Print this page
OS-1566 dataset quota for ZFS datasets
| 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 dsl_pool_mos_diduse_space(tx->tx_pool,
107 107 used, compressed, uncompressed);
108 108 return;
109 109 }
110 110 dmu_buf_will_dirty(ds->ds_dbuf, tx);
111 111
112 112 mutex_enter(&ds->ds_dir->dd_lock);
113 113 mutex_enter(&ds->ds_lock);
114 114 delta = parent_delta(ds, used);
115 115 ds->ds_phys->ds_referenced_bytes += used;
116 116 ds->ds_phys->ds_compressed_bytes += compressed;
117 117 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
118 118 ds->ds_phys->ds_unique_bytes += used;
119 119 mutex_exit(&ds->ds_lock);
120 120 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
121 121 compressed, uncompressed, tx);
122 122 dsl_dir_transfer_space(ds->ds_dir, used - delta,
123 123 DD_USED_REFRSRV, DD_USED_HEAD, tx);
124 124 mutex_exit(&ds->ds_dir->dd_lock);
125 125 }
126 126
127 127 int
128 128 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
129 129 boolean_t async)
130 130 {
131 131 if (BP_IS_HOLE(bp))
132 132 return (0);
133 133
134 134 ASSERT(dmu_tx_is_syncing(tx));
135 135 ASSERT(bp->blk_birth <= tx->tx_txg);
136 136
137 137 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
138 138 int compressed = BP_GET_PSIZE(bp);
139 139 int uncompressed = BP_GET_UCSIZE(bp);
140 140
141 141 ASSERT(used > 0);
142 142 if (ds == NULL) {
143 143 dsl_free(tx->tx_pool, tx->tx_txg, bp);
144 144 dsl_pool_mos_diduse_space(tx->tx_pool,
145 145 -used, -compressed, -uncompressed);
146 146 return (used);
147 147 }
148 148 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
149 149
150 150 ASSERT(!dsl_dataset_is_snapshot(ds));
151 151 dmu_buf_will_dirty(ds->ds_dbuf, tx);
152 152
153 153 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
154 154 int64_t delta;
155 155
156 156 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
157 157 dsl_free(tx->tx_pool, tx->tx_txg, bp);
158 158
159 159 mutex_enter(&ds->ds_dir->dd_lock);
160 160 mutex_enter(&ds->ds_lock);
161 161 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
162 162 !DS_UNIQUE_IS_ACCURATE(ds));
163 163 delta = parent_delta(ds, -used);
164 164 ds->ds_phys->ds_unique_bytes -= used;
165 165 mutex_exit(&ds->ds_lock);
166 166 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
167 167 delta, -compressed, -uncompressed, tx);
168 168 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
169 169 DD_USED_REFRSRV, DD_USED_HEAD, tx);
170 170 mutex_exit(&ds->ds_dir->dd_lock);
171 171 } else {
172 172 dprintf_bp(bp, "putting on dead list: %s", "");
173 173 if (async) {
174 174 /*
175 175 * We are here as part of zio's write done callback,
176 176 * which means we're a zio interrupt thread. We can't
177 177 * call dsl_deadlist_insert() now because it may block
178 178 * waiting for I/O. Instead, put bp on the deferred
179 179 * queue and let dsl_pool_sync() finish the job.
180 180 */
181 181 bplist_append(&ds->ds_pending_deadlist, bp);
182 182 } else {
183 183 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
184 184 }
185 185 ASSERT3U(ds->ds_prev->ds_object, ==,
186 186 ds->ds_phys->ds_prev_snap_obj);
187 187 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
188 188 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
189 189 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
190 190 ds->ds_object && bp->blk_birth >
191 191 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
192 192 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
193 193 mutex_enter(&ds->ds_prev->ds_lock);
194 194 ds->ds_prev->ds_phys->ds_unique_bytes += used;
195 195 mutex_exit(&ds->ds_prev->ds_lock);
196 196 }
197 197 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
198 198 dsl_dir_transfer_space(ds->ds_dir, used,
199 199 DD_USED_HEAD, DD_USED_SNAP, tx);
200 200 }
201 201 }
202 202 mutex_enter(&ds->ds_lock);
203 203 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
204 204 ds->ds_phys->ds_referenced_bytes -= used;
205 205 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
206 206 ds->ds_phys->ds_compressed_bytes -= compressed;
207 207 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
208 208 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
209 209 mutex_exit(&ds->ds_lock);
210 210
211 211 return (used);
212 212 }
213 213
214 214 uint64_t
215 215 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
216 216 {
217 217 uint64_t trysnap = 0;
218 218
219 219 if (ds == NULL)
220 220 return (0);
221 221 /*
222 222 * The snapshot creation could fail, but that would cause an
223 223 * incorrect FALSE return, which would only result in an
224 224 * overestimation of the amount of space that an operation would
225 225 * consume, which is OK.
226 226 *
227 227 * There's also a small window where we could miss a pending
228 228 * snapshot, because we could set the sync task in the quiescing
229 229 * phase. So this should only be used as a guess.
230 230 */
231 231 if (ds->ds_trysnap_txg >
232 232 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
233 233 trysnap = ds->ds_trysnap_txg;
234 234 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
235 235 }
236 236
237 237 boolean_t
238 238 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
239 239 uint64_t blk_birth)
240 240 {
241 241 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
242 242 return (B_FALSE);
243 243
244 244 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
245 245
246 246 return (B_TRUE);
247 247 }
248 248
249 249 /* ARGSUSED */
250 250 static void
251 251 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
252 252 {
253 253 dsl_dataset_t *ds = dsv;
254 254
255 255 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
256 256
257 257 unique_remove(ds->ds_fsid_guid);
258 258
259 259 if (ds->ds_objset != NULL)
260 260 dmu_objset_evict(ds->ds_objset);
261 261
262 262 if (ds->ds_prev) {
263 263 dsl_dataset_drop_ref(ds->ds_prev, ds);
264 264 ds->ds_prev = NULL;
265 265 }
266 266
267 267 bplist_destroy(&ds->ds_pending_deadlist);
268 268 if (db != NULL) {
269 269 dsl_deadlist_close(&ds->ds_deadlist);
270 270 } else {
271 271 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
272 272 ASSERT(!ds->ds_deadlist.dl_oldfmt);
273 273 }
274 274 if (ds->ds_dir)
275 275 dsl_dir_close(ds->ds_dir, ds);
276 276
277 277 ASSERT(!list_link_active(&ds->ds_synced_link));
278 278
279 279 mutex_destroy(&ds->ds_lock);
280 280 mutex_destroy(&ds->ds_recvlock);
281 281 mutex_destroy(&ds->ds_opening_lock);
282 282 rw_destroy(&ds->ds_rwlock);
283 283 cv_destroy(&ds->ds_exclusive_cv);
284 284
285 285 kmem_free(ds, sizeof (dsl_dataset_t));
286 286 }
287 287
288 288 static int
289 289 dsl_dataset_get_snapname(dsl_dataset_t *ds)
290 290 {
291 291 dsl_dataset_phys_t *headphys;
292 292 int err;
293 293 dmu_buf_t *headdbuf;
294 294 dsl_pool_t *dp = ds->ds_dir->dd_pool;
295 295 objset_t *mos = dp->dp_meta_objset;
296 296
297 297 if (ds->ds_snapname[0])
298 298 return (0);
299 299 if (ds->ds_phys->ds_next_snap_obj == 0)
300 300 return (0);
301 301
302 302 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
303 303 FTAG, &headdbuf);
304 304 if (err)
305 305 return (err);
306 306 headphys = headdbuf->db_data;
307 307 err = zap_value_search(dp->dp_meta_objset,
308 308 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
309 309 dmu_buf_rele(headdbuf, FTAG);
310 310 return (err);
311 311 }
312 312
313 313 static int
314 314 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
315 315 {
316 316 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
317 317 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
318 318 matchtype_t mt;
319 319 int err;
320 320
321 321 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
322 322 mt = MT_FIRST;
323 323 else
324 324 mt = MT_EXACT;
325 325
326 326 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
327 327 value, mt, NULL, 0, NULL);
328 328 if (err == ENOTSUP && mt == MT_FIRST)
329 329 err = zap_lookup(mos, snapobj, name, 8, 1, value);
330 330 return (err);
331 331 }
332 332
333 333 static int
334 334 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
335 335 {
336 336 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
337 337 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
338 338 matchtype_t mt;
339 339 int err;
340 340
|
↓ open down ↓ |
340 lines elided |
↑ open up ↑ |
341 341 dsl_dir_snap_cmtime_update(ds->ds_dir);
342 342
343 343 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
344 344 mt = MT_FIRST;
345 345 else
346 346 mt = MT_EXACT;
347 347
348 348 err = zap_remove_norm(mos, snapobj, name, mt, tx);
349 349 if (err == ENOTSUP && mt == MT_FIRST)
350 350 err = zap_remove(mos, snapobj, name, tx);
351 +
352 + if (err == 0)
353 + dsl_snapcount_adjust(ds->ds_dir, tx, -1, B_TRUE);
354 +
351 355 return (err);
352 356 }
353 357
354 358 static int
355 359 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
356 360 dsl_dataset_t **dsp)
357 361 {
358 362 objset_t *mos = dp->dp_meta_objset;
359 363 dmu_buf_t *dbuf;
360 364 dsl_dataset_t *ds;
361 365 int err;
362 366 dmu_object_info_t doi;
363 367
364 368 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
365 369 dsl_pool_sync_context(dp));
366 370
367 371 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
368 372 if (err)
369 373 return (err);
370 374
371 375 /* Make sure dsobj has the correct object type. */
372 376 dmu_object_info_from_db(dbuf, &doi);
373 377 if (doi.doi_type != DMU_OT_DSL_DATASET)
374 378 return (EINVAL);
375 379
376 380 ds = dmu_buf_get_user(dbuf);
377 381 if (ds == NULL) {
378 382 dsl_dataset_t *winner;
379 383
380 384 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
381 385 ds->ds_dbuf = dbuf;
382 386 ds->ds_object = dsobj;
383 387 ds->ds_phys = dbuf->db_data;
384 388
385 389 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
386 390 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
387 391 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
388 392 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
389 393
390 394 rw_init(&ds->ds_rwlock, 0, 0, 0);
391 395 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
392 396
393 397 bplist_create(&ds->ds_pending_deadlist);
394 398 dsl_deadlist_open(&ds->ds_deadlist,
395 399 mos, ds->ds_phys->ds_deadlist_obj);
396 400
397 401 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
398 402 offsetof(dmu_sendarg_t, dsa_link));
399 403
400 404 if (err == 0) {
401 405 err = dsl_dir_open_obj(dp,
402 406 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
403 407 }
404 408 if (err) {
405 409 mutex_destroy(&ds->ds_lock);
406 410 mutex_destroy(&ds->ds_recvlock);
407 411 mutex_destroy(&ds->ds_opening_lock);
408 412 rw_destroy(&ds->ds_rwlock);
409 413 cv_destroy(&ds->ds_exclusive_cv);
410 414 bplist_destroy(&ds->ds_pending_deadlist);
411 415 dsl_deadlist_close(&ds->ds_deadlist);
412 416 kmem_free(ds, sizeof (dsl_dataset_t));
413 417 dmu_buf_rele(dbuf, tag);
414 418 return (err);
415 419 }
416 420
417 421 if (!dsl_dataset_is_snapshot(ds)) {
418 422 ds->ds_snapname[0] = '\0';
419 423 if (ds->ds_phys->ds_prev_snap_obj) {
420 424 err = dsl_dataset_get_ref(dp,
421 425 ds->ds_phys->ds_prev_snap_obj,
422 426 ds, &ds->ds_prev);
423 427 }
424 428 } else {
425 429 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
426 430 err = dsl_dataset_get_snapname(ds);
427 431 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
428 432 err = zap_count(
429 433 ds->ds_dir->dd_pool->dp_meta_objset,
430 434 ds->ds_phys->ds_userrefs_obj,
431 435 &ds->ds_userrefs);
432 436 }
433 437 }
434 438
435 439 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
436 440 /*
437 441 * In sync context, we're called with either no lock
438 442 * or with the write lock. If we're not syncing,
439 443 * we're always called with the read lock held.
440 444 */
441 445 boolean_t need_lock =
442 446 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
443 447 dsl_pool_sync_context(dp);
444 448
445 449 if (need_lock)
446 450 rw_enter(&dp->dp_config_rwlock, RW_READER);
447 451
448 452 err = dsl_prop_get_ds(ds,
449 453 "refreservation", sizeof (uint64_t), 1,
450 454 &ds->ds_reserved, NULL);
451 455 if (err == 0) {
452 456 err = dsl_prop_get_ds(ds,
453 457 "refquota", sizeof (uint64_t), 1,
454 458 &ds->ds_quota, NULL);
455 459 }
456 460
457 461 if (need_lock)
458 462 rw_exit(&dp->dp_config_rwlock);
459 463 } else {
460 464 ds->ds_reserved = ds->ds_quota = 0;
461 465 }
462 466
463 467 if (err == 0) {
464 468 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
465 469 dsl_dataset_evict);
466 470 }
467 471 if (err || winner) {
468 472 bplist_destroy(&ds->ds_pending_deadlist);
469 473 dsl_deadlist_close(&ds->ds_deadlist);
470 474 if (ds->ds_prev)
471 475 dsl_dataset_drop_ref(ds->ds_prev, ds);
472 476 dsl_dir_close(ds->ds_dir, ds);
473 477 mutex_destroy(&ds->ds_lock);
474 478 mutex_destroy(&ds->ds_recvlock);
475 479 mutex_destroy(&ds->ds_opening_lock);
476 480 rw_destroy(&ds->ds_rwlock);
477 481 cv_destroy(&ds->ds_exclusive_cv);
478 482 kmem_free(ds, sizeof (dsl_dataset_t));
479 483 if (err) {
480 484 dmu_buf_rele(dbuf, tag);
481 485 return (err);
482 486 }
483 487 ds = winner;
484 488 } else {
485 489 ds->ds_fsid_guid =
486 490 unique_insert(ds->ds_phys->ds_fsid_guid);
487 491 }
488 492 }
489 493 ASSERT3P(ds->ds_dbuf, ==, dbuf);
490 494 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
491 495 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
492 496 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
493 497 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
494 498 mutex_enter(&ds->ds_lock);
495 499 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
496 500 mutex_exit(&ds->ds_lock);
497 501 dmu_buf_rele(ds->ds_dbuf, tag);
498 502 return (ENOENT);
499 503 }
500 504 mutex_exit(&ds->ds_lock);
501 505 *dsp = ds;
502 506 return (0);
503 507 }
504 508
505 509 static int
506 510 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
507 511 {
508 512 dsl_pool_t *dp = ds->ds_dir->dd_pool;
509 513
510 514 /*
511 515 * In syncing context we don't want the rwlock lock: there
512 516 * may be an existing writer waiting for sync phase to
513 517 * finish. We don't need to worry about such writers, since
514 518 * sync phase is single-threaded, so the writer can't be
515 519 * doing anything while we are active.
516 520 */
517 521 if (dsl_pool_sync_context(dp)) {
518 522 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
519 523 return (0);
520 524 }
521 525
522 526 /*
523 527 * Normal users will hold the ds_rwlock as a READER until they
524 528 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
525 529 * drop their READER lock after they set the ds_owner field.
526 530 *
527 531 * If the dataset is being destroyed, the destroy thread will
528 532 * obtain a WRITER lock for exclusive access after it's done its
529 533 * open-context work and then change the ds_owner to
530 534 * dsl_reaper once destruction is assured. So threads
531 535 * may block here temporarily, until the "destructability" of
532 536 * the dataset is determined.
533 537 */
534 538 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
535 539 mutex_enter(&ds->ds_lock);
536 540 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
537 541 rw_exit(&dp->dp_config_rwlock);
538 542 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
539 543 if (DSL_DATASET_IS_DESTROYED(ds)) {
540 544 mutex_exit(&ds->ds_lock);
541 545 dsl_dataset_drop_ref(ds, tag);
542 546 rw_enter(&dp->dp_config_rwlock, RW_READER);
543 547 return (ENOENT);
544 548 }
545 549 /*
546 550 * The dp_config_rwlock lives above the ds_lock. And
547 551 * we need to check DSL_DATASET_IS_DESTROYED() while
548 552 * holding the ds_lock, so we have to drop and reacquire
549 553 * the ds_lock here.
550 554 */
551 555 mutex_exit(&ds->ds_lock);
552 556 rw_enter(&dp->dp_config_rwlock, RW_READER);
553 557 mutex_enter(&ds->ds_lock);
554 558 }
555 559 mutex_exit(&ds->ds_lock);
556 560 return (0);
557 561 }
558 562
559 563 int
560 564 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
561 565 dsl_dataset_t **dsp)
562 566 {
563 567 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
564 568
565 569 if (err)
566 570 return (err);
567 571 return (dsl_dataset_hold_ref(*dsp, tag));
568 572 }
569 573
570 574 int
571 575 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
572 576 void *tag, dsl_dataset_t **dsp)
573 577 {
574 578 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
575 579 if (err)
576 580 return (err);
577 581 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
578 582 dsl_dataset_rele(*dsp, tag);
579 583 *dsp = NULL;
580 584 return (EBUSY);
581 585 }
582 586 return (0);
583 587 }
584 588
585 589 int
586 590 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
587 591 {
588 592 dsl_dir_t *dd;
589 593 dsl_pool_t *dp;
590 594 const char *snapname;
591 595 uint64_t obj;
592 596 int err = 0;
593 597
594 598 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
595 599 if (err)
596 600 return (err);
597 601
598 602 dp = dd->dd_pool;
599 603 obj = dd->dd_phys->dd_head_dataset_obj;
600 604 rw_enter(&dp->dp_config_rwlock, RW_READER);
601 605 if (obj)
602 606 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
603 607 else
604 608 err = ENOENT;
605 609 if (err)
606 610 goto out;
607 611
608 612 err = dsl_dataset_hold_ref(*dsp, tag);
609 613
610 614 /* we may be looking for a snapshot */
611 615 if (err == 0 && snapname != NULL) {
612 616 dsl_dataset_t *ds = NULL;
613 617
614 618 if (*snapname++ != '@') {
615 619 dsl_dataset_rele(*dsp, tag);
616 620 err = ENOENT;
617 621 goto out;
618 622 }
619 623
620 624 dprintf("looking for snapshot '%s'\n", snapname);
621 625 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
622 626 if (err == 0)
623 627 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
624 628 dsl_dataset_rele(*dsp, tag);
625 629
626 630 ASSERT3U((err == 0), ==, (ds != NULL));
627 631
628 632 if (ds) {
629 633 mutex_enter(&ds->ds_lock);
630 634 if (ds->ds_snapname[0] == 0)
631 635 (void) strlcpy(ds->ds_snapname, snapname,
632 636 sizeof (ds->ds_snapname));
633 637 mutex_exit(&ds->ds_lock);
634 638 err = dsl_dataset_hold_ref(ds, tag);
635 639 *dsp = err ? NULL : ds;
636 640 }
637 641 }
638 642 out:
639 643 rw_exit(&dp->dp_config_rwlock);
640 644 dsl_dir_close(dd, FTAG);
641 645 return (err);
642 646 }
643 647
644 648 int
645 649 dsl_dataset_own(const char *name, boolean_t inconsistentok,
646 650 void *tag, dsl_dataset_t **dsp)
647 651 {
648 652 int err = dsl_dataset_hold(name, tag, dsp);
649 653 if (err)
650 654 return (err);
651 655 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
652 656 dsl_dataset_rele(*dsp, tag);
653 657 return (EBUSY);
654 658 }
655 659 return (0);
656 660 }
657 661
658 662 void
659 663 dsl_dataset_name(dsl_dataset_t *ds, char *name)
660 664 {
661 665 if (ds == NULL) {
662 666 (void) strcpy(name, "mos");
663 667 } else {
664 668 dsl_dir_name(ds->ds_dir, name);
665 669 VERIFY(0 == dsl_dataset_get_snapname(ds));
666 670 if (ds->ds_snapname[0]) {
667 671 (void) strcat(name, "@");
668 672 /*
669 673 * We use a "recursive" mutex so that we
670 674 * can call dprintf_ds() with ds_lock held.
671 675 */
672 676 if (!MUTEX_HELD(&ds->ds_lock)) {
673 677 mutex_enter(&ds->ds_lock);
674 678 (void) strcat(name, ds->ds_snapname);
675 679 mutex_exit(&ds->ds_lock);
676 680 } else {
677 681 (void) strcat(name, ds->ds_snapname);
678 682 }
679 683 }
680 684 }
681 685 }
682 686
683 687 static int
684 688 dsl_dataset_namelen(dsl_dataset_t *ds)
685 689 {
686 690 int result;
687 691
688 692 if (ds == NULL) {
689 693 result = 3; /* "mos" */
690 694 } else {
691 695 result = dsl_dir_namelen(ds->ds_dir);
692 696 VERIFY(0 == dsl_dataset_get_snapname(ds));
693 697 if (ds->ds_snapname[0]) {
694 698 ++result; /* adding one for the @-sign */
695 699 if (!MUTEX_HELD(&ds->ds_lock)) {
696 700 mutex_enter(&ds->ds_lock);
697 701 result += strlen(ds->ds_snapname);
698 702 mutex_exit(&ds->ds_lock);
699 703 } else {
700 704 result += strlen(ds->ds_snapname);
701 705 }
702 706 }
703 707 }
704 708
705 709 return (result);
706 710 }
707 711
708 712 void
709 713 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
710 714 {
711 715 dmu_buf_rele(ds->ds_dbuf, tag);
712 716 }
713 717
714 718 void
715 719 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
716 720 {
717 721 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
718 722 rw_exit(&ds->ds_rwlock);
719 723 }
720 724 dsl_dataset_drop_ref(ds, tag);
721 725 }
722 726
723 727 void
724 728 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
725 729 {
726 730 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
727 731 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
728 732
729 733 mutex_enter(&ds->ds_lock);
730 734 ds->ds_owner = NULL;
731 735 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
732 736 rw_exit(&ds->ds_rwlock);
733 737 cv_broadcast(&ds->ds_exclusive_cv);
734 738 }
735 739 mutex_exit(&ds->ds_lock);
736 740 if (ds->ds_dbuf)
737 741 dsl_dataset_drop_ref(ds, tag);
738 742 else
739 743 dsl_dataset_evict(NULL, ds);
740 744 }
741 745
742 746 boolean_t
743 747 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
744 748 {
745 749 boolean_t gotit = FALSE;
746 750
747 751 mutex_enter(&ds->ds_lock);
748 752 if (ds->ds_owner == NULL &&
749 753 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
750 754 ds->ds_owner = tag;
751 755 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
752 756 rw_exit(&ds->ds_rwlock);
753 757 gotit = TRUE;
754 758 }
755 759 mutex_exit(&ds->ds_lock);
756 760 return (gotit);
757 761 }
758 762
759 763 void
760 764 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
761 765 {
762 766 ASSERT3P(owner, ==, ds->ds_owner);
763 767 if (!RW_WRITE_HELD(&ds->ds_rwlock))
764 768 rw_enter(&ds->ds_rwlock, RW_WRITER);
765 769 }
766 770
767 771 uint64_t
768 772 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
769 773 uint64_t flags, dmu_tx_t *tx)
770 774 {
771 775 dsl_pool_t *dp = dd->dd_pool;
772 776 dmu_buf_t *dbuf;
773 777 dsl_dataset_phys_t *dsphys;
774 778 uint64_t dsobj;
775 779 objset_t *mos = dp->dp_meta_objset;
776 780
777 781 if (origin == NULL)
778 782 origin = dp->dp_origin_snap;
779 783
780 784 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
781 785 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
782 786 ASSERT(dmu_tx_is_syncing(tx));
783 787 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
784 788
785 789 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
786 790 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
787 791 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
788 792 dmu_buf_will_dirty(dbuf, tx);
789 793 dsphys = dbuf->db_data;
790 794 bzero(dsphys, sizeof (dsl_dataset_phys_t));
791 795 dsphys->ds_dir_obj = dd->dd_object;
792 796 dsphys->ds_flags = flags;
793 797 dsphys->ds_fsid_guid = unique_create();
794 798 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
795 799 sizeof (dsphys->ds_guid));
796 800 dsphys->ds_snapnames_zapobj =
797 801 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
798 802 DMU_OT_NONE, 0, tx);
799 803 dsphys->ds_creation_time = gethrestime_sec();
800 804 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
801 805
802 806 if (origin == NULL) {
803 807 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
804 808 } else {
805 809 dsl_dataset_t *ohds;
806 810
807 811 dsphys->ds_prev_snap_obj = origin->ds_object;
808 812 dsphys->ds_prev_snap_txg =
809 813 origin->ds_phys->ds_creation_txg;
810 814 dsphys->ds_referenced_bytes =
811 815 origin->ds_phys->ds_referenced_bytes;
812 816 dsphys->ds_compressed_bytes =
813 817 origin->ds_phys->ds_compressed_bytes;
814 818 dsphys->ds_uncompressed_bytes =
815 819 origin->ds_phys->ds_uncompressed_bytes;
816 820 dsphys->ds_bp = origin->ds_phys->ds_bp;
817 821 dsphys->ds_flags |= origin->ds_phys->ds_flags;
818 822
819 823 dmu_buf_will_dirty(origin->ds_dbuf, tx);
820 824 origin->ds_phys->ds_num_children++;
821 825
822 826 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
823 827 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
824 828 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
825 829 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
826 830 dsl_dataset_rele(ohds, FTAG);
827 831
828 832 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
829 833 if (origin->ds_phys->ds_next_clones_obj == 0) {
830 834 origin->ds_phys->ds_next_clones_obj =
831 835 zap_create(mos,
832 836 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
833 837 }
834 838 VERIFY(0 == zap_add_int(mos,
835 839 origin->ds_phys->ds_next_clones_obj,
836 840 dsobj, tx));
837 841 }
838 842
839 843 dmu_buf_will_dirty(dd->dd_dbuf, tx);
840 844 dd->dd_phys->dd_origin_obj = origin->ds_object;
841 845 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
842 846 if (origin->ds_dir->dd_phys->dd_clones == 0) {
843 847 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
844 848 origin->ds_dir->dd_phys->dd_clones =
845 849 zap_create(mos,
846 850 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
847 851 }
848 852 VERIFY3U(0, ==, zap_add_int(mos,
849 853 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
850 854 }
851 855 }
852 856
853 857 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
854 858 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
855 859
856 860 dmu_buf_rele(dbuf, FTAG);
857 861
858 862 dmu_buf_will_dirty(dd->dd_dbuf, tx);
859 863 dd->dd_phys->dd_head_dataset_obj = dsobj;
860 864
861 865 return (dsobj);
862 866 }
863 867
864 868 uint64_t
865 869 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
866 870 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
867 871 {
868 872 dsl_pool_t *dp = pdd->dd_pool;
869 873 uint64_t dsobj, ddobj;
870 874 dsl_dir_t *dd;
871 875
872 876 ASSERT(lastname[0] != '@');
873 877
874 878 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
875 879 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
876 880
877 881 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
878 882
879 883 dsl_deleg_set_create_perms(dd, tx, cr);
880 884
881 885 dsl_dir_close(dd, FTAG);
882 886
883 887 /*
884 888 * If we are creating a clone, make sure we zero out any stale
885 889 * data from the origin snapshots zil header.
886 890 */
887 891 if (origin != NULL) {
888 892 dsl_dataset_t *ds;
889 893 objset_t *os;
890 894
891 895 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
892 896 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
893 897 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
894 898 dsl_dataset_dirty(ds, tx);
895 899 dsl_dataset_rele(ds, FTAG);
896 900 }
897 901
898 902 return (dsobj);
899 903 }
900 904
901 905 /*
902 906 * The snapshots must all be in the same pool.
903 907 */
904 908 int
905 909 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer,
906 910 nvlist_t *errlist)
907 911 {
908 912 int err;
909 913 dsl_sync_task_t *dst;
910 914 spa_t *spa;
911 915 nvpair_t *pair;
912 916 dsl_sync_task_group_t *dstg;
913 917
914 918 pair = nvlist_next_nvpair(snaps, NULL);
915 919 if (pair == NULL)
916 920 return (0);
917 921
918 922 err = spa_open(nvpair_name(pair), &spa, FTAG);
919 923 if (err)
920 924 return (err);
921 925 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
922 926
923 927 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
924 928 pair = nvlist_next_nvpair(snaps, pair)) {
925 929 dsl_dataset_t *ds;
926 930
927 931 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
928 932 if (err == 0) {
929 933 struct dsl_ds_destroyarg *dsda;
930 934
931 935 dsl_dataset_make_exclusive(ds, dstg);
932 936 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
933 937 KM_SLEEP);
934 938 dsda->ds = ds;
935 939 dsda->defer = defer;
936 940 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
937 941 dsl_dataset_destroy_sync, dsda, dstg, 0);
938 942 } else if (err == ENOENT) {
939 943 err = 0;
940 944 } else {
941 945 fnvlist_add_int32(errlist, nvpair_name(pair), err);
942 946 break;
943 947 }
944 948 }
945 949
946 950 if (err == 0)
947 951 err = dsl_sync_task_group_wait(dstg);
948 952
949 953 for (dst = list_head(&dstg->dstg_tasks); dst;
950 954 dst = list_next(&dstg->dstg_tasks, dst)) {
951 955 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
952 956 dsl_dataset_t *ds = dsda->ds;
953 957
954 958 /*
955 959 * Return the snapshots that triggered the error.
956 960 */
957 961 if (dst->dst_err != 0) {
958 962 char name[ZFS_MAXNAMELEN];
959 963 dsl_dataset_name(ds, name);
960 964 fnvlist_add_int32(errlist, name, dst->dst_err);
961 965 }
962 966 ASSERT3P(dsda->rm_origin, ==, NULL);
963 967 dsl_dataset_disown(ds, dstg);
964 968 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
965 969 }
966 970
967 971 dsl_sync_task_group_destroy(dstg);
968 972 spa_close(spa, FTAG);
969 973 return (err);
970 974
971 975 }
972 976
973 977 static boolean_t
974 978 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
975 979 {
976 980 boolean_t might_destroy = B_FALSE;
977 981
978 982 mutex_enter(&ds->ds_lock);
979 983 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
980 984 DS_IS_DEFER_DESTROY(ds))
981 985 might_destroy = B_TRUE;
982 986 mutex_exit(&ds->ds_lock);
983 987
984 988 return (might_destroy);
985 989 }
986 990
987 991 /*
988 992 * If we're removing a clone, and these three conditions are true:
989 993 * 1) the clone's origin has no other children
990 994 * 2) the clone's origin has no user references
991 995 * 3) the clone's origin has been marked for deferred destruction
992 996 * Then, prepare to remove the origin as part of this sync task group.
993 997 */
994 998 static int
995 999 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
996 1000 {
997 1001 dsl_dataset_t *ds = dsda->ds;
998 1002 dsl_dataset_t *origin = ds->ds_prev;
999 1003
1000 1004 if (dsl_dataset_might_destroy_origin(origin)) {
1001 1005 char *name;
1002 1006 int namelen;
1003 1007 int error;
1004 1008
1005 1009 namelen = dsl_dataset_namelen(origin) + 1;
1006 1010 name = kmem_alloc(namelen, KM_SLEEP);
1007 1011 dsl_dataset_name(origin, name);
1008 1012 #ifdef _KERNEL
1009 1013 error = zfs_unmount_snap(name, NULL);
1010 1014 if (error) {
1011 1015 kmem_free(name, namelen);
1012 1016 return (error);
1013 1017 }
1014 1018 #endif
1015 1019 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1016 1020 kmem_free(name, namelen);
1017 1021 if (error)
1018 1022 return (error);
1019 1023 dsda->rm_origin = origin;
1020 1024 dsl_dataset_make_exclusive(origin, tag);
1021 1025 }
1022 1026
1023 1027 return (0);
1024 1028 }
1025 1029
1026 1030 /*
1027 1031 * ds must be opened as OWNER. On return (whether successful or not),
1028 1032 * ds will be closed and caller can no longer dereference it.
1029 1033 */
1030 1034 int
1031 1035 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1032 1036 {
1033 1037 int err;
1034 1038 dsl_sync_task_group_t *dstg;
1035 1039 objset_t *os;
1036 1040 dsl_dir_t *dd;
1037 1041 uint64_t obj;
1038 1042 struct dsl_ds_destroyarg dsda = { 0 };
1039 1043
1040 1044 dsda.ds = ds;
1041 1045
1042 1046 if (dsl_dataset_is_snapshot(ds)) {
1043 1047 /* Destroying a snapshot is simpler */
1044 1048 dsl_dataset_make_exclusive(ds, tag);
1045 1049
1046 1050 dsda.defer = defer;
1047 1051 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1048 1052 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1049 1053 &dsda, tag, 0);
1050 1054 ASSERT3P(dsda.rm_origin, ==, NULL);
1051 1055 goto out;
1052 1056 } else if (defer) {
1053 1057 err = EINVAL;
1054 1058 goto out;
1055 1059 }
1056 1060
1057 1061 dd = ds->ds_dir;
1058 1062
1059 1063 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1060 1064 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1061 1065 /*
1062 1066 * Check for errors and mark this ds as inconsistent, in
1063 1067 * case we crash while freeing the objects.
1064 1068 */
1065 1069 err = dsl_sync_task_do(dd->dd_pool,
1066 1070 dsl_dataset_destroy_begin_check,
1067 1071 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1068 1072 if (err)
1069 1073 goto out;
1070 1074
1071 1075 err = dmu_objset_from_ds(ds, &os);
1072 1076 if (err)
1073 1077 goto out;
1074 1078
1075 1079 /*
1076 1080 * Remove all objects while in the open context so that
1077 1081 * there is less work to do in the syncing context.
1078 1082 */
1079 1083 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1080 1084 ds->ds_phys->ds_prev_snap_txg)) {
1081 1085 /*
1082 1086 * Ignore errors, if there is not enough disk space
1083 1087 * we will deal with it in dsl_dataset_destroy_sync().
1084 1088 */
1085 1089 (void) dmu_free_object(os, obj);
1086 1090 }
1087 1091 if (err != ESRCH)
1088 1092 goto out;
1089 1093
1090 1094 /*
1091 1095 * Sync out all in-flight IO.
1092 1096 */
1093 1097 txg_wait_synced(dd->dd_pool, 0);
1094 1098
1095 1099 /*
1096 1100 * If we managed to free all the objects in open
1097 1101 * context, the user space accounting should be zero.
1098 1102 */
1099 1103 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1100 1104 dmu_objset_userused_enabled(os)) {
1101 1105 uint64_t count;
1102 1106
1103 1107 ASSERT(zap_count(os, DMU_USERUSED_OBJECT,
1104 1108 &count) != 0 || count == 0);
1105 1109 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT,
1106 1110 &count) != 0 || count == 0);
1107 1111 }
1108 1112 }
1109 1113
1110 1114 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1111 1115 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1112 1116 rw_exit(&dd->dd_pool->dp_config_rwlock);
1113 1117
1114 1118 if (err)
1115 1119 goto out;
1116 1120
1117 1121 /*
1118 1122 * Blow away the dsl_dir + head dataset.
1119 1123 */
1120 1124 dsl_dataset_make_exclusive(ds, tag);
1121 1125 /*
1122 1126 * If we're removing a clone, we might also need to remove its
1123 1127 * origin.
1124 1128 */
1125 1129 do {
1126 1130 dsda.need_prep = B_FALSE;
1127 1131 if (dsl_dir_is_clone(dd)) {
1128 1132 err = dsl_dataset_origin_rm_prep(&dsda, tag);
|
↓ open down ↓ |
768 lines elided |
↑ open up ↑ |
1129 1133 if (err) {
1130 1134 dsl_dir_close(dd, FTAG);
1131 1135 goto out;
1132 1136 }
1133 1137 }
1134 1138
1135 1139 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1136 1140 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1137 1141 dsl_dataset_destroy_sync, &dsda, tag, 0);
1138 1142 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1139 - dsl_dir_destroy_sync, dd, FTAG, 0);
1143 + dsl_dir_destroy_sync, dd, tag, 0);
1140 1144 err = dsl_sync_task_group_wait(dstg);
1141 1145 dsl_sync_task_group_destroy(dstg);
1142 1146
1143 1147 /*
1144 1148 * We could be racing against 'zfs release' or 'zfs destroy -d'
1145 1149 * on the origin snap, in which case we can get EBUSY if we
1146 1150 * needed to destroy the origin snap but were not ready to
1147 1151 * do so.
1148 1152 */
1149 1153 if (dsda.need_prep) {
1150 1154 ASSERT(err == EBUSY);
1151 1155 ASSERT(dsl_dir_is_clone(dd));
1152 1156 ASSERT(dsda.rm_origin == NULL);
1153 1157 }
1154 1158 } while (dsda.need_prep);
1155 1159
1156 1160 if (dsda.rm_origin != NULL)
1157 1161 dsl_dataset_disown(dsda.rm_origin, tag);
1158 1162
1159 1163 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1160 1164 if (err)
1161 1165 dsl_dir_close(dd, FTAG);
1162 1166 out:
1163 1167 dsl_dataset_disown(ds, tag);
1164 1168 return (err);
1165 1169 }
1166 1170
1167 1171 blkptr_t *
1168 1172 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1169 1173 {
1170 1174 return (&ds->ds_phys->ds_bp);
1171 1175 }
1172 1176
1173 1177 void
1174 1178 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1175 1179 {
1176 1180 ASSERT(dmu_tx_is_syncing(tx));
1177 1181 /* If it's the meta-objset, set dp_meta_rootbp */
1178 1182 if (ds == NULL) {
1179 1183 tx->tx_pool->dp_meta_rootbp = *bp;
1180 1184 } else {
1181 1185 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1182 1186 ds->ds_phys->ds_bp = *bp;
1183 1187 }
1184 1188 }
1185 1189
1186 1190 spa_t *
1187 1191 dsl_dataset_get_spa(dsl_dataset_t *ds)
1188 1192 {
1189 1193 return (ds->ds_dir->dd_pool->dp_spa);
1190 1194 }
1191 1195
1192 1196 void
1193 1197 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1194 1198 {
1195 1199 dsl_pool_t *dp;
1196 1200
1197 1201 if (ds == NULL) /* this is the meta-objset */
1198 1202 return;
1199 1203
1200 1204 ASSERT(ds->ds_objset != NULL);
1201 1205
1202 1206 if (ds->ds_phys->ds_next_snap_obj != 0)
1203 1207 panic("dirtying snapshot!");
1204 1208
1205 1209 dp = ds->ds_dir->dd_pool;
1206 1210
1207 1211 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1208 1212 /* up the hold count until we can be written out */
1209 1213 dmu_buf_add_ref(ds->ds_dbuf, ds);
1210 1214 }
1211 1215 }
1212 1216
1213 1217 boolean_t
1214 1218 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1215 1219 {
1216 1220 for (int t = 0; t < TXG_SIZE; t++) {
1217 1221 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1218 1222 ds, t))
1219 1223 return (B_TRUE);
1220 1224 }
1221 1225 return (B_FALSE);
1222 1226 }
1223 1227
1224 1228 /*
1225 1229 * The unique space in the head dataset can be calculated by subtracting
1226 1230 * the space used in the most recent snapshot, that is still being used
1227 1231 * in this file system, from the space currently in use. To figure out
1228 1232 * the space in the most recent snapshot still in use, we need to take
1229 1233 * the total space used in the snapshot and subtract out the space that
1230 1234 * has been freed up since the snapshot was taken.
1231 1235 */
1232 1236 static void
1233 1237 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1234 1238 {
1235 1239 uint64_t mrs_used;
1236 1240 uint64_t dlused, dlcomp, dluncomp;
1237 1241
1238 1242 ASSERT(!dsl_dataset_is_snapshot(ds));
1239 1243
1240 1244 if (ds->ds_phys->ds_prev_snap_obj != 0)
1241 1245 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1242 1246 else
1243 1247 mrs_used = 0;
1244 1248
1245 1249 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1246 1250
1247 1251 ASSERT3U(dlused, <=, mrs_used);
1248 1252 ds->ds_phys->ds_unique_bytes =
1249 1253 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1250 1254
1251 1255 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1252 1256 SPA_VERSION_UNIQUE_ACCURATE)
1253 1257 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1254 1258 }
1255 1259
1256 1260 struct killarg {
1257 1261 dsl_dataset_t *ds;
1258 1262 dmu_tx_t *tx;
1259 1263 };
1260 1264
1261 1265 /* ARGSUSED */
1262 1266 static int
1263 1267 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1264 1268 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1265 1269 {
1266 1270 struct killarg *ka = arg;
1267 1271 dmu_tx_t *tx = ka->tx;
1268 1272
1269 1273 if (bp == NULL)
1270 1274 return (0);
1271 1275
1272 1276 if (zb->zb_level == ZB_ZIL_LEVEL) {
1273 1277 ASSERT(zilog != NULL);
1274 1278 /*
1275 1279 * It's a block in the intent log. It has no
1276 1280 * accounting, so just free it.
1277 1281 */
1278 1282 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1279 1283 } else {
1280 1284 ASSERT(zilog == NULL);
1281 1285 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1282 1286 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1283 1287 }
1284 1288
1285 1289 return (0);
1286 1290 }
1287 1291
1288 1292 /* ARGSUSED */
1289 1293 static int
1290 1294 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1291 1295 {
1292 1296 dsl_dataset_t *ds = arg1;
1293 1297 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1294 1298 uint64_t count;
1295 1299 int err;
1296 1300
1297 1301 /*
1298 1302 * Can't delete a head dataset if there are snapshots of it.
1299 1303 * (Except if the only snapshots are from the branch we cloned
1300 1304 * from.)
1301 1305 */
1302 1306 if (ds->ds_prev != NULL &&
1303 1307 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1304 1308 return (EBUSY);
1305 1309
1306 1310 /*
1307 1311 * This is really a dsl_dir thing, but check it here so that
1308 1312 * we'll be less likely to leave this dataset inconsistent &
1309 1313 * nearly destroyed.
1310 1314 */
1311 1315 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1312 1316 if (err)
1313 1317 return (err);
1314 1318 if (count != 0)
1315 1319 return (EEXIST);
1316 1320
1317 1321 return (0);
1318 1322 }
1319 1323
1320 1324 /* ARGSUSED */
1321 1325 static void
1322 1326 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1323 1327 {
1324 1328 dsl_dataset_t *ds = arg1;
1325 1329
1326 1330 /* Mark it as inconsistent on-disk, in case we crash */
1327 1331 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1328 1332 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1329 1333
1330 1334 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1331 1335 }
1332 1336
1333 1337 static int
1334 1338 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1335 1339 dmu_tx_t *tx)
1336 1340 {
1337 1341 dsl_dataset_t *ds = dsda->ds;
1338 1342 dsl_dataset_t *ds_prev = ds->ds_prev;
1339 1343
1340 1344 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1341 1345 struct dsl_ds_destroyarg ndsda = {0};
1342 1346
1343 1347 /*
1344 1348 * If we're not prepared to remove the origin, don't remove
1345 1349 * the clone either.
1346 1350 */
1347 1351 if (dsda->rm_origin == NULL) {
1348 1352 dsda->need_prep = B_TRUE;
1349 1353 return (EBUSY);
1350 1354 }
1351 1355
1352 1356 ndsda.ds = ds_prev;
1353 1357 ndsda.is_origin_rm = B_TRUE;
1354 1358 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1355 1359 }
1356 1360
1357 1361 /*
1358 1362 * If we're not going to remove the origin after all,
1359 1363 * undo the open context setup.
1360 1364 */
1361 1365 if (dsda->rm_origin != NULL) {
1362 1366 dsl_dataset_disown(dsda->rm_origin, tag);
1363 1367 dsda->rm_origin = NULL;
1364 1368 }
1365 1369
1366 1370 return (0);
1367 1371 }
1368 1372
1369 1373 /*
1370 1374 * If you add new checks here, you may need to add
1371 1375 * additional checks to the "temporary" case in
1372 1376 * snapshot_check() in dmu_objset.c.
1373 1377 */
1374 1378 /* ARGSUSED */
1375 1379 int
1376 1380 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1377 1381 {
1378 1382 struct dsl_ds_destroyarg *dsda = arg1;
1379 1383 dsl_dataset_t *ds = dsda->ds;
1380 1384
1381 1385 /* we have an owner hold, so noone else can destroy us */
1382 1386 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1383 1387
1384 1388 /*
1385 1389 * Only allow deferred destroy on pools that support it.
1386 1390 * NOTE: deferred destroy is only supported on snapshots.
1387 1391 */
1388 1392 if (dsda->defer) {
1389 1393 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1390 1394 SPA_VERSION_USERREFS)
1391 1395 return (ENOTSUP);
1392 1396 ASSERT(dsl_dataset_is_snapshot(ds));
1393 1397 return (0);
1394 1398 }
1395 1399
1396 1400 /*
1397 1401 * Can't delete a head dataset if there are snapshots of it.
1398 1402 * (Except if the only snapshots are from the branch we cloned
1399 1403 * from.)
1400 1404 */
1401 1405 if (ds->ds_prev != NULL &&
1402 1406 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1403 1407 return (EBUSY);
1404 1408
1405 1409 /*
1406 1410 * If we made changes this txg, traverse_dsl_dataset won't find
1407 1411 * them. Try again.
1408 1412 */
1409 1413 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1410 1414 return (EAGAIN);
1411 1415
1412 1416 if (dsl_dataset_is_snapshot(ds)) {
1413 1417 /*
1414 1418 * If this snapshot has an elevated user reference count,
1415 1419 * we can't destroy it yet.
1416 1420 */
1417 1421 if (ds->ds_userrefs > 0 && !dsda->releasing)
1418 1422 return (EBUSY);
1419 1423
1420 1424 mutex_enter(&ds->ds_lock);
1421 1425 /*
1422 1426 * Can't delete a branch point. However, if we're destroying
1423 1427 * a clone and removing its origin due to it having a user
1424 1428 * hold count of 0 and having been marked for deferred destroy,
1425 1429 * it's OK for the origin to have a single clone.
1426 1430 */
1427 1431 if (ds->ds_phys->ds_num_children >
1428 1432 (dsda->is_origin_rm ? 2 : 1)) {
1429 1433 mutex_exit(&ds->ds_lock);
1430 1434 return (EEXIST);
1431 1435 }
1432 1436 mutex_exit(&ds->ds_lock);
1433 1437 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1434 1438 return (dsl_dataset_origin_check(dsda, arg2, tx));
1435 1439 }
1436 1440
1437 1441 /* XXX we should do some i/o error checking... */
1438 1442 return (0);
1439 1443 }
1440 1444
1441 1445 struct refsarg {
1442 1446 kmutex_t lock;
1443 1447 boolean_t gone;
1444 1448 kcondvar_t cv;
1445 1449 };
1446 1450
1447 1451 /* ARGSUSED */
1448 1452 static void
1449 1453 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1450 1454 {
1451 1455 struct refsarg *arg = argv;
1452 1456
1453 1457 mutex_enter(&arg->lock);
1454 1458 arg->gone = TRUE;
1455 1459 cv_signal(&arg->cv);
1456 1460 mutex_exit(&arg->lock);
1457 1461 }
1458 1462
1459 1463 static void
1460 1464 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1461 1465 {
1462 1466 struct refsarg arg;
1463 1467
1464 1468 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1465 1469 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1466 1470 arg.gone = FALSE;
1467 1471 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1468 1472 dsl_dataset_refs_gone);
1469 1473 dmu_buf_rele(ds->ds_dbuf, tag);
1470 1474 mutex_enter(&arg.lock);
1471 1475 while (!arg.gone)
1472 1476 cv_wait(&arg.cv, &arg.lock);
1473 1477 ASSERT(arg.gone);
1474 1478 mutex_exit(&arg.lock);
1475 1479 ds->ds_dbuf = NULL;
1476 1480 ds->ds_phys = NULL;
1477 1481 mutex_destroy(&arg.lock);
1478 1482 cv_destroy(&arg.cv);
1479 1483 }
1480 1484
1481 1485 static void
1482 1486 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1483 1487 {
1484 1488 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1485 1489 uint64_t count;
1486 1490 int err;
1487 1491
1488 1492 ASSERT(ds->ds_phys->ds_num_children >= 2);
1489 1493 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1490 1494 /*
1491 1495 * The err should not be ENOENT, but a bug in a previous version
1492 1496 * of the code could cause upgrade_clones_cb() to not set
1493 1497 * ds_next_snap_obj when it should, leading to a missing entry.
1494 1498 * If we knew that the pool was created after
1495 1499 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1496 1500 * ENOENT. However, at least we can check that we don't have
1497 1501 * too many entries in the next_clones_obj even after failing to
1498 1502 * remove this one.
1499 1503 */
1500 1504 if (err != ENOENT) {
1501 1505 VERIFY0(err);
1502 1506 }
1503 1507 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1504 1508 &count));
1505 1509 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1506 1510 }
1507 1511
1508 1512 static void
1509 1513 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1510 1514 {
1511 1515 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1512 1516 zap_cursor_t zc;
1513 1517 zap_attribute_t za;
1514 1518
1515 1519 /*
1516 1520 * If it is the old version, dd_clones doesn't exist so we can't
1517 1521 * find the clones, but deadlist_remove_key() is a no-op so it
1518 1522 * doesn't matter.
1519 1523 */
1520 1524 if (ds->ds_dir->dd_phys->dd_clones == 0)
1521 1525 return;
1522 1526
1523 1527 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1524 1528 zap_cursor_retrieve(&zc, &za) == 0;
1525 1529 zap_cursor_advance(&zc)) {
1526 1530 dsl_dataset_t *clone;
1527 1531
1528 1532 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1529 1533 za.za_first_integer, FTAG, &clone));
1530 1534 if (clone->ds_dir->dd_origin_txg > mintxg) {
1531 1535 dsl_deadlist_remove_key(&clone->ds_deadlist,
1532 1536 mintxg, tx);
1533 1537 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1534 1538 }
1535 1539 dsl_dataset_rele(clone, FTAG);
1536 1540 }
1537 1541 zap_cursor_fini(&zc);
1538 1542 }
1539 1543
1540 1544 struct process_old_arg {
1541 1545 dsl_dataset_t *ds;
1542 1546 dsl_dataset_t *ds_prev;
1543 1547 boolean_t after_branch_point;
1544 1548 zio_t *pio;
1545 1549 uint64_t used, comp, uncomp;
1546 1550 };
1547 1551
1548 1552 static int
1549 1553 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1550 1554 {
1551 1555 struct process_old_arg *poa = arg;
1552 1556 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1553 1557
1554 1558 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1555 1559 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1556 1560 if (poa->ds_prev && !poa->after_branch_point &&
1557 1561 bp->blk_birth >
1558 1562 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1559 1563 poa->ds_prev->ds_phys->ds_unique_bytes +=
1560 1564 bp_get_dsize_sync(dp->dp_spa, bp);
1561 1565 }
1562 1566 } else {
1563 1567 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1564 1568 poa->comp += BP_GET_PSIZE(bp);
1565 1569 poa->uncomp += BP_GET_UCSIZE(bp);
1566 1570 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1567 1571 }
1568 1572 return (0);
1569 1573 }
1570 1574
1571 1575 static void
1572 1576 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1573 1577 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1574 1578 {
1575 1579 struct process_old_arg poa = { 0 };
1576 1580 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1577 1581 objset_t *mos = dp->dp_meta_objset;
1578 1582
1579 1583 ASSERT(ds->ds_deadlist.dl_oldfmt);
1580 1584 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1581 1585
1582 1586 poa.ds = ds;
1583 1587 poa.ds_prev = ds_prev;
1584 1588 poa.after_branch_point = after_branch_point;
1585 1589 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1586 1590 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1587 1591 process_old_cb, &poa, tx));
1588 1592 VERIFY0(zio_wait(poa.pio));
1589 1593 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1590 1594
1591 1595 /* change snapused */
1592 1596 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1593 1597 -poa.used, -poa.comp, -poa.uncomp, tx);
1594 1598
1595 1599 /* swap next's deadlist to our deadlist */
1596 1600 dsl_deadlist_close(&ds->ds_deadlist);
1597 1601 dsl_deadlist_close(&ds_next->ds_deadlist);
1598 1602 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1599 1603 ds->ds_phys->ds_deadlist_obj);
1600 1604 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1601 1605 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1602 1606 ds_next->ds_phys->ds_deadlist_obj);
1603 1607 }
1604 1608
1605 1609 static int
1606 1610 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1607 1611 {
1608 1612 int err;
1609 1613 struct killarg ka;
1610 1614
1611 1615 /*
1612 1616 * Free everything that we point to (that's born after
1613 1617 * the previous snapshot, if we are a clone)
1614 1618 *
1615 1619 * NB: this should be very quick, because we already
1616 1620 * freed all the objects in open context.
1617 1621 */
1618 1622 ka.ds = ds;
1619 1623 ka.tx = tx;
1620 1624 err = traverse_dataset(ds,
1621 1625 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1622 1626 kill_blkptr, &ka);
1623 1627 ASSERT0(err);
1624 1628 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1625 1629
1626 1630 return (err);
1627 1631 }
1628 1632
1629 1633 void
1630 1634 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1631 1635 {
1632 1636 struct dsl_ds_destroyarg *dsda = arg1;
1633 1637 dsl_dataset_t *ds = dsda->ds;
1634 1638 int err;
1635 1639 int after_branch_point = FALSE;
1636 1640 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1637 1641 objset_t *mos = dp->dp_meta_objset;
1638 1642 dsl_dataset_t *ds_prev = NULL;
1639 1643 boolean_t wont_destroy;
1640 1644 uint64_t obj;
1641 1645
1642 1646 wont_destroy = (dsda->defer &&
1643 1647 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1644 1648
1645 1649 ASSERT(ds->ds_owner || wont_destroy);
1646 1650 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1647 1651 ASSERT(ds->ds_prev == NULL ||
1648 1652 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1649 1653 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1650 1654
1651 1655 if (wont_destroy) {
1652 1656 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1653 1657 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1654 1658 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1655 1659 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
1656 1660 return;
1657 1661 }
1658 1662
1659 1663 /* We need to log before removing it from the namespace. */
1660 1664 spa_history_log_internal_ds(ds, "destroy", tx, "");
1661 1665
1662 1666 /* signal any waiters that this dataset is going away */
1663 1667 mutex_enter(&ds->ds_lock);
1664 1668 ds->ds_owner = dsl_reaper;
1665 1669 cv_broadcast(&ds->ds_exclusive_cv);
1666 1670 mutex_exit(&ds->ds_lock);
1667 1671
1668 1672 /* Remove our reservation */
1669 1673 if (ds->ds_reserved != 0) {
1670 1674 dsl_prop_setarg_t psa;
1671 1675 uint64_t value = 0;
1672 1676
1673 1677 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1674 1678 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1675 1679 &value);
1676 1680 psa.psa_effective_value = 0; /* predict default value */
1677 1681
1678 1682 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1679 1683 ASSERT0(ds->ds_reserved);
1680 1684 }
1681 1685
1682 1686 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1683 1687
1684 1688 dsl_scan_ds_destroyed(ds, tx);
1685 1689
1686 1690 obj = ds->ds_object;
1687 1691
1688 1692 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1689 1693 if (ds->ds_prev) {
1690 1694 ds_prev = ds->ds_prev;
1691 1695 } else {
1692 1696 VERIFY(0 == dsl_dataset_hold_obj(dp,
1693 1697 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1694 1698 }
1695 1699 after_branch_point =
1696 1700 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1697 1701
1698 1702 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1699 1703 if (after_branch_point &&
1700 1704 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1701 1705 remove_from_next_clones(ds_prev, obj, tx);
1702 1706 if (ds->ds_phys->ds_next_snap_obj != 0) {
1703 1707 VERIFY(0 == zap_add_int(mos,
1704 1708 ds_prev->ds_phys->ds_next_clones_obj,
1705 1709 ds->ds_phys->ds_next_snap_obj, tx));
1706 1710 }
1707 1711 }
1708 1712 if (after_branch_point &&
1709 1713 ds->ds_phys->ds_next_snap_obj == 0) {
1710 1714 /* This clone is toast. */
1711 1715 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1712 1716 ds_prev->ds_phys->ds_num_children--;
1713 1717
1714 1718 /*
1715 1719 * If the clone's origin has no other clones, no
1716 1720 * user holds, and has been marked for deferred
1717 1721 * deletion, then we should have done the necessary
1718 1722 * destroy setup for it.
1719 1723 */
1720 1724 if (ds_prev->ds_phys->ds_num_children == 1 &&
1721 1725 ds_prev->ds_userrefs == 0 &&
1722 1726 DS_IS_DEFER_DESTROY(ds_prev)) {
1723 1727 ASSERT3P(dsda->rm_origin, !=, NULL);
1724 1728 } else {
1725 1729 ASSERT3P(dsda->rm_origin, ==, NULL);
1726 1730 }
1727 1731 } else if (!after_branch_point) {
1728 1732 ds_prev->ds_phys->ds_next_snap_obj =
1729 1733 ds->ds_phys->ds_next_snap_obj;
1730 1734 }
1731 1735 }
1732 1736
1733 1737 if (dsl_dataset_is_snapshot(ds)) {
1734 1738 dsl_dataset_t *ds_next;
1735 1739 uint64_t old_unique;
1736 1740 uint64_t used = 0, comp = 0, uncomp = 0;
1737 1741
1738 1742 VERIFY(0 == dsl_dataset_hold_obj(dp,
1739 1743 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1740 1744 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1741 1745
1742 1746 old_unique = ds_next->ds_phys->ds_unique_bytes;
1743 1747
1744 1748 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1745 1749 ds_next->ds_phys->ds_prev_snap_obj =
1746 1750 ds->ds_phys->ds_prev_snap_obj;
1747 1751 ds_next->ds_phys->ds_prev_snap_txg =
1748 1752 ds->ds_phys->ds_prev_snap_txg;
1749 1753 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1750 1754 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1751 1755
1752 1756
1753 1757 if (ds_next->ds_deadlist.dl_oldfmt) {
1754 1758 process_old_deadlist(ds, ds_prev, ds_next,
1755 1759 after_branch_point, tx);
1756 1760 } else {
1757 1761 /* Adjust prev's unique space. */
1758 1762 if (ds_prev && !after_branch_point) {
1759 1763 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1760 1764 ds_prev->ds_phys->ds_prev_snap_txg,
1761 1765 ds->ds_phys->ds_prev_snap_txg,
1762 1766 &used, &comp, &uncomp);
1763 1767 ds_prev->ds_phys->ds_unique_bytes += used;
1764 1768 }
1765 1769
1766 1770 /* Adjust snapused. */
1767 1771 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1768 1772 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1769 1773 &used, &comp, &uncomp);
1770 1774 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1771 1775 -used, -comp, -uncomp, tx);
1772 1776
1773 1777 /* Move blocks to be freed to pool's free list. */
1774 1778 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1775 1779 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1776 1780 tx);
1777 1781 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1778 1782 DD_USED_HEAD, used, comp, uncomp, tx);
1779 1783
1780 1784 /* Merge our deadlist into next's and free it. */
1781 1785 dsl_deadlist_merge(&ds_next->ds_deadlist,
1782 1786 ds->ds_phys->ds_deadlist_obj, tx);
1783 1787 }
1784 1788 dsl_deadlist_close(&ds->ds_deadlist);
1785 1789 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1786 1790
1787 1791 /* Collapse range in clone heads */
1788 1792 dsl_dataset_remove_clones_key(ds,
1789 1793 ds->ds_phys->ds_creation_txg, tx);
1790 1794
1791 1795 if (dsl_dataset_is_snapshot(ds_next)) {
1792 1796 dsl_dataset_t *ds_nextnext;
1793 1797
1794 1798 /*
1795 1799 * Update next's unique to include blocks which
1796 1800 * were previously shared by only this snapshot
1797 1801 * and it. Those blocks will be born after the
1798 1802 * prev snap and before this snap, and will have
1799 1803 * died after the next snap and before the one
1800 1804 * after that (ie. be on the snap after next's
1801 1805 * deadlist).
1802 1806 */
1803 1807 VERIFY(0 == dsl_dataset_hold_obj(dp,
1804 1808 ds_next->ds_phys->ds_next_snap_obj,
1805 1809 FTAG, &ds_nextnext));
1806 1810 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1807 1811 ds->ds_phys->ds_prev_snap_txg,
1808 1812 ds->ds_phys->ds_creation_txg,
1809 1813 &used, &comp, &uncomp);
1810 1814 ds_next->ds_phys->ds_unique_bytes += used;
1811 1815 dsl_dataset_rele(ds_nextnext, FTAG);
1812 1816 ASSERT3P(ds_next->ds_prev, ==, NULL);
1813 1817
1814 1818 /* Collapse range in this head. */
1815 1819 dsl_dataset_t *hds;
1816 1820 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1817 1821 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1818 1822 FTAG, &hds));
1819 1823 dsl_deadlist_remove_key(&hds->ds_deadlist,
1820 1824 ds->ds_phys->ds_creation_txg, tx);
1821 1825 dsl_dataset_rele(hds, FTAG);
1822 1826
1823 1827 } else {
1824 1828 ASSERT3P(ds_next->ds_prev, ==, ds);
1825 1829 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1826 1830 ds_next->ds_prev = NULL;
1827 1831 if (ds_prev) {
1828 1832 VERIFY(0 == dsl_dataset_get_ref(dp,
1829 1833 ds->ds_phys->ds_prev_snap_obj,
1830 1834 ds_next, &ds_next->ds_prev));
1831 1835 }
1832 1836
1833 1837 dsl_dataset_recalc_head_uniq(ds_next);
1834 1838
1835 1839 /*
1836 1840 * Reduce the amount of our unconsmed refreservation
1837 1841 * being charged to our parent by the amount of
1838 1842 * new unique data we have gained.
1839 1843 */
1840 1844 if (old_unique < ds_next->ds_reserved) {
1841 1845 int64_t mrsdelta;
1842 1846 uint64_t new_unique =
1843 1847 ds_next->ds_phys->ds_unique_bytes;
1844 1848
1845 1849 ASSERT(old_unique <= new_unique);
1846 1850 mrsdelta = MIN(new_unique - old_unique,
1847 1851 ds_next->ds_reserved - old_unique);
1848 1852 dsl_dir_diduse_space(ds->ds_dir,
1849 1853 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1850 1854 }
1851 1855 }
1852 1856 dsl_dataset_rele(ds_next, FTAG);
1853 1857 } else {
1854 1858 zfeature_info_t *async_destroy =
1855 1859 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1856 1860 objset_t *os;
1857 1861
1858 1862 /*
1859 1863 * There's no next snapshot, so this is a head dataset.
1860 1864 * Destroy the deadlist. Unless it's a clone, the
1861 1865 * deadlist should be empty. (If it's a clone, it's
1862 1866 * safe to ignore the deadlist contents.)
1863 1867 */
1864 1868 dsl_deadlist_close(&ds->ds_deadlist);
1865 1869 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1866 1870 ds->ds_phys->ds_deadlist_obj = 0;
1867 1871
1868 1872 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
1869 1873
1870 1874 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1871 1875 err = old_synchronous_dataset_destroy(ds, tx);
1872 1876 } else {
1873 1877 /*
1874 1878 * Move the bptree into the pool's list of trees to
1875 1879 * clean up and update space accounting information.
1876 1880 */
1877 1881 uint64_t used, comp, uncomp;
1878 1882
1879 1883 zil_destroy_sync(dmu_objset_zil(os), tx);
1880 1884
1881 1885 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1882 1886 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1883 1887 dp->dp_bptree_obj = bptree_alloc(mos, tx);
1884 1888 VERIFY(zap_add(mos,
1885 1889 DMU_POOL_DIRECTORY_OBJECT,
1886 1890 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1887 1891 &dp->dp_bptree_obj, tx) == 0);
1888 1892 }
1889 1893
1890 1894 used = ds->ds_dir->dd_phys->dd_used_bytes;
1891 1895 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1892 1896 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1893 1897
1894 1898 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1895 1899 ds->ds_phys->ds_unique_bytes == used);
1896 1900
1897 1901 bptree_add(mos, dp->dp_bptree_obj,
1898 1902 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1899 1903 used, comp, uncomp, tx);
1900 1904 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1901 1905 -used, -comp, -uncomp, tx);
1902 1906 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1903 1907 used, comp, uncomp, tx);
1904 1908 }
1905 1909
1906 1910 if (ds->ds_prev != NULL) {
1907 1911 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1908 1912 VERIFY3U(0, ==, zap_remove_int(mos,
1909 1913 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1910 1914 ds->ds_object, tx));
1911 1915 }
1912 1916 dsl_dataset_rele(ds->ds_prev, ds);
1913 1917 ds->ds_prev = ds_prev = NULL;
1914 1918 }
1915 1919 }
1916 1920
1917 1921 /*
1918 1922 * This must be done after the dsl_traverse(), because it will
1919 1923 * re-open the objset.
1920 1924 */
1921 1925 if (ds->ds_objset) {
1922 1926 dmu_objset_evict(ds->ds_objset);
1923 1927 ds->ds_objset = NULL;
1924 1928 }
1925 1929
1926 1930 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1927 1931 /* Erase the link in the dir */
1928 1932 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1929 1933 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1930 1934 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1931 1935 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1932 1936 ASSERT(err == 0);
1933 1937 } else {
1934 1938 /* remove from snapshot namespace */
1935 1939 dsl_dataset_t *ds_head;
1936 1940 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1937 1941 VERIFY(0 == dsl_dataset_hold_obj(dp,
1938 1942 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1939 1943 VERIFY(0 == dsl_dataset_get_snapname(ds));
1940 1944 #ifdef ZFS_DEBUG
1941 1945 {
1942 1946 uint64_t val;
1943 1947
1944 1948 err = dsl_dataset_snap_lookup(ds_head,
1945 1949 ds->ds_snapname, &val);
1946 1950 ASSERT0(err);
1947 1951 ASSERT3U(val, ==, obj);
1948 1952 }
1949 1953 #endif
1950 1954 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1951 1955 ASSERT(err == 0);
1952 1956 dsl_dataset_rele(ds_head, FTAG);
1953 1957 }
1954 1958
1955 1959 if (ds_prev && ds->ds_prev != ds_prev)
1956 1960 dsl_dataset_rele(ds_prev, FTAG);
1957 1961
1958 1962 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1959 1963
1960 1964 if (ds->ds_phys->ds_next_clones_obj != 0) {
1961 1965 uint64_t count;
1962 1966 ASSERT(0 == zap_count(mos,
1963 1967 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1964 1968 VERIFY(0 == dmu_object_free(mos,
1965 1969 ds->ds_phys->ds_next_clones_obj, tx));
1966 1970 }
1967 1971 if (ds->ds_phys->ds_props_obj != 0)
1968 1972 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1969 1973 if (ds->ds_phys->ds_userrefs_obj != 0)
1970 1974 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1971 1975 dsl_dir_close(ds->ds_dir, ds);
1972 1976 ds->ds_dir = NULL;
1973 1977 dsl_dataset_drain_refs(ds, tag);
1974 1978 VERIFY(0 == dmu_object_free(mos, obj, tx));
1975 1979
1976 1980 if (dsda->rm_origin) {
1977 1981 /*
1978 1982 * Remove the origin of the clone we just destroyed.
1979 1983 */
1980 1984 struct dsl_ds_destroyarg ndsda = {0};
1981 1985
1982 1986 ndsda.ds = dsda->rm_origin;
1983 1987 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1984 1988 }
1985 1989 }
1986 1990
1987 1991 static int
1988 1992 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1989 1993 {
1990 1994 uint64_t asize;
1991 1995
1992 1996 if (!dmu_tx_is_syncing(tx))
1993 1997 return (0);
1994 1998
1995 1999 /*
1996 2000 * If there's an fs-only reservation, any blocks that might become
1997 2001 * owned by the snapshot dataset must be accommodated by space
1998 2002 * outside of the reservation.
1999 2003 */
2000 2004 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2001 2005 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2002 2006 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2003 2007 return (ENOSPC);
2004 2008
|
↓ open down ↓ |
855 lines elided |
↑ open up ↑ |
2005 2009 /*
2006 2010 * Propagate any reserved space for this snapshot to other
2007 2011 * snapshot checks in this sync group.
2008 2012 */
2009 2013 if (asize > 0)
2010 2014 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2011 2015
2012 2016 return (0);
2013 2017 }
2014 2018
2019 +/*
2020 + * Check if adding additional snapshot(s) would exceed any snapshot quotas.
2021 + * Note that all snapshot quotas up to the root dataset (i.e. the pool itself)
2022 + * or the given ancestor must be satisfied. Note that it is valid for the
2023 + * count to exceed the quota. This can happen if a recursive snapshot is taken
2024 + * from a dataset above this one.
2025 + */
2015 2026 int
2027 +dsl_snapcount_check(dsl_dir_t *dd, dmu_tx_t *tx, uint64_t cnt,
2028 + dsl_dir_t *ancestor)
2029 +{
2030 + uint64_t quota;
2031 + int err = 0;
2032 +
2033 + /*
2034 + * As with dsl_dataset_set_reservation_check(), don't run this check in
2035 + * open context.
2036 + */
2037 + if (!dmu_tx_is_syncing(tx))
2038 + return (0);
2039 +
2040 + /*
2041 + * If renaming a dataset with no snapshots, count adjustment is 0.
2042 + * Likewise when taking a recursive snapshot below the top-level (see
2043 + * the comment in snapshot_check() for more details).
2044 + */
2045 + if (cnt == 0)
2046 + return (0);
2047 +
2048 + /*
2049 + * If an ancestor has been provided, stop checking the quota once we
2050 + * hit that dir. We need this during rename so that we don't overcount
2051 + * the check once we recurse up to the common ancestor.
2052 + */
2053 + if (ancestor == dd)
2054 + return (0);
2055 +
2056 + /*
2057 + * If there's no value for this property, there's no need to enforce a
2058 + * snapshot quota.
2059 + */
2060 + err = dsl_prop_get_dd(dd, zfs_prop_to_name(ZFS_PROP_SNAPSHOT_QUOTA),
2061 + 8, 1, "a, NULL, B_FALSE);
2062 + if (err == ENOENT)
2063 + return (0);
2064 + else if (err != 0)
2065 + return (err);
2066 +
2067 +#ifdef _KERNEL
2068 + extern void __dtrace_probe_zfs__ss__quota(uint64_t, uint64_t, char *);
2069 + __dtrace_probe_zfs__ss__quota(
2070 + (uint64_t)dd->dd_phys->dd_snapshot_count, (uint64_t)quota,
2071 + dd->dd_myname);
2072 +#endif
2073 +
2074 + if (quota > 0 && (dd->dd_phys->dd_snapshot_count + cnt) > quota)
2075 + return (EDQUOT);
2076 +
2077 + if (dd->dd_parent != NULL)
2078 + err = dsl_snapcount_check(dd->dd_parent, tx, cnt, ancestor);
2079 +
2080 + return (err);
2081 +}
2082 +
2083 +/*
2084 + * Adjust the snapshot count for the specified dsl_dir_t and all parents.
2085 + * When a new snapshot is created, increment the count on all parents, and when
2086 + * a snapshot is destroyed, decrement the count.
2087 + */
2088 +void
2089 +dsl_snapcount_adjust(dsl_dir_t *dd, dmu_tx_t *tx, int64_t delta,
2090 + boolean_t first)
2091 +{
2092 + /*
2093 + * On initial entry we need to check if this feature is active, but
2094 + * we don't want to re-check this on each recursive call. Note: the
2095 + * feature cannot be active if its not enabled. If the feature is not
2096 + * active, don't touch the on-disk count fields.
2097 + */
2098 + if (first) {
2099 + dsl_dataset_t *ds = NULL;
2100 + spa_t *spa;
2101 + zfeature_info_t *quota_feat =
2102 + &spa_feature_table[SPA_FEATURE_DS_SS_QUOTA];
2103 +
2104 + VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2105 + dd->dd_phys->dd_head_dataset_obj, FTAG, &ds));
2106 + spa = dsl_dataset_get_spa(ds);
2107 + dsl_dataset_rele(ds, FTAG);
2108 + if (!spa_feature_is_active(spa, quota_feat))
2109 + return;
2110 + }
2111 +
2112 + /*
2113 + * As with dsl_dataset_set_reservation_check(), wdon't want to run
2114 + * this check in open context.
2115 + */
2116 + if (!dmu_tx_is_syncing(tx))
2117 + return;
2118 +
2119 + /* if renaming a dataset with no snapshots, count adjustment is 0 */
2120 + if (delta == 0)
2121 + return;
2122 +
2123 + /* Increment count for parent */
2124 + dmu_buf_will_dirty(dd->dd_dbuf, tx);
2125 +
2126 + mutex_enter(&dd->dd_lock);
2127 +
2128 + /*
2129 + * Counts may be incorrect if dealing with an existing pool and
2130 + * there has never been a quota set in the dataset hierarchy.
2131 + * This is not an error.
2132 + */
2133 + if (delta < 0 && dd->dd_phys->dd_snapshot_count < (delta * -1)) {
2134 +#ifdef _KERNEL
2135 + extern void __dtrace_probe_zfs__sscnt__adj__neg(char *);
2136 + __dtrace_probe_zfs__sscnt__adj__neg(dd->dd_myname);
2137 +#endif
2138 + mutex_exit(&dd->dd_lock);
2139 + return;
2140 + }
2141 +
2142 + dd->dd_phys->dd_snapshot_count += delta;
2143 +
2144 + /* Roll up this additional count into our ancestors */
2145 +
2146 + if (dd->dd_parent != NULL)
2147 + dsl_snapcount_adjust(dd->dd_parent, tx, delta, B_FALSE);
2148 +
2149 + mutex_exit(&dd->dd_lock);
2150 +}
2151 +
2152 +int
2016 2153 dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname,
2017 - dmu_tx_t *tx)
2154 + uint64_t cnt, dmu_tx_t *tx)
2018 2155 {
2019 2156 int err;
2020 2157 uint64_t value;
2021 2158
2022 2159 /*
2023 2160 * We don't allow multiple snapshots of the same txg. If there
2024 2161 * is already one, try again.
2025 2162 */
2026 2163 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2027 2164 return (EAGAIN);
2028 2165
2029 2166 /*
2030 2167 * Check for conflicting snapshot name.
2031 2168 */
2032 2169 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2033 2170 if (err == 0)
2034 2171 return (EEXIST);
|
↓ open down ↓ |
7 lines elided |
↑ open up ↑ |
2035 2172 if (err != ENOENT)
2036 2173 return (err);
2037 2174
2038 2175 /*
2039 2176 * Check that the dataset's name is not too long. Name consists
2040 2177 * of the dataset's length + 1 for the @-sign + snapshot name's length
2041 2178 */
2042 2179 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2043 2180 return (ENAMETOOLONG);
2044 2181
2182 + err = dsl_snapcount_check(ds->ds_dir, tx, cnt, NULL);
2183 + if (err)
2184 + return (err);
2185 +
2045 2186 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2046 2187 if (err)
2047 2188 return (err);
2048 2189
2049 2190 ds->ds_trysnap_txg = tx->tx_txg;
2050 2191 return (0);
2051 2192 }
2052 2193
2053 2194 void
2054 2195 dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname,
2055 2196 dmu_tx_t *tx)
|
↓ open down ↓ |
1 lines elided |
↑ open up ↑ |
2056 2197 {
2057 2198 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2058 2199 dmu_buf_t *dbuf;
2059 2200 dsl_dataset_phys_t *dsphys;
2060 2201 uint64_t dsobj, crtxg;
2061 2202 objset_t *mos = dp->dp_meta_objset;
2062 2203 int err;
2063 2204
2064 2205 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2065 2206
2207 + dsl_snapcount_adjust(ds->ds_dir, tx, 1, B_TRUE);
2208 +
2066 2209 /*
2067 2210 * The origin's ds_creation_txg has to be < TXG_INITIAL
2068 2211 */
2069 2212 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2070 2213 crtxg = 1;
2071 2214 else
2072 2215 crtxg = tx->tx_txg;
2073 2216
2074 2217 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2075 2218 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2076 2219 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2077 2220 dmu_buf_will_dirty(dbuf, tx);
2078 2221 dsphys = dbuf->db_data;
2079 2222 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2080 2223 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2081 2224 dsphys->ds_fsid_guid = unique_create();
2082 2225 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2083 2226 sizeof (dsphys->ds_guid));
2084 2227 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2085 2228 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2086 2229 dsphys->ds_next_snap_obj = ds->ds_object;
2087 2230 dsphys->ds_num_children = 1;
2088 2231 dsphys->ds_creation_time = gethrestime_sec();
2089 2232 dsphys->ds_creation_txg = crtxg;
2090 2233 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2091 2234 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2092 2235 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2093 2236 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2094 2237 dsphys->ds_flags = ds->ds_phys->ds_flags;
2095 2238 dsphys->ds_bp = ds->ds_phys->ds_bp;
2096 2239 dmu_buf_rele(dbuf, FTAG);
2097 2240
2098 2241 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2099 2242 if (ds->ds_prev) {
2100 2243 uint64_t next_clones_obj =
2101 2244 ds->ds_prev->ds_phys->ds_next_clones_obj;
2102 2245 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2103 2246 ds->ds_object ||
2104 2247 ds->ds_prev->ds_phys->ds_num_children > 1);
2105 2248 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2106 2249 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2107 2250 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2108 2251 ds->ds_prev->ds_phys->ds_creation_txg);
2109 2252 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2110 2253 } else if (next_clones_obj != 0) {
2111 2254 remove_from_next_clones(ds->ds_prev,
2112 2255 dsphys->ds_next_snap_obj, tx);
2113 2256 VERIFY3U(0, ==, zap_add_int(mos,
2114 2257 next_clones_obj, dsobj, tx));
2115 2258 }
2116 2259 }
2117 2260
2118 2261 /*
2119 2262 * If we have a reference-reservation on this dataset, we will
2120 2263 * need to increase the amount of refreservation being charged
2121 2264 * since our unique space is going to zero.
2122 2265 */
2123 2266 if (ds->ds_reserved) {
2124 2267 int64_t delta;
2125 2268 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2126 2269 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2127 2270 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2128 2271 delta, 0, 0, tx);
2129 2272 }
2130 2273
2131 2274 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2132 2275 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2133 2276 ds->ds_dir->dd_myname, snapname, dsobj,
2134 2277 ds->ds_phys->ds_prev_snap_txg);
2135 2278 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2136 2279 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2137 2280 dsl_deadlist_close(&ds->ds_deadlist);
2138 2281 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2139 2282 dsl_deadlist_add_key(&ds->ds_deadlist,
2140 2283 ds->ds_phys->ds_prev_snap_txg, tx);
2141 2284
2142 2285 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2143 2286 ds->ds_phys->ds_prev_snap_obj = dsobj;
2144 2287 ds->ds_phys->ds_prev_snap_txg = crtxg;
2145 2288 ds->ds_phys->ds_unique_bytes = 0;
2146 2289 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2147 2290 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2148 2291
2149 2292 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2150 2293 snapname, 8, 1, &dsobj, tx);
2151 2294 ASSERT(err == 0);
2152 2295
2153 2296 if (ds->ds_prev)
2154 2297 dsl_dataset_drop_ref(ds->ds_prev, ds);
2155 2298 VERIFY(0 == dsl_dataset_get_ref(dp,
2156 2299 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2157 2300
2158 2301 dsl_scan_ds_snapshotted(ds, tx);
2159 2302
2160 2303 dsl_dir_snap_cmtime_update(ds->ds_dir);
2161 2304
2162 2305 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
2163 2306 }
2164 2307
2165 2308 void
2166 2309 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2167 2310 {
2168 2311 ASSERT(dmu_tx_is_syncing(tx));
2169 2312 ASSERT(ds->ds_objset != NULL);
2170 2313 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2171 2314
2172 2315 /*
2173 2316 * in case we had to change ds_fsid_guid when we opened it,
2174 2317 * sync it out now.
2175 2318 */
2176 2319 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2177 2320 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2178 2321
2179 2322 dmu_objset_sync(ds->ds_objset, zio, tx);
2180 2323 }
2181 2324
2182 2325 static void
2183 2326 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2184 2327 {
2185 2328 uint64_t count = 0;
2186 2329 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2187 2330 zap_cursor_t zc;
2188 2331 zap_attribute_t za;
2189 2332 nvlist_t *propval;
2190 2333 nvlist_t *val;
2191 2334
2192 2335 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2193 2336 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2194 2337 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2195 2338
2196 2339 /*
2197 2340 * There may me missing entries in ds_next_clones_obj
2198 2341 * due to a bug in a previous version of the code.
2199 2342 * Only trust it if it has the right number of entries.
2200 2343 */
2201 2344 if (ds->ds_phys->ds_next_clones_obj != 0) {
2202 2345 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2203 2346 &count));
2204 2347 }
2205 2348 if (count != ds->ds_phys->ds_num_children - 1) {
2206 2349 goto fail;
2207 2350 }
2208 2351 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2209 2352 zap_cursor_retrieve(&zc, &za) == 0;
2210 2353 zap_cursor_advance(&zc)) {
2211 2354 dsl_dataset_t *clone;
2212 2355 char buf[ZFS_MAXNAMELEN];
2213 2356 /*
2214 2357 * Even though we hold the dp_config_rwlock, the dataset
2215 2358 * may fail to open, returning ENOENT. If there is a
2216 2359 * thread concurrently attempting to destroy this
2217 2360 * dataset, it will have the ds_rwlock held for
2218 2361 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2219 2362 * dsl_dataset_hold_ref() will fail its
2220 2363 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2221 2364 * dp_config_rwlock, and wait for the destroy progress
2222 2365 * and signal ds_exclusive_cv. If the destroy was
2223 2366 * successful, we will see that
2224 2367 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2225 2368 */
2226 2369 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2227 2370 za.za_first_integer, FTAG, &clone) != 0)
2228 2371 continue;
2229 2372 dsl_dir_name(clone->ds_dir, buf);
2230 2373 VERIFY(nvlist_add_boolean(val, buf) == 0);
2231 2374 dsl_dataset_rele(clone, FTAG);
2232 2375 }
2233 2376 zap_cursor_fini(&zc);
2234 2377 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2235 2378 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2236 2379 propval) == 0);
2237 2380 fail:
2238 2381 nvlist_free(val);
2239 2382 nvlist_free(propval);
2240 2383 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2241 2384 }
2242 2385
2243 2386 void
2244 2387 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2245 2388 {
2246 2389 uint64_t refd, avail, uobjs, aobjs, ratio;
2247 2390
2248 2391 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2249 2392 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2250 2393 ds->ds_phys->ds_compressed_bytes);
2251 2394
2252 2395 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2253 2396
2254 2397 if (dsl_dataset_is_snapshot(ds)) {
2255 2398 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2256 2399 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2257 2400 ds->ds_phys->ds_unique_bytes);
2258 2401 get_clones_stat(ds, nv);
2259 2402 } else {
2260 2403 dsl_dir_stats(ds->ds_dir, nv);
2261 2404 }
2262 2405
2263 2406 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2264 2407 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2265 2408 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2266 2409
2267 2410 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2268 2411 ds->ds_phys->ds_creation_time);
2269 2412 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2270 2413 ds->ds_phys->ds_creation_txg);
2271 2414 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2272 2415 ds->ds_quota);
2273 2416 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2274 2417 ds->ds_reserved);
2275 2418 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2276 2419 ds->ds_phys->ds_guid);
2277 2420 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2278 2421 ds->ds_phys->ds_unique_bytes);
2279 2422 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2280 2423 ds->ds_object);
2281 2424 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2282 2425 ds->ds_userrefs);
2283 2426 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2284 2427 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2285 2428
2286 2429 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2287 2430 uint64_t written, comp, uncomp;
2288 2431 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2289 2432 dsl_dataset_t *prev;
2290 2433
2291 2434 rw_enter(&dp->dp_config_rwlock, RW_READER);
2292 2435 int err = dsl_dataset_hold_obj(dp,
2293 2436 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2294 2437 rw_exit(&dp->dp_config_rwlock);
2295 2438 if (err == 0) {
2296 2439 err = dsl_dataset_space_written(prev, ds, &written,
2297 2440 &comp, &uncomp);
2298 2441 dsl_dataset_rele(prev, FTAG);
2299 2442 if (err == 0) {
2300 2443 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2301 2444 written);
2302 2445 }
2303 2446 }
2304 2447 }
2305 2448 }
2306 2449
2307 2450 void
2308 2451 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2309 2452 {
2310 2453 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2311 2454 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2312 2455 stat->dds_guid = ds->ds_phys->ds_guid;
2313 2456 stat->dds_origin[0] = '\0';
2314 2457 if (dsl_dataset_is_snapshot(ds)) {
2315 2458 stat->dds_is_snapshot = B_TRUE;
2316 2459 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2317 2460 } else {
2318 2461 stat->dds_is_snapshot = B_FALSE;
2319 2462 stat->dds_num_clones = 0;
2320 2463
2321 2464 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2322 2465 if (dsl_dir_is_clone(ds->ds_dir)) {
2323 2466 dsl_dataset_t *ods;
2324 2467
2325 2468 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2326 2469 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2327 2470 dsl_dataset_name(ods, stat->dds_origin);
2328 2471 dsl_dataset_drop_ref(ods, FTAG);
2329 2472 }
2330 2473 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2331 2474 }
2332 2475 }
2333 2476
2334 2477 uint64_t
2335 2478 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2336 2479 {
2337 2480 return (ds->ds_fsid_guid);
2338 2481 }
2339 2482
2340 2483 void
2341 2484 dsl_dataset_space(dsl_dataset_t *ds,
2342 2485 uint64_t *refdbytesp, uint64_t *availbytesp,
2343 2486 uint64_t *usedobjsp, uint64_t *availobjsp)
2344 2487 {
2345 2488 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2346 2489 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2347 2490 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2348 2491 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2349 2492 if (ds->ds_quota != 0) {
2350 2493 /*
2351 2494 * Adjust available bytes according to refquota
2352 2495 */
2353 2496 if (*refdbytesp < ds->ds_quota)
2354 2497 *availbytesp = MIN(*availbytesp,
2355 2498 ds->ds_quota - *refdbytesp);
2356 2499 else
2357 2500 *availbytesp = 0;
2358 2501 }
2359 2502 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2360 2503 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2361 2504 }
2362 2505
2363 2506 boolean_t
2364 2507 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2365 2508 {
2366 2509 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2367 2510
2368 2511 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2369 2512 dsl_pool_sync_context(dp));
2370 2513 if (ds->ds_prev == NULL)
2371 2514 return (B_FALSE);
2372 2515 if (ds->ds_phys->ds_bp.blk_birth >
2373 2516 ds->ds_prev->ds_phys->ds_creation_txg) {
2374 2517 objset_t *os, *os_prev;
2375 2518 /*
2376 2519 * It may be that only the ZIL differs, because it was
2377 2520 * reset in the head. Don't count that as being
2378 2521 * modified.
2379 2522 */
2380 2523 if (dmu_objset_from_ds(ds, &os) != 0)
2381 2524 return (B_TRUE);
2382 2525 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2383 2526 return (B_TRUE);
2384 2527 return (bcmp(&os->os_phys->os_meta_dnode,
2385 2528 &os_prev->os_phys->os_meta_dnode,
2386 2529 sizeof (os->os_phys->os_meta_dnode)) != 0);
2387 2530 }
2388 2531 return (B_FALSE);
2389 2532 }
2390 2533
2391 2534 /* ARGSUSED */
2392 2535 static int
2393 2536 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2394 2537 {
2395 2538 dsl_dataset_t *ds = arg1;
2396 2539 char *newsnapname = arg2;
2397 2540 dsl_dir_t *dd = ds->ds_dir;
2398 2541 dsl_dataset_t *hds;
2399 2542 uint64_t val;
2400 2543 int err;
2401 2544
2402 2545 err = dsl_dataset_hold_obj(dd->dd_pool,
2403 2546 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2404 2547 if (err)
2405 2548 return (err);
2406 2549
2407 2550 /* new name better not be in use */
2408 2551 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2409 2552 dsl_dataset_rele(hds, FTAG);
2410 2553
2411 2554 if (err == 0)
2412 2555 err = EEXIST;
2413 2556 else if (err == ENOENT)
2414 2557 err = 0;
2415 2558
2416 2559 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2417 2560 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2418 2561 err = ENAMETOOLONG;
2419 2562
2420 2563 return (err);
2421 2564 }
2422 2565
2423 2566 static void
2424 2567 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2425 2568 {
2426 2569 dsl_dataset_t *ds = arg1;
2427 2570 const char *newsnapname = arg2;
2428 2571 dsl_dir_t *dd = ds->ds_dir;
2429 2572 objset_t *mos = dd->dd_pool->dp_meta_objset;
2430 2573 dsl_dataset_t *hds;
2431 2574 int err;
2432 2575
2433 2576 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2434 2577
2435 2578 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2436 2579 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2437 2580
2438 2581 VERIFY(0 == dsl_dataset_get_snapname(ds));
2439 2582 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2440 2583 ASSERT0(err);
2441 2584 mutex_enter(&ds->ds_lock);
2442 2585 (void) strcpy(ds->ds_snapname, newsnapname);
2443 2586 mutex_exit(&ds->ds_lock);
2444 2587 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2445 2588 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2446 2589 ASSERT0(err);
2447 2590
2448 2591 spa_history_log_internal_ds(ds, "rename", tx,
2449 2592 "-> @%s", newsnapname);
2450 2593 dsl_dataset_rele(hds, FTAG);
2451 2594 }
2452 2595
2453 2596 struct renamesnaparg {
2454 2597 dsl_sync_task_group_t *dstg;
2455 2598 char failed[MAXPATHLEN];
2456 2599 char *oldsnap;
2457 2600 char *newsnap;
2458 2601 };
2459 2602
2460 2603 static int
2461 2604 dsl_snapshot_rename_one(const char *name, void *arg)
2462 2605 {
2463 2606 struct renamesnaparg *ra = arg;
2464 2607 dsl_dataset_t *ds = NULL;
2465 2608 char *snapname;
2466 2609 int err;
2467 2610
2468 2611 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2469 2612 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2470 2613
2471 2614 /*
2472 2615 * For recursive snapshot renames the parent won't be changing
2473 2616 * so we just pass name for both the to/from argument.
2474 2617 */
2475 2618 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2476 2619 if (err != 0) {
2477 2620 strfree(snapname);
2478 2621 return (err == ENOENT ? 0 : err);
2479 2622 }
2480 2623
2481 2624 #ifdef _KERNEL
2482 2625 /*
2483 2626 * For all filesystems undergoing rename, we'll need to unmount it.
2484 2627 */
2485 2628 (void) zfs_unmount_snap(snapname, NULL);
2486 2629 #endif
2487 2630 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2488 2631 strfree(snapname);
2489 2632 if (err != 0)
2490 2633 return (err == ENOENT ? 0 : err);
2491 2634
2492 2635 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2493 2636 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2494 2637
2495 2638 return (0);
2496 2639 }
2497 2640
2498 2641 static int
2499 2642 dsl_recursive_rename(char *oldname, const char *newname)
2500 2643 {
2501 2644 int err;
2502 2645 struct renamesnaparg *ra;
2503 2646 dsl_sync_task_t *dst;
2504 2647 spa_t *spa;
2505 2648 char *cp, *fsname = spa_strdup(oldname);
2506 2649 int len = strlen(oldname) + 1;
2507 2650
2508 2651 /* truncate the snapshot name to get the fsname */
2509 2652 cp = strchr(fsname, '@');
2510 2653 *cp = '\0';
2511 2654
2512 2655 err = spa_open(fsname, &spa, FTAG);
2513 2656 if (err) {
2514 2657 kmem_free(fsname, len);
2515 2658 return (err);
2516 2659 }
2517 2660 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2518 2661 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2519 2662
2520 2663 ra->oldsnap = strchr(oldname, '@') + 1;
2521 2664 ra->newsnap = strchr(newname, '@') + 1;
2522 2665 *ra->failed = '\0';
2523 2666
2524 2667 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2525 2668 DS_FIND_CHILDREN);
2526 2669 kmem_free(fsname, len);
2527 2670
2528 2671 if (err == 0) {
2529 2672 err = dsl_sync_task_group_wait(ra->dstg);
2530 2673 }
2531 2674
2532 2675 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2533 2676 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2534 2677 dsl_dataset_t *ds = dst->dst_arg1;
2535 2678 if (dst->dst_err) {
2536 2679 dsl_dir_name(ds->ds_dir, ra->failed);
2537 2680 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2538 2681 (void) strlcat(ra->failed, ra->newsnap,
2539 2682 sizeof (ra->failed));
2540 2683 }
2541 2684 dsl_dataset_rele(ds, ra->dstg);
2542 2685 }
2543 2686
2544 2687 if (err)
2545 2688 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2546 2689
2547 2690 dsl_sync_task_group_destroy(ra->dstg);
2548 2691 kmem_free(ra, sizeof (struct renamesnaparg));
2549 2692 spa_close(spa, FTAG);
2550 2693 return (err);
2551 2694 }
2552 2695
2553 2696 static int
2554 2697 dsl_valid_rename(const char *oldname, void *arg)
2555 2698 {
2556 2699 int delta = *(int *)arg;
2557 2700
2558 2701 if (strlen(oldname) + delta >= MAXNAMELEN)
2559 2702 return (ENAMETOOLONG);
2560 2703
2561 2704 return (0);
2562 2705 }
2563 2706
2564 2707 #pragma weak dmu_objset_rename = dsl_dataset_rename
2565 2708 int
2566 2709 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2567 2710 {
2568 2711 dsl_dir_t *dd;
2569 2712 dsl_dataset_t *ds;
2570 2713 const char *tail;
2571 2714 int err;
2572 2715
2573 2716 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2574 2717 if (err)
2575 2718 return (err);
2576 2719
2577 2720 if (tail == NULL) {
2578 2721 int delta = strlen(newname) - strlen(oldname);
2579 2722
2580 2723 /* if we're growing, validate child name lengths */
2581 2724 if (delta > 0)
2582 2725 err = dmu_objset_find(oldname, dsl_valid_rename,
2583 2726 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2584 2727
2585 2728 if (err == 0)
2586 2729 err = dsl_dir_rename(dd, newname);
2587 2730 dsl_dir_close(dd, FTAG);
2588 2731 return (err);
2589 2732 }
2590 2733
2591 2734 if (tail[0] != '@') {
2592 2735 /* the name ended in a nonexistent component */
2593 2736 dsl_dir_close(dd, FTAG);
2594 2737 return (ENOENT);
2595 2738 }
2596 2739
2597 2740 dsl_dir_close(dd, FTAG);
2598 2741
2599 2742 /* new name must be snapshot in same filesystem */
2600 2743 tail = strchr(newname, '@');
2601 2744 if (tail == NULL)
2602 2745 return (EINVAL);
2603 2746 tail++;
2604 2747 if (strncmp(oldname, newname, tail - newname) != 0)
2605 2748 return (EXDEV);
2606 2749
2607 2750 if (recursive) {
2608 2751 err = dsl_recursive_rename(oldname, newname);
2609 2752 } else {
2610 2753 err = dsl_dataset_hold(oldname, FTAG, &ds);
2611 2754 if (err)
2612 2755 return (err);
2613 2756
2614 2757 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2615 2758 dsl_dataset_snapshot_rename_check,
2616 2759 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2617 2760
2618 2761 dsl_dataset_rele(ds, FTAG);
2619 2762 }
2620 2763
2621 2764 return (err);
2622 2765 }
2623 2766
2624 2767 struct promotenode {
2625 2768 list_node_t link;
2626 2769 dsl_dataset_t *ds;
2627 2770 };
2628 2771
2629 2772 struct promotearg {
2630 2773 list_t shared_snaps, origin_snaps, clone_snaps;
2631 2774 dsl_dataset_t *origin_origin;
2632 2775 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2633 2776 char *err_ds;
2634 2777 };
2635 2778
2636 2779 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2637 2780 static boolean_t snaplist_unstable(list_t *l);
2638 2781
2639 2782 static int
2640 2783 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2641 2784 {
2642 2785 dsl_dataset_t *hds = arg1;
2643 2786 struct promotearg *pa = arg2;
2644 2787 struct promotenode *snap = list_head(&pa->shared_snaps);
2645 2788 dsl_dataset_t *origin_ds = snap->ds;
2646 2789 int err;
2647 2790 uint64_t unused;
2648 2791
2649 2792 /* Check that it is a real clone */
2650 2793 if (!dsl_dir_is_clone(hds->ds_dir))
2651 2794 return (EINVAL);
2652 2795
2653 2796 /* Since this is so expensive, don't do the preliminary check */
2654 2797 if (!dmu_tx_is_syncing(tx))
2655 2798 return (0);
2656 2799
2657 2800 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2658 2801 return (EXDEV);
2659 2802
2660 2803 /* compute origin's new unique space */
2661 2804 snap = list_tail(&pa->clone_snaps);
2662 2805 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2663 2806 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2664 2807 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2665 2808 &pa->unique, &unused, &unused);
2666 2809
2667 2810 /*
2668 2811 * Walk the snapshots that we are moving
2669 2812 *
2670 2813 * Compute space to transfer. Consider the incremental changes
2671 2814 * to used for each snapshot:
2672 2815 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2673 2816 * So each snapshot gave birth to:
2674 2817 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2675 2818 * So a sequence would look like:
2676 2819 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2677 2820 * Which simplifies to:
2678 2821 * uN + kN + kN-1 + ... + k1 + k0
2679 2822 * Note however, if we stop before we reach the ORIGIN we get:
2680 2823 * uN + kN + kN-1 + ... + kM - uM-1
2681 2824 */
2682 2825 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2683 2826 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2684 2827 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2685 2828 for (snap = list_head(&pa->shared_snaps); snap;
2686 2829 snap = list_next(&pa->shared_snaps, snap)) {
2687 2830 uint64_t val, dlused, dlcomp, dluncomp;
2688 2831 dsl_dataset_t *ds = snap->ds;
2689 2832
2690 2833 /* Check that the snapshot name does not conflict */
2691 2834 VERIFY(0 == dsl_dataset_get_snapname(ds));
2692 2835 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2693 2836 if (err == 0) {
2694 2837 err = EEXIST;
2695 2838 goto out;
2696 2839 }
2697 2840 if (err != ENOENT)
2698 2841 goto out;
2699 2842
2700 2843 /* The very first snapshot does not have a deadlist */
2701 2844 if (ds->ds_phys->ds_prev_snap_obj == 0)
2702 2845 continue;
2703 2846
2704 2847 dsl_deadlist_space(&ds->ds_deadlist,
2705 2848 &dlused, &dlcomp, &dluncomp);
2706 2849 pa->used += dlused;
2707 2850 pa->comp += dlcomp;
2708 2851 pa->uncomp += dluncomp;
2709 2852 }
2710 2853
|
↓ open down ↓ |
635 lines elided |
↑ open up ↑ |
2711 2854 /*
2712 2855 * If we are a clone of a clone then we never reached ORIGIN,
2713 2856 * so we need to subtract out the clone origin's used space.
2714 2857 */
2715 2858 if (pa->origin_origin) {
2716 2859 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2717 2860 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2718 2861 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2719 2862 }
2720 2863
2721 - /* Check that there is enough space here */
2864 + /* Check that there is enough space and quota headroom here */
2722 2865 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2723 - pa->used);
2866 + origin_ds->ds_dir, pa->used, tx);
2724 2867 if (err)
2725 2868 return (err);
2726 2869
2727 2870 /*
2728 2871 * Compute the amounts of space that will be used by snapshots
2729 2872 * after the promotion (for both origin and clone). For each,
2730 2873 * it is the amount of space that will be on all of their
2731 2874 * deadlists (that was not born before their new origin).
2732 2875 */
2733 2876 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2734 2877 uint64_t space;
2735 2878
2736 2879 /*
2737 2880 * Note, typically this will not be a clone of a clone,
2738 2881 * so dd_origin_txg will be < TXG_INITIAL, so
2739 2882 * these snaplist_space() -> dsl_deadlist_space_range()
2740 2883 * calls will be fast because they do not have to
2741 2884 * iterate over all bps.
2742 2885 */
2743 2886 snap = list_head(&pa->origin_snaps);
2744 2887 err = snaplist_space(&pa->shared_snaps,
2745 2888 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2746 2889 if (err)
2747 2890 return (err);
2748 2891
2749 2892 err = snaplist_space(&pa->clone_snaps,
2750 2893 snap->ds->ds_dir->dd_origin_txg, &space);
2751 2894 if (err)
2752 2895 return (err);
2753 2896 pa->cloneusedsnap += space;
2754 2897 }
2755 2898 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2756 2899 err = snaplist_space(&pa->origin_snaps,
2757 2900 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2758 2901 if (err)
2759 2902 return (err);
2760 2903 }
2761 2904
2762 2905 return (0);
2763 2906 out:
2764 2907 pa->err_ds = snap->ds->ds_snapname;
2765 2908 return (err);
2766 2909 }
2767 2910
2768 2911 static void
2769 2912 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2770 2913 {
2771 2914 dsl_dataset_t *hds = arg1;
2772 2915 struct promotearg *pa = arg2;
2773 2916 struct promotenode *snap = list_head(&pa->shared_snaps);
2774 2917 dsl_dataset_t *origin_ds = snap->ds;
2775 2918 dsl_dataset_t *origin_head;
2776 2919 dsl_dir_t *dd = hds->ds_dir;
2777 2920 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2778 2921 dsl_dir_t *odd = NULL;
2779 2922 uint64_t oldnext_obj;
2780 2923 int64_t delta;
2781 2924
2782 2925 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2783 2926
2784 2927 snap = list_head(&pa->origin_snaps);
2785 2928 origin_head = snap->ds;
2786 2929
2787 2930 /*
2788 2931 * We need to explicitly open odd, since origin_ds's dd will be
2789 2932 * changing.
2790 2933 */
2791 2934 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2792 2935 NULL, FTAG, &odd));
2793 2936
2794 2937 /* change origin's next snap */
2795 2938 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2796 2939 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2797 2940 snap = list_tail(&pa->clone_snaps);
2798 2941 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2799 2942 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2800 2943
2801 2944 /* change the origin's next clone */
2802 2945 if (origin_ds->ds_phys->ds_next_clones_obj) {
2803 2946 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2804 2947 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2805 2948 origin_ds->ds_phys->ds_next_clones_obj,
2806 2949 oldnext_obj, tx));
2807 2950 }
2808 2951
2809 2952 /* change origin */
2810 2953 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2811 2954 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2812 2955 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2813 2956 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2814 2957 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2815 2958 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2816 2959 origin_head->ds_dir->dd_origin_txg =
2817 2960 origin_ds->ds_phys->ds_creation_txg;
2818 2961
2819 2962 /* change dd_clone entries */
2820 2963 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2821 2964 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2822 2965 odd->dd_phys->dd_clones, hds->ds_object, tx));
2823 2966 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2824 2967 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2825 2968 hds->ds_object, tx));
2826 2969
2827 2970 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2828 2971 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2829 2972 origin_head->ds_object, tx));
2830 2973 if (dd->dd_phys->dd_clones == 0) {
2831 2974 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2832 2975 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2833 2976 }
2834 2977 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2835 2978 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2836 2979
2837 2980 }
2838 2981
2839 2982 /* move snapshots to this dir */
2840 2983 for (snap = list_head(&pa->shared_snaps); snap;
2841 2984 snap = list_next(&pa->shared_snaps, snap)) {
2842 2985 dsl_dataset_t *ds = snap->ds;
2843 2986
2844 2987 /* unregister props as dsl_dir is changing */
2845 2988 if (ds->ds_objset) {
|
↓ open down ↓ |
112 lines elided |
↑ open up ↑ |
2846 2989 dmu_objset_evict(ds->ds_objset);
2847 2990 ds->ds_objset = NULL;
2848 2991 }
2849 2992 /* move snap name entry */
2850 2993 VERIFY(0 == dsl_dataset_get_snapname(ds));
2851 2994 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2852 2995 ds->ds_snapname, tx));
2853 2996 VERIFY(0 == zap_add(dp->dp_meta_objset,
2854 2997 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2855 2998 8, 1, &ds->ds_object, tx));
2999 + dsl_snapcount_adjust(hds->ds_dir, tx, 1, B_TRUE);
2856 3000
2857 3001 /* change containing dsl_dir */
2858 3002 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2859 3003 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2860 3004 ds->ds_phys->ds_dir_obj = dd->dd_object;
2861 3005 ASSERT3P(ds->ds_dir, ==, odd);
2862 3006 dsl_dir_close(ds->ds_dir, ds);
2863 3007 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2864 3008 NULL, ds, &ds->ds_dir));
2865 3009
2866 3010 /* move any clone references */
2867 3011 if (ds->ds_phys->ds_next_clones_obj &&
2868 3012 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2869 3013 zap_cursor_t zc;
2870 3014 zap_attribute_t za;
2871 3015
2872 3016 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2873 3017 ds->ds_phys->ds_next_clones_obj);
2874 3018 zap_cursor_retrieve(&zc, &za) == 0;
2875 3019 zap_cursor_advance(&zc)) {
2876 3020 dsl_dataset_t *cnds;
2877 3021 uint64_t o;
2878 3022
2879 3023 if (za.za_first_integer == oldnext_obj) {
2880 3024 /*
2881 3025 * We've already moved the
2882 3026 * origin's reference.
2883 3027 */
2884 3028 continue;
2885 3029 }
2886 3030
2887 3031 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2888 3032 za.za_first_integer, FTAG, &cnds));
2889 3033 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2890 3034
2891 3035 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2892 3036 odd->dd_phys->dd_clones, o, tx), ==, 0);
2893 3037 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2894 3038 dd->dd_phys->dd_clones, o, tx), ==, 0);
2895 3039 dsl_dataset_rele(cnds, FTAG);
2896 3040 }
2897 3041 zap_cursor_fini(&zc);
2898 3042 }
2899 3043
2900 3044 ASSERT0(dsl_prop_numcb(ds));
2901 3045 }
2902 3046
2903 3047 /*
2904 3048 * Change space accounting.
2905 3049 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2906 3050 * both be valid, or both be 0 (resulting in delta == 0). This
2907 3051 * is true for each of {clone,origin} independently.
2908 3052 */
2909 3053
2910 3054 delta = pa->cloneusedsnap -
2911 3055 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2912 3056 ASSERT3S(delta, >=, 0);
2913 3057 ASSERT3U(pa->used, >=, delta);
2914 3058 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2915 3059 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2916 3060 pa->used - delta, pa->comp, pa->uncomp, tx);
2917 3061
2918 3062 delta = pa->originusedsnap -
2919 3063 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2920 3064 ASSERT3S(delta, <=, 0);
2921 3065 ASSERT3U(pa->used, >=, -delta);
2922 3066 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2923 3067 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2924 3068 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2925 3069
2926 3070 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2927 3071
2928 3072 /* log history record */
2929 3073 spa_history_log_internal_ds(hds, "promote", tx, "");
2930 3074
2931 3075 dsl_dir_close(odd, FTAG);
2932 3076 }
2933 3077
2934 3078 static char *snaplist_tag = "snaplist";
2935 3079 /*
2936 3080 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2937 3081 * (exclusive) and last_obj (inclusive). The list will be in reverse
2938 3082 * order (last_obj will be the list_head()). If first_obj == 0, do all
2939 3083 * snapshots back to this dataset's origin.
2940 3084 */
2941 3085 static int
2942 3086 snaplist_make(dsl_pool_t *dp, boolean_t own,
2943 3087 uint64_t first_obj, uint64_t last_obj, list_t *l)
2944 3088 {
2945 3089 uint64_t obj = last_obj;
2946 3090
2947 3091 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2948 3092
2949 3093 list_create(l, sizeof (struct promotenode),
2950 3094 offsetof(struct promotenode, link));
2951 3095
2952 3096 while (obj != first_obj) {
2953 3097 dsl_dataset_t *ds;
2954 3098 struct promotenode *snap;
2955 3099 int err;
2956 3100
2957 3101 if (own) {
2958 3102 err = dsl_dataset_own_obj(dp, obj,
2959 3103 0, snaplist_tag, &ds);
2960 3104 if (err == 0)
2961 3105 dsl_dataset_make_exclusive(ds, snaplist_tag);
2962 3106 } else {
2963 3107 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2964 3108 }
2965 3109 if (err == ENOENT) {
2966 3110 /* lost race with snapshot destroy */
2967 3111 struct promotenode *last = list_tail(l);
2968 3112 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2969 3113 obj = last->ds->ds_phys->ds_prev_snap_obj;
2970 3114 continue;
2971 3115 } else if (err) {
2972 3116 return (err);
2973 3117 }
2974 3118
2975 3119 if (first_obj == 0)
2976 3120 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2977 3121
2978 3122 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2979 3123 snap->ds = ds;
2980 3124 list_insert_tail(l, snap);
2981 3125 obj = ds->ds_phys->ds_prev_snap_obj;
2982 3126 }
2983 3127
2984 3128 return (0);
2985 3129 }
2986 3130
2987 3131 static int
2988 3132 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2989 3133 {
2990 3134 struct promotenode *snap;
2991 3135
2992 3136 *spacep = 0;
2993 3137 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2994 3138 uint64_t used, comp, uncomp;
2995 3139 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2996 3140 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2997 3141 *spacep += used;
2998 3142 }
2999 3143 return (0);
3000 3144 }
3001 3145
3002 3146 static void
3003 3147 snaplist_destroy(list_t *l, boolean_t own)
3004 3148 {
3005 3149 struct promotenode *snap;
3006 3150
3007 3151 if (!l || !list_link_active(&l->list_head))
3008 3152 return;
3009 3153
3010 3154 while ((snap = list_tail(l)) != NULL) {
3011 3155 list_remove(l, snap);
3012 3156 if (own)
3013 3157 dsl_dataset_disown(snap->ds, snaplist_tag);
3014 3158 else
3015 3159 dsl_dataset_rele(snap->ds, snaplist_tag);
3016 3160 kmem_free(snap, sizeof (struct promotenode));
3017 3161 }
3018 3162 list_destroy(l);
3019 3163 }
3020 3164
3021 3165 /*
3022 3166 * Promote a clone. Nomenclature note:
3023 3167 * "clone" or "cds": the original clone which is being promoted
3024 3168 * "origin" or "ods": the snapshot which is originally clone's origin
3025 3169 * "origin head" or "ohds": the dataset which is the head
3026 3170 * (filesystem/volume) for the origin
3027 3171 * "origin origin": the origin of the origin's filesystem (typically
3028 3172 * NULL, indicating that the clone is not a clone of a clone).
3029 3173 */
3030 3174 int
3031 3175 dsl_dataset_promote(const char *name, char *conflsnap)
3032 3176 {
3033 3177 dsl_dataset_t *ds;
3034 3178 dsl_dir_t *dd;
3035 3179 dsl_pool_t *dp;
3036 3180 dmu_object_info_t doi;
3037 3181 struct promotearg pa = { 0 };
3038 3182 struct promotenode *snap;
3039 3183 int err;
3040 3184
3041 3185 err = dsl_dataset_hold(name, FTAG, &ds);
3042 3186 if (err)
3043 3187 return (err);
3044 3188 dd = ds->ds_dir;
3045 3189 dp = dd->dd_pool;
3046 3190
3047 3191 err = dmu_object_info(dp->dp_meta_objset,
3048 3192 ds->ds_phys->ds_snapnames_zapobj, &doi);
3049 3193 if (err) {
3050 3194 dsl_dataset_rele(ds, FTAG);
3051 3195 return (err);
3052 3196 }
3053 3197
3054 3198 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3055 3199 dsl_dataset_rele(ds, FTAG);
3056 3200 return (EINVAL);
3057 3201 }
3058 3202
3059 3203 /*
3060 3204 * We are going to inherit all the snapshots taken before our
3061 3205 * origin (i.e., our new origin will be our parent's origin).
3062 3206 * Take ownership of them so that we can rename them into our
3063 3207 * namespace.
3064 3208 */
3065 3209 rw_enter(&dp->dp_config_rwlock, RW_READER);
3066 3210
3067 3211 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3068 3212 &pa.shared_snaps);
3069 3213 if (err != 0)
3070 3214 goto out;
3071 3215
3072 3216 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3073 3217 if (err != 0)
3074 3218 goto out;
3075 3219
3076 3220 snap = list_head(&pa.shared_snaps);
3077 3221 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3078 3222 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3079 3223 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3080 3224 if (err != 0)
3081 3225 goto out;
3082 3226
3083 3227 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3084 3228 err = dsl_dataset_hold_obj(dp,
3085 3229 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3086 3230 FTAG, &pa.origin_origin);
3087 3231 if (err != 0)
3088 3232 goto out;
3089 3233 }
3090 3234
3091 3235 out:
3092 3236 rw_exit(&dp->dp_config_rwlock);
3093 3237
3094 3238 /*
3095 3239 * Add in 128x the snapnames zapobj size, since we will be moving
3096 3240 * a bunch of snapnames to the promoted ds, and dirtying their
3097 3241 * bonus buffers.
3098 3242 */
3099 3243 if (err == 0) {
3100 3244 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3101 3245 dsl_dataset_promote_sync, ds, &pa,
3102 3246 2 + 2 * doi.doi_physical_blocks_512);
3103 3247 if (err && pa.err_ds && conflsnap)
3104 3248 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3105 3249 }
3106 3250
3107 3251 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3108 3252 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3109 3253 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3110 3254 if (pa.origin_origin)
3111 3255 dsl_dataset_rele(pa.origin_origin, FTAG);
3112 3256 dsl_dataset_rele(ds, FTAG);
3113 3257 return (err);
3114 3258 }
3115 3259
3116 3260 struct cloneswaparg {
3117 3261 dsl_dataset_t *cds; /* clone dataset */
3118 3262 dsl_dataset_t *ohds; /* origin's head dataset */
3119 3263 boolean_t force;
3120 3264 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3121 3265 };
3122 3266
3123 3267 /* ARGSUSED */
3124 3268 static int
3125 3269 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3126 3270 {
3127 3271 struct cloneswaparg *csa = arg1;
3128 3272
3129 3273 /* they should both be heads */
3130 3274 if (dsl_dataset_is_snapshot(csa->cds) ||
3131 3275 dsl_dataset_is_snapshot(csa->ohds))
3132 3276 return (EINVAL);
3133 3277
3134 3278 /* the branch point should be just before them */
3135 3279 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3136 3280 return (EINVAL);
3137 3281
3138 3282 /* cds should be the clone (unless they are unrelated) */
3139 3283 if (csa->cds->ds_prev != NULL &&
3140 3284 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3141 3285 csa->ohds->ds_object !=
3142 3286 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3143 3287 return (EINVAL);
3144 3288
3145 3289 /* the clone should be a child of the origin */
3146 3290 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3147 3291 return (EINVAL);
3148 3292
3149 3293 /* ohds shouldn't be modified unless 'force' */
3150 3294 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3151 3295 return (ETXTBSY);
3152 3296
3153 3297 /* adjust amount of any unconsumed refreservation */
3154 3298 csa->unused_refres_delta =
3155 3299 (int64_t)MIN(csa->ohds->ds_reserved,
3156 3300 csa->ohds->ds_phys->ds_unique_bytes) -
3157 3301 (int64_t)MIN(csa->ohds->ds_reserved,
3158 3302 csa->cds->ds_phys->ds_unique_bytes);
3159 3303
3160 3304 if (csa->unused_refres_delta > 0 &&
3161 3305 csa->unused_refres_delta >
3162 3306 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3163 3307 return (ENOSPC);
3164 3308
3165 3309 if (csa->ohds->ds_quota != 0 &&
3166 3310 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3167 3311 return (EDQUOT);
3168 3312
3169 3313 return (0);
3170 3314 }
3171 3315
3172 3316 /* ARGSUSED */
3173 3317 static void
3174 3318 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3175 3319 {
3176 3320 struct cloneswaparg *csa = arg1;
3177 3321 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3178 3322
3179 3323 ASSERT(csa->cds->ds_reserved == 0);
3180 3324 ASSERT(csa->ohds->ds_quota == 0 ||
3181 3325 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3182 3326
3183 3327 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3184 3328 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3185 3329
3186 3330 if (csa->cds->ds_objset != NULL) {
3187 3331 dmu_objset_evict(csa->cds->ds_objset);
3188 3332 csa->cds->ds_objset = NULL;
3189 3333 }
3190 3334
3191 3335 if (csa->ohds->ds_objset != NULL) {
3192 3336 dmu_objset_evict(csa->ohds->ds_objset);
3193 3337 csa->ohds->ds_objset = NULL;
3194 3338 }
3195 3339
3196 3340 /*
3197 3341 * Reset origin's unique bytes, if it exists.
3198 3342 */
3199 3343 if (csa->cds->ds_prev) {
3200 3344 dsl_dataset_t *origin = csa->cds->ds_prev;
3201 3345 uint64_t comp, uncomp;
3202 3346
3203 3347 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3204 3348 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3205 3349 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3206 3350 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3207 3351 }
3208 3352
3209 3353 /* swap blkptrs */
3210 3354 {
3211 3355 blkptr_t tmp;
3212 3356 tmp = csa->ohds->ds_phys->ds_bp;
3213 3357 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3214 3358 csa->cds->ds_phys->ds_bp = tmp;
3215 3359 }
3216 3360
3217 3361 /* set dd_*_bytes */
3218 3362 {
3219 3363 int64_t dused, dcomp, duncomp;
3220 3364 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3221 3365 uint64_t odl_used, odl_comp, odl_uncomp;
3222 3366
3223 3367 ASSERT3U(csa->cds->ds_dir->dd_phys->
3224 3368 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3225 3369
3226 3370 dsl_deadlist_space(&csa->cds->ds_deadlist,
3227 3371 &cdl_used, &cdl_comp, &cdl_uncomp);
3228 3372 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3229 3373 &odl_used, &odl_comp, &odl_uncomp);
3230 3374
3231 3375 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3232 3376 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3233 3377 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3234 3378 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3235 3379 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3236 3380 cdl_uncomp -
3237 3381 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3238 3382
3239 3383 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3240 3384 dused, dcomp, duncomp, tx);
3241 3385 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3242 3386 -dused, -dcomp, -duncomp, tx);
3243 3387
3244 3388 /*
3245 3389 * The difference in the space used by snapshots is the
3246 3390 * difference in snapshot space due to the head's
3247 3391 * deadlist (since that's the only thing that's
3248 3392 * changing that affects the snapused).
3249 3393 */
3250 3394 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3251 3395 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3252 3396 &cdl_used, &cdl_comp, &cdl_uncomp);
3253 3397 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3254 3398 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3255 3399 &odl_used, &odl_comp, &odl_uncomp);
3256 3400 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3257 3401 DD_USED_HEAD, DD_USED_SNAP, tx);
3258 3402 }
3259 3403
3260 3404 /* swap ds_*_bytes */
3261 3405 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3262 3406 csa->cds->ds_phys->ds_referenced_bytes);
3263 3407 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3264 3408 csa->cds->ds_phys->ds_compressed_bytes);
3265 3409 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3266 3410 csa->cds->ds_phys->ds_uncompressed_bytes);
3267 3411 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3268 3412 csa->cds->ds_phys->ds_unique_bytes);
3269 3413
3270 3414 /* apply any parent delta for change in unconsumed refreservation */
3271 3415 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3272 3416 csa->unused_refres_delta, 0, 0, tx);
3273 3417
3274 3418 /*
3275 3419 * Swap deadlists.
3276 3420 */
3277 3421 dsl_deadlist_close(&csa->cds->ds_deadlist);
3278 3422 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3279 3423 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3280 3424 csa->cds->ds_phys->ds_deadlist_obj);
3281 3425 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3282 3426 csa->cds->ds_phys->ds_deadlist_obj);
3283 3427 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3284 3428 csa->ohds->ds_phys->ds_deadlist_obj);
3285 3429
3286 3430 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3287 3431
3288 3432 spa_history_log_internal_ds(csa->cds, "clone swap", tx,
3289 3433 "parent=%s", csa->ohds->ds_dir->dd_myname);
3290 3434 }
3291 3435
3292 3436 /*
3293 3437 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3294 3438 * recv" into an existing fs to swizzle the file system to the new
3295 3439 * version, and by "zfs rollback". Can also be used to swap two
3296 3440 * independent head datasets if neither has any snapshots.
3297 3441 */
3298 3442 int
3299 3443 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3300 3444 boolean_t force)
3301 3445 {
3302 3446 struct cloneswaparg csa;
3303 3447 int error;
3304 3448
3305 3449 ASSERT(clone->ds_owner);
3306 3450 ASSERT(origin_head->ds_owner);
3307 3451 retry:
3308 3452 /*
3309 3453 * Need exclusive access for the swap. If we're swapping these
3310 3454 * datasets back after an error, we already hold the locks.
3311 3455 */
3312 3456 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3313 3457 rw_enter(&clone->ds_rwlock, RW_WRITER);
3314 3458 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3315 3459 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3316 3460 rw_exit(&clone->ds_rwlock);
3317 3461 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3318 3462 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3319 3463 rw_exit(&origin_head->ds_rwlock);
3320 3464 goto retry;
3321 3465 }
3322 3466 }
3323 3467 csa.cds = clone;
3324 3468 csa.ohds = origin_head;
3325 3469 csa.force = force;
3326 3470 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3327 3471 dsl_dataset_clone_swap_check,
3328 3472 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3329 3473 return (error);
3330 3474 }
3331 3475
3332 3476 /*
3333 3477 * Given a pool name and a dataset object number in that pool,
3334 3478 * return the name of that dataset.
3335 3479 */
3336 3480 int
3337 3481 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3338 3482 {
3339 3483 spa_t *spa;
3340 3484 dsl_pool_t *dp;
3341 3485 dsl_dataset_t *ds;
3342 3486 int error;
3343 3487
3344 3488 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3345 3489 return (error);
3346 3490 dp = spa_get_dsl(spa);
3347 3491 rw_enter(&dp->dp_config_rwlock, RW_READER);
3348 3492 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3349 3493 dsl_dataset_name(ds, buf);
3350 3494 dsl_dataset_rele(ds, FTAG);
3351 3495 }
3352 3496 rw_exit(&dp->dp_config_rwlock);
3353 3497 spa_close(spa, FTAG);
3354 3498
3355 3499 return (error);
3356 3500 }
3357 3501
3358 3502 int
3359 3503 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3360 3504 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3361 3505 {
3362 3506 int error = 0;
3363 3507
3364 3508 ASSERT3S(asize, >, 0);
3365 3509
3366 3510 /*
3367 3511 * *ref_rsrv is the portion of asize that will come from any
3368 3512 * unconsumed refreservation space.
3369 3513 */
3370 3514 *ref_rsrv = 0;
3371 3515
3372 3516 mutex_enter(&ds->ds_lock);
3373 3517 /*
3374 3518 * Make a space adjustment for reserved bytes.
3375 3519 */
3376 3520 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3377 3521 ASSERT3U(*used, >=,
3378 3522 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3379 3523 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3380 3524 *ref_rsrv =
3381 3525 asize - MIN(asize, parent_delta(ds, asize + inflight));
3382 3526 }
3383 3527
3384 3528 if (!check_quota || ds->ds_quota == 0) {
3385 3529 mutex_exit(&ds->ds_lock);
3386 3530 return (0);
3387 3531 }
3388 3532 /*
3389 3533 * If they are requesting more space, and our current estimate
3390 3534 * is over quota, they get to try again unless the actual
3391 3535 * on-disk is over quota and there are no pending changes (which
3392 3536 * may free up space for us).
3393 3537 */
3394 3538 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3395 3539 if (inflight > 0 ||
3396 3540 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3397 3541 error = ERESTART;
3398 3542 else
3399 3543 error = EDQUOT;
3400 3544 }
3401 3545 mutex_exit(&ds->ds_lock);
3402 3546
3403 3547 return (error);
3404 3548 }
3405 3549
3406 3550 /* ARGSUSED */
3407 3551 static int
3408 3552 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3409 3553 {
3410 3554 dsl_dataset_t *ds = arg1;
3411 3555 dsl_prop_setarg_t *psa = arg2;
3412 3556 int err;
3413 3557
3414 3558 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3415 3559 return (ENOTSUP);
3416 3560
3417 3561 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3418 3562 return (err);
3419 3563
3420 3564 if (psa->psa_effective_value == 0)
3421 3565 return (0);
3422 3566
3423 3567 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3424 3568 psa->psa_effective_value < ds->ds_reserved)
3425 3569 return (ENOSPC);
3426 3570
3427 3571 return (0);
3428 3572 }
3429 3573
3430 3574 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3431 3575
3432 3576 void
3433 3577 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3434 3578 {
3435 3579 dsl_dataset_t *ds = arg1;
3436 3580 dsl_prop_setarg_t *psa = arg2;
3437 3581 uint64_t effective_value = psa->psa_effective_value;
3438 3582
3439 3583 dsl_prop_set_sync(ds, psa, tx);
3440 3584 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3441 3585
3442 3586 if (ds->ds_quota != effective_value) {
3443 3587 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3444 3588 ds->ds_quota = effective_value;
3445 3589 }
3446 3590 }
3447 3591
3448 3592 int
3449 3593 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3450 3594 {
3451 3595 dsl_dataset_t *ds;
3452 3596 dsl_prop_setarg_t psa;
3453 3597 int err;
3454 3598
3455 3599 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3456 3600
3457 3601 err = dsl_dataset_hold(dsname, FTAG, &ds);
3458 3602 if (err)
3459 3603 return (err);
3460 3604
3461 3605 /*
3462 3606 * If someone removes a file, then tries to set the quota, we
3463 3607 * want to make sure the file freeing takes effect.
3464 3608 */
3465 3609 txg_wait_open(ds->ds_dir->dd_pool, 0);
3466 3610
3467 3611 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3468 3612 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3469 3613 ds, &psa, 0);
3470 3614
3471 3615 dsl_dataset_rele(ds, FTAG);
3472 3616 return (err);
3473 3617 }
3474 3618
3475 3619 static int
3476 3620 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3477 3621 {
3478 3622 dsl_dataset_t *ds = arg1;
3479 3623 dsl_prop_setarg_t *psa = arg2;
3480 3624 uint64_t effective_value;
3481 3625 uint64_t unique;
3482 3626 int err;
3483 3627
3484 3628 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3485 3629 SPA_VERSION_REFRESERVATION)
3486 3630 return (ENOTSUP);
3487 3631
3488 3632 if (dsl_dataset_is_snapshot(ds))
3489 3633 return (EINVAL);
3490 3634
3491 3635 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3492 3636 return (err);
3493 3637
3494 3638 effective_value = psa->psa_effective_value;
3495 3639
3496 3640 /*
3497 3641 * If we are doing the preliminary check in open context, the
3498 3642 * space estimates may be inaccurate.
3499 3643 */
3500 3644 if (!dmu_tx_is_syncing(tx))
3501 3645 return (0);
3502 3646
3503 3647 mutex_enter(&ds->ds_lock);
3504 3648 if (!DS_UNIQUE_IS_ACCURATE(ds))
3505 3649 dsl_dataset_recalc_head_uniq(ds);
3506 3650 unique = ds->ds_phys->ds_unique_bytes;
3507 3651 mutex_exit(&ds->ds_lock);
3508 3652
3509 3653 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3510 3654 uint64_t delta = MAX(unique, effective_value) -
3511 3655 MAX(unique, ds->ds_reserved);
3512 3656
3513 3657 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3514 3658 return (ENOSPC);
3515 3659 if (ds->ds_quota > 0 &&
3516 3660 effective_value > ds->ds_quota)
3517 3661 return (ENOSPC);
3518 3662 }
3519 3663
3520 3664 return (0);
3521 3665 }
3522 3666
3523 3667 static void
3524 3668 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3525 3669 {
3526 3670 dsl_dataset_t *ds = arg1;
3527 3671 dsl_prop_setarg_t *psa = arg2;
3528 3672 uint64_t effective_value = psa->psa_effective_value;
3529 3673 uint64_t unique;
3530 3674 int64_t delta;
3531 3675
3532 3676 dsl_prop_set_sync(ds, psa, tx);
3533 3677 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3534 3678
3535 3679 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3536 3680
3537 3681 mutex_enter(&ds->ds_dir->dd_lock);
3538 3682 mutex_enter(&ds->ds_lock);
3539 3683 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3540 3684 unique = ds->ds_phys->ds_unique_bytes;
3541 3685 delta = MAX(0, (int64_t)(effective_value - unique)) -
3542 3686 MAX(0, (int64_t)(ds->ds_reserved - unique));
3543 3687 ds->ds_reserved = effective_value;
3544 3688 mutex_exit(&ds->ds_lock);
3545 3689
3546 3690 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3547 3691 mutex_exit(&ds->ds_dir->dd_lock);
3548 3692 }
3549 3693
3550 3694 int
3551 3695 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3552 3696 uint64_t reservation)
3553 3697 {
3554 3698 dsl_dataset_t *ds;
3555 3699 dsl_prop_setarg_t psa;
3556 3700 int err;
3557 3701
3558 3702 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3559 3703 &reservation);
3560 3704
3561 3705 err = dsl_dataset_hold(dsname, FTAG, &ds);
3562 3706 if (err)
3563 3707 return (err);
3564 3708
3565 3709 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3566 3710 dsl_dataset_set_reservation_check,
3567 3711 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3568 3712
3569 3713 dsl_dataset_rele(ds, FTAG);
3570 3714 return (err);
3571 3715 }
3572 3716
3573 3717 typedef struct zfs_hold_cleanup_arg {
3574 3718 dsl_pool_t *dp;
3575 3719 uint64_t dsobj;
3576 3720 char htag[MAXNAMELEN];
3577 3721 } zfs_hold_cleanup_arg_t;
3578 3722
3579 3723 static void
3580 3724 dsl_dataset_user_release_onexit(void *arg)
3581 3725 {
3582 3726 zfs_hold_cleanup_arg_t *ca = arg;
3583 3727
3584 3728 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3585 3729 B_TRUE);
3586 3730 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3587 3731 }
3588 3732
3589 3733 void
3590 3734 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3591 3735 minor_t minor)
3592 3736 {
3593 3737 zfs_hold_cleanup_arg_t *ca;
3594 3738
3595 3739 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3596 3740 ca->dp = ds->ds_dir->dd_pool;
3597 3741 ca->dsobj = ds->ds_object;
3598 3742 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3599 3743 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3600 3744 dsl_dataset_user_release_onexit, ca, NULL));
3601 3745 }
3602 3746
3603 3747 /*
3604 3748 * If you add new checks here, you may need to add
3605 3749 * additional checks to the "temporary" case in
3606 3750 * snapshot_check() in dmu_objset.c.
3607 3751 */
3608 3752 static int
3609 3753 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3610 3754 {
3611 3755 dsl_dataset_t *ds = arg1;
3612 3756 struct dsl_ds_holdarg *ha = arg2;
3613 3757 const char *htag = ha->htag;
3614 3758 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3615 3759 int error = 0;
3616 3760
3617 3761 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3618 3762 return (ENOTSUP);
3619 3763
3620 3764 if (!dsl_dataset_is_snapshot(ds))
3621 3765 return (EINVAL);
3622 3766
3623 3767 /* tags must be unique */
3624 3768 mutex_enter(&ds->ds_lock);
3625 3769 if (ds->ds_phys->ds_userrefs_obj) {
3626 3770 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3627 3771 8, 1, tx);
3628 3772 if (error == 0)
3629 3773 error = EEXIST;
3630 3774 else if (error == ENOENT)
3631 3775 error = 0;
3632 3776 }
3633 3777 mutex_exit(&ds->ds_lock);
3634 3778
3635 3779 if (error == 0 && ha->temphold &&
3636 3780 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3637 3781 error = E2BIG;
3638 3782
3639 3783 return (error);
3640 3784 }
3641 3785
3642 3786 void
3643 3787 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3644 3788 {
3645 3789 dsl_dataset_t *ds = arg1;
3646 3790 struct dsl_ds_holdarg *ha = arg2;
3647 3791 const char *htag = ha->htag;
3648 3792 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3649 3793 objset_t *mos = dp->dp_meta_objset;
3650 3794 uint64_t now = gethrestime_sec();
3651 3795 uint64_t zapobj;
3652 3796
3653 3797 mutex_enter(&ds->ds_lock);
3654 3798 if (ds->ds_phys->ds_userrefs_obj == 0) {
3655 3799 /*
3656 3800 * This is the first user hold for this dataset. Create
3657 3801 * the userrefs zap object.
3658 3802 */
3659 3803 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3660 3804 zapobj = ds->ds_phys->ds_userrefs_obj =
3661 3805 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3662 3806 } else {
3663 3807 zapobj = ds->ds_phys->ds_userrefs_obj;
3664 3808 }
3665 3809 ds->ds_userrefs++;
3666 3810 mutex_exit(&ds->ds_lock);
3667 3811
3668 3812 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3669 3813
3670 3814 if (ha->temphold) {
3671 3815 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3672 3816 htag, &now, tx));
3673 3817 }
3674 3818
3675 3819 spa_history_log_internal_ds(ds, "hold", tx,
3676 3820 "tag = %s temp = %d holds now = %llu",
3677 3821 htag, (int)ha->temphold, ds->ds_userrefs);
3678 3822 }
3679 3823
3680 3824 static int
3681 3825 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3682 3826 {
3683 3827 struct dsl_ds_holdarg *ha = arg;
3684 3828 dsl_dataset_t *ds;
3685 3829 int error;
3686 3830 char *name;
3687 3831
3688 3832 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3689 3833 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3690 3834 error = dsl_dataset_hold(name, ha->dstg, &ds);
3691 3835 strfree(name);
3692 3836 if (error == 0) {
3693 3837 ha->gotone = B_TRUE;
3694 3838 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3695 3839 dsl_dataset_user_hold_sync, ds, ha, 0);
3696 3840 } else if (error == ENOENT && ha->recursive) {
3697 3841 error = 0;
3698 3842 } else {
3699 3843 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3700 3844 }
3701 3845 return (error);
3702 3846 }
3703 3847
3704 3848 int
3705 3849 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3706 3850 boolean_t temphold)
3707 3851 {
3708 3852 struct dsl_ds_holdarg *ha;
3709 3853 int error;
3710 3854
3711 3855 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3712 3856 ha->htag = htag;
3713 3857 ha->temphold = temphold;
3714 3858 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3715 3859 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3716 3860 ds, ha, 0);
3717 3861 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3718 3862
3719 3863 return (error);
3720 3864 }
3721 3865
3722 3866 int
3723 3867 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3724 3868 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3725 3869 {
3726 3870 struct dsl_ds_holdarg *ha;
3727 3871 dsl_sync_task_t *dst;
3728 3872 spa_t *spa;
3729 3873 int error;
3730 3874 minor_t minor = 0;
3731 3875
3732 3876 if (cleanup_fd != -1) {
3733 3877 /* Currently we only support cleanup-on-exit of tempholds. */
3734 3878 if (!temphold)
3735 3879 return (EINVAL);
3736 3880 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3737 3881 if (error)
3738 3882 return (error);
3739 3883 }
3740 3884
3741 3885 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3742 3886
3743 3887 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3744 3888
3745 3889 error = spa_open(dsname, &spa, FTAG);
3746 3890 if (error) {
3747 3891 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3748 3892 if (cleanup_fd != -1)
3749 3893 zfs_onexit_fd_rele(cleanup_fd);
3750 3894 return (error);
3751 3895 }
3752 3896
3753 3897 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3754 3898 ha->htag = htag;
3755 3899 ha->snapname = snapname;
3756 3900 ha->recursive = recursive;
3757 3901 ha->temphold = temphold;
3758 3902
3759 3903 if (recursive) {
3760 3904 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3761 3905 ha, DS_FIND_CHILDREN);
3762 3906 } else {
3763 3907 error = dsl_dataset_user_hold_one(dsname, ha);
3764 3908 }
3765 3909 if (error == 0)
3766 3910 error = dsl_sync_task_group_wait(ha->dstg);
3767 3911
3768 3912 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3769 3913 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3770 3914 dsl_dataset_t *ds = dst->dst_arg1;
3771 3915
3772 3916 if (dst->dst_err) {
3773 3917 dsl_dataset_name(ds, ha->failed);
3774 3918 *strchr(ha->failed, '@') = '\0';
3775 3919 } else if (error == 0 && minor != 0 && temphold) {
3776 3920 /*
3777 3921 * If this hold is to be released upon process exit,
3778 3922 * register that action now.
3779 3923 */
3780 3924 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3781 3925 }
3782 3926 dsl_dataset_rele(ds, ha->dstg);
3783 3927 }
3784 3928
3785 3929 if (error == 0 && recursive && !ha->gotone)
3786 3930 error = ENOENT;
3787 3931
3788 3932 if (error)
3789 3933 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3790 3934
3791 3935 dsl_sync_task_group_destroy(ha->dstg);
3792 3936
3793 3937 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3794 3938 spa_close(spa, FTAG);
3795 3939 if (cleanup_fd != -1)
3796 3940 zfs_onexit_fd_rele(cleanup_fd);
3797 3941 return (error);
3798 3942 }
3799 3943
3800 3944 struct dsl_ds_releasearg {
3801 3945 dsl_dataset_t *ds;
3802 3946 const char *htag;
3803 3947 boolean_t own; /* do we own or just hold ds? */
3804 3948 };
3805 3949
3806 3950 static int
3807 3951 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3808 3952 boolean_t *might_destroy)
3809 3953 {
3810 3954 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3811 3955 uint64_t zapobj;
3812 3956 uint64_t tmp;
3813 3957 int error;
3814 3958
3815 3959 *might_destroy = B_FALSE;
3816 3960
3817 3961 mutex_enter(&ds->ds_lock);
3818 3962 zapobj = ds->ds_phys->ds_userrefs_obj;
3819 3963 if (zapobj == 0) {
3820 3964 /* The tag can't possibly exist */
3821 3965 mutex_exit(&ds->ds_lock);
3822 3966 return (ESRCH);
3823 3967 }
3824 3968
3825 3969 /* Make sure the tag exists */
3826 3970 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3827 3971 if (error) {
3828 3972 mutex_exit(&ds->ds_lock);
3829 3973 if (error == ENOENT)
3830 3974 error = ESRCH;
3831 3975 return (error);
3832 3976 }
3833 3977
3834 3978 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3835 3979 DS_IS_DEFER_DESTROY(ds))
3836 3980 *might_destroy = B_TRUE;
3837 3981
3838 3982 mutex_exit(&ds->ds_lock);
3839 3983 return (0);
3840 3984 }
3841 3985
3842 3986 static int
3843 3987 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3844 3988 {
3845 3989 struct dsl_ds_releasearg *ra = arg1;
3846 3990 dsl_dataset_t *ds = ra->ds;
3847 3991 boolean_t might_destroy;
3848 3992 int error;
3849 3993
3850 3994 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3851 3995 return (ENOTSUP);
3852 3996
3853 3997 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3854 3998 if (error)
3855 3999 return (error);
3856 4000
3857 4001 if (might_destroy) {
3858 4002 struct dsl_ds_destroyarg dsda = {0};
3859 4003
3860 4004 if (dmu_tx_is_syncing(tx)) {
3861 4005 /*
3862 4006 * If we're not prepared to remove the snapshot,
3863 4007 * we can't allow the release to happen right now.
3864 4008 */
3865 4009 if (!ra->own)
3866 4010 return (EBUSY);
3867 4011 }
3868 4012 dsda.ds = ds;
3869 4013 dsda.releasing = B_TRUE;
3870 4014 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3871 4015 }
3872 4016
3873 4017 return (0);
3874 4018 }
3875 4019
3876 4020 static void
3877 4021 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3878 4022 {
3879 4023 struct dsl_ds_releasearg *ra = arg1;
3880 4024 dsl_dataset_t *ds = ra->ds;
3881 4025 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3882 4026 objset_t *mos = dp->dp_meta_objset;
3883 4027 uint64_t zapobj;
3884 4028 uint64_t refs;
3885 4029 int error;
3886 4030
3887 4031 mutex_enter(&ds->ds_lock);
3888 4032 ds->ds_userrefs--;
3889 4033 refs = ds->ds_userrefs;
3890 4034 mutex_exit(&ds->ds_lock);
3891 4035 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3892 4036 VERIFY(error == 0 || error == ENOENT);
3893 4037 zapobj = ds->ds_phys->ds_userrefs_obj;
3894 4038 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3895 4039
3896 4040 spa_history_log_internal_ds(ds, "release", tx,
3897 4041 "tag = %s refs now = %lld", ra->htag, (longlong_t)refs);
3898 4042
3899 4043 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3900 4044 DS_IS_DEFER_DESTROY(ds)) {
3901 4045 struct dsl_ds_destroyarg dsda = {0};
3902 4046
3903 4047 ASSERT(ra->own);
3904 4048 dsda.ds = ds;
3905 4049 dsda.releasing = B_TRUE;
3906 4050 /* We already did the destroy_check */
3907 4051 dsl_dataset_destroy_sync(&dsda, tag, tx);
3908 4052 }
3909 4053 }
3910 4054
3911 4055 static int
3912 4056 dsl_dataset_user_release_one(const char *dsname, void *arg)
3913 4057 {
3914 4058 struct dsl_ds_holdarg *ha = arg;
3915 4059 struct dsl_ds_releasearg *ra;
3916 4060 dsl_dataset_t *ds;
3917 4061 int error;
3918 4062 void *dtag = ha->dstg;
3919 4063 char *name;
3920 4064 boolean_t own = B_FALSE;
3921 4065 boolean_t might_destroy;
3922 4066
3923 4067 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3924 4068 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3925 4069 error = dsl_dataset_hold(name, dtag, &ds);
3926 4070 strfree(name);
3927 4071 if (error == ENOENT && ha->recursive)
3928 4072 return (0);
3929 4073 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3930 4074 if (error)
3931 4075 return (error);
3932 4076
3933 4077 ha->gotone = B_TRUE;
3934 4078
3935 4079 ASSERT(dsl_dataset_is_snapshot(ds));
3936 4080
3937 4081 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3938 4082 if (error) {
3939 4083 dsl_dataset_rele(ds, dtag);
3940 4084 return (error);
3941 4085 }
3942 4086
3943 4087 if (might_destroy) {
3944 4088 #ifdef _KERNEL
3945 4089 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3946 4090 error = zfs_unmount_snap(name, NULL);
3947 4091 strfree(name);
3948 4092 if (error) {
3949 4093 dsl_dataset_rele(ds, dtag);
3950 4094 return (error);
3951 4095 }
3952 4096 #endif
3953 4097 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3954 4098 dsl_dataset_rele(ds, dtag);
3955 4099 return (EBUSY);
3956 4100 } else {
3957 4101 own = B_TRUE;
3958 4102 dsl_dataset_make_exclusive(ds, dtag);
3959 4103 }
3960 4104 }
3961 4105
3962 4106 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3963 4107 ra->ds = ds;
3964 4108 ra->htag = ha->htag;
3965 4109 ra->own = own;
3966 4110 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3967 4111 dsl_dataset_user_release_sync, ra, dtag, 0);
3968 4112
3969 4113 return (0);
3970 4114 }
3971 4115
3972 4116 int
3973 4117 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3974 4118 boolean_t recursive)
3975 4119 {
3976 4120 struct dsl_ds_holdarg *ha;
3977 4121 dsl_sync_task_t *dst;
3978 4122 spa_t *spa;
3979 4123 int error;
3980 4124
3981 4125 top:
3982 4126 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3983 4127
3984 4128 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3985 4129
3986 4130 error = spa_open(dsname, &spa, FTAG);
3987 4131 if (error) {
3988 4132 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3989 4133 return (error);
3990 4134 }
3991 4135
3992 4136 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3993 4137 ha->htag = htag;
3994 4138 ha->snapname = snapname;
3995 4139 ha->recursive = recursive;
3996 4140 if (recursive) {
3997 4141 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3998 4142 ha, DS_FIND_CHILDREN);
3999 4143 } else {
4000 4144 error = dsl_dataset_user_release_one(dsname, ha);
4001 4145 }
4002 4146 if (error == 0)
4003 4147 error = dsl_sync_task_group_wait(ha->dstg);
4004 4148
4005 4149 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4006 4150 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4007 4151 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4008 4152 dsl_dataset_t *ds = ra->ds;
4009 4153
4010 4154 if (dst->dst_err)
4011 4155 dsl_dataset_name(ds, ha->failed);
4012 4156
4013 4157 if (ra->own)
4014 4158 dsl_dataset_disown(ds, ha->dstg);
4015 4159 else
4016 4160 dsl_dataset_rele(ds, ha->dstg);
4017 4161
4018 4162 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4019 4163 }
4020 4164
4021 4165 if (error == 0 && recursive && !ha->gotone)
4022 4166 error = ENOENT;
4023 4167
4024 4168 if (error && error != EBUSY)
4025 4169 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4026 4170
4027 4171 dsl_sync_task_group_destroy(ha->dstg);
4028 4172 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4029 4173 spa_close(spa, FTAG);
4030 4174
4031 4175 /*
4032 4176 * We can get EBUSY if we were racing with deferred destroy and
4033 4177 * dsl_dataset_user_release_check() hadn't done the necessary
4034 4178 * open context setup. We can also get EBUSY if we're racing
4035 4179 * with destroy and that thread is the ds_owner. Either way
4036 4180 * the busy condition should be transient, and we should retry
4037 4181 * the release operation.
4038 4182 */
4039 4183 if (error == EBUSY)
4040 4184 goto top;
4041 4185
4042 4186 return (error);
4043 4187 }
4044 4188
4045 4189 /*
4046 4190 * Called at spa_load time (with retry == B_FALSE) to release a stale
4047 4191 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4048 4192 */
4049 4193 int
4050 4194 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4051 4195 boolean_t retry)
4052 4196 {
4053 4197 dsl_dataset_t *ds;
4054 4198 char *snap;
4055 4199 char *name;
4056 4200 int namelen;
4057 4201 int error;
4058 4202
4059 4203 do {
4060 4204 rw_enter(&dp->dp_config_rwlock, RW_READER);
4061 4205 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4062 4206 rw_exit(&dp->dp_config_rwlock);
4063 4207 if (error)
4064 4208 return (error);
4065 4209 namelen = dsl_dataset_namelen(ds)+1;
4066 4210 name = kmem_alloc(namelen, KM_SLEEP);
4067 4211 dsl_dataset_name(ds, name);
4068 4212 dsl_dataset_rele(ds, FTAG);
4069 4213
4070 4214 snap = strchr(name, '@');
4071 4215 *snap = '\0';
4072 4216 ++snap;
4073 4217 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4074 4218 kmem_free(name, namelen);
4075 4219
4076 4220 /*
4077 4221 * The object can't have been destroyed because we have a hold,
4078 4222 * but it might have been renamed, resulting in ENOENT. Retry
4079 4223 * if we've been requested to do so.
4080 4224 *
4081 4225 * It would be nice if we could use the dsobj all the way
4082 4226 * through and avoid ENOENT entirely. But we might need to
4083 4227 * unmount the snapshot, and there's currently no way to lookup
4084 4228 * a vfsp using a ZFS object id.
4085 4229 */
4086 4230 } while ((error == ENOENT) && retry);
4087 4231
4088 4232 return (error);
4089 4233 }
4090 4234
4091 4235 int
4092 4236 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4093 4237 {
4094 4238 dsl_dataset_t *ds;
4095 4239 int err;
4096 4240
4097 4241 err = dsl_dataset_hold(dsname, FTAG, &ds);
4098 4242 if (err)
4099 4243 return (err);
4100 4244
4101 4245 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4102 4246 if (ds->ds_phys->ds_userrefs_obj != 0) {
4103 4247 zap_attribute_t *za;
4104 4248 zap_cursor_t zc;
4105 4249
4106 4250 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4107 4251 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4108 4252 ds->ds_phys->ds_userrefs_obj);
4109 4253 zap_cursor_retrieve(&zc, za) == 0;
4110 4254 zap_cursor_advance(&zc)) {
4111 4255 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4112 4256 za->za_first_integer));
4113 4257 }
4114 4258 zap_cursor_fini(&zc);
4115 4259 kmem_free(za, sizeof (zap_attribute_t));
4116 4260 }
4117 4261 dsl_dataset_rele(ds, FTAG);
4118 4262 return (0);
4119 4263 }
4120 4264
4121 4265 /*
4122 4266 * Note, this function is used as the callback for dmu_objset_find(). We
4123 4267 * always return 0 so that we will continue to find and process
4124 4268 * inconsistent datasets, even if we encounter an error trying to
4125 4269 * process one of them.
4126 4270 */
4127 4271 /* ARGSUSED */
4128 4272 int
4129 4273 dsl_destroy_inconsistent(const char *dsname, void *arg)
4130 4274 {
4131 4275 dsl_dataset_t *ds;
4132 4276
4133 4277 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4134 4278 if (DS_IS_INCONSISTENT(ds))
4135 4279 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4136 4280 else
4137 4281 dsl_dataset_disown(ds, FTAG);
4138 4282 }
4139 4283 return (0);
4140 4284 }
4141 4285
4142 4286 /*
4143 4287 * Return (in *usedp) the amount of space written in new that is not
4144 4288 * present in oldsnap. New may be a snapshot or the head. Old must be
4145 4289 * a snapshot before new, in new's filesystem (or its origin). If not then
4146 4290 * fail and return EINVAL.
4147 4291 *
4148 4292 * The written space is calculated by considering two components: First, we
4149 4293 * ignore any freed space, and calculate the written as new's used space
4150 4294 * minus old's used space. Next, we add in the amount of space that was freed
4151 4295 * between the two snapshots, thus reducing new's used space relative to old's.
4152 4296 * Specifically, this is the space that was born before old->ds_creation_txg,
4153 4297 * and freed before new (ie. on new's deadlist or a previous deadlist).
4154 4298 *
4155 4299 * space freed [---------------------]
4156 4300 * snapshots ---O-------O--------O-------O------
4157 4301 * oldsnap new
4158 4302 */
4159 4303 int
4160 4304 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4161 4305 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4162 4306 {
4163 4307 int err = 0;
4164 4308 uint64_t snapobj;
4165 4309 dsl_pool_t *dp = new->ds_dir->dd_pool;
4166 4310
4167 4311 *usedp = 0;
4168 4312 *usedp += new->ds_phys->ds_referenced_bytes;
4169 4313 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4170 4314
4171 4315 *compp = 0;
4172 4316 *compp += new->ds_phys->ds_compressed_bytes;
4173 4317 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4174 4318
4175 4319 *uncompp = 0;
4176 4320 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4177 4321 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4178 4322
4179 4323 rw_enter(&dp->dp_config_rwlock, RW_READER);
4180 4324 snapobj = new->ds_object;
4181 4325 while (snapobj != oldsnap->ds_object) {
4182 4326 dsl_dataset_t *snap;
4183 4327 uint64_t used, comp, uncomp;
4184 4328
4185 4329 if (snapobj == new->ds_object) {
4186 4330 snap = new;
4187 4331 } else {
4188 4332 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4189 4333 if (err != 0)
4190 4334 break;
4191 4335 }
4192 4336
4193 4337 if (snap->ds_phys->ds_prev_snap_txg ==
4194 4338 oldsnap->ds_phys->ds_creation_txg) {
4195 4339 /*
4196 4340 * The blocks in the deadlist can not be born after
4197 4341 * ds_prev_snap_txg, so get the whole deadlist space,
4198 4342 * which is more efficient (especially for old-format
4199 4343 * deadlists). Unfortunately the deadlist code
4200 4344 * doesn't have enough information to make this
4201 4345 * optimization itself.
4202 4346 */
4203 4347 dsl_deadlist_space(&snap->ds_deadlist,
4204 4348 &used, &comp, &uncomp);
4205 4349 } else {
4206 4350 dsl_deadlist_space_range(&snap->ds_deadlist,
4207 4351 0, oldsnap->ds_phys->ds_creation_txg,
4208 4352 &used, &comp, &uncomp);
4209 4353 }
4210 4354 *usedp += used;
4211 4355 *compp += comp;
4212 4356 *uncompp += uncomp;
4213 4357
4214 4358 /*
4215 4359 * If we get to the beginning of the chain of snapshots
4216 4360 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4217 4361 * was not a snapshot of/before new.
4218 4362 */
4219 4363 snapobj = snap->ds_phys->ds_prev_snap_obj;
4220 4364 if (snap != new)
4221 4365 dsl_dataset_rele(snap, FTAG);
4222 4366 if (snapobj == 0) {
4223 4367 err = EINVAL;
4224 4368 break;
4225 4369 }
4226 4370
4227 4371 }
4228 4372 rw_exit(&dp->dp_config_rwlock);
4229 4373 return (err);
4230 4374 }
4231 4375
4232 4376 /*
4233 4377 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4234 4378 * lastsnap, and all snapshots in between are deleted.
4235 4379 *
4236 4380 * blocks that would be freed [---------------------------]
4237 4381 * snapshots ---O-------O--------O-------O--------O
4238 4382 * firstsnap lastsnap
4239 4383 *
4240 4384 * This is the set of blocks that were born after the snap before firstsnap,
4241 4385 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4242 4386 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4243 4387 * We calculate this by iterating over the relevant deadlists (from the snap
4244 4388 * after lastsnap, backward to the snap after firstsnap), summing up the
4245 4389 * space on the deadlist that was born after the snap before firstsnap.
4246 4390 */
4247 4391 int
4248 4392 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4249 4393 dsl_dataset_t *lastsnap,
4250 4394 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4251 4395 {
4252 4396 int err = 0;
4253 4397 uint64_t snapobj;
4254 4398 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4255 4399
4256 4400 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4257 4401 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4258 4402
4259 4403 /*
4260 4404 * Check that the snapshots are in the same dsl_dir, and firstsnap
4261 4405 * is before lastsnap.
4262 4406 */
4263 4407 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4264 4408 firstsnap->ds_phys->ds_creation_txg >
4265 4409 lastsnap->ds_phys->ds_creation_txg)
4266 4410 return (EINVAL);
4267 4411
4268 4412 *usedp = *compp = *uncompp = 0;
4269 4413
4270 4414 rw_enter(&dp->dp_config_rwlock, RW_READER);
4271 4415 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4272 4416 while (snapobj != firstsnap->ds_object) {
4273 4417 dsl_dataset_t *ds;
4274 4418 uint64_t used, comp, uncomp;
4275 4419
4276 4420 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4277 4421 if (err != 0)
4278 4422 break;
4279 4423
4280 4424 dsl_deadlist_space_range(&ds->ds_deadlist,
4281 4425 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4282 4426 &used, &comp, &uncomp);
4283 4427 *usedp += used;
4284 4428 *compp += comp;
4285 4429 *uncompp += uncomp;
4286 4430
4287 4431 snapobj = ds->ds_phys->ds_prev_snap_obj;
4288 4432 ASSERT3U(snapobj, !=, 0);
4289 4433 dsl_dataset_rele(ds, FTAG);
4290 4434 }
4291 4435 rw_exit(&dp->dp_config_rwlock);
4292 4436 return (err);
4293 4437 }
|
↓ open down ↓ |
1428 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX