Print this page
2948 do not generate 0 as random 64-bit guid
| 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) 2012 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
49 49 static char *dsl_reaper = "the grim reaper";
50 50
51 51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
52 52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
53 53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
54 54
55 55 #define SWITCH64(x, y) \
56 56 { \
57 57 uint64_t __tmp = (x); \
58 58 (x) = (y); \
59 59 (y) = __tmp; \
60 60 }
61 61
62 62 #define DS_REF_MAX (1ULL << 62)
63 63
64 64 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
65 65
66 66 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67 67
68 68
69 69 /*
70 70 * Figure out how much of this delta should be propogated to the dsl_dir
71 71 * layer. If there's a refreservation, that space has already been
72 72 * partially accounted for in our ancestors.
73 73 */
74 74 static int64_t
75 75 parent_delta(dsl_dataset_t *ds, int64_t delta)
76 76 {
77 77 uint64_t old_bytes, new_bytes;
78 78
79 79 if (ds->ds_reserved == 0)
80 80 return (delta);
81 81
82 82 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
83 83 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
84 84
85 85 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
86 86 return (new_bytes - old_bytes);
87 87 }
88 88
89 89 void
90 90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
91 91 {
92 92 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
93 93 int compressed = BP_GET_PSIZE(bp);
94 94 int uncompressed = BP_GET_UCSIZE(bp);
95 95 int64_t delta;
96 96
97 97 dprintf_bp(bp, "ds=%p", ds);
98 98
99 99 ASSERT(dmu_tx_is_syncing(tx));
100 100 /* It could have been compressed away to nothing */
101 101 if (BP_IS_HOLE(bp))
102 102 return;
103 103 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
104 104 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
105 105 if (ds == NULL) {
106 106 /*
107 107 * Account for the meta-objset space in its placeholder
108 108 * dsl_dir.
109 109 */
110 110 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
111 111 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
112 112 used, compressed, uncompressed, tx);
113 113 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
114 114 return;
115 115 }
116 116 dmu_buf_will_dirty(ds->ds_dbuf, tx);
117 117
118 118 mutex_enter(&ds->ds_dir->dd_lock);
119 119 mutex_enter(&ds->ds_lock);
120 120 delta = parent_delta(ds, used);
121 121 ds->ds_phys->ds_referenced_bytes += used;
122 122 ds->ds_phys->ds_compressed_bytes += compressed;
123 123 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 124 ds->ds_phys->ds_unique_bytes += used;
125 125 mutex_exit(&ds->ds_lock);
126 126 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
127 127 compressed, uncompressed, tx);
128 128 dsl_dir_transfer_space(ds->ds_dir, used - delta,
129 129 DD_USED_REFRSRV, DD_USED_HEAD, tx);
130 130 mutex_exit(&ds->ds_dir->dd_lock);
131 131 }
132 132
133 133 int
134 134 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
135 135 boolean_t async)
136 136 {
137 137 if (BP_IS_HOLE(bp))
138 138 return (0);
139 139
140 140 ASSERT(dmu_tx_is_syncing(tx));
141 141 ASSERT(bp->blk_birth <= tx->tx_txg);
142 142
143 143 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
144 144 int compressed = BP_GET_PSIZE(bp);
145 145 int uncompressed = BP_GET_UCSIZE(bp);
146 146
147 147 ASSERT(used > 0);
148 148 if (ds == NULL) {
149 149 /*
150 150 * Account for the meta-objset space in its placeholder
151 151 * dataset.
152 152 */
153 153 dsl_free(tx->tx_pool, tx->tx_txg, bp);
154 154
155 155 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
156 156 -used, -compressed, -uncompressed, tx);
157 157 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
158 158 return (used);
159 159 }
160 160 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
161 161
162 162 ASSERT(!dsl_dataset_is_snapshot(ds));
163 163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
164 164
165 165 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
166 166 int64_t delta;
167 167
168 168 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
169 169 dsl_free(tx->tx_pool, tx->tx_txg, bp);
170 170
171 171 mutex_enter(&ds->ds_dir->dd_lock);
172 172 mutex_enter(&ds->ds_lock);
173 173 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
174 174 !DS_UNIQUE_IS_ACCURATE(ds));
175 175 delta = parent_delta(ds, -used);
176 176 ds->ds_phys->ds_unique_bytes -= used;
177 177 mutex_exit(&ds->ds_lock);
178 178 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
179 179 delta, -compressed, -uncompressed, tx);
180 180 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
181 181 DD_USED_REFRSRV, DD_USED_HEAD, tx);
182 182 mutex_exit(&ds->ds_dir->dd_lock);
183 183 } else {
184 184 dprintf_bp(bp, "putting on dead list: %s", "");
185 185 if (async) {
186 186 /*
187 187 * We are here as part of zio's write done callback,
188 188 * which means we're a zio interrupt thread. We can't
189 189 * call dsl_deadlist_insert() now because it may block
190 190 * waiting for I/O. Instead, put bp on the deferred
191 191 * queue and let dsl_pool_sync() finish the job.
192 192 */
193 193 bplist_append(&ds->ds_pending_deadlist, bp);
194 194 } else {
195 195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 196 }
197 197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 198 ds->ds_phys->ds_prev_snap_obj);
199 199 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
200 200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 201 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
202 202 ds->ds_object && bp->blk_birth >
203 203 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
204 204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 205 mutex_enter(&ds->ds_prev->ds_lock);
206 206 ds->ds_prev->ds_phys->ds_unique_bytes += used;
207 207 mutex_exit(&ds->ds_prev->ds_lock);
208 208 }
209 209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 210 dsl_dir_transfer_space(ds->ds_dir, used,
211 211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 212 }
213 213 }
214 214 mutex_enter(&ds->ds_lock);
215 215 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
216 216 ds->ds_phys->ds_referenced_bytes -= used;
217 217 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
218 218 ds->ds_phys->ds_compressed_bytes -= compressed;
219 219 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
220 220 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
221 221 mutex_exit(&ds->ds_lock);
222 222
223 223 return (used);
224 224 }
225 225
226 226 uint64_t
227 227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 228 {
229 229 uint64_t trysnap = 0;
230 230
231 231 if (ds == NULL)
232 232 return (0);
233 233 /*
234 234 * The snapshot creation could fail, but that would cause an
235 235 * incorrect FALSE return, which would only result in an
236 236 * overestimation of the amount of space that an operation would
237 237 * consume, which is OK.
238 238 *
239 239 * There's also a small window where we could miss a pending
240 240 * snapshot, because we could set the sync task in the quiescing
241 241 * phase. So this should only be used as a guess.
242 242 */
243 243 if (ds->ds_trysnap_txg >
244 244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 245 trysnap = ds->ds_trysnap_txg;
246 246 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
247 247 }
248 248
249 249 boolean_t
250 250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 251 uint64_t blk_birth)
252 252 {
253 253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
254 254 return (B_FALSE);
255 255
256 256 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
257 257
258 258 return (B_TRUE);
259 259 }
260 260
261 261 /* ARGSUSED */
262 262 static void
263 263 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
264 264 {
265 265 dsl_dataset_t *ds = dsv;
266 266
267 267 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
268 268
269 269 unique_remove(ds->ds_fsid_guid);
270 270
271 271 if (ds->ds_objset != NULL)
272 272 dmu_objset_evict(ds->ds_objset);
273 273
274 274 if (ds->ds_prev) {
275 275 dsl_dataset_drop_ref(ds->ds_prev, ds);
276 276 ds->ds_prev = NULL;
277 277 }
278 278
279 279 bplist_destroy(&ds->ds_pending_deadlist);
280 280 if (db != NULL) {
281 281 dsl_deadlist_close(&ds->ds_deadlist);
282 282 } else {
283 283 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
284 284 ASSERT(!ds->ds_deadlist.dl_oldfmt);
285 285 }
286 286 if (ds->ds_dir)
287 287 dsl_dir_close(ds->ds_dir, ds);
288 288
289 289 ASSERT(!list_link_active(&ds->ds_synced_link));
290 290
291 291 mutex_destroy(&ds->ds_lock);
292 292 mutex_destroy(&ds->ds_recvlock);
293 293 mutex_destroy(&ds->ds_opening_lock);
294 294 rw_destroy(&ds->ds_rwlock);
295 295 cv_destroy(&ds->ds_exclusive_cv);
296 296
297 297 kmem_free(ds, sizeof (dsl_dataset_t));
298 298 }
299 299
300 300 static int
301 301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
302 302 {
303 303 dsl_dataset_phys_t *headphys;
304 304 int err;
305 305 dmu_buf_t *headdbuf;
306 306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 307 objset_t *mos = dp->dp_meta_objset;
308 308
309 309 if (ds->ds_snapname[0])
310 310 return (0);
311 311 if (ds->ds_phys->ds_next_snap_obj == 0)
312 312 return (0);
313 313
314 314 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
315 315 FTAG, &headdbuf);
316 316 if (err)
317 317 return (err);
318 318 headphys = headdbuf->db_data;
319 319 err = zap_value_search(dp->dp_meta_objset,
320 320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 321 dmu_buf_rele(headdbuf, FTAG);
322 322 return (err);
323 323 }
324 324
325 325 static int
326 326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
327 327 {
328 328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
330 330 matchtype_t mt;
331 331 int err;
332 332
333 333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
334 334 mt = MT_FIRST;
335 335 else
336 336 mt = MT_EXACT;
337 337
338 338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 339 value, mt, NULL, 0, NULL);
340 340 if (err == ENOTSUP && mt == MT_FIRST)
341 341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
342 342 return (err);
343 343 }
344 344
345 345 static int
346 346 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
347 347 {
348 348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
350 350 matchtype_t mt;
351 351 int err;
352 352
353 353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354 354
355 355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
356 356 mt = MT_FIRST;
357 357 else
358 358 mt = MT_EXACT;
359 359
360 360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
361 361 if (err == ENOTSUP && mt == MT_FIRST)
362 362 err = zap_remove(mos, snapobj, name, tx);
363 363 return (err);
364 364 }
365 365
366 366 static int
367 367 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
368 368 dsl_dataset_t **dsp)
369 369 {
370 370 objset_t *mos = dp->dp_meta_objset;
371 371 dmu_buf_t *dbuf;
372 372 dsl_dataset_t *ds;
373 373 int err;
374 374 dmu_object_info_t doi;
375 375
376 376 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
377 377 dsl_pool_sync_context(dp));
378 378
379 379 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
380 380 if (err)
381 381 return (err);
382 382
383 383 /* Make sure dsobj has the correct object type. */
384 384 dmu_object_info_from_db(dbuf, &doi);
385 385 if (doi.doi_type != DMU_OT_DSL_DATASET)
386 386 return (EINVAL);
387 387
388 388 ds = dmu_buf_get_user(dbuf);
389 389 if (ds == NULL) {
390 390 dsl_dataset_t *winner;
391 391
392 392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
393 393 ds->ds_dbuf = dbuf;
394 394 ds->ds_object = dsobj;
395 395 ds->ds_phys = dbuf->db_data;
396 396
397 397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
398 398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
399 399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
400 400 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
401 401
402 402 rw_init(&ds->ds_rwlock, 0, 0, 0);
403 403 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
404 404
405 405 bplist_create(&ds->ds_pending_deadlist);
406 406 dsl_deadlist_open(&ds->ds_deadlist,
407 407 mos, ds->ds_phys->ds_deadlist_obj);
408 408
409 409 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
410 410 offsetof(dmu_sendarg_t, dsa_link));
411 411
412 412 if (err == 0) {
413 413 err = dsl_dir_open_obj(dp,
414 414 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
415 415 }
416 416 if (err) {
417 417 mutex_destroy(&ds->ds_lock);
418 418 mutex_destroy(&ds->ds_recvlock);
419 419 mutex_destroy(&ds->ds_opening_lock);
420 420 rw_destroy(&ds->ds_rwlock);
421 421 cv_destroy(&ds->ds_exclusive_cv);
422 422 bplist_destroy(&ds->ds_pending_deadlist);
423 423 dsl_deadlist_close(&ds->ds_deadlist);
424 424 kmem_free(ds, sizeof (dsl_dataset_t));
425 425 dmu_buf_rele(dbuf, tag);
426 426 return (err);
427 427 }
428 428
429 429 if (!dsl_dataset_is_snapshot(ds)) {
430 430 ds->ds_snapname[0] = '\0';
431 431 if (ds->ds_phys->ds_prev_snap_obj) {
432 432 err = dsl_dataset_get_ref(dp,
433 433 ds->ds_phys->ds_prev_snap_obj,
434 434 ds, &ds->ds_prev);
435 435 }
436 436 } else {
437 437 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
438 438 err = dsl_dataset_get_snapname(ds);
439 439 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
440 440 err = zap_count(
441 441 ds->ds_dir->dd_pool->dp_meta_objset,
442 442 ds->ds_phys->ds_userrefs_obj,
443 443 &ds->ds_userrefs);
444 444 }
445 445 }
446 446
447 447 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
448 448 /*
449 449 * In sync context, we're called with either no lock
450 450 * or with the write lock. If we're not syncing,
451 451 * we're always called with the read lock held.
452 452 */
453 453 boolean_t need_lock =
454 454 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
455 455 dsl_pool_sync_context(dp);
456 456
457 457 if (need_lock)
458 458 rw_enter(&dp->dp_config_rwlock, RW_READER);
459 459
460 460 err = dsl_prop_get_ds(ds,
461 461 "refreservation", sizeof (uint64_t), 1,
462 462 &ds->ds_reserved, NULL);
463 463 if (err == 0) {
464 464 err = dsl_prop_get_ds(ds,
465 465 "refquota", sizeof (uint64_t), 1,
466 466 &ds->ds_quota, NULL);
467 467 }
468 468
469 469 if (need_lock)
470 470 rw_exit(&dp->dp_config_rwlock);
471 471 } else {
472 472 ds->ds_reserved = ds->ds_quota = 0;
473 473 }
474 474
475 475 if (err == 0) {
476 476 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
477 477 dsl_dataset_evict);
478 478 }
479 479 if (err || winner) {
480 480 bplist_destroy(&ds->ds_pending_deadlist);
481 481 dsl_deadlist_close(&ds->ds_deadlist);
482 482 if (ds->ds_prev)
483 483 dsl_dataset_drop_ref(ds->ds_prev, ds);
484 484 dsl_dir_close(ds->ds_dir, ds);
485 485 mutex_destroy(&ds->ds_lock);
486 486 mutex_destroy(&ds->ds_recvlock);
487 487 mutex_destroy(&ds->ds_opening_lock);
488 488 rw_destroy(&ds->ds_rwlock);
489 489 cv_destroy(&ds->ds_exclusive_cv);
490 490 kmem_free(ds, sizeof (dsl_dataset_t));
491 491 if (err) {
492 492 dmu_buf_rele(dbuf, tag);
493 493 return (err);
494 494 }
495 495 ds = winner;
496 496 } else {
497 497 ds->ds_fsid_guid =
498 498 unique_insert(ds->ds_phys->ds_fsid_guid);
499 499 }
500 500 }
501 501 ASSERT3P(ds->ds_dbuf, ==, dbuf);
502 502 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
503 503 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
504 504 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
505 505 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
506 506 mutex_enter(&ds->ds_lock);
507 507 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
508 508 mutex_exit(&ds->ds_lock);
509 509 dmu_buf_rele(ds->ds_dbuf, tag);
510 510 return (ENOENT);
511 511 }
512 512 mutex_exit(&ds->ds_lock);
513 513 *dsp = ds;
514 514 return (0);
515 515 }
516 516
517 517 static int
518 518 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
519 519 {
520 520 dsl_pool_t *dp = ds->ds_dir->dd_pool;
521 521
522 522 /*
523 523 * In syncing context we don't want the rwlock lock: there
524 524 * may be an existing writer waiting for sync phase to
525 525 * finish. We don't need to worry about such writers, since
526 526 * sync phase is single-threaded, so the writer can't be
527 527 * doing anything while we are active.
528 528 */
529 529 if (dsl_pool_sync_context(dp)) {
530 530 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
531 531 return (0);
532 532 }
533 533
534 534 /*
535 535 * Normal users will hold the ds_rwlock as a READER until they
536 536 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
537 537 * drop their READER lock after they set the ds_owner field.
538 538 *
539 539 * If the dataset is being destroyed, the destroy thread will
540 540 * obtain a WRITER lock for exclusive access after it's done its
541 541 * open-context work and then change the ds_owner to
542 542 * dsl_reaper once destruction is assured. So threads
543 543 * may block here temporarily, until the "destructability" of
544 544 * the dataset is determined.
545 545 */
546 546 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
547 547 mutex_enter(&ds->ds_lock);
548 548 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
549 549 rw_exit(&dp->dp_config_rwlock);
550 550 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
551 551 if (DSL_DATASET_IS_DESTROYED(ds)) {
552 552 mutex_exit(&ds->ds_lock);
553 553 dsl_dataset_drop_ref(ds, tag);
554 554 rw_enter(&dp->dp_config_rwlock, RW_READER);
555 555 return (ENOENT);
556 556 }
557 557 /*
558 558 * The dp_config_rwlock lives above the ds_lock. And
559 559 * we need to check DSL_DATASET_IS_DESTROYED() while
560 560 * holding the ds_lock, so we have to drop and reacquire
561 561 * the ds_lock here.
562 562 */
563 563 mutex_exit(&ds->ds_lock);
564 564 rw_enter(&dp->dp_config_rwlock, RW_READER);
565 565 mutex_enter(&ds->ds_lock);
566 566 }
567 567 mutex_exit(&ds->ds_lock);
568 568 return (0);
569 569 }
570 570
571 571 int
572 572 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
573 573 dsl_dataset_t **dsp)
574 574 {
575 575 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
576 576
577 577 if (err)
578 578 return (err);
579 579 return (dsl_dataset_hold_ref(*dsp, tag));
580 580 }
581 581
582 582 int
583 583 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
584 584 void *tag, dsl_dataset_t **dsp)
585 585 {
586 586 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
587 587 if (err)
588 588 return (err);
589 589 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
590 590 dsl_dataset_rele(*dsp, tag);
591 591 *dsp = NULL;
592 592 return (EBUSY);
593 593 }
594 594 return (0);
595 595 }
596 596
597 597 int
598 598 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
599 599 {
600 600 dsl_dir_t *dd;
601 601 dsl_pool_t *dp;
602 602 const char *snapname;
603 603 uint64_t obj;
604 604 int err = 0;
605 605
606 606 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
607 607 if (err)
608 608 return (err);
609 609
610 610 dp = dd->dd_pool;
611 611 obj = dd->dd_phys->dd_head_dataset_obj;
612 612 rw_enter(&dp->dp_config_rwlock, RW_READER);
613 613 if (obj)
614 614 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
615 615 else
616 616 err = ENOENT;
617 617 if (err)
618 618 goto out;
619 619
620 620 err = dsl_dataset_hold_ref(*dsp, tag);
621 621
622 622 /* we may be looking for a snapshot */
623 623 if (err == 0 && snapname != NULL) {
624 624 dsl_dataset_t *ds = NULL;
625 625
626 626 if (*snapname++ != '@') {
627 627 dsl_dataset_rele(*dsp, tag);
628 628 err = ENOENT;
629 629 goto out;
630 630 }
631 631
632 632 dprintf("looking for snapshot '%s'\n", snapname);
633 633 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
634 634 if (err == 0)
635 635 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
636 636 dsl_dataset_rele(*dsp, tag);
637 637
638 638 ASSERT3U((err == 0), ==, (ds != NULL));
639 639
640 640 if (ds) {
641 641 mutex_enter(&ds->ds_lock);
642 642 if (ds->ds_snapname[0] == 0)
643 643 (void) strlcpy(ds->ds_snapname, snapname,
644 644 sizeof (ds->ds_snapname));
645 645 mutex_exit(&ds->ds_lock);
646 646 err = dsl_dataset_hold_ref(ds, tag);
647 647 *dsp = err ? NULL : ds;
648 648 }
649 649 }
650 650 out:
651 651 rw_exit(&dp->dp_config_rwlock);
652 652 dsl_dir_close(dd, FTAG);
653 653 return (err);
654 654 }
655 655
656 656 int
657 657 dsl_dataset_own(const char *name, boolean_t inconsistentok,
658 658 void *tag, dsl_dataset_t **dsp)
659 659 {
660 660 int err = dsl_dataset_hold(name, tag, dsp);
661 661 if (err)
662 662 return (err);
663 663 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
664 664 dsl_dataset_rele(*dsp, tag);
665 665 return (EBUSY);
666 666 }
667 667 return (0);
668 668 }
669 669
670 670 void
671 671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
672 672 {
673 673 if (ds == NULL) {
674 674 (void) strcpy(name, "mos");
675 675 } else {
676 676 dsl_dir_name(ds->ds_dir, name);
677 677 VERIFY(0 == dsl_dataset_get_snapname(ds));
678 678 if (ds->ds_snapname[0]) {
679 679 (void) strcat(name, "@");
680 680 /*
681 681 * We use a "recursive" mutex so that we
682 682 * can call dprintf_ds() with ds_lock held.
683 683 */
684 684 if (!MUTEX_HELD(&ds->ds_lock)) {
685 685 mutex_enter(&ds->ds_lock);
686 686 (void) strcat(name, ds->ds_snapname);
687 687 mutex_exit(&ds->ds_lock);
688 688 } else {
689 689 (void) strcat(name, ds->ds_snapname);
690 690 }
691 691 }
692 692 }
693 693 }
694 694
695 695 static int
696 696 dsl_dataset_namelen(dsl_dataset_t *ds)
697 697 {
698 698 int result;
699 699
700 700 if (ds == NULL) {
701 701 result = 3; /* "mos" */
702 702 } else {
703 703 result = dsl_dir_namelen(ds->ds_dir);
704 704 VERIFY(0 == dsl_dataset_get_snapname(ds));
705 705 if (ds->ds_snapname[0]) {
706 706 ++result; /* adding one for the @-sign */
707 707 if (!MUTEX_HELD(&ds->ds_lock)) {
708 708 mutex_enter(&ds->ds_lock);
709 709 result += strlen(ds->ds_snapname);
710 710 mutex_exit(&ds->ds_lock);
711 711 } else {
712 712 result += strlen(ds->ds_snapname);
713 713 }
714 714 }
715 715 }
716 716
717 717 return (result);
718 718 }
719 719
720 720 void
721 721 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
722 722 {
723 723 dmu_buf_rele(ds->ds_dbuf, tag);
724 724 }
725 725
726 726 void
727 727 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
728 728 {
729 729 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
730 730 rw_exit(&ds->ds_rwlock);
731 731 }
732 732 dsl_dataset_drop_ref(ds, tag);
733 733 }
734 734
735 735 void
736 736 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
737 737 {
738 738 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
739 739 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
740 740
741 741 mutex_enter(&ds->ds_lock);
742 742 ds->ds_owner = NULL;
743 743 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
744 744 rw_exit(&ds->ds_rwlock);
745 745 cv_broadcast(&ds->ds_exclusive_cv);
746 746 }
747 747 mutex_exit(&ds->ds_lock);
748 748 if (ds->ds_dbuf)
749 749 dsl_dataset_drop_ref(ds, tag);
750 750 else
751 751 dsl_dataset_evict(NULL, ds);
752 752 }
753 753
754 754 boolean_t
755 755 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
756 756 {
757 757 boolean_t gotit = FALSE;
758 758
759 759 mutex_enter(&ds->ds_lock);
760 760 if (ds->ds_owner == NULL &&
761 761 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
762 762 ds->ds_owner = tag;
763 763 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
764 764 rw_exit(&ds->ds_rwlock);
765 765 gotit = TRUE;
766 766 }
767 767 mutex_exit(&ds->ds_lock);
768 768 return (gotit);
769 769 }
770 770
771 771 void
772 772 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
773 773 {
774 774 ASSERT3P(owner, ==, ds->ds_owner);
775 775 if (!RW_WRITE_HELD(&ds->ds_rwlock))
776 776 rw_enter(&ds->ds_rwlock, RW_WRITER);
777 777 }
778 778
779 779 uint64_t
780 780 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
781 781 uint64_t flags, dmu_tx_t *tx)
782 782 {
783 783 dsl_pool_t *dp = dd->dd_pool;
784 784 dmu_buf_t *dbuf;
785 785 dsl_dataset_phys_t *dsphys;
786 786 uint64_t dsobj;
787 787 objset_t *mos = dp->dp_meta_objset;
788 788
789 789 if (origin == NULL)
790 790 origin = dp->dp_origin_snap;
791 791
792 792 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
793 793 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
794 794 ASSERT(dmu_tx_is_syncing(tx));
795 795 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
|
↓ open down ↓ |
795 lines elided |
↑ open up ↑ |
796 796
797 797 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
798 798 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
799 799 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
800 800 dmu_buf_will_dirty(dbuf, tx);
801 801 dsphys = dbuf->db_data;
802 802 bzero(dsphys, sizeof (dsl_dataset_phys_t));
803 803 dsphys->ds_dir_obj = dd->dd_object;
804 804 dsphys->ds_flags = flags;
805 805 dsphys->ds_fsid_guid = unique_create();
806 - (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
807 - sizeof (dsphys->ds_guid));
806 + do {
807 + (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
808 + sizeof (dsphys->ds_guid));
809 + } while (dsphys->ds_guid == 0);
808 810 dsphys->ds_snapnames_zapobj =
809 811 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
810 812 DMU_OT_NONE, 0, tx);
811 813 dsphys->ds_creation_time = gethrestime_sec();
812 814 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
813 815
814 816 if (origin == NULL) {
815 817 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
816 818 } else {
817 819 dsl_dataset_t *ohds;
818 820
819 821 dsphys->ds_prev_snap_obj = origin->ds_object;
820 822 dsphys->ds_prev_snap_txg =
821 823 origin->ds_phys->ds_creation_txg;
822 824 dsphys->ds_referenced_bytes =
823 825 origin->ds_phys->ds_referenced_bytes;
824 826 dsphys->ds_compressed_bytes =
825 827 origin->ds_phys->ds_compressed_bytes;
826 828 dsphys->ds_uncompressed_bytes =
827 829 origin->ds_phys->ds_uncompressed_bytes;
828 830 dsphys->ds_bp = origin->ds_phys->ds_bp;
829 831 dsphys->ds_flags |= origin->ds_phys->ds_flags;
830 832
831 833 dmu_buf_will_dirty(origin->ds_dbuf, tx);
832 834 origin->ds_phys->ds_num_children++;
833 835
834 836 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
835 837 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
836 838 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
837 839 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
838 840 dsl_dataset_rele(ohds, FTAG);
839 841
840 842 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
841 843 if (origin->ds_phys->ds_next_clones_obj == 0) {
842 844 origin->ds_phys->ds_next_clones_obj =
843 845 zap_create(mos,
844 846 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
845 847 }
846 848 VERIFY(0 == zap_add_int(mos,
847 849 origin->ds_phys->ds_next_clones_obj,
848 850 dsobj, tx));
849 851 }
850 852
851 853 dmu_buf_will_dirty(dd->dd_dbuf, tx);
852 854 dd->dd_phys->dd_origin_obj = origin->ds_object;
853 855 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
854 856 if (origin->ds_dir->dd_phys->dd_clones == 0) {
855 857 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
856 858 origin->ds_dir->dd_phys->dd_clones =
857 859 zap_create(mos,
858 860 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
859 861 }
860 862 VERIFY3U(0, ==, zap_add_int(mos,
861 863 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
862 864 }
863 865 }
864 866
865 867 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
866 868 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867 869
868 870 dmu_buf_rele(dbuf, FTAG);
869 871
870 872 dmu_buf_will_dirty(dd->dd_dbuf, tx);
871 873 dd->dd_phys->dd_head_dataset_obj = dsobj;
872 874
873 875 return (dsobj);
874 876 }
875 877
876 878 uint64_t
877 879 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
878 880 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
879 881 {
880 882 dsl_pool_t *dp = pdd->dd_pool;
881 883 uint64_t dsobj, ddobj;
882 884 dsl_dir_t *dd;
883 885
884 886 ASSERT(lastname[0] != '@');
885 887
886 888 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
887 889 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
888 890
889 891 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
890 892
891 893 dsl_deleg_set_create_perms(dd, tx, cr);
892 894
893 895 dsl_dir_close(dd, FTAG);
894 896
895 897 /*
896 898 * If we are creating a clone, make sure we zero out any stale
897 899 * data from the origin snapshots zil header.
898 900 */
899 901 if (origin != NULL) {
900 902 dsl_dataset_t *ds;
901 903 objset_t *os;
902 904
903 905 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
904 906 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
905 907 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
906 908 dsl_dataset_dirty(ds, tx);
907 909 dsl_dataset_rele(ds, FTAG);
908 910 }
909 911
910 912 return (dsobj);
911 913 }
912 914
913 915 /*
914 916 * The snapshots must all be in the same pool.
915 917 */
916 918 int
917 919 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
918 920 {
919 921 int err;
920 922 dsl_sync_task_t *dst;
921 923 spa_t *spa;
922 924 nvpair_t *pair;
923 925 dsl_sync_task_group_t *dstg;
924 926
925 927 pair = nvlist_next_nvpair(snaps, NULL);
926 928 if (pair == NULL)
927 929 return (0);
928 930
929 931 err = spa_open(nvpair_name(pair), &spa, FTAG);
930 932 if (err)
931 933 return (err);
932 934 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
933 935
934 936 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
935 937 pair = nvlist_next_nvpair(snaps, pair)) {
936 938 dsl_dataset_t *ds;
937 939
938 940 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
939 941 if (err == 0) {
940 942 struct dsl_ds_destroyarg *dsda;
941 943
942 944 dsl_dataset_make_exclusive(ds, dstg);
943 945 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
944 946 KM_SLEEP);
945 947 dsda->ds = ds;
946 948 dsda->defer = defer;
947 949 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
948 950 dsl_dataset_destroy_sync, dsda, dstg, 0);
949 951 } else if (err == ENOENT) {
950 952 err = 0;
951 953 } else {
952 954 (void) strcpy(failed, nvpair_name(pair));
953 955 break;
954 956 }
955 957 }
956 958
957 959 if (err == 0)
958 960 err = dsl_sync_task_group_wait(dstg);
959 961
960 962 for (dst = list_head(&dstg->dstg_tasks); dst;
961 963 dst = list_next(&dstg->dstg_tasks, dst)) {
962 964 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
963 965 dsl_dataset_t *ds = dsda->ds;
964 966
965 967 /*
966 968 * Return the file system name that triggered the error
967 969 */
968 970 if (dst->dst_err) {
969 971 dsl_dataset_name(ds, failed);
970 972 }
971 973 ASSERT3P(dsda->rm_origin, ==, NULL);
972 974 dsl_dataset_disown(ds, dstg);
973 975 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
974 976 }
975 977
976 978 dsl_sync_task_group_destroy(dstg);
977 979 spa_close(spa, FTAG);
978 980 return (err);
979 981
980 982 }
981 983
982 984 static boolean_t
983 985 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
984 986 {
985 987 boolean_t might_destroy = B_FALSE;
986 988
987 989 mutex_enter(&ds->ds_lock);
988 990 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
989 991 DS_IS_DEFER_DESTROY(ds))
990 992 might_destroy = B_TRUE;
991 993 mutex_exit(&ds->ds_lock);
992 994
993 995 return (might_destroy);
994 996 }
995 997
996 998 /*
997 999 * If we're removing a clone, and these three conditions are true:
998 1000 * 1) the clone's origin has no other children
999 1001 * 2) the clone's origin has no user references
1000 1002 * 3) the clone's origin has been marked for deferred destruction
1001 1003 * Then, prepare to remove the origin as part of this sync task group.
1002 1004 */
1003 1005 static int
1004 1006 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1005 1007 {
1006 1008 dsl_dataset_t *ds = dsda->ds;
1007 1009 dsl_dataset_t *origin = ds->ds_prev;
1008 1010
1009 1011 if (dsl_dataset_might_destroy_origin(origin)) {
1010 1012 char *name;
1011 1013 int namelen;
1012 1014 int error;
1013 1015
1014 1016 namelen = dsl_dataset_namelen(origin) + 1;
1015 1017 name = kmem_alloc(namelen, KM_SLEEP);
1016 1018 dsl_dataset_name(origin, name);
1017 1019 #ifdef _KERNEL
1018 1020 error = zfs_unmount_snap(name, NULL);
1019 1021 if (error) {
1020 1022 kmem_free(name, namelen);
1021 1023 return (error);
1022 1024 }
1023 1025 #endif
1024 1026 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1025 1027 kmem_free(name, namelen);
1026 1028 if (error)
1027 1029 return (error);
1028 1030 dsda->rm_origin = origin;
1029 1031 dsl_dataset_make_exclusive(origin, tag);
1030 1032 }
1031 1033
1032 1034 return (0);
1033 1035 }
1034 1036
1035 1037 /*
1036 1038 * ds must be opened as OWNER. On return (whether successful or not),
1037 1039 * ds will be closed and caller can no longer dereference it.
1038 1040 */
1039 1041 int
1040 1042 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1041 1043 {
1042 1044 int err;
1043 1045 dsl_sync_task_group_t *dstg;
1044 1046 objset_t *os;
1045 1047 dsl_dir_t *dd;
1046 1048 uint64_t obj;
1047 1049 struct dsl_ds_destroyarg dsda = { 0 };
1048 1050 dsl_dataset_t dummy_ds = { 0 };
1049 1051
1050 1052 dsda.ds = ds;
1051 1053
1052 1054 if (dsl_dataset_is_snapshot(ds)) {
1053 1055 /* Destroying a snapshot is simpler */
1054 1056 dsl_dataset_make_exclusive(ds, tag);
1055 1057
1056 1058 dsda.defer = defer;
1057 1059 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1058 1060 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1059 1061 &dsda, tag, 0);
1060 1062 ASSERT3P(dsda.rm_origin, ==, NULL);
1061 1063 goto out;
1062 1064 } else if (defer) {
1063 1065 err = EINVAL;
1064 1066 goto out;
1065 1067 }
1066 1068
1067 1069 dd = ds->ds_dir;
1068 1070 dummy_ds.ds_dir = dd;
1069 1071 dummy_ds.ds_object = ds->ds_object;
1070 1072
1071 1073 /*
1072 1074 * Check for errors and mark this ds as inconsistent, in
1073 1075 * case we crash while freeing the objects.
1074 1076 */
1075 1077 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1076 1078 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1077 1079 if (err)
1078 1080 goto out;
1079 1081
1080 1082 err = dmu_objset_from_ds(ds, &os);
1081 1083 if (err)
1082 1084 goto out;
1083 1085
1084 1086 /*
1085 1087 * If async destruction is not enabled try to remove all objects
1086 1088 * while in the open context so that there is less work to do in
1087 1089 * the syncing context.
1088 1090 */
1089 1091 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1090 1092 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1091 1093 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1092 1094 ds->ds_phys->ds_prev_snap_txg)) {
1093 1095 /*
1094 1096 * Ignore errors, if there is not enough disk space
1095 1097 * we will deal with it in dsl_dataset_destroy_sync().
1096 1098 */
1097 1099 (void) dmu_free_object(os, obj);
1098 1100 }
1099 1101 if (err != ESRCH)
1100 1102 goto out;
1101 1103 }
1102 1104
1103 1105 /*
1104 1106 * Only the ZIL knows how to free log blocks.
1105 1107 */
1106 1108 zil_destroy(dmu_objset_zil(os), B_FALSE);
1107 1109
1108 1110 /*
1109 1111 * Sync out all in-flight IO.
1110 1112 */
1111 1113 txg_wait_synced(dd->dd_pool, 0);
1112 1114
1113 1115 /*
1114 1116 * If we managed to free all the objects in open
1115 1117 * context, the user space accounting should be zero.
1116 1118 */
1117 1119 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1118 1120 dmu_objset_userused_enabled(os)) {
1119 1121 uint64_t count;
1120 1122
1121 1123 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1122 1124 count == 0);
1123 1125 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1124 1126 count == 0);
1125 1127 }
1126 1128
1127 1129 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1128 1130 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1129 1131 rw_exit(&dd->dd_pool->dp_config_rwlock);
1130 1132
1131 1133 if (err)
1132 1134 goto out;
1133 1135
1134 1136 /*
1135 1137 * Blow away the dsl_dir + head dataset.
1136 1138 */
1137 1139 dsl_dataset_make_exclusive(ds, tag);
1138 1140 /*
1139 1141 * If we're removing a clone, we might also need to remove its
1140 1142 * origin.
1141 1143 */
1142 1144 do {
1143 1145 dsda.need_prep = B_FALSE;
1144 1146 if (dsl_dir_is_clone(dd)) {
1145 1147 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1146 1148 if (err) {
1147 1149 dsl_dir_close(dd, FTAG);
1148 1150 goto out;
1149 1151 }
1150 1152 }
1151 1153
1152 1154 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1153 1155 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1154 1156 dsl_dataset_destroy_sync, &dsda, tag, 0);
1155 1157 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1156 1158 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1157 1159 err = dsl_sync_task_group_wait(dstg);
1158 1160 dsl_sync_task_group_destroy(dstg);
1159 1161
1160 1162 /*
1161 1163 * We could be racing against 'zfs release' or 'zfs destroy -d'
1162 1164 * on the origin snap, in which case we can get EBUSY if we
1163 1165 * needed to destroy the origin snap but were not ready to
1164 1166 * do so.
1165 1167 */
1166 1168 if (dsda.need_prep) {
1167 1169 ASSERT(err == EBUSY);
1168 1170 ASSERT(dsl_dir_is_clone(dd));
1169 1171 ASSERT(dsda.rm_origin == NULL);
1170 1172 }
1171 1173 } while (dsda.need_prep);
1172 1174
1173 1175 if (dsda.rm_origin != NULL)
1174 1176 dsl_dataset_disown(dsda.rm_origin, tag);
1175 1177
1176 1178 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1177 1179 if (err)
1178 1180 dsl_dir_close(dd, FTAG);
1179 1181 out:
1180 1182 dsl_dataset_disown(ds, tag);
1181 1183 return (err);
1182 1184 }
1183 1185
1184 1186 blkptr_t *
1185 1187 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1186 1188 {
1187 1189 return (&ds->ds_phys->ds_bp);
1188 1190 }
1189 1191
1190 1192 void
1191 1193 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1192 1194 {
1193 1195 ASSERT(dmu_tx_is_syncing(tx));
1194 1196 /* If it's the meta-objset, set dp_meta_rootbp */
1195 1197 if (ds == NULL) {
1196 1198 tx->tx_pool->dp_meta_rootbp = *bp;
1197 1199 } else {
1198 1200 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1199 1201 ds->ds_phys->ds_bp = *bp;
1200 1202 }
1201 1203 }
1202 1204
1203 1205 spa_t *
1204 1206 dsl_dataset_get_spa(dsl_dataset_t *ds)
1205 1207 {
1206 1208 return (ds->ds_dir->dd_pool->dp_spa);
1207 1209 }
1208 1210
1209 1211 void
1210 1212 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1211 1213 {
1212 1214 dsl_pool_t *dp;
1213 1215
1214 1216 if (ds == NULL) /* this is the meta-objset */
1215 1217 return;
1216 1218
1217 1219 ASSERT(ds->ds_objset != NULL);
1218 1220
1219 1221 if (ds->ds_phys->ds_next_snap_obj != 0)
1220 1222 panic("dirtying snapshot!");
1221 1223
1222 1224 dp = ds->ds_dir->dd_pool;
1223 1225
1224 1226 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1225 1227 /* up the hold count until we can be written out */
1226 1228 dmu_buf_add_ref(ds->ds_dbuf, ds);
1227 1229 }
1228 1230 }
1229 1231
1230 1232 /*
1231 1233 * The unique space in the head dataset can be calculated by subtracting
1232 1234 * the space used in the most recent snapshot, that is still being used
1233 1235 * in this file system, from the space currently in use. To figure out
1234 1236 * the space in the most recent snapshot still in use, we need to take
1235 1237 * the total space used in the snapshot and subtract out the space that
1236 1238 * has been freed up since the snapshot was taken.
1237 1239 */
1238 1240 static void
1239 1241 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1240 1242 {
1241 1243 uint64_t mrs_used;
1242 1244 uint64_t dlused, dlcomp, dluncomp;
1243 1245
1244 1246 ASSERT(!dsl_dataset_is_snapshot(ds));
1245 1247
1246 1248 if (ds->ds_phys->ds_prev_snap_obj != 0)
1247 1249 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1248 1250 else
1249 1251 mrs_used = 0;
1250 1252
1251 1253 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1252 1254
1253 1255 ASSERT3U(dlused, <=, mrs_used);
1254 1256 ds->ds_phys->ds_unique_bytes =
1255 1257 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1256 1258
1257 1259 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1258 1260 SPA_VERSION_UNIQUE_ACCURATE)
1259 1261 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1260 1262 }
1261 1263
1262 1264 struct killarg {
1263 1265 dsl_dataset_t *ds;
1264 1266 dmu_tx_t *tx;
1265 1267 };
1266 1268
1267 1269 /* ARGSUSED */
1268 1270 static int
1269 1271 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1270 1272 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1271 1273 {
1272 1274 struct killarg *ka = arg;
1273 1275 dmu_tx_t *tx = ka->tx;
1274 1276
1275 1277 if (bp == NULL)
1276 1278 return (0);
1277 1279
1278 1280 if (zb->zb_level == ZB_ZIL_LEVEL) {
1279 1281 ASSERT(zilog != NULL);
1280 1282 /*
1281 1283 * It's a block in the intent log. It has no
1282 1284 * accounting, so just free it.
1283 1285 */
1284 1286 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1285 1287 } else {
1286 1288 ASSERT(zilog == NULL);
1287 1289 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1288 1290 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1289 1291 }
1290 1292
1291 1293 return (0);
1292 1294 }
1293 1295
1294 1296 /* ARGSUSED */
1295 1297 static int
1296 1298 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1297 1299 {
1298 1300 dsl_dataset_t *ds = arg1;
1299 1301 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1300 1302 uint64_t count;
1301 1303 int err;
1302 1304
1303 1305 /*
1304 1306 * Can't delete a head dataset if there are snapshots of it.
1305 1307 * (Except if the only snapshots are from the branch we cloned
1306 1308 * from.)
1307 1309 */
1308 1310 if (ds->ds_prev != NULL &&
1309 1311 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1310 1312 return (EBUSY);
1311 1313
1312 1314 /*
1313 1315 * This is really a dsl_dir thing, but check it here so that
1314 1316 * we'll be less likely to leave this dataset inconsistent &
1315 1317 * nearly destroyed.
1316 1318 */
1317 1319 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1318 1320 if (err)
1319 1321 return (err);
1320 1322 if (count != 0)
1321 1323 return (EEXIST);
1322 1324
1323 1325 return (0);
1324 1326 }
1325 1327
1326 1328 /* ARGSUSED */
1327 1329 static void
1328 1330 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1329 1331 {
1330 1332 dsl_dataset_t *ds = arg1;
1331 1333 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1332 1334
1333 1335 /* Mark it as inconsistent on-disk, in case we crash */
1334 1336 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1335 1337 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1336 1338
1337 1339 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1338 1340 "dataset = %llu", ds->ds_object);
1339 1341 }
1340 1342
1341 1343 static int
1342 1344 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1343 1345 dmu_tx_t *tx)
1344 1346 {
1345 1347 dsl_dataset_t *ds = dsda->ds;
1346 1348 dsl_dataset_t *ds_prev = ds->ds_prev;
1347 1349
1348 1350 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1349 1351 struct dsl_ds_destroyarg ndsda = {0};
1350 1352
1351 1353 /*
1352 1354 * If we're not prepared to remove the origin, don't remove
1353 1355 * the clone either.
1354 1356 */
1355 1357 if (dsda->rm_origin == NULL) {
1356 1358 dsda->need_prep = B_TRUE;
1357 1359 return (EBUSY);
1358 1360 }
1359 1361
1360 1362 ndsda.ds = ds_prev;
1361 1363 ndsda.is_origin_rm = B_TRUE;
1362 1364 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1363 1365 }
1364 1366
1365 1367 /*
1366 1368 * If we're not going to remove the origin after all,
1367 1369 * undo the open context setup.
1368 1370 */
1369 1371 if (dsda->rm_origin != NULL) {
1370 1372 dsl_dataset_disown(dsda->rm_origin, tag);
1371 1373 dsda->rm_origin = NULL;
1372 1374 }
1373 1375
1374 1376 return (0);
1375 1377 }
1376 1378
1377 1379 /*
1378 1380 * If you add new checks here, you may need to add
1379 1381 * additional checks to the "temporary" case in
1380 1382 * snapshot_check() in dmu_objset.c.
1381 1383 */
1382 1384 /* ARGSUSED */
1383 1385 int
1384 1386 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1385 1387 {
1386 1388 struct dsl_ds_destroyarg *dsda = arg1;
1387 1389 dsl_dataset_t *ds = dsda->ds;
1388 1390
1389 1391 /* we have an owner hold, so noone else can destroy us */
1390 1392 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1391 1393
1392 1394 /*
1393 1395 * Only allow deferred destroy on pools that support it.
1394 1396 * NOTE: deferred destroy is only supported on snapshots.
1395 1397 */
1396 1398 if (dsda->defer) {
1397 1399 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1398 1400 SPA_VERSION_USERREFS)
1399 1401 return (ENOTSUP);
1400 1402 ASSERT(dsl_dataset_is_snapshot(ds));
1401 1403 return (0);
1402 1404 }
1403 1405
1404 1406 /*
1405 1407 * Can't delete a head dataset if there are snapshots of it.
1406 1408 * (Except if the only snapshots are from the branch we cloned
1407 1409 * from.)
1408 1410 */
1409 1411 if (ds->ds_prev != NULL &&
1410 1412 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1411 1413 return (EBUSY);
1412 1414
1413 1415 /*
1414 1416 * If we made changes this txg, traverse_dsl_dataset won't find
1415 1417 * them. Try again.
1416 1418 */
1417 1419 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1418 1420 return (EAGAIN);
1419 1421
1420 1422 if (dsl_dataset_is_snapshot(ds)) {
1421 1423 /*
1422 1424 * If this snapshot has an elevated user reference count,
1423 1425 * we can't destroy it yet.
1424 1426 */
1425 1427 if (ds->ds_userrefs > 0 && !dsda->releasing)
1426 1428 return (EBUSY);
1427 1429
1428 1430 mutex_enter(&ds->ds_lock);
1429 1431 /*
1430 1432 * Can't delete a branch point. However, if we're destroying
1431 1433 * a clone and removing its origin due to it having a user
1432 1434 * hold count of 0 and having been marked for deferred destroy,
1433 1435 * it's OK for the origin to have a single clone.
1434 1436 */
1435 1437 if (ds->ds_phys->ds_num_children >
1436 1438 (dsda->is_origin_rm ? 2 : 1)) {
1437 1439 mutex_exit(&ds->ds_lock);
1438 1440 return (EEXIST);
1439 1441 }
1440 1442 mutex_exit(&ds->ds_lock);
1441 1443 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1442 1444 return (dsl_dataset_origin_check(dsda, arg2, tx));
1443 1445 }
1444 1446
1445 1447 /* XXX we should do some i/o error checking... */
1446 1448 return (0);
1447 1449 }
1448 1450
1449 1451 struct refsarg {
1450 1452 kmutex_t lock;
1451 1453 boolean_t gone;
1452 1454 kcondvar_t cv;
1453 1455 };
1454 1456
1455 1457 /* ARGSUSED */
1456 1458 static void
1457 1459 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1458 1460 {
1459 1461 struct refsarg *arg = argv;
1460 1462
1461 1463 mutex_enter(&arg->lock);
1462 1464 arg->gone = TRUE;
1463 1465 cv_signal(&arg->cv);
1464 1466 mutex_exit(&arg->lock);
1465 1467 }
1466 1468
1467 1469 static void
1468 1470 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1469 1471 {
1470 1472 struct refsarg arg;
1471 1473
1472 1474 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1473 1475 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1474 1476 arg.gone = FALSE;
1475 1477 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1476 1478 dsl_dataset_refs_gone);
1477 1479 dmu_buf_rele(ds->ds_dbuf, tag);
1478 1480 mutex_enter(&arg.lock);
1479 1481 while (!arg.gone)
1480 1482 cv_wait(&arg.cv, &arg.lock);
1481 1483 ASSERT(arg.gone);
1482 1484 mutex_exit(&arg.lock);
1483 1485 ds->ds_dbuf = NULL;
1484 1486 ds->ds_phys = NULL;
1485 1487 mutex_destroy(&arg.lock);
1486 1488 cv_destroy(&arg.cv);
1487 1489 }
1488 1490
1489 1491 static void
1490 1492 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1491 1493 {
1492 1494 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1493 1495 uint64_t count;
1494 1496 int err;
1495 1497
1496 1498 ASSERT(ds->ds_phys->ds_num_children >= 2);
1497 1499 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1498 1500 /*
1499 1501 * The err should not be ENOENT, but a bug in a previous version
1500 1502 * of the code could cause upgrade_clones_cb() to not set
1501 1503 * ds_next_snap_obj when it should, leading to a missing entry.
1502 1504 * If we knew that the pool was created after
1503 1505 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1504 1506 * ENOENT. However, at least we can check that we don't have
1505 1507 * too many entries in the next_clones_obj even after failing to
1506 1508 * remove this one.
1507 1509 */
1508 1510 if (err != ENOENT) {
1509 1511 VERIFY3U(err, ==, 0);
1510 1512 }
1511 1513 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1512 1514 &count));
1513 1515 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1514 1516 }
1515 1517
1516 1518 static void
1517 1519 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1518 1520 {
1519 1521 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1520 1522 zap_cursor_t zc;
1521 1523 zap_attribute_t za;
1522 1524
1523 1525 /*
1524 1526 * If it is the old version, dd_clones doesn't exist so we can't
1525 1527 * find the clones, but deadlist_remove_key() is a no-op so it
1526 1528 * doesn't matter.
1527 1529 */
1528 1530 if (ds->ds_dir->dd_phys->dd_clones == 0)
1529 1531 return;
1530 1532
1531 1533 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1532 1534 zap_cursor_retrieve(&zc, &za) == 0;
1533 1535 zap_cursor_advance(&zc)) {
1534 1536 dsl_dataset_t *clone;
1535 1537
1536 1538 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1537 1539 za.za_first_integer, FTAG, &clone));
1538 1540 if (clone->ds_dir->dd_origin_txg > mintxg) {
1539 1541 dsl_deadlist_remove_key(&clone->ds_deadlist,
1540 1542 mintxg, tx);
1541 1543 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1542 1544 }
1543 1545 dsl_dataset_rele(clone, FTAG);
1544 1546 }
1545 1547 zap_cursor_fini(&zc);
1546 1548 }
1547 1549
1548 1550 struct process_old_arg {
1549 1551 dsl_dataset_t *ds;
1550 1552 dsl_dataset_t *ds_prev;
1551 1553 boolean_t after_branch_point;
1552 1554 zio_t *pio;
1553 1555 uint64_t used, comp, uncomp;
1554 1556 };
1555 1557
1556 1558 static int
1557 1559 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1558 1560 {
1559 1561 struct process_old_arg *poa = arg;
1560 1562 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1561 1563
1562 1564 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1563 1565 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1564 1566 if (poa->ds_prev && !poa->after_branch_point &&
1565 1567 bp->blk_birth >
1566 1568 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1567 1569 poa->ds_prev->ds_phys->ds_unique_bytes +=
1568 1570 bp_get_dsize_sync(dp->dp_spa, bp);
1569 1571 }
1570 1572 } else {
1571 1573 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1572 1574 poa->comp += BP_GET_PSIZE(bp);
1573 1575 poa->uncomp += BP_GET_UCSIZE(bp);
1574 1576 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1575 1577 }
1576 1578 return (0);
1577 1579 }
1578 1580
1579 1581 static void
1580 1582 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1581 1583 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1582 1584 {
1583 1585 struct process_old_arg poa = { 0 };
1584 1586 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1585 1587 objset_t *mos = dp->dp_meta_objset;
1586 1588
1587 1589 ASSERT(ds->ds_deadlist.dl_oldfmt);
1588 1590 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1589 1591
1590 1592 poa.ds = ds;
1591 1593 poa.ds_prev = ds_prev;
1592 1594 poa.after_branch_point = after_branch_point;
1593 1595 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1594 1596 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1595 1597 process_old_cb, &poa, tx));
1596 1598 VERIFY3U(zio_wait(poa.pio), ==, 0);
1597 1599 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1598 1600
1599 1601 /* change snapused */
1600 1602 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1601 1603 -poa.used, -poa.comp, -poa.uncomp, tx);
1602 1604
1603 1605 /* swap next's deadlist to our deadlist */
1604 1606 dsl_deadlist_close(&ds->ds_deadlist);
1605 1607 dsl_deadlist_close(&ds_next->ds_deadlist);
1606 1608 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1607 1609 ds->ds_phys->ds_deadlist_obj);
1608 1610 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1609 1611 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1610 1612 ds_next->ds_phys->ds_deadlist_obj);
1611 1613 }
1612 1614
1613 1615 static int
1614 1616 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1615 1617 {
1616 1618 int err;
1617 1619 struct killarg ka;
1618 1620
1619 1621 /*
1620 1622 * Free everything that we point to (that's born after
1621 1623 * the previous snapshot, if we are a clone)
1622 1624 *
1623 1625 * NB: this should be very quick, because we already
1624 1626 * freed all the objects in open context.
1625 1627 */
1626 1628 ka.ds = ds;
1627 1629 ka.tx = tx;
1628 1630 err = traverse_dataset(ds,
1629 1631 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1630 1632 kill_blkptr, &ka);
1631 1633 ASSERT3U(err, ==, 0);
1632 1634 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1633 1635
1634 1636 return (err);
1635 1637 }
1636 1638
1637 1639 void
1638 1640 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1639 1641 {
1640 1642 struct dsl_ds_destroyarg *dsda = arg1;
1641 1643 dsl_dataset_t *ds = dsda->ds;
1642 1644 int err;
1643 1645 int after_branch_point = FALSE;
1644 1646 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1645 1647 objset_t *mos = dp->dp_meta_objset;
1646 1648 dsl_dataset_t *ds_prev = NULL;
1647 1649 boolean_t wont_destroy;
1648 1650 uint64_t obj;
1649 1651
1650 1652 wont_destroy = (dsda->defer &&
1651 1653 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1652 1654
1653 1655 ASSERT(ds->ds_owner || wont_destroy);
1654 1656 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1655 1657 ASSERT(ds->ds_prev == NULL ||
1656 1658 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1657 1659 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1658 1660
1659 1661 if (wont_destroy) {
1660 1662 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1661 1663 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1662 1664 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1663 1665 return;
1664 1666 }
1665 1667
1666 1668 /* signal any waiters that this dataset is going away */
1667 1669 mutex_enter(&ds->ds_lock);
1668 1670 ds->ds_owner = dsl_reaper;
1669 1671 cv_broadcast(&ds->ds_exclusive_cv);
1670 1672 mutex_exit(&ds->ds_lock);
1671 1673
1672 1674 /* Remove our reservation */
1673 1675 if (ds->ds_reserved != 0) {
1674 1676 dsl_prop_setarg_t psa;
1675 1677 uint64_t value = 0;
1676 1678
1677 1679 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1678 1680 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1679 1681 &value);
1680 1682 psa.psa_effective_value = 0; /* predict default value */
1681 1683
1682 1684 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1683 1685 ASSERT3U(ds->ds_reserved, ==, 0);
1684 1686 }
1685 1687
1686 1688 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1687 1689
1688 1690 dsl_scan_ds_destroyed(ds, tx);
1689 1691
1690 1692 obj = ds->ds_object;
1691 1693
1692 1694 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1693 1695 if (ds->ds_prev) {
1694 1696 ds_prev = ds->ds_prev;
1695 1697 } else {
1696 1698 VERIFY(0 == dsl_dataset_hold_obj(dp,
1697 1699 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1698 1700 }
1699 1701 after_branch_point =
1700 1702 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1701 1703
1702 1704 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1703 1705 if (after_branch_point &&
1704 1706 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1705 1707 remove_from_next_clones(ds_prev, obj, tx);
1706 1708 if (ds->ds_phys->ds_next_snap_obj != 0) {
1707 1709 VERIFY(0 == zap_add_int(mos,
1708 1710 ds_prev->ds_phys->ds_next_clones_obj,
1709 1711 ds->ds_phys->ds_next_snap_obj, tx));
1710 1712 }
1711 1713 }
1712 1714 if (after_branch_point &&
1713 1715 ds->ds_phys->ds_next_snap_obj == 0) {
1714 1716 /* This clone is toast. */
1715 1717 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1716 1718 ds_prev->ds_phys->ds_num_children--;
1717 1719
1718 1720 /*
1719 1721 * If the clone's origin has no other clones, no
1720 1722 * user holds, and has been marked for deferred
1721 1723 * deletion, then we should have done the necessary
1722 1724 * destroy setup for it.
1723 1725 */
1724 1726 if (ds_prev->ds_phys->ds_num_children == 1 &&
1725 1727 ds_prev->ds_userrefs == 0 &&
1726 1728 DS_IS_DEFER_DESTROY(ds_prev)) {
1727 1729 ASSERT3P(dsda->rm_origin, !=, NULL);
1728 1730 } else {
1729 1731 ASSERT3P(dsda->rm_origin, ==, NULL);
1730 1732 }
1731 1733 } else if (!after_branch_point) {
1732 1734 ds_prev->ds_phys->ds_next_snap_obj =
1733 1735 ds->ds_phys->ds_next_snap_obj;
1734 1736 }
1735 1737 }
1736 1738
1737 1739 if (dsl_dataset_is_snapshot(ds)) {
1738 1740 dsl_dataset_t *ds_next;
1739 1741 uint64_t old_unique;
1740 1742 uint64_t used = 0, comp = 0, uncomp = 0;
1741 1743
1742 1744 VERIFY(0 == dsl_dataset_hold_obj(dp,
1743 1745 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1744 1746 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1745 1747
1746 1748 old_unique = ds_next->ds_phys->ds_unique_bytes;
1747 1749
1748 1750 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1749 1751 ds_next->ds_phys->ds_prev_snap_obj =
1750 1752 ds->ds_phys->ds_prev_snap_obj;
1751 1753 ds_next->ds_phys->ds_prev_snap_txg =
1752 1754 ds->ds_phys->ds_prev_snap_txg;
1753 1755 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1754 1756 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1755 1757
1756 1758
1757 1759 if (ds_next->ds_deadlist.dl_oldfmt) {
1758 1760 process_old_deadlist(ds, ds_prev, ds_next,
1759 1761 after_branch_point, tx);
1760 1762 } else {
1761 1763 /* Adjust prev's unique space. */
1762 1764 if (ds_prev && !after_branch_point) {
1763 1765 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1764 1766 ds_prev->ds_phys->ds_prev_snap_txg,
1765 1767 ds->ds_phys->ds_prev_snap_txg,
1766 1768 &used, &comp, &uncomp);
1767 1769 ds_prev->ds_phys->ds_unique_bytes += used;
1768 1770 }
1769 1771
1770 1772 /* Adjust snapused. */
1771 1773 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1772 1774 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1773 1775 &used, &comp, &uncomp);
1774 1776 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1775 1777 -used, -comp, -uncomp, tx);
1776 1778
1777 1779 /* Move blocks to be freed to pool's free list. */
1778 1780 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1779 1781 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1780 1782 tx);
1781 1783 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1782 1784 DD_USED_HEAD, used, comp, uncomp, tx);
1783 1785
1784 1786 /* Merge our deadlist into next's and free it. */
1785 1787 dsl_deadlist_merge(&ds_next->ds_deadlist,
1786 1788 ds->ds_phys->ds_deadlist_obj, tx);
1787 1789 }
1788 1790 dsl_deadlist_close(&ds->ds_deadlist);
1789 1791 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1790 1792
1791 1793 /* Collapse range in clone heads */
1792 1794 dsl_dataset_remove_clones_key(ds,
1793 1795 ds->ds_phys->ds_creation_txg, tx);
1794 1796
1795 1797 if (dsl_dataset_is_snapshot(ds_next)) {
1796 1798 dsl_dataset_t *ds_nextnext;
1797 1799
1798 1800 /*
1799 1801 * Update next's unique to include blocks which
1800 1802 * were previously shared by only this snapshot
1801 1803 * and it. Those blocks will be born after the
1802 1804 * prev snap and before this snap, and will have
1803 1805 * died after the next snap and before the one
1804 1806 * after that (ie. be on the snap after next's
1805 1807 * deadlist).
1806 1808 */
1807 1809 VERIFY(0 == dsl_dataset_hold_obj(dp,
1808 1810 ds_next->ds_phys->ds_next_snap_obj,
1809 1811 FTAG, &ds_nextnext));
1810 1812 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1811 1813 ds->ds_phys->ds_prev_snap_txg,
1812 1814 ds->ds_phys->ds_creation_txg,
1813 1815 &used, &comp, &uncomp);
1814 1816 ds_next->ds_phys->ds_unique_bytes += used;
1815 1817 dsl_dataset_rele(ds_nextnext, FTAG);
1816 1818 ASSERT3P(ds_next->ds_prev, ==, NULL);
1817 1819
1818 1820 /* Collapse range in this head. */
1819 1821 dsl_dataset_t *hds;
1820 1822 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1821 1823 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1822 1824 FTAG, &hds));
1823 1825 dsl_deadlist_remove_key(&hds->ds_deadlist,
1824 1826 ds->ds_phys->ds_creation_txg, tx);
1825 1827 dsl_dataset_rele(hds, FTAG);
1826 1828
1827 1829 } else {
1828 1830 ASSERT3P(ds_next->ds_prev, ==, ds);
1829 1831 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1830 1832 ds_next->ds_prev = NULL;
1831 1833 if (ds_prev) {
1832 1834 VERIFY(0 == dsl_dataset_get_ref(dp,
1833 1835 ds->ds_phys->ds_prev_snap_obj,
1834 1836 ds_next, &ds_next->ds_prev));
1835 1837 }
1836 1838
1837 1839 dsl_dataset_recalc_head_uniq(ds_next);
1838 1840
1839 1841 /*
1840 1842 * Reduce the amount of our unconsmed refreservation
1841 1843 * being charged to our parent by the amount of
1842 1844 * new unique data we have gained.
1843 1845 */
1844 1846 if (old_unique < ds_next->ds_reserved) {
1845 1847 int64_t mrsdelta;
1846 1848 uint64_t new_unique =
1847 1849 ds_next->ds_phys->ds_unique_bytes;
1848 1850
1849 1851 ASSERT(old_unique <= new_unique);
1850 1852 mrsdelta = MIN(new_unique - old_unique,
1851 1853 ds_next->ds_reserved - old_unique);
1852 1854 dsl_dir_diduse_space(ds->ds_dir,
1853 1855 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1854 1856 }
1855 1857 }
1856 1858 dsl_dataset_rele(ds_next, FTAG);
1857 1859 } else {
1858 1860 zfeature_info_t *async_destroy =
1859 1861 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1860 1862
1861 1863 /*
1862 1864 * There's no next snapshot, so this is a head dataset.
1863 1865 * Destroy the deadlist. Unless it's a clone, the
1864 1866 * deadlist should be empty. (If it's a clone, it's
1865 1867 * safe to ignore the deadlist contents.)
1866 1868 */
1867 1869 dsl_deadlist_close(&ds->ds_deadlist);
1868 1870 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1869 1871 ds->ds_phys->ds_deadlist_obj = 0;
1870 1872
1871 1873 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1872 1874 err = old_synchronous_dataset_destroy(ds, tx);
1873 1875 } else {
1874 1876 /*
1875 1877 * Move the bptree into the pool's list of trees to
1876 1878 * clean up and update space accounting information.
1877 1879 */
1878 1880 uint64_t used, comp, uncomp;
1879 1881
1880 1882 ASSERT(err == 0 || err == EBUSY);
1881 1883 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1882 1884 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1883 1885 dp->dp_bptree_obj = bptree_alloc(
1884 1886 dp->dp_meta_objset, tx);
1885 1887 VERIFY(zap_add(dp->dp_meta_objset,
1886 1888 DMU_POOL_DIRECTORY_OBJECT,
1887 1889 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1888 1890 &dp->dp_bptree_obj, tx) == 0);
1889 1891 }
1890 1892
1891 1893 used = ds->ds_dir->dd_phys->dd_used_bytes;
1892 1894 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1893 1895 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1894 1896
1895 1897 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1896 1898 ds->ds_phys->ds_unique_bytes == used);
1897 1899
1898 1900 bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
1899 1901 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1900 1902 used, comp, uncomp, tx);
1901 1903 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1902 1904 -used, -comp, -uncomp, tx);
1903 1905 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1904 1906 used, comp, uncomp, tx);
1905 1907 }
1906 1908
1907 1909 if (ds->ds_prev != NULL) {
1908 1910 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1909 1911 VERIFY3U(0, ==, zap_remove_int(mos,
1910 1912 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1911 1913 ds->ds_object, tx));
1912 1914 }
1913 1915 dsl_dataset_rele(ds->ds_prev, ds);
1914 1916 ds->ds_prev = ds_prev = NULL;
1915 1917 }
1916 1918 }
1917 1919
1918 1920 /*
1919 1921 * This must be done after the dsl_traverse(), because it will
1920 1922 * re-open the objset.
1921 1923 */
1922 1924 if (ds->ds_objset) {
1923 1925 dmu_objset_evict(ds->ds_objset);
1924 1926 ds->ds_objset = NULL;
1925 1927 }
1926 1928
1927 1929 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1928 1930 /* Erase the link in the dir */
1929 1931 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1930 1932 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1931 1933 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1932 1934 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1933 1935 ASSERT(err == 0);
1934 1936 } else {
1935 1937 /* remove from snapshot namespace */
1936 1938 dsl_dataset_t *ds_head;
1937 1939 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1938 1940 VERIFY(0 == dsl_dataset_hold_obj(dp,
1939 1941 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1940 1942 VERIFY(0 == dsl_dataset_get_snapname(ds));
1941 1943 #ifdef ZFS_DEBUG
1942 1944 {
1943 1945 uint64_t val;
1944 1946
1945 1947 err = dsl_dataset_snap_lookup(ds_head,
1946 1948 ds->ds_snapname, &val);
1947 1949 ASSERT3U(err, ==, 0);
1948 1950 ASSERT3U(val, ==, obj);
1949 1951 }
1950 1952 #endif
1951 1953 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1952 1954 ASSERT(err == 0);
1953 1955 dsl_dataset_rele(ds_head, FTAG);
1954 1956 }
1955 1957
1956 1958 if (ds_prev && ds->ds_prev != ds_prev)
1957 1959 dsl_dataset_rele(ds_prev, FTAG);
1958 1960
1959 1961 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1960 1962 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1961 1963 "dataset = %llu", ds->ds_object);
1962 1964
1963 1965 if (ds->ds_phys->ds_next_clones_obj != 0) {
1964 1966 uint64_t count;
1965 1967 ASSERT(0 == zap_count(mos,
1966 1968 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1967 1969 VERIFY(0 == dmu_object_free(mos,
1968 1970 ds->ds_phys->ds_next_clones_obj, tx));
1969 1971 }
1970 1972 if (ds->ds_phys->ds_props_obj != 0)
1971 1973 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1972 1974 if (ds->ds_phys->ds_userrefs_obj != 0)
1973 1975 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1974 1976 dsl_dir_close(ds->ds_dir, ds);
1975 1977 ds->ds_dir = NULL;
1976 1978 dsl_dataset_drain_refs(ds, tag);
1977 1979 VERIFY(0 == dmu_object_free(mos, obj, tx));
1978 1980
1979 1981 if (dsda->rm_origin) {
1980 1982 /*
1981 1983 * Remove the origin of the clone we just destroyed.
1982 1984 */
1983 1985 struct dsl_ds_destroyarg ndsda = {0};
1984 1986
1985 1987 ndsda.ds = dsda->rm_origin;
1986 1988 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1987 1989 }
1988 1990 }
1989 1991
1990 1992 static int
1991 1993 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1992 1994 {
1993 1995 uint64_t asize;
1994 1996
1995 1997 if (!dmu_tx_is_syncing(tx))
1996 1998 return (0);
1997 1999
1998 2000 /*
1999 2001 * If there's an fs-only reservation, any blocks that might become
2000 2002 * owned by the snapshot dataset must be accommodated by space
2001 2003 * outside of the reservation.
2002 2004 */
2003 2005 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2004 2006 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2005 2007 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2006 2008 return (ENOSPC);
2007 2009
2008 2010 /*
2009 2011 * Propogate any reserved space for this snapshot to other
2010 2012 * snapshot checks in this sync group.
2011 2013 */
2012 2014 if (asize > 0)
2013 2015 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2014 2016
2015 2017 return (0);
2016 2018 }
2017 2019
2018 2020 int
2019 2021 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
2020 2022 {
2021 2023 dsl_dataset_t *ds = arg1;
2022 2024 const char *snapname = arg2;
2023 2025 int err;
2024 2026 uint64_t value;
2025 2027
2026 2028 /*
2027 2029 * We don't allow multiple snapshots of the same txg. If there
2028 2030 * is already one, try again.
2029 2031 */
2030 2032 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2031 2033 return (EAGAIN);
2032 2034
2033 2035 /*
2034 2036 * Check for conflicting name snapshot name.
2035 2037 */
2036 2038 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2037 2039 if (err == 0)
2038 2040 return (EEXIST);
2039 2041 if (err != ENOENT)
2040 2042 return (err);
2041 2043
2042 2044 /*
2043 2045 * Check that the dataset's name is not too long. Name consists
2044 2046 * of the dataset's length + 1 for the @-sign + snapshot name's length
2045 2047 */
2046 2048 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2047 2049 return (ENAMETOOLONG);
2048 2050
2049 2051 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2050 2052 if (err)
2051 2053 return (err);
2052 2054
2053 2055 ds->ds_trysnap_txg = tx->tx_txg;
2054 2056 return (0);
2055 2057 }
2056 2058
2057 2059 void
2058 2060 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2059 2061 {
2060 2062 dsl_dataset_t *ds = arg1;
2061 2063 const char *snapname = arg2;
2062 2064 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2063 2065 dmu_buf_t *dbuf;
2064 2066 dsl_dataset_phys_t *dsphys;
2065 2067 uint64_t dsobj, crtxg;
2066 2068 objset_t *mos = dp->dp_meta_objset;
2067 2069 int err;
2068 2070
2069 2071 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2070 2072
2071 2073 /*
2072 2074 * The origin's ds_creation_txg has to be < TXG_INITIAL
2073 2075 */
2074 2076 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2075 2077 crtxg = 1;
2076 2078 else
|
↓ open down ↓ |
1259 lines elided |
↑ open up ↑ |
2077 2079 crtxg = tx->tx_txg;
2078 2080
2079 2081 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2080 2082 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2081 2083 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2082 2084 dmu_buf_will_dirty(dbuf, tx);
2083 2085 dsphys = dbuf->db_data;
2084 2086 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2085 2087 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2086 2088 dsphys->ds_fsid_guid = unique_create();
2087 - (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2088 - sizeof (dsphys->ds_guid));
2089 + do {
2090 + (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2091 + sizeof (dsphys->ds_guid));
2092 + } while (dsphys->ds_guid == 0);
2089 2093 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2090 2094 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2091 2095 dsphys->ds_next_snap_obj = ds->ds_object;
2092 2096 dsphys->ds_num_children = 1;
2093 2097 dsphys->ds_creation_time = gethrestime_sec();
2094 2098 dsphys->ds_creation_txg = crtxg;
2095 2099 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2096 2100 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2097 2101 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2098 2102 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2099 2103 dsphys->ds_flags = ds->ds_phys->ds_flags;
2100 2104 dsphys->ds_bp = ds->ds_phys->ds_bp;
2101 2105 dmu_buf_rele(dbuf, FTAG);
2102 2106
2103 2107 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2104 2108 if (ds->ds_prev) {
2105 2109 uint64_t next_clones_obj =
2106 2110 ds->ds_prev->ds_phys->ds_next_clones_obj;
2107 2111 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2108 2112 ds->ds_object ||
2109 2113 ds->ds_prev->ds_phys->ds_num_children > 1);
2110 2114 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2111 2115 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2112 2116 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2113 2117 ds->ds_prev->ds_phys->ds_creation_txg);
2114 2118 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2115 2119 } else if (next_clones_obj != 0) {
2116 2120 remove_from_next_clones(ds->ds_prev,
2117 2121 dsphys->ds_next_snap_obj, tx);
2118 2122 VERIFY3U(0, ==, zap_add_int(mos,
2119 2123 next_clones_obj, dsobj, tx));
2120 2124 }
2121 2125 }
2122 2126
2123 2127 /*
2124 2128 * If we have a reference-reservation on this dataset, we will
2125 2129 * need to increase the amount of refreservation being charged
2126 2130 * since our unique space is going to zero.
2127 2131 */
2128 2132 if (ds->ds_reserved) {
2129 2133 int64_t delta;
2130 2134 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2131 2135 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2132 2136 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2133 2137 delta, 0, 0, tx);
2134 2138 }
2135 2139
2136 2140 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2137 2141 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2138 2142 ds->ds_dir->dd_myname, snapname, dsobj,
2139 2143 ds->ds_phys->ds_prev_snap_txg);
2140 2144 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2141 2145 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2142 2146 dsl_deadlist_close(&ds->ds_deadlist);
2143 2147 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2144 2148 dsl_deadlist_add_key(&ds->ds_deadlist,
2145 2149 ds->ds_phys->ds_prev_snap_txg, tx);
2146 2150
2147 2151 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2148 2152 ds->ds_phys->ds_prev_snap_obj = dsobj;
2149 2153 ds->ds_phys->ds_prev_snap_txg = crtxg;
2150 2154 ds->ds_phys->ds_unique_bytes = 0;
2151 2155 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2152 2156 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2153 2157
2154 2158 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2155 2159 snapname, 8, 1, &dsobj, tx);
2156 2160 ASSERT(err == 0);
2157 2161
2158 2162 if (ds->ds_prev)
2159 2163 dsl_dataset_drop_ref(ds->ds_prev, ds);
2160 2164 VERIFY(0 == dsl_dataset_get_ref(dp,
2161 2165 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2162 2166
2163 2167 dsl_scan_ds_snapshotted(ds, tx);
2164 2168
2165 2169 dsl_dir_snap_cmtime_update(ds->ds_dir);
2166 2170
2167 2171 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2168 2172 "dataset = %llu", dsobj);
2169 2173 }
2170 2174
2171 2175 void
2172 2176 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2173 2177 {
2174 2178 ASSERT(dmu_tx_is_syncing(tx));
2175 2179 ASSERT(ds->ds_objset != NULL);
2176 2180 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2177 2181
2178 2182 /*
2179 2183 * in case we had to change ds_fsid_guid when we opened it,
2180 2184 * sync it out now.
2181 2185 */
2182 2186 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2183 2187 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2184 2188
2185 2189 dsl_dir_dirty(ds->ds_dir, tx);
2186 2190 dmu_objset_sync(ds->ds_objset, zio, tx);
2187 2191 }
2188 2192
2189 2193 static void
2190 2194 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2191 2195 {
2192 2196 uint64_t count = 0;
2193 2197 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2194 2198 zap_cursor_t zc;
2195 2199 zap_attribute_t za;
2196 2200 nvlist_t *propval;
2197 2201 nvlist_t *val;
2198 2202
2199 2203 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2200 2204 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2201 2205 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2202 2206
2203 2207 /*
2204 2208 * There may me missing entries in ds_next_clones_obj
2205 2209 * due to a bug in a previous version of the code.
2206 2210 * Only trust it if it has the right number of entries.
2207 2211 */
2208 2212 if (ds->ds_phys->ds_next_clones_obj != 0) {
2209 2213 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2210 2214 &count));
2211 2215 }
2212 2216 if (count != ds->ds_phys->ds_num_children - 1) {
2213 2217 goto fail;
2214 2218 }
2215 2219 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2216 2220 zap_cursor_retrieve(&zc, &za) == 0;
2217 2221 zap_cursor_advance(&zc)) {
2218 2222 dsl_dataset_t *clone;
2219 2223 char buf[ZFS_MAXNAMELEN];
2220 2224 /*
2221 2225 * Even though we hold the dp_config_rwlock, the dataset
2222 2226 * may fail to open, returning ENOENT. If there is a
2223 2227 * thread concurrently attempting to destroy this
2224 2228 * dataset, it will have the ds_rwlock held for
2225 2229 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2226 2230 * dsl_dataset_hold_ref() will fail its
2227 2231 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2228 2232 * dp_config_rwlock, and wait for the destroy progress
2229 2233 * and signal ds_exclusive_cv. If the destroy was
2230 2234 * successful, we will see that
2231 2235 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2232 2236 */
2233 2237 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2234 2238 za.za_first_integer, FTAG, &clone) != 0)
2235 2239 continue;
2236 2240 dsl_dir_name(clone->ds_dir, buf);
2237 2241 VERIFY(nvlist_add_boolean(val, buf) == 0);
2238 2242 dsl_dataset_rele(clone, FTAG);
2239 2243 }
2240 2244 zap_cursor_fini(&zc);
2241 2245 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2242 2246 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2243 2247 propval) == 0);
2244 2248 fail:
2245 2249 nvlist_free(val);
2246 2250 nvlist_free(propval);
2247 2251 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2248 2252 }
2249 2253
2250 2254 void
2251 2255 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2252 2256 {
2253 2257 uint64_t refd, avail, uobjs, aobjs, ratio;
2254 2258
2255 2259 dsl_dir_stats(ds->ds_dir, nv);
2256 2260
2257 2261 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2258 2262 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2259 2263 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2260 2264
2261 2265 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2262 2266 ds->ds_phys->ds_creation_time);
2263 2267 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2264 2268 ds->ds_phys->ds_creation_txg);
2265 2269 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2266 2270 ds->ds_quota);
2267 2271 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2268 2272 ds->ds_reserved);
2269 2273 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2270 2274 ds->ds_phys->ds_guid);
2271 2275 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2272 2276 ds->ds_phys->ds_unique_bytes);
2273 2277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2274 2278 ds->ds_object);
2275 2279 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2276 2280 ds->ds_userrefs);
2277 2281 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2278 2282 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2279 2283
2280 2284 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2281 2285 uint64_t written, comp, uncomp;
2282 2286 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2283 2287 dsl_dataset_t *prev;
2284 2288
2285 2289 rw_enter(&dp->dp_config_rwlock, RW_READER);
2286 2290 int err = dsl_dataset_hold_obj(dp,
2287 2291 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2288 2292 rw_exit(&dp->dp_config_rwlock);
2289 2293 if (err == 0) {
2290 2294 err = dsl_dataset_space_written(prev, ds, &written,
2291 2295 &comp, &uncomp);
2292 2296 dsl_dataset_rele(prev, FTAG);
2293 2297 if (err == 0) {
2294 2298 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2295 2299 written);
2296 2300 }
2297 2301 }
2298 2302 }
2299 2303
2300 2304 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2301 2305 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2302 2306 ds->ds_phys->ds_compressed_bytes);
2303 2307 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2304 2308
2305 2309 if (ds->ds_phys->ds_next_snap_obj) {
2306 2310 /*
2307 2311 * This is a snapshot; override the dd's space used with
2308 2312 * our unique space and compression ratio.
2309 2313 */
2310 2314 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2311 2315 ds->ds_phys->ds_unique_bytes);
2312 2316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2313 2317
2314 2318 get_clones_stat(ds, nv);
2315 2319 }
2316 2320 }
2317 2321
2318 2322 void
2319 2323 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2320 2324 {
2321 2325 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2322 2326 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2323 2327 stat->dds_guid = ds->ds_phys->ds_guid;
2324 2328 if (ds->ds_phys->ds_next_snap_obj) {
2325 2329 stat->dds_is_snapshot = B_TRUE;
2326 2330 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2327 2331 } else {
2328 2332 stat->dds_is_snapshot = B_FALSE;
2329 2333 stat->dds_num_clones = 0;
2330 2334 }
2331 2335
2332 2336 /* clone origin is really a dsl_dir thing... */
2333 2337 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2334 2338 if (dsl_dir_is_clone(ds->ds_dir)) {
2335 2339 dsl_dataset_t *ods;
2336 2340
2337 2341 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2338 2342 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2339 2343 dsl_dataset_name(ods, stat->dds_origin);
2340 2344 dsl_dataset_drop_ref(ods, FTAG);
2341 2345 } else {
2342 2346 stat->dds_origin[0] = '\0';
2343 2347 }
2344 2348 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2345 2349 }
2346 2350
2347 2351 uint64_t
2348 2352 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2349 2353 {
2350 2354 return (ds->ds_fsid_guid);
2351 2355 }
2352 2356
2353 2357 void
2354 2358 dsl_dataset_space(dsl_dataset_t *ds,
2355 2359 uint64_t *refdbytesp, uint64_t *availbytesp,
2356 2360 uint64_t *usedobjsp, uint64_t *availobjsp)
2357 2361 {
2358 2362 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2359 2363 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2360 2364 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2361 2365 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2362 2366 if (ds->ds_quota != 0) {
2363 2367 /*
2364 2368 * Adjust available bytes according to refquota
2365 2369 */
2366 2370 if (*refdbytesp < ds->ds_quota)
2367 2371 *availbytesp = MIN(*availbytesp,
2368 2372 ds->ds_quota - *refdbytesp);
2369 2373 else
2370 2374 *availbytesp = 0;
2371 2375 }
2372 2376 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2373 2377 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2374 2378 }
2375 2379
2376 2380 boolean_t
2377 2381 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2378 2382 {
2379 2383 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2380 2384
2381 2385 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2382 2386 dsl_pool_sync_context(dp));
2383 2387 if (ds->ds_prev == NULL)
2384 2388 return (B_FALSE);
2385 2389 if (ds->ds_phys->ds_bp.blk_birth >
2386 2390 ds->ds_prev->ds_phys->ds_creation_txg) {
2387 2391 objset_t *os, *os_prev;
2388 2392 /*
2389 2393 * It may be that only the ZIL differs, because it was
2390 2394 * reset in the head. Don't count that as being
2391 2395 * modified.
2392 2396 */
2393 2397 if (dmu_objset_from_ds(ds, &os) != 0)
2394 2398 return (B_TRUE);
2395 2399 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2396 2400 return (B_TRUE);
2397 2401 return (bcmp(&os->os_phys->os_meta_dnode,
2398 2402 &os_prev->os_phys->os_meta_dnode,
2399 2403 sizeof (os->os_phys->os_meta_dnode)) != 0);
2400 2404 }
2401 2405 return (B_FALSE);
2402 2406 }
2403 2407
2404 2408 /* ARGSUSED */
2405 2409 static int
2406 2410 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2407 2411 {
2408 2412 dsl_dataset_t *ds = arg1;
2409 2413 char *newsnapname = arg2;
2410 2414 dsl_dir_t *dd = ds->ds_dir;
2411 2415 dsl_dataset_t *hds;
2412 2416 uint64_t val;
2413 2417 int err;
2414 2418
2415 2419 err = dsl_dataset_hold_obj(dd->dd_pool,
2416 2420 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2417 2421 if (err)
2418 2422 return (err);
2419 2423
2420 2424 /* new name better not be in use */
2421 2425 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2422 2426 dsl_dataset_rele(hds, FTAG);
2423 2427
2424 2428 if (err == 0)
2425 2429 err = EEXIST;
2426 2430 else if (err == ENOENT)
2427 2431 err = 0;
2428 2432
2429 2433 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2430 2434 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2431 2435 err = ENAMETOOLONG;
2432 2436
2433 2437 return (err);
2434 2438 }
2435 2439
2436 2440 static void
2437 2441 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2438 2442 {
2439 2443 dsl_dataset_t *ds = arg1;
2440 2444 const char *newsnapname = arg2;
2441 2445 dsl_dir_t *dd = ds->ds_dir;
2442 2446 objset_t *mos = dd->dd_pool->dp_meta_objset;
2443 2447 dsl_dataset_t *hds;
2444 2448 int err;
2445 2449
2446 2450 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2447 2451
2448 2452 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2449 2453 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2450 2454
2451 2455 VERIFY(0 == dsl_dataset_get_snapname(ds));
2452 2456 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2453 2457 ASSERT3U(err, ==, 0);
2454 2458 mutex_enter(&ds->ds_lock);
2455 2459 (void) strcpy(ds->ds_snapname, newsnapname);
2456 2460 mutex_exit(&ds->ds_lock);
2457 2461 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2458 2462 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2459 2463 ASSERT3U(err, ==, 0);
2460 2464
2461 2465 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2462 2466 "dataset = %llu", ds->ds_object);
2463 2467 dsl_dataset_rele(hds, FTAG);
2464 2468 }
2465 2469
2466 2470 struct renamesnaparg {
2467 2471 dsl_sync_task_group_t *dstg;
2468 2472 char failed[MAXPATHLEN];
2469 2473 char *oldsnap;
2470 2474 char *newsnap;
2471 2475 };
2472 2476
2473 2477 static int
2474 2478 dsl_snapshot_rename_one(const char *name, void *arg)
2475 2479 {
2476 2480 struct renamesnaparg *ra = arg;
2477 2481 dsl_dataset_t *ds = NULL;
2478 2482 char *snapname;
2479 2483 int err;
2480 2484
2481 2485 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2482 2486 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2483 2487
2484 2488 /*
2485 2489 * For recursive snapshot renames the parent won't be changing
2486 2490 * so we just pass name for both the to/from argument.
2487 2491 */
2488 2492 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2489 2493 if (err != 0) {
2490 2494 strfree(snapname);
2491 2495 return (err == ENOENT ? 0 : err);
2492 2496 }
2493 2497
2494 2498 #ifdef _KERNEL
2495 2499 /*
2496 2500 * For all filesystems undergoing rename, we'll need to unmount it.
2497 2501 */
2498 2502 (void) zfs_unmount_snap(snapname, NULL);
2499 2503 #endif
2500 2504 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2501 2505 strfree(snapname);
2502 2506 if (err != 0)
2503 2507 return (err == ENOENT ? 0 : err);
2504 2508
2505 2509 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2506 2510 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2507 2511
2508 2512 return (0);
2509 2513 }
2510 2514
2511 2515 static int
2512 2516 dsl_recursive_rename(char *oldname, const char *newname)
2513 2517 {
2514 2518 int err;
2515 2519 struct renamesnaparg *ra;
2516 2520 dsl_sync_task_t *dst;
2517 2521 spa_t *spa;
2518 2522 char *cp, *fsname = spa_strdup(oldname);
2519 2523 int len = strlen(oldname) + 1;
2520 2524
2521 2525 /* truncate the snapshot name to get the fsname */
2522 2526 cp = strchr(fsname, '@');
2523 2527 *cp = '\0';
2524 2528
2525 2529 err = spa_open(fsname, &spa, FTAG);
2526 2530 if (err) {
2527 2531 kmem_free(fsname, len);
2528 2532 return (err);
2529 2533 }
2530 2534 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2531 2535 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2532 2536
2533 2537 ra->oldsnap = strchr(oldname, '@') + 1;
2534 2538 ra->newsnap = strchr(newname, '@') + 1;
2535 2539 *ra->failed = '\0';
2536 2540
2537 2541 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2538 2542 DS_FIND_CHILDREN);
2539 2543 kmem_free(fsname, len);
2540 2544
2541 2545 if (err == 0) {
2542 2546 err = dsl_sync_task_group_wait(ra->dstg);
2543 2547 }
2544 2548
2545 2549 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2546 2550 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2547 2551 dsl_dataset_t *ds = dst->dst_arg1;
2548 2552 if (dst->dst_err) {
2549 2553 dsl_dir_name(ds->ds_dir, ra->failed);
2550 2554 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2551 2555 (void) strlcat(ra->failed, ra->newsnap,
2552 2556 sizeof (ra->failed));
2553 2557 }
2554 2558 dsl_dataset_rele(ds, ra->dstg);
2555 2559 }
2556 2560
2557 2561 if (err)
2558 2562 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2559 2563
2560 2564 dsl_sync_task_group_destroy(ra->dstg);
2561 2565 kmem_free(ra, sizeof (struct renamesnaparg));
2562 2566 spa_close(spa, FTAG);
2563 2567 return (err);
2564 2568 }
2565 2569
2566 2570 static int
2567 2571 dsl_valid_rename(const char *oldname, void *arg)
2568 2572 {
2569 2573 int delta = *(int *)arg;
2570 2574
2571 2575 if (strlen(oldname) + delta >= MAXNAMELEN)
2572 2576 return (ENAMETOOLONG);
2573 2577
2574 2578 return (0);
2575 2579 }
2576 2580
2577 2581 #pragma weak dmu_objset_rename = dsl_dataset_rename
2578 2582 int
2579 2583 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2580 2584 {
2581 2585 dsl_dir_t *dd;
2582 2586 dsl_dataset_t *ds;
2583 2587 const char *tail;
2584 2588 int err;
2585 2589
2586 2590 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2587 2591 if (err)
2588 2592 return (err);
2589 2593
2590 2594 if (tail == NULL) {
2591 2595 int delta = strlen(newname) - strlen(oldname);
2592 2596
2593 2597 /* if we're growing, validate child name lengths */
2594 2598 if (delta > 0)
2595 2599 err = dmu_objset_find(oldname, dsl_valid_rename,
2596 2600 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2597 2601
2598 2602 if (err == 0)
2599 2603 err = dsl_dir_rename(dd, newname);
2600 2604 dsl_dir_close(dd, FTAG);
2601 2605 return (err);
2602 2606 }
2603 2607
2604 2608 if (tail[0] != '@') {
2605 2609 /* the name ended in a nonexistent component */
2606 2610 dsl_dir_close(dd, FTAG);
2607 2611 return (ENOENT);
2608 2612 }
2609 2613
2610 2614 dsl_dir_close(dd, FTAG);
2611 2615
2612 2616 /* new name must be snapshot in same filesystem */
2613 2617 tail = strchr(newname, '@');
2614 2618 if (tail == NULL)
2615 2619 return (EINVAL);
2616 2620 tail++;
2617 2621 if (strncmp(oldname, newname, tail - newname) != 0)
2618 2622 return (EXDEV);
2619 2623
2620 2624 if (recursive) {
2621 2625 err = dsl_recursive_rename(oldname, newname);
2622 2626 } else {
2623 2627 err = dsl_dataset_hold(oldname, FTAG, &ds);
2624 2628 if (err)
2625 2629 return (err);
2626 2630
2627 2631 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2628 2632 dsl_dataset_snapshot_rename_check,
2629 2633 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2630 2634
2631 2635 dsl_dataset_rele(ds, FTAG);
2632 2636 }
2633 2637
2634 2638 return (err);
2635 2639 }
2636 2640
2637 2641 struct promotenode {
2638 2642 list_node_t link;
2639 2643 dsl_dataset_t *ds;
2640 2644 };
2641 2645
2642 2646 struct promotearg {
2643 2647 list_t shared_snaps, origin_snaps, clone_snaps;
2644 2648 dsl_dataset_t *origin_origin;
2645 2649 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2646 2650 char *err_ds;
2647 2651 };
2648 2652
2649 2653 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2650 2654 static boolean_t snaplist_unstable(list_t *l);
2651 2655
2652 2656 static int
2653 2657 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2654 2658 {
2655 2659 dsl_dataset_t *hds = arg1;
2656 2660 struct promotearg *pa = arg2;
2657 2661 struct promotenode *snap = list_head(&pa->shared_snaps);
2658 2662 dsl_dataset_t *origin_ds = snap->ds;
2659 2663 int err;
2660 2664 uint64_t unused;
2661 2665
2662 2666 /* Check that it is a real clone */
2663 2667 if (!dsl_dir_is_clone(hds->ds_dir))
2664 2668 return (EINVAL);
2665 2669
2666 2670 /* Since this is so expensive, don't do the preliminary check */
2667 2671 if (!dmu_tx_is_syncing(tx))
2668 2672 return (0);
2669 2673
2670 2674 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2671 2675 return (EXDEV);
2672 2676
2673 2677 /* compute origin's new unique space */
2674 2678 snap = list_tail(&pa->clone_snaps);
2675 2679 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2676 2680 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2677 2681 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2678 2682 &pa->unique, &unused, &unused);
2679 2683
2680 2684 /*
2681 2685 * Walk the snapshots that we are moving
2682 2686 *
2683 2687 * Compute space to transfer. Consider the incremental changes
2684 2688 * to used for each snapshot:
2685 2689 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2686 2690 * So each snapshot gave birth to:
2687 2691 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2688 2692 * So a sequence would look like:
2689 2693 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2690 2694 * Which simplifies to:
2691 2695 * uN + kN + kN-1 + ... + k1 + k0
2692 2696 * Note however, if we stop before we reach the ORIGIN we get:
2693 2697 * uN + kN + kN-1 + ... + kM - uM-1
2694 2698 */
2695 2699 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2696 2700 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2697 2701 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2698 2702 for (snap = list_head(&pa->shared_snaps); snap;
2699 2703 snap = list_next(&pa->shared_snaps, snap)) {
2700 2704 uint64_t val, dlused, dlcomp, dluncomp;
2701 2705 dsl_dataset_t *ds = snap->ds;
2702 2706
2703 2707 /* Check that the snapshot name does not conflict */
2704 2708 VERIFY(0 == dsl_dataset_get_snapname(ds));
2705 2709 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2706 2710 if (err == 0) {
2707 2711 err = EEXIST;
2708 2712 goto out;
2709 2713 }
2710 2714 if (err != ENOENT)
2711 2715 goto out;
2712 2716
2713 2717 /* The very first snapshot does not have a deadlist */
2714 2718 if (ds->ds_phys->ds_prev_snap_obj == 0)
2715 2719 continue;
2716 2720
2717 2721 dsl_deadlist_space(&ds->ds_deadlist,
2718 2722 &dlused, &dlcomp, &dluncomp);
2719 2723 pa->used += dlused;
2720 2724 pa->comp += dlcomp;
2721 2725 pa->uncomp += dluncomp;
2722 2726 }
2723 2727
2724 2728 /*
2725 2729 * If we are a clone of a clone then we never reached ORIGIN,
2726 2730 * so we need to subtract out the clone origin's used space.
2727 2731 */
2728 2732 if (pa->origin_origin) {
2729 2733 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2730 2734 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2731 2735 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2732 2736 }
2733 2737
2734 2738 /* Check that there is enough space here */
2735 2739 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2736 2740 pa->used);
2737 2741 if (err)
2738 2742 return (err);
2739 2743
2740 2744 /*
2741 2745 * Compute the amounts of space that will be used by snapshots
2742 2746 * after the promotion (for both origin and clone). For each,
2743 2747 * it is the amount of space that will be on all of their
2744 2748 * deadlists (that was not born before their new origin).
2745 2749 */
2746 2750 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2747 2751 uint64_t space;
2748 2752
2749 2753 /*
2750 2754 * Note, typically this will not be a clone of a clone,
2751 2755 * so dd_origin_txg will be < TXG_INITIAL, so
2752 2756 * these snaplist_space() -> dsl_deadlist_space_range()
2753 2757 * calls will be fast because they do not have to
2754 2758 * iterate over all bps.
2755 2759 */
2756 2760 snap = list_head(&pa->origin_snaps);
2757 2761 err = snaplist_space(&pa->shared_snaps,
2758 2762 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2759 2763 if (err)
2760 2764 return (err);
2761 2765
2762 2766 err = snaplist_space(&pa->clone_snaps,
2763 2767 snap->ds->ds_dir->dd_origin_txg, &space);
2764 2768 if (err)
2765 2769 return (err);
2766 2770 pa->cloneusedsnap += space;
2767 2771 }
2768 2772 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2769 2773 err = snaplist_space(&pa->origin_snaps,
2770 2774 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2771 2775 if (err)
2772 2776 return (err);
2773 2777 }
2774 2778
2775 2779 return (0);
2776 2780 out:
2777 2781 pa->err_ds = snap->ds->ds_snapname;
2778 2782 return (err);
2779 2783 }
2780 2784
2781 2785 static void
2782 2786 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2783 2787 {
2784 2788 dsl_dataset_t *hds = arg1;
2785 2789 struct promotearg *pa = arg2;
2786 2790 struct promotenode *snap = list_head(&pa->shared_snaps);
2787 2791 dsl_dataset_t *origin_ds = snap->ds;
2788 2792 dsl_dataset_t *origin_head;
2789 2793 dsl_dir_t *dd = hds->ds_dir;
2790 2794 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2791 2795 dsl_dir_t *odd = NULL;
2792 2796 uint64_t oldnext_obj;
2793 2797 int64_t delta;
2794 2798
2795 2799 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2796 2800
2797 2801 snap = list_head(&pa->origin_snaps);
2798 2802 origin_head = snap->ds;
2799 2803
2800 2804 /*
2801 2805 * We need to explicitly open odd, since origin_ds's dd will be
2802 2806 * changing.
2803 2807 */
2804 2808 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2805 2809 NULL, FTAG, &odd));
2806 2810
2807 2811 /* change origin's next snap */
2808 2812 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2809 2813 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2810 2814 snap = list_tail(&pa->clone_snaps);
2811 2815 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2812 2816 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2813 2817
2814 2818 /* change the origin's next clone */
2815 2819 if (origin_ds->ds_phys->ds_next_clones_obj) {
2816 2820 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2817 2821 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2818 2822 origin_ds->ds_phys->ds_next_clones_obj,
2819 2823 oldnext_obj, tx));
2820 2824 }
2821 2825
2822 2826 /* change origin */
2823 2827 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2824 2828 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2825 2829 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2826 2830 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2827 2831 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2828 2832 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2829 2833 origin_head->ds_dir->dd_origin_txg =
2830 2834 origin_ds->ds_phys->ds_creation_txg;
2831 2835
2832 2836 /* change dd_clone entries */
2833 2837 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2834 2838 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2835 2839 odd->dd_phys->dd_clones, hds->ds_object, tx));
2836 2840 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2837 2841 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2838 2842 hds->ds_object, tx));
2839 2843
2840 2844 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2841 2845 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2842 2846 origin_head->ds_object, tx));
2843 2847 if (dd->dd_phys->dd_clones == 0) {
2844 2848 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2845 2849 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2846 2850 }
2847 2851 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2848 2852 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2849 2853
2850 2854 }
2851 2855
2852 2856 /* move snapshots to this dir */
2853 2857 for (snap = list_head(&pa->shared_snaps); snap;
2854 2858 snap = list_next(&pa->shared_snaps, snap)) {
2855 2859 dsl_dataset_t *ds = snap->ds;
2856 2860
2857 2861 /* unregister props as dsl_dir is changing */
2858 2862 if (ds->ds_objset) {
2859 2863 dmu_objset_evict(ds->ds_objset);
2860 2864 ds->ds_objset = NULL;
2861 2865 }
2862 2866 /* move snap name entry */
2863 2867 VERIFY(0 == dsl_dataset_get_snapname(ds));
2864 2868 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2865 2869 ds->ds_snapname, tx));
2866 2870 VERIFY(0 == zap_add(dp->dp_meta_objset,
2867 2871 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2868 2872 8, 1, &ds->ds_object, tx));
2869 2873
2870 2874 /* change containing dsl_dir */
2871 2875 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2872 2876 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2873 2877 ds->ds_phys->ds_dir_obj = dd->dd_object;
2874 2878 ASSERT3P(ds->ds_dir, ==, odd);
2875 2879 dsl_dir_close(ds->ds_dir, ds);
2876 2880 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2877 2881 NULL, ds, &ds->ds_dir));
2878 2882
2879 2883 /* move any clone references */
2880 2884 if (ds->ds_phys->ds_next_clones_obj &&
2881 2885 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2882 2886 zap_cursor_t zc;
2883 2887 zap_attribute_t za;
2884 2888
2885 2889 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2886 2890 ds->ds_phys->ds_next_clones_obj);
2887 2891 zap_cursor_retrieve(&zc, &za) == 0;
2888 2892 zap_cursor_advance(&zc)) {
2889 2893 dsl_dataset_t *cnds;
2890 2894 uint64_t o;
2891 2895
2892 2896 if (za.za_first_integer == oldnext_obj) {
2893 2897 /*
2894 2898 * We've already moved the
2895 2899 * origin's reference.
2896 2900 */
2897 2901 continue;
2898 2902 }
2899 2903
2900 2904 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2901 2905 za.za_first_integer, FTAG, &cnds));
2902 2906 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2903 2907
2904 2908 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2905 2909 odd->dd_phys->dd_clones, o, tx), ==, 0);
2906 2910 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2907 2911 dd->dd_phys->dd_clones, o, tx), ==, 0);
2908 2912 dsl_dataset_rele(cnds, FTAG);
2909 2913 }
2910 2914 zap_cursor_fini(&zc);
2911 2915 }
2912 2916
2913 2917 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2914 2918 }
2915 2919
2916 2920 /*
2917 2921 * Change space accounting.
2918 2922 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2919 2923 * both be valid, or both be 0 (resulting in delta == 0). This
2920 2924 * is true for each of {clone,origin} independently.
2921 2925 */
2922 2926
2923 2927 delta = pa->cloneusedsnap -
2924 2928 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2925 2929 ASSERT3S(delta, >=, 0);
2926 2930 ASSERT3U(pa->used, >=, delta);
2927 2931 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2928 2932 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2929 2933 pa->used - delta, pa->comp, pa->uncomp, tx);
2930 2934
2931 2935 delta = pa->originusedsnap -
2932 2936 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2933 2937 ASSERT3S(delta, <=, 0);
2934 2938 ASSERT3U(pa->used, >=, -delta);
2935 2939 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2936 2940 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2937 2941 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2938 2942
2939 2943 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2940 2944
2941 2945 /* log history record */
2942 2946 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2943 2947 "dataset = %llu", hds->ds_object);
2944 2948
2945 2949 dsl_dir_close(odd, FTAG);
2946 2950 }
2947 2951
2948 2952 static char *snaplist_tag = "snaplist";
2949 2953 /*
2950 2954 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2951 2955 * (exclusive) and last_obj (inclusive). The list will be in reverse
2952 2956 * order (last_obj will be the list_head()). If first_obj == 0, do all
2953 2957 * snapshots back to this dataset's origin.
2954 2958 */
2955 2959 static int
2956 2960 snaplist_make(dsl_pool_t *dp, boolean_t own,
2957 2961 uint64_t first_obj, uint64_t last_obj, list_t *l)
2958 2962 {
2959 2963 uint64_t obj = last_obj;
2960 2964
2961 2965 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2962 2966
2963 2967 list_create(l, sizeof (struct promotenode),
2964 2968 offsetof(struct promotenode, link));
2965 2969
2966 2970 while (obj != first_obj) {
2967 2971 dsl_dataset_t *ds;
2968 2972 struct promotenode *snap;
2969 2973 int err;
2970 2974
2971 2975 if (own) {
2972 2976 err = dsl_dataset_own_obj(dp, obj,
2973 2977 0, snaplist_tag, &ds);
2974 2978 if (err == 0)
2975 2979 dsl_dataset_make_exclusive(ds, snaplist_tag);
2976 2980 } else {
2977 2981 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2978 2982 }
2979 2983 if (err == ENOENT) {
2980 2984 /* lost race with snapshot destroy */
2981 2985 struct promotenode *last = list_tail(l);
2982 2986 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2983 2987 obj = last->ds->ds_phys->ds_prev_snap_obj;
2984 2988 continue;
2985 2989 } else if (err) {
2986 2990 return (err);
2987 2991 }
2988 2992
2989 2993 if (first_obj == 0)
2990 2994 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2991 2995
2992 2996 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2993 2997 snap->ds = ds;
2994 2998 list_insert_tail(l, snap);
2995 2999 obj = ds->ds_phys->ds_prev_snap_obj;
2996 3000 }
2997 3001
2998 3002 return (0);
2999 3003 }
3000 3004
3001 3005 static int
3002 3006 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3003 3007 {
3004 3008 struct promotenode *snap;
3005 3009
3006 3010 *spacep = 0;
3007 3011 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3008 3012 uint64_t used, comp, uncomp;
3009 3013 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3010 3014 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3011 3015 *spacep += used;
3012 3016 }
3013 3017 return (0);
3014 3018 }
3015 3019
3016 3020 static void
3017 3021 snaplist_destroy(list_t *l, boolean_t own)
3018 3022 {
3019 3023 struct promotenode *snap;
3020 3024
3021 3025 if (!l || !list_link_active(&l->list_head))
3022 3026 return;
3023 3027
3024 3028 while ((snap = list_tail(l)) != NULL) {
3025 3029 list_remove(l, snap);
3026 3030 if (own)
3027 3031 dsl_dataset_disown(snap->ds, snaplist_tag);
3028 3032 else
3029 3033 dsl_dataset_rele(snap->ds, snaplist_tag);
3030 3034 kmem_free(snap, sizeof (struct promotenode));
3031 3035 }
3032 3036 list_destroy(l);
3033 3037 }
3034 3038
3035 3039 /*
3036 3040 * Promote a clone. Nomenclature note:
3037 3041 * "clone" or "cds": the original clone which is being promoted
3038 3042 * "origin" or "ods": the snapshot which is originally clone's origin
3039 3043 * "origin head" or "ohds": the dataset which is the head
3040 3044 * (filesystem/volume) for the origin
3041 3045 * "origin origin": the origin of the origin's filesystem (typically
3042 3046 * NULL, indicating that the clone is not a clone of a clone).
3043 3047 */
3044 3048 int
3045 3049 dsl_dataset_promote(const char *name, char *conflsnap)
3046 3050 {
3047 3051 dsl_dataset_t *ds;
3048 3052 dsl_dir_t *dd;
3049 3053 dsl_pool_t *dp;
3050 3054 dmu_object_info_t doi;
3051 3055 struct promotearg pa = { 0 };
3052 3056 struct promotenode *snap;
3053 3057 int err;
3054 3058
3055 3059 err = dsl_dataset_hold(name, FTAG, &ds);
3056 3060 if (err)
3057 3061 return (err);
3058 3062 dd = ds->ds_dir;
3059 3063 dp = dd->dd_pool;
3060 3064
3061 3065 err = dmu_object_info(dp->dp_meta_objset,
3062 3066 ds->ds_phys->ds_snapnames_zapobj, &doi);
3063 3067 if (err) {
3064 3068 dsl_dataset_rele(ds, FTAG);
3065 3069 return (err);
3066 3070 }
3067 3071
3068 3072 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3069 3073 dsl_dataset_rele(ds, FTAG);
3070 3074 return (EINVAL);
3071 3075 }
3072 3076
3073 3077 /*
3074 3078 * We are going to inherit all the snapshots taken before our
3075 3079 * origin (i.e., our new origin will be our parent's origin).
3076 3080 * Take ownership of them so that we can rename them into our
3077 3081 * namespace.
3078 3082 */
3079 3083 rw_enter(&dp->dp_config_rwlock, RW_READER);
3080 3084
3081 3085 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3082 3086 &pa.shared_snaps);
3083 3087 if (err != 0)
3084 3088 goto out;
3085 3089
3086 3090 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3087 3091 if (err != 0)
3088 3092 goto out;
3089 3093
3090 3094 snap = list_head(&pa.shared_snaps);
3091 3095 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3092 3096 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3093 3097 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3094 3098 if (err != 0)
3095 3099 goto out;
3096 3100
3097 3101 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3098 3102 err = dsl_dataset_hold_obj(dp,
3099 3103 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3100 3104 FTAG, &pa.origin_origin);
3101 3105 if (err != 0)
3102 3106 goto out;
3103 3107 }
3104 3108
3105 3109 out:
3106 3110 rw_exit(&dp->dp_config_rwlock);
3107 3111
3108 3112 /*
3109 3113 * Add in 128x the snapnames zapobj size, since we will be moving
3110 3114 * a bunch of snapnames to the promoted ds, and dirtying their
3111 3115 * bonus buffers.
3112 3116 */
3113 3117 if (err == 0) {
3114 3118 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3115 3119 dsl_dataset_promote_sync, ds, &pa,
3116 3120 2 + 2 * doi.doi_physical_blocks_512);
3117 3121 if (err && pa.err_ds && conflsnap)
3118 3122 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3119 3123 }
3120 3124
3121 3125 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3122 3126 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3123 3127 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3124 3128 if (pa.origin_origin)
3125 3129 dsl_dataset_rele(pa.origin_origin, FTAG);
3126 3130 dsl_dataset_rele(ds, FTAG);
3127 3131 return (err);
3128 3132 }
3129 3133
3130 3134 struct cloneswaparg {
3131 3135 dsl_dataset_t *cds; /* clone dataset */
3132 3136 dsl_dataset_t *ohds; /* origin's head dataset */
3133 3137 boolean_t force;
3134 3138 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3135 3139 };
3136 3140
3137 3141 /* ARGSUSED */
3138 3142 static int
3139 3143 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3140 3144 {
3141 3145 struct cloneswaparg *csa = arg1;
3142 3146
3143 3147 /* they should both be heads */
3144 3148 if (dsl_dataset_is_snapshot(csa->cds) ||
3145 3149 dsl_dataset_is_snapshot(csa->ohds))
3146 3150 return (EINVAL);
3147 3151
3148 3152 /* the branch point should be just before them */
3149 3153 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3150 3154 return (EINVAL);
3151 3155
3152 3156 /* cds should be the clone (unless they are unrelated) */
3153 3157 if (csa->cds->ds_prev != NULL &&
3154 3158 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3155 3159 csa->ohds->ds_object !=
3156 3160 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3157 3161 return (EINVAL);
3158 3162
3159 3163 /* the clone should be a child of the origin */
3160 3164 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3161 3165 return (EINVAL);
3162 3166
3163 3167 /* ohds shouldn't be modified unless 'force' */
3164 3168 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3165 3169 return (ETXTBSY);
3166 3170
3167 3171 /* adjust amount of any unconsumed refreservation */
3168 3172 csa->unused_refres_delta =
3169 3173 (int64_t)MIN(csa->ohds->ds_reserved,
3170 3174 csa->ohds->ds_phys->ds_unique_bytes) -
3171 3175 (int64_t)MIN(csa->ohds->ds_reserved,
3172 3176 csa->cds->ds_phys->ds_unique_bytes);
3173 3177
3174 3178 if (csa->unused_refres_delta > 0 &&
3175 3179 csa->unused_refres_delta >
3176 3180 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3177 3181 return (ENOSPC);
3178 3182
3179 3183 if (csa->ohds->ds_quota != 0 &&
3180 3184 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3181 3185 return (EDQUOT);
3182 3186
3183 3187 return (0);
3184 3188 }
3185 3189
3186 3190 /* ARGSUSED */
3187 3191 static void
3188 3192 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3189 3193 {
3190 3194 struct cloneswaparg *csa = arg1;
3191 3195 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3192 3196
3193 3197 ASSERT(csa->cds->ds_reserved == 0);
3194 3198 ASSERT(csa->ohds->ds_quota == 0 ||
3195 3199 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3196 3200
3197 3201 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3198 3202 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3199 3203
3200 3204 if (csa->cds->ds_objset != NULL) {
3201 3205 dmu_objset_evict(csa->cds->ds_objset);
3202 3206 csa->cds->ds_objset = NULL;
3203 3207 }
3204 3208
3205 3209 if (csa->ohds->ds_objset != NULL) {
3206 3210 dmu_objset_evict(csa->ohds->ds_objset);
3207 3211 csa->ohds->ds_objset = NULL;
3208 3212 }
3209 3213
3210 3214 /*
3211 3215 * Reset origin's unique bytes, if it exists.
3212 3216 */
3213 3217 if (csa->cds->ds_prev) {
3214 3218 dsl_dataset_t *origin = csa->cds->ds_prev;
3215 3219 uint64_t comp, uncomp;
3216 3220
3217 3221 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3218 3222 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3219 3223 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3220 3224 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3221 3225 }
3222 3226
3223 3227 /* swap blkptrs */
3224 3228 {
3225 3229 blkptr_t tmp;
3226 3230 tmp = csa->ohds->ds_phys->ds_bp;
3227 3231 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3228 3232 csa->cds->ds_phys->ds_bp = tmp;
3229 3233 }
3230 3234
3231 3235 /* set dd_*_bytes */
3232 3236 {
3233 3237 int64_t dused, dcomp, duncomp;
3234 3238 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3235 3239 uint64_t odl_used, odl_comp, odl_uncomp;
3236 3240
3237 3241 ASSERT3U(csa->cds->ds_dir->dd_phys->
3238 3242 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3239 3243
3240 3244 dsl_deadlist_space(&csa->cds->ds_deadlist,
3241 3245 &cdl_used, &cdl_comp, &cdl_uncomp);
3242 3246 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3243 3247 &odl_used, &odl_comp, &odl_uncomp);
3244 3248
3245 3249 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3246 3250 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3247 3251 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3248 3252 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3249 3253 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3250 3254 cdl_uncomp -
3251 3255 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3252 3256
3253 3257 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3254 3258 dused, dcomp, duncomp, tx);
3255 3259 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3256 3260 -dused, -dcomp, -duncomp, tx);
3257 3261
3258 3262 /*
3259 3263 * The difference in the space used by snapshots is the
3260 3264 * difference in snapshot space due to the head's
3261 3265 * deadlist (since that's the only thing that's
3262 3266 * changing that affects the snapused).
3263 3267 */
3264 3268 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3265 3269 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3266 3270 &cdl_used, &cdl_comp, &cdl_uncomp);
3267 3271 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3268 3272 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3269 3273 &odl_used, &odl_comp, &odl_uncomp);
3270 3274 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3271 3275 DD_USED_HEAD, DD_USED_SNAP, tx);
3272 3276 }
3273 3277
3274 3278 /* swap ds_*_bytes */
3275 3279 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3276 3280 csa->cds->ds_phys->ds_referenced_bytes);
3277 3281 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3278 3282 csa->cds->ds_phys->ds_compressed_bytes);
3279 3283 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3280 3284 csa->cds->ds_phys->ds_uncompressed_bytes);
3281 3285 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3282 3286 csa->cds->ds_phys->ds_unique_bytes);
3283 3287
3284 3288 /* apply any parent delta for change in unconsumed refreservation */
3285 3289 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3286 3290 csa->unused_refres_delta, 0, 0, tx);
3287 3291
3288 3292 /*
3289 3293 * Swap deadlists.
3290 3294 */
3291 3295 dsl_deadlist_close(&csa->cds->ds_deadlist);
3292 3296 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3293 3297 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3294 3298 csa->cds->ds_phys->ds_deadlist_obj);
3295 3299 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3296 3300 csa->cds->ds_phys->ds_deadlist_obj);
3297 3301 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3298 3302 csa->ohds->ds_phys->ds_deadlist_obj);
3299 3303
3300 3304 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3301 3305 }
3302 3306
3303 3307 /*
3304 3308 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3305 3309 * recv" into an existing fs to swizzle the file system to the new
3306 3310 * version, and by "zfs rollback". Can also be used to swap two
3307 3311 * independent head datasets if neither has any snapshots.
3308 3312 */
3309 3313 int
3310 3314 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3311 3315 boolean_t force)
3312 3316 {
3313 3317 struct cloneswaparg csa;
3314 3318 int error;
3315 3319
3316 3320 ASSERT(clone->ds_owner);
3317 3321 ASSERT(origin_head->ds_owner);
3318 3322 retry:
3319 3323 /*
3320 3324 * Need exclusive access for the swap. If we're swapping these
3321 3325 * datasets back after an error, we already hold the locks.
3322 3326 */
3323 3327 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3324 3328 rw_enter(&clone->ds_rwlock, RW_WRITER);
3325 3329 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3326 3330 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3327 3331 rw_exit(&clone->ds_rwlock);
3328 3332 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3329 3333 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3330 3334 rw_exit(&origin_head->ds_rwlock);
3331 3335 goto retry;
3332 3336 }
3333 3337 }
3334 3338 csa.cds = clone;
3335 3339 csa.ohds = origin_head;
3336 3340 csa.force = force;
3337 3341 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3338 3342 dsl_dataset_clone_swap_check,
3339 3343 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3340 3344 return (error);
3341 3345 }
3342 3346
3343 3347 /*
3344 3348 * Given a pool name and a dataset object number in that pool,
3345 3349 * return the name of that dataset.
3346 3350 */
3347 3351 int
3348 3352 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3349 3353 {
3350 3354 spa_t *spa;
3351 3355 dsl_pool_t *dp;
3352 3356 dsl_dataset_t *ds;
3353 3357 int error;
3354 3358
3355 3359 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3356 3360 return (error);
3357 3361 dp = spa_get_dsl(spa);
3358 3362 rw_enter(&dp->dp_config_rwlock, RW_READER);
3359 3363 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3360 3364 dsl_dataset_name(ds, buf);
3361 3365 dsl_dataset_rele(ds, FTAG);
3362 3366 }
3363 3367 rw_exit(&dp->dp_config_rwlock);
3364 3368 spa_close(spa, FTAG);
3365 3369
3366 3370 return (error);
3367 3371 }
3368 3372
3369 3373 int
3370 3374 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3371 3375 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3372 3376 {
3373 3377 int error = 0;
3374 3378
3375 3379 ASSERT3S(asize, >, 0);
3376 3380
3377 3381 /*
3378 3382 * *ref_rsrv is the portion of asize that will come from any
3379 3383 * unconsumed refreservation space.
3380 3384 */
3381 3385 *ref_rsrv = 0;
3382 3386
3383 3387 mutex_enter(&ds->ds_lock);
3384 3388 /*
3385 3389 * Make a space adjustment for reserved bytes.
3386 3390 */
3387 3391 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3388 3392 ASSERT3U(*used, >=,
3389 3393 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3390 3394 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3391 3395 *ref_rsrv =
3392 3396 asize - MIN(asize, parent_delta(ds, asize + inflight));
3393 3397 }
3394 3398
3395 3399 if (!check_quota || ds->ds_quota == 0) {
3396 3400 mutex_exit(&ds->ds_lock);
3397 3401 return (0);
3398 3402 }
3399 3403 /*
3400 3404 * If they are requesting more space, and our current estimate
3401 3405 * is over quota, they get to try again unless the actual
3402 3406 * on-disk is over quota and there are no pending changes (which
3403 3407 * may free up space for us).
3404 3408 */
3405 3409 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3406 3410 if (inflight > 0 ||
3407 3411 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3408 3412 error = ERESTART;
3409 3413 else
3410 3414 error = EDQUOT;
3411 3415 }
3412 3416 mutex_exit(&ds->ds_lock);
3413 3417
3414 3418 return (error);
3415 3419 }
3416 3420
3417 3421 /* ARGSUSED */
3418 3422 static int
3419 3423 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3420 3424 {
3421 3425 dsl_dataset_t *ds = arg1;
3422 3426 dsl_prop_setarg_t *psa = arg2;
3423 3427 int err;
3424 3428
3425 3429 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3426 3430 return (ENOTSUP);
3427 3431
3428 3432 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3429 3433 return (err);
3430 3434
3431 3435 if (psa->psa_effective_value == 0)
3432 3436 return (0);
3433 3437
3434 3438 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3435 3439 psa->psa_effective_value < ds->ds_reserved)
3436 3440 return (ENOSPC);
3437 3441
3438 3442 return (0);
3439 3443 }
3440 3444
3441 3445 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3442 3446
3443 3447 void
3444 3448 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3445 3449 {
3446 3450 dsl_dataset_t *ds = arg1;
3447 3451 dsl_prop_setarg_t *psa = arg2;
3448 3452 uint64_t effective_value = psa->psa_effective_value;
3449 3453
3450 3454 dsl_prop_set_sync(ds, psa, tx);
3451 3455 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3452 3456
3453 3457 if (ds->ds_quota != effective_value) {
3454 3458 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3455 3459 ds->ds_quota = effective_value;
3456 3460
3457 3461 spa_history_log_internal(LOG_DS_REFQUOTA,
3458 3462 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3459 3463 (longlong_t)ds->ds_quota, ds->ds_object);
3460 3464 }
3461 3465 }
3462 3466
3463 3467 int
3464 3468 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3465 3469 {
3466 3470 dsl_dataset_t *ds;
3467 3471 dsl_prop_setarg_t psa;
3468 3472 int err;
3469 3473
3470 3474 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3471 3475
3472 3476 err = dsl_dataset_hold(dsname, FTAG, &ds);
3473 3477 if (err)
3474 3478 return (err);
3475 3479
3476 3480 /*
3477 3481 * If someone removes a file, then tries to set the quota, we
3478 3482 * want to make sure the file freeing takes effect.
3479 3483 */
3480 3484 txg_wait_open(ds->ds_dir->dd_pool, 0);
3481 3485
3482 3486 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3483 3487 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3484 3488 ds, &psa, 0);
3485 3489
3486 3490 dsl_dataset_rele(ds, FTAG);
3487 3491 return (err);
3488 3492 }
3489 3493
3490 3494 static int
3491 3495 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3492 3496 {
3493 3497 dsl_dataset_t *ds = arg1;
3494 3498 dsl_prop_setarg_t *psa = arg2;
3495 3499 uint64_t effective_value;
3496 3500 uint64_t unique;
3497 3501 int err;
3498 3502
3499 3503 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3500 3504 SPA_VERSION_REFRESERVATION)
3501 3505 return (ENOTSUP);
3502 3506
3503 3507 if (dsl_dataset_is_snapshot(ds))
3504 3508 return (EINVAL);
3505 3509
3506 3510 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3507 3511 return (err);
3508 3512
3509 3513 effective_value = psa->psa_effective_value;
3510 3514
3511 3515 /*
3512 3516 * If we are doing the preliminary check in open context, the
3513 3517 * space estimates may be inaccurate.
3514 3518 */
3515 3519 if (!dmu_tx_is_syncing(tx))
3516 3520 return (0);
3517 3521
3518 3522 mutex_enter(&ds->ds_lock);
3519 3523 if (!DS_UNIQUE_IS_ACCURATE(ds))
3520 3524 dsl_dataset_recalc_head_uniq(ds);
3521 3525 unique = ds->ds_phys->ds_unique_bytes;
3522 3526 mutex_exit(&ds->ds_lock);
3523 3527
3524 3528 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3525 3529 uint64_t delta = MAX(unique, effective_value) -
3526 3530 MAX(unique, ds->ds_reserved);
3527 3531
3528 3532 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3529 3533 return (ENOSPC);
3530 3534 if (ds->ds_quota > 0 &&
3531 3535 effective_value > ds->ds_quota)
3532 3536 return (ENOSPC);
3533 3537 }
3534 3538
3535 3539 return (0);
3536 3540 }
3537 3541
3538 3542 static void
3539 3543 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3540 3544 {
3541 3545 dsl_dataset_t *ds = arg1;
3542 3546 dsl_prop_setarg_t *psa = arg2;
3543 3547 uint64_t effective_value = psa->psa_effective_value;
3544 3548 uint64_t unique;
3545 3549 int64_t delta;
3546 3550
3547 3551 dsl_prop_set_sync(ds, psa, tx);
3548 3552 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3549 3553
3550 3554 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3551 3555
3552 3556 mutex_enter(&ds->ds_dir->dd_lock);
3553 3557 mutex_enter(&ds->ds_lock);
3554 3558 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3555 3559 unique = ds->ds_phys->ds_unique_bytes;
3556 3560 delta = MAX(0, (int64_t)(effective_value - unique)) -
3557 3561 MAX(0, (int64_t)(ds->ds_reserved - unique));
3558 3562 ds->ds_reserved = effective_value;
3559 3563 mutex_exit(&ds->ds_lock);
3560 3564
3561 3565 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3562 3566 mutex_exit(&ds->ds_dir->dd_lock);
3563 3567
3564 3568 spa_history_log_internal(LOG_DS_REFRESERV,
3565 3569 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3566 3570 (longlong_t)effective_value, ds->ds_object);
3567 3571 }
3568 3572
3569 3573 int
3570 3574 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3571 3575 uint64_t reservation)
3572 3576 {
3573 3577 dsl_dataset_t *ds;
3574 3578 dsl_prop_setarg_t psa;
3575 3579 int err;
3576 3580
3577 3581 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3578 3582 &reservation);
3579 3583
3580 3584 err = dsl_dataset_hold(dsname, FTAG, &ds);
3581 3585 if (err)
3582 3586 return (err);
3583 3587
3584 3588 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3585 3589 dsl_dataset_set_reservation_check,
3586 3590 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3587 3591
3588 3592 dsl_dataset_rele(ds, FTAG);
3589 3593 return (err);
3590 3594 }
3591 3595
3592 3596 typedef struct zfs_hold_cleanup_arg {
3593 3597 dsl_pool_t *dp;
3594 3598 uint64_t dsobj;
3595 3599 char htag[MAXNAMELEN];
3596 3600 } zfs_hold_cleanup_arg_t;
3597 3601
3598 3602 static void
3599 3603 dsl_dataset_user_release_onexit(void *arg)
3600 3604 {
3601 3605 zfs_hold_cleanup_arg_t *ca = arg;
3602 3606
3603 3607 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3604 3608 B_TRUE);
3605 3609 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3606 3610 }
3607 3611
3608 3612 void
3609 3613 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3610 3614 minor_t minor)
3611 3615 {
3612 3616 zfs_hold_cleanup_arg_t *ca;
3613 3617
3614 3618 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3615 3619 ca->dp = ds->ds_dir->dd_pool;
3616 3620 ca->dsobj = ds->ds_object;
3617 3621 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3618 3622 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3619 3623 dsl_dataset_user_release_onexit, ca, NULL));
3620 3624 }
3621 3625
3622 3626 /*
3623 3627 * If you add new checks here, you may need to add
3624 3628 * additional checks to the "temporary" case in
3625 3629 * snapshot_check() in dmu_objset.c.
3626 3630 */
3627 3631 static int
3628 3632 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3629 3633 {
3630 3634 dsl_dataset_t *ds = arg1;
3631 3635 struct dsl_ds_holdarg *ha = arg2;
3632 3636 char *htag = ha->htag;
3633 3637 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3634 3638 int error = 0;
3635 3639
3636 3640 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3637 3641 return (ENOTSUP);
3638 3642
3639 3643 if (!dsl_dataset_is_snapshot(ds))
3640 3644 return (EINVAL);
3641 3645
3642 3646 /* tags must be unique */
3643 3647 mutex_enter(&ds->ds_lock);
3644 3648 if (ds->ds_phys->ds_userrefs_obj) {
3645 3649 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3646 3650 8, 1, tx);
3647 3651 if (error == 0)
3648 3652 error = EEXIST;
3649 3653 else if (error == ENOENT)
3650 3654 error = 0;
3651 3655 }
3652 3656 mutex_exit(&ds->ds_lock);
3653 3657
3654 3658 if (error == 0 && ha->temphold &&
3655 3659 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3656 3660 error = E2BIG;
3657 3661
3658 3662 return (error);
3659 3663 }
3660 3664
3661 3665 void
3662 3666 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3663 3667 {
3664 3668 dsl_dataset_t *ds = arg1;
3665 3669 struct dsl_ds_holdarg *ha = arg2;
3666 3670 char *htag = ha->htag;
3667 3671 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3668 3672 objset_t *mos = dp->dp_meta_objset;
3669 3673 uint64_t now = gethrestime_sec();
3670 3674 uint64_t zapobj;
3671 3675
3672 3676 mutex_enter(&ds->ds_lock);
3673 3677 if (ds->ds_phys->ds_userrefs_obj == 0) {
3674 3678 /*
3675 3679 * This is the first user hold for this dataset. Create
3676 3680 * the userrefs zap object.
3677 3681 */
3678 3682 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3679 3683 zapobj = ds->ds_phys->ds_userrefs_obj =
3680 3684 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3681 3685 } else {
3682 3686 zapobj = ds->ds_phys->ds_userrefs_obj;
3683 3687 }
3684 3688 ds->ds_userrefs++;
3685 3689 mutex_exit(&ds->ds_lock);
3686 3690
3687 3691 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3688 3692
3689 3693 if (ha->temphold) {
3690 3694 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3691 3695 htag, &now, tx));
3692 3696 }
3693 3697
3694 3698 spa_history_log_internal(LOG_DS_USER_HOLD,
3695 3699 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3696 3700 (int)ha->temphold, ds->ds_object);
3697 3701 }
3698 3702
3699 3703 static int
3700 3704 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3701 3705 {
3702 3706 struct dsl_ds_holdarg *ha = arg;
3703 3707 dsl_dataset_t *ds;
3704 3708 int error;
3705 3709 char *name;
3706 3710
3707 3711 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3708 3712 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3709 3713 error = dsl_dataset_hold(name, ha->dstg, &ds);
3710 3714 strfree(name);
3711 3715 if (error == 0) {
3712 3716 ha->gotone = B_TRUE;
3713 3717 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3714 3718 dsl_dataset_user_hold_sync, ds, ha, 0);
3715 3719 } else if (error == ENOENT && ha->recursive) {
3716 3720 error = 0;
3717 3721 } else {
3718 3722 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3719 3723 }
3720 3724 return (error);
3721 3725 }
3722 3726
3723 3727 int
3724 3728 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3725 3729 boolean_t temphold)
3726 3730 {
3727 3731 struct dsl_ds_holdarg *ha;
3728 3732 int error;
3729 3733
3730 3734 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3731 3735 ha->htag = htag;
3732 3736 ha->temphold = temphold;
3733 3737 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3734 3738 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3735 3739 ds, ha, 0);
3736 3740 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3737 3741
3738 3742 return (error);
3739 3743 }
3740 3744
3741 3745 int
3742 3746 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3743 3747 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3744 3748 {
3745 3749 struct dsl_ds_holdarg *ha;
3746 3750 dsl_sync_task_t *dst;
3747 3751 spa_t *spa;
3748 3752 int error;
3749 3753 minor_t minor = 0;
3750 3754
3751 3755 if (cleanup_fd != -1) {
3752 3756 /* Currently we only support cleanup-on-exit of tempholds. */
3753 3757 if (!temphold)
3754 3758 return (EINVAL);
3755 3759 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3756 3760 if (error)
3757 3761 return (error);
3758 3762 }
3759 3763
3760 3764 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3761 3765
3762 3766 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3763 3767
3764 3768 error = spa_open(dsname, &spa, FTAG);
3765 3769 if (error) {
3766 3770 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3767 3771 if (cleanup_fd != -1)
3768 3772 zfs_onexit_fd_rele(cleanup_fd);
3769 3773 return (error);
3770 3774 }
3771 3775
3772 3776 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3773 3777 ha->htag = htag;
3774 3778 ha->snapname = snapname;
3775 3779 ha->recursive = recursive;
3776 3780 ha->temphold = temphold;
3777 3781
3778 3782 if (recursive) {
3779 3783 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3780 3784 ha, DS_FIND_CHILDREN);
3781 3785 } else {
3782 3786 error = dsl_dataset_user_hold_one(dsname, ha);
3783 3787 }
3784 3788 if (error == 0)
3785 3789 error = dsl_sync_task_group_wait(ha->dstg);
3786 3790
3787 3791 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3788 3792 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3789 3793 dsl_dataset_t *ds = dst->dst_arg1;
3790 3794
3791 3795 if (dst->dst_err) {
3792 3796 dsl_dataset_name(ds, ha->failed);
3793 3797 *strchr(ha->failed, '@') = '\0';
3794 3798 } else if (error == 0 && minor != 0 && temphold) {
3795 3799 /*
3796 3800 * If this hold is to be released upon process exit,
3797 3801 * register that action now.
3798 3802 */
3799 3803 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3800 3804 }
3801 3805 dsl_dataset_rele(ds, ha->dstg);
3802 3806 }
3803 3807
3804 3808 if (error == 0 && recursive && !ha->gotone)
3805 3809 error = ENOENT;
3806 3810
3807 3811 if (error)
3808 3812 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3809 3813
3810 3814 dsl_sync_task_group_destroy(ha->dstg);
3811 3815
3812 3816 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3813 3817 spa_close(spa, FTAG);
3814 3818 if (cleanup_fd != -1)
3815 3819 zfs_onexit_fd_rele(cleanup_fd);
3816 3820 return (error);
3817 3821 }
3818 3822
3819 3823 struct dsl_ds_releasearg {
3820 3824 dsl_dataset_t *ds;
3821 3825 const char *htag;
3822 3826 boolean_t own; /* do we own or just hold ds? */
3823 3827 };
3824 3828
3825 3829 static int
3826 3830 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3827 3831 boolean_t *might_destroy)
3828 3832 {
3829 3833 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3830 3834 uint64_t zapobj;
3831 3835 uint64_t tmp;
3832 3836 int error;
3833 3837
3834 3838 *might_destroy = B_FALSE;
3835 3839
3836 3840 mutex_enter(&ds->ds_lock);
3837 3841 zapobj = ds->ds_phys->ds_userrefs_obj;
3838 3842 if (zapobj == 0) {
3839 3843 /* The tag can't possibly exist */
3840 3844 mutex_exit(&ds->ds_lock);
3841 3845 return (ESRCH);
3842 3846 }
3843 3847
3844 3848 /* Make sure the tag exists */
3845 3849 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3846 3850 if (error) {
3847 3851 mutex_exit(&ds->ds_lock);
3848 3852 if (error == ENOENT)
3849 3853 error = ESRCH;
3850 3854 return (error);
3851 3855 }
3852 3856
3853 3857 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3854 3858 DS_IS_DEFER_DESTROY(ds))
3855 3859 *might_destroy = B_TRUE;
3856 3860
3857 3861 mutex_exit(&ds->ds_lock);
3858 3862 return (0);
3859 3863 }
3860 3864
3861 3865 static int
3862 3866 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3863 3867 {
3864 3868 struct dsl_ds_releasearg *ra = arg1;
3865 3869 dsl_dataset_t *ds = ra->ds;
3866 3870 boolean_t might_destroy;
3867 3871 int error;
3868 3872
3869 3873 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3870 3874 return (ENOTSUP);
3871 3875
3872 3876 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3873 3877 if (error)
3874 3878 return (error);
3875 3879
3876 3880 if (might_destroy) {
3877 3881 struct dsl_ds_destroyarg dsda = {0};
3878 3882
3879 3883 if (dmu_tx_is_syncing(tx)) {
3880 3884 /*
3881 3885 * If we're not prepared to remove the snapshot,
3882 3886 * we can't allow the release to happen right now.
3883 3887 */
3884 3888 if (!ra->own)
3885 3889 return (EBUSY);
3886 3890 }
3887 3891 dsda.ds = ds;
3888 3892 dsda.releasing = B_TRUE;
3889 3893 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3890 3894 }
3891 3895
3892 3896 return (0);
3893 3897 }
3894 3898
3895 3899 static void
3896 3900 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3897 3901 {
3898 3902 struct dsl_ds_releasearg *ra = arg1;
3899 3903 dsl_dataset_t *ds = ra->ds;
3900 3904 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3901 3905 objset_t *mos = dp->dp_meta_objset;
3902 3906 uint64_t zapobj;
3903 3907 uint64_t dsobj = ds->ds_object;
3904 3908 uint64_t refs;
3905 3909 int error;
3906 3910
3907 3911 mutex_enter(&ds->ds_lock);
3908 3912 ds->ds_userrefs--;
3909 3913 refs = ds->ds_userrefs;
3910 3914 mutex_exit(&ds->ds_lock);
3911 3915 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3912 3916 VERIFY(error == 0 || error == ENOENT);
3913 3917 zapobj = ds->ds_phys->ds_userrefs_obj;
3914 3918 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3915 3919 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3916 3920 DS_IS_DEFER_DESTROY(ds)) {
3917 3921 struct dsl_ds_destroyarg dsda = {0};
3918 3922
3919 3923 ASSERT(ra->own);
3920 3924 dsda.ds = ds;
3921 3925 dsda.releasing = B_TRUE;
3922 3926 /* We already did the destroy_check */
3923 3927 dsl_dataset_destroy_sync(&dsda, tag, tx);
3924 3928 }
3925 3929
3926 3930 spa_history_log_internal(LOG_DS_USER_RELEASE,
3927 3931 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3928 3932 ra->htag, (longlong_t)refs, dsobj);
3929 3933 }
3930 3934
3931 3935 static int
3932 3936 dsl_dataset_user_release_one(const char *dsname, void *arg)
3933 3937 {
3934 3938 struct dsl_ds_holdarg *ha = arg;
3935 3939 struct dsl_ds_releasearg *ra;
3936 3940 dsl_dataset_t *ds;
3937 3941 int error;
3938 3942 void *dtag = ha->dstg;
3939 3943 char *name;
3940 3944 boolean_t own = B_FALSE;
3941 3945 boolean_t might_destroy;
3942 3946
3943 3947 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3944 3948 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3945 3949 error = dsl_dataset_hold(name, dtag, &ds);
3946 3950 strfree(name);
3947 3951 if (error == ENOENT && ha->recursive)
3948 3952 return (0);
3949 3953 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3950 3954 if (error)
3951 3955 return (error);
3952 3956
3953 3957 ha->gotone = B_TRUE;
3954 3958
3955 3959 ASSERT(dsl_dataset_is_snapshot(ds));
3956 3960
3957 3961 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3958 3962 if (error) {
3959 3963 dsl_dataset_rele(ds, dtag);
3960 3964 return (error);
3961 3965 }
3962 3966
3963 3967 if (might_destroy) {
3964 3968 #ifdef _KERNEL
3965 3969 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3966 3970 error = zfs_unmount_snap(name, NULL);
3967 3971 strfree(name);
3968 3972 if (error) {
3969 3973 dsl_dataset_rele(ds, dtag);
3970 3974 return (error);
3971 3975 }
3972 3976 #endif
3973 3977 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3974 3978 dsl_dataset_rele(ds, dtag);
3975 3979 return (EBUSY);
3976 3980 } else {
3977 3981 own = B_TRUE;
3978 3982 dsl_dataset_make_exclusive(ds, dtag);
3979 3983 }
3980 3984 }
3981 3985
3982 3986 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3983 3987 ra->ds = ds;
3984 3988 ra->htag = ha->htag;
3985 3989 ra->own = own;
3986 3990 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3987 3991 dsl_dataset_user_release_sync, ra, dtag, 0);
3988 3992
3989 3993 return (0);
3990 3994 }
3991 3995
3992 3996 int
3993 3997 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3994 3998 boolean_t recursive)
3995 3999 {
3996 4000 struct dsl_ds_holdarg *ha;
3997 4001 dsl_sync_task_t *dst;
3998 4002 spa_t *spa;
3999 4003 int error;
4000 4004
4001 4005 top:
4002 4006 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
4003 4007
4004 4008 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4005 4009
4006 4010 error = spa_open(dsname, &spa, FTAG);
4007 4011 if (error) {
4008 4012 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4009 4013 return (error);
4010 4014 }
4011 4015
4012 4016 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4013 4017 ha->htag = htag;
4014 4018 ha->snapname = snapname;
4015 4019 ha->recursive = recursive;
4016 4020 if (recursive) {
4017 4021 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4018 4022 ha, DS_FIND_CHILDREN);
4019 4023 } else {
4020 4024 error = dsl_dataset_user_release_one(dsname, ha);
4021 4025 }
4022 4026 if (error == 0)
4023 4027 error = dsl_sync_task_group_wait(ha->dstg);
4024 4028
4025 4029 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4026 4030 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4027 4031 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4028 4032 dsl_dataset_t *ds = ra->ds;
4029 4033
4030 4034 if (dst->dst_err)
4031 4035 dsl_dataset_name(ds, ha->failed);
4032 4036
4033 4037 if (ra->own)
4034 4038 dsl_dataset_disown(ds, ha->dstg);
4035 4039 else
4036 4040 dsl_dataset_rele(ds, ha->dstg);
4037 4041
4038 4042 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4039 4043 }
4040 4044
4041 4045 if (error == 0 && recursive && !ha->gotone)
4042 4046 error = ENOENT;
4043 4047
4044 4048 if (error && error != EBUSY)
4045 4049 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4046 4050
4047 4051 dsl_sync_task_group_destroy(ha->dstg);
4048 4052 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4049 4053 spa_close(spa, FTAG);
4050 4054
4051 4055 /*
4052 4056 * We can get EBUSY if we were racing with deferred destroy and
4053 4057 * dsl_dataset_user_release_check() hadn't done the necessary
4054 4058 * open context setup. We can also get EBUSY if we're racing
4055 4059 * with destroy and that thread is the ds_owner. Either way
4056 4060 * the busy condition should be transient, and we should retry
4057 4061 * the release operation.
4058 4062 */
4059 4063 if (error == EBUSY)
4060 4064 goto top;
4061 4065
4062 4066 return (error);
4063 4067 }
4064 4068
4065 4069 /*
4066 4070 * Called at spa_load time (with retry == B_FALSE) to release a stale
4067 4071 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4068 4072 */
4069 4073 int
4070 4074 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4071 4075 boolean_t retry)
4072 4076 {
4073 4077 dsl_dataset_t *ds;
4074 4078 char *snap;
4075 4079 char *name;
4076 4080 int namelen;
4077 4081 int error;
4078 4082
4079 4083 do {
4080 4084 rw_enter(&dp->dp_config_rwlock, RW_READER);
4081 4085 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4082 4086 rw_exit(&dp->dp_config_rwlock);
4083 4087 if (error)
4084 4088 return (error);
4085 4089 namelen = dsl_dataset_namelen(ds)+1;
4086 4090 name = kmem_alloc(namelen, KM_SLEEP);
4087 4091 dsl_dataset_name(ds, name);
4088 4092 dsl_dataset_rele(ds, FTAG);
4089 4093
4090 4094 snap = strchr(name, '@');
4091 4095 *snap = '\0';
4092 4096 ++snap;
4093 4097 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4094 4098 kmem_free(name, namelen);
4095 4099
4096 4100 /*
4097 4101 * The object can't have been destroyed because we have a hold,
4098 4102 * but it might have been renamed, resulting in ENOENT. Retry
4099 4103 * if we've been requested to do so.
4100 4104 *
4101 4105 * It would be nice if we could use the dsobj all the way
4102 4106 * through and avoid ENOENT entirely. But we might need to
4103 4107 * unmount the snapshot, and there's currently no way to lookup
4104 4108 * a vfsp using a ZFS object id.
4105 4109 */
4106 4110 } while ((error == ENOENT) && retry);
4107 4111
4108 4112 return (error);
4109 4113 }
4110 4114
4111 4115 int
4112 4116 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4113 4117 {
4114 4118 dsl_dataset_t *ds;
4115 4119 int err;
4116 4120
4117 4121 err = dsl_dataset_hold(dsname, FTAG, &ds);
4118 4122 if (err)
4119 4123 return (err);
4120 4124
4121 4125 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4122 4126 if (ds->ds_phys->ds_userrefs_obj != 0) {
4123 4127 zap_attribute_t *za;
4124 4128 zap_cursor_t zc;
4125 4129
4126 4130 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4127 4131 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4128 4132 ds->ds_phys->ds_userrefs_obj);
4129 4133 zap_cursor_retrieve(&zc, za) == 0;
4130 4134 zap_cursor_advance(&zc)) {
4131 4135 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4132 4136 za->za_first_integer));
4133 4137 }
4134 4138 zap_cursor_fini(&zc);
4135 4139 kmem_free(za, sizeof (zap_attribute_t));
4136 4140 }
4137 4141 dsl_dataset_rele(ds, FTAG);
4138 4142 return (0);
4139 4143 }
4140 4144
4141 4145 /*
4142 4146 * Note, this function is used as the callback for dmu_objset_find(). We
4143 4147 * always return 0 so that we will continue to find and process
4144 4148 * inconsistent datasets, even if we encounter an error trying to
4145 4149 * process one of them.
4146 4150 */
4147 4151 /* ARGSUSED */
4148 4152 int
4149 4153 dsl_destroy_inconsistent(const char *dsname, void *arg)
4150 4154 {
4151 4155 dsl_dataset_t *ds;
4152 4156
4153 4157 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4154 4158 if (DS_IS_INCONSISTENT(ds))
4155 4159 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4156 4160 else
4157 4161 dsl_dataset_disown(ds, FTAG);
4158 4162 }
4159 4163 return (0);
4160 4164 }
4161 4165
4162 4166 /*
4163 4167 * Return (in *usedp) the amount of space written in new that is not
4164 4168 * present in oldsnap. New may be a snapshot or the head. Old must be
4165 4169 * a snapshot before new, in new's filesystem (or its origin). If not then
4166 4170 * fail and return EINVAL.
4167 4171 *
4168 4172 * The written space is calculated by considering two components: First, we
4169 4173 * ignore any freed space, and calculate the written as new's used space
4170 4174 * minus old's used space. Next, we add in the amount of space that was freed
4171 4175 * between the two snapshots, thus reducing new's used space relative to old's.
4172 4176 * Specifically, this is the space that was born before old->ds_creation_txg,
4173 4177 * and freed before new (ie. on new's deadlist or a previous deadlist).
4174 4178 *
4175 4179 * space freed [---------------------]
4176 4180 * snapshots ---O-------O--------O-------O------
4177 4181 * oldsnap new
4178 4182 */
4179 4183 int
4180 4184 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4181 4185 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4182 4186 {
4183 4187 int err = 0;
4184 4188 uint64_t snapobj;
4185 4189 dsl_pool_t *dp = new->ds_dir->dd_pool;
4186 4190
4187 4191 *usedp = 0;
4188 4192 *usedp += new->ds_phys->ds_referenced_bytes;
4189 4193 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4190 4194
4191 4195 *compp = 0;
4192 4196 *compp += new->ds_phys->ds_compressed_bytes;
4193 4197 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4194 4198
4195 4199 *uncompp = 0;
4196 4200 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4197 4201 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4198 4202
4199 4203 rw_enter(&dp->dp_config_rwlock, RW_READER);
4200 4204 snapobj = new->ds_object;
4201 4205 while (snapobj != oldsnap->ds_object) {
4202 4206 dsl_dataset_t *snap;
4203 4207 uint64_t used, comp, uncomp;
4204 4208
4205 4209 if (snapobj == new->ds_object) {
4206 4210 snap = new;
4207 4211 } else {
4208 4212 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4209 4213 if (err != 0)
4210 4214 break;
4211 4215 }
4212 4216
4213 4217 if (snap->ds_phys->ds_prev_snap_txg ==
4214 4218 oldsnap->ds_phys->ds_creation_txg) {
4215 4219 /*
4216 4220 * The blocks in the deadlist can not be born after
4217 4221 * ds_prev_snap_txg, so get the whole deadlist space,
4218 4222 * which is more efficient (especially for old-format
4219 4223 * deadlists). Unfortunately the deadlist code
4220 4224 * doesn't have enough information to make this
4221 4225 * optimization itself.
4222 4226 */
4223 4227 dsl_deadlist_space(&snap->ds_deadlist,
4224 4228 &used, &comp, &uncomp);
4225 4229 } else {
4226 4230 dsl_deadlist_space_range(&snap->ds_deadlist,
4227 4231 0, oldsnap->ds_phys->ds_creation_txg,
4228 4232 &used, &comp, &uncomp);
4229 4233 }
4230 4234 *usedp += used;
4231 4235 *compp += comp;
4232 4236 *uncompp += uncomp;
4233 4237
4234 4238 /*
4235 4239 * If we get to the beginning of the chain of snapshots
4236 4240 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4237 4241 * was not a snapshot of/before new.
4238 4242 */
4239 4243 snapobj = snap->ds_phys->ds_prev_snap_obj;
4240 4244 if (snap != new)
4241 4245 dsl_dataset_rele(snap, FTAG);
4242 4246 if (snapobj == 0) {
4243 4247 err = EINVAL;
4244 4248 break;
4245 4249 }
4246 4250
4247 4251 }
4248 4252 rw_exit(&dp->dp_config_rwlock);
4249 4253 return (err);
4250 4254 }
4251 4255
4252 4256 /*
4253 4257 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4254 4258 * lastsnap, and all snapshots in between are deleted.
4255 4259 *
4256 4260 * blocks that would be freed [---------------------------]
4257 4261 * snapshots ---O-------O--------O-------O--------O
4258 4262 * firstsnap lastsnap
4259 4263 *
4260 4264 * This is the set of blocks that were born after the snap before firstsnap,
4261 4265 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4262 4266 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4263 4267 * We calculate this by iterating over the relevant deadlists (from the snap
4264 4268 * after lastsnap, backward to the snap after firstsnap), summing up the
4265 4269 * space on the deadlist that was born after the snap before firstsnap.
4266 4270 */
4267 4271 int
4268 4272 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4269 4273 dsl_dataset_t *lastsnap,
4270 4274 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4271 4275 {
4272 4276 int err = 0;
4273 4277 uint64_t snapobj;
4274 4278 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4275 4279
4276 4280 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4277 4281 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4278 4282
4279 4283 /*
4280 4284 * Check that the snapshots are in the same dsl_dir, and firstsnap
4281 4285 * is before lastsnap.
4282 4286 */
4283 4287 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4284 4288 firstsnap->ds_phys->ds_creation_txg >
4285 4289 lastsnap->ds_phys->ds_creation_txg)
4286 4290 return (EINVAL);
4287 4291
4288 4292 *usedp = *compp = *uncompp = 0;
4289 4293
4290 4294 rw_enter(&dp->dp_config_rwlock, RW_READER);
4291 4295 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4292 4296 while (snapobj != firstsnap->ds_object) {
4293 4297 dsl_dataset_t *ds;
4294 4298 uint64_t used, comp, uncomp;
4295 4299
4296 4300 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4297 4301 if (err != 0)
4298 4302 break;
4299 4303
4300 4304 dsl_deadlist_space_range(&ds->ds_deadlist,
4301 4305 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4302 4306 &used, &comp, &uncomp);
4303 4307 *usedp += used;
4304 4308 *compp += comp;
4305 4309 *uncompp += uncomp;
4306 4310
4307 4311 snapobj = ds->ds_phys->ds_prev_snap_obj;
4308 4312 ASSERT3U(snapobj, !=, 0);
4309 4313 dsl_dataset_rele(ds, FTAG);
4310 4314 }
4311 4315 rw_exit(&dp->dp_config_rwlock);
4312 4316 return (err);
4313 4317 }
|
↓ open down ↓ |
2215 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX