Print this page
6562 Refquota on receive doesn't account for overage
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Toomas Soome <tsoome@me.com>
| 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]
|
↓ open down ↓ |
17 lines elided |
↑ open up ↑ |
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 * Copyright (c) 2014 Integros [integros.com]
28 + * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
28 29 */
29 30
30 31 #include <sys/dmu_objset.h>
31 32 #include <sys/dsl_dataset.h>
32 33 #include <sys/dsl_dir.h>
33 34 #include <sys/dsl_prop.h>
34 35 #include <sys/dsl_synctask.h>
35 36 #include <sys/dmu_traverse.h>
36 37 #include <sys/dmu_impl.h>
37 38 #include <sys/dmu_tx.h>
38 39 #include <sys/arc.h>
39 40 #include <sys/zio.h>
40 41 #include <sys/zap.h>
41 42 #include <sys/zfeature.h>
42 43 #include <sys/unique.h>
43 44 #include <sys/zfs_context.h>
44 45 #include <sys/zfs_ioctl.h>
45 46 #include <sys/spa.h>
46 47 #include <sys/zfs_znode.h>
47 48 #include <sys/zfs_onexit.h>
48 49 #include <sys/zvol.h>
49 50 #include <sys/dsl_scan.h>
50 51 #include <sys/dsl_deadlist.h>
51 52 #include <sys/dsl_destroy.h>
52 53 #include <sys/dsl_userhold.h>
53 54 #include <sys/dsl_bookmark.h>
54 55 #include <sys/dmu_send.h>
55 56 #include <sys/zio_checksum.h>
56 57 #include <sys/zio_compress.h>
57 58 #include <zfs_fletcher.h>
58 59
59 60 /*
60 61 * The SPA supports block sizes up to 16MB. However, very large blocks
61 62 * can have an impact on i/o latency (e.g. tying up a spinning disk for
62 63 * ~300ms), and also potentially on the memory allocator. Therefore,
63 64 * we do not allow the recordsize to be set larger than zfs_max_recordsize
64 65 * (default 1MB). Larger blocks can be created by changing this tunable,
65 66 * and pools with larger blocks can always be imported and used, regardless
66 67 * of this setting.
67 68 */
68 69 int zfs_max_recordsize = 1 * 1024 * 1024;
69 70
70 71 #define SWITCH64(x, y) \
|
↓ open down ↓ |
33 lines elided |
↑ open up ↑ |
71 72 { \
72 73 uint64_t __tmp = (x); \
73 74 (x) = (y); \
74 75 (y) = __tmp; \
75 76 }
76 77
77 78 #define DS_REF_MAX (1ULL << 62)
78 79
79 80 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
80 81
82 +extern int spa_asize_inflation;
83 +
81 84 /*
82 85 * Figure out how much of this delta should be propogated to the dsl_dir
83 86 * layer. If there's a refreservation, that space has already been
84 87 * partially accounted for in our ancestors.
85 88 */
86 89 static int64_t
87 90 parent_delta(dsl_dataset_t *ds, int64_t delta)
88 91 {
89 92 dsl_dataset_phys_t *ds_phys;
90 93 uint64_t old_bytes, new_bytes;
91 94
92 95 if (ds->ds_reserved == 0)
93 96 return (delta);
94 97
95 98 ds_phys = dsl_dataset_phys(ds);
96 99 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
97 100 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
98 101
99 102 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
100 103 return (new_bytes - old_bytes);
101 104 }
102 105
103 106 void
104 107 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
105 108 {
106 109 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
107 110 int compressed = BP_GET_PSIZE(bp);
108 111 int uncompressed = BP_GET_UCSIZE(bp);
109 112 int64_t delta;
110 113
111 114 dprintf_bp(bp, "ds=%p", ds);
112 115
113 116 ASSERT(dmu_tx_is_syncing(tx));
114 117 /* It could have been compressed away to nothing */
115 118 if (BP_IS_HOLE(bp))
116 119 return;
117 120 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
118 121 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
119 122 if (ds == NULL) {
120 123 dsl_pool_mos_diduse_space(tx->tx_pool,
121 124 used, compressed, uncompressed);
122 125 return;
123 126 }
124 127
125 128 dmu_buf_will_dirty(ds->ds_dbuf, tx);
126 129 mutex_enter(&ds->ds_lock);
127 130 delta = parent_delta(ds, used);
128 131 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
129 132 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
130 133 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
131 134 dsl_dataset_phys(ds)->ds_unique_bytes += used;
132 135
133 136 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
134 137 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
135 138 B_TRUE;
136 139 }
137 140
138 141 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
139 142 if (f != SPA_FEATURE_NONE)
140 143 ds->ds_feature_activation_needed[f] = B_TRUE;
141 144
142 145 mutex_exit(&ds->ds_lock);
143 146 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
144 147 compressed, uncompressed, tx);
145 148 dsl_dir_transfer_space(ds->ds_dir, used - delta,
146 149 DD_USED_REFRSRV, DD_USED_HEAD, tx);
147 150 }
148 151
149 152 int
150 153 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
151 154 boolean_t async)
152 155 {
153 156 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
154 157 int compressed = BP_GET_PSIZE(bp);
155 158 int uncompressed = BP_GET_UCSIZE(bp);
156 159
157 160 if (BP_IS_HOLE(bp))
158 161 return (0);
159 162
160 163 ASSERT(dmu_tx_is_syncing(tx));
161 164 ASSERT(bp->blk_birth <= tx->tx_txg);
162 165
163 166 if (ds == NULL) {
164 167 dsl_free(tx->tx_pool, tx->tx_txg, bp);
165 168 dsl_pool_mos_diduse_space(tx->tx_pool,
166 169 -used, -compressed, -uncompressed);
167 170 return (used);
168 171 }
169 172 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
170 173
171 174 ASSERT(!ds->ds_is_snapshot);
172 175 dmu_buf_will_dirty(ds->ds_dbuf, tx);
173 176
174 177 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
175 178 int64_t delta;
176 179
177 180 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
178 181 dsl_free(tx->tx_pool, tx->tx_txg, bp);
179 182
180 183 mutex_enter(&ds->ds_lock);
181 184 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
182 185 !DS_UNIQUE_IS_ACCURATE(ds));
183 186 delta = parent_delta(ds, -used);
184 187 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
185 188 mutex_exit(&ds->ds_lock);
186 189 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
187 190 delta, -compressed, -uncompressed, tx);
188 191 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
189 192 DD_USED_REFRSRV, DD_USED_HEAD, tx);
190 193 } else {
191 194 dprintf_bp(bp, "putting on dead list: %s", "");
192 195 if (async) {
193 196 /*
194 197 * We are here as part of zio's write done callback,
195 198 * which means we're a zio interrupt thread. We can't
196 199 * call dsl_deadlist_insert() now because it may block
197 200 * waiting for I/O. Instead, put bp on the deferred
198 201 * queue and let dsl_pool_sync() finish the job.
199 202 */
200 203 bplist_append(&ds->ds_pending_deadlist, bp);
201 204 } else {
202 205 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
203 206 }
204 207 ASSERT3U(ds->ds_prev->ds_object, ==,
205 208 dsl_dataset_phys(ds)->ds_prev_snap_obj);
206 209 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
207 210 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
208 211 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
209 212 ds->ds_object && bp->blk_birth >
210 213 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
211 214 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
212 215 mutex_enter(&ds->ds_prev->ds_lock);
213 216 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
214 217 mutex_exit(&ds->ds_prev->ds_lock);
215 218 }
216 219 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
217 220 dsl_dir_transfer_space(ds->ds_dir, used,
218 221 DD_USED_HEAD, DD_USED_SNAP, tx);
219 222 }
220 223 }
221 224 mutex_enter(&ds->ds_lock);
222 225 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
223 226 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
224 227 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
225 228 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
226 229 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
227 230 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
228 231 mutex_exit(&ds->ds_lock);
229 232
230 233 return (used);
231 234 }
232 235
233 236 uint64_t
234 237 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
235 238 {
236 239 uint64_t trysnap = 0;
237 240
238 241 if (ds == NULL)
239 242 return (0);
240 243 /*
241 244 * The snapshot creation could fail, but that would cause an
242 245 * incorrect FALSE return, which would only result in an
243 246 * overestimation of the amount of space that an operation would
244 247 * consume, which is OK.
245 248 *
246 249 * There's also a small window where we could miss a pending
247 250 * snapshot, because we could set the sync task in the quiescing
248 251 * phase. So this should only be used as a guess.
249 252 */
250 253 if (ds->ds_trysnap_txg >
251 254 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
252 255 trysnap = ds->ds_trysnap_txg;
253 256 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
254 257 }
255 258
256 259 boolean_t
257 260 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
258 261 uint64_t blk_birth)
259 262 {
260 263 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
261 264 (bp != NULL && BP_IS_HOLE(bp)))
262 265 return (B_FALSE);
263 266
264 267 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
265 268
266 269 return (B_TRUE);
267 270 }
268 271
269 272 static void
270 273 dsl_dataset_evict(void *dbu)
271 274 {
272 275 dsl_dataset_t *ds = dbu;
273 276
274 277 ASSERT(ds->ds_owner == NULL);
275 278
276 279 ds->ds_dbuf = NULL;
277 280
278 281 unique_remove(ds->ds_fsid_guid);
279 282
280 283 if (ds->ds_objset != NULL)
281 284 dmu_objset_evict(ds->ds_objset);
282 285
283 286 if (ds->ds_prev) {
284 287 dsl_dataset_rele(ds->ds_prev, ds);
285 288 ds->ds_prev = NULL;
286 289 }
287 290
288 291 bplist_destroy(&ds->ds_pending_deadlist);
289 292 if (ds->ds_deadlist.dl_os != NULL)
290 293 dsl_deadlist_close(&ds->ds_deadlist);
291 294 if (ds->ds_dir)
292 295 dsl_dir_async_rele(ds->ds_dir, ds);
293 296
294 297 ASSERT(!list_link_active(&ds->ds_synced_link));
295 298
296 299 list_destroy(&ds->ds_prop_cbs);
297 300 mutex_destroy(&ds->ds_lock);
298 301 mutex_destroy(&ds->ds_opening_lock);
299 302 mutex_destroy(&ds->ds_sendstream_lock);
300 303 refcount_destroy(&ds->ds_longholds);
301 304
302 305 kmem_free(ds, sizeof (dsl_dataset_t));
303 306 }
304 307
305 308 int
306 309 dsl_dataset_get_snapname(dsl_dataset_t *ds)
307 310 {
308 311 dsl_dataset_phys_t *headphys;
309 312 int err;
310 313 dmu_buf_t *headdbuf;
311 314 dsl_pool_t *dp = ds->ds_dir->dd_pool;
312 315 objset_t *mos = dp->dp_meta_objset;
313 316
314 317 if (ds->ds_snapname[0])
315 318 return (0);
316 319 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
317 320 return (0);
318 321
319 322 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
320 323 FTAG, &headdbuf);
321 324 if (err != 0)
322 325 return (err);
323 326 headphys = headdbuf->db_data;
324 327 err = zap_value_search(dp->dp_meta_objset,
325 328 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
326 329 dmu_buf_rele(headdbuf, FTAG);
327 330 return (err);
328 331 }
329 332
330 333 int
331 334 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
332 335 {
333 336 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
334 337 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
335 338 matchtype_t mt;
336 339 int err;
337 340
338 341 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
339 342 mt = MT_FIRST;
340 343 else
341 344 mt = MT_EXACT;
342 345
343 346 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
344 347 value, mt, NULL, 0, NULL);
345 348 if (err == ENOTSUP && mt == MT_FIRST)
346 349 err = zap_lookup(mos, snapobj, name, 8, 1, value);
347 350 return (err);
348 351 }
349 352
350 353 int
351 354 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
352 355 boolean_t adj_cnt)
353 356 {
354 357 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
355 358 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
356 359 matchtype_t mt;
357 360 int err;
358 361
359 362 dsl_dir_snap_cmtime_update(ds->ds_dir);
360 363
361 364 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
362 365 mt = MT_FIRST;
363 366 else
364 367 mt = MT_EXACT;
365 368
366 369 err = zap_remove_norm(mos, snapobj, name, mt, tx);
367 370 if (err == ENOTSUP && mt == MT_FIRST)
368 371 err = zap_remove(mos, snapobj, name, tx);
369 372
370 373 if (err == 0 && adj_cnt)
371 374 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
372 375 DD_FIELD_SNAPSHOT_COUNT, tx);
373 376
374 377 return (err);
375 378 }
376 379
377 380 boolean_t
378 381 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
379 382 {
380 383 dmu_buf_t *dbuf = ds->ds_dbuf;
381 384 boolean_t result = B_FALSE;
382 385
383 386 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
384 387 ds->ds_object, DMU_BONUS_BLKID, tag)) {
385 388
386 389 if (ds == dmu_buf_get_user(dbuf))
387 390 result = B_TRUE;
388 391 else
389 392 dmu_buf_rele(dbuf, tag);
390 393 }
391 394
392 395 return (result);
393 396 }
394 397
395 398 int
396 399 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
397 400 dsl_dataset_t **dsp)
398 401 {
399 402 objset_t *mos = dp->dp_meta_objset;
400 403 dmu_buf_t *dbuf;
401 404 dsl_dataset_t *ds;
402 405 int err;
403 406 dmu_object_info_t doi;
404 407
405 408 ASSERT(dsl_pool_config_held(dp));
406 409
407 410 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
408 411 if (err != 0)
409 412 return (err);
410 413
411 414 /* Make sure dsobj has the correct object type. */
412 415 dmu_object_info_from_db(dbuf, &doi);
413 416 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
414 417 dmu_buf_rele(dbuf, tag);
415 418 return (SET_ERROR(EINVAL));
416 419 }
417 420
418 421 ds = dmu_buf_get_user(dbuf);
419 422 if (ds == NULL) {
420 423 dsl_dataset_t *winner = NULL;
421 424
422 425 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
423 426 ds->ds_dbuf = dbuf;
424 427 ds->ds_object = dsobj;
425 428 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
426 429
427 430 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
428 431 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
429 432 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
430 433 refcount_create(&ds->ds_longholds);
431 434
432 435 bplist_create(&ds->ds_pending_deadlist);
433 436 dsl_deadlist_open(&ds->ds_deadlist,
434 437 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
435 438
436 439 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
437 440 offsetof(dmu_sendarg_t, dsa_link));
438 441
439 442 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
440 443 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
441 444
442 445 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
443 446 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
444 447 if (!(spa_feature_table[f].fi_flags &
445 448 ZFEATURE_FLAG_PER_DATASET))
446 449 continue;
447 450 err = zap_contains(mos, dsobj,
448 451 spa_feature_table[f].fi_guid);
449 452 if (err == 0) {
450 453 ds->ds_feature_inuse[f] = B_TRUE;
451 454 } else {
452 455 ASSERT3U(err, ==, ENOENT);
453 456 err = 0;
454 457 }
455 458 }
456 459 }
457 460
458 461 err = dsl_dir_hold_obj(dp,
459 462 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
460 463 if (err != 0) {
461 464 mutex_destroy(&ds->ds_lock);
462 465 mutex_destroy(&ds->ds_opening_lock);
463 466 mutex_destroy(&ds->ds_sendstream_lock);
464 467 refcount_destroy(&ds->ds_longholds);
465 468 bplist_destroy(&ds->ds_pending_deadlist);
466 469 dsl_deadlist_close(&ds->ds_deadlist);
467 470 kmem_free(ds, sizeof (dsl_dataset_t));
468 471 dmu_buf_rele(dbuf, tag);
469 472 return (err);
470 473 }
471 474
472 475 if (!ds->ds_is_snapshot) {
473 476 ds->ds_snapname[0] = '\0';
474 477 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
475 478 err = dsl_dataset_hold_obj(dp,
476 479 dsl_dataset_phys(ds)->ds_prev_snap_obj,
477 480 ds, &ds->ds_prev);
478 481 }
479 482 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
480 483 int zaperr = zap_lookup(mos, ds->ds_object,
481 484 DS_FIELD_BOOKMARK_NAMES,
482 485 sizeof (ds->ds_bookmarks), 1,
483 486 &ds->ds_bookmarks);
484 487 if (zaperr != ENOENT)
485 488 VERIFY0(zaperr);
486 489 }
487 490 } else {
488 491 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
489 492 err = dsl_dataset_get_snapname(ds);
490 493 if (err == 0 &&
491 494 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
492 495 err = zap_count(
493 496 ds->ds_dir->dd_pool->dp_meta_objset,
494 497 dsl_dataset_phys(ds)->ds_userrefs_obj,
495 498 &ds->ds_userrefs);
496 499 }
497 500 }
498 501
499 502 if (err == 0 && !ds->ds_is_snapshot) {
500 503 err = dsl_prop_get_int_ds(ds,
501 504 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
502 505 &ds->ds_reserved);
503 506 if (err == 0) {
504 507 err = dsl_prop_get_int_ds(ds,
505 508 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
506 509 &ds->ds_quota);
507 510 }
508 511 } else {
509 512 ds->ds_reserved = ds->ds_quota = 0;
510 513 }
511 514
512 515 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
513 516 if (err == 0)
514 517 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
515 518
516 519 if (err != 0 || winner != NULL) {
517 520 bplist_destroy(&ds->ds_pending_deadlist);
518 521 dsl_deadlist_close(&ds->ds_deadlist);
519 522 if (ds->ds_prev)
520 523 dsl_dataset_rele(ds->ds_prev, ds);
521 524 dsl_dir_rele(ds->ds_dir, ds);
522 525 mutex_destroy(&ds->ds_lock);
523 526 mutex_destroy(&ds->ds_opening_lock);
524 527 mutex_destroy(&ds->ds_sendstream_lock);
525 528 refcount_destroy(&ds->ds_longholds);
526 529 kmem_free(ds, sizeof (dsl_dataset_t));
527 530 if (err != 0) {
528 531 dmu_buf_rele(dbuf, tag);
529 532 return (err);
530 533 }
531 534 ds = winner;
532 535 } else {
533 536 ds->ds_fsid_guid =
534 537 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
535 538 }
536 539 }
537 540 ASSERT3P(ds->ds_dbuf, ==, dbuf);
538 541 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
539 542 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
540 543 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
541 544 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
542 545 *dsp = ds;
543 546 return (0);
544 547 }
545 548
546 549 int
547 550 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
548 551 void *tag, dsl_dataset_t **dsp)
549 552 {
550 553 dsl_dir_t *dd;
551 554 const char *snapname;
552 555 uint64_t obj;
553 556 int err = 0;
554 557 dsl_dataset_t *ds;
555 558
556 559 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
557 560 if (err != 0)
558 561 return (err);
559 562
560 563 ASSERT(dsl_pool_config_held(dp));
561 564 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
562 565 if (obj != 0)
563 566 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
564 567 else
565 568 err = SET_ERROR(ENOENT);
566 569
567 570 /* we may be looking for a snapshot */
568 571 if (err == 0 && snapname != NULL) {
569 572 dsl_dataset_t *snap_ds;
570 573
571 574 if (*snapname++ != '@') {
572 575 dsl_dataset_rele(ds, tag);
573 576 dsl_dir_rele(dd, FTAG);
574 577 return (SET_ERROR(ENOENT));
575 578 }
576 579
577 580 dprintf("looking for snapshot '%s'\n", snapname);
578 581 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
579 582 if (err == 0)
580 583 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
581 584 dsl_dataset_rele(ds, tag);
582 585
583 586 if (err == 0) {
584 587 mutex_enter(&snap_ds->ds_lock);
585 588 if (snap_ds->ds_snapname[0] == 0)
586 589 (void) strlcpy(snap_ds->ds_snapname, snapname,
587 590 sizeof (snap_ds->ds_snapname));
588 591 mutex_exit(&snap_ds->ds_lock);
589 592 ds = snap_ds;
590 593 }
591 594 }
592 595 if (err == 0)
593 596 *dsp = ds;
594 597 dsl_dir_rele(dd, FTAG);
595 598 return (err);
596 599 }
597 600
598 601 int
599 602 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
600 603 void *tag, dsl_dataset_t **dsp)
601 604 {
602 605 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
603 606 if (err != 0)
604 607 return (err);
605 608 if (!dsl_dataset_tryown(*dsp, tag)) {
606 609 dsl_dataset_rele(*dsp, tag);
607 610 *dsp = NULL;
608 611 return (SET_ERROR(EBUSY));
609 612 }
610 613 return (0);
611 614 }
612 615
613 616 int
614 617 dsl_dataset_own(dsl_pool_t *dp, const char *name,
615 618 void *tag, dsl_dataset_t **dsp)
616 619 {
617 620 int err = dsl_dataset_hold(dp, name, tag, dsp);
618 621 if (err != 0)
619 622 return (err);
620 623 if (!dsl_dataset_tryown(*dsp, tag)) {
621 624 dsl_dataset_rele(*dsp, tag);
622 625 return (SET_ERROR(EBUSY));
623 626 }
624 627 return (0);
625 628 }
626 629
627 630 /*
628 631 * See the comment above dsl_pool_hold() for details. In summary, a long
629 632 * hold is used to prevent destruction of a dataset while the pool hold
630 633 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
631 634 *
632 635 * The dataset and pool must be held when this function is called. After it
633 636 * is called, the pool hold may be released while the dataset is still held
634 637 * and accessed.
635 638 */
636 639 void
637 640 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
638 641 {
639 642 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
640 643 (void) refcount_add(&ds->ds_longholds, tag);
641 644 }
642 645
643 646 void
644 647 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
645 648 {
646 649 (void) refcount_remove(&ds->ds_longholds, tag);
647 650 }
648 651
649 652 /* Return B_TRUE if there are any long holds on this dataset. */
650 653 boolean_t
651 654 dsl_dataset_long_held(dsl_dataset_t *ds)
652 655 {
653 656 return (!refcount_is_zero(&ds->ds_longholds));
654 657 }
655 658
656 659 void
657 660 dsl_dataset_name(dsl_dataset_t *ds, char *name)
658 661 {
659 662 if (ds == NULL) {
660 663 (void) strcpy(name, "mos");
661 664 } else {
662 665 dsl_dir_name(ds->ds_dir, name);
663 666 VERIFY0(dsl_dataset_get_snapname(ds));
664 667 if (ds->ds_snapname[0]) {
665 668 (void) strcat(name, "@");
666 669 /*
667 670 * We use a "recursive" mutex so that we
668 671 * can call dprintf_ds() with ds_lock held.
669 672 */
670 673 if (!MUTEX_HELD(&ds->ds_lock)) {
671 674 mutex_enter(&ds->ds_lock);
672 675 (void) strcat(name, ds->ds_snapname);
673 676 mutex_exit(&ds->ds_lock);
674 677 } else {
675 678 (void) strcat(name, ds->ds_snapname);
676 679 }
677 680 }
678 681 }
679 682 }
680 683
681 684 void
682 685 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
683 686 {
684 687 dmu_buf_rele(ds->ds_dbuf, tag);
685 688 }
686 689
687 690 void
688 691 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
689 692 {
690 693 ASSERT3P(ds->ds_owner, ==, tag);
691 694 ASSERT(ds->ds_dbuf != NULL);
692 695
693 696 mutex_enter(&ds->ds_lock);
694 697 ds->ds_owner = NULL;
695 698 mutex_exit(&ds->ds_lock);
696 699 dsl_dataset_long_rele(ds, tag);
697 700 dsl_dataset_rele(ds, tag);
698 701 }
699 702
700 703 boolean_t
701 704 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
702 705 {
703 706 boolean_t gotit = FALSE;
704 707
705 708 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
706 709 mutex_enter(&ds->ds_lock);
707 710 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
708 711 ds->ds_owner = tag;
709 712 dsl_dataset_long_hold(ds, tag);
710 713 gotit = TRUE;
711 714 }
712 715 mutex_exit(&ds->ds_lock);
713 716 return (gotit);
714 717 }
715 718
716 719 boolean_t
717 720 dsl_dataset_has_owner(dsl_dataset_t *ds)
718 721 {
719 722 boolean_t rv;
720 723 mutex_enter(&ds->ds_lock);
721 724 rv = (ds->ds_owner != NULL);
722 725 mutex_exit(&ds->ds_lock);
723 726 return (rv);
724 727 }
725 728
726 729 static void
727 730 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
728 731 {
729 732 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730 733 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
731 734 uint64_t zero = 0;
732 735
733 736 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
734 737
735 738 spa_feature_incr(spa, f, tx);
736 739 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
737 740
738 741 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
739 742 sizeof (zero), 1, &zero, tx));
740 743 }
741 744
742 745 void
743 746 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
744 747 {
745 748 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
746 749 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
747 750
748 751 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
749 752
750 753 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
751 754 spa_feature_decr(spa, f, tx);
752 755 }
753 756
754 757 uint64_t
755 758 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
756 759 uint64_t flags, dmu_tx_t *tx)
757 760 {
758 761 dsl_pool_t *dp = dd->dd_pool;
759 762 dmu_buf_t *dbuf;
760 763 dsl_dataset_phys_t *dsphys;
761 764 uint64_t dsobj;
762 765 objset_t *mos = dp->dp_meta_objset;
763 766
764 767 if (origin == NULL)
765 768 origin = dp->dp_origin_snap;
766 769
767 770 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
768 771 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
769 772 ASSERT(dmu_tx_is_syncing(tx));
770 773 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
771 774
772 775 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
773 776 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
774 777 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
775 778 dmu_buf_will_dirty(dbuf, tx);
776 779 dsphys = dbuf->db_data;
777 780 bzero(dsphys, sizeof (dsl_dataset_phys_t));
778 781 dsphys->ds_dir_obj = dd->dd_object;
779 782 dsphys->ds_flags = flags;
780 783 dsphys->ds_fsid_guid = unique_create();
781 784 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
782 785 sizeof (dsphys->ds_guid));
783 786 dsphys->ds_snapnames_zapobj =
784 787 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
785 788 DMU_OT_NONE, 0, tx);
786 789 dsphys->ds_creation_time = gethrestime_sec();
787 790 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
788 791
789 792 if (origin == NULL) {
790 793 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
791 794 } else {
792 795 dsl_dataset_t *ohds; /* head of the origin snapshot */
793 796
794 797 dsphys->ds_prev_snap_obj = origin->ds_object;
795 798 dsphys->ds_prev_snap_txg =
796 799 dsl_dataset_phys(origin)->ds_creation_txg;
797 800 dsphys->ds_referenced_bytes =
798 801 dsl_dataset_phys(origin)->ds_referenced_bytes;
799 802 dsphys->ds_compressed_bytes =
800 803 dsl_dataset_phys(origin)->ds_compressed_bytes;
801 804 dsphys->ds_uncompressed_bytes =
802 805 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
803 806 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
804 807
805 808 /*
806 809 * Inherit flags that describe the dataset's contents
807 810 * (INCONSISTENT) or properties (Case Insensitive).
808 811 */
809 812 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
810 813 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
811 814
812 815 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
813 816 if (origin->ds_feature_inuse[f])
814 817 dsl_dataset_activate_feature(dsobj, f, tx);
815 818 }
816 819
817 820 dmu_buf_will_dirty(origin->ds_dbuf, tx);
818 821 dsl_dataset_phys(origin)->ds_num_children++;
819 822
820 823 VERIFY0(dsl_dataset_hold_obj(dp,
821 824 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
822 825 FTAG, &ohds));
823 826 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
824 827 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
825 828 dsl_dataset_rele(ohds, FTAG);
826 829
827 830 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
828 831 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
829 832 dsl_dataset_phys(origin)->ds_next_clones_obj =
830 833 zap_create(mos,
831 834 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
832 835 }
833 836 VERIFY0(zap_add_int(mos,
834 837 dsl_dataset_phys(origin)->ds_next_clones_obj,
835 838 dsobj, tx));
836 839 }
837 840
838 841 dmu_buf_will_dirty(dd->dd_dbuf, tx);
839 842 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
840 843 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
841 844 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
842 845 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
843 846 dsl_dir_phys(origin->ds_dir)->dd_clones =
844 847 zap_create(mos,
845 848 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
846 849 }
847 850 VERIFY0(zap_add_int(mos,
848 851 dsl_dir_phys(origin->ds_dir)->dd_clones,
849 852 dsobj, tx));
850 853 }
851 854 }
852 855
853 856 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
854 857 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
855 858
856 859 dmu_buf_rele(dbuf, FTAG);
857 860
858 861 dmu_buf_will_dirty(dd->dd_dbuf, tx);
859 862 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
860 863
861 864 return (dsobj);
862 865 }
863 866
864 867 static void
865 868 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
866 869 {
867 870 objset_t *os;
868 871
869 872 VERIFY0(dmu_objset_from_ds(ds, &os));
870 873 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
871 874 dsl_dataset_dirty(ds, tx);
872 875 }
873 876
874 877 uint64_t
875 878 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
876 879 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
877 880 {
878 881 dsl_pool_t *dp = pdd->dd_pool;
879 882 uint64_t dsobj, ddobj;
880 883 dsl_dir_t *dd;
881 884
882 885 ASSERT(dmu_tx_is_syncing(tx));
883 886 ASSERT(lastname[0] != '@');
884 887
885 888 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
886 889 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
887 890
888 891 dsobj = dsl_dataset_create_sync_dd(dd, origin,
889 892 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
890 893
891 894 dsl_deleg_set_create_perms(dd, tx, cr);
892 895
893 896 /*
894 897 * Since we're creating a new node we know it's a leaf, so we can
895 898 * initialize the counts if the limit feature is active.
896 899 */
897 900 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
898 901 uint64_t cnt = 0;
899 902 objset_t *os = dd->dd_pool->dp_meta_objset;
900 903
901 904 dsl_dir_zapify(dd, tx);
902 905 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
903 906 sizeof (cnt), 1, &cnt, tx));
904 907 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
905 908 sizeof (cnt), 1, &cnt, tx));
906 909 }
907 910
908 911 dsl_dir_rele(dd, FTAG);
909 912
910 913 /*
911 914 * If we are creating a clone, make sure we zero out any stale
912 915 * data from the origin snapshots zil header.
913 916 */
914 917 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
915 918 dsl_dataset_t *ds;
916 919
917 920 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
918 921 dsl_dataset_zero_zil(ds, tx);
919 922 dsl_dataset_rele(ds, FTAG);
920 923 }
921 924
922 925 return (dsobj);
923 926 }
924 927
925 928 /*
926 929 * The unique space in the head dataset can be calculated by subtracting
927 930 * the space used in the most recent snapshot, that is still being used
928 931 * in this file system, from the space currently in use. To figure out
929 932 * the space in the most recent snapshot still in use, we need to take
930 933 * the total space used in the snapshot and subtract out the space that
931 934 * has been freed up since the snapshot was taken.
932 935 */
933 936 void
934 937 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
935 938 {
936 939 uint64_t mrs_used;
937 940 uint64_t dlused, dlcomp, dluncomp;
938 941
939 942 ASSERT(!ds->ds_is_snapshot);
940 943
941 944 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
942 945 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
943 946 else
944 947 mrs_used = 0;
945 948
946 949 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
947 950
948 951 ASSERT3U(dlused, <=, mrs_used);
949 952 dsl_dataset_phys(ds)->ds_unique_bytes =
950 953 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
951 954
952 955 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
953 956 SPA_VERSION_UNIQUE_ACCURATE)
954 957 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
955 958 }
956 959
957 960 void
958 961 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
959 962 dmu_tx_t *tx)
960 963 {
961 964 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
962 965 uint64_t count;
963 966 int err;
964 967
965 968 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
966 969 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
967 970 obj, tx);
968 971 /*
969 972 * The err should not be ENOENT, but a bug in a previous version
970 973 * of the code could cause upgrade_clones_cb() to not set
971 974 * ds_next_snap_obj when it should, leading to a missing entry.
972 975 * If we knew that the pool was created after
973 976 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
974 977 * ENOENT. However, at least we can check that we don't have
975 978 * too many entries in the next_clones_obj even after failing to
976 979 * remove this one.
977 980 */
978 981 if (err != ENOENT)
979 982 VERIFY0(err);
980 983 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
981 984 &count));
982 985 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
983 986 }
984 987
985 988
986 989 blkptr_t *
987 990 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
988 991 {
989 992 return (&dsl_dataset_phys(ds)->ds_bp);
990 993 }
991 994
992 995 void
993 996 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
994 997 {
995 998 ASSERT(dmu_tx_is_syncing(tx));
996 999 /* If it's the meta-objset, set dp_meta_rootbp */
997 1000 if (ds == NULL) {
998 1001 tx->tx_pool->dp_meta_rootbp = *bp;
999 1002 } else {
1000 1003 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1001 1004 dsl_dataset_phys(ds)->ds_bp = *bp;
1002 1005 }
1003 1006 }
1004 1007
1005 1008 spa_t *
1006 1009 dsl_dataset_get_spa(dsl_dataset_t *ds)
1007 1010 {
1008 1011 return (ds->ds_dir->dd_pool->dp_spa);
1009 1012 }
1010 1013
1011 1014 void
1012 1015 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1013 1016 {
1014 1017 dsl_pool_t *dp;
1015 1018
1016 1019 if (ds == NULL) /* this is the meta-objset */
1017 1020 return;
1018 1021
1019 1022 ASSERT(ds->ds_objset != NULL);
1020 1023
1021 1024 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1022 1025 panic("dirtying snapshot!");
1023 1026
1024 1027 dp = ds->ds_dir->dd_pool;
1025 1028
1026 1029 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1027 1030 /* up the hold count until we can be written out */
1028 1031 dmu_buf_add_ref(ds->ds_dbuf, ds);
1029 1032 }
1030 1033 }
1031 1034
1032 1035 boolean_t
1033 1036 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1034 1037 {
1035 1038 for (int t = 0; t < TXG_SIZE; t++) {
1036 1039 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1037 1040 ds, t))
1038 1041 return (B_TRUE);
1039 1042 }
1040 1043 return (B_FALSE);
1041 1044 }
1042 1045
1043 1046 static int
1044 1047 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1045 1048 {
1046 1049 uint64_t asize;
1047 1050
1048 1051 if (!dmu_tx_is_syncing(tx))
1049 1052 return (0);
1050 1053
1051 1054 /*
1052 1055 * If there's an fs-only reservation, any blocks that might become
1053 1056 * owned by the snapshot dataset must be accommodated by space
1054 1057 * outside of the reservation.
1055 1058 */
1056 1059 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1057 1060 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1058 1061 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1059 1062 return (SET_ERROR(ENOSPC));
1060 1063
1061 1064 /*
1062 1065 * Propagate any reserved space for this snapshot to other
1063 1066 * snapshot checks in this sync group.
1064 1067 */
1065 1068 if (asize > 0)
1066 1069 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1067 1070
1068 1071 return (0);
1069 1072 }
1070 1073
1071 1074 typedef struct dsl_dataset_snapshot_arg {
1072 1075 nvlist_t *ddsa_snaps;
1073 1076 nvlist_t *ddsa_props;
1074 1077 nvlist_t *ddsa_errors;
1075 1078 cred_t *ddsa_cr;
1076 1079 } dsl_dataset_snapshot_arg_t;
1077 1080
1078 1081 int
1079 1082 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1080 1083 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1081 1084 {
1082 1085 int error;
1083 1086 uint64_t value;
1084 1087
1085 1088 ds->ds_trysnap_txg = tx->tx_txg;
1086 1089
1087 1090 if (!dmu_tx_is_syncing(tx))
1088 1091 return (0);
1089 1092
1090 1093 /*
1091 1094 * We don't allow multiple snapshots of the same txg. If there
1092 1095 * is already one, try again.
1093 1096 */
1094 1097 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1095 1098 return (SET_ERROR(EAGAIN));
1096 1099
1097 1100 /*
1098 1101 * Check for conflicting snapshot name.
1099 1102 */
1100 1103 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1101 1104 if (error == 0)
1102 1105 return (SET_ERROR(EEXIST));
1103 1106 if (error != ENOENT)
1104 1107 return (error);
1105 1108
1106 1109 /*
1107 1110 * We don't allow taking snapshots of inconsistent datasets, such as
1108 1111 * those into which we are currently receiving. However, if we are
1109 1112 * creating this snapshot as part of a receive, this check will be
1110 1113 * executed atomically with respect to the completion of the receive
1111 1114 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1112 1115 * case we ignore this, knowing it will be fixed up for us shortly in
1113 1116 * dmu_recv_end_sync().
1114 1117 */
1115 1118 if (!recv && DS_IS_INCONSISTENT(ds))
1116 1119 return (SET_ERROR(EBUSY));
1117 1120
1118 1121 /*
1119 1122 * Skip the check for temporary snapshots or if we have already checked
1120 1123 * the counts in dsl_dataset_snapshot_check. This means we really only
1121 1124 * check the count here when we're receiving a stream.
1122 1125 */
1123 1126 if (cnt != 0 && cr != NULL) {
1124 1127 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1125 1128 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1126 1129 if (error != 0)
1127 1130 return (error);
1128 1131 }
1129 1132
1130 1133 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1131 1134 if (error != 0)
1132 1135 return (error);
1133 1136
1134 1137 return (0);
1135 1138 }
1136 1139
1137 1140 static int
1138 1141 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1139 1142 {
1140 1143 dsl_dataset_snapshot_arg_t *ddsa = arg;
1141 1144 dsl_pool_t *dp = dmu_tx_pool(tx);
1142 1145 nvpair_t *pair;
1143 1146 int rv = 0;
1144 1147
1145 1148 /*
1146 1149 * Pre-compute how many total new snapshots will be created for each
1147 1150 * level in the tree and below. This is needed for validating the
1148 1151 * snapshot limit when either taking a recursive snapshot or when
1149 1152 * taking multiple snapshots.
1150 1153 *
1151 1154 * The problem is that the counts are not actually adjusted when
1152 1155 * we are checking, only when we finally sync. For a single snapshot,
1153 1156 * this is easy, the count will increase by 1 at each node up the tree,
1154 1157 * but its more complicated for the recursive/multiple snapshot case.
1155 1158 *
1156 1159 * The dsl_fs_ss_limit_check function does recursively check the count
1157 1160 * at each level up the tree but since it is validating each snapshot
1158 1161 * independently we need to be sure that we are validating the complete
1159 1162 * count for the entire set of snapshots. We do this by rolling up the
1160 1163 * counts for each component of the name into an nvlist and then
1161 1164 * checking each of those cases with the aggregated count.
1162 1165 *
1163 1166 * This approach properly handles not only the recursive snapshot
1164 1167 * case (where we get all of those on the ddsa_snaps list) but also
1165 1168 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1166 1169 * validate the limit on 'a' using a count of 2).
1167 1170 *
1168 1171 * We validate the snapshot names in the third loop and only report
1169 1172 * name errors once.
1170 1173 */
1171 1174 if (dmu_tx_is_syncing(tx)) {
1172 1175 nvlist_t *cnt_track = NULL;
1173 1176 cnt_track = fnvlist_alloc();
1174 1177
1175 1178 /* Rollup aggregated counts into the cnt_track list */
1176 1179 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1177 1180 pair != NULL;
1178 1181 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1179 1182 char *pdelim;
1180 1183 uint64_t val;
1181 1184 char nm[MAXPATHLEN];
1182 1185
1183 1186 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1184 1187 pdelim = strchr(nm, '@');
1185 1188 if (pdelim == NULL)
1186 1189 continue;
1187 1190 *pdelim = '\0';
1188 1191
1189 1192 do {
1190 1193 if (nvlist_lookup_uint64(cnt_track, nm,
1191 1194 &val) == 0) {
1192 1195 /* update existing entry */
1193 1196 fnvlist_add_uint64(cnt_track, nm,
1194 1197 val + 1);
1195 1198 } else {
1196 1199 /* add to list */
1197 1200 fnvlist_add_uint64(cnt_track, nm, 1);
1198 1201 }
1199 1202
1200 1203 pdelim = strrchr(nm, '/');
1201 1204 if (pdelim != NULL)
1202 1205 *pdelim = '\0';
1203 1206 } while (pdelim != NULL);
1204 1207 }
1205 1208
1206 1209 /* Check aggregated counts at each level */
1207 1210 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1208 1211 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1209 1212 int error = 0;
1210 1213 char *name;
1211 1214 uint64_t cnt = 0;
1212 1215 dsl_dataset_t *ds;
1213 1216
1214 1217 name = nvpair_name(pair);
1215 1218 cnt = fnvpair_value_uint64(pair);
1216 1219 ASSERT(cnt > 0);
1217 1220
1218 1221 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1219 1222 if (error == 0) {
1220 1223 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1221 1224 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1222 1225 ddsa->ddsa_cr);
1223 1226 dsl_dataset_rele(ds, FTAG);
1224 1227 }
1225 1228
1226 1229 if (error != 0) {
1227 1230 if (ddsa->ddsa_errors != NULL)
1228 1231 fnvlist_add_int32(ddsa->ddsa_errors,
1229 1232 name, error);
1230 1233 rv = error;
1231 1234 /* only report one error for this check */
1232 1235 break;
1233 1236 }
1234 1237 }
1235 1238 nvlist_free(cnt_track);
1236 1239 }
1237 1240
1238 1241 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1239 1242 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1240 1243 int error = 0;
1241 1244 dsl_dataset_t *ds;
1242 1245 char *name, *atp;
1243 1246 char dsname[MAXNAMELEN];
1244 1247
1245 1248 name = nvpair_name(pair);
1246 1249 if (strlen(name) >= MAXNAMELEN)
1247 1250 error = SET_ERROR(ENAMETOOLONG);
1248 1251 if (error == 0) {
1249 1252 atp = strchr(name, '@');
1250 1253 if (atp == NULL)
1251 1254 error = SET_ERROR(EINVAL);
1252 1255 if (error == 0)
1253 1256 (void) strlcpy(dsname, name, atp - name + 1);
1254 1257 }
1255 1258 if (error == 0)
1256 1259 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1257 1260 if (error == 0) {
1258 1261 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1259 1262 error = dsl_dataset_snapshot_check_impl(ds,
1260 1263 atp + 1, tx, B_FALSE, 0, NULL);
1261 1264 dsl_dataset_rele(ds, FTAG);
1262 1265 }
1263 1266
1264 1267 if (error != 0) {
1265 1268 if (ddsa->ddsa_errors != NULL) {
1266 1269 fnvlist_add_int32(ddsa->ddsa_errors,
1267 1270 name, error);
1268 1271 }
1269 1272 rv = error;
1270 1273 }
1271 1274 }
1272 1275
1273 1276 return (rv);
1274 1277 }
1275 1278
1276 1279 void
1277 1280 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1278 1281 dmu_tx_t *tx)
1279 1282 {
1280 1283 static zil_header_t zero_zil;
1281 1284
1282 1285 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1283 1286 dmu_buf_t *dbuf;
1284 1287 dsl_dataset_phys_t *dsphys;
1285 1288 uint64_t dsobj, crtxg;
1286 1289 objset_t *mos = dp->dp_meta_objset;
1287 1290 objset_t *os;
1288 1291
1289 1292 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1290 1293
1291 1294 /*
1292 1295 * If we are on an old pool, the zil must not be active, in which
1293 1296 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1294 1297 */
1295 1298 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1296 1299 dmu_objset_from_ds(ds, &os) != 0 ||
1297 1300 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1298 1301 sizeof (zero_zil)) == 0);
1299 1302
1300 1303 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1301 1304
1302 1305 /*
1303 1306 * The origin's ds_creation_txg has to be < TXG_INITIAL
1304 1307 */
1305 1308 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1306 1309 crtxg = 1;
1307 1310 else
1308 1311 crtxg = tx->tx_txg;
1309 1312
1310 1313 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1311 1314 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1312 1315 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1313 1316 dmu_buf_will_dirty(dbuf, tx);
1314 1317 dsphys = dbuf->db_data;
1315 1318 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1316 1319 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1317 1320 dsphys->ds_fsid_guid = unique_create();
1318 1321 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1319 1322 sizeof (dsphys->ds_guid));
1320 1323 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1321 1324 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1322 1325 dsphys->ds_next_snap_obj = ds->ds_object;
1323 1326 dsphys->ds_num_children = 1;
1324 1327 dsphys->ds_creation_time = gethrestime_sec();
1325 1328 dsphys->ds_creation_txg = crtxg;
1326 1329 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1327 1330 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1328 1331 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1329 1332 dsphys->ds_uncompressed_bytes =
1330 1333 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1331 1334 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1332 1335 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1333 1336 dmu_buf_rele(dbuf, FTAG);
1334 1337
1335 1338 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1336 1339 if (ds->ds_feature_inuse[f])
1337 1340 dsl_dataset_activate_feature(dsobj, f, tx);
1338 1341 }
1339 1342
1340 1343 ASSERT3U(ds->ds_prev != 0, ==,
1341 1344 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1342 1345 if (ds->ds_prev) {
1343 1346 uint64_t next_clones_obj =
1344 1347 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1345 1348 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1346 1349 ds->ds_object ||
1347 1350 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1348 1351 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1349 1352 ds->ds_object) {
1350 1353 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1351 1354 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1352 1355 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1353 1356 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1354 1357 } else if (next_clones_obj != 0) {
1355 1358 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1356 1359 dsphys->ds_next_snap_obj, tx);
1357 1360 VERIFY0(zap_add_int(mos,
1358 1361 next_clones_obj, dsobj, tx));
1359 1362 }
1360 1363 }
1361 1364
1362 1365 /*
1363 1366 * If we have a reference-reservation on this dataset, we will
1364 1367 * need to increase the amount of refreservation being charged
1365 1368 * since our unique space is going to zero.
1366 1369 */
1367 1370 if (ds->ds_reserved) {
1368 1371 int64_t delta;
1369 1372 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1370 1373 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1371 1374 ds->ds_reserved);
1372 1375 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1373 1376 delta, 0, 0, tx);
1374 1377 }
1375 1378
1376 1379 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1377 1380 dsl_dataset_phys(ds)->ds_deadlist_obj =
1378 1381 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1379 1382 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1380 1383 dsl_deadlist_close(&ds->ds_deadlist);
1381 1384 dsl_deadlist_open(&ds->ds_deadlist, mos,
1382 1385 dsl_dataset_phys(ds)->ds_deadlist_obj);
1383 1386 dsl_deadlist_add_key(&ds->ds_deadlist,
1384 1387 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1385 1388
1386 1389 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1387 1390 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1388 1391 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1389 1392 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1390 1393 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1391 1394 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1392 1395
1393 1396 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1394 1397 snapname, 8, 1, &dsobj, tx));
1395 1398
1396 1399 if (ds->ds_prev)
1397 1400 dsl_dataset_rele(ds->ds_prev, ds);
1398 1401 VERIFY0(dsl_dataset_hold_obj(dp,
1399 1402 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1400 1403
1401 1404 dsl_scan_ds_snapshotted(ds, tx);
1402 1405
1403 1406 dsl_dir_snap_cmtime_update(ds->ds_dir);
1404 1407
1405 1408 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1406 1409 }
1407 1410
1408 1411 static void
1409 1412 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1410 1413 {
1411 1414 dsl_dataset_snapshot_arg_t *ddsa = arg;
1412 1415 dsl_pool_t *dp = dmu_tx_pool(tx);
1413 1416 nvpair_t *pair;
1414 1417
1415 1418 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1416 1419 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1417 1420 dsl_dataset_t *ds;
1418 1421 char *name, *atp;
1419 1422 char dsname[MAXNAMELEN];
1420 1423
1421 1424 name = nvpair_name(pair);
1422 1425 atp = strchr(name, '@');
1423 1426 (void) strlcpy(dsname, name, atp - name + 1);
1424 1427 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1425 1428
1426 1429 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1427 1430 if (ddsa->ddsa_props != NULL) {
1428 1431 dsl_props_set_sync_impl(ds->ds_prev,
1429 1432 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1430 1433 }
1431 1434 dsl_dataset_rele(ds, FTAG);
1432 1435 }
1433 1436 }
1434 1437
1435 1438 /*
1436 1439 * The snapshots must all be in the same pool.
1437 1440 * All-or-nothing: if there are any failures, nothing will be modified.
1438 1441 */
1439 1442 int
1440 1443 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1441 1444 {
1442 1445 dsl_dataset_snapshot_arg_t ddsa;
1443 1446 nvpair_t *pair;
1444 1447 boolean_t needsuspend;
1445 1448 int error;
1446 1449 spa_t *spa;
1447 1450 char *firstname;
1448 1451 nvlist_t *suspended = NULL;
1449 1452
1450 1453 pair = nvlist_next_nvpair(snaps, NULL);
1451 1454 if (pair == NULL)
1452 1455 return (0);
1453 1456 firstname = nvpair_name(pair);
1454 1457
1455 1458 error = spa_open(firstname, &spa, FTAG);
1456 1459 if (error != 0)
1457 1460 return (error);
1458 1461 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1459 1462 spa_close(spa, FTAG);
1460 1463
1461 1464 if (needsuspend) {
1462 1465 suspended = fnvlist_alloc();
1463 1466 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1464 1467 pair = nvlist_next_nvpair(snaps, pair)) {
1465 1468 char fsname[MAXNAMELEN];
1466 1469 char *snapname = nvpair_name(pair);
1467 1470 char *atp;
1468 1471 void *cookie;
1469 1472
1470 1473 atp = strchr(snapname, '@');
1471 1474 if (atp == NULL) {
1472 1475 error = SET_ERROR(EINVAL);
1473 1476 break;
1474 1477 }
1475 1478 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1476 1479
1477 1480 error = zil_suspend(fsname, &cookie);
1478 1481 if (error != 0)
1479 1482 break;
1480 1483 fnvlist_add_uint64(suspended, fsname,
1481 1484 (uintptr_t)cookie);
1482 1485 }
1483 1486 }
1484 1487
1485 1488 ddsa.ddsa_snaps = snaps;
1486 1489 ddsa.ddsa_props = props;
1487 1490 ddsa.ddsa_errors = errors;
1488 1491 ddsa.ddsa_cr = CRED();
1489 1492
1490 1493 if (error == 0) {
1491 1494 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1492 1495 dsl_dataset_snapshot_sync, &ddsa,
1493 1496 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1494 1497 }
1495 1498
1496 1499 if (suspended != NULL) {
1497 1500 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1498 1501 pair = nvlist_next_nvpair(suspended, pair)) {
1499 1502 zil_resume((void *)(uintptr_t)
1500 1503 fnvpair_value_uint64(pair));
1501 1504 }
1502 1505 fnvlist_free(suspended);
1503 1506 }
1504 1507
1505 1508 return (error);
1506 1509 }
1507 1510
1508 1511 typedef struct dsl_dataset_snapshot_tmp_arg {
1509 1512 const char *ddsta_fsname;
1510 1513 const char *ddsta_snapname;
1511 1514 minor_t ddsta_cleanup_minor;
1512 1515 const char *ddsta_htag;
1513 1516 } dsl_dataset_snapshot_tmp_arg_t;
1514 1517
1515 1518 static int
1516 1519 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1517 1520 {
1518 1521 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1519 1522 dsl_pool_t *dp = dmu_tx_pool(tx);
1520 1523 dsl_dataset_t *ds;
1521 1524 int error;
1522 1525
1523 1526 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1524 1527 if (error != 0)
1525 1528 return (error);
1526 1529
1527 1530 /* NULL cred means no limit check for tmp snapshot */
1528 1531 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1529 1532 tx, B_FALSE, 0, NULL);
1530 1533 if (error != 0) {
1531 1534 dsl_dataset_rele(ds, FTAG);
1532 1535 return (error);
1533 1536 }
1534 1537
1535 1538 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1536 1539 dsl_dataset_rele(ds, FTAG);
1537 1540 return (SET_ERROR(ENOTSUP));
1538 1541 }
1539 1542 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1540 1543 B_TRUE, tx);
1541 1544 if (error != 0) {
1542 1545 dsl_dataset_rele(ds, FTAG);
1543 1546 return (error);
1544 1547 }
1545 1548
1546 1549 dsl_dataset_rele(ds, FTAG);
1547 1550 return (0);
1548 1551 }
1549 1552
1550 1553 static void
1551 1554 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1552 1555 {
1553 1556 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1554 1557 dsl_pool_t *dp = dmu_tx_pool(tx);
1555 1558 dsl_dataset_t *ds;
1556 1559
1557 1560 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1558 1561
1559 1562 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1560 1563 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1561 1564 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1562 1565 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1563 1566
1564 1567 dsl_dataset_rele(ds, FTAG);
1565 1568 }
1566 1569
1567 1570 int
1568 1571 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1569 1572 minor_t cleanup_minor, const char *htag)
1570 1573 {
1571 1574 dsl_dataset_snapshot_tmp_arg_t ddsta;
1572 1575 int error;
1573 1576 spa_t *spa;
1574 1577 boolean_t needsuspend;
1575 1578 void *cookie;
1576 1579
1577 1580 ddsta.ddsta_fsname = fsname;
1578 1581 ddsta.ddsta_snapname = snapname;
1579 1582 ddsta.ddsta_cleanup_minor = cleanup_minor;
1580 1583 ddsta.ddsta_htag = htag;
1581 1584
1582 1585 error = spa_open(fsname, &spa, FTAG);
1583 1586 if (error != 0)
1584 1587 return (error);
1585 1588 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1586 1589 spa_close(spa, FTAG);
1587 1590
1588 1591 if (needsuspend) {
1589 1592 error = zil_suspend(fsname, &cookie);
1590 1593 if (error != 0)
1591 1594 return (error);
1592 1595 }
1593 1596
1594 1597 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1595 1598 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1596 1599
1597 1600 if (needsuspend)
1598 1601 zil_resume(cookie);
1599 1602 return (error);
1600 1603 }
1601 1604
1602 1605
1603 1606 void
1604 1607 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1605 1608 {
1606 1609 ASSERT(dmu_tx_is_syncing(tx));
1607 1610 ASSERT(ds->ds_objset != NULL);
1608 1611 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1609 1612
1610 1613 /*
1611 1614 * in case we had to change ds_fsid_guid when we opened it,
1612 1615 * sync it out now.
1613 1616 */
1614 1617 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1615 1618 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1616 1619
1617 1620 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1618 1621 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1619 1622 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1620 1623 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1621 1624 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1622 1625 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1623 1626 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1624 1627 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1625 1628 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1626 1629 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1627 1630 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1628 1631 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1629 1632 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1630 1633 }
1631 1634
1632 1635 dmu_objset_sync(ds->ds_objset, zio, tx);
1633 1636
1634 1637 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1635 1638 if (ds->ds_feature_activation_needed[f]) {
1636 1639 if (ds->ds_feature_inuse[f])
1637 1640 continue;
1638 1641 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1639 1642 ds->ds_feature_inuse[f] = B_TRUE;
1640 1643 }
1641 1644 }
1642 1645 }
1643 1646
1644 1647 static void
1645 1648 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1646 1649 {
1647 1650 uint64_t count = 0;
1648 1651 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1649 1652 zap_cursor_t zc;
1650 1653 zap_attribute_t za;
1651 1654 nvlist_t *propval = fnvlist_alloc();
1652 1655 nvlist_t *val = fnvlist_alloc();
1653 1656
1654 1657 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1655 1658
1656 1659 /*
1657 1660 * There may be missing entries in ds_next_clones_obj
1658 1661 * due to a bug in a previous version of the code.
1659 1662 * Only trust it if it has the right number of entries.
1660 1663 */
1661 1664 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1662 1665 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1663 1666 &count));
1664 1667 }
1665 1668 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1666 1669 goto fail;
1667 1670 for (zap_cursor_init(&zc, mos,
1668 1671 dsl_dataset_phys(ds)->ds_next_clones_obj);
1669 1672 zap_cursor_retrieve(&zc, &za) == 0;
1670 1673 zap_cursor_advance(&zc)) {
1671 1674 dsl_dataset_t *clone;
1672 1675 char buf[ZFS_MAXNAMELEN];
1673 1676 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1674 1677 za.za_first_integer, FTAG, &clone));
1675 1678 dsl_dir_name(clone->ds_dir, buf);
1676 1679 fnvlist_add_boolean(val, buf);
1677 1680 dsl_dataset_rele(clone, FTAG);
1678 1681 }
1679 1682 zap_cursor_fini(&zc);
1680 1683 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1681 1684 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1682 1685 fail:
1683 1686 nvlist_free(val);
1684 1687 nvlist_free(propval);
1685 1688 }
1686 1689
1687 1690 static void
1688 1691 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1689 1692 {
1690 1693 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1691 1694
1692 1695 if (dsl_dataset_has_resume_receive_state(ds)) {
1693 1696 char *str;
1694 1697 void *packed;
1695 1698 uint8_t *compressed;
1696 1699 uint64_t val;
1697 1700 nvlist_t *token_nv = fnvlist_alloc();
1698 1701 size_t packed_size, compressed_size;
1699 1702
1700 1703 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1701 1704 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1702 1705 fnvlist_add_uint64(token_nv, "fromguid", val);
1703 1706 }
1704 1707 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1705 1708 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1706 1709 fnvlist_add_uint64(token_nv, "object", val);
1707 1710 }
1708 1711 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1709 1712 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1710 1713 fnvlist_add_uint64(token_nv, "offset", val);
1711 1714 }
1712 1715 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1713 1716 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1714 1717 fnvlist_add_uint64(token_nv, "bytes", val);
1715 1718 }
1716 1719 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1717 1720 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1718 1721 fnvlist_add_uint64(token_nv, "toguid", val);
1719 1722 }
1720 1723 char buf[256];
1721 1724 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1722 1725 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1723 1726 fnvlist_add_string(token_nv, "toname", buf);
1724 1727 }
1725 1728 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1726 1729 DS_FIELD_RESUME_EMBEDOK) == 0) {
1727 1730 fnvlist_add_boolean(token_nv, "embedok");
1728 1731 }
1729 1732 packed = fnvlist_pack(token_nv, &packed_size);
1730 1733 fnvlist_free(token_nv);
1731 1734 compressed = kmem_alloc(packed_size, KM_SLEEP);
1732 1735
1733 1736 compressed_size = gzip_compress(packed, compressed,
1734 1737 packed_size, packed_size, 6);
1735 1738
1736 1739 zio_cksum_t cksum;
1737 1740 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1738 1741
1739 1742 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1740 1743 for (int i = 0; i < compressed_size; i++) {
1741 1744 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1742 1745 }
1743 1746 str[compressed_size * 2] = '\0';
1744 1747 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1745 1748 ZFS_SEND_RESUME_TOKEN_VERSION,
1746 1749 (longlong_t)cksum.zc_word[0],
1747 1750 (longlong_t)packed_size, str);
1748 1751 dsl_prop_nvlist_add_string(nv,
1749 1752 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1750 1753 kmem_free(packed, packed_size);
1751 1754 kmem_free(str, compressed_size * 2 + 1);
1752 1755 kmem_free(compressed, packed_size);
1753 1756 strfree(propval);
1754 1757 }
1755 1758 }
1756 1759
1757 1760 void
1758 1761 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1759 1762 {
1760 1763 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1761 1764 uint64_t refd, avail, uobjs, aobjs, ratio;
1762 1765
1763 1766 ASSERT(dsl_pool_config_held(dp));
1764 1767
1765 1768 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1766 1769 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1767 1770 dsl_dataset_phys(ds)->ds_compressed_bytes);
1768 1771
1769 1772 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1770 1773 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1771 1774 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1772 1775
1773 1776 if (ds->ds_is_snapshot) {
1774 1777 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1775 1778 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1776 1779 dsl_dataset_phys(ds)->ds_unique_bytes);
1777 1780 get_clones_stat(ds, nv);
1778 1781 } else {
1779 1782 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1780 1783 char buf[MAXNAMELEN];
1781 1784 dsl_dataset_name(ds->ds_prev, buf);
1782 1785 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1783 1786 }
1784 1787
1785 1788 dsl_dir_stats(ds->ds_dir, nv);
1786 1789 }
1787 1790
1788 1791 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1789 1792 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1790 1793 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1791 1794
1792 1795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1793 1796 dsl_dataset_phys(ds)->ds_creation_time);
1794 1797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1795 1798 dsl_dataset_phys(ds)->ds_creation_txg);
1796 1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1797 1800 ds->ds_quota);
1798 1801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1799 1802 ds->ds_reserved);
1800 1803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1801 1804 dsl_dataset_phys(ds)->ds_guid);
1802 1805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1803 1806 dsl_dataset_phys(ds)->ds_unique_bytes);
1804 1807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1805 1808 ds->ds_object);
1806 1809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1807 1810 ds->ds_userrefs);
1808 1811 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1809 1812 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1810 1813
1811 1814 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1812 1815 uint64_t written, comp, uncomp;
1813 1816 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1814 1817 dsl_dataset_t *prev;
1815 1818
1816 1819 int err = dsl_dataset_hold_obj(dp,
1817 1820 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1818 1821 if (err == 0) {
1819 1822 err = dsl_dataset_space_written(prev, ds, &written,
1820 1823 &comp, &uncomp);
1821 1824 dsl_dataset_rele(prev, FTAG);
1822 1825 if (err == 0) {
1823 1826 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1824 1827 written);
1825 1828 }
1826 1829 }
1827 1830 }
1828 1831
1829 1832 if (!dsl_dataset_is_snapshot(ds)) {
1830 1833 /*
1831 1834 * A failed "newfs" (e.g. full) resumable receive leaves
1832 1835 * the stats set on this dataset. Check here for the prop.
1833 1836 */
1834 1837 get_receive_resume_stats(ds, nv);
1835 1838
1836 1839 /*
1837 1840 * A failed incremental resumable receive leaves the
1838 1841 * stats set on our child named "%recv". Check the child
1839 1842 * for the prop.
1840 1843 */
1841 1844 char recvname[ZFS_MAXNAMELEN];
1842 1845 dsl_dataset_t *recv_ds;
1843 1846 dsl_dataset_name(ds, recvname);
1844 1847 (void) strcat(recvname, "/");
1845 1848 (void) strcat(recvname, recv_clone_name);
1846 1849 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1847 1850 get_receive_resume_stats(recv_ds, nv);
1848 1851 dsl_dataset_rele(recv_ds, FTAG);
1849 1852 }
1850 1853 }
1851 1854 }
1852 1855
1853 1856 void
1854 1857 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1855 1858 {
1856 1859 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1857 1860 ASSERT(dsl_pool_config_held(dp));
1858 1861
1859 1862 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1860 1863 stat->dds_inconsistent =
1861 1864 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1862 1865 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1863 1866 stat->dds_origin[0] = '\0';
1864 1867 if (ds->ds_is_snapshot) {
1865 1868 stat->dds_is_snapshot = B_TRUE;
1866 1869 stat->dds_num_clones =
1867 1870 dsl_dataset_phys(ds)->ds_num_children - 1;
1868 1871 } else {
1869 1872 stat->dds_is_snapshot = B_FALSE;
1870 1873 stat->dds_num_clones = 0;
1871 1874
1872 1875 if (dsl_dir_is_clone(ds->ds_dir)) {
1873 1876 dsl_dataset_t *ods;
1874 1877
1875 1878 VERIFY0(dsl_dataset_hold_obj(dp,
1876 1879 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1877 1880 FTAG, &ods));
1878 1881 dsl_dataset_name(ods, stat->dds_origin);
1879 1882 dsl_dataset_rele(ods, FTAG);
1880 1883 }
1881 1884 }
1882 1885 }
1883 1886
1884 1887 uint64_t
1885 1888 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1886 1889 {
1887 1890 return (ds->ds_fsid_guid);
1888 1891 }
1889 1892
1890 1893 void
1891 1894 dsl_dataset_space(dsl_dataset_t *ds,
1892 1895 uint64_t *refdbytesp, uint64_t *availbytesp,
1893 1896 uint64_t *usedobjsp, uint64_t *availobjsp)
1894 1897 {
1895 1898 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1896 1899 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1897 1900 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1898 1901 *availbytesp +=
1899 1902 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1900 1903 if (ds->ds_quota != 0) {
1901 1904 /*
1902 1905 * Adjust available bytes according to refquota
1903 1906 */
1904 1907 if (*refdbytesp < ds->ds_quota)
1905 1908 *availbytesp = MIN(*availbytesp,
1906 1909 ds->ds_quota - *refdbytesp);
1907 1910 else
1908 1911 *availbytesp = 0;
1909 1912 }
1910 1913 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1911 1914 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1912 1915 }
1913 1916
1914 1917 boolean_t
1915 1918 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1916 1919 {
1917 1920 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1918 1921
1919 1922 ASSERT(dsl_pool_config_held(dp));
1920 1923 if (snap == NULL)
1921 1924 return (B_FALSE);
1922 1925 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1923 1926 dsl_dataset_phys(snap)->ds_creation_txg) {
1924 1927 objset_t *os, *os_snap;
1925 1928 /*
1926 1929 * It may be that only the ZIL differs, because it was
1927 1930 * reset in the head. Don't count that as being
1928 1931 * modified.
1929 1932 */
1930 1933 if (dmu_objset_from_ds(ds, &os) != 0)
1931 1934 return (B_TRUE);
1932 1935 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1933 1936 return (B_TRUE);
1934 1937 return (bcmp(&os->os_phys->os_meta_dnode,
1935 1938 &os_snap->os_phys->os_meta_dnode,
1936 1939 sizeof (os->os_phys->os_meta_dnode)) != 0);
1937 1940 }
1938 1941 return (B_FALSE);
1939 1942 }
1940 1943
1941 1944 typedef struct dsl_dataset_rename_snapshot_arg {
1942 1945 const char *ddrsa_fsname;
1943 1946 const char *ddrsa_oldsnapname;
1944 1947 const char *ddrsa_newsnapname;
1945 1948 boolean_t ddrsa_recursive;
1946 1949 dmu_tx_t *ddrsa_tx;
1947 1950 } dsl_dataset_rename_snapshot_arg_t;
1948 1951
1949 1952 /* ARGSUSED */
1950 1953 static int
1951 1954 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1952 1955 dsl_dataset_t *hds, void *arg)
1953 1956 {
1954 1957 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1955 1958 int error;
1956 1959 uint64_t val;
1957 1960
1958 1961 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1959 1962 if (error != 0) {
1960 1963 /* ignore nonexistent snapshots */
1961 1964 return (error == ENOENT ? 0 : error);
1962 1965 }
1963 1966
1964 1967 /* new name should not exist */
1965 1968 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1966 1969 if (error == 0)
1967 1970 error = SET_ERROR(EEXIST);
1968 1971 else if (error == ENOENT)
1969 1972 error = 0;
1970 1973
1971 1974 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1972 1975 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1973 1976 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1974 1977 error = SET_ERROR(ENAMETOOLONG);
1975 1978
1976 1979 return (error);
1977 1980 }
1978 1981
1979 1982 static int
1980 1983 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1981 1984 {
1982 1985 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1983 1986 dsl_pool_t *dp = dmu_tx_pool(tx);
1984 1987 dsl_dataset_t *hds;
1985 1988 int error;
1986 1989
1987 1990 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1988 1991 if (error != 0)
1989 1992 return (error);
1990 1993
1991 1994 if (ddrsa->ddrsa_recursive) {
1992 1995 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1993 1996 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1994 1997 DS_FIND_CHILDREN);
1995 1998 } else {
1996 1999 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1997 2000 }
1998 2001 dsl_dataset_rele(hds, FTAG);
1999 2002 return (error);
2000 2003 }
2001 2004
2002 2005 static int
2003 2006 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2004 2007 dsl_dataset_t *hds, void *arg)
2005 2008 {
2006 2009 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2007 2010 dsl_dataset_t *ds;
2008 2011 uint64_t val;
2009 2012 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2010 2013 int error;
2011 2014
2012 2015 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2013 2016 ASSERT(error == 0 || error == ENOENT);
2014 2017 if (error == ENOENT) {
2015 2018 /* ignore nonexistent snapshots */
2016 2019 return (0);
2017 2020 }
2018 2021
2019 2022 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2020 2023
2021 2024 /* log before we change the name */
2022 2025 spa_history_log_internal_ds(ds, "rename", tx,
2023 2026 "-> @%s", ddrsa->ddrsa_newsnapname);
2024 2027
2025 2028 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2026 2029 B_FALSE));
2027 2030 mutex_enter(&ds->ds_lock);
2028 2031 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2029 2032 mutex_exit(&ds->ds_lock);
2030 2033 VERIFY0(zap_add(dp->dp_meta_objset,
2031 2034 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2032 2035 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2033 2036
2034 2037 dsl_dataset_rele(ds, FTAG);
2035 2038 return (0);
2036 2039 }
2037 2040
2038 2041 static void
2039 2042 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2040 2043 {
2041 2044 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2042 2045 dsl_pool_t *dp = dmu_tx_pool(tx);
2043 2046 dsl_dataset_t *hds;
2044 2047
2045 2048 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2046 2049 ddrsa->ddrsa_tx = tx;
2047 2050 if (ddrsa->ddrsa_recursive) {
2048 2051 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2049 2052 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2050 2053 DS_FIND_CHILDREN));
2051 2054 } else {
2052 2055 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2053 2056 }
2054 2057 dsl_dataset_rele(hds, FTAG);
2055 2058 }
2056 2059
2057 2060 int
2058 2061 dsl_dataset_rename_snapshot(const char *fsname,
2059 2062 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2060 2063 {
2061 2064 dsl_dataset_rename_snapshot_arg_t ddrsa;
2062 2065
2063 2066 ddrsa.ddrsa_fsname = fsname;
2064 2067 ddrsa.ddrsa_oldsnapname = oldsnapname;
2065 2068 ddrsa.ddrsa_newsnapname = newsnapname;
2066 2069 ddrsa.ddrsa_recursive = recursive;
2067 2070
2068 2071 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2069 2072 dsl_dataset_rename_snapshot_sync, &ddrsa,
2070 2073 1, ZFS_SPACE_CHECK_RESERVED));
2071 2074 }
2072 2075
2073 2076 /*
2074 2077 * If we're doing an ownership handoff, we need to make sure that there is
2075 2078 * only one long hold on the dataset. We're not allowed to change anything here
2076 2079 * so we don't permanently release the long hold or regular hold here. We want
2077 2080 * to do this only when syncing to avoid the dataset unexpectedly going away
2078 2081 * when we release the long hold.
2079 2082 */
2080 2083 static int
2081 2084 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2082 2085 {
2083 2086 boolean_t held;
2084 2087
2085 2088 if (!dmu_tx_is_syncing(tx))
2086 2089 return (0);
2087 2090
2088 2091 if (owner != NULL) {
2089 2092 VERIFY3P(ds->ds_owner, ==, owner);
2090 2093 dsl_dataset_long_rele(ds, owner);
2091 2094 }
2092 2095
2093 2096 held = dsl_dataset_long_held(ds);
2094 2097
2095 2098 if (owner != NULL)
2096 2099 dsl_dataset_long_hold(ds, owner);
2097 2100
2098 2101 if (held)
2099 2102 return (SET_ERROR(EBUSY));
2100 2103
2101 2104 return (0);
2102 2105 }
2103 2106
2104 2107 typedef struct dsl_dataset_rollback_arg {
2105 2108 const char *ddra_fsname;
2106 2109 void *ddra_owner;
2107 2110 nvlist_t *ddra_result;
2108 2111 } dsl_dataset_rollback_arg_t;
2109 2112
2110 2113 static int
2111 2114 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2112 2115 {
2113 2116 dsl_dataset_rollback_arg_t *ddra = arg;
2114 2117 dsl_pool_t *dp = dmu_tx_pool(tx);
2115 2118 dsl_dataset_t *ds;
2116 2119 int64_t unused_refres_delta;
2117 2120 int error;
2118 2121
2119 2122 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2120 2123 if (error != 0)
2121 2124 return (error);
2122 2125
2123 2126 /* must not be a snapshot */
2124 2127 if (ds->ds_is_snapshot) {
2125 2128 dsl_dataset_rele(ds, FTAG);
2126 2129 return (SET_ERROR(EINVAL));
2127 2130 }
2128 2131
2129 2132 /* must have a most recent snapshot */
2130 2133 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2131 2134 dsl_dataset_rele(ds, FTAG);
2132 2135 return (SET_ERROR(EINVAL));
2133 2136 }
2134 2137
2135 2138 /* must not have any bookmarks after the most recent snapshot */
2136 2139 nvlist_t *proprequest = fnvlist_alloc();
2137 2140 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2138 2141 nvlist_t *bookmarks = fnvlist_alloc();
2139 2142 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2140 2143 fnvlist_free(proprequest);
2141 2144 if (error != 0)
2142 2145 return (error);
2143 2146 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2144 2147 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2145 2148 nvlist_t *valuenv =
2146 2149 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2147 2150 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2148 2151 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2149 2152 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2150 2153 fnvlist_free(bookmarks);
2151 2154 dsl_dataset_rele(ds, FTAG);
2152 2155 return (SET_ERROR(EEXIST));
2153 2156 }
2154 2157 }
2155 2158 fnvlist_free(bookmarks);
2156 2159
2157 2160 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2158 2161 if (error != 0) {
2159 2162 dsl_dataset_rele(ds, FTAG);
2160 2163 return (error);
2161 2164 }
2162 2165
2163 2166 /*
2164 2167 * Check if the snap we are rolling back to uses more than
2165 2168 * the refquota.
2166 2169 */
2167 2170 if (ds->ds_quota != 0 &&
2168 2171 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2169 2172 dsl_dataset_rele(ds, FTAG);
2170 2173 return (SET_ERROR(EDQUOT));
2171 2174 }
2172 2175
2173 2176 /*
2174 2177 * When we do the clone swap, we will temporarily use more space
2175 2178 * due to the refreservation (the head will no longer have any
2176 2179 * unique space, so the entire amount of the refreservation will need
2177 2180 * to be free). We will immediately destroy the clone, freeing
2178 2181 * this space, but the freeing happens over many txg's.
2179 2182 */
2180 2183 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2181 2184 dsl_dataset_phys(ds)->ds_unique_bytes);
2182 2185
2183 2186 if (unused_refres_delta > 0 &&
2184 2187 unused_refres_delta >
2185 2188 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2186 2189 dsl_dataset_rele(ds, FTAG);
2187 2190 return (SET_ERROR(ENOSPC));
2188 2191 }
2189 2192
2190 2193 dsl_dataset_rele(ds, FTAG);
2191 2194 return (0);
2192 2195 }
2193 2196
2194 2197 static void
2195 2198 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2196 2199 {
2197 2200 dsl_dataset_rollback_arg_t *ddra = arg;
2198 2201 dsl_pool_t *dp = dmu_tx_pool(tx);
2199 2202 dsl_dataset_t *ds, *clone;
2200 2203 uint64_t cloneobj;
2201 2204 char namebuf[ZFS_MAXNAMELEN];
2202 2205
2203 2206 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2204 2207
2205 2208 dsl_dataset_name(ds->ds_prev, namebuf);
2206 2209 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2207 2210
2208 2211 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2209 2212 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2210 2213
2211 2214 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2212 2215
2213 2216 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2214 2217 dsl_dataset_zero_zil(ds, tx);
2215 2218
2216 2219 dsl_destroy_head_sync_impl(clone, tx);
2217 2220
2218 2221 dsl_dataset_rele(clone, FTAG);
2219 2222 dsl_dataset_rele(ds, FTAG);
2220 2223 }
2221 2224
2222 2225 /*
2223 2226 * Rolls back the given filesystem or volume to the most recent snapshot.
2224 2227 * The name of the most recent snapshot will be returned under key "target"
2225 2228 * in the result nvlist.
2226 2229 *
2227 2230 * If owner != NULL:
2228 2231 * - The existing dataset MUST be owned by the specified owner at entry
2229 2232 * - Upon return, dataset will still be held by the same owner, whether we
2230 2233 * succeed or not.
2231 2234 *
2232 2235 * This mode is required any time the existing filesystem is mounted. See
2233 2236 * notes above zfs_suspend_fs() for further details.
2234 2237 */
2235 2238 int
2236 2239 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2237 2240 {
2238 2241 dsl_dataset_rollback_arg_t ddra;
2239 2242
2240 2243 ddra.ddra_fsname = fsname;
2241 2244 ddra.ddra_owner = owner;
2242 2245 ddra.ddra_result = result;
2243 2246
2244 2247 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2245 2248 dsl_dataset_rollback_sync, &ddra,
2246 2249 1, ZFS_SPACE_CHECK_RESERVED));
2247 2250 }
2248 2251
2249 2252 struct promotenode {
2250 2253 list_node_t link;
2251 2254 dsl_dataset_t *ds;
2252 2255 };
2253 2256
2254 2257 typedef struct dsl_dataset_promote_arg {
2255 2258 const char *ddpa_clonename;
2256 2259 dsl_dataset_t *ddpa_clone;
2257 2260 list_t shared_snaps, origin_snaps, clone_snaps;
2258 2261 dsl_dataset_t *origin_origin; /* origin of the origin */
2259 2262 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2260 2263 char *err_ds;
2261 2264 cred_t *cr;
2262 2265 } dsl_dataset_promote_arg_t;
2263 2266
2264 2267 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2265 2268 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2266 2269 void *tag);
2267 2270 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2268 2271
2269 2272 static int
2270 2273 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2271 2274 {
2272 2275 dsl_dataset_promote_arg_t *ddpa = arg;
2273 2276 dsl_pool_t *dp = dmu_tx_pool(tx);
2274 2277 dsl_dataset_t *hds;
2275 2278 struct promotenode *snap;
2276 2279 dsl_dataset_t *origin_ds;
2277 2280 int err;
2278 2281 uint64_t unused;
2279 2282 uint64_t ss_mv_cnt;
2280 2283 size_t max_snap_len;
2281 2284
2282 2285 err = promote_hold(ddpa, dp, FTAG);
2283 2286 if (err != 0)
2284 2287 return (err);
2285 2288
2286 2289 hds = ddpa->ddpa_clone;
2287 2290 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2288 2291
2289 2292 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2290 2293 promote_rele(ddpa, FTAG);
2291 2294 return (SET_ERROR(EXDEV));
2292 2295 }
2293 2296
2294 2297 /*
2295 2298 * Compute and check the amount of space to transfer. Since this is
2296 2299 * so expensive, don't do the preliminary check.
2297 2300 */
2298 2301 if (!dmu_tx_is_syncing(tx)) {
2299 2302 promote_rele(ddpa, FTAG);
2300 2303 return (0);
2301 2304 }
2302 2305
2303 2306 snap = list_head(&ddpa->shared_snaps);
2304 2307 origin_ds = snap->ds;
2305 2308
2306 2309 /* compute origin's new unique space */
2307 2310 snap = list_tail(&ddpa->clone_snaps);
2308 2311 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2309 2312 origin_ds->ds_object);
2310 2313 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2311 2314 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2312 2315 &ddpa->unique, &unused, &unused);
2313 2316
2314 2317 /*
2315 2318 * Walk the snapshots that we are moving
2316 2319 *
2317 2320 * Compute space to transfer. Consider the incremental changes
2318 2321 * to used by each snapshot:
2319 2322 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2320 2323 * So each snapshot gave birth to:
2321 2324 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2322 2325 * So a sequence would look like:
2323 2326 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2324 2327 * Which simplifies to:
2325 2328 * uN + kN + kN-1 + ... + k1 + k0
2326 2329 * Note however, if we stop before we reach the ORIGIN we get:
2327 2330 * uN + kN + kN-1 + ... + kM - uM-1
2328 2331 */
2329 2332 ss_mv_cnt = 0;
2330 2333 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2331 2334 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2332 2335 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2333 2336 for (snap = list_head(&ddpa->shared_snaps); snap;
2334 2337 snap = list_next(&ddpa->shared_snaps, snap)) {
2335 2338 uint64_t val, dlused, dlcomp, dluncomp;
2336 2339 dsl_dataset_t *ds = snap->ds;
2337 2340
2338 2341 ss_mv_cnt++;
2339 2342
2340 2343 /*
2341 2344 * If there are long holds, we won't be able to evict
2342 2345 * the objset.
2343 2346 */
2344 2347 if (dsl_dataset_long_held(ds)) {
2345 2348 err = SET_ERROR(EBUSY);
2346 2349 goto out;
2347 2350 }
2348 2351
2349 2352 /* Check that the snapshot name does not conflict */
2350 2353 VERIFY0(dsl_dataset_get_snapname(ds));
2351 2354 if (strlen(ds->ds_snapname) >= max_snap_len) {
2352 2355 err = SET_ERROR(ENAMETOOLONG);
2353 2356 goto out;
2354 2357 }
2355 2358 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2356 2359 if (err == 0) {
2357 2360 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2358 2361 err = SET_ERROR(EEXIST);
2359 2362 goto out;
2360 2363 }
2361 2364 if (err != ENOENT)
2362 2365 goto out;
2363 2366
2364 2367 /* The very first snapshot does not have a deadlist */
2365 2368 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2366 2369 continue;
2367 2370
2368 2371 dsl_deadlist_space(&ds->ds_deadlist,
2369 2372 &dlused, &dlcomp, &dluncomp);
2370 2373 ddpa->used += dlused;
2371 2374 ddpa->comp += dlcomp;
2372 2375 ddpa->uncomp += dluncomp;
2373 2376 }
2374 2377
2375 2378 /*
2376 2379 * If we are a clone of a clone then we never reached ORIGIN,
2377 2380 * so we need to subtract out the clone origin's used space.
2378 2381 */
2379 2382 if (ddpa->origin_origin) {
2380 2383 ddpa->used -=
2381 2384 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2382 2385 ddpa->comp -=
2383 2386 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2384 2387 ddpa->uncomp -=
2385 2388 dsl_dataset_phys(ddpa->origin_origin)->
2386 2389 ds_uncompressed_bytes;
2387 2390 }
2388 2391
2389 2392 /* Check that there is enough space and limit headroom here */
2390 2393 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2391 2394 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2392 2395 if (err != 0)
2393 2396 goto out;
2394 2397
2395 2398 /*
2396 2399 * Compute the amounts of space that will be used by snapshots
2397 2400 * after the promotion (for both origin and clone). For each,
2398 2401 * it is the amount of space that will be on all of their
2399 2402 * deadlists (that was not born before their new origin).
2400 2403 */
2401 2404 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2402 2405 uint64_t space;
2403 2406
2404 2407 /*
2405 2408 * Note, typically this will not be a clone of a clone,
2406 2409 * so dd_origin_txg will be < TXG_INITIAL, so
2407 2410 * these snaplist_space() -> dsl_deadlist_space_range()
2408 2411 * calls will be fast because they do not have to
2409 2412 * iterate over all bps.
2410 2413 */
2411 2414 snap = list_head(&ddpa->origin_snaps);
2412 2415 err = snaplist_space(&ddpa->shared_snaps,
2413 2416 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2414 2417 if (err != 0)
2415 2418 goto out;
2416 2419
2417 2420 err = snaplist_space(&ddpa->clone_snaps,
2418 2421 snap->ds->ds_dir->dd_origin_txg, &space);
2419 2422 if (err != 0)
2420 2423 goto out;
2421 2424 ddpa->cloneusedsnap += space;
2422 2425 }
2423 2426 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2424 2427 DD_FLAG_USED_BREAKDOWN) {
2425 2428 err = snaplist_space(&ddpa->origin_snaps,
2426 2429 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2427 2430 &ddpa->originusedsnap);
2428 2431 if (err != 0)
2429 2432 goto out;
2430 2433 }
2431 2434
2432 2435 out:
2433 2436 promote_rele(ddpa, FTAG);
2434 2437 return (err);
2435 2438 }
2436 2439
2437 2440 static void
2438 2441 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2439 2442 {
2440 2443 dsl_dataset_promote_arg_t *ddpa = arg;
2441 2444 dsl_pool_t *dp = dmu_tx_pool(tx);
2442 2445 dsl_dataset_t *hds;
2443 2446 struct promotenode *snap;
2444 2447 dsl_dataset_t *origin_ds;
2445 2448 dsl_dataset_t *origin_head;
2446 2449 dsl_dir_t *dd;
2447 2450 dsl_dir_t *odd = NULL;
2448 2451 uint64_t oldnext_obj;
2449 2452 int64_t delta;
2450 2453
2451 2454 VERIFY0(promote_hold(ddpa, dp, FTAG));
2452 2455 hds = ddpa->ddpa_clone;
2453 2456
2454 2457 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2455 2458
2456 2459 snap = list_head(&ddpa->shared_snaps);
2457 2460 origin_ds = snap->ds;
2458 2461 dd = hds->ds_dir;
2459 2462
2460 2463 snap = list_head(&ddpa->origin_snaps);
2461 2464 origin_head = snap->ds;
2462 2465
2463 2466 /*
2464 2467 * We need to explicitly open odd, since origin_ds's dd will be
2465 2468 * changing.
2466 2469 */
2467 2470 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2468 2471 NULL, FTAG, &odd));
2469 2472
2470 2473 /* change origin's next snap */
2471 2474 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2472 2475 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2473 2476 snap = list_tail(&ddpa->clone_snaps);
2474 2477 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2475 2478 origin_ds->ds_object);
2476 2479 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2477 2480
2478 2481 /* change the origin's next clone */
2479 2482 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2480 2483 dsl_dataset_remove_from_next_clones(origin_ds,
2481 2484 snap->ds->ds_object, tx);
2482 2485 VERIFY0(zap_add_int(dp->dp_meta_objset,
2483 2486 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2484 2487 oldnext_obj, tx));
2485 2488 }
2486 2489
2487 2490 /* change origin */
2488 2491 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2489 2492 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2490 2493 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2491 2494 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2492 2495 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2493 2496 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2494 2497 origin_head->ds_dir->dd_origin_txg =
2495 2498 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2496 2499
2497 2500 /* change dd_clone entries */
2498 2501 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2499 2502 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2500 2503 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2501 2504 VERIFY0(zap_add_int(dp->dp_meta_objset,
2502 2505 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2503 2506 hds->ds_object, tx));
2504 2507
2505 2508 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2506 2509 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2507 2510 origin_head->ds_object, tx));
2508 2511 if (dsl_dir_phys(dd)->dd_clones == 0) {
2509 2512 dsl_dir_phys(dd)->dd_clones =
2510 2513 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2511 2514 DMU_OT_NONE, 0, tx);
2512 2515 }
2513 2516 VERIFY0(zap_add_int(dp->dp_meta_objset,
2514 2517 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2515 2518 }
2516 2519
2517 2520 /* move snapshots to this dir */
2518 2521 for (snap = list_head(&ddpa->shared_snaps); snap;
2519 2522 snap = list_next(&ddpa->shared_snaps, snap)) {
2520 2523 dsl_dataset_t *ds = snap->ds;
2521 2524
2522 2525 /*
2523 2526 * Property callbacks are registered to a particular
2524 2527 * dsl_dir. Since ours is changing, evict the objset
2525 2528 * so that they will be unregistered from the old dsl_dir.
2526 2529 */
2527 2530 if (ds->ds_objset) {
2528 2531 dmu_objset_evict(ds->ds_objset);
2529 2532 ds->ds_objset = NULL;
2530 2533 }
2531 2534
2532 2535 /* move snap name entry */
2533 2536 VERIFY0(dsl_dataset_get_snapname(ds));
2534 2537 VERIFY0(dsl_dataset_snap_remove(origin_head,
2535 2538 ds->ds_snapname, tx, B_TRUE));
2536 2539 VERIFY0(zap_add(dp->dp_meta_objset,
2537 2540 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2538 2541 8, 1, &ds->ds_object, tx));
2539 2542 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2540 2543 DD_FIELD_SNAPSHOT_COUNT, tx);
2541 2544
2542 2545 /* change containing dsl_dir */
2543 2546 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2544 2547 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2545 2548 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2546 2549 ASSERT3P(ds->ds_dir, ==, odd);
2547 2550 dsl_dir_rele(ds->ds_dir, ds);
2548 2551 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2549 2552 NULL, ds, &ds->ds_dir));
2550 2553
2551 2554 /* move any clone references */
2552 2555 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2553 2556 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2554 2557 zap_cursor_t zc;
2555 2558 zap_attribute_t za;
2556 2559
2557 2560 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2558 2561 dsl_dataset_phys(ds)->ds_next_clones_obj);
2559 2562 zap_cursor_retrieve(&zc, &za) == 0;
2560 2563 zap_cursor_advance(&zc)) {
2561 2564 dsl_dataset_t *cnds;
2562 2565 uint64_t o;
2563 2566
2564 2567 if (za.za_first_integer == oldnext_obj) {
2565 2568 /*
2566 2569 * We've already moved the
2567 2570 * origin's reference.
2568 2571 */
2569 2572 continue;
2570 2573 }
2571 2574
2572 2575 VERIFY0(dsl_dataset_hold_obj(dp,
2573 2576 za.za_first_integer, FTAG, &cnds));
2574 2577 o = dsl_dir_phys(cnds->ds_dir)->
2575 2578 dd_head_dataset_obj;
2576 2579
2577 2580 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2578 2581 dsl_dir_phys(odd)->dd_clones, o, tx));
2579 2582 VERIFY0(zap_add_int(dp->dp_meta_objset,
2580 2583 dsl_dir_phys(dd)->dd_clones, o, tx));
2581 2584 dsl_dataset_rele(cnds, FTAG);
2582 2585 }
2583 2586 zap_cursor_fini(&zc);
2584 2587 }
2585 2588
2586 2589 ASSERT(!dsl_prop_hascb(ds));
2587 2590 }
2588 2591
2589 2592 /*
2590 2593 * Change space accounting.
2591 2594 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2592 2595 * both be valid, or both be 0 (resulting in delta == 0). This
2593 2596 * is true for each of {clone,origin} independently.
2594 2597 */
2595 2598
2596 2599 delta = ddpa->cloneusedsnap -
2597 2600 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2598 2601 ASSERT3S(delta, >=, 0);
2599 2602 ASSERT3U(ddpa->used, >=, delta);
2600 2603 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2601 2604 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2602 2605 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2603 2606
2604 2607 delta = ddpa->originusedsnap -
2605 2608 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2606 2609 ASSERT3S(delta, <=, 0);
2607 2610 ASSERT3U(ddpa->used, >=, -delta);
2608 2611 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2609 2612 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2610 2613 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2611 2614
2612 2615 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2613 2616
2614 2617 /* log history record */
2615 2618 spa_history_log_internal_ds(hds, "promote", tx, "");
2616 2619
2617 2620 dsl_dir_rele(odd, FTAG);
2618 2621 promote_rele(ddpa, FTAG);
2619 2622 }
2620 2623
2621 2624 /*
2622 2625 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2623 2626 * (exclusive) and last_obj (inclusive). The list will be in reverse
2624 2627 * order (last_obj will be the list_head()). If first_obj == 0, do all
2625 2628 * snapshots back to this dataset's origin.
2626 2629 */
2627 2630 static int
2628 2631 snaplist_make(dsl_pool_t *dp,
2629 2632 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2630 2633 {
2631 2634 uint64_t obj = last_obj;
2632 2635
2633 2636 list_create(l, sizeof (struct promotenode),
2634 2637 offsetof(struct promotenode, link));
2635 2638
2636 2639 while (obj != first_obj) {
2637 2640 dsl_dataset_t *ds;
2638 2641 struct promotenode *snap;
2639 2642 int err;
2640 2643
2641 2644 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2642 2645 ASSERT(err != ENOENT);
2643 2646 if (err != 0)
2644 2647 return (err);
2645 2648
2646 2649 if (first_obj == 0)
2647 2650 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2648 2651
2649 2652 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2650 2653 snap->ds = ds;
2651 2654 list_insert_tail(l, snap);
2652 2655 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2653 2656 }
2654 2657
2655 2658 return (0);
2656 2659 }
2657 2660
2658 2661 static int
2659 2662 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2660 2663 {
2661 2664 struct promotenode *snap;
2662 2665
2663 2666 *spacep = 0;
2664 2667 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2665 2668 uint64_t used, comp, uncomp;
2666 2669 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2667 2670 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2668 2671 *spacep += used;
2669 2672 }
2670 2673 return (0);
2671 2674 }
2672 2675
2673 2676 static void
2674 2677 snaplist_destroy(list_t *l, void *tag)
2675 2678 {
2676 2679 struct promotenode *snap;
2677 2680
2678 2681 if (l == NULL || !list_link_active(&l->list_head))
2679 2682 return;
2680 2683
2681 2684 while ((snap = list_tail(l)) != NULL) {
2682 2685 list_remove(l, snap);
2683 2686 dsl_dataset_rele(snap->ds, tag);
2684 2687 kmem_free(snap, sizeof (*snap));
2685 2688 }
2686 2689 list_destroy(l);
2687 2690 }
2688 2691
2689 2692 static int
2690 2693 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2691 2694 {
2692 2695 int error;
2693 2696 dsl_dir_t *dd;
2694 2697 struct promotenode *snap;
2695 2698
2696 2699 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2697 2700 &ddpa->ddpa_clone);
2698 2701 if (error != 0)
2699 2702 return (error);
2700 2703 dd = ddpa->ddpa_clone->ds_dir;
2701 2704
2702 2705 if (ddpa->ddpa_clone->ds_is_snapshot ||
2703 2706 !dsl_dir_is_clone(dd)) {
2704 2707 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2705 2708 return (SET_ERROR(EINVAL));
2706 2709 }
2707 2710
2708 2711 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2709 2712 &ddpa->shared_snaps, tag);
2710 2713 if (error != 0)
2711 2714 goto out;
2712 2715
2713 2716 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2714 2717 &ddpa->clone_snaps, tag);
2715 2718 if (error != 0)
2716 2719 goto out;
2717 2720
2718 2721 snap = list_head(&ddpa->shared_snaps);
2719 2722 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2720 2723 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2721 2724 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2722 2725 &ddpa->origin_snaps, tag);
2723 2726 if (error != 0)
2724 2727 goto out;
2725 2728
2726 2729 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2727 2730 error = dsl_dataset_hold_obj(dp,
2728 2731 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2729 2732 tag, &ddpa->origin_origin);
2730 2733 if (error != 0)
2731 2734 goto out;
2732 2735 }
2733 2736 out:
2734 2737 if (error != 0)
2735 2738 promote_rele(ddpa, tag);
2736 2739 return (error);
2737 2740 }
2738 2741
2739 2742 static void
2740 2743 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2741 2744 {
2742 2745 snaplist_destroy(&ddpa->shared_snaps, tag);
2743 2746 snaplist_destroy(&ddpa->clone_snaps, tag);
2744 2747 snaplist_destroy(&ddpa->origin_snaps, tag);
2745 2748 if (ddpa->origin_origin != NULL)
2746 2749 dsl_dataset_rele(ddpa->origin_origin, tag);
2747 2750 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2748 2751 }
2749 2752
2750 2753 /*
2751 2754 * Promote a clone.
2752 2755 *
2753 2756 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2754 2757 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2755 2758 */
2756 2759 int
2757 2760 dsl_dataset_promote(const char *name, char *conflsnap)
2758 2761 {
2759 2762 dsl_dataset_promote_arg_t ddpa = { 0 };
2760 2763 uint64_t numsnaps;
2761 2764 int error;
2762 2765 objset_t *os;
2763 2766
2764 2767 /*
2765 2768 * We will modify space proportional to the number of
2766 2769 * snapshots. Compute numsnaps.
2767 2770 */
2768 2771 error = dmu_objset_hold(name, FTAG, &os);
2769 2772 if (error != 0)
2770 2773 return (error);
2771 2774 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2772 2775 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2773 2776 &numsnaps);
2774 2777 dmu_objset_rele(os, FTAG);
2775 2778 if (error != 0)
2776 2779 return (error);
2777 2780
2778 2781 ddpa.ddpa_clonename = name;
2779 2782 ddpa.err_ds = conflsnap;
2780 2783 ddpa.cr = CRED();
|
↓ open down ↓ |
2690 lines elided |
↑ open up ↑ |
2781 2784
2782 2785 return (dsl_sync_task(name, dsl_dataset_promote_check,
2783 2786 dsl_dataset_promote_sync, &ddpa,
2784 2787 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2785 2788 }
2786 2789
2787 2790 int
2788 2791 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2789 2792 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2790 2793 {
2794 + /*
2795 + * "slack" factor for received datasets with refquota set on them.
2796 + * See the bottom of this function for details on its use.
2797 + */
2798 + uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
2791 2799 int64_t unused_refres_delta;
2792 2800
2793 2801 /* they should both be heads */
2794 2802 if (clone->ds_is_snapshot ||
2795 2803 origin_head->ds_is_snapshot)
2796 2804 return (SET_ERROR(EINVAL));
2797 2805
2798 2806 /* if we are not forcing, the branch point should be just before them */
2799 2807 if (!force && clone->ds_prev != origin_head->ds_prev)
2800 2808 return (SET_ERROR(EINVAL));
2801 2809
2802 2810 /* clone should be the clone (unless they are unrelated) */
2803 2811 if (clone->ds_prev != NULL &&
2804 2812 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2805 2813 origin_head->ds_dir != clone->ds_prev->ds_dir)
2806 2814 return (SET_ERROR(EINVAL));
2807 2815
2808 2816 /* the clone should be a child of the origin */
2809 2817 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2810 2818 return (SET_ERROR(EINVAL));
2811 2819
2812 2820 /* origin_head shouldn't be modified unless 'force' */
2813 2821 if (!force &&
2814 2822 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2815 2823 return (SET_ERROR(ETXTBSY));
2816 2824
2817 2825 /* origin_head should have no long holds (e.g. is not mounted) */
2818 2826 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2819 2827 return (SET_ERROR(EBUSY));
2820 2828
2821 2829 /* check amount of any unconsumed refreservation */
2822 2830 unused_refres_delta =
|
↓ open down ↓ |
22 lines elided |
↑ open up ↑ |
2823 2831 (int64_t)MIN(origin_head->ds_reserved,
2824 2832 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2825 2833 (int64_t)MIN(origin_head->ds_reserved,
2826 2834 dsl_dataset_phys(clone)->ds_unique_bytes);
2827 2835
2828 2836 if (unused_refres_delta > 0 &&
2829 2837 unused_refres_delta >
2830 2838 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2831 2839 return (SET_ERROR(ENOSPC));
2832 2840
2833 - /* clone can't be over the head's refquota */
2841 + /*
2842 + * The clone can't be too much over the head's refquota.
2843 + *
2844 + * To ensure that the entire refquota can be used, we allow one
2845 + * transaction to exceed the the refquota. Therefore, this check
2846 + * needs to also allow for the space referenced to be more than the
2847 + * refquota. The maximum amount of space that one transaction can use
2848 + * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
2849 + * overage ensures that we are able to receive a filesystem that
2850 + * exceeds the refquota on the source system.
2851 + *
2852 + * So that overage is the refquota_slack we use below.
2853 + */
2834 2854 if (origin_head->ds_quota != 0 &&
2835 2855 dsl_dataset_phys(clone)->ds_referenced_bytes >
2836 - origin_head->ds_quota)
2856 + origin_head->ds_quota + refquota_slack)
2837 2857 return (SET_ERROR(EDQUOT));
2838 2858
2839 2859 return (0);
2840 2860 }
2841 2861
2842 2862 void
2843 2863 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2844 2864 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2845 2865 {
2846 2866 dsl_pool_t *dp = dmu_tx_pool(tx);
2847 2867 int64_t unused_refres_delta;
2848 2868
2849 2869 ASSERT(clone->ds_reserved == 0);
2870 + /*
2871 + * NOTE: On DEBUG kernels there could be a race between this and
2872 + * the check function if spa_asize_inflation is adjusted...
2873 + */
2850 2874 ASSERT(origin_head->ds_quota == 0 ||
2851 - dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2875 + dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
2876 + DMU_MAX_ACCESS * spa_asize_inflation);
2852 2877 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2853 2878
2854 2879 /*
2855 2880 * Swap per-dataset feature flags.
2856 2881 */
2857 2882 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2858 2883 if (!(spa_feature_table[f].fi_flags &
2859 2884 ZFEATURE_FLAG_PER_DATASET)) {
2860 2885 ASSERT(!clone->ds_feature_inuse[f]);
2861 2886 ASSERT(!origin_head->ds_feature_inuse[f]);
2862 2887 continue;
2863 2888 }
2864 2889
2865 2890 boolean_t clone_inuse = clone->ds_feature_inuse[f];
2866 2891 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2867 2892
2868 2893 if (clone_inuse) {
2869 2894 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2870 2895 clone->ds_feature_inuse[f] = B_FALSE;
2871 2896 }
2872 2897 if (origin_head_inuse) {
2873 2898 dsl_dataset_deactivate_feature(origin_head->ds_object,
2874 2899 f, tx);
2875 2900 origin_head->ds_feature_inuse[f] = B_FALSE;
2876 2901 }
2877 2902 if (clone_inuse) {
2878 2903 dsl_dataset_activate_feature(origin_head->ds_object,
2879 2904 f, tx);
2880 2905 origin_head->ds_feature_inuse[f] = B_TRUE;
2881 2906 }
2882 2907 if (origin_head_inuse) {
2883 2908 dsl_dataset_activate_feature(clone->ds_object, f, tx);
2884 2909 clone->ds_feature_inuse[f] = B_TRUE;
2885 2910 }
2886 2911 }
2887 2912
2888 2913 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2889 2914 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2890 2915
2891 2916 if (clone->ds_objset != NULL) {
2892 2917 dmu_objset_evict(clone->ds_objset);
2893 2918 clone->ds_objset = NULL;
2894 2919 }
2895 2920
2896 2921 if (origin_head->ds_objset != NULL) {
2897 2922 dmu_objset_evict(origin_head->ds_objset);
2898 2923 origin_head->ds_objset = NULL;
2899 2924 }
2900 2925
2901 2926 unused_refres_delta =
2902 2927 (int64_t)MIN(origin_head->ds_reserved,
2903 2928 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2904 2929 (int64_t)MIN(origin_head->ds_reserved,
2905 2930 dsl_dataset_phys(clone)->ds_unique_bytes);
2906 2931
2907 2932 /*
2908 2933 * Reset origin's unique bytes, if it exists.
2909 2934 */
2910 2935 if (clone->ds_prev) {
2911 2936 dsl_dataset_t *origin = clone->ds_prev;
2912 2937 uint64_t comp, uncomp;
2913 2938
2914 2939 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2915 2940 dsl_deadlist_space_range(&clone->ds_deadlist,
2916 2941 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2917 2942 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2918 2943 }
2919 2944
2920 2945 /* swap blkptrs */
2921 2946 {
2922 2947 blkptr_t tmp;
2923 2948 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2924 2949 dsl_dataset_phys(origin_head)->ds_bp =
2925 2950 dsl_dataset_phys(clone)->ds_bp;
2926 2951 dsl_dataset_phys(clone)->ds_bp = tmp;
2927 2952 }
2928 2953
2929 2954 /* set dd_*_bytes */
2930 2955 {
2931 2956 int64_t dused, dcomp, duncomp;
2932 2957 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2933 2958 uint64_t odl_used, odl_comp, odl_uncomp;
2934 2959
2935 2960 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2936 2961 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2937 2962
2938 2963 dsl_deadlist_space(&clone->ds_deadlist,
2939 2964 &cdl_used, &cdl_comp, &cdl_uncomp);
2940 2965 dsl_deadlist_space(&origin_head->ds_deadlist,
2941 2966 &odl_used, &odl_comp, &odl_uncomp);
2942 2967
2943 2968 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2944 2969 cdl_used -
2945 2970 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2946 2971 odl_used);
2947 2972 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2948 2973 cdl_comp -
2949 2974 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2950 2975 odl_comp);
2951 2976 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2952 2977 cdl_uncomp -
2953 2978 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2954 2979 odl_uncomp);
2955 2980
2956 2981 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2957 2982 dused, dcomp, duncomp, tx);
2958 2983 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2959 2984 -dused, -dcomp, -duncomp, tx);
2960 2985
2961 2986 /*
2962 2987 * The difference in the space used by snapshots is the
2963 2988 * difference in snapshot space due to the head's
2964 2989 * deadlist (since that's the only thing that's
2965 2990 * changing that affects the snapused).
2966 2991 */
2967 2992 dsl_deadlist_space_range(&clone->ds_deadlist,
2968 2993 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2969 2994 &cdl_used, &cdl_comp, &cdl_uncomp);
2970 2995 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2971 2996 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2972 2997 &odl_used, &odl_comp, &odl_uncomp);
2973 2998 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2974 2999 DD_USED_HEAD, DD_USED_SNAP, tx);
2975 3000 }
2976 3001
2977 3002 /* swap ds_*_bytes */
2978 3003 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2979 3004 dsl_dataset_phys(clone)->ds_referenced_bytes);
2980 3005 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2981 3006 dsl_dataset_phys(clone)->ds_compressed_bytes);
2982 3007 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2983 3008 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2984 3009 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2985 3010 dsl_dataset_phys(clone)->ds_unique_bytes);
2986 3011
2987 3012 /* apply any parent delta for change in unconsumed refreservation */
2988 3013 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2989 3014 unused_refres_delta, 0, 0, tx);
2990 3015
2991 3016 /*
2992 3017 * Swap deadlists.
2993 3018 */
2994 3019 dsl_deadlist_close(&clone->ds_deadlist);
2995 3020 dsl_deadlist_close(&origin_head->ds_deadlist);
2996 3021 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2997 3022 dsl_dataset_phys(clone)->ds_deadlist_obj);
2998 3023 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2999 3024 dsl_dataset_phys(clone)->ds_deadlist_obj);
3000 3025 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3001 3026 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3002 3027
3003 3028 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3004 3029
3005 3030 spa_history_log_internal_ds(clone, "clone swap", tx,
3006 3031 "parent=%s", origin_head->ds_dir->dd_myname);
3007 3032 }
3008 3033
3009 3034 /*
3010 3035 * Given a pool name and a dataset object number in that pool,
3011 3036 * return the name of that dataset.
3012 3037 */
3013 3038 int
3014 3039 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3015 3040 {
3016 3041 dsl_pool_t *dp;
3017 3042 dsl_dataset_t *ds;
3018 3043 int error;
3019 3044
3020 3045 error = dsl_pool_hold(pname, FTAG, &dp);
3021 3046 if (error != 0)
3022 3047 return (error);
3023 3048
3024 3049 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3025 3050 if (error == 0) {
3026 3051 dsl_dataset_name(ds, buf);
3027 3052 dsl_dataset_rele(ds, FTAG);
3028 3053 }
3029 3054 dsl_pool_rele(dp, FTAG);
3030 3055
3031 3056 return (error);
3032 3057 }
3033 3058
3034 3059 int
3035 3060 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3036 3061 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3037 3062 {
3038 3063 int error = 0;
3039 3064
3040 3065 ASSERT3S(asize, >, 0);
3041 3066
3042 3067 /*
3043 3068 * *ref_rsrv is the portion of asize that will come from any
3044 3069 * unconsumed refreservation space.
3045 3070 */
3046 3071 *ref_rsrv = 0;
3047 3072
3048 3073 mutex_enter(&ds->ds_lock);
3049 3074 /*
3050 3075 * Make a space adjustment for reserved bytes.
3051 3076 */
3052 3077 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3053 3078 ASSERT3U(*used, >=,
3054 3079 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3055 3080 *used -=
3056 3081 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3057 3082 *ref_rsrv =
3058 3083 asize - MIN(asize, parent_delta(ds, asize + inflight));
3059 3084 }
3060 3085
3061 3086 if (!check_quota || ds->ds_quota == 0) {
3062 3087 mutex_exit(&ds->ds_lock);
3063 3088 return (0);
3064 3089 }
3065 3090 /*
3066 3091 * If they are requesting more space, and our current estimate
3067 3092 * is over quota, they get to try again unless the actual
3068 3093 * on-disk is over quota and there are no pending changes (which
3069 3094 * may free up space for us).
3070 3095 */
3071 3096 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3072 3097 ds->ds_quota) {
3073 3098 if (inflight > 0 ||
3074 3099 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3075 3100 error = SET_ERROR(ERESTART);
3076 3101 else
3077 3102 error = SET_ERROR(EDQUOT);
3078 3103 }
3079 3104 mutex_exit(&ds->ds_lock);
3080 3105
3081 3106 return (error);
3082 3107 }
3083 3108
3084 3109 typedef struct dsl_dataset_set_qr_arg {
3085 3110 const char *ddsqra_name;
3086 3111 zprop_source_t ddsqra_source;
3087 3112 uint64_t ddsqra_value;
3088 3113 } dsl_dataset_set_qr_arg_t;
3089 3114
3090 3115
3091 3116 /* ARGSUSED */
3092 3117 static int
3093 3118 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3094 3119 {
3095 3120 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3096 3121 dsl_pool_t *dp = dmu_tx_pool(tx);
3097 3122 dsl_dataset_t *ds;
3098 3123 int error;
3099 3124 uint64_t newval;
3100 3125
3101 3126 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3102 3127 return (SET_ERROR(ENOTSUP));
3103 3128
3104 3129 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3105 3130 if (error != 0)
3106 3131 return (error);
3107 3132
3108 3133 if (ds->ds_is_snapshot) {
3109 3134 dsl_dataset_rele(ds, FTAG);
3110 3135 return (SET_ERROR(EINVAL));
3111 3136 }
3112 3137
3113 3138 error = dsl_prop_predict(ds->ds_dir,
3114 3139 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3115 3140 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3116 3141 if (error != 0) {
3117 3142 dsl_dataset_rele(ds, FTAG);
3118 3143 return (error);
3119 3144 }
3120 3145
3121 3146 if (newval == 0) {
3122 3147 dsl_dataset_rele(ds, FTAG);
3123 3148 return (0);
3124 3149 }
3125 3150
3126 3151 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3127 3152 newval < ds->ds_reserved) {
3128 3153 dsl_dataset_rele(ds, FTAG);
3129 3154 return (SET_ERROR(ENOSPC));
3130 3155 }
3131 3156
3132 3157 dsl_dataset_rele(ds, FTAG);
3133 3158 return (0);
3134 3159 }
3135 3160
3136 3161 static void
3137 3162 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3138 3163 {
3139 3164 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3140 3165 dsl_pool_t *dp = dmu_tx_pool(tx);
3141 3166 dsl_dataset_t *ds;
3142 3167 uint64_t newval;
3143 3168
3144 3169 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3145 3170
3146 3171 dsl_prop_set_sync_impl(ds,
3147 3172 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3148 3173 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3149 3174 &ddsqra->ddsqra_value, tx);
3150 3175
3151 3176 VERIFY0(dsl_prop_get_int_ds(ds,
3152 3177 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3153 3178
3154 3179 if (ds->ds_quota != newval) {
3155 3180 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3156 3181 ds->ds_quota = newval;
3157 3182 }
3158 3183 dsl_dataset_rele(ds, FTAG);
3159 3184 }
3160 3185
3161 3186 int
3162 3187 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3163 3188 uint64_t refquota)
3164 3189 {
3165 3190 dsl_dataset_set_qr_arg_t ddsqra;
3166 3191
3167 3192 ddsqra.ddsqra_name = dsname;
3168 3193 ddsqra.ddsqra_source = source;
3169 3194 ddsqra.ddsqra_value = refquota;
3170 3195
3171 3196 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3172 3197 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3173 3198 }
3174 3199
3175 3200 static int
3176 3201 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3177 3202 {
3178 3203 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3179 3204 dsl_pool_t *dp = dmu_tx_pool(tx);
3180 3205 dsl_dataset_t *ds;
3181 3206 int error;
3182 3207 uint64_t newval, unique;
3183 3208
3184 3209 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3185 3210 return (SET_ERROR(ENOTSUP));
3186 3211
3187 3212 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3188 3213 if (error != 0)
3189 3214 return (error);
3190 3215
3191 3216 if (ds->ds_is_snapshot) {
3192 3217 dsl_dataset_rele(ds, FTAG);
3193 3218 return (SET_ERROR(EINVAL));
3194 3219 }
3195 3220
3196 3221 error = dsl_prop_predict(ds->ds_dir,
3197 3222 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3198 3223 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3199 3224 if (error != 0) {
3200 3225 dsl_dataset_rele(ds, FTAG);
3201 3226 return (error);
3202 3227 }
3203 3228
3204 3229 /*
3205 3230 * If we are doing the preliminary check in open context, the
3206 3231 * space estimates may be inaccurate.
3207 3232 */
3208 3233 if (!dmu_tx_is_syncing(tx)) {
3209 3234 dsl_dataset_rele(ds, FTAG);
3210 3235 return (0);
3211 3236 }
3212 3237
3213 3238 mutex_enter(&ds->ds_lock);
3214 3239 if (!DS_UNIQUE_IS_ACCURATE(ds))
3215 3240 dsl_dataset_recalc_head_uniq(ds);
3216 3241 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3217 3242 mutex_exit(&ds->ds_lock);
3218 3243
3219 3244 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3220 3245 uint64_t delta = MAX(unique, newval) -
3221 3246 MAX(unique, ds->ds_reserved);
3222 3247
3223 3248 if (delta >
3224 3249 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3225 3250 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3226 3251 dsl_dataset_rele(ds, FTAG);
3227 3252 return (SET_ERROR(ENOSPC));
3228 3253 }
3229 3254 }
3230 3255
3231 3256 dsl_dataset_rele(ds, FTAG);
3232 3257 return (0);
3233 3258 }
3234 3259
3235 3260 void
3236 3261 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3237 3262 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3238 3263 {
3239 3264 uint64_t newval;
3240 3265 uint64_t unique;
3241 3266 int64_t delta;
3242 3267
3243 3268 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3244 3269 source, sizeof (value), 1, &value, tx);
3245 3270
3246 3271 VERIFY0(dsl_prop_get_int_ds(ds,
3247 3272 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3248 3273
3249 3274 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3250 3275 mutex_enter(&ds->ds_dir->dd_lock);
3251 3276 mutex_enter(&ds->ds_lock);
3252 3277 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3253 3278 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3254 3279 delta = MAX(0, (int64_t)(newval - unique)) -
3255 3280 MAX(0, (int64_t)(ds->ds_reserved - unique));
3256 3281 ds->ds_reserved = newval;
3257 3282 mutex_exit(&ds->ds_lock);
3258 3283
3259 3284 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3260 3285 mutex_exit(&ds->ds_dir->dd_lock);
3261 3286 }
3262 3287
3263 3288 static void
3264 3289 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3265 3290 {
3266 3291 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3267 3292 dsl_pool_t *dp = dmu_tx_pool(tx);
3268 3293 dsl_dataset_t *ds;
3269 3294
3270 3295 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3271 3296 dsl_dataset_set_refreservation_sync_impl(ds,
3272 3297 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3273 3298 dsl_dataset_rele(ds, FTAG);
3274 3299 }
3275 3300
3276 3301 int
3277 3302 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3278 3303 uint64_t refreservation)
3279 3304 {
3280 3305 dsl_dataset_set_qr_arg_t ddsqra;
3281 3306
3282 3307 ddsqra.ddsqra_name = dsname;
3283 3308 ddsqra.ddsqra_source = source;
3284 3309 ddsqra.ddsqra_value = refreservation;
3285 3310
3286 3311 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3287 3312 dsl_dataset_set_refreservation_sync, &ddsqra,
3288 3313 0, ZFS_SPACE_CHECK_NONE));
3289 3314 }
3290 3315
3291 3316 /*
3292 3317 * Return (in *usedp) the amount of space written in new that is not
3293 3318 * present in oldsnap. New may be a snapshot or the head. Old must be
3294 3319 * a snapshot before new, in new's filesystem (or its origin). If not then
3295 3320 * fail and return EINVAL.
3296 3321 *
3297 3322 * The written space is calculated by considering two components: First, we
3298 3323 * ignore any freed space, and calculate the written as new's used space
3299 3324 * minus old's used space. Next, we add in the amount of space that was freed
3300 3325 * between the two snapshots, thus reducing new's used space relative to old's.
3301 3326 * Specifically, this is the space that was born before old->ds_creation_txg,
3302 3327 * and freed before new (ie. on new's deadlist or a previous deadlist).
3303 3328 *
3304 3329 * space freed [---------------------]
3305 3330 * snapshots ---O-------O--------O-------O------
3306 3331 * oldsnap new
3307 3332 */
3308 3333 int
3309 3334 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3310 3335 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3311 3336 {
3312 3337 int err = 0;
3313 3338 uint64_t snapobj;
3314 3339 dsl_pool_t *dp = new->ds_dir->dd_pool;
3315 3340
3316 3341 ASSERT(dsl_pool_config_held(dp));
3317 3342
3318 3343 *usedp = 0;
3319 3344 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3320 3345 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3321 3346
3322 3347 *compp = 0;
3323 3348 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3324 3349 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3325 3350
3326 3351 *uncompp = 0;
3327 3352 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3328 3353 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3329 3354
3330 3355 snapobj = new->ds_object;
3331 3356 while (snapobj != oldsnap->ds_object) {
3332 3357 dsl_dataset_t *snap;
3333 3358 uint64_t used, comp, uncomp;
3334 3359
3335 3360 if (snapobj == new->ds_object) {
3336 3361 snap = new;
3337 3362 } else {
3338 3363 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3339 3364 if (err != 0)
3340 3365 break;
3341 3366 }
3342 3367
3343 3368 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3344 3369 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3345 3370 /*
3346 3371 * The blocks in the deadlist can not be born after
3347 3372 * ds_prev_snap_txg, so get the whole deadlist space,
3348 3373 * which is more efficient (especially for old-format
3349 3374 * deadlists). Unfortunately the deadlist code
3350 3375 * doesn't have enough information to make this
3351 3376 * optimization itself.
3352 3377 */
3353 3378 dsl_deadlist_space(&snap->ds_deadlist,
3354 3379 &used, &comp, &uncomp);
3355 3380 } else {
3356 3381 dsl_deadlist_space_range(&snap->ds_deadlist,
3357 3382 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3358 3383 &used, &comp, &uncomp);
3359 3384 }
3360 3385 *usedp += used;
3361 3386 *compp += comp;
3362 3387 *uncompp += uncomp;
3363 3388
3364 3389 /*
3365 3390 * If we get to the beginning of the chain of snapshots
3366 3391 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3367 3392 * was not a snapshot of/before new.
3368 3393 */
3369 3394 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3370 3395 if (snap != new)
3371 3396 dsl_dataset_rele(snap, FTAG);
3372 3397 if (snapobj == 0) {
3373 3398 err = SET_ERROR(EINVAL);
3374 3399 break;
3375 3400 }
3376 3401
3377 3402 }
3378 3403 return (err);
3379 3404 }
3380 3405
3381 3406 /*
3382 3407 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3383 3408 * lastsnap, and all snapshots in between are deleted.
3384 3409 *
3385 3410 * blocks that would be freed [---------------------------]
3386 3411 * snapshots ---O-------O--------O-------O--------O
3387 3412 * firstsnap lastsnap
3388 3413 *
3389 3414 * This is the set of blocks that were born after the snap before firstsnap,
3390 3415 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3391 3416 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3392 3417 * We calculate this by iterating over the relevant deadlists (from the snap
3393 3418 * after lastsnap, backward to the snap after firstsnap), summing up the
3394 3419 * space on the deadlist that was born after the snap before firstsnap.
3395 3420 */
3396 3421 int
3397 3422 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3398 3423 dsl_dataset_t *lastsnap,
3399 3424 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3400 3425 {
3401 3426 int err = 0;
3402 3427 uint64_t snapobj;
3403 3428 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3404 3429
3405 3430 ASSERT(firstsnap->ds_is_snapshot);
3406 3431 ASSERT(lastsnap->ds_is_snapshot);
3407 3432
3408 3433 /*
3409 3434 * Check that the snapshots are in the same dsl_dir, and firstsnap
3410 3435 * is before lastsnap.
3411 3436 */
3412 3437 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3413 3438 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3414 3439 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3415 3440 return (SET_ERROR(EINVAL));
3416 3441
3417 3442 *usedp = *compp = *uncompp = 0;
3418 3443
3419 3444 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3420 3445 while (snapobj != firstsnap->ds_object) {
3421 3446 dsl_dataset_t *ds;
3422 3447 uint64_t used, comp, uncomp;
3423 3448
3424 3449 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3425 3450 if (err != 0)
3426 3451 break;
3427 3452
3428 3453 dsl_deadlist_space_range(&ds->ds_deadlist,
3429 3454 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3430 3455 &used, &comp, &uncomp);
3431 3456 *usedp += used;
3432 3457 *compp += comp;
3433 3458 *uncompp += uncomp;
3434 3459
3435 3460 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3436 3461 ASSERT3U(snapobj, !=, 0);
3437 3462 dsl_dataset_rele(ds, FTAG);
3438 3463 }
3439 3464 return (err);
3440 3465 }
3441 3466
3442 3467 /*
3443 3468 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3444 3469 * For example, they could both be snapshots of the same filesystem, and
3445 3470 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3446 3471 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3447 3472 * filesystem. Or 'earlier' could be the origin's origin.
3448 3473 *
3449 3474 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3450 3475 */
3451 3476 boolean_t
3452 3477 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3453 3478 uint64_t earlier_txg)
3454 3479 {
3455 3480 dsl_pool_t *dp = later->ds_dir->dd_pool;
3456 3481 int error;
3457 3482 boolean_t ret;
3458 3483
3459 3484 ASSERT(dsl_pool_config_held(dp));
3460 3485 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3461 3486
3462 3487 if (earlier_txg == 0)
3463 3488 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3464 3489
3465 3490 if (later->ds_is_snapshot &&
3466 3491 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3467 3492 return (B_FALSE);
3468 3493
3469 3494 if (later->ds_dir == earlier->ds_dir)
3470 3495 return (B_TRUE);
3471 3496 if (!dsl_dir_is_clone(later->ds_dir))
3472 3497 return (B_FALSE);
3473 3498
3474 3499 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3475 3500 return (B_TRUE);
3476 3501 dsl_dataset_t *origin;
3477 3502 error = dsl_dataset_hold_obj(dp,
3478 3503 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3479 3504 if (error != 0)
3480 3505 return (B_FALSE);
3481 3506 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3482 3507 dsl_dataset_rele(origin, FTAG);
3483 3508 return (ret);
3484 3509 }
3485 3510
3486 3511 void
3487 3512 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3488 3513 {
3489 3514 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3490 3515 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3491 3516 }
3492 3517
3493 3518 boolean_t
3494 3519 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3495 3520 {
3496 3521 dmu_object_info_t doi;
3497 3522
3498 3523 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3499 3524 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3500 3525 }
3501 3526
3502 3527 boolean_t
3503 3528 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3504 3529 {
3505 3530 return (dsl_dataset_is_zapified(ds) &&
3506 3531 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3507 3532 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3508 3533 }
|
↓ open down ↓ |
647 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX