Print this page
3743 zfs needs a refcount audit
Submitted by: Will Andrews <willa@spectralogic.com>
Submitted by: Justin Gibbs <justing@spectralogic.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.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]
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) 2013 by Delphix. All rights reserved.
24 24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 25 */
26 26
27 27 #include <sys/dmu_objset.h>
28 28 #include <sys/dsl_dataset.h>
29 29 #include <sys/dsl_dir.h>
30 30 #include <sys/dsl_prop.h>
31 31 #include <sys/dsl_synctask.h>
32 32 #include <sys/dmu_traverse.h>
33 33 #include <sys/dmu_impl.h>
34 34 #include <sys/dmu_tx.h>
35 35 #include <sys/arc.h>
36 36 #include <sys/zio.h>
37 37 #include <sys/zap.h>
38 38 #include <sys/zfeature.h>
39 39 #include <sys/unique.h>
40 40 #include <sys/zfs_context.h>
41 41 #include <sys/zfs_ioctl.h>
42 42 #include <sys/spa.h>
43 43 #include <sys/zfs_znode.h>
44 44 #include <sys/zfs_onexit.h>
45 45 #include <sys/zvol.h>
46 46 #include <sys/dsl_scan.h>
47 47 #include <sys/dsl_deadlist.h>
48 48 #include <sys/dsl_destroy.h>
49 49 #include <sys/dsl_userhold.h>
50 50
51 51 #define SWITCH64(x, y) \
52 52 { \
53 53 uint64_t __tmp = (x); \
54 54 (x) = (y); \
55 55 (y) = __tmp; \
56 56 }
57 57
58 58 #define DS_REF_MAX (1ULL << 62)
59 59
60 60 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
61 61
62 62 /*
63 63 * Figure out how much of this delta should be propogated to the dsl_dir
64 64 * layer. If there's a refreservation, that space has already been
65 65 * partially accounted for in our ancestors.
66 66 */
67 67 static int64_t
68 68 parent_delta(dsl_dataset_t *ds, int64_t delta)
69 69 {
70 70 uint64_t old_bytes, new_bytes;
71 71
72 72 if (ds->ds_reserved == 0)
73 73 return (delta);
74 74
75 75 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
76 76 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
77 77
78 78 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
79 79 return (new_bytes - old_bytes);
80 80 }
81 81
82 82 void
83 83 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
84 84 {
85 85 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
86 86 int compressed = BP_GET_PSIZE(bp);
87 87 int uncompressed = BP_GET_UCSIZE(bp);
88 88 int64_t delta;
89 89
90 90 dprintf_bp(bp, "ds=%p", ds);
91 91
92 92 ASSERT(dmu_tx_is_syncing(tx));
93 93 /* It could have been compressed away to nothing */
94 94 if (BP_IS_HOLE(bp))
95 95 return;
96 96 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
97 97 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
98 98 if (ds == NULL) {
99 99 dsl_pool_mos_diduse_space(tx->tx_pool,
100 100 used, compressed, uncompressed);
101 101 return;
102 102 }
103 103 dmu_buf_will_dirty(ds->ds_dbuf, tx);
104 104
105 105 mutex_enter(&ds->ds_dir->dd_lock);
106 106 mutex_enter(&ds->ds_lock);
107 107 delta = parent_delta(ds, used);
108 108 ds->ds_phys->ds_referenced_bytes += used;
109 109 ds->ds_phys->ds_compressed_bytes += compressed;
110 110 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
111 111 ds->ds_phys->ds_unique_bytes += used;
112 112 mutex_exit(&ds->ds_lock);
113 113 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
114 114 compressed, uncompressed, tx);
115 115 dsl_dir_transfer_space(ds->ds_dir, used - delta,
116 116 DD_USED_REFRSRV, DD_USED_HEAD, tx);
117 117 mutex_exit(&ds->ds_dir->dd_lock);
118 118 }
119 119
120 120 int
121 121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
122 122 boolean_t async)
123 123 {
124 124 if (BP_IS_HOLE(bp))
125 125 return (0);
126 126
127 127 ASSERT(dmu_tx_is_syncing(tx));
128 128 ASSERT(bp->blk_birth <= tx->tx_txg);
129 129
130 130 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
131 131 int compressed = BP_GET_PSIZE(bp);
132 132 int uncompressed = BP_GET_UCSIZE(bp);
133 133
134 134 ASSERT(used > 0);
135 135 if (ds == NULL) {
136 136 dsl_free(tx->tx_pool, tx->tx_txg, bp);
137 137 dsl_pool_mos_diduse_space(tx->tx_pool,
138 138 -used, -compressed, -uncompressed);
139 139 return (used);
140 140 }
141 141 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
142 142
143 143 ASSERT(!dsl_dataset_is_snapshot(ds));
144 144 dmu_buf_will_dirty(ds->ds_dbuf, tx);
145 145
146 146 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
147 147 int64_t delta;
148 148
149 149 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
150 150 dsl_free(tx->tx_pool, tx->tx_txg, bp);
151 151
152 152 mutex_enter(&ds->ds_dir->dd_lock);
153 153 mutex_enter(&ds->ds_lock);
154 154 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
155 155 !DS_UNIQUE_IS_ACCURATE(ds));
156 156 delta = parent_delta(ds, -used);
157 157 ds->ds_phys->ds_unique_bytes -= used;
158 158 mutex_exit(&ds->ds_lock);
159 159 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
160 160 delta, -compressed, -uncompressed, tx);
161 161 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
162 162 DD_USED_REFRSRV, DD_USED_HEAD, tx);
163 163 mutex_exit(&ds->ds_dir->dd_lock);
164 164 } else {
165 165 dprintf_bp(bp, "putting on dead list: %s", "");
166 166 if (async) {
167 167 /*
168 168 * We are here as part of zio's write done callback,
169 169 * which means we're a zio interrupt thread. We can't
170 170 * call dsl_deadlist_insert() now because it may block
171 171 * waiting for I/O. Instead, put bp on the deferred
172 172 * queue and let dsl_pool_sync() finish the job.
173 173 */
174 174 bplist_append(&ds->ds_pending_deadlist, bp);
175 175 } else {
176 176 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
177 177 }
178 178 ASSERT3U(ds->ds_prev->ds_object, ==,
179 179 ds->ds_phys->ds_prev_snap_obj);
180 180 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
181 181 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
182 182 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
183 183 ds->ds_object && bp->blk_birth >
184 184 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
185 185 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
186 186 mutex_enter(&ds->ds_prev->ds_lock);
187 187 ds->ds_prev->ds_phys->ds_unique_bytes += used;
188 188 mutex_exit(&ds->ds_prev->ds_lock);
189 189 }
190 190 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
191 191 dsl_dir_transfer_space(ds->ds_dir, used,
192 192 DD_USED_HEAD, DD_USED_SNAP, tx);
193 193 }
194 194 }
195 195 mutex_enter(&ds->ds_lock);
196 196 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
197 197 ds->ds_phys->ds_referenced_bytes -= used;
198 198 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
199 199 ds->ds_phys->ds_compressed_bytes -= compressed;
200 200 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
201 201 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
202 202 mutex_exit(&ds->ds_lock);
203 203
204 204 return (used);
205 205 }
206 206
207 207 uint64_t
208 208 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
209 209 {
210 210 uint64_t trysnap = 0;
211 211
212 212 if (ds == NULL)
213 213 return (0);
214 214 /*
215 215 * The snapshot creation could fail, but that would cause an
216 216 * incorrect FALSE return, which would only result in an
217 217 * overestimation of the amount of space that an operation would
218 218 * consume, which is OK.
219 219 *
220 220 * There's also a small window where we could miss a pending
221 221 * snapshot, because we could set the sync task in the quiescing
222 222 * phase. So this should only be used as a guess.
223 223 */
224 224 if (ds->ds_trysnap_txg >
225 225 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
226 226 trysnap = ds->ds_trysnap_txg;
227 227 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
228 228 }
229 229
230 230 boolean_t
231 231 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
232 232 uint64_t blk_birth)
233 233 {
234 234 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
235 235 return (B_FALSE);
236 236
237 237 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
238 238
239 239 return (B_TRUE);
240 240 }
241 241
242 242 /* ARGSUSED */
243 243 static void
244 244 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
245 245 {
246 246 dsl_dataset_t *ds = dsv;
247 247
248 248 ASSERT(ds->ds_owner == NULL);
249 249
250 250 unique_remove(ds->ds_fsid_guid);
251 251
252 252 if (ds->ds_objset != NULL)
253 253 dmu_objset_evict(ds->ds_objset);
254 254
255 255 if (ds->ds_prev) {
256 256 dsl_dataset_rele(ds->ds_prev, ds);
257 257 ds->ds_prev = NULL;
258 258 }
259 259
260 260 bplist_destroy(&ds->ds_pending_deadlist);
261 261 if (ds->ds_phys->ds_deadlist_obj != 0)
262 262 dsl_deadlist_close(&ds->ds_deadlist);
263 263 if (ds->ds_dir)
264 264 dsl_dir_rele(ds->ds_dir, ds);
265 265
266 266 ASSERT(!list_link_active(&ds->ds_synced_link));
267 267
268 268 mutex_destroy(&ds->ds_lock);
269 269 mutex_destroy(&ds->ds_opening_lock);
270 270 refcount_destroy(&ds->ds_longholds);
271 271
272 272 kmem_free(ds, sizeof (dsl_dataset_t));
273 273 }
274 274
275 275 int
276 276 dsl_dataset_get_snapname(dsl_dataset_t *ds)
277 277 {
278 278 dsl_dataset_phys_t *headphys;
279 279 int err;
280 280 dmu_buf_t *headdbuf;
281 281 dsl_pool_t *dp = ds->ds_dir->dd_pool;
282 282 objset_t *mos = dp->dp_meta_objset;
283 283
284 284 if (ds->ds_snapname[0])
285 285 return (0);
286 286 if (ds->ds_phys->ds_next_snap_obj == 0)
287 287 return (0);
288 288
289 289 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
290 290 FTAG, &headdbuf);
291 291 if (err != 0)
292 292 return (err);
293 293 headphys = headdbuf->db_data;
294 294 err = zap_value_search(dp->dp_meta_objset,
295 295 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
296 296 dmu_buf_rele(headdbuf, FTAG);
297 297 return (err);
298 298 }
299 299
300 300 int
301 301 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
302 302 {
303 303 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
304 304 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
305 305 matchtype_t mt;
306 306 int err;
307 307
308 308 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
309 309 mt = MT_FIRST;
310 310 else
311 311 mt = MT_EXACT;
312 312
313 313 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
314 314 value, mt, NULL, 0, NULL);
315 315 if (err == ENOTSUP && mt == MT_FIRST)
316 316 err = zap_lookup(mos, snapobj, name, 8, 1, value);
317 317 return (err);
318 318 }
319 319
320 320 int
321 321 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
322 322 {
323 323 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
324 324 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
325 325 matchtype_t mt;
326 326 int err;
327 327
328 328 dsl_dir_snap_cmtime_update(ds->ds_dir);
329 329
330 330 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
331 331 mt = MT_FIRST;
332 332 else
333 333 mt = MT_EXACT;
334 334
335 335 err = zap_remove_norm(mos, snapobj, name, mt, tx);
336 336 if (err == ENOTSUP && mt == MT_FIRST)
337 337 err = zap_remove(mos, snapobj, name, tx);
338 338 return (err);
339 339 }
340 340
341 341 int
342 342 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
343 343 dsl_dataset_t **dsp)
344 344 {
345 345 objset_t *mos = dp->dp_meta_objset;
346 346 dmu_buf_t *dbuf;
347 347 dsl_dataset_t *ds;
348 348 int err;
|
↓ open down ↓ |
348 lines elided |
↑ open up ↑ |
349 349 dmu_object_info_t doi;
350 350
351 351 ASSERT(dsl_pool_config_held(dp));
352 352
353 353 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
354 354 if (err != 0)
355 355 return (err);
356 356
357 357 /* Make sure dsobj has the correct object type. */
358 358 dmu_object_info_from_db(dbuf, &doi);
359 - if (doi.doi_type != DMU_OT_DSL_DATASET)
359 + if (doi.doi_type != DMU_OT_DSL_DATASET) {
360 + dmu_buf_rele(dbuf, tag);
360 361 return (SET_ERROR(EINVAL));
362 + }
361 363
362 364 ds = dmu_buf_get_user(dbuf);
363 365 if (ds == NULL) {
364 366 dsl_dataset_t *winner = NULL;
365 367
366 368 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
367 369 ds->ds_dbuf = dbuf;
368 370 ds->ds_object = dsobj;
369 371 ds->ds_phys = dbuf->db_data;
370 372
371 373 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
372 374 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
373 375 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
374 376 refcount_create(&ds->ds_longholds);
375 377
376 378 bplist_create(&ds->ds_pending_deadlist);
377 379 dsl_deadlist_open(&ds->ds_deadlist,
378 380 mos, ds->ds_phys->ds_deadlist_obj);
379 381
380 382 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
381 383 offsetof(dmu_sendarg_t, dsa_link));
382 384
383 385 if (err == 0) {
384 386 err = dsl_dir_hold_obj(dp,
385 387 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
386 388 }
387 389 if (err != 0) {
388 390 mutex_destroy(&ds->ds_lock);
389 391 mutex_destroy(&ds->ds_opening_lock);
390 392 refcount_destroy(&ds->ds_longholds);
391 393 bplist_destroy(&ds->ds_pending_deadlist);
392 394 dsl_deadlist_close(&ds->ds_deadlist);
393 395 kmem_free(ds, sizeof (dsl_dataset_t));
394 396 dmu_buf_rele(dbuf, tag);
395 397 return (err);
396 398 }
397 399
398 400 if (!dsl_dataset_is_snapshot(ds)) {
399 401 ds->ds_snapname[0] = '\0';
400 402 if (ds->ds_phys->ds_prev_snap_obj != 0) {
401 403 err = dsl_dataset_hold_obj(dp,
402 404 ds->ds_phys->ds_prev_snap_obj,
403 405 ds, &ds->ds_prev);
404 406 }
405 407 } else {
406 408 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
407 409 err = dsl_dataset_get_snapname(ds);
408 410 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
409 411 err = zap_count(
410 412 ds->ds_dir->dd_pool->dp_meta_objset,
411 413 ds->ds_phys->ds_userrefs_obj,
412 414 &ds->ds_userrefs);
413 415 }
414 416 }
415 417
416 418 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
417 419 err = dsl_prop_get_int_ds(ds,
418 420 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
419 421 &ds->ds_reserved);
420 422 if (err == 0) {
421 423 err = dsl_prop_get_int_ds(ds,
422 424 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
423 425 &ds->ds_quota);
424 426 }
425 427 } else {
426 428 ds->ds_reserved = ds->ds_quota = 0;
427 429 }
428 430
429 431 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
430 432 &ds->ds_phys, dsl_dataset_evict)) != NULL) {
431 433 bplist_destroy(&ds->ds_pending_deadlist);
432 434 dsl_deadlist_close(&ds->ds_deadlist);
433 435 if (ds->ds_prev)
434 436 dsl_dataset_rele(ds->ds_prev, ds);
435 437 dsl_dir_rele(ds->ds_dir, ds);
436 438 mutex_destroy(&ds->ds_lock);
437 439 mutex_destroy(&ds->ds_opening_lock);
438 440 refcount_destroy(&ds->ds_longholds);
439 441 kmem_free(ds, sizeof (dsl_dataset_t));
440 442 if (err != 0) {
441 443 dmu_buf_rele(dbuf, tag);
442 444 return (err);
443 445 }
444 446 ds = winner;
445 447 } else {
446 448 ds->ds_fsid_guid =
447 449 unique_insert(ds->ds_phys->ds_fsid_guid);
448 450 }
449 451 }
450 452 ASSERT3P(ds->ds_dbuf, ==, dbuf);
451 453 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
452 454 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
453 455 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
454 456 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
455 457 *dsp = ds;
456 458 return (0);
457 459 }
458 460
459 461 int
460 462 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
461 463 void *tag, dsl_dataset_t **dsp)
462 464 {
463 465 dsl_dir_t *dd;
464 466 const char *snapname;
465 467 uint64_t obj;
466 468 int err = 0;
467 469
468 470 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
469 471 if (err != 0)
470 472 return (err);
471 473
472 474 ASSERT(dsl_pool_config_held(dp));
473 475 obj = dd->dd_phys->dd_head_dataset_obj;
474 476 if (obj != 0)
475 477 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
476 478 else
477 479 err = SET_ERROR(ENOENT);
478 480
479 481 /* we may be looking for a snapshot */
480 482 if (err == 0 && snapname != NULL) {
481 483 dsl_dataset_t *ds;
482 484
483 485 if (*snapname++ != '@') {
484 486 dsl_dataset_rele(*dsp, tag);
485 487 dsl_dir_rele(dd, FTAG);
486 488 return (SET_ERROR(ENOENT));
487 489 }
488 490
489 491 dprintf("looking for snapshot '%s'\n", snapname);
490 492 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
491 493 if (err == 0)
492 494 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
493 495 dsl_dataset_rele(*dsp, tag);
494 496
495 497 if (err == 0) {
496 498 mutex_enter(&ds->ds_lock);
497 499 if (ds->ds_snapname[0] == 0)
498 500 (void) strlcpy(ds->ds_snapname, snapname,
499 501 sizeof (ds->ds_snapname));
500 502 mutex_exit(&ds->ds_lock);
501 503 *dsp = ds;
502 504 }
503 505 }
504 506
505 507 dsl_dir_rele(dd, FTAG);
506 508 return (err);
507 509 }
508 510
509 511 int
510 512 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
511 513 void *tag, dsl_dataset_t **dsp)
512 514 {
513 515 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
514 516 if (err != 0)
515 517 return (err);
516 518 if (!dsl_dataset_tryown(*dsp, tag)) {
517 519 dsl_dataset_rele(*dsp, tag);
518 520 *dsp = NULL;
519 521 return (SET_ERROR(EBUSY));
520 522 }
521 523 return (0);
522 524 }
523 525
524 526 int
525 527 dsl_dataset_own(dsl_pool_t *dp, const char *name,
526 528 void *tag, dsl_dataset_t **dsp)
527 529 {
528 530 int err = dsl_dataset_hold(dp, name, tag, dsp);
529 531 if (err != 0)
530 532 return (err);
531 533 if (!dsl_dataset_tryown(*dsp, tag)) {
532 534 dsl_dataset_rele(*dsp, tag);
533 535 return (SET_ERROR(EBUSY));
534 536 }
535 537 return (0);
536 538 }
537 539
538 540 /*
539 541 * See the comment above dsl_pool_hold() for details. In summary, a long
540 542 * hold is used to prevent destruction of a dataset while the pool hold
541 543 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
542 544 *
543 545 * The dataset and pool must be held when this function is called. After it
544 546 * is called, the pool hold may be released while the dataset is still held
545 547 * and accessed.
546 548 */
547 549 void
548 550 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
549 551 {
550 552 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
551 553 (void) refcount_add(&ds->ds_longholds, tag);
552 554 }
553 555
554 556 void
555 557 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
556 558 {
557 559 (void) refcount_remove(&ds->ds_longholds, tag);
558 560 }
559 561
560 562 /* Return B_TRUE if there are any long holds on this dataset. */
561 563 boolean_t
562 564 dsl_dataset_long_held(dsl_dataset_t *ds)
563 565 {
564 566 return (!refcount_is_zero(&ds->ds_longholds));
565 567 }
566 568
567 569 void
568 570 dsl_dataset_name(dsl_dataset_t *ds, char *name)
569 571 {
570 572 if (ds == NULL) {
571 573 (void) strcpy(name, "mos");
572 574 } else {
573 575 dsl_dir_name(ds->ds_dir, name);
574 576 VERIFY0(dsl_dataset_get_snapname(ds));
575 577 if (ds->ds_snapname[0]) {
576 578 (void) strcat(name, "@");
577 579 /*
578 580 * We use a "recursive" mutex so that we
579 581 * can call dprintf_ds() with ds_lock held.
580 582 */
581 583 if (!MUTEX_HELD(&ds->ds_lock)) {
582 584 mutex_enter(&ds->ds_lock);
583 585 (void) strcat(name, ds->ds_snapname);
584 586 mutex_exit(&ds->ds_lock);
585 587 } else {
586 588 (void) strcat(name, ds->ds_snapname);
587 589 }
588 590 }
589 591 }
590 592 }
591 593
592 594 static int
593 595 dsl_dataset_namelen(dsl_dataset_t *ds)
594 596 {
595 597 int result;
596 598
597 599 if (ds == NULL) {
598 600 result = 3; /* "mos" */
599 601 } else {
600 602 result = dsl_dir_namelen(ds->ds_dir);
601 603 VERIFY0(dsl_dataset_get_snapname(ds));
602 604 if (ds->ds_snapname[0]) {
603 605 ++result; /* adding one for the @-sign */
604 606 if (!MUTEX_HELD(&ds->ds_lock)) {
605 607 mutex_enter(&ds->ds_lock);
606 608 result += strlen(ds->ds_snapname);
607 609 mutex_exit(&ds->ds_lock);
608 610 } else {
609 611 result += strlen(ds->ds_snapname);
610 612 }
611 613 }
612 614 }
613 615
614 616 return (result);
615 617 }
616 618
617 619 void
618 620 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
619 621 {
620 622 dmu_buf_rele(ds->ds_dbuf, tag);
621 623 }
622 624
623 625 void
624 626 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
625 627 {
626 628 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
627 629
628 630 mutex_enter(&ds->ds_lock);
629 631 ds->ds_owner = NULL;
630 632 mutex_exit(&ds->ds_lock);
631 633 dsl_dataset_long_rele(ds, tag);
632 634 if (ds->ds_dbuf != NULL)
633 635 dsl_dataset_rele(ds, tag);
634 636 else
635 637 dsl_dataset_evict(NULL, ds);
636 638 }
637 639
638 640 boolean_t
639 641 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
640 642 {
641 643 boolean_t gotit = FALSE;
642 644
643 645 mutex_enter(&ds->ds_lock);
644 646 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
645 647 ds->ds_owner = tag;
646 648 dsl_dataset_long_hold(ds, tag);
647 649 gotit = TRUE;
648 650 }
649 651 mutex_exit(&ds->ds_lock);
650 652 return (gotit);
651 653 }
652 654
653 655 uint64_t
654 656 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
655 657 uint64_t flags, dmu_tx_t *tx)
656 658 {
657 659 dsl_pool_t *dp = dd->dd_pool;
658 660 dmu_buf_t *dbuf;
659 661 dsl_dataset_phys_t *dsphys;
660 662 uint64_t dsobj;
661 663 objset_t *mos = dp->dp_meta_objset;
662 664
663 665 if (origin == NULL)
664 666 origin = dp->dp_origin_snap;
665 667
666 668 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
667 669 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
668 670 ASSERT(dmu_tx_is_syncing(tx));
669 671 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
670 672
671 673 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
672 674 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
673 675 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
674 676 dmu_buf_will_dirty(dbuf, tx);
675 677 dsphys = dbuf->db_data;
676 678 bzero(dsphys, sizeof (dsl_dataset_phys_t));
677 679 dsphys->ds_dir_obj = dd->dd_object;
678 680 dsphys->ds_flags = flags;
679 681 dsphys->ds_fsid_guid = unique_create();
680 682 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
681 683 sizeof (dsphys->ds_guid));
682 684 dsphys->ds_snapnames_zapobj =
683 685 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
684 686 DMU_OT_NONE, 0, tx);
685 687 dsphys->ds_creation_time = gethrestime_sec();
686 688 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
687 689
688 690 if (origin == NULL) {
689 691 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
690 692 } else {
691 693 dsl_dataset_t *ohds; /* head of the origin snapshot */
692 694
693 695 dsphys->ds_prev_snap_obj = origin->ds_object;
694 696 dsphys->ds_prev_snap_txg =
695 697 origin->ds_phys->ds_creation_txg;
696 698 dsphys->ds_referenced_bytes =
697 699 origin->ds_phys->ds_referenced_bytes;
698 700 dsphys->ds_compressed_bytes =
699 701 origin->ds_phys->ds_compressed_bytes;
700 702 dsphys->ds_uncompressed_bytes =
701 703 origin->ds_phys->ds_uncompressed_bytes;
702 704 dsphys->ds_bp = origin->ds_phys->ds_bp;
703 705 dsphys->ds_flags |= origin->ds_phys->ds_flags;
704 706
705 707 dmu_buf_will_dirty(origin->ds_dbuf, tx);
706 708 origin->ds_phys->ds_num_children++;
707 709
708 710 VERIFY0(dsl_dataset_hold_obj(dp,
709 711 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
710 712 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
711 713 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
712 714 dsl_dataset_rele(ohds, FTAG);
713 715
714 716 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
715 717 if (origin->ds_phys->ds_next_clones_obj == 0) {
716 718 origin->ds_phys->ds_next_clones_obj =
717 719 zap_create(mos,
718 720 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
719 721 }
720 722 VERIFY0(zap_add_int(mos,
721 723 origin->ds_phys->ds_next_clones_obj, dsobj, tx));
722 724 }
723 725
724 726 dmu_buf_will_dirty(dd->dd_dbuf, tx);
725 727 dd->dd_phys->dd_origin_obj = origin->ds_object;
726 728 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
727 729 if (origin->ds_dir->dd_phys->dd_clones == 0) {
728 730 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
729 731 origin->ds_dir->dd_phys->dd_clones =
730 732 zap_create(mos,
731 733 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
732 734 }
733 735 VERIFY0(zap_add_int(mos,
734 736 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
735 737 }
736 738 }
737 739
738 740 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
739 741 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
740 742
741 743 dmu_buf_rele(dbuf, FTAG);
742 744
743 745 dmu_buf_will_dirty(dd->dd_dbuf, tx);
744 746 dd->dd_phys->dd_head_dataset_obj = dsobj;
745 747
746 748 return (dsobj);
747 749 }
748 750
749 751 static void
750 752 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
751 753 {
752 754 objset_t *os;
753 755
754 756 VERIFY0(dmu_objset_from_ds(ds, &os));
755 757 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
756 758 dsl_dataset_dirty(ds, tx);
757 759 }
758 760
759 761 uint64_t
760 762 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
761 763 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
762 764 {
763 765 dsl_pool_t *dp = pdd->dd_pool;
764 766 uint64_t dsobj, ddobj;
765 767 dsl_dir_t *dd;
766 768
767 769 ASSERT(dmu_tx_is_syncing(tx));
768 770 ASSERT(lastname[0] != '@');
769 771
770 772 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
771 773 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
772 774
773 775 dsobj = dsl_dataset_create_sync_dd(dd, origin,
774 776 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
775 777
776 778 dsl_deleg_set_create_perms(dd, tx, cr);
777 779
778 780 dsl_dir_rele(dd, FTAG);
779 781
780 782 /*
781 783 * If we are creating a clone, make sure we zero out any stale
782 784 * data from the origin snapshots zil header.
783 785 */
784 786 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
785 787 dsl_dataset_t *ds;
786 788
787 789 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
788 790 dsl_dataset_zero_zil(ds, tx);
789 791 dsl_dataset_rele(ds, FTAG);
790 792 }
791 793
792 794 return (dsobj);
793 795 }
794 796
795 797 /*
796 798 * The unique space in the head dataset can be calculated by subtracting
797 799 * the space used in the most recent snapshot, that is still being used
798 800 * in this file system, from the space currently in use. To figure out
799 801 * the space in the most recent snapshot still in use, we need to take
800 802 * the total space used in the snapshot and subtract out the space that
801 803 * has been freed up since the snapshot was taken.
802 804 */
803 805 void
804 806 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
805 807 {
806 808 uint64_t mrs_used;
807 809 uint64_t dlused, dlcomp, dluncomp;
808 810
809 811 ASSERT(!dsl_dataset_is_snapshot(ds));
810 812
811 813 if (ds->ds_phys->ds_prev_snap_obj != 0)
812 814 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
813 815 else
814 816 mrs_used = 0;
815 817
816 818 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
817 819
818 820 ASSERT3U(dlused, <=, mrs_used);
819 821 ds->ds_phys->ds_unique_bytes =
820 822 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
821 823
822 824 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
823 825 SPA_VERSION_UNIQUE_ACCURATE)
824 826 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
825 827 }
826 828
827 829 void
828 830 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
829 831 dmu_tx_t *tx)
830 832 {
831 833 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
832 834 uint64_t count;
833 835 int err;
834 836
835 837 ASSERT(ds->ds_phys->ds_num_children >= 2);
836 838 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
837 839 /*
838 840 * The err should not be ENOENT, but a bug in a previous version
839 841 * of the code could cause upgrade_clones_cb() to not set
840 842 * ds_next_snap_obj when it should, leading to a missing entry.
841 843 * If we knew that the pool was created after
842 844 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
843 845 * ENOENT. However, at least we can check that we don't have
844 846 * too many entries in the next_clones_obj even after failing to
845 847 * remove this one.
846 848 */
847 849 if (err != ENOENT)
848 850 VERIFY0(err);
849 851 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
850 852 &count));
851 853 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
852 854 }
853 855
854 856
855 857 blkptr_t *
856 858 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
857 859 {
858 860 return (&ds->ds_phys->ds_bp);
859 861 }
860 862
861 863 void
862 864 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
863 865 {
864 866 ASSERT(dmu_tx_is_syncing(tx));
865 867 /* If it's the meta-objset, set dp_meta_rootbp */
866 868 if (ds == NULL) {
867 869 tx->tx_pool->dp_meta_rootbp = *bp;
868 870 } else {
869 871 dmu_buf_will_dirty(ds->ds_dbuf, tx);
870 872 ds->ds_phys->ds_bp = *bp;
871 873 }
872 874 }
873 875
874 876 spa_t *
875 877 dsl_dataset_get_spa(dsl_dataset_t *ds)
876 878 {
877 879 return (ds->ds_dir->dd_pool->dp_spa);
878 880 }
879 881
880 882 void
881 883 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
882 884 {
883 885 dsl_pool_t *dp;
884 886
885 887 if (ds == NULL) /* this is the meta-objset */
886 888 return;
887 889
888 890 ASSERT(ds->ds_objset != NULL);
889 891
890 892 if (ds->ds_phys->ds_next_snap_obj != 0)
891 893 panic("dirtying snapshot!");
892 894
893 895 dp = ds->ds_dir->dd_pool;
894 896
895 897 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
896 898 /* up the hold count until we can be written out */
897 899 dmu_buf_add_ref(ds->ds_dbuf, ds);
898 900 }
899 901 }
900 902
901 903 boolean_t
902 904 dsl_dataset_is_dirty(dsl_dataset_t *ds)
903 905 {
904 906 for (int t = 0; t < TXG_SIZE; t++) {
905 907 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
906 908 ds, t))
907 909 return (B_TRUE);
908 910 }
909 911 return (B_FALSE);
910 912 }
911 913
912 914 static int
913 915 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
914 916 {
915 917 uint64_t asize;
916 918
917 919 if (!dmu_tx_is_syncing(tx))
918 920 return (0);
919 921
920 922 /*
921 923 * If there's an fs-only reservation, any blocks that might become
922 924 * owned by the snapshot dataset must be accommodated by space
923 925 * outside of the reservation.
924 926 */
925 927 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
926 928 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
927 929 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
928 930 return (SET_ERROR(ENOSPC));
929 931
930 932 /*
931 933 * Propagate any reserved space for this snapshot to other
932 934 * snapshot checks in this sync group.
933 935 */
934 936 if (asize > 0)
935 937 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
936 938
937 939 return (0);
938 940 }
939 941
940 942 typedef struct dsl_dataset_snapshot_arg {
941 943 nvlist_t *ddsa_snaps;
942 944 nvlist_t *ddsa_props;
943 945 nvlist_t *ddsa_errors;
944 946 } dsl_dataset_snapshot_arg_t;
945 947
946 948 int
947 949 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
948 950 dmu_tx_t *tx)
949 951 {
950 952 int error;
951 953 uint64_t value;
952 954
953 955 ds->ds_trysnap_txg = tx->tx_txg;
954 956
955 957 if (!dmu_tx_is_syncing(tx))
956 958 return (0);
957 959
958 960 /*
959 961 * We don't allow multiple snapshots of the same txg. If there
960 962 * is already one, try again.
961 963 */
962 964 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
963 965 return (SET_ERROR(EAGAIN));
964 966
965 967 /*
966 968 * Check for conflicting snapshot name.
967 969 */
968 970 error = dsl_dataset_snap_lookup(ds, snapname, &value);
969 971 if (error == 0)
970 972 return (SET_ERROR(EEXIST));
971 973 if (error != ENOENT)
972 974 return (error);
973 975
974 976 error = dsl_dataset_snapshot_reserve_space(ds, tx);
975 977 if (error != 0)
976 978 return (error);
977 979
978 980 return (0);
979 981 }
980 982
981 983 static int
982 984 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
983 985 {
984 986 dsl_dataset_snapshot_arg_t *ddsa = arg;
985 987 dsl_pool_t *dp = dmu_tx_pool(tx);
986 988 nvpair_t *pair;
987 989 int rv = 0;
988 990
989 991 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
990 992 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
991 993 int error = 0;
992 994 dsl_dataset_t *ds;
993 995 char *name, *atp;
994 996 char dsname[MAXNAMELEN];
995 997
996 998 name = nvpair_name(pair);
997 999 if (strlen(name) >= MAXNAMELEN)
998 1000 error = SET_ERROR(ENAMETOOLONG);
999 1001 if (error == 0) {
1000 1002 atp = strchr(name, '@');
1001 1003 if (atp == NULL)
1002 1004 error = SET_ERROR(EINVAL);
1003 1005 if (error == 0)
1004 1006 (void) strlcpy(dsname, name, atp - name + 1);
1005 1007 }
1006 1008 if (error == 0)
1007 1009 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1008 1010 if (error == 0) {
1009 1011 error = dsl_dataset_snapshot_check_impl(ds,
1010 1012 atp + 1, tx);
1011 1013 dsl_dataset_rele(ds, FTAG);
1012 1014 }
1013 1015
1014 1016 if (error != 0) {
1015 1017 if (ddsa->ddsa_errors != NULL) {
1016 1018 fnvlist_add_int32(ddsa->ddsa_errors,
1017 1019 name, error);
1018 1020 }
1019 1021 rv = error;
1020 1022 }
1021 1023 }
1022 1024 return (rv);
1023 1025 }
1024 1026
1025 1027 void
1026 1028 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1027 1029 dmu_tx_t *tx)
1028 1030 {
1029 1031 static zil_header_t zero_zil;
1030 1032
1031 1033 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1032 1034 dmu_buf_t *dbuf;
1033 1035 dsl_dataset_phys_t *dsphys;
1034 1036 uint64_t dsobj, crtxg;
1035 1037 objset_t *mos = dp->dp_meta_objset;
1036 1038 objset_t *os;
1037 1039
1038 1040 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1039 1041
1040 1042 /*
1041 1043 * If we are on an old pool, the zil must not be active, in which
1042 1044 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1043 1045 */
1044 1046 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1045 1047 dmu_objset_from_ds(ds, &os) != 0 ||
1046 1048 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1047 1049 sizeof (zero_zil)) == 0);
1048 1050
1049 1051
1050 1052 /*
1051 1053 * The origin's ds_creation_txg has to be < TXG_INITIAL
1052 1054 */
1053 1055 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1054 1056 crtxg = 1;
1055 1057 else
1056 1058 crtxg = tx->tx_txg;
1057 1059
1058 1060 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1059 1061 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1060 1062 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1061 1063 dmu_buf_will_dirty(dbuf, tx);
1062 1064 dsphys = dbuf->db_data;
1063 1065 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1064 1066 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1065 1067 dsphys->ds_fsid_guid = unique_create();
1066 1068 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1067 1069 sizeof (dsphys->ds_guid));
1068 1070 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1069 1071 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1070 1072 dsphys->ds_next_snap_obj = ds->ds_object;
1071 1073 dsphys->ds_num_children = 1;
1072 1074 dsphys->ds_creation_time = gethrestime_sec();
1073 1075 dsphys->ds_creation_txg = crtxg;
1074 1076 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1075 1077 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1076 1078 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1077 1079 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1078 1080 dsphys->ds_flags = ds->ds_phys->ds_flags;
1079 1081 dsphys->ds_bp = ds->ds_phys->ds_bp;
1080 1082 dmu_buf_rele(dbuf, FTAG);
1081 1083
1082 1084 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1083 1085 if (ds->ds_prev) {
1084 1086 uint64_t next_clones_obj =
1085 1087 ds->ds_prev->ds_phys->ds_next_clones_obj;
1086 1088 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1087 1089 ds->ds_object ||
1088 1090 ds->ds_prev->ds_phys->ds_num_children > 1);
1089 1091 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1090 1092 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1091 1093 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1092 1094 ds->ds_prev->ds_phys->ds_creation_txg);
1093 1095 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1094 1096 } else if (next_clones_obj != 0) {
1095 1097 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1096 1098 dsphys->ds_next_snap_obj, tx);
1097 1099 VERIFY0(zap_add_int(mos,
1098 1100 next_clones_obj, dsobj, tx));
1099 1101 }
1100 1102 }
1101 1103
1102 1104 /*
1103 1105 * If we have a reference-reservation on this dataset, we will
1104 1106 * need to increase the amount of refreservation being charged
1105 1107 * since our unique space is going to zero.
1106 1108 */
1107 1109 if (ds->ds_reserved) {
1108 1110 int64_t delta;
1109 1111 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1110 1112 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1111 1113 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1112 1114 delta, 0, 0, tx);
1113 1115 }
1114 1116
1115 1117 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1116 1118 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1117 1119 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1118 1120 dsl_deadlist_close(&ds->ds_deadlist);
1119 1121 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1120 1122 dsl_deadlist_add_key(&ds->ds_deadlist,
1121 1123 ds->ds_phys->ds_prev_snap_txg, tx);
1122 1124
1123 1125 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1124 1126 ds->ds_phys->ds_prev_snap_obj = dsobj;
1125 1127 ds->ds_phys->ds_prev_snap_txg = crtxg;
1126 1128 ds->ds_phys->ds_unique_bytes = 0;
1127 1129 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1128 1130 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1129 1131
1130 1132 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1131 1133 snapname, 8, 1, &dsobj, tx));
1132 1134
1133 1135 if (ds->ds_prev)
1134 1136 dsl_dataset_rele(ds->ds_prev, ds);
1135 1137 VERIFY0(dsl_dataset_hold_obj(dp,
1136 1138 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1137 1139
1138 1140 dsl_scan_ds_snapshotted(ds, tx);
1139 1141
1140 1142 dsl_dir_snap_cmtime_update(ds->ds_dir);
1141 1143
1142 1144 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1143 1145 }
1144 1146
1145 1147 static void
1146 1148 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1147 1149 {
1148 1150 dsl_dataset_snapshot_arg_t *ddsa = arg;
1149 1151 dsl_pool_t *dp = dmu_tx_pool(tx);
1150 1152 nvpair_t *pair;
1151 1153
1152 1154 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1153 1155 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1154 1156 dsl_dataset_t *ds;
1155 1157 char *name, *atp;
1156 1158 char dsname[MAXNAMELEN];
1157 1159
1158 1160 name = nvpair_name(pair);
1159 1161 atp = strchr(name, '@');
1160 1162 (void) strlcpy(dsname, name, atp - name + 1);
1161 1163 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1162 1164
1163 1165 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1164 1166 if (ddsa->ddsa_props != NULL) {
1165 1167 dsl_props_set_sync_impl(ds->ds_prev,
1166 1168 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1167 1169 }
1168 1170 dsl_dataset_rele(ds, FTAG);
1169 1171 }
1170 1172 }
1171 1173
1172 1174 /*
1173 1175 * The snapshots must all be in the same pool.
1174 1176 * All-or-nothing: if there are any failures, nothing will be modified.
1175 1177 */
1176 1178 int
1177 1179 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1178 1180 {
1179 1181 dsl_dataset_snapshot_arg_t ddsa;
1180 1182 nvpair_t *pair;
1181 1183 boolean_t needsuspend;
1182 1184 int error;
1183 1185 spa_t *spa;
1184 1186 char *firstname;
1185 1187 nvlist_t *suspended = NULL;
1186 1188
1187 1189 pair = nvlist_next_nvpair(snaps, NULL);
1188 1190 if (pair == NULL)
1189 1191 return (0);
1190 1192 firstname = nvpair_name(pair);
1191 1193
1192 1194 error = spa_open(firstname, &spa, FTAG);
1193 1195 if (error != 0)
1194 1196 return (error);
1195 1197 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1196 1198 spa_close(spa, FTAG);
1197 1199
1198 1200 if (needsuspend) {
1199 1201 suspended = fnvlist_alloc();
1200 1202 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1201 1203 pair = nvlist_next_nvpair(snaps, pair)) {
1202 1204 char fsname[MAXNAMELEN];
1203 1205 char *snapname = nvpair_name(pair);
1204 1206 char *atp;
1205 1207 void *cookie;
1206 1208
1207 1209 atp = strchr(snapname, '@');
1208 1210 if (atp == NULL) {
1209 1211 error = SET_ERROR(EINVAL);
1210 1212 break;
1211 1213 }
1212 1214 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1213 1215
1214 1216 error = zil_suspend(fsname, &cookie);
1215 1217 if (error != 0)
1216 1218 break;
1217 1219 fnvlist_add_uint64(suspended, fsname,
1218 1220 (uintptr_t)cookie);
1219 1221 }
1220 1222 }
1221 1223
1222 1224 ddsa.ddsa_snaps = snaps;
1223 1225 ddsa.ddsa_props = props;
1224 1226 ddsa.ddsa_errors = errors;
1225 1227
1226 1228 if (error == 0) {
1227 1229 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1228 1230 dsl_dataset_snapshot_sync, &ddsa,
1229 1231 fnvlist_num_pairs(snaps) * 3);
1230 1232 }
1231 1233
1232 1234 if (suspended != NULL) {
1233 1235 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1234 1236 pair = nvlist_next_nvpair(suspended, pair)) {
1235 1237 zil_resume((void *)(uintptr_t)
1236 1238 fnvpair_value_uint64(pair));
1237 1239 }
1238 1240 fnvlist_free(suspended);
1239 1241 }
1240 1242
1241 1243 return (error);
1242 1244 }
1243 1245
1244 1246 typedef struct dsl_dataset_snapshot_tmp_arg {
1245 1247 const char *ddsta_fsname;
1246 1248 const char *ddsta_snapname;
1247 1249 minor_t ddsta_cleanup_minor;
1248 1250 const char *ddsta_htag;
1249 1251 } dsl_dataset_snapshot_tmp_arg_t;
1250 1252
1251 1253 static int
1252 1254 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1253 1255 {
1254 1256 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1255 1257 dsl_pool_t *dp = dmu_tx_pool(tx);
1256 1258 dsl_dataset_t *ds;
1257 1259 int error;
1258 1260
1259 1261 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1260 1262 if (error != 0)
1261 1263 return (error);
1262 1264
1263 1265 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx);
1264 1266 if (error != 0) {
1265 1267 dsl_dataset_rele(ds, FTAG);
1266 1268 return (error);
1267 1269 }
1268 1270
1269 1271 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1270 1272 dsl_dataset_rele(ds, FTAG);
1271 1273 return (SET_ERROR(ENOTSUP));
1272 1274 }
1273 1275 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1274 1276 B_TRUE, tx);
1275 1277 if (error != 0) {
1276 1278 dsl_dataset_rele(ds, FTAG);
1277 1279 return (error);
1278 1280 }
1279 1281
1280 1282 dsl_dataset_rele(ds, FTAG);
1281 1283 return (0);
1282 1284 }
1283 1285
1284 1286 static void
1285 1287 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1286 1288 {
1287 1289 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1288 1290 dsl_pool_t *dp = dmu_tx_pool(tx);
1289 1291 dsl_dataset_t *ds;
1290 1292
1291 1293 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1292 1294
1293 1295 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1294 1296 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1295 1297 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1296 1298 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1297 1299
1298 1300 dsl_dataset_rele(ds, FTAG);
1299 1301 }
1300 1302
1301 1303 int
1302 1304 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1303 1305 minor_t cleanup_minor, const char *htag)
1304 1306 {
1305 1307 dsl_dataset_snapshot_tmp_arg_t ddsta;
1306 1308 int error;
1307 1309 spa_t *spa;
1308 1310 boolean_t needsuspend;
1309 1311 void *cookie;
1310 1312
1311 1313 ddsta.ddsta_fsname = fsname;
1312 1314 ddsta.ddsta_snapname = snapname;
1313 1315 ddsta.ddsta_cleanup_minor = cleanup_minor;
1314 1316 ddsta.ddsta_htag = htag;
1315 1317
1316 1318 error = spa_open(fsname, &spa, FTAG);
1317 1319 if (error != 0)
1318 1320 return (error);
1319 1321 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1320 1322 spa_close(spa, FTAG);
1321 1323
1322 1324 if (needsuspend) {
1323 1325 error = zil_suspend(fsname, &cookie);
1324 1326 if (error != 0)
1325 1327 return (error);
1326 1328 }
1327 1329
1328 1330 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1329 1331 dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1330 1332
1331 1333 if (needsuspend)
1332 1334 zil_resume(cookie);
1333 1335 return (error);
1334 1336 }
1335 1337
1336 1338
1337 1339 void
1338 1340 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1339 1341 {
1340 1342 ASSERT(dmu_tx_is_syncing(tx));
1341 1343 ASSERT(ds->ds_objset != NULL);
1342 1344 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1343 1345
1344 1346 /*
1345 1347 * in case we had to change ds_fsid_guid when we opened it,
1346 1348 * sync it out now.
1347 1349 */
1348 1350 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1349 1351 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1350 1352
1351 1353 dmu_objset_sync(ds->ds_objset, zio, tx);
1352 1354 }
1353 1355
1354 1356 static void
1355 1357 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1356 1358 {
1357 1359 uint64_t count = 0;
1358 1360 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1359 1361 zap_cursor_t zc;
1360 1362 zap_attribute_t za;
1361 1363 nvlist_t *propval = fnvlist_alloc();
1362 1364 nvlist_t *val = fnvlist_alloc();
1363 1365
1364 1366 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1365 1367
1366 1368 /*
1367 1369 * There may be missing entries in ds_next_clones_obj
1368 1370 * due to a bug in a previous version of the code.
1369 1371 * Only trust it if it has the right number of entries.
1370 1372 */
1371 1373 if (ds->ds_phys->ds_next_clones_obj != 0) {
1372 1374 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1373 1375 &count));
1374 1376 }
1375 1377 if (count != ds->ds_phys->ds_num_children - 1)
1376 1378 goto fail;
1377 1379 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1378 1380 zap_cursor_retrieve(&zc, &za) == 0;
1379 1381 zap_cursor_advance(&zc)) {
1380 1382 dsl_dataset_t *clone;
1381 1383 char buf[ZFS_MAXNAMELEN];
1382 1384 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1383 1385 za.za_first_integer, FTAG, &clone));
1384 1386 dsl_dir_name(clone->ds_dir, buf);
1385 1387 fnvlist_add_boolean(val, buf);
1386 1388 dsl_dataset_rele(clone, FTAG);
1387 1389 }
1388 1390 zap_cursor_fini(&zc);
1389 1391 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1390 1392 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1391 1393 fail:
1392 1394 nvlist_free(val);
1393 1395 nvlist_free(propval);
1394 1396 }
1395 1397
1396 1398 void
1397 1399 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1398 1400 {
1399 1401 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1400 1402 uint64_t refd, avail, uobjs, aobjs, ratio;
1401 1403
1402 1404 ASSERT(dsl_pool_config_held(dp));
1403 1405
1404 1406 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1405 1407 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1406 1408 ds->ds_phys->ds_compressed_bytes);
1407 1409
1408 1410 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1409 1411 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1410 1412 ds->ds_phys->ds_uncompressed_bytes);
1411 1413
1412 1414 if (dsl_dataset_is_snapshot(ds)) {
1413 1415 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1414 1416 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1415 1417 ds->ds_phys->ds_unique_bytes);
1416 1418 get_clones_stat(ds, nv);
1417 1419 } else {
1418 1420 dsl_dir_stats(ds->ds_dir, nv);
1419 1421 }
1420 1422
1421 1423 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1422 1424 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1423 1425 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1424 1426
1425 1427 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1426 1428 ds->ds_phys->ds_creation_time);
1427 1429 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1428 1430 ds->ds_phys->ds_creation_txg);
1429 1431 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1430 1432 ds->ds_quota);
1431 1433 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1432 1434 ds->ds_reserved);
1433 1435 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1434 1436 ds->ds_phys->ds_guid);
1435 1437 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1436 1438 ds->ds_phys->ds_unique_bytes);
1437 1439 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1438 1440 ds->ds_object);
1439 1441 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1440 1442 ds->ds_userrefs);
1441 1443 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1442 1444 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1443 1445
1444 1446 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1445 1447 uint64_t written, comp, uncomp;
1446 1448 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1447 1449 dsl_dataset_t *prev;
1448 1450
1449 1451 int err = dsl_dataset_hold_obj(dp,
1450 1452 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1451 1453 if (err == 0) {
1452 1454 err = dsl_dataset_space_written(prev, ds, &written,
1453 1455 &comp, &uncomp);
1454 1456 dsl_dataset_rele(prev, FTAG);
1455 1457 if (err == 0) {
1456 1458 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1457 1459 written);
1458 1460 }
1459 1461 }
1460 1462 }
1461 1463 }
1462 1464
1463 1465 void
1464 1466 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1465 1467 {
1466 1468 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1467 1469 ASSERT(dsl_pool_config_held(dp));
1468 1470
1469 1471 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1470 1472 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1471 1473 stat->dds_guid = ds->ds_phys->ds_guid;
1472 1474 stat->dds_origin[0] = '\0';
1473 1475 if (dsl_dataset_is_snapshot(ds)) {
1474 1476 stat->dds_is_snapshot = B_TRUE;
1475 1477 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1476 1478 } else {
1477 1479 stat->dds_is_snapshot = B_FALSE;
1478 1480 stat->dds_num_clones = 0;
1479 1481
1480 1482 if (dsl_dir_is_clone(ds->ds_dir)) {
1481 1483 dsl_dataset_t *ods;
1482 1484
1483 1485 VERIFY0(dsl_dataset_hold_obj(dp,
1484 1486 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1485 1487 dsl_dataset_name(ods, stat->dds_origin);
1486 1488 dsl_dataset_rele(ods, FTAG);
1487 1489 }
1488 1490 }
1489 1491 }
1490 1492
1491 1493 uint64_t
1492 1494 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1493 1495 {
1494 1496 return (ds->ds_fsid_guid);
1495 1497 }
1496 1498
1497 1499 void
1498 1500 dsl_dataset_space(dsl_dataset_t *ds,
1499 1501 uint64_t *refdbytesp, uint64_t *availbytesp,
1500 1502 uint64_t *usedobjsp, uint64_t *availobjsp)
1501 1503 {
1502 1504 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
1503 1505 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1504 1506 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1505 1507 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1506 1508 if (ds->ds_quota != 0) {
1507 1509 /*
1508 1510 * Adjust available bytes according to refquota
1509 1511 */
1510 1512 if (*refdbytesp < ds->ds_quota)
1511 1513 *availbytesp = MIN(*availbytesp,
1512 1514 ds->ds_quota - *refdbytesp);
1513 1515 else
1514 1516 *availbytesp = 0;
1515 1517 }
1516 1518 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1517 1519 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1518 1520 }
1519 1521
1520 1522 boolean_t
1521 1523 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1522 1524 {
1523 1525 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1524 1526
1525 1527 ASSERT(dsl_pool_config_held(dp));
1526 1528 if (ds->ds_prev == NULL)
1527 1529 return (B_FALSE);
1528 1530 if (ds->ds_phys->ds_bp.blk_birth >
1529 1531 ds->ds_prev->ds_phys->ds_creation_txg) {
1530 1532 objset_t *os, *os_prev;
1531 1533 /*
1532 1534 * It may be that only the ZIL differs, because it was
1533 1535 * reset in the head. Don't count that as being
1534 1536 * modified.
1535 1537 */
1536 1538 if (dmu_objset_from_ds(ds, &os) != 0)
1537 1539 return (B_TRUE);
1538 1540 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
1539 1541 return (B_TRUE);
1540 1542 return (bcmp(&os->os_phys->os_meta_dnode,
1541 1543 &os_prev->os_phys->os_meta_dnode,
1542 1544 sizeof (os->os_phys->os_meta_dnode)) != 0);
1543 1545 }
1544 1546 return (B_FALSE);
1545 1547 }
1546 1548
1547 1549 typedef struct dsl_dataset_rename_snapshot_arg {
1548 1550 const char *ddrsa_fsname;
1549 1551 const char *ddrsa_oldsnapname;
1550 1552 const char *ddrsa_newsnapname;
1551 1553 boolean_t ddrsa_recursive;
1552 1554 dmu_tx_t *ddrsa_tx;
1553 1555 } dsl_dataset_rename_snapshot_arg_t;
1554 1556
1555 1557 /* ARGSUSED */
1556 1558 static int
1557 1559 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1558 1560 dsl_dataset_t *hds, void *arg)
1559 1561 {
1560 1562 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1561 1563 int error;
1562 1564 uint64_t val;
1563 1565
1564 1566 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1565 1567 if (error != 0) {
1566 1568 /* ignore nonexistent snapshots */
1567 1569 return (error == ENOENT ? 0 : error);
1568 1570 }
1569 1571
1570 1572 /* new name should not exist */
1571 1573 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1572 1574 if (error == 0)
1573 1575 error = SET_ERROR(EEXIST);
1574 1576 else if (error == ENOENT)
1575 1577 error = 0;
1576 1578
1577 1579 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1578 1580 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1579 1581 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1580 1582 error = SET_ERROR(ENAMETOOLONG);
1581 1583
1582 1584 return (error);
1583 1585 }
1584 1586
1585 1587 static int
1586 1588 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1587 1589 {
1588 1590 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1589 1591 dsl_pool_t *dp = dmu_tx_pool(tx);
1590 1592 dsl_dataset_t *hds;
1591 1593 int error;
1592 1594
1593 1595 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1594 1596 if (error != 0)
1595 1597 return (error);
1596 1598
1597 1599 if (ddrsa->ddrsa_recursive) {
1598 1600 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1599 1601 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1600 1602 DS_FIND_CHILDREN);
1601 1603 } else {
1602 1604 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1603 1605 }
1604 1606 dsl_dataset_rele(hds, FTAG);
1605 1607 return (error);
1606 1608 }
1607 1609
1608 1610 static int
1609 1611 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1610 1612 dsl_dataset_t *hds, void *arg)
1611 1613 {
1612 1614 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1613 1615 dsl_dataset_t *ds;
1614 1616 uint64_t val;
1615 1617 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1616 1618 int error;
1617 1619
1618 1620 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1619 1621 ASSERT(error == 0 || error == ENOENT);
1620 1622 if (error == ENOENT) {
1621 1623 /* ignore nonexistent snapshots */
1622 1624 return (0);
1623 1625 }
1624 1626
1625 1627 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1626 1628
1627 1629 /* log before we change the name */
1628 1630 spa_history_log_internal_ds(ds, "rename", tx,
1629 1631 "-> @%s", ddrsa->ddrsa_newsnapname);
1630 1632
1631 1633 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1632 1634 mutex_enter(&ds->ds_lock);
1633 1635 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1634 1636 mutex_exit(&ds->ds_lock);
1635 1637 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1636 1638 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1637 1639
1638 1640 dsl_dataset_rele(ds, FTAG);
1639 1641 return (0);
1640 1642 }
1641 1643
1642 1644 static void
1643 1645 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1644 1646 {
1645 1647 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1646 1648 dsl_pool_t *dp = dmu_tx_pool(tx);
1647 1649 dsl_dataset_t *hds;
1648 1650
1649 1651 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1650 1652 ddrsa->ddrsa_tx = tx;
1651 1653 if (ddrsa->ddrsa_recursive) {
1652 1654 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1653 1655 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1654 1656 DS_FIND_CHILDREN));
1655 1657 } else {
1656 1658 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1657 1659 }
1658 1660 dsl_dataset_rele(hds, FTAG);
1659 1661 }
1660 1662
1661 1663 int
1662 1664 dsl_dataset_rename_snapshot(const char *fsname,
1663 1665 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1664 1666 {
1665 1667 dsl_dataset_rename_snapshot_arg_t ddrsa;
1666 1668
1667 1669 ddrsa.ddrsa_fsname = fsname;
1668 1670 ddrsa.ddrsa_oldsnapname = oldsnapname;
1669 1671 ddrsa.ddrsa_newsnapname = newsnapname;
1670 1672 ddrsa.ddrsa_recursive = recursive;
1671 1673
1672 1674 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1673 1675 dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1674 1676 }
1675 1677
1676 1678 static int
1677 1679 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1678 1680 {
1679 1681 const char *fsname = arg;
1680 1682 dsl_pool_t *dp = dmu_tx_pool(tx);
1681 1683 dsl_dataset_t *ds;
1682 1684 int64_t unused_refres_delta;
1683 1685 int error;
1684 1686
1685 1687 error = dsl_dataset_hold(dp, fsname, FTAG, &ds);
1686 1688 if (error != 0)
1687 1689 return (error);
1688 1690
1689 1691 /* must not be a snapshot */
1690 1692 if (dsl_dataset_is_snapshot(ds)) {
1691 1693 dsl_dataset_rele(ds, FTAG);
1692 1694 return (SET_ERROR(EINVAL));
1693 1695 }
1694 1696
1695 1697 /* must have a most recent snapshot */
1696 1698 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1697 1699 dsl_dataset_rele(ds, FTAG);
1698 1700 return (SET_ERROR(EINVAL));
1699 1701 }
1700 1702
1701 1703 if (dsl_dataset_long_held(ds)) {
1702 1704 dsl_dataset_rele(ds, FTAG);
1703 1705 return (SET_ERROR(EBUSY));
1704 1706 }
1705 1707
1706 1708 /*
1707 1709 * Check if the snap we are rolling back to uses more than
1708 1710 * the refquota.
1709 1711 */
1710 1712 if (ds->ds_quota != 0 &&
1711 1713 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1712 1714 dsl_dataset_rele(ds, FTAG);
1713 1715 return (SET_ERROR(EDQUOT));
1714 1716 }
1715 1717
1716 1718 /*
1717 1719 * When we do the clone swap, we will temporarily use more space
1718 1720 * due to the refreservation (the head will no longer have any
1719 1721 * unique space, so the entire amount of the refreservation will need
1720 1722 * to be free). We will immediately destroy the clone, freeing
1721 1723 * this space, but the freeing happens over many txg's.
1722 1724 */
1723 1725 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1724 1726 ds->ds_phys->ds_unique_bytes);
1725 1727
1726 1728 if (unused_refres_delta > 0 &&
1727 1729 unused_refres_delta >
1728 1730 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1729 1731 dsl_dataset_rele(ds, FTAG);
1730 1732 return (SET_ERROR(ENOSPC));
1731 1733 }
1732 1734
1733 1735 dsl_dataset_rele(ds, FTAG);
1734 1736 return (0);
1735 1737 }
1736 1738
1737 1739 static void
1738 1740 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1739 1741 {
1740 1742 const char *fsname = arg;
1741 1743 dsl_pool_t *dp = dmu_tx_pool(tx);
1742 1744 dsl_dataset_t *ds, *clone;
1743 1745 uint64_t cloneobj;
1744 1746
1745 1747 VERIFY0(dsl_dataset_hold(dp, fsname, FTAG, &ds));
1746 1748
1747 1749 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1748 1750 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1749 1751
1750 1752 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1751 1753
1752 1754 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1753 1755 dsl_dataset_zero_zil(ds, tx);
1754 1756
1755 1757 dsl_destroy_head_sync_impl(clone, tx);
1756 1758
1757 1759 dsl_dataset_rele(clone, FTAG);
1758 1760 dsl_dataset_rele(ds, FTAG);
1759 1761 }
1760 1762
1761 1763 int
1762 1764 dsl_dataset_rollback(const char *fsname)
1763 1765 {
1764 1766 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1765 1767 dsl_dataset_rollback_sync, (void *)fsname, 1));
1766 1768 }
1767 1769
1768 1770 struct promotenode {
1769 1771 list_node_t link;
1770 1772 dsl_dataset_t *ds;
1771 1773 };
1772 1774
1773 1775 typedef struct dsl_dataset_promote_arg {
1774 1776 const char *ddpa_clonename;
1775 1777 dsl_dataset_t *ddpa_clone;
1776 1778 list_t shared_snaps, origin_snaps, clone_snaps;
1777 1779 dsl_dataset_t *origin_origin; /* origin of the origin */
1778 1780 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1779 1781 char *err_ds;
1780 1782 } dsl_dataset_promote_arg_t;
1781 1783
1782 1784 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1783 1785 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1784 1786 void *tag);
1785 1787 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1786 1788
1787 1789 static int
1788 1790 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1789 1791 {
1790 1792 dsl_dataset_promote_arg_t *ddpa = arg;
1791 1793 dsl_pool_t *dp = dmu_tx_pool(tx);
1792 1794 dsl_dataset_t *hds;
1793 1795 struct promotenode *snap;
1794 1796 dsl_dataset_t *origin_ds;
1795 1797 int err;
1796 1798 uint64_t unused;
1797 1799
1798 1800 err = promote_hold(ddpa, dp, FTAG);
1799 1801 if (err != 0)
1800 1802 return (err);
1801 1803
1802 1804 hds = ddpa->ddpa_clone;
1803 1805
1804 1806 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1805 1807 promote_rele(ddpa, FTAG);
1806 1808 return (SET_ERROR(EXDEV));
1807 1809 }
1808 1810
1809 1811 /*
1810 1812 * Compute and check the amount of space to transfer. Since this is
1811 1813 * so expensive, don't do the preliminary check.
1812 1814 */
1813 1815 if (!dmu_tx_is_syncing(tx)) {
1814 1816 promote_rele(ddpa, FTAG);
1815 1817 return (0);
1816 1818 }
1817 1819
1818 1820 snap = list_head(&ddpa->shared_snaps);
1819 1821 origin_ds = snap->ds;
1820 1822
1821 1823 /* compute origin's new unique space */
1822 1824 snap = list_tail(&ddpa->clone_snaps);
1823 1825 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1824 1826 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1825 1827 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1826 1828 &ddpa->unique, &unused, &unused);
1827 1829
1828 1830 /*
1829 1831 * Walk the snapshots that we are moving
1830 1832 *
1831 1833 * Compute space to transfer. Consider the incremental changes
1832 1834 * to used by each snapshot:
1833 1835 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1834 1836 * So each snapshot gave birth to:
1835 1837 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1836 1838 * So a sequence would look like:
1837 1839 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1838 1840 * Which simplifies to:
1839 1841 * uN + kN + kN-1 + ... + k1 + k0
1840 1842 * Note however, if we stop before we reach the ORIGIN we get:
1841 1843 * uN + kN + kN-1 + ... + kM - uM-1
1842 1844 */
1843 1845 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1844 1846 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1845 1847 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1846 1848 for (snap = list_head(&ddpa->shared_snaps); snap;
1847 1849 snap = list_next(&ddpa->shared_snaps, snap)) {
1848 1850 uint64_t val, dlused, dlcomp, dluncomp;
1849 1851 dsl_dataset_t *ds = snap->ds;
1850 1852
1851 1853 /*
1852 1854 * If there are long holds, we won't be able to evict
1853 1855 * the objset.
1854 1856 */
1855 1857 if (dsl_dataset_long_held(ds)) {
1856 1858 err = SET_ERROR(EBUSY);
1857 1859 goto out;
1858 1860 }
1859 1861
1860 1862 /* Check that the snapshot name does not conflict */
1861 1863 VERIFY0(dsl_dataset_get_snapname(ds));
1862 1864 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1863 1865 if (err == 0) {
1864 1866 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1865 1867 err = SET_ERROR(EEXIST);
1866 1868 goto out;
1867 1869 }
1868 1870 if (err != ENOENT)
1869 1871 goto out;
1870 1872
1871 1873 /* The very first snapshot does not have a deadlist */
1872 1874 if (ds->ds_phys->ds_prev_snap_obj == 0)
1873 1875 continue;
1874 1876
1875 1877 dsl_deadlist_space(&ds->ds_deadlist,
1876 1878 &dlused, &dlcomp, &dluncomp);
1877 1879 ddpa->used += dlused;
1878 1880 ddpa->comp += dlcomp;
1879 1881 ddpa->uncomp += dluncomp;
1880 1882 }
1881 1883
1882 1884 /*
1883 1885 * If we are a clone of a clone then we never reached ORIGIN,
1884 1886 * so we need to subtract out the clone origin's used space.
1885 1887 */
1886 1888 if (ddpa->origin_origin) {
1887 1889 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1888 1890 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1889 1891 ddpa->uncomp -=
1890 1892 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1891 1893 }
1892 1894
1893 1895 /* Check that there is enough space here */
1894 1896 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1895 1897 ddpa->used);
1896 1898 if (err != 0)
1897 1899 goto out;
1898 1900
1899 1901 /*
1900 1902 * Compute the amounts of space that will be used by snapshots
1901 1903 * after the promotion (for both origin and clone). For each,
1902 1904 * it is the amount of space that will be on all of their
1903 1905 * deadlists (that was not born before their new origin).
1904 1906 */
1905 1907 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1906 1908 uint64_t space;
1907 1909
1908 1910 /*
1909 1911 * Note, typically this will not be a clone of a clone,
1910 1912 * so dd_origin_txg will be < TXG_INITIAL, so
1911 1913 * these snaplist_space() -> dsl_deadlist_space_range()
1912 1914 * calls will be fast because they do not have to
1913 1915 * iterate over all bps.
1914 1916 */
1915 1917 snap = list_head(&ddpa->origin_snaps);
1916 1918 err = snaplist_space(&ddpa->shared_snaps,
1917 1919 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1918 1920 if (err != 0)
1919 1921 goto out;
1920 1922
1921 1923 err = snaplist_space(&ddpa->clone_snaps,
1922 1924 snap->ds->ds_dir->dd_origin_txg, &space);
1923 1925 if (err != 0)
1924 1926 goto out;
1925 1927 ddpa->cloneusedsnap += space;
1926 1928 }
1927 1929 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1928 1930 err = snaplist_space(&ddpa->origin_snaps,
1929 1931 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
1930 1932 if (err != 0)
1931 1933 goto out;
1932 1934 }
1933 1935
1934 1936 out:
1935 1937 promote_rele(ddpa, FTAG);
1936 1938 return (err);
1937 1939 }
1938 1940
1939 1941 static void
1940 1942 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
1941 1943 {
1942 1944 dsl_dataset_promote_arg_t *ddpa = arg;
1943 1945 dsl_pool_t *dp = dmu_tx_pool(tx);
1944 1946 dsl_dataset_t *hds;
1945 1947 struct promotenode *snap;
1946 1948 dsl_dataset_t *origin_ds;
1947 1949 dsl_dataset_t *origin_head;
1948 1950 dsl_dir_t *dd;
1949 1951 dsl_dir_t *odd = NULL;
1950 1952 uint64_t oldnext_obj;
1951 1953 int64_t delta;
1952 1954
1953 1955 VERIFY0(promote_hold(ddpa, dp, FTAG));
1954 1956 hds = ddpa->ddpa_clone;
1955 1957
1956 1958 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
1957 1959
1958 1960 snap = list_head(&ddpa->shared_snaps);
1959 1961 origin_ds = snap->ds;
1960 1962 dd = hds->ds_dir;
1961 1963
1962 1964 snap = list_head(&ddpa->origin_snaps);
1963 1965 origin_head = snap->ds;
1964 1966
1965 1967 /*
1966 1968 * We need to explicitly open odd, since origin_ds's dd will be
1967 1969 * changing.
1968 1970 */
1969 1971 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
1970 1972 NULL, FTAG, &odd));
1971 1973
1972 1974 /* change origin's next snap */
1973 1975 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
1974 1976 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
1975 1977 snap = list_tail(&ddpa->clone_snaps);
1976 1978 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1977 1979 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
1978 1980
1979 1981 /* change the origin's next clone */
1980 1982 if (origin_ds->ds_phys->ds_next_clones_obj) {
1981 1983 dsl_dataset_remove_from_next_clones(origin_ds,
1982 1984 snap->ds->ds_object, tx);
1983 1985 VERIFY0(zap_add_int(dp->dp_meta_objset,
1984 1986 origin_ds->ds_phys->ds_next_clones_obj,
1985 1987 oldnext_obj, tx));
1986 1988 }
1987 1989
1988 1990 /* change origin */
1989 1991 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1990 1992 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
1991 1993 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
1992 1994 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
1993 1995 dmu_buf_will_dirty(odd->dd_dbuf, tx);
1994 1996 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
1995 1997 origin_head->ds_dir->dd_origin_txg =
1996 1998 origin_ds->ds_phys->ds_creation_txg;
1997 1999
1998 2000 /* change dd_clone entries */
1999 2001 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2000 2002 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2001 2003 odd->dd_phys->dd_clones, hds->ds_object, tx));
2002 2004 VERIFY0(zap_add_int(dp->dp_meta_objset,
2003 2005 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2004 2006 hds->ds_object, tx));
2005 2007
2006 2008 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2007 2009 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2008 2010 origin_head->ds_object, tx));
2009 2011 if (dd->dd_phys->dd_clones == 0) {
2010 2012 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2011 2013 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2012 2014 }
2013 2015 VERIFY0(zap_add_int(dp->dp_meta_objset,
2014 2016 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2015 2017 }
2016 2018
2017 2019 /* move snapshots to this dir */
2018 2020 for (snap = list_head(&ddpa->shared_snaps); snap;
2019 2021 snap = list_next(&ddpa->shared_snaps, snap)) {
2020 2022 dsl_dataset_t *ds = snap->ds;
2021 2023
2022 2024 /*
2023 2025 * Property callbacks are registered to a particular
2024 2026 * dsl_dir. Since ours is changing, evict the objset
2025 2027 * so that they will be unregistered from the old dsl_dir.
2026 2028 */
2027 2029 if (ds->ds_objset) {
2028 2030 dmu_objset_evict(ds->ds_objset);
2029 2031 ds->ds_objset = NULL;
2030 2032 }
2031 2033
2032 2034 /* move snap name entry */
2033 2035 VERIFY0(dsl_dataset_get_snapname(ds));
2034 2036 VERIFY0(dsl_dataset_snap_remove(origin_head,
2035 2037 ds->ds_snapname, tx));
2036 2038 VERIFY0(zap_add(dp->dp_meta_objset,
2037 2039 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2038 2040 8, 1, &ds->ds_object, tx));
2039 2041
2040 2042 /* change containing dsl_dir */
2041 2043 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2042 2044 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2043 2045 ds->ds_phys->ds_dir_obj = dd->dd_object;
2044 2046 ASSERT3P(ds->ds_dir, ==, odd);
2045 2047 dsl_dir_rele(ds->ds_dir, ds);
2046 2048 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2047 2049 NULL, ds, &ds->ds_dir));
2048 2050
2049 2051 /* move any clone references */
2050 2052 if (ds->ds_phys->ds_next_clones_obj &&
2051 2053 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2052 2054 zap_cursor_t zc;
2053 2055 zap_attribute_t za;
2054 2056
2055 2057 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2056 2058 ds->ds_phys->ds_next_clones_obj);
2057 2059 zap_cursor_retrieve(&zc, &za) == 0;
2058 2060 zap_cursor_advance(&zc)) {
2059 2061 dsl_dataset_t *cnds;
2060 2062 uint64_t o;
2061 2063
2062 2064 if (za.za_first_integer == oldnext_obj) {
2063 2065 /*
2064 2066 * We've already moved the
2065 2067 * origin's reference.
2066 2068 */
2067 2069 continue;
2068 2070 }
2069 2071
2070 2072 VERIFY0(dsl_dataset_hold_obj(dp,
2071 2073 za.za_first_integer, FTAG, &cnds));
2072 2074 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2073 2075
2074 2076 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2075 2077 odd->dd_phys->dd_clones, o, tx));
2076 2078 VERIFY0(zap_add_int(dp->dp_meta_objset,
2077 2079 dd->dd_phys->dd_clones, o, tx));
2078 2080 dsl_dataset_rele(cnds, FTAG);
2079 2081 }
2080 2082 zap_cursor_fini(&zc);
2081 2083 }
2082 2084
2083 2085 ASSERT(!dsl_prop_hascb(ds));
2084 2086 }
2085 2087
2086 2088 /*
2087 2089 * Change space accounting.
2088 2090 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2089 2091 * both be valid, or both be 0 (resulting in delta == 0). This
2090 2092 * is true for each of {clone,origin} independently.
2091 2093 */
2092 2094
2093 2095 delta = ddpa->cloneusedsnap -
2094 2096 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2095 2097 ASSERT3S(delta, >=, 0);
2096 2098 ASSERT3U(ddpa->used, >=, delta);
2097 2099 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2098 2100 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2099 2101 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2100 2102
2101 2103 delta = ddpa->originusedsnap -
2102 2104 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2103 2105 ASSERT3S(delta, <=, 0);
2104 2106 ASSERT3U(ddpa->used, >=, -delta);
2105 2107 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2106 2108 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2107 2109 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2108 2110
2109 2111 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2110 2112
2111 2113 /* log history record */
2112 2114 spa_history_log_internal_ds(hds, "promote", tx, "");
2113 2115
2114 2116 dsl_dir_rele(odd, FTAG);
2115 2117 promote_rele(ddpa, FTAG);
2116 2118 }
2117 2119
2118 2120 /*
2119 2121 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2120 2122 * (exclusive) and last_obj (inclusive). The list will be in reverse
2121 2123 * order (last_obj will be the list_head()). If first_obj == 0, do all
2122 2124 * snapshots back to this dataset's origin.
2123 2125 */
2124 2126 static int
2125 2127 snaplist_make(dsl_pool_t *dp,
2126 2128 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2127 2129 {
2128 2130 uint64_t obj = last_obj;
2129 2131
2130 2132 list_create(l, sizeof (struct promotenode),
2131 2133 offsetof(struct promotenode, link));
2132 2134
2133 2135 while (obj != first_obj) {
2134 2136 dsl_dataset_t *ds;
2135 2137 struct promotenode *snap;
2136 2138 int err;
2137 2139
2138 2140 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2139 2141 ASSERT(err != ENOENT);
2140 2142 if (err != 0)
2141 2143 return (err);
2142 2144
2143 2145 if (first_obj == 0)
2144 2146 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2145 2147
2146 2148 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2147 2149 snap->ds = ds;
2148 2150 list_insert_tail(l, snap);
2149 2151 obj = ds->ds_phys->ds_prev_snap_obj;
2150 2152 }
2151 2153
2152 2154 return (0);
2153 2155 }
2154 2156
2155 2157 static int
2156 2158 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2157 2159 {
2158 2160 struct promotenode *snap;
2159 2161
2160 2162 *spacep = 0;
2161 2163 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2162 2164 uint64_t used, comp, uncomp;
2163 2165 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2164 2166 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2165 2167 *spacep += used;
2166 2168 }
2167 2169 return (0);
2168 2170 }
2169 2171
2170 2172 static void
2171 2173 snaplist_destroy(list_t *l, void *tag)
2172 2174 {
2173 2175 struct promotenode *snap;
2174 2176
2175 2177 if (l == NULL || !list_link_active(&l->list_head))
2176 2178 return;
2177 2179
2178 2180 while ((snap = list_tail(l)) != NULL) {
2179 2181 list_remove(l, snap);
2180 2182 dsl_dataset_rele(snap->ds, tag);
2181 2183 kmem_free(snap, sizeof (*snap));
2182 2184 }
2183 2185 list_destroy(l);
2184 2186 }
2185 2187
2186 2188 static int
2187 2189 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2188 2190 {
2189 2191 int error;
2190 2192 dsl_dir_t *dd;
2191 2193 struct promotenode *snap;
2192 2194
2193 2195 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2194 2196 &ddpa->ddpa_clone);
2195 2197 if (error != 0)
2196 2198 return (error);
2197 2199 dd = ddpa->ddpa_clone->ds_dir;
2198 2200
2199 2201 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2200 2202 !dsl_dir_is_clone(dd)) {
2201 2203 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2202 2204 return (SET_ERROR(EINVAL));
2203 2205 }
2204 2206
2205 2207 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2206 2208 &ddpa->shared_snaps, tag);
2207 2209 if (error != 0)
2208 2210 goto out;
2209 2211
2210 2212 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2211 2213 &ddpa->clone_snaps, tag);
2212 2214 if (error != 0)
2213 2215 goto out;
2214 2216
2215 2217 snap = list_head(&ddpa->shared_snaps);
2216 2218 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2217 2219 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2218 2220 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2219 2221 &ddpa->origin_snaps, tag);
2220 2222 if (error != 0)
2221 2223 goto out;
2222 2224
2223 2225 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2224 2226 error = dsl_dataset_hold_obj(dp,
2225 2227 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2226 2228 tag, &ddpa->origin_origin);
2227 2229 if (error != 0)
2228 2230 goto out;
2229 2231 }
2230 2232 out:
2231 2233 if (error != 0)
2232 2234 promote_rele(ddpa, tag);
2233 2235 return (error);
2234 2236 }
2235 2237
2236 2238 static void
2237 2239 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2238 2240 {
2239 2241 snaplist_destroy(&ddpa->shared_snaps, tag);
2240 2242 snaplist_destroy(&ddpa->clone_snaps, tag);
2241 2243 snaplist_destroy(&ddpa->origin_snaps, tag);
2242 2244 if (ddpa->origin_origin != NULL)
2243 2245 dsl_dataset_rele(ddpa->origin_origin, tag);
2244 2246 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2245 2247 }
2246 2248
2247 2249 /*
2248 2250 * Promote a clone.
2249 2251 *
2250 2252 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2251 2253 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2252 2254 */
2253 2255 int
2254 2256 dsl_dataset_promote(const char *name, char *conflsnap)
2255 2257 {
2256 2258 dsl_dataset_promote_arg_t ddpa = { 0 };
2257 2259 uint64_t numsnaps;
2258 2260 int error;
2259 2261 objset_t *os;
2260 2262
2261 2263 /*
2262 2264 * We will modify space proportional to the number of
2263 2265 * snapshots. Compute numsnaps.
2264 2266 */
2265 2267 error = dmu_objset_hold(name, FTAG, &os);
2266 2268 if (error != 0)
2267 2269 return (error);
2268 2270 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2269 2271 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2270 2272 dmu_objset_rele(os, FTAG);
2271 2273 if (error != 0)
2272 2274 return (error);
2273 2275
2274 2276 ddpa.ddpa_clonename = name;
2275 2277 ddpa.err_ds = conflsnap;
2276 2278
2277 2279 return (dsl_sync_task(name, dsl_dataset_promote_check,
2278 2280 dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2279 2281 }
2280 2282
2281 2283 int
2282 2284 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2283 2285 dsl_dataset_t *origin_head, boolean_t force)
2284 2286 {
2285 2287 int64_t unused_refres_delta;
2286 2288
2287 2289 /* they should both be heads */
2288 2290 if (dsl_dataset_is_snapshot(clone) ||
2289 2291 dsl_dataset_is_snapshot(origin_head))
2290 2292 return (SET_ERROR(EINVAL));
2291 2293
2292 2294 /* the branch point should be just before them */
2293 2295 if (clone->ds_prev != origin_head->ds_prev)
2294 2296 return (SET_ERROR(EINVAL));
2295 2297
2296 2298 /* clone should be the clone (unless they are unrelated) */
2297 2299 if (clone->ds_prev != NULL &&
2298 2300 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2299 2301 origin_head->ds_object !=
2300 2302 clone->ds_prev->ds_phys->ds_next_snap_obj)
2301 2303 return (SET_ERROR(EINVAL));
2302 2304
2303 2305 /* the clone should be a child of the origin */
2304 2306 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2305 2307 return (SET_ERROR(EINVAL));
2306 2308
2307 2309 /* origin_head shouldn't be modified unless 'force' */
2308 2310 if (!force && dsl_dataset_modified_since_lastsnap(origin_head))
2309 2311 return (SET_ERROR(ETXTBSY));
2310 2312
2311 2313 /* origin_head should have no long holds (e.g. is not mounted) */
2312 2314 if (dsl_dataset_long_held(origin_head))
2313 2315 return (SET_ERROR(EBUSY));
2314 2316
2315 2317 /* check amount of any unconsumed refreservation */
2316 2318 unused_refres_delta =
2317 2319 (int64_t)MIN(origin_head->ds_reserved,
2318 2320 origin_head->ds_phys->ds_unique_bytes) -
2319 2321 (int64_t)MIN(origin_head->ds_reserved,
2320 2322 clone->ds_phys->ds_unique_bytes);
2321 2323
2322 2324 if (unused_refres_delta > 0 &&
2323 2325 unused_refres_delta >
2324 2326 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2325 2327 return (SET_ERROR(ENOSPC));
2326 2328
2327 2329 /* clone can't be over the head's refquota */
2328 2330 if (origin_head->ds_quota != 0 &&
2329 2331 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2330 2332 return (SET_ERROR(EDQUOT));
2331 2333
2332 2334 return (0);
2333 2335 }
2334 2336
2335 2337 void
2336 2338 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2337 2339 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2338 2340 {
2339 2341 dsl_pool_t *dp = dmu_tx_pool(tx);
2340 2342 int64_t unused_refres_delta;
2341 2343
2342 2344 ASSERT(clone->ds_reserved == 0);
2343 2345 ASSERT(origin_head->ds_quota == 0 ||
2344 2346 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2345 2347
2346 2348 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2347 2349 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2348 2350
2349 2351 if (clone->ds_objset != NULL) {
2350 2352 dmu_objset_evict(clone->ds_objset);
2351 2353 clone->ds_objset = NULL;
2352 2354 }
2353 2355
2354 2356 if (origin_head->ds_objset != NULL) {
2355 2357 dmu_objset_evict(origin_head->ds_objset);
2356 2358 origin_head->ds_objset = NULL;
2357 2359 }
2358 2360
2359 2361 unused_refres_delta =
2360 2362 (int64_t)MIN(origin_head->ds_reserved,
2361 2363 origin_head->ds_phys->ds_unique_bytes) -
2362 2364 (int64_t)MIN(origin_head->ds_reserved,
2363 2365 clone->ds_phys->ds_unique_bytes);
2364 2366
2365 2367 /*
2366 2368 * Reset origin's unique bytes, if it exists.
2367 2369 */
2368 2370 if (clone->ds_prev) {
2369 2371 dsl_dataset_t *origin = clone->ds_prev;
2370 2372 uint64_t comp, uncomp;
2371 2373
2372 2374 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2373 2375 dsl_deadlist_space_range(&clone->ds_deadlist,
2374 2376 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2375 2377 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2376 2378 }
2377 2379
2378 2380 /* swap blkptrs */
2379 2381 {
2380 2382 blkptr_t tmp;
2381 2383 tmp = origin_head->ds_phys->ds_bp;
2382 2384 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2383 2385 clone->ds_phys->ds_bp = tmp;
2384 2386 }
2385 2387
2386 2388 /* set dd_*_bytes */
2387 2389 {
2388 2390 int64_t dused, dcomp, duncomp;
2389 2391 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2390 2392 uint64_t odl_used, odl_comp, odl_uncomp;
2391 2393
2392 2394 ASSERT3U(clone->ds_dir->dd_phys->
2393 2395 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2394 2396
2395 2397 dsl_deadlist_space(&clone->ds_deadlist,
2396 2398 &cdl_used, &cdl_comp, &cdl_uncomp);
2397 2399 dsl_deadlist_space(&origin_head->ds_deadlist,
2398 2400 &odl_used, &odl_comp, &odl_uncomp);
2399 2401
2400 2402 dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2401 2403 (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2402 2404 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2403 2405 (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2404 2406 duncomp = clone->ds_phys->ds_uncompressed_bytes +
2405 2407 cdl_uncomp -
2406 2408 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2407 2409
2408 2410 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2409 2411 dused, dcomp, duncomp, tx);
2410 2412 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2411 2413 -dused, -dcomp, -duncomp, tx);
2412 2414
2413 2415 /*
2414 2416 * The difference in the space used by snapshots is the
2415 2417 * difference in snapshot space due to the head's
2416 2418 * deadlist (since that's the only thing that's
2417 2419 * changing that affects the snapused).
2418 2420 */
2419 2421 dsl_deadlist_space_range(&clone->ds_deadlist,
2420 2422 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2421 2423 &cdl_used, &cdl_comp, &cdl_uncomp);
2422 2424 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2423 2425 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2424 2426 &odl_used, &odl_comp, &odl_uncomp);
2425 2427 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2426 2428 DD_USED_HEAD, DD_USED_SNAP, tx);
2427 2429 }
2428 2430
2429 2431 /* swap ds_*_bytes */
2430 2432 SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2431 2433 clone->ds_phys->ds_referenced_bytes);
2432 2434 SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2433 2435 clone->ds_phys->ds_compressed_bytes);
2434 2436 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2435 2437 clone->ds_phys->ds_uncompressed_bytes);
2436 2438 SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2437 2439 clone->ds_phys->ds_unique_bytes);
2438 2440
2439 2441 /* apply any parent delta for change in unconsumed refreservation */
2440 2442 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2441 2443 unused_refres_delta, 0, 0, tx);
2442 2444
2443 2445 /*
2444 2446 * Swap deadlists.
2445 2447 */
2446 2448 dsl_deadlist_close(&clone->ds_deadlist);
2447 2449 dsl_deadlist_close(&origin_head->ds_deadlist);
2448 2450 SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2449 2451 clone->ds_phys->ds_deadlist_obj);
2450 2452 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2451 2453 clone->ds_phys->ds_deadlist_obj);
2452 2454 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2453 2455 origin_head->ds_phys->ds_deadlist_obj);
2454 2456
2455 2457 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2456 2458
2457 2459 spa_history_log_internal_ds(clone, "clone swap", tx,
2458 2460 "parent=%s", origin_head->ds_dir->dd_myname);
2459 2461 }
2460 2462
2461 2463 /*
2462 2464 * Given a pool name and a dataset object number in that pool,
2463 2465 * return the name of that dataset.
2464 2466 */
2465 2467 int
2466 2468 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2467 2469 {
2468 2470 dsl_pool_t *dp;
2469 2471 dsl_dataset_t *ds;
2470 2472 int error;
2471 2473
2472 2474 error = dsl_pool_hold(pname, FTAG, &dp);
2473 2475 if (error != 0)
2474 2476 return (error);
2475 2477
2476 2478 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2477 2479 if (error == 0) {
2478 2480 dsl_dataset_name(ds, buf);
2479 2481 dsl_dataset_rele(ds, FTAG);
2480 2482 }
2481 2483 dsl_pool_rele(dp, FTAG);
2482 2484
2483 2485 return (error);
2484 2486 }
2485 2487
2486 2488 int
2487 2489 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2488 2490 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2489 2491 {
2490 2492 int error = 0;
2491 2493
2492 2494 ASSERT3S(asize, >, 0);
2493 2495
2494 2496 /*
2495 2497 * *ref_rsrv is the portion of asize that will come from any
2496 2498 * unconsumed refreservation space.
2497 2499 */
2498 2500 *ref_rsrv = 0;
2499 2501
2500 2502 mutex_enter(&ds->ds_lock);
2501 2503 /*
2502 2504 * Make a space adjustment for reserved bytes.
2503 2505 */
2504 2506 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2505 2507 ASSERT3U(*used, >=,
2506 2508 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2507 2509 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2508 2510 *ref_rsrv =
2509 2511 asize - MIN(asize, parent_delta(ds, asize + inflight));
2510 2512 }
2511 2513
2512 2514 if (!check_quota || ds->ds_quota == 0) {
2513 2515 mutex_exit(&ds->ds_lock);
2514 2516 return (0);
2515 2517 }
2516 2518 /*
2517 2519 * If they are requesting more space, and our current estimate
2518 2520 * is over quota, they get to try again unless the actual
2519 2521 * on-disk is over quota and there are no pending changes (which
2520 2522 * may free up space for us).
2521 2523 */
2522 2524 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2523 2525 if (inflight > 0 ||
2524 2526 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2525 2527 error = SET_ERROR(ERESTART);
2526 2528 else
2527 2529 error = SET_ERROR(EDQUOT);
2528 2530 }
2529 2531 mutex_exit(&ds->ds_lock);
2530 2532
2531 2533 return (error);
2532 2534 }
2533 2535
2534 2536 typedef struct dsl_dataset_set_qr_arg {
2535 2537 const char *ddsqra_name;
2536 2538 zprop_source_t ddsqra_source;
2537 2539 uint64_t ddsqra_value;
2538 2540 } dsl_dataset_set_qr_arg_t;
2539 2541
2540 2542
2541 2543 /* ARGSUSED */
2542 2544 static int
2543 2545 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2544 2546 {
2545 2547 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2546 2548 dsl_pool_t *dp = dmu_tx_pool(tx);
2547 2549 dsl_dataset_t *ds;
2548 2550 int error;
2549 2551 uint64_t newval;
2550 2552
2551 2553 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2552 2554 return (SET_ERROR(ENOTSUP));
2553 2555
2554 2556 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2555 2557 if (error != 0)
2556 2558 return (error);
2557 2559
2558 2560 if (dsl_dataset_is_snapshot(ds)) {
2559 2561 dsl_dataset_rele(ds, FTAG);
2560 2562 return (SET_ERROR(EINVAL));
2561 2563 }
2562 2564
2563 2565 error = dsl_prop_predict(ds->ds_dir,
2564 2566 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2565 2567 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2566 2568 if (error != 0) {
2567 2569 dsl_dataset_rele(ds, FTAG);
2568 2570 return (error);
2569 2571 }
2570 2572
2571 2573 if (newval == 0) {
2572 2574 dsl_dataset_rele(ds, FTAG);
2573 2575 return (0);
2574 2576 }
2575 2577
2576 2578 if (newval < ds->ds_phys->ds_referenced_bytes ||
2577 2579 newval < ds->ds_reserved) {
2578 2580 dsl_dataset_rele(ds, FTAG);
2579 2581 return (SET_ERROR(ENOSPC));
2580 2582 }
2581 2583
2582 2584 dsl_dataset_rele(ds, FTAG);
2583 2585 return (0);
2584 2586 }
2585 2587
2586 2588 static void
2587 2589 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2588 2590 {
2589 2591 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2590 2592 dsl_pool_t *dp = dmu_tx_pool(tx);
2591 2593 dsl_dataset_t *ds;
2592 2594 uint64_t newval;
2593 2595
2594 2596 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2595 2597
2596 2598 dsl_prop_set_sync_impl(ds,
2597 2599 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2598 2600 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2599 2601 &ddsqra->ddsqra_value, tx);
2600 2602
2601 2603 VERIFY0(dsl_prop_get_int_ds(ds,
2602 2604 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2603 2605
2604 2606 if (ds->ds_quota != newval) {
2605 2607 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2606 2608 ds->ds_quota = newval;
2607 2609 }
2608 2610 dsl_dataset_rele(ds, FTAG);
2609 2611 }
2610 2612
2611 2613 int
2612 2614 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2613 2615 uint64_t refquota)
2614 2616 {
2615 2617 dsl_dataset_set_qr_arg_t ddsqra;
2616 2618
2617 2619 ddsqra.ddsqra_name = dsname;
2618 2620 ddsqra.ddsqra_source = source;
2619 2621 ddsqra.ddsqra_value = refquota;
2620 2622
2621 2623 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2622 2624 dsl_dataset_set_refquota_sync, &ddsqra, 0));
2623 2625 }
2624 2626
2625 2627 static int
2626 2628 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2627 2629 {
2628 2630 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2629 2631 dsl_pool_t *dp = dmu_tx_pool(tx);
2630 2632 dsl_dataset_t *ds;
2631 2633 int error;
2632 2634 uint64_t newval, unique;
2633 2635
2634 2636 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2635 2637 return (SET_ERROR(ENOTSUP));
2636 2638
2637 2639 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2638 2640 if (error != 0)
2639 2641 return (error);
2640 2642
2641 2643 if (dsl_dataset_is_snapshot(ds)) {
2642 2644 dsl_dataset_rele(ds, FTAG);
2643 2645 return (SET_ERROR(EINVAL));
2644 2646 }
2645 2647
2646 2648 error = dsl_prop_predict(ds->ds_dir,
2647 2649 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2648 2650 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2649 2651 if (error != 0) {
2650 2652 dsl_dataset_rele(ds, FTAG);
2651 2653 return (error);
2652 2654 }
2653 2655
2654 2656 /*
2655 2657 * If we are doing the preliminary check in open context, the
2656 2658 * space estimates may be inaccurate.
2657 2659 */
2658 2660 if (!dmu_tx_is_syncing(tx)) {
2659 2661 dsl_dataset_rele(ds, FTAG);
2660 2662 return (0);
2661 2663 }
2662 2664
2663 2665 mutex_enter(&ds->ds_lock);
2664 2666 if (!DS_UNIQUE_IS_ACCURATE(ds))
2665 2667 dsl_dataset_recalc_head_uniq(ds);
2666 2668 unique = ds->ds_phys->ds_unique_bytes;
2667 2669 mutex_exit(&ds->ds_lock);
2668 2670
2669 2671 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2670 2672 uint64_t delta = MAX(unique, newval) -
2671 2673 MAX(unique, ds->ds_reserved);
2672 2674
2673 2675 if (delta >
2674 2676 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2675 2677 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2676 2678 dsl_dataset_rele(ds, FTAG);
2677 2679 return (SET_ERROR(ENOSPC));
2678 2680 }
2679 2681 }
2680 2682
2681 2683 dsl_dataset_rele(ds, FTAG);
2682 2684 return (0);
2683 2685 }
2684 2686
2685 2687 void
2686 2688 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2687 2689 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2688 2690 {
2689 2691 uint64_t newval;
2690 2692 uint64_t unique;
2691 2693 int64_t delta;
2692 2694
2693 2695 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2694 2696 source, sizeof (value), 1, &value, tx);
2695 2697
2696 2698 VERIFY0(dsl_prop_get_int_ds(ds,
2697 2699 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2698 2700
2699 2701 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2700 2702 mutex_enter(&ds->ds_dir->dd_lock);
2701 2703 mutex_enter(&ds->ds_lock);
2702 2704 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2703 2705 unique = ds->ds_phys->ds_unique_bytes;
2704 2706 delta = MAX(0, (int64_t)(newval - unique)) -
2705 2707 MAX(0, (int64_t)(ds->ds_reserved - unique));
2706 2708 ds->ds_reserved = newval;
2707 2709 mutex_exit(&ds->ds_lock);
2708 2710
2709 2711 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2710 2712 mutex_exit(&ds->ds_dir->dd_lock);
2711 2713 }
2712 2714
2713 2715 static void
2714 2716 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2715 2717 {
2716 2718 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2717 2719 dsl_pool_t *dp = dmu_tx_pool(tx);
2718 2720 dsl_dataset_t *ds;
2719 2721
2720 2722 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2721 2723 dsl_dataset_set_refreservation_sync_impl(ds,
2722 2724 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2723 2725 dsl_dataset_rele(ds, FTAG);
2724 2726 }
2725 2727
2726 2728 int
2727 2729 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2728 2730 uint64_t refreservation)
2729 2731 {
2730 2732 dsl_dataset_set_qr_arg_t ddsqra;
2731 2733
2732 2734 ddsqra.ddsqra_name = dsname;
2733 2735 ddsqra.ddsqra_source = source;
2734 2736 ddsqra.ddsqra_value = refreservation;
2735 2737
2736 2738 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2737 2739 dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2738 2740 }
2739 2741
2740 2742 /*
2741 2743 * Return (in *usedp) the amount of space written in new that is not
2742 2744 * present in oldsnap. New may be a snapshot or the head. Old must be
2743 2745 * a snapshot before new, in new's filesystem (or its origin). If not then
2744 2746 * fail and return EINVAL.
2745 2747 *
2746 2748 * The written space is calculated by considering two components: First, we
2747 2749 * ignore any freed space, and calculate the written as new's used space
2748 2750 * minus old's used space. Next, we add in the amount of space that was freed
2749 2751 * between the two snapshots, thus reducing new's used space relative to old's.
2750 2752 * Specifically, this is the space that was born before old->ds_creation_txg,
2751 2753 * and freed before new (ie. on new's deadlist or a previous deadlist).
2752 2754 *
2753 2755 * space freed [---------------------]
2754 2756 * snapshots ---O-------O--------O-------O------
2755 2757 * oldsnap new
2756 2758 */
2757 2759 int
2758 2760 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2759 2761 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2760 2762 {
2761 2763 int err = 0;
2762 2764 uint64_t snapobj;
2763 2765 dsl_pool_t *dp = new->ds_dir->dd_pool;
2764 2766
2765 2767 ASSERT(dsl_pool_config_held(dp));
2766 2768
2767 2769 *usedp = 0;
2768 2770 *usedp += new->ds_phys->ds_referenced_bytes;
2769 2771 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2770 2772
2771 2773 *compp = 0;
2772 2774 *compp += new->ds_phys->ds_compressed_bytes;
2773 2775 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
2774 2776
2775 2777 *uncompp = 0;
2776 2778 *uncompp += new->ds_phys->ds_uncompressed_bytes;
2777 2779 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2778 2780
2779 2781 snapobj = new->ds_object;
2780 2782 while (snapobj != oldsnap->ds_object) {
2781 2783 dsl_dataset_t *snap;
2782 2784 uint64_t used, comp, uncomp;
2783 2785
2784 2786 if (snapobj == new->ds_object) {
2785 2787 snap = new;
2786 2788 } else {
2787 2789 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2788 2790 if (err != 0)
2789 2791 break;
2790 2792 }
2791 2793
2792 2794 if (snap->ds_phys->ds_prev_snap_txg ==
2793 2795 oldsnap->ds_phys->ds_creation_txg) {
2794 2796 /*
2795 2797 * The blocks in the deadlist can not be born after
2796 2798 * ds_prev_snap_txg, so get the whole deadlist space,
2797 2799 * which is more efficient (especially for old-format
2798 2800 * deadlists). Unfortunately the deadlist code
2799 2801 * doesn't have enough information to make this
2800 2802 * optimization itself.
2801 2803 */
2802 2804 dsl_deadlist_space(&snap->ds_deadlist,
2803 2805 &used, &comp, &uncomp);
2804 2806 } else {
2805 2807 dsl_deadlist_space_range(&snap->ds_deadlist,
2806 2808 0, oldsnap->ds_phys->ds_creation_txg,
2807 2809 &used, &comp, &uncomp);
2808 2810 }
2809 2811 *usedp += used;
2810 2812 *compp += comp;
2811 2813 *uncompp += uncomp;
2812 2814
2813 2815 /*
2814 2816 * If we get to the beginning of the chain of snapshots
2815 2817 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2816 2818 * was not a snapshot of/before new.
2817 2819 */
2818 2820 snapobj = snap->ds_phys->ds_prev_snap_obj;
2819 2821 if (snap != new)
2820 2822 dsl_dataset_rele(snap, FTAG);
2821 2823 if (snapobj == 0) {
2822 2824 err = SET_ERROR(EINVAL);
2823 2825 break;
2824 2826 }
2825 2827
2826 2828 }
2827 2829 return (err);
2828 2830 }
2829 2831
2830 2832 /*
2831 2833 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2832 2834 * lastsnap, and all snapshots in between are deleted.
2833 2835 *
2834 2836 * blocks that would be freed [---------------------------]
2835 2837 * snapshots ---O-------O--------O-------O--------O
2836 2838 * firstsnap lastsnap
2837 2839 *
2838 2840 * This is the set of blocks that were born after the snap before firstsnap,
2839 2841 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2840 2842 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2841 2843 * We calculate this by iterating over the relevant deadlists (from the snap
2842 2844 * after lastsnap, backward to the snap after firstsnap), summing up the
2843 2845 * space on the deadlist that was born after the snap before firstsnap.
2844 2846 */
2845 2847 int
2846 2848 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2847 2849 dsl_dataset_t *lastsnap,
2848 2850 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2849 2851 {
2850 2852 int err = 0;
2851 2853 uint64_t snapobj;
2852 2854 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2853 2855
2854 2856 ASSERT(dsl_dataset_is_snapshot(firstsnap));
2855 2857 ASSERT(dsl_dataset_is_snapshot(lastsnap));
2856 2858
2857 2859 /*
2858 2860 * Check that the snapshots are in the same dsl_dir, and firstsnap
2859 2861 * is before lastsnap.
2860 2862 */
2861 2863 if (firstsnap->ds_dir != lastsnap->ds_dir ||
2862 2864 firstsnap->ds_phys->ds_creation_txg >
2863 2865 lastsnap->ds_phys->ds_creation_txg)
2864 2866 return (SET_ERROR(EINVAL));
2865 2867
2866 2868 *usedp = *compp = *uncompp = 0;
2867 2869
2868 2870 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2869 2871 while (snapobj != firstsnap->ds_object) {
2870 2872 dsl_dataset_t *ds;
2871 2873 uint64_t used, comp, uncomp;
2872 2874
2873 2875 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2874 2876 if (err != 0)
2875 2877 break;
2876 2878
2877 2879 dsl_deadlist_space_range(&ds->ds_deadlist,
2878 2880 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2879 2881 &used, &comp, &uncomp);
2880 2882 *usedp += used;
2881 2883 *compp += comp;
2882 2884 *uncompp += uncomp;
2883 2885
2884 2886 snapobj = ds->ds_phys->ds_prev_snap_obj;
2885 2887 ASSERT3U(snapobj, !=, 0);
2886 2888 dsl_dataset_rele(ds, FTAG);
2887 2889 }
2888 2890 return (err);
2889 2891 }
2890 2892
2891 2893 /*
2892 2894 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2893 2895 * For example, they could both be snapshots of the same filesystem, and
2894 2896 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
2895 2897 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
2896 2898 * filesystem. Or 'earlier' could be the origin's origin.
2897 2899 */
2898 2900 boolean_t
2899 2901 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2900 2902 {
2901 2903 dsl_pool_t *dp = later->ds_dir->dd_pool;
2902 2904 int error;
2903 2905 boolean_t ret;
2904 2906
2905 2907 ASSERT(dsl_pool_config_held(dp));
2906 2908
2907 2909 if (earlier->ds_phys->ds_creation_txg >=
2908 2910 later->ds_phys->ds_creation_txg)
2909 2911 return (B_FALSE);
2910 2912
2911 2913 if (later->ds_dir == earlier->ds_dir)
2912 2914 return (B_TRUE);
2913 2915 if (!dsl_dir_is_clone(later->ds_dir))
2914 2916 return (B_FALSE);
2915 2917
2916 2918 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2917 2919 return (B_TRUE);
2918 2920 dsl_dataset_t *origin;
2919 2921 error = dsl_dataset_hold_obj(dp,
2920 2922 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2921 2923 if (error != 0)
2922 2924 return (B_FALSE);
2923 2925 ret = dsl_dataset_is_before(origin, earlier);
2924 2926 dsl_dataset_rele(origin, FTAG);
2925 2927 return (ret);
2926 2928 }
|
↓ open down ↓ |
2556 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX