so expensive, don't do the preliminary check.
2296 2302 */
2297 2303 if (!dmu_tx_is_syncing(tx)) {
2298 2304 promote_rele(ddpa, FTAG);
2299 2305 return (0);
2300 2306 }
2301 2307
2302 2308 snap = list_head(&ddpa->shared_snaps);
2303 2309 origin_ds = snap->ds;
2304 2310
2305 2311 /* compute origin's new unique space */
2306 2312 snap = list_tail(&ddpa->clone_snaps);
2307 2313 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2308 2314 origin_ds->ds_object);
2309 2315 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2310 2316 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2311 2317 &ddpa->unique, &unused, &unused);
2312 2318
2313 2319 /*
2314 2320 * Walk the snapshots that we are moving
2315 2321 *
2316 2322 * Compute space to transfer. Consider the incremental changes
2317 2323 * to used by each snapshot:
2318 2324 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2319 2325 * So each snapshot gave birth to:
2320 2326 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2321 2327 * So a sequence would look like:
2322 2328 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2323 2329 * Which simplifies to:
2324 2330 * uN + kN + kN-1 + ... + k1 + k0
2325 2331 * Note however, if we stop before we reach the ORIGIN we get:
2326 2332 * uN + kN + kN-1 + ... + kM - uM-1
2327 2333 */
2328 2334 ss_mv_cnt = 0;
2329 2335 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2330 2336 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2331 2337 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2332 2338 for (snap = list_head(&ddpa->shared_snaps); snap;
2333 2339 snap = list_next(&ddpa->shared_snaps, snap)) {
2334 2340 uint64_t val, dlused, dlcomp, dluncomp;
2335 2341 dsl_dataset_t *ds = snap->ds;
2336 2342
2337 2343 ss_mv_cnt++;
2338 2344
2339 2345 /*
2340 2346 * If there are long holds, we won't be able to evict
2341 2347 * the objset.
2342 2348 */
2343 2349 if (dsl_dataset_long_held(ds)) {
2344 2350 err = SET_ERROR(EBUSY);
2345 2351 goto out;
2346 2352 }
2347 2353
2348 2354 /* Check that the snapshot name does not conflict */
2349 2355 VERIFY0(dsl_dataset_get_snapname(ds));
2350 2356 if (strlen(ds->ds_snapname) >= max_snap_len) {
2351 2357 err = SET_ERROR(ENAMETOOLONG);
2352 2358 goto out;
2353 2359 }
2354 2360 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2355 2361 if (err == 0) {
2356 2362 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2357 2363 err = SET_ERROR(EEXIST);
2358 2364 goto out;
2359 2365 }
2360 2366 if (err != ENOENT)
2361 2367 goto out;
2362 2368
2363 2369 /* The very first snapshot does not have a deadlist */
2364 2370 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2365 2371 continue;
2366 2372
2367 2373 dsl_deadlist_space(&ds->ds_deadlist,
2368 2374 &dlused, &dlcomp, &dluncomp);
2369 2375 ddpa->used += dlused;
2370 2376 ddpa->comp += dlcomp;
2371 2377 ddpa->uncomp += dluncomp;
2372 2378 }
2373 2379
2374 2380 /*
2375 2381 * If we are a clone of a clone then we never reached ORIGIN,
2376 2382 * so we need to subtract out the clone origin's used space.
2377 2383 */
2378 2384 if (ddpa->origin_origin) {
2379 2385 ddpa->used -=
2380 2386 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2381 2387 ddpa->comp -=
2382 2388 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2383 2389 ddpa->uncomp -=
2384 2390 dsl_dataset_phys(ddpa->origin_origin)->
2385 2391 ds_uncompressed_bytes;
2386 2392 }
2387 2393
2388 2394 /* Check that there is enough space and limit headroom here */
2389 2395 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2390 2396 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2391 2397 if (err != 0)
2392 2398 goto out;
2393 2399
2394 2400 /*
2395 2401 * Compute the amounts of space that will be used by snapshots
2396 2402 * after the promotion (for both origin and clone). For each,
2397 2403 * it is the amount of space that will be on all of their
2398 2404 * deadlists (that was not born before their new origin).
2399 2405 */
2400 2406 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2401 2407 uint64_t space;
2402 2408
2403 2409 /*
2404 2410 * Note, typically this will not be a clone of a clone,
2405 2411 * so dd_origin_txg will be < TXG_INITIAL, so
2406 2412 * these snaplist_space() -> dsl_deadlist_space_range()
2407 2413 * calls will be fast because they do not have to
2408 2414 * iterate over all bps.
2409 2415 */
2410 2416 snap = list_head(&ddpa->origin_snaps);
2411 2417 err = snaplist_space(&ddpa->shared_snaps,
2412 2418 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2413 2419 if (err != 0)
2414 2420 goto out;
2415 2421
2416 2422 err = snaplist_space(&ddpa->clone_snaps,
2417 2423 snap->ds->ds_dir->dd_origin_txg, &space);
2418 2424 if (err != 0)
2419 2425 goto out;
2420 2426 ddpa->cloneusedsnap += space;
2421 2427 }
2422 2428 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2423 2429 DD_FLAG_USED_BREAKDOWN) {
2424 2430 err = snaplist_space(&ddpa->origin_snaps,
2425 2431 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2426 2432 &ddpa->originusedsnap);
2427 2433 if (err != 0)
2428 2434 goto out;
2429 2435 }
2430 2436
2431 2437 out:
2432 2438 promote_rele(ddpa, FTAG);
2433 2439 return (err);
2434 2440 }
2435 2441
2436 2442 static void
2437 2443 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2438 2444 {
2439 2445 dsl_dataset_promote_arg_t *ddpa = arg;
2440 2446 dsl_pool_t *dp = dmu_tx_pool(tx);
2441 2447 dsl_dataset_t *hds;
2442 2448 struct promotenode *snap;
2443 2449 dsl_dataset_t *origin_ds;
2444 2450 dsl_dataset_t *origin_head;
2445 2451 dsl_dir_t *dd;
2446 2452 dsl_dir_t *odd = NULL;
2447 2453 uint64_t oldnext_obj;
2448 2454 int64_t delta;
2449 2455
2450 2456 VERIFY0(promote_hold(ddpa, dp, FTAG));
2451 2457 hds = ddpa->ddpa_clone;
2452 2458
2453 2459 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2454 2460
2455 2461 snap = list_head(&ddpa->shared_snaps);
2456 2462 origin_ds = snap->ds;
2457 2463 dd = hds->ds_dir;
2458 2464
2459 2465 snap = list_head(&ddpa->origin_snaps);
2460 2466 origin_head = snap->ds;
2461 2467
2462 2468 /*
2463 2469 * We need to explicitly open odd, since origin_ds's dd will be
2464 2470 * changing.
2465 2471 */
2466 2472 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2467 2473 NULL, FTAG, &odd));
2468 2474
2469 2475 /* change origin's next snap */
2470 2476 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2471 2477 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2472 2478 snap = list_tail(&ddpa->clone_snaps);
2473 2479 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2474 2480 origin_ds->ds_object);
2475 2481 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2476 2482
2477 2483 /* change the origin's next clone */
2478 2484 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2479 2485 dsl_dataset_remove_from_next_clones(origin_ds,
2480 2486 snap->ds->ds_object, tx);
2481 2487 VERIFY0(zap_add_int(dp->dp_meta_objset,
2482 2488 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2483 2489 oldnext_obj, tx));
2484 2490 }
2485 2491
2486 2492 /* change origin */
2487 2493 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2488 2494 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2489 2495 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2490 2496 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2491 2497 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2492 2498 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2493 2499 origin_head->ds_dir->dd_origin_txg =
2494 2500 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2495 2501
2496 2502 /* change dd_clone entries */
2497 2503 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2498 2504 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2499 2505 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2500 2506 VERIFY0(zap_add_int(dp->dp_meta_objset,
2501 2507 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2502 2508 hds->ds_object, tx));
2503 2509
2504 2510 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2505 2511 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2506 2512 origin_head->ds_object, tx));
2507 2513 if (dsl_dir_phys(dd)->dd_clones == 0) {
2508 2514 dsl_dir_phys(dd)->dd_clones =
2509 2515 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2510 2516 DMU_OT_NONE, 0, tx);
2511 2517 }
2512 2518 VERIFY0(zap_add_int(dp->dp_meta_objset,
2513 2519 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2514 2520 }
2515 2521
2516 2522 /* move snapshots to this dir */
2517 2523 for (snap = list_head(&ddpa->shared_snaps); snap;
2518 2524 snap = list_next(&ddpa->shared_snaps, snap)) {
2519 2525 dsl_dataset_t *ds = snap->ds;
2520 2526
2521 2527 /*
2522 2528 * Property callbacks are registered to a particular
2523 2529 * dsl_dir. Since ours is changing, evict the objset
2524 2530 * so that they will be unregistered from the old dsl_dir.
2525 2531 */
2526 2532 if (ds->ds_objset) {
2527 2533 dmu_objset_evict(ds->ds_objset);
2528 2534 ds->ds_objset = NULL;
2529 2535 }
2530 2536
2531 2537 /* move snap name entry */
2532 2538 VERIFY0(dsl_dataset_get_snapname(ds));
2533 2539 VERIFY0(dsl_dataset_snap_remove(origin_head,
2534 2540 ds->ds_snapname, tx, B_TRUE));
2535 2541 VERIFY0(zap_add(dp->dp_meta_objset,
2536 2542 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2537 2543 8, 1, &ds->ds_object, tx));
2538 2544 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2539 2545 DD_FIELD_SNAPSHOT_COUNT, tx);
2540 2546
2541 2547 /* change containing dsl_dir */
2542 2548 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2543 2549 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2544 2550 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2545 2551 ASSERT3P(ds->ds_dir, ==, odd);
2546 2552 dsl_dir_rele(ds->ds_dir, ds);
2547 2553 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2548 2554 NULL, ds, &ds->ds_dir));
2549 2555
2550 2556 /* move any clone references */
2551 2557 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2552 2558 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2553 2559 zap_cursor_t zc;
2554 2560 zap_attribute_t za;
2555 2561
2556 2562 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2557 2563 dsl_dataset_phys(ds)->ds_next_clones_obj);
2558 2564 zap_cursor_retrieve(&zc, &za) == 0;
2559 2565 zap_cursor_advance(&zc)) {
2560 2566 dsl_dataset_t *cnds;
2561 2567 uint64_t o;
2562 2568
2563 2569 if (za.za_first_integer == oldnext_obj) {
2564 2570 /*
2565 2571 * We've already moved the
2566 2572 * origin's reference.
2567 2573 */
2568 2574 continue;
2569 2575 }
2570 2576
2571 2577 VERIFY0(dsl_dataset_hold_obj(dp,
2572 2578 za.za_first_integer, FTAG, &cnds));
2573 2579 o = dsl_dir_phys(cnds->ds_dir)->
2574 2580 dd_head_dataset_obj;
2575 2581
2576 2582 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2577 2583 dsl_dir_phys(odd)->dd_clones, o, tx));
2578 2584 VERIFY0(zap_add_int(dp->dp_meta_objset,
2579 2585 dsl_dir_phys(dd)->dd_clones, o, tx));
2580 2586 dsl_dataset_rele(cnds, FTAG);
2581 2587 }
2582 2588 zap_cursor_fini(&zc);
2583 2589 }
2584 2590
2585 2591 ASSERT(!dsl_prop_hascb(ds));
2586 2592 }
2587 2593
2588 2594 /*
2589 2595 * Change space accounting.
2590 2596 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2591 2597 * both be valid, or both be 0 (resulting in delta == 0). This
2592 2598 * is true for each of {clone,origin} independently.
2593 2599 */
2594 2600
2595 2601 delta = ddpa->cloneusedsnap -
2596 2602 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2597 2603 ASSERT3S(delta, >=, 0);
2598 2604 ASSERT3U(ddpa->used, >=, delta);
2599 2605 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2600 2606 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2601 2607 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2602 2608
2603 2609 delta = ddpa->originusedsnap -
2604 2610 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2605 2611 ASSERT3S(delta, <=, 0);
2606 2612 ASSERT3U(ddpa->used, >=, -delta);
2607 2613 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2608 2614 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2609 2615 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2610 2616
2611 2617 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2612 2618
2613 2619 /* log history record */
2614 2620 spa_history_log_internal_ds(hds, "promote", tx, "");
2615 2621
2616 2622 dsl_dir_rele(odd, FTAG);
2617 2623 promote_rele(ddpa, FTAG);
2618 2624 }
2619 2625
2620 2626 /*
2621 2627 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2622 2628 * (exclusive) and last_obj (inclusive). The list will be in reverse
2623 2629 * order (last_obj will be the list_head()). If first_obj == 0, do all
2624 2630 * snapshots back to this dataset's origin.
2625 2631 */
2626 2632 static int
2627 2633 snaplist_make(dsl_pool_t *dp,
2628 2634 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2629 2635 {
2630 2636 uint64_t obj = last_obj;
2631 2637
2632 2638 list_create(l, sizeof (struct promotenode),
2633 2639 offsetof(struct promotenode, link));
2634 2640
2635 2641 while (obj != first_obj) {
2636 2642 dsl_dataset_t *ds;
2637 2643 struct promotenode *snap;
2638 2644 int err;
2639 2645
2640 2646 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2641 2647 ASSERT(err != ENOENT);
2642 2648 if (err != 0)
2643 2649 return (err);
2644 2650
2645 2651 if (first_obj == 0)
2646 2652 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2647 2653
2648 2654 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2649 2655 snap->ds = ds;
2650 2656 list_insert_tail(l, snap);
2651 2657 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2652 2658 }
2653 2659
2654 2660 return (0);
2655 2661 }
2656 2662
2657 2663 static int
2658 2664 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2659 2665 {
2660 2666 struct promotenode *snap;
2661 2667
2662 2668 *spacep = 0;
2663 2669 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2664 2670 uint64_t used, comp, uncomp;
2665 2671 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2666 2672 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2667 2673 *spacep += used;
2668 2674 }
2669 2675 return (0);
2670 2676 }
2671 2677
2672 2678 static void
2673 2679 snaplist_destroy(list_t *l, void *tag)
2674 2680 {
2675 2681 struct promotenode *snap;
2676 2682
2677 2683 if (l == NULL || !list_link_active(&l->list_head))
2678 2684 return;
2679 2685
2680 2686 while ((snap = list_tail(l)) != NULL) {
2681 2687 list_remove(l, snap);
2682 2688 dsl_dataset_rele(snap->ds, tag);
2683 2689 kmem_free(snap, sizeof (*snap));
2684 2690 }
2685 2691 list_destroy(l);
2686 2692 }
2687 2693
2688 2694 static int
2689 2695 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2690 2696 {
2691 2697 int error;
2692 2698 dsl_dir_t *dd;
2693 2699 struct promotenode *snap;
2694 2700
2695 2701 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2696 2702 &ddpa->ddpa_clone);
2697 2703 if (error != 0)
2698 2704 return (error);
2699 2705 dd = ddpa->ddpa_clone->ds_dir;
2700 2706
2701 2707 if (ddpa->ddpa_clone->ds_is_snapshot ||
2702 2708 !dsl_dir_is_clone(dd)) {
2703 2709 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2704 2710 return (SET_ERROR(EINVAL));
2705 2711 }
2706 2712
2707 2713 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2708 2714 &ddpa->shared_snaps, tag);
2709 2715 if (error != 0)
2710 2716 goto out;
2711 2717
2712 2718 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2713 2719 &ddpa->clone_snaps, tag);
2714 2720 if (error != 0)
2715 2721 goto out;
2716 2722
2717 2723 snap = list_head(&ddpa->shared_snaps);
2718 2724 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2719 2725 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2720 2726 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2721 2727 &ddpa->origin_snaps, tag);
2722 2728 if (error != 0)
2723 2729 goto out;
2724 2730
2725 2731 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2726 2732 error = dsl_dataset_hold_obj(dp,
2727 2733 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2728 2734 tag, &ddpa->origin_origin);
2729 2735 if (error != 0)
2730 2736 goto out;
2731 2737 }
2732 2738 out:
2733 2739 if (error != 0)
2734 2740 promote_rele(ddpa, tag);
2735 2741 return (error);
2736 2742 }
2737 2743
2738 2744 static void
2739 2745 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2740 2746 {
2741 2747 snaplist_destroy(&ddpa->shared_snaps, tag);
2742 2748 snaplist_destroy(&ddpa->clone_snaps, tag);
2743 2749 snaplist_destroy(&ddpa->origin_snaps, tag);
2744 2750 if (ddpa->origin_origin != NULL)
2745 2751 dsl_dataset_rele(ddpa->origin_origin, tag);
2746 2752 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2747 2753 }
2748 2754
2749 2755 /*
2750 2756 * Promote a clone.
2751 2757 *
2752 2758 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2753 2759 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2754 2760 */
2755 2761 int
2756 2762 dsl_dataset_promote(const char *name, char *conflsnap)
2757 2763 {
2758 2764 dsl_dataset_promote_arg_t ddpa = { 0 };
2759 2765 uint64_t numsnaps;
2760 2766 int error;
2761 2767 objset_t *os;
2762 2768
2763 2769 /*
2764 2770 * We will modify space proportional to the number of
2765 2771 * snapshots. Compute numsnaps.
2766 2772 */
2767 2773 error = dmu_objset_hold(name, FTAG, &os);
2768 2774 if (error != 0)
2769 2775 return (error);
2770 2776 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2771 2777 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2772 2778 &numsnaps);
2773 2779 dmu_objset_rele(os, FTAG);
2774 2780 if (error != 0)
2775 2781 return (error);
2776 2782
2777 2783 ddpa.ddpa_clonename = name;
2778 2784 ddpa.err_ds = conflsnap;
2779 2785 ddpa.cr = CRED();
2780 2786
2781 2787 return (dsl_sync_task(name, dsl_dataset_promote_check,
2782 2788 dsl_dataset_promote_sync, &ddpa,
2783 2789 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2784 2790 }
2785 2791
2786 2792 int
2787 2793 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2788 2794 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2789 2795 {
2790 2796 int64_t unused_refres_delta;
2791 2797
2792 2798 /* they should both be heads */
2793 2799 if (clone->ds_is_snapshot ||
2794 2800 origin_head->ds_is_snapshot)
2795 2801 return (SET_ERROR(EINVAL));
2796 2802
2797 2803 /* if we are not forcing, the branch point should be just before them */
2798 2804 if (!force && clone->ds_prev != origin_head->ds_prev)
2799 2805 return (SET_ERROR(EINVAL));
2800 2806
2801 2807 /* clone should be the clone (unless they are unrelated) */
2802 2808 if (clone->ds_prev != NULL &&
2803 2809 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2804 2810 origin_head->ds_dir != clone->ds_prev->ds_dir)
2805 2811 return (SET_ERROR(EINVAL));
2806 2812
2807 2813 /* the clone should be a child of the origin */
2808 2814 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2809 2815 return (SET_ERROR(EINVAL));
2810 2816
2811 2817 /* origin_head shouldn't be modified unless 'force' */
2812 2818 if (!force &&
2813 2819 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2814 2820 return (SET_ERROR(ETXTBSY));
2815 2821
2816 2822 /* origin_head should have no long holds (e.g. is not mounted) */
2817 2823 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2818 2824 return (SET_ERROR(EBUSY));
2819 2825
2820 2826 /* check amount of any unconsumed refreservation */
2821 2827 unused_refres_delta =
2822 2828 (int64_t)MIN(origin_head->ds_reserved,
2823 2829 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2824 2830 (int64_t)MIN(origin_head->ds_reserved,
2825 2831 dsl_dataset_phys(clone)->ds_unique_bytes);
2826 2832
2827 2833 if (unused_refres_delta > 0 &&
2828 2834 unused_refres_delta >
2829 2835 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2830 2836 return (SET_ERROR(ENOSPC));
2831 2837
2832 2838 /* clone can't be over the head's refquota */
2833 2839 if (origin_head->ds_quota != 0 &&
2834 2840 dsl_dataset_phys(clone)->ds_referenced_bytes >
2835 2841 origin_head->ds_quota)
2836 2842 return (SET_ERROR(EDQUOT));
2837 2843
2838 2844 return (0);
2839 2845 }
2840 2846
2841 2847 void
2842 2848 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2843 2849 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2844 2850 {
2845 2851 dsl_pool_t *dp = dmu_tx_pool(tx);
2846 2852 int64_t unused_refres_delta;
2847 2853
2848 2854 ASSERT(clone->ds_reserved == 0);
2849 2855 ASSERT(origin_head->ds_quota == 0 ||
2850 2856 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2851 2857 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2852 2858
2853 2859 /*
2854 2860 * Swap per-dataset feature flags.
2855 2861 */
2856 2862 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2857 2863 if (!(spa_feature_table[f].fi_flags &
2858 2864 ZFEATURE_FLAG_PER_DATASET)) {
2859 2865 ASSERT(!clone->ds_feature_inuse[f]);
2860 2866 ASSERT(!origin_head->ds_feature_inuse[f]);
2861 2867 continue;
2862 2868 }
2863 2869
2864 2870 boolean_t clone_inuse = clone->ds_feature_inuse[f];
2865 2871 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2866 2872
2867 2873 if (clone_inuse) {
2868 2874 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2869 2875 clone->ds_feature_inuse[f] = B_FALSE;
2870 2876 }
2871 2877 if (origin_head_inuse) {
2872 2878 dsl_dataset_deactivate_feature(origin_head->ds_object,
2873 2879 f, tx);
2874 2880 origin_head->ds_feature_inuse[f] = B_FALSE;
2875 2881 }
2876 2882 if (clone_inuse) {
2877 2883 dsl_dataset_activate_feature(origin_head->ds_object,
2878 2884 f, tx);
2879 2885 origin_head->ds_feature_inuse[f] = B_TRUE;
2880 2886 }
2881 2887 if (origin_head_inuse) {
2882 2888 dsl_dataset_activate_feature(clone->ds_object, f, tx);
2883 2889 clone->ds_feature_inuse[f] = B_TRUE;
2884 2890 }
2885 2891 }
2886 2892
2887 2893 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2888 2894 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2889 2895
2890 2896 if (clone->ds_objset != NULL) {
2891 2897 dmu_objset_evict(clone->ds_objset);
2892 2898 clone->ds_objset = NULL;
2893 2899 }
2894 2900
2895 2901 if (origin_head->ds_objset != NULL) {
2896 2902 dmu_objset_evict(origin_head->ds_objset);
2897 2903 origin_head->ds_objset = NULL;
2898 2904 }
2899 2905
2900 2906 unused_refres_delta =
2901 2907 (int64_t)MIN(origin_head->ds_reserved,
2902 2908 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2903 2909 (int64_t)MIN(origin_head->ds_reserved,
2904 2910 dsl_dataset_phys(clone)->ds_unique_bytes);
2905 2911
2906 2912 /*
2907 2913 * Reset origin's unique bytes, if it exists.
2908 2914 */
2909 2915 if (clone->ds_prev) {
2910 2916 dsl_dataset_t *origin = clone->ds_prev;
2911 2917 uint64_t comp, uncomp;
2912 2918
2913 2919 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2914 2920 dsl_deadlist_space_range(&clone->ds_deadlist,
2915 2921 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2916 2922 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2917 2923 }
2918 2924
2919 2925 /* swap blkptrs */
2920 2926 {
2921 2927 blkptr_t tmp;
2922 2928 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2923 2929 dsl_dataset_phys(origin_head)->ds_bp =
2924 2930 dsl_dataset_phys(clone)->ds_bp;
2925 2931 dsl_dataset_phys(clone)->ds_bp = tmp;
2926 2932 }
2927 2933
2928 2934 /* set dd_*_bytes */
2929 2935 {
2930 2936 int64_t dused, dcomp, duncomp;
2931 2937 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2932 2938 uint64_t odl_used, odl_comp, odl_uncomp;
2933 2939
2934 2940 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2935 2941 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2936 2942
2937 2943 dsl_deadlist_space(&clone->ds_deadlist,
2938 2944 &cdl_used, &cdl_comp, &cdl_uncomp);
2939 2945 dsl_deadlist_space(&origin_head->ds_deadlist,
2940 2946 &odl_used, &odl_comp, &odl_uncomp);
2941 2947
2942 2948 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2943 2949 cdl_used -
2944 2950 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2945 2951 odl_used);
2946 2952 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2947 2953 cdl_comp -
2948 2954 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2949 2955 odl_comp);
2950 2956 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2951 2957 cdl_uncomp -
2952 2958 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2953 2959 odl_uncomp);
2954 2960
2955 2961 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2956 2962 dused, dcomp, duncomp, tx);
2957 2963 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2958 2964 -dused, -dcomp, -duncomp, tx);
2959 2965
2960 2966 /*
2961 2967 * The difference in the space used by snapshots is the
2962 2968 * difference in snapshot space due to the head's
2963 2969 * deadlist (since that's the only thing that's
2964 2970 * changing that affects the snapused).
2965 2971 */
2966 2972 dsl_deadlist_space_range(&clone->ds_deadlist,
2967 2973 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2968 2974 &cdl_used, &cdl_comp, &cdl_uncomp);
2969 2975 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2970 2976 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2971 2977 &odl_used, &odl_comp, &odl_uncomp);
2972 2978 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2973 2979 DD_USED_HEAD, DD_USED_SNAP, tx);
2974 2980 }
2975 2981
2976 2982 /* swap ds_*_bytes */
2977 2983 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2978 2984 dsl_dataset_phys(clone)->ds_referenced_bytes);
2979 2985 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2980 2986 dsl_dataset_phys(clone)->ds_compressed_bytes);
2981 2987 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2982 2988 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2983 2989 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2984 2990 dsl_dataset_phys(clone)->ds_unique_bytes);
2985 2991
2986 2992 /* apply any parent delta for change in unconsumed refreservation */
2987 2993 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2988 2994 unused_refres_delta, 0, 0, tx);
2989 2995
2990 2996 /*
2991 2997 * Swap deadlists.
2992 2998 */
2993 2999 dsl_deadlist_close(&clone->ds_deadlist);
2994 3000 dsl_deadlist_close(&origin_head->ds_deadlist);
2995 3001 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2996 3002 dsl_dataset_phys(clone)->ds_deadlist_obj);
2997 3003 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2998 3004 dsl_dataset_phys(clone)->ds_deadlist_obj);
2999 3005 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3000 3006 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3001 3007
3002 3008 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3003 3009
3004 3010 spa_history_log_internal_ds(clone, "clone swap", tx,
3005 3011 "parent=%s", origin_head->ds_dir->dd_myname);
3006 3012 }
3007 3013
3008 3014 /*
3009 3015 * Given a pool name and a dataset object number in that pool,
3010 3016 * return the name of that dataset.
3011 3017 */
3012 3018 int
3013 3019 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3014 3020 {
3015 3021 dsl_pool_t *dp;
3016 3022 dsl_dataset_t *ds;
3017 3023 int error;
3018 3024
3019 3025 error = dsl_pool_hold(pname, FTAG, &dp);
3020 3026 if (error != 0)
3021 3027 return (error);
3022 3028
3023 3029 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3024 3030 if (error == 0) {
3025 3031 dsl_dataset_name(ds, buf);
3026 3032 dsl_dataset_rele(ds, FTAG);
3027 3033 }
3028 3034 dsl_pool_rele(dp, FTAG);
3029 3035
3030 3036 return (error);
3031 3037 }
3032 3038
3033 3039 int
3034 3040 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3035 3041 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3036 3042 {
3037 3043 int error = 0;
3038 3044
3039 3045 ASSERT3S(asize, >, 0);
3040 3046
3041 3047 /*
3042 3048 * *ref_rsrv is the portion of asize that will come from any
3043 3049 * unconsumed refreservation space.
3044 3050 */
3045 3051 *ref_rsrv = 0;
3046 3052
3047 3053 mutex_enter(&ds->ds_lock);
3048 3054 /*
3049 3055 * Make a space adjustment for reserved bytes.
3050 3056 */
3051 3057 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3052 3058 ASSERT3U(*used, >=,
3053 3059 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3054 3060 *used -=
3055 3061 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3056 3062 *ref_rsrv =
3057 3063 asize - MIN(asize, parent_delta(ds, asize + inflight));
3058 3064 }
3059 3065
3060 3066 if (!check_quota || ds->ds_quota == 0) {
3061 3067 mutex_exit(&ds->ds_lock);
3062 3068 return (0);
3063 3069 }
3064 3070 /*
3065 3071 * If they are requesting more space, and our current estimate
3066 3072 * is over quota, they get to try again unless the actual
3067 3073 * on-disk is over quota and there are no pending changes (which
3068 3074 * may free up space for us).
3069 3075 */
3070 3076 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3071 3077 ds->ds_quota) {
3072 3078 if (inflight > 0 ||
3073 3079 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3074 3080 error = SET_ERROR(ERESTART);
3075 3081 else
3076 3082 error = SET_ERROR(EDQUOT);
3077 3083 }
3078 3084 mutex_exit(&ds->ds_lock);
3079 3085
3080 3086 return (error);
3081 3087 }
3082 3088
3083 3089 typedef struct dsl_dataset_set_qr_arg {
3084 3090 const char *ddsqra_name;
3085 3091 zprop_source_t ddsqra_source;
3086 3092 uint64_t ddsqra_value;
3087 3093 } dsl_dataset_set_qr_arg_t;
3088 3094
3089 3095
3090 3096 /* ARGSUSED */
3091 3097 static int
3092 3098 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3093 3099 {
3094 3100 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3095 3101 dsl_pool_t *dp = dmu_tx_pool(tx);
3096 3102 dsl_dataset_t *ds;
3097 3103 int error;
3098 3104 uint64_t newval;
3099 3105
3100 3106 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3101 3107 return (SET_ERROR(ENOTSUP));
3102 3108
3103 3109 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3104 3110 if (error != 0)
3105 3111 return (error);
3106 3112
3107 3113 if (ds->ds_is_snapshot) {
3108 3114 dsl_dataset_rele(ds, FTAG);
3109 3115 return (SET_ERROR(EINVAL));
3110 3116 }
3111 3117
3112 3118 error = dsl_prop_predict(ds->ds_dir,
3113 3119 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3114 3120 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3115 3121 if (error != 0) {
3116 3122 dsl_dataset_rele(ds, FTAG);
3117 3123 return (error);
3118 3124 }
3119 3125
3120 3126 if (newval == 0) {
3121 3127 dsl_dataset_rele(ds, FTAG);
3122 3128 return (0);
3123 3129 }
3124 3130
3125 3131 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3126 3132 newval < ds->ds_reserved) {
3127 3133 dsl_dataset_rele(ds, FTAG);
3128 3134 return (SET_ERROR(ENOSPC));
3129 3135 }
3130 3136
3131 3137 dsl_dataset_rele(ds, FTAG);
3132 3138 return (0);
3133 3139 }
3134 3140
3135 3141 static void
3136 3142 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3137 3143 {
3138 3144 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3139 3145 dsl_pool_t *dp = dmu_tx_pool(tx);
3140 3146 dsl_dataset_t *ds;
3141 3147 uint64_t newval;
3142 3148
3143 3149 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3144 3150
3145 3151 dsl_prop_set_sync_impl(ds,
3146 3152 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3147 3153 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3148 3154 &ddsqra->ddsqra_value, tx);
3149 3155
3150 3156 VERIFY0(dsl_prop_get_int_ds(ds,
3151 3157 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3152 3158
3153 3159 if (ds->ds_quota != newval) {
3154 3160 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3155 3161 ds->ds_quota = newval;
3156 3162 }
3157 3163 dsl_dataset_rele(ds, FTAG);
3158 3164 }
3159 3165
3160 3166 int
3161 3167 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3162 3168 uint64_t refquota)
3163 3169 {
3164 3170 dsl_dataset_set_qr_arg_t ddsqra;
3165 3171
3166 3172 ddsqra.ddsqra_name = dsname;
3167 3173 ddsqra.ddsqra_source = source;
3168 3174 ddsqra.ddsqra_value = refquota;
3169 3175
3170 3176 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3171 3177 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3172 3178 }
3173 3179
3174 3180 static int
3175 3181 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3176 3182 {
3177 3183 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3178 3184 dsl_pool_t *dp = dmu_tx_pool(tx);
3179 3185 dsl_dataset_t *ds;
3180 3186 int error;
3181 3187 uint64_t newval, unique;
3182 3188
3183 3189 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3184 3190 return (SET_ERROR(ENOTSUP));
3185 3191
3186 3192 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3187 3193 if (error != 0)
3188 3194 return (error);
3189 3195
3190 3196 if (ds->ds_is_snapshot) {
3191 3197 dsl_dataset_rele(ds, FTAG);
3192 3198 return (SET_ERROR(EINVAL));
3193 3199 }
3194 3200
3195 3201 error = dsl_prop_predict(ds->ds_dir,
3196 3202 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3197 3203 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3198 3204 if (error != 0) {
3199 3205 dsl_dataset_rele(ds, FTAG);
3200 3206 return (error);
3201 3207 }
3202 3208
3203 3209 /*
3204 3210 * If we are doing the preliminary check in open context, the
3205 3211 * space estimates may be inaccurate.
3206 3212 */
3207 3213 if (!dmu_tx_is_syncing(tx)) {
3208 3214 dsl_dataset_rele(ds, FTAG);
3209 3215 return (0);
3210 3216 }
3211 3217
3212 3218 mutex_enter(&ds->ds_lock);
3213 3219 if (!DS_UNIQUE_IS_ACCURATE(ds))
3214 3220 dsl_dataset_recalc_head_uniq(ds);
3215 3221 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3216 3222 mutex_exit(&ds->ds_lock);
3217 3223
3218 3224 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3219 3225 uint64_t delta = MAX(unique, newval) -
3220 3226 MAX(unique, ds->ds_reserved);
3221 3227
3222 3228 if (delta >
3223 3229 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3224 3230 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3225 3231 dsl_dataset_rele(ds, FTAG);
3226 3232 return (SET_ERROR(ENOSPC));
3227 3233 }
3228 3234 }
3229 3235
3230 3236 dsl_dataset_rele(ds, FTAG);
3231 3237 return (0);
3232 3238 }
3233 3239
3234 3240 void
3235 3241 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3236 3242 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3237 3243 {
3238 3244 uint64_t newval;
3239 3245 uint64_t unique;
3240 3246 int64_t delta;
3241 3247
3242 3248 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3243 3249 source, sizeof (value), 1, &value, tx);
3244 3250
3245 3251 VERIFY0(dsl_prop_get_int_ds(ds,
3246 3252 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3247 3253
3248 3254 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3249 3255 mutex_enter(&ds->ds_dir->dd_lock);
3250 3256 mutex_enter(&ds->ds_lock);
3251 3257 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3252 3258 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3253 3259 delta = MAX(0, (int64_t)(newval - unique)) -
3254 3260 MAX(0, (int64_t)(ds->ds_reserved - unique));
3255 3261 ds->ds_reserved = newval;
3256 3262 mutex_exit(&ds->ds_lock);
3257 3263
3258 3264 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3259 3265 mutex_exit(&ds->ds_dir->dd_lock);
3260 3266 }
3261 3267
3262 3268 static void
3263 3269 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3264 3270 {
3265 3271 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3266 3272 dsl_pool_t *dp = dmu_tx_pool(tx);
3267 3273 dsl_dataset_t *ds;
3268 3274
3269 3275 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3270 3276 dsl_dataset_set_refreservation_sync_impl(ds,
3271 3277 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3272 3278 dsl_dataset_rele(ds, FTAG);
3273 3279 }
3274 3280
3275 3281 int
3276 3282 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3277 3283 uint64_t refreservation)
3278 3284 {
3279 3285 dsl_dataset_set_qr_arg_t ddsqra;
3280 3286
3281 3287 ddsqra.ddsqra_name = dsname;
3282 3288 ddsqra.ddsqra_source = source;
3283 3289 ddsqra.ddsqra_value = refreservation;
3284 3290
3285 3291 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3286 3292 dsl_dataset_set_refreservation_sync, &ddsqra,
3287 3293 0, ZFS_SPACE_CHECK_NONE));
3288 3294 }
3289 3295
3290 3296 /*
3291 3297 * Return (in *usedp) the amount of space written in new that is not
3292 3298 * present in oldsnap. New may be a snapshot or the head. Old must be
3293 3299 * a snapshot before new, in new's filesystem (or its origin). If not then
3294 3300 * fail and return EINVAL.
3295 3301 *
3296 3302 * The written space is calculated by considering two components: First, we
3297 3303 * ignore any freed space, and calculate the written as new's used space
3298 3304 * minus old's used space. Next, we add in the amount of space that was freed
3299 3305 * between the two snapshots, thus reducing new's used space relative to old's.
3300 3306 * Specifically, this is the space that was born before old->ds_creation_txg,
3301 3307 * and freed before new (ie. on new's deadlist or a previous deadlist).
3302 3308 *
3303 3309 * space freed [---------------------]
3304 3310 * snapshots ---O-------O--------O-------O------
3305 3311 * oldsnap new
3306 3312 */
3307 3313 int
3308 3314 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3309 3315 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3310 3316 {
3311 3317 int err = 0;
3312 3318 uint64_t snapobj;
3313 3319 dsl_pool_t *dp = new->ds_dir->dd_pool;
3314 3320
3315 3321 ASSERT(dsl_pool_config_held(dp));
3316 3322
3317 3323 *usedp = 0;
3318 3324 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3319 3325 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3320 3326
3321 3327 *compp = 0;
3322 3328 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3323 3329 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3324 3330
3325 3331 *uncompp = 0;
3326 3332 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3327 3333 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3328 3334
3329 3335 snapobj = new->ds_object;
3330 3336 while (snapobj != oldsnap->ds_object) {
3331 3337 dsl_dataset_t *snap;
3332 3338 uint64_t used, comp, uncomp;
3333 3339
3334 3340 if (snapobj == new->ds_object) {
3335 3341 snap = new;
3336 3342 } else {
3337 3343 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3338 3344 if (err != 0)
3339 3345 break;
3340 3346 }
3341 3347
3342 3348 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3343 3349 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3344 3350 /*
3345 3351 * The blocks in the deadlist can not be born after
3346 3352 * ds_prev_snap_txg, so get the whole deadlist space,
3347 3353 * which is more efficient (especially for old-format
3348 3354 * deadlists). Unfortunately the deadlist code
3349 3355 * doesn't have enough information to make this
3350 3356 * optimization itself.
3351 3357 */
3352 3358 dsl_deadlist_space(&snap->ds_deadlist,
3353 3359 &used, &comp, &uncomp);
3354 3360 } else {
3355 3361 dsl_deadlist_space_range(&snap->ds_deadlist,
3356 3362 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3357 3363 &used, &comp, &uncomp);
3358 3364 }
3359 3365 *usedp += used;
3360 3366 *compp += comp;
3361 3367 *uncompp += uncomp;
3362 3368
3363 3369 /*
3364 3370 * If we get to the beginning of the chain of snapshots
3365 3371 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3366 3372 * was not a snapshot of/before new.
3367 3373 */
3368 3374 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3369 3375 if (snap != new)
3370 3376 dsl_dataset_rele(snap, FTAG);
3371 3377 if (snapobj == 0) {
3372 3378 err = SET_ERROR(EINVAL);
3373 3379 break;
3374 3380 }
3375 3381
3376 3382 }
3377 3383 return (err);
3378 3384 }
3379 3385
3380 3386 /*
3381 3387 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3382 3388 * lastsnap, and all snapshots in between are deleted.
3383 3389 *
3384 3390 * blocks that would be freed [---------------------------]
3385 3391 * snapshots ---O-------O--------O-------O--------O
3386 3392 * firstsnap lastsnap
3387 3393 *
3388 3394 * This is the set of blocks that were born after the snap before firstsnap,
3389 3395 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3390 3396 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3391 3397 * We calculate this by iterating over the relevant deadlists (from the snap
3392 3398 * after lastsnap, backward to the snap after firstsnap), summing up the
3393 3399 * space on the deadlist that was born after the snap before firstsnap.
3394 3400 */
3395 3401 int
3396 3402 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3397 3403 dsl_dataset_t *lastsnap,
3398 3404 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3399 3405 {
3400 3406 int err = 0;
3401 3407 uint64_t snapobj;
3402 3408 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3403 3409
3404 3410 ASSERT(firstsnap->ds_is_snapshot);
3405 3411 ASSERT(lastsnap->ds_is_snapshot);
3406 3412
3407 3413 /*
3408 3414 * Check that the snapshots are in the same dsl_dir, and firstsnap
3409 3415 * is before lastsnap.
3410 3416 */
3411 3417 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3412 3418 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3413 3419 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3414 3420 return (SET_ERROR(EINVAL));
3415 3421
3416 3422 *usedp = *compp = *uncompp = 0;
3417 3423
3418 3424 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3419 3425 while (snapobj != firstsnap->ds_object) {
3420 3426 dsl_dataset_t *ds;
3421 3427 uint64_t used, comp, uncomp;
3422 3428
3423 3429 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3424 3430 if (err != 0)
3425 3431 break;
3426 3432
3427 3433 dsl_deadlist_space_range(&ds->ds_deadlist,
3428 3434 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3429 3435 &used, &comp, &uncomp);
3430 3436 *usedp += used;
3431 3437 *compp += comp;
3432 3438 *uncompp += uncomp;
3433 3439
3434 3440 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3435 3441 ASSERT3U(snapobj, !=, 0);
3436 3442 dsl_dataset_rele(ds, FTAG);
3437 3443 }
3438 3444 return (err);
3439 3445 }
3440 3446
3441 3447 /*
3442 3448 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3443 3449 * For example, they could both be snapshots of the same filesystem, and
3444 3450 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3445 3451 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3446 3452 * filesystem. Or 'earlier' could be the origin's origin.
3447 3453 *
3448 3454 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3449 3455 */
3450 3456 boolean_t
3451 3457 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3452 3458 uint64_t earlier_txg)
3453 3459 {
3454 3460 dsl_pool_t *dp = later->ds_dir->dd_pool;
3455 3461 int error;
3456 3462 boolean_t ret;
3457 3463
3458 3464 ASSERT(dsl_pool_config_held(dp));
3459 3465 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3460 3466
3461 3467 if (earlier_txg == 0)
3462 3468 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3463 3469
3464 3470 if (later->ds_is_snapshot &&
3465 3471 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3466 3472 return (B_FALSE);
3467 3473
3468 3474 if (later->ds_dir == earlier->ds_dir)
3469 3475 return (B_TRUE);
3470 3476 if (!dsl_dir_is_clone(later->ds_dir))
3471 3477 return (B_FALSE);
3472 3478
3473 3479 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3474 3480 return (B_TRUE);
3475 3481 dsl_dataset_t *origin;
3476 3482 error = dsl_dataset_hold_obj(dp,
3477 3483 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3478 3484 if (error != 0)
3479 3485 return (B_FALSE);
3480 3486 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3481 3487 dsl_dataset_rele(origin, FTAG);
3482 3488 return (ret);
3483 3489 }
3484 3490
3485 3491 void
3486 3492 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3487 3493 {
3488 3494 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3489 3495 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3490 3496 }
3491 3497
3492 3498 boolean_t
3493 3499 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3494 3500 {
3495 3501 dmu_object_info_t doi;
3496 3502
3497 3503 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3498 3504 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3499 3505 }
3500 3506
3501 3507 boolean_t
3502 3508 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3503 3509 {
3504 3510 return (dsl_dataset_is_zapified(ds) &&
3505 3511 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3506 3512 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3507 3513 }
|
↓ open down ↓ |
2824 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX