Print this page
2882 implement libzfs_core
2883 changing "canmount" property to "on" should not always remount dataset
2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Chris Siden <christopher.siden@delphix.com>
Reviewed by: Garrett D'Amore <garrett@damore.org>
Reviewed by: Bill Pijewski <wdp@joyent.com>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Split |
Close |
Expand all |
Collapse all |
--- old/usr/src/uts/common/fs/zfs/dsl_dataset.c
+++ new/usr/src/uts/common/fs/zfs/dsl_dataset.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 25 */
26 26
27 27 #include <sys/dmu_objset.h>
28 28 #include <sys/dsl_dataset.h>
29 29 #include <sys/dsl_dir.h>
30 30 #include <sys/dsl_prop.h>
31 31 #include <sys/dsl_synctask.h>
32 32 #include <sys/dmu_traverse.h>
33 33 #include <sys/dmu_impl.h>
34 34 #include <sys/dmu_tx.h>
35 35 #include <sys/arc.h>
36 36 #include <sys/zio.h>
37 37 #include <sys/zap.h>
38 38 #include <sys/zfeature.h>
39 39 #include <sys/unique.h>
40 40 #include <sys/zfs_context.h>
41 41 #include <sys/zfs_ioctl.h>
42 42 #include <sys/spa.h>
43 43 #include <sys/zfs_znode.h>
44 44 #include <sys/zfs_onexit.h>
45 45 #include <sys/zvol.h>
46 46 #include <sys/dsl_scan.h>
47 47 #include <sys/dsl_deadlist.h>
48 48
49 49 static char *dsl_reaper = "the grim reaper";
50 50
51 51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
52 52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
53 53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
54 54
55 55 #define SWITCH64(x, y) \
56 56 { \
57 57 uint64_t __tmp = (x); \
58 58 (x) = (y); \
59 59 (y) = __tmp; \
60 60 }
61 61
62 62 #define DS_REF_MAX (1ULL << 62)
63 63
64 64 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
65 65
66 66 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67 67
68 68
69 69 /*
70 70 * Figure out how much of this delta should be propogated to the dsl_dir
71 71 * layer. If there's a refreservation, that space has already been
72 72 * partially accounted for in our ancestors.
73 73 */
74 74 static int64_t
75 75 parent_delta(dsl_dataset_t *ds, int64_t delta)
76 76 {
77 77 uint64_t old_bytes, new_bytes;
78 78
79 79 if (ds->ds_reserved == 0)
80 80 return (delta);
81 81
82 82 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
83 83 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
84 84
85 85 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
86 86 return (new_bytes - old_bytes);
87 87 }
88 88
89 89 void
90 90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
91 91 {
92 92 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
93 93 int compressed = BP_GET_PSIZE(bp);
94 94 int uncompressed = BP_GET_UCSIZE(bp);
95 95 int64_t delta;
96 96
97 97 dprintf_bp(bp, "ds=%p", ds);
98 98
99 99 ASSERT(dmu_tx_is_syncing(tx));
100 100 /* It could have been compressed away to nothing */
101 101 if (BP_IS_HOLE(bp))
102 102 return;
103 103 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
104 104 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
105 105 if (ds == NULL) {
106 106 /*
107 107 * Account for the meta-objset space in its placeholder
108 108 * dsl_dir.
109 109 */
110 110 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
111 111 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
112 112 used, compressed, uncompressed, tx);
113 113 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
114 114 return;
115 115 }
116 116 dmu_buf_will_dirty(ds->ds_dbuf, tx);
117 117
118 118 mutex_enter(&ds->ds_dir->dd_lock);
119 119 mutex_enter(&ds->ds_lock);
120 120 delta = parent_delta(ds, used);
121 121 ds->ds_phys->ds_referenced_bytes += used;
122 122 ds->ds_phys->ds_compressed_bytes += compressed;
123 123 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 124 ds->ds_phys->ds_unique_bytes += used;
125 125 mutex_exit(&ds->ds_lock);
126 126 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
127 127 compressed, uncompressed, tx);
128 128 dsl_dir_transfer_space(ds->ds_dir, used - delta,
129 129 DD_USED_REFRSRV, DD_USED_HEAD, tx);
130 130 mutex_exit(&ds->ds_dir->dd_lock);
131 131 }
132 132
133 133 int
134 134 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
135 135 boolean_t async)
136 136 {
137 137 if (BP_IS_HOLE(bp))
138 138 return (0);
139 139
140 140 ASSERT(dmu_tx_is_syncing(tx));
141 141 ASSERT(bp->blk_birth <= tx->tx_txg);
142 142
143 143 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
144 144 int compressed = BP_GET_PSIZE(bp);
145 145 int uncompressed = BP_GET_UCSIZE(bp);
146 146
147 147 ASSERT(used > 0);
148 148 if (ds == NULL) {
149 149 /*
150 150 * Account for the meta-objset space in its placeholder
151 151 * dataset.
152 152 */
153 153 dsl_free(tx->tx_pool, tx->tx_txg, bp);
154 154
155 155 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
156 156 -used, -compressed, -uncompressed, tx);
157 157 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
158 158 return (used);
159 159 }
160 160 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
161 161
162 162 ASSERT(!dsl_dataset_is_snapshot(ds));
163 163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
164 164
165 165 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
166 166 int64_t delta;
167 167
168 168 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
169 169 dsl_free(tx->tx_pool, tx->tx_txg, bp);
170 170
171 171 mutex_enter(&ds->ds_dir->dd_lock);
172 172 mutex_enter(&ds->ds_lock);
173 173 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
174 174 !DS_UNIQUE_IS_ACCURATE(ds));
175 175 delta = parent_delta(ds, -used);
176 176 ds->ds_phys->ds_unique_bytes -= used;
177 177 mutex_exit(&ds->ds_lock);
178 178 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
179 179 delta, -compressed, -uncompressed, tx);
180 180 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
181 181 DD_USED_REFRSRV, DD_USED_HEAD, tx);
182 182 mutex_exit(&ds->ds_dir->dd_lock);
183 183 } else {
184 184 dprintf_bp(bp, "putting on dead list: %s", "");
185 185 if (async) {
186 186 /*
187 187 * We are here as part of zio's write done callback,
188 188 * which means we're a zio interrupt thread. We can't
189 189 * call dsl_deadlist_insert() now because it may block
190 190 * waiting for I/O. Instead, put bp on the deferred
191 191 * queue and let dsl_pool_sync() finish the job.
192 192 */
193 193 bplist_append(&ds->ds_pending_deadlist, bp);
194 194 } else {
195 195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 196 }
197 197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 198 ds->ds_phys->ds_prev_snap_obj);
199 199 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
200 200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 201 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
202 202 ds->ds_object && bp->blk_birth >
203 203 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
204 204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 205 mutex_enter(&ds->ds_prev->ds_lock);
206 206 ds->ds_prev->ds_phys->ds_unique_bytes += used;
207 207 mutex_exit(&ds->ds_prev->ds_lock);
208 208 }
209 209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 210 dsl_dir_transfer_space(ds->ds_dir, used,
211 211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 212 }
213 213 }
214 214 mutex_enter(&ds->ds_lock);
215 215 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
216 216 ds->ds_phys->ds_referenced_bytes -= used;
217 217 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
218 218 ds->ds_phys->ds_compressed_bytes -= compressed;
219 219 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
220 220 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
221 221 mutex_exit(&ds->ds_lock);
222 222
223 223 return (used);
224 224 }
225 225
226 226 uint64_t
227 227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 228 {
229 229 uint64_t trysnap = 0;
230 230
231 231 if (ds == NULL)
232 232 return (0);
233 233 /*
234 234 * The snapshot creation could fail, but that would cause an
235 235 * incorrect FALSE return, which would only result in an
236 236 * overestimation of the amount of space that an operation would
237 237 * consume, which is OK.
238 238 *
239 239 * There's also a small window where we could miss a pending
240 240 * snapshot, because we could set the sync task in the quiescing
241 241 * phase. So this should only be used as a guess.
242 242 */
243 243 if (ds->ds_trysnap_txg >
244 244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 245 trysnap = ds->ds_trysnap_txg;
246 246 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
247 247 }
248 248
249 249 boolean_t
250 250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 251 uint64_t blk_birth)
252 252 {
253 253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
254 254 return (B_FALSE);
255 255
256 256 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
257 257
258 258 return (B_TRUE);
259 259 }
260 260
261 261 /* ARGSUSED */
262 262 static void
263 263 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
264 264 {
265 265 dsl_dataset_t *ds = dsv;
266 266
267 267 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
268 268
269 269 unique_remove(ds->ds_fsid_guid);
270 270
271 271 if (ds->ds_objset != NULL)
272 272 dmu_objset_evict(ds->ds_objset);
273 273
274 274 if (ds->ds_prev) {
275 275 dsl_dataset_drop_ref(ds->ds_prev, ds);
276 276 ds->ds_prev = NULL;
277 277 }
278 278
279 279 bplist_destroy(&ds->ds_pending_deadlist);
280 280 if (db != NULL) {
281 281 dsl_deadlist_close(&ds->ds_deadlist);
282 282 } else {
283 283 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
284 284 ASSERT(!ds->ds_deadlist.dl_oldfmt);
285 285 }
286 286 if (ds->ds_dir)
287 287 dsl_dir_close(ds->ds_dir, ds);
288 288
289 289 ASSERT(!list_link_active(&ds->ds_synced_link));
290 290
291 291 mutex_destroy(&ds->ds_lock);
292 292 mutex_destroy(&ds->ds_recvlock);
293 293 mutex_destroy(&ds->ds_opening_lock);
294 294 rw_destroy(&ds->ds_rwlock);
295 295 cv_destroy(&ds->ds_exclusive_cv);
296 296
297 297 kmem_free(ds, sizeof (dsl_dataset_t));
298 298 }
299 299
300 300 static int
301 301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
302 302 {
303 303 dsl_dataset_phys_t *headphys;
304 304 int err;
305 305 dmu_buf_t *headdbuf;
306 306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 307 objset_t *mos = dp->dp_meta_objset;
308 308
309 309 if (ds->ds_snapname[0])
310 310 return (0);
311 311 if (ds->ds_phys->ds_next_snap_obj == 0)
312 312 return (0);
313 313
314 314 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
315 315 FTAG, &headdbuf);
316 316 if (err)
317 317 return (err);
318 318 headphys = headdbuf->db_data;
319 319 err = zap_value_search(dp->dp_meta_objset,
320 320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 321 dmu_buf_rele(headdbuf, FTAG);
322 322 return (err);
323 323 }
324 324
325 325 static int
326 326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
327 327 {
328 328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
330 330 matchtype_t mt;
331 331 int err;
332 332
333 333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
334 334 mt = MT_FIRST;
335 335 else
336 336 mt = MT_EXACT;
337 337
338 338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 339 value, mt, NULL, 0, NULL);
340 340 if (err == ENOTSUP && mt == MT_FIRST)
341 341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
342 342 return (err);
343 343 }
344 344
345 345 static int
346 346 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
347 347 {
348 348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
350 350 matchtype_t mt;
351 351 int err;
352 352
353 353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354 354
355 355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
356 356 mt = MT_FIRST;
357 357 else
358 358 mt = MT_EXACT;
359 359
360 360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
361 361 if (err == ENOTSUP && mt == MT_FIRST)
362 362 err = zap_remove(mos, snapobj, name, tx);
363 363 return (err);
364 364 }
365 365
366 366 static int
367 367 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
368 368 dsl_dataset_t **dsp)
369 369 {
370 370 objset_t *mos = dp->dp_meta_objset;
371 371 dmu_buf_t *dbuf;
372 372 dsl_dataset_t *ds;
373 373 int err;
374 374 dmu_object_info_t doi;
375 375
376 376 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
377 377 dsl_pool_sync_context(dp));
378 378
379 379 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
380 380 if (err)
381 381 return (err);
382 382
383 383 /* Make sure dsobj has the correct object type. */
384 384 dmu_object_info_from_db(dbuf, &doi);
385 385 if (doi.doi_type != DMU_OT_DSL_DATASET)
386 386 return (EINVAL);
387 387
388 388 ds = dmu_buf_get_user(dbuf);
389 389 if (ds == NULL) {
390 390 dsl_dataset_t *winner;
391 391
392 392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
393 393 ds->ds_dbuf = dbuf;
394 394 ds->ds_object = dsobj;
395 395 ds->ds_phys = dbuf->db_data;
396 396
397 397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
398 398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
399 399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
400 400 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
401 401
402 402 rw_init(&ds->ds_rwlock, 0, 0, 0);
403 403 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
404 404
405 405 bplist_create(&ds->ds_pending_deadlist);
406 406 dsl_deadlist_open(&ds->ds_deadlist,
407 407 mos, ds->ds_phys->ds_deadlist_obj);
408 408
409 409 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
410 410 offsetof(dmu_sendarg_t, dsa_link));
411 411
412 412 if (err == 0) {
413 413 err = dsl_dir_open_obj(dp,
414 414 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
415 415 }
416 416 if (err) {
417 417 mutex_destroy(&ds->ds_lock);
418 418 mutex_destroy(&ds->ds_recvlock);
419 419 mutex_destroy(&ds->ds_opening_lock);
420 420 rw_destroy(&ds->ds_rwlock);
421 421 cv_destroy(&ds->ds_exclusive_cv);
422 422 bplist_destroy(&ds->ds_pending_deadlist);
423 423 dsl_deadlist_close(&ds->ds_deadlist);
424 424 kmem_free(ds, sizeof (dsl_dataset_t));
425 425 dmu_buf_rele(dbuf, tag);
426 426 return (err);
427 427 }
428 428
429 429 if (!dsl_dataset_is_snapshot(ds)) {
430 430 ds->ds_snapname[0] = '\0';
431 431 if (ds->ds_phys->ds_prev_snap_obj) {
432 432 err = dsl_dataset_get_ref(dp,
433 433 ds->ds_phys->ds_prev_snap_obj,
434 434 ds, &ds->ds_prev);
435 435 }
436 436 } else {
437 437 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
438 438 err = dsl_dataset_get_snapname(ds);
439 439 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
440 440 err = zap_count(
441 441 ds->ds_dir->dd_pool->dp_meta_objset,
442 442 ds->ds_phys->ds_userrefs_obj,
443 443 &ds->ds_userrefs);
444 444 }
445 445 }
446 446
447 447 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
448 448 /*
449 449 * In sync context, we're called with either no lock
450 450 * or with the write lock. If we're not syncing,
451 451 * we're always called with the read lock held.
452 452 */
453 453 boolean_t need_lock =
454 454 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
455 455 dsl_pool_sync_context(dp);
456 456
457 457 if (need_lock)
458 458 rw_enter(&dp->dp_config_rwlock, RW_READER);
459 459
460 460 err = dsl_prop_get_ds(ds,
461 461 "refreservation", sizeof (uint64_t), 1,
462 462 &ds->ds_reserved, NULL);
463 463 if (err == 0) {
464 464 err = dsl_prop_get_ds(ds,
465 465 "refquota", sizeof (uint64_t), 1,
466 466 &ds->ds_quota, NULL);
467 467 }
468 468
469 469 if (need_lock)
470 470 rw_exit(&dp->dp_config_rwlock);
471 471 } else {
472 472 ds->ds_reserved = ds->ds_quota = 0;
473 473 }
474 474
475 475 if (err == 0) {
476 476 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
477 477 dsl_dataset_evict);
478 478 }
479 479 if (err || winner) {
480 480 bplist_destroy(&ds->ds_pending_deadlist);
481 481 dsl_deadlist_close(&ds->ds_deadlist);
482 482 if (ds->ds_prev)
483 483 dsl_dataset_drop_ref(ds->ds_prev, ds);
484 484 dsl_dir_close(ds->ds_dir, ds);
485 485 mutex_destroy(&ds->ds_lock);
486 486 mutex_destroy(&ds->ds_recvlock);
487 487 mutex_destroy(&ds->ds_opening_lock);
488 488 rw_destroy(&ds->ds_rwlock);
489 489 cv_destroy(&ds->ds_exclusive_cv);
490 490 kmem_free(ds, sizeof (dsl_dataset_t));
491 491 if (err) {
492 492 dmu_buf_rele(dbuf, tag);
493 493 return (err);
494 494 }
495 495 ds = winner;
496 496 } else {
497 497 ds->ds_fsid_guid =
498 498 unique_insert(ds->ds_phys->ds_fsid_guid);
499 499 }
500 500 }
501 501 ASSERT3P(ds->ds_dbuf, ==, dbuf);
502 502 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
503 503 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
504 504 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
505 505 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
506 506 mutex_enter(&ds->ds_lock);
507 507 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
508 508 mutex_exit(&ds->ds_lock);
509 509 dmu_buf_rele(ds->ds_dbuf, tag);
510 510 return (ENOENT);
511 511 }
512 512 mutex_exit(&ds->ds_lock);
513 513 *dsp = ds;
514 514 return (0);
515 515 }
516 516
517 517 static int
518 518 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
519 519 {
520 520 dsl_pool_t *dp = ds->ds_dir->dd_pool;
521 521
522 522 /*
523 523 * In syncing context we don't want the rwlock lock: there
524 524 * may be an existing writer waiting for sync phase to
525 525 * finish. We don't need to worry about such writers, since
526 526 * sync phase is single-threaded, so the writer can't be
527 527 * doing anything while we are active.
528 528 */
529 529 if (dsl_pool_sync_context(dp)) {
530 530 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
531 531 return (0);
532 532 }
533 533
534 534 /*
535 535 * Normal users will hold the ds_rwlock as a READER until they
536 536 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
537 537 * drop their READER lock after they set the ds_owner field.
538 538 *
539 539 * If the dataset is being destroyed, the destroy thread will
540 540 * obtain a WRITER lock for exclusive access after it's done its
541 541 * open-context work and then change the ds_owner to
542 542 * dsl_reaper once destruction is assured. So threads
543 543 * may block here temporarily, until the "destructability" of
544 544 * the dataset is determined.
545 545 */
546 546 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
547 547 mutex_enter(&ds->ds_lock);
548 548 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
549 549 rw_exit(&dp->dp_config_rwlock);
550 550 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
551 551 if (DSL_DATASET_IS_DESTROYED(ds)) {
552 552 mutex_exit(&ds->ds_lock);
553 553 dsl_dataset_drop_ref(ds, tag);
554 554 rw_enter(&dp->dp_config_rwlock, RW_READER);
555 555 return (ENOENT);
556 556 }
557 557 /*
558 558 * The dp_config_rwlock lives above the ds_lock. And
559 559 * we need to check DSL_DATASET_IS_DESTROYED() while
560 560 * holding the ds_lock, so we have to drop and reacquire
561 561 * the ds_lock here.
562 562 */
563 563 mutex_exit(&ds->ds_lock);
564 564 rw_enter(&dp->dp_config_rwlock, RW_READER);
565 565 mutex_enter(&ds->ds_lock);
566 566 }
567 567 mutex_exit(&ds->ds_lock);
568 568 return (0);
569 569 }
570 570
571 571 int
572 572 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
573 573 dsl_dataset_t **dsp)
574 574 {
575 575 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
576 576
577 577 if (err)
578 578 return (err);
579 579 return (dsl_dataset_hold_ref(*dsp, tag));
580 580 }
581 581
582 582 int
583 583 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
584 584 void *tag, dsl_dataset_t **dsp)
585 585 {
586 586 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
587 587 if (err)
588 588 return (err);
589 589 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
590 590 dsl_dataset_rele(*dsp, tag);
591 591 *dsp = NULL;
592 592 return (EBUSY);
593 593 }
594 594 return (0);
595 595 }
596 596
597 597 int
598 598 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
599 599 {
600 600 dsl_dir_t *dd;
601 601 dsl_pool_t *dp;
602 602 const char *snapname;
603 603 uint64_t obj;
604 604 int err = 0;
605 605
606 606 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
607 607 if (err)
608 608 return (err);
609 609
610 610 dp = dd->dd_pool;
611 611 obj = dd->dd_phys->dd_head_dataset_obj;
612 612 rw_enter(&dp->dp_config_rwlock, RW_READER);
613 613 if (obj)
614 614 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
615 615 else
616 616 err = ENOENT;
617 617 if (err)
618 618 goto out;
619 619
620 620 err = dsl_dataset_hold_ref(*dsp, tag);
621 621
622 622 /* we may be looking for a snapshot */
623 623 if (err == 0 && snapname != NULL) {
624 624 dsl_dataset_t *ds = NULL;
625 625
626 626 if (*snapname++ != '@') {
627 627 dsl_dataset_rele(*dsp, tag);
628 628 err = ENOENT;
629 629 goto out;
630 630 }
631 631
632 632 dprintf("looking for snapshot '%s'\n", snapname);
633 633 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
634 634 if (err == 0)
635 635 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
636 636 dsl_dataset_rele(*dsp, tag);
637 637
638 638 ASSERT3U((err == 0), ==, (ds != NULL));
639 639
640 640 if (ds) {
641 641 mutex_enter(&ds->ds_lock);
642 642 if (ds->ds_snapname[0] == 0)
643 643 (void) strlcpy(ds->ds_snapname, snapname,
644 644 sizeof (ds->ds_snapname));
645 645 mutex_exit(&ds->ds_lock);
646 646 err = dsl_dataset_hold_ref(ds, tag);
647 647 *dsp = err ? NULL : ds;
648 648 }
649 649 }
650 650 out:
651 651 rw_exit(&dp->dp_config_rwlock);
652 652 dsl_dir_close(dd, FTAG);
653 653 return (err);
654 654 }
655 655
656 656 int
657 657 dsl_dataset_own(const char *name, boolean_t inconsistentok,
658 658 void *tag, dsl_dataset_t **dsp)
659 659 {
660 660 int err = dsl_dataset_hold(name, tag, dsp);
661 661 if (err)
662 662 return (err);
663 663 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
664 664 dsl_dataset_rele(*dsp, tag);
665 665 return (EBUSY);
666 666 }
667 667 return (0);
668 668 }
669 669
670 670 void
671 671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
672 672 {
673 673 if (ds == NULL) {
674 674 (void) strcpy(name, "mos");
675 675 } else {
676 676 dsl_dir_name(ds->ds_dir, name);
677 677 VERIFY(0 == dsl_dataset_get_snapname(ds));
678 678 if (ds->ds_snapname[0]) {
679 679 (void) strcat(name, "@");
680 680 /*
681 681 * We use a "recursive" mutex so that we
682 682 * can call dprintf_ds() with ds_lock held.
683 683 */
684 684 if (!MUTEX_HELD(&ds->ds_lock)) {
685 685 mutex_enter(&ds->ds_lock);
686 686 (void) strcat(name, ds->ds_snapname);
687 687 mutex_exit(&ds->ds_lock);
688 688 } else {
689 689 (void) strcat(name, ds->ds_snapname);
690 690 }
691 691 }
692 692 }
693 693 }
694 694
695 695 static int
696 696 dsl_dataset_namelen(dsl_dataset_t *ds)
697 697 {
698 698 int result;
699 699
700 700 if (ds == NULL) {
701 701 result = 3; /* "mos" */
702 702 } else {
703 703 result = dsl_dir_namelen(ds->ds_dir);
704 704 VERIFY(0 == dsl_dataset_get_snapname(ds));
705 705 if (ds->ds_snapname[0]) {
706 706 ++result; /* adding one for the @-sign */
707 707 if (!MUTEX_HELD(&ds->ds_lock)) {
708 708 mutex_enter(&ds->ds_lock);
709 709 result += strlen(ds->ds_snapname);
710 710 mutex_exit(&ds->ds_lock);
711 711 } else {
712 712 result += strlen(ds->ds_snapname);
713 713 }
714 714 }
715 715 }
716 716
717 717 return (result);
718 718 }
719 719
720 720 void
721 721 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
722 722 {
723 723 dmu_buf_rele(ds->ds_dbuf, tag);
724 724 }
725 725
726 726 void
727 727 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
728 728 {
729 729 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
730 730 rw_exit(&ds->ds_rwlock);
731 731 }
732 732 dsl_dataset_drop_ref(ds, tag);
733 733 }
734 734
735 735 void
736 736 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
737 737 {
738 738 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
739 739 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
740 740
741 741 mutex_enter(&ds->ds_lock);
742 742 ds->ds_owner = NULL;
743 743 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
744 744 rw_exit(&ds->ds_rwlock);
745 745 cv_broadcast(&ds->ds_exclusive_cv);
746 746 }
747 747 mutex_exit(&ds->ds_lock);
748 748 if (ds->ds_dbuf)
749 749 dsl_dataset_drop_ref(ds, tag);
750 750 else
751 751 dsl_dataset_evict(NULL, ds);
752 752 }
753 753
754 754 boolean_t
755 755 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
756 756 {
757 757 boolean_t gotit = FALSE;
758 758
759 759 mutex_enter(&ds->ds_lock);
760 760 if (ds->ds_owner == NULL &&
761 761 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
762 762 ds->ds_owner = tag;
763 763 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
764 764 rw_exit(&ds->ds_rwlock);
765 765 gotit = TRUE;
766 766 }
767 767 mutex_exit(&ds->ds_lock);
768 768 return (gotit);
769 769 }
770 770
771 771 void
772 772 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
773 773 {
774 774 ASSERT3P(owner, ==, ds->ds_owner);
775 775 if (!RW_WRITE_HELD(&ds->ds_rwlock))
776 776 rw_enter(&ds->ds_rwlock, RW_WRITER);
777 777 }
778 778
779 779 uint64_t
780 780 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
781 781 uint64_t flags, dmu_tx_t *tx)
782 782 {
783 783 dsl_pool_t *dp = dd->dd_pool;
784 784 dmu_buf_t *dbuf;
785 785 dsl_dataset_phys_t *dsphys;
786 786 uint64_t dsobj;
787 787 objset_t *mos = dp->dp_meta_objset;
788 788
789 789 if (origin == NULL)
790 790 origin = dp->dp_origin_snap;
791 791
792 792 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
793 793 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
794 794 ASSERT(dmu_tx_is_syncing(tx));
795 795 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
796 796
797 797 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
798 798 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
799 799 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
800 800 dmu_buf_will_dirty(dbuf, tx);
801 801 dsphys = dbuf->db_data;
802 802 bzero(dsphys, sizeof (dsl_dataset_phys_t));
803 803 dsphys->ds_dir_obj = dd->dd_object;
804 804 dsphys->ds_flags = flags;
805 805 dsphys->ds_fsid_guid = unique_create();
806 806 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
807 807 sizeof (dsphys->ds_guid));
808 808 dsphys->ds_snapnames_zapobj =
809 809 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
810 810 DMU_OT_NONE, 0, tx);
811 811 dsphys->ds_creation_time = gethrestime_sec();
812 812 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
813 813
814 814 if (origin == NULL) {
815 815 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
816 816 } else {
817 817 dsl_dataset_t *ohds;
818 818
819 819 dsphys->ds_prev_snap_obj = origin->ds_object;
820 820 dsphys->ds_prev_snap_txg =
821 821 origin->ds_phys->ds_creation_txg;
822 822 dsphys->ds_referenced_bytes =
823 823 origin->ds_phys->ds_referenced_bytes;
824 824 dsphys->ds_compressed_bytes =
825 825 origin->ds_phys->ds_compressed_bytes;
826 826 dsphys->ds_uncompressed_bytes =
827 827 origin->ds_phys->ds_uncompressed_bytes;
828 828 dsphys->ds_bp = origin->ds_phys->ds_bp;
829 829 dsphys->ds_flags |= origin->ds_phys->ds_flags;
830 830
831 831 dmu_buf_will_dirty(origin->ds_dbuf, tx);
832 832 origin->ds_phys->ds_num_children++;
833 833
834 834 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
835 835 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
836 836 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
837 837 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
838 838 dsl_dataset_rele(ohds, FTAG);
839 839
840 840 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
841 841 if (origin->ds_phys->ds_next_clones_obj == 0) {
842 842 origin->ds_phys->ds_next_clones_obj =
843 843 zap_create(mos,
844 844 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
845 845 }
846 846 VERIFY(0 == zap_add_int(mos,
847 847 origin->ds_phys->ds_next_clones_obj,
848 848 dsobj, tx));
849 849 }
850 850
851 851 dmu_buf_will_dirty(dd->dd_dbuf, tx);
852 852 dd->dd_phys->dd_origin_obj = origin->ds_object;
853 853 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
854 854 if (origin->ds_dir->dd_phys->dd_clones == 0) {
855 855 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
856 856 origin->ds_dir->dd_phys->dd_clones =
857 857 zap_create(mos,
858 858 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
859 859 }
860 860 VERIFY3U(0, ==, zap_add_int(mos,
861 861 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
862 862 }
863 863 }
864 864
865 865 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
866 866 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867 867
868 868 dmu_buf_rele(dbuf, FTAG);
869 869
870 870 dmu_buf_will_dirty(dd->dd_dbuf, tx);
871 871 dd->dd_phys->dd_head_dataset_obj = dsobj;
872 872
873 873 return (dsobj);
874 874 }
875 875
876 876 uint64_t
877 877 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
878 878 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
879 879 {
880 880 dsl_pool_t *dp = pdd->dd_pool;
881 881 uint64_t dsobj, ddobj;
882 882 dsl_dir_t *dd;
883 883
884 884 ASSERT(lastname[0] != '@');
885 885
886 886 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
887 887 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
888 888
889 889 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
890 890
891 891 dsl_deleg_set_create_perms(dd, tx, cr);
892 892
893 893 dsl_dir_close(dd, FTAG);
894 894
895 895 /*
896 896 * If we are creating a clone, make sure we zero out any stale
897 897 * data from the origin snapshots zil header.
898 898 */
899 899 if (origin != NULL) {
900 900 dsl_dataset_t *ds;
901 901 objset_t *os;
902 902
903 903 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
904 904 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
905 905 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
906 906 dsl_dataset_dirty(ds, tx);
↓ open down ↓ |
906 lines elided |
↑ open up ↑ |
907 907 dsl_dataset_rele(ds, FTAG);
908 908 }
909 909
910 910 return (dsobj);
911 911 }
912 912
913 913 /*
914 914 * The snapshots must all be in the same pool.
915 915 */
916 916 int
917 -dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
917 +dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer,
918 + nvlist_t *errlist)
918 919 {
919 920 int err;
920 921 dsl_sync_task_t *dst;
921 922 spa_t *spa;
922 923 nvpair_t *pair;
923 924 dsl_sync_task_group_t *dstg;
924 925
925 926 pair = nvlist_next_nvpair(snaps, NULL);
926 927 if (pair == NULL)
927 928 return (0);
928 929
929 930 err = spa_open(nvpair_name(pair), &spa, FTAG);
930 931 if (err)
931 932 return (err);
932 933 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
933 934
934 935 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
935 936 pair = nvlist_next_nvpair(snaps, pair)) {
936 937 dsl_dataset_t *ds;
937 938
938 939 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
939 940 if (err == 0) {
940 941 struct dsl_ds_destroyarg *dsda;
941 942
↓ open down ↓ |
14 lines elided |
↑ open up ↑ |
942 943 dsl_dataset_make_exclusive(ds, dstg);
943 944 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
944 945 KM_SLEEP);
945 946 dsda->ds = ds;
946 947 dsda->defer = defer;
947 948 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
948 949 dsl_dataset_destroy_sync, dsda, dstg, 0);
949 950 } else if (err == ENOENT) {
950 951 err = 0;
951 952 } else {
952 - (void) strcpy(failed, nvpair_name(pair));
953 + fnvlist_add_int32(errlist, nvpair_name(pair), err);
953 954 break;
954 955 }
955 956 }
956 957
957 958 if (err == 0)
958 959 err = dsl_sync_task_group_wait(dstg);
959 960
960 961 for (dst = list_head(&dstg->dstg_tasks); dst;
961 962 dst = list_next(&dstg->dstg_tasks, dst)) {
962 963 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
963 964 dsl_dataset_t *ds = dsda->ds;
964 965
965 966 /*
966 - * Return the file system name that triggered the error
967 + * Return the snapshots that triggered the error.
967 968 */
968 - if (dst->dst_err) {
969 - dsl_dataset_name(ds, failed);
969 + if (dst->dst_err != 0) {
970 + char name[ZFS_MAXNAMELEN];
971 + dsl_dataset_name(ds, name);
972 + fnvlist_add_int32(errlist, name, dst->dst_err);
970 973 }
971 974 ASSERT3P(dsda->rm_origin, ==, NULL);
972 975 dsl_dataset_disown(ds, dstg);
973 976 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
974 977 }
975 978
976 979 dsl_sync_task_group_destroy(dstg);
977 980 spa_close(spa, FTAG);
978 981 return (err);
979 982
980 983 }
981 984
982 985 static boolean_t
983 986 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
984 987 {
985 988 boolean_t might_destroy = B_FALSE;
986 989
987 990 mutex_enter(&ds->ds_lock);
988 991 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
989 992 DS_IS_DEFER_DESTROY(ds))
990 993 might_destroy = B_TRUE;
991 994 mutex_exit(&ds->ds_lock);
992 995
993 996 return (might_destroy);
994 997 }
995 998
996 999 /*
997 1000 * If we're removing a clone, and these three conditions are true:
998 1001 * 1) the clone's origin has no other children
999 1002 * 2) the clone's origin has no user references
1000 1003 * 3) the clone's origin has been marked for deferred destruction
1001 1004 * Then, prepare to remove the origin as part of this sync task group.
1002 1005 */
1003 1006 static int
1004 1007 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1005 1008 {
1006 1009 dsl_dataset_t *ds = dsda->ds;
1007 1010 dsl_dataset_t *origin = ds->ds_prev;
1008 1011
1009 1012 if (dsl_dataset_might_destroy_origin(origin)) {
1010 1013 char *name;
1011 1014 int namelen;
1012 1015 int error;
1013 1016
1014 1017 namelen = dsl_dataset_namelen(origin) + 1;
1015 1018 name = kmem_alloc(namelen, KM_SLEEP);
1016 1019 dsl_dataset_name(origin, name);
1017 1020 #ifdef _KERNEL
1018 1021 error = zfs_unmount_snap(name, NULL);
1019 1022 if (error) {
1020 1023 kmem_free(name, namelen);
1021 1024 return (error);
1022 1025 }
1023 1026 #endif
1024 1027 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1025 1028 kmem_free(name, namelen);
1026 1029 if (error)
1027 1030 return (error);
1028 1031 dsda->rm_origin = origin;
1029 1032 dsl_dataset_make_exclusive(origin, tag);
1030 1033 }
1031 1034
1032 1035 return (0);
1033 1036 }
1034 1037
1035 1038 /*
1036 1039 * ds must be opened as OWNER. On return (whether successful or not),
1037 1040 * ds will be closed and caller can no longer dereference it.
↓ open down ↓ |
58 lines elided |
↑ open up ↑ |
1038 1041 */
1039 1042 int
1040 1043 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1041 1044 {
1042 1045 int err;
1043 1046 dsl_sync_task_group_t *dstg;
1044 1047 objset_t *os;
1045 1048 dsl_dir_t *dd;
1046 1049 uint64_t obj;
1047 1050 struct dsl_ds_destroyarg dsda = { 0 };
1048 - dsl_dataset_t dummy_ds = { 0 };
1049 1051
1050 1052 dsda.ds = ds;
1051 1053
1052 1054 if (dsl_dataset_is_snapshot(ds)) {
1053 1055 /* Destroying a snapshot is simpler */
1054 1056 dsl_dataset_make_exclusive(ds, tag);
1055 1057
1056 1058 dsda.defer = defer;
1057 1059 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1058 1060 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1059 1061 &dsda, tag, 0);
1060 1062 ASSERT3P(dsda.rm_origin, ==, NULL);
1061 1063 goto out;
1062 1064 } else if (defer) {
1063 1065 err = EINVAL;
1064 1066 goto out;
1065 1067 }
1066 1068
1067 1069 dd = ds->ds_dir;
1068 - dummy_ds.ds_dir = dd;
1069 - dummy_ds.ds_object = ds->ds_object;
1070 1070
1071 1071 /*
1072 1072 * Check for errors and mark this ds as inconsistent, in
1073 1073 * case we crash while freeing the objects.
1074 1074 */
1075 1075 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1076 1076 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1077 1077 if (err)
1078 1078 goto out;
1079 1079
1080 1080 err = dmu_objset_from_ds(ds, &os);
1081 1081 if (err)
1082 1082 goto out;
1083 1083
1084 1084 /*
1085 1085 * If async destruction is not enabled try to remove all objects
1086 1086 * while in the open context so that there is less work to do in
1087 1087 * the syncing context.
1088 1088 */
1089 1089 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1090 1090 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1091 1091 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1092 1092 ds->ds_phys->ds_prev_snap_txg)) {
1093 1093 /*
1094 1094 * Ignore errors, if there is not enough disk space
1095 1095 * we will deal with it in dsl_dataset_destroy_sync().
1096 1096 */
1097 1097 (void) dmu_free_object(os, obj);
1098 1098 }
1099 1099 if (err != ESRCH)
1100 1100 goto out;
1101 1101 }
1102 1102
1103 1103 /*
1104 1104 * Only the ZIL knows how to free log blocks.
1105 1105 */
1106 1106 zil_destroy(dmu_objset_zil(os), B_FALSE);
1107 1107
1108 1108 /*
1109 1109 * Sync out all in-flight IO.
1110 1110 */
1111 1111 txg_wait_synced(dd->dd_pool, 0);
1112 1112
1113 1113 /*
1114 1114 * If we managed to free all the objects in open
1115 1115 * context, the user space accounting should be zero.
1116 1116 */
1117 1117 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1118 1118 dmu_objset_userused_enabled(os)) {
1119 1119 uint64_t count;
1120 1120
1121 1121 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1122 1122 count == 0);
1123 1123 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1124 1124 count == 0);
1125 1125 }
1126 1126
1127 1127 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1128 1128 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1129 1129 rw_exit(&dd->dd_pool->dp_config_rwlock);
1130 1130
1131 1131 if (err)
1132 1132 goto out;
1133 1133
1134 1134 /*
1135 1135 * Blow away the dsl_dir + head dataset.
1136 1136 */
1137 1137 dsl_dataset_make_exclusive(ds, tag);
1138 1138 /*
1139 1139 * If we're removing a clone, we might also need to remove its
1140 1140 * origin.
1141 1141 */
1142 1142 do {
1143 1143 dsda.need_prep = B_FALSE;
1144 1144 if (dsl_dir_is_clone(dd)) {
1145 1145 err = dsl_dataset_origin_rm_prep(&dsda, tag);
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
1146 1146 if (err) {
1147 1147 dsl_dir_close(dd, FTAG);
1148 1148 goto out;
1149 1149 }
1150 1150 }
1151 1151
1152 1152 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1153 1153 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1154 1154 dsl_dataset_destroy_sync, &dsda, tag, 0);
1155 1155 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1156 - dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1156 + dsl_dir_destroy_sync, dd, FTAG, 0);
1157 1157 err = dsl_sync_task_group_wait(dstg);
1158 1158 dsl_sync_task_group_destroy(dstg);
1159 1159
1160 1160 /*
1161 1161 * We could be racing against 'zfs release' or 'zfs destroy -d'
1162 1162 * on the origin snap, in which case we can get EBUSY if we
1163 1163 * needed to destroy the origin snap but were not ready to
1164 1164 * do so.
1165 1165 */
1166 1166 if (dsda.need_prep) {
1167 1167 ASSERT(err == EBUSY);
1168 1168 ASSERT(dsl_dir_is_clone(dd));
1169 1169 ASSERT(dsda.rm_origin == NULL);
1170 1170 }
1171 1171 } while (dsda.need_prep);
1172 1172
1173 1173 if (dsda.rm_origin != NULL)
1174 1174 dsl_dataset_disown(dsda.rm_origin, tag);
1175 1175
1176 1176 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1177 1177 if (err)
1178 1178 dsl_dir_close(dd, FTAG);
1179 1179 out:
1180 1180 dsl_dataset_disown(ds, tag);
1181 1181 return (err);
1182 1182 }
1183 1183
1184 1184 blkptr_t *
1185 1185 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1186 1186 {
1187 1187 return (&ds->ds_phys->ds_bp);
1188 1188 }
1189 1189
1190 1190 void
1191 1191 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1192 1192 {
1193 1193 ASSERT(dmu_tx_is_syncing(tx));
1194 1194 /* If it's the meta-objset, set dp_meta_rootbp */
1195 1195 if (ds == NULL) {
1196 1196 tx->tx_pool->dp_meta_rootbp = *bp;
1197 1197 } else {
1198 1198 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1199 1199 ds->ds_phys->ds_bp = *bp;
1200 1200 }
1201 1201 }
1202 1202
1203 1203 spa_t *
1204 1204 dsl_dataset_get_spa(dsl_dataset_t *ds)
1205 1205 {
1206 1206 return (ds->ds_dir->dd_pool->dp_spa);
1207 1207 }
1208 1208
1209 1209 void
1210 1210 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1211 1211 {
1212 1212 dsl_pool_t *dp;
1213 1213
1214 1214 if (ds == NULL) /* this is the meta-objset */
1215 1215 return;
1216 1216
1217 1217 ASSERT(ds->ds_objset != NULL);
1218 1218
1219 1219 if (ds->ds_phys->ds_next_snap_obj != 0)
1220 1220 panic("dirtying snapshot!");
1221 1221
1222 1222 dp = ds->ds_dir->dd_pool;
1223 1223
1224 1224 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1225 1225 /* up the hold count until we can be written out */
1226 1226 dmu_buf_add_ref(ds->ds_dbuf, ds);
1227 1227 }
1228 1228 }
1229 1229
1230 1230 /*
1231 1231 * The unique space in the head dataset can be calculated by subtracting
1232 1232 * the space used in the most recent snapshot, that is still being used
1233 1233 * in this file system, from the space currently in use. To figure out
1234 1234 * the space in the most recent snapshot still in use, we need to take
1235 1235 * the total space used in the snapshot and subtract out the space that
1236 1236 * has been freed up since the snapshot was taken.
1237 1237 */
1238 1238 static void
1239 1239 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1240 1240 {
1241 1241 uint64_t mrs_used;
1242 1242 uint64_t dlused, dlcomp, dluncomp;
1243 1243
1244 1244 ASSERT(!dsl_dataset_is_snapshot(ds));
1245 1245
1246 1246 if (ds->ds_phys->ds_prev_snap_obj != 0)
1247 1247 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1248 1248 else
1249 1249 mrs_used = 0;
1250 1250
1251 1251 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1252 1252
1253 1253 ASSERT3U(dlused, <=, mrs_used);
1254 1254 ds->ds_phys->ds_unique_bytes =
1255 1255 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1256 1256
1257 1257 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1258 1258 SPA_VERSION_UNIQUE_ACCURATE)
1259 1259 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1260 1260 }
1261 1261
1262 1262 struct killarg {
1263 1263 dsl_dataset_t *ds;
1264 1264 dmu_tx_t *tx;
1265 1265 };
1266 1266
1267 1267 /* ARGSUSED */
1268 1268 static int
1269 1269 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1270 1270 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1271 1271 {
1272 1272 struct killarg *ka = arg;
1273 1273 dmu_tx_t *tx = ka->tx;
1274 1274
1275 1275 if (bp == NULL)
1276 1276 return (0);
1277 1277
1278 1278 if (zb->zb_level == ZB_ZIL_LEVEL) {
1279 1279 ASSERT(zilog != NULL);
1280 1280 /*
1281 1281 * It's a block in the intent log. It has no
1282 1282 * accounting, so just free it.
1283 1283 */
1284 1284 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1285 1285 } else {
1286 1286 ASSERT(zilog == NULL);
1287 1287 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1288 1288 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1289 1289 }
1290 1290
1291 1291 return (0);
1292 1292 }
1293 1293
1294 1294 /* ARGSUSED */
1295 1295 static int
1296 1296 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1297 1297 {
1298 1298 dsl_dataset_t *ds = arg1;
1299 1299 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1300 1300 uint64_t count;
1301 1301 int err;
1302 1302
1303 1303 /*
1304 1304 * Can't delete a head dataset if there are snapshots of it.
1305 1305 * (Except if the only snapshots are from the branch we cloned
1306 1306 * from.)
1307 1307 */
1308 1308 if (ds->ds_prev != NULL &&
1309 1309 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1310 1310 return (EBUSY);
1311 1311
1312 1312 /*
1313 1313 * This is really a dsl_dir thing, but check it here so that
1314 1314 * we'll be less likely to leave this dataset inconsistent &
1315 1315 * nearly destroyed.
1316 1316 */
1317 1317 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1318 1318 if (err)
1319 1319 return (err);
1320 1320 if (count != 0)
↓ open down ↓ |
154 lines elided |
↑ open up ↑ |
1321 1321 return (EEXIST);
1322 1322
1323 1323 return (0);
1324 1324 }
1325 1325
1326 1326 /* ARGSUSED */
1327 1327 static void
1328 1328 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1329 1329 {
1330 1330 dsl_dataset_t *ds = arg1;
1331 - dsl_pool_t *dp = ds->ds_dir->dd_pool;
1332 1331
1333 1332 /* Mark it as inconsistent on-disk, in case we crash */
1334 1333 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1335 1334 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1336 1335
1337 - spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1338 - "dataset = %llu", ds->ds_object);
1336 + spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1339 1337 }
1340 1338
1341 1339 static int
1342 1340 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1343 1341 dmu_tx_t *tx)
1344 1342 {
1345 1343 dsl_dataset_t *ds = dsda->ds;
1346 1344 dsl_dataset_t *ds_prev = ds->ds_prev;
1347 1345
1348 1346 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1349 1347 struct dsl_ds_destroyarg ndsda = {0};
1350 1348
1351 1349 /*
1352 1350 * If we're not prepared to remove the origin, don't remove
1353 1351 * the clone either.
1354 1352 */
1355 1353 if (dsda->rm_origin == NULL) {
1356 1354 dsda->need_prep = B_TRUE;
1357 1355 return (EBUSY);
1358 1356 }
1359 1357
1360 1358 ndsda.ds = ds_prev;
1361 1359 ndsda.is_origin_rm = B_TRUE;
1362 1360 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1363 1361 }
1364 1362
1365 1363 /*
1366 1364 * If we're not going to remove the origin after all,
1367 1365 * undo the open context setup.
1368 1366 */
1369 1367 if (dsda->rm_origin != NULL) {
1370 1368 dsl_dataset_disown(dsda->rm_origin, tag);
1371 1369 dsda->rm_origin = NULL;
1372 1370 }
1373 1371
1374 1372 return (0);
1375 1373 }
1376 1374
1377 1375 /*
1378 1376 * If you add new checks here, you may need to add
1379 1377 * additional checks to the "temporary" case in
1380 1378 * snapshot_check() in dmu_objset.c.
1381 1379 */
1382 1380 /* ARGSUSED */
1383 1381 int
1384 1382 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1385 1383 {
1386 1384 struct dsl_ds_destroyarg *dsda = arg1;
1387 1385 dsl_dataset_t *ds = dsda->ds;
1388 1386
1389 1387 /* we have an owner hold, so noone else can destroy us */
1390 1388 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1391 1389
1392 1390 /*
1393 1391 * Only allow deferred destroy on pools that support it.
1394 1392 * NOTE: deferred destroy is only supported on snapshots.
1395 1393 */
1396 1394 if (dsda->defer) {
1397 1395 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1398 1396 SPA_VERSION_USERREFS)
1399 1397 return (ENOTSUP);
1400 1398 ASSERT(dsl_dataset_is_snapshot(ds));
1401 1399 return (0);
1402 1400 }
1403 1401
1404 1402 /*
1405 1403 * Can't delete a head dataset if there are snapshots of it.
1406 1404 * (Except if the only snapshots are from the branch we cloned
1407 1405 * from.)
1408 1406 */
1409 1407 if (ds->ds_prev != NULL &&
1410 1408 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1411 1409 return (EBUSY);
1412 1410
1413 1411 /*
1414 1412 * If we made changes this txg, traverse_dsl_dataset won't find
1415 1413 * them. Try again.
1416 1414 */
1417 1415 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1418 1416 return (EAGAIN);
1419 1417
1420 1418 if (dsl_dataset_is_snapshot(ds)) {
1421 1419 /*
1422 1420 * If this snapshot has an elevated user reference count,
1423 1421 * we can't destroy it yet.
1424 1422 */
1425 1423 if (ds->ds_userrefs > 0 && !dsda->releasing)
1426 1424 return (EBUSY);
1427 1425
1428 1426 mutex_enter(&ds->ds_lock);
1429 1427 /*
1430 1428 * Can't delete a branch point. However, if we're destroying
1431 1429 * a clone and removing its origin due to it having a user
1432 1430 * hold count of 0 and having been marked for deferred destroy,
1433 1431 * it's OK for the origin to have a single clone.
1434 1432 */
1435 1433 if (ds->ds_phys->ds_num_children >
1436 1434 (dsda->is_origin_rm ? 2 : 1)) {
1437 1435 mutex_exit(&ds->ds_lock);
1438 1436 return (EEXIST);
1439 1437 }
1440 1438 mutex_exit(&ds->ds_lock);
1441 1439 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1442 1440 return (dsl_dataset_origin_check(dsda, arg2, tx));
1443 1441 }
1444 1442
1445 1443 /* XXX we should do some i/o error checking... */
1446 1444 return (0);
1447 1445 }
1448 1446
1449 1447 struct refsarg {
1450 1448 kmutex_t lock;
1451 1449 boolean_t gone;
1452 1450 kcondvar_t cv;
1453 1451 };
1454 1452
1455 1453 /* ARGSUSED */
1456 1454 static void
1457 1455 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1458 1456 {
1459 1457 struct refsarg *arg = argv;
1460 1458
1461 1459 mutex_enter(&arg->lock);
1462 1460 arg->gone = TRUE;
1463 1461 cv_signal(&arg->cv);
1464 1462 mutex_exit(&arg->lock);
1465 1463 }
1466 1464
1467 1465 static void
1468 1466 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1469 1467 {
1470 1468 struct refsarg arg;
1471 1469
1472 1470 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1473 1471 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1474 1472 arg.gone = FALSE;
1475 1473 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1476 1474 dsl_dataset_refs_gone);
1477 1475 dmu_buf_rele(ds->ds_dbuf, tag);
1478 1476 mutex_enter(&arg.lock);
1479 1477 while (!arg.gone)
1480 1478 cv_wait(&arg.cv, &arg.lock);
1481 1479 ASSERT(arg.gone);
1482 1480 mutex_exit(&arg.lock);
1483 1481 ds->ds_dbuf = NULL;
1484 1482 ds->ds_phys = NULL;
1485 1483 mutex_destroy(&arg.lock);
1486 1484 cv_destroy(&arg.cv);
1487 1485 }
1488 1486
1489 1487 static void
1490 1488 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1491 1489 {
1492 1490 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1493 1491 uint64_t count;
1494 1492 int err;
1495 1493
1496 1494 ASSERT(ds->ds_phys->ds_num_children >= 2);
1497 1495 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1498 1496 /*
1499 1497 * The err should not be ENOENT, but a bug in a previous version
1500 1498 * of the code could cause upgrade_clones_cb() to not set
1501 1499 * ds_next_snap_obj when it should, leading to a missing entry.
1502 1500 * If we knew that the pool was created after
1503 1501 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1504 1502 * ENOENT. However, at least we can check that we don't have
1505 1503 * too many entries in the next_clones_obj even after failing to
1506 1504 * remove this one.
1507 1505 */
1508 1506 if (err != ENOENT) {
1509 1507 VERIFY3U(err, ==, 0);
1510 1508 }
1511 1509 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1512 1510 &count));
1513 1511 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1514 1512 }
1515 1513
1516 1514 static void
1517 1515 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1518 1516 {
1519 1517 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1520 1518 zap_cursor_t zc;
1521 1519 zap_attribute_t za;
1522 1520
1523 1521 /*
1524 1522 * If it is the old version, dd_clones doesn't exist so we can't
1525 1523 * find the clones, but deadlist_remove_key() is a no-op so it
1526 1524 * doesn't matter.
1527 1525 */
1528 1526 if (ds->ds_dir->dd_phys->dd_clones == 0)
1529 1527 return;
1530 1528
1531 1529 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1532 1530 zap_cursor_retrieve(&zc, &za) == 0;
1533 1531 zap_cursor_advance(&zc)) {
1534 1532 dsl_dataset_t *clone;
1535 1533
1536 1534 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1537 1535 za.za_first_integer, FTAG, &clone));
1538 1536 if (clone->ds_dir->dd_origin_txg > mintxg) {
1539 1537 dsl_deadlist_remove_key(&clone->ds_deadlist,
1540 1538 mintxg, tx);
1541 1539 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1542 1540 }
1543 1541 dsl_dataset_rele(clone, FTAG);
1544 1542 }
1545 1543 zap_cursor_fini(&zc);
1546 1544 }
1547 1545
1548 1546 struct process_old_arg {
1549 1547 dsl_dataset_t *ds;
1550 1548 dsl_dataset_t *ds_prev;
1551 1549 boolean_t after_branch_point;
1552 1550 zio_t *pio;
1553 1551 uint64_t used, comp, uncomp;
1554 1552 };
1555 1553
1556 1554 static int
1557 1555 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1558 1556 {
1559 1557 struct process_old_arg *poa = arg;
1560 1558 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1561 1559
1562 1560 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1563 1561 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1564 1562 if (poa->ds_prev && !poa->after_branch_point &&
1565 1563 bp->blk_birth >
1566 1564 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1567 1565 poa->ds_prev->ds_phys->ds_unique_bytes +=
1568 1566 bp_get_dsize_sync(dp->dp_spa, bp);
1569 1567 }
1570 1568 } else {
1571 1569 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1572 1570 poa->comp += BP_GET_PSIZE(bp);
1573 1571 poa->uncomp += BP_GET_UCSIZE(bp);
1574 1572 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1575 1573 }
1576 1574 return (0);
1577 1575 }
1578 1576
1579 1577 static void
1580 1578 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1581 1579 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1582 1580 {
1583 1581 struct process_old_arg poa = { 0 };
1584 1582 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1585 1583 objset_t *mos = dp->dp_meta_objset;
1586 1584
1587 1585 ASSERT(ds->ds_deadlist.dl_oldfmt);
1588 1586 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1589 1587
1590 1588 poa.ds = ds;
1591 1589 poa.ds_prev = ds_prev;
1592 1590 poa.after_branch_point = after_branch_point;
1593 1591 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1594 1592 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1595 1593 process_old_cb, &poa, tx));
1596 1594 VERIFY3U(zio_wait(poa.pio), ==, 0);
1597 1595 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1598 1596
1599 1597 /* change snapused */
1600 1598 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1601 1599 -poa.used, -poa.comp, -poa.uncomp, tx);
1602 1600
1603 1601 /* swap next's deadlist to our deadlist */
1604 1602 dsl_deadlist_close(&ds->ds_deadlist);
1605 1603 dsl_deadlist_close(&ds_next->ds_deadlist);
1606 1604 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1607 1605 ds->ds_phys->ds_deadlist_obj);
1608 1606 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1609 1607 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1610 1608 ds_next->ds_phys->ds_deadlist_obj);
1611 1609 }
1612 1610
1613 1611 static int
1614 1612 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1615 1613 {
1616 1614 int err;
1617 1615 struct killarg ka;
1618 1616
1619 1617 /*
1620 1618 * Free everything that we point to (that's born after
1621 1619 * the previous snapshot, if we are a clone)
1622 1620 *
1623 1621 * NB: this should be very quick, because we already
1624 1622 * freed all the objects in open context.
1625 1623 */
1626 1624 ka.ds = ds;
1627 1625 ka.tx = tx;
1628 1626 err = traverse_dataset(ds,
1629 1627 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1630 1628 kill_blkptr, &ka);
1631 1629 ASSERT3U(err, ==, 0);
1632 1630 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1633 1631
1634 1632 return (err);
1635 1633 }
1636 1634
1637 1635 void
1638 1636 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1639 1637 {
1640 1638 struct dsl_ds_destroyarg *dsda = arg1;
1641 1639 dsl_dataset_t *ds = dsda->ds;
1642 1640 int err;
1643 1641 int after_branch_point = FALSE;
1644 1642 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1645 1643 objset_t *mos = dp->dp_meta_objset;
1646 1644 dsl_dataset_t *ds_prev = NULL;
1647 1645 boolean_t wont_destroy;
1648 1646 uint64_t obj;
1649 1647
1650 1648 wont_destroy = (dsda->defer &&
1651 1649 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1652 1650
↓ open down ↓ |
304 lines elided |
↑ open up ↑ |
1653 1651 ASSERT(ds->ds_owner || wont_destroy);
1654 1652 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1655 1653 ASSERT(ds->ds_prev == NULL ||
1656 1654 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1657 1655 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1658 1656
1659 1657 if (wont_destroy) {
1660 1658 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1661 1659 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1662 1660 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1661 + spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
1663 1662 return;
1664 1663 }
1665 1664
1665 + /* We need to log before removing it from the namespace. */
1666 + spa_history_log_internal_ds(ds, "destroy", tx, "");
1667 +
1666 1668 /* signal any waiters that this dataset is going away */
1667 1669 mutex_enter(&ds->ds_lock);
1668 1670 ds->ds_owner = dsl_reaper;
1669 1671 cv_broadcast(&ds->ds_exclusive_cv);
1670 1672 mutex_exit(&ds->ds_lock);
1671 1673
1672 1674 /* Remove our reservation */
1673 1675 if (ds->ds_reserved != 0) {
1674 1676 dsl_prop_setarg_t psa;
1675 1677 uint64_t value = 0;
1676 1678
1677 1679 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1678 1680 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1679 1681 &value);
1680 1682 psa.psa_effective_value = 0; /* predict default value */
1681 1683
1682 1684 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1683 1685 ASSERT3U(ds->ds_reserved, ==, 0);
1684 1686 }
1685 1687
1686 1688 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1687 1689
1688 1690 dsl_scan_ds_destroyed(ds, tx);
1689 1691
1690 1692 obj = ds->ds_object;
1691 1693
1692 1694 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1693 1695 if (ds->ds_prev) {
1694 1696 ds_prev = ds->ds_prev;
1695 1697 } else {
1696 1698 VERIFY(0 == dsl_dataset_hold_obj(dp,
1697 1699 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1698 1700 }
1699 1701 after_branch_point =
1700 1702 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1701 1703
1702 1704 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1703 1705 if (after_branch_point &&
1704 1706 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1705 1707 remove_from_next_clones(ds_prev, obj, tx);
1706 1708 if (ds->ds_phys->ds_next_snap_obj != 0) {
1707 1709 VERIFY(0 == zap_add_int(mos,
1708 1710 ds_prev->ds_phys->ds_next_clones_obj,
1709 1711 ds->ds_phys->ds_next_snap_obj, tx));
1710 1712 }
1711 1713 }
1712 1714 if (after_branch_point &&
1713 1715 ds->ds_phys->ds_next_snap_obj == 0) {
1714 1716 /* This clone is toast. */
1715 1717 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1716 1718 ds_prev->ds_phys->ds_num_children--;
1717 1719
1718 1720 /*
1719 1721 * If the clone's origin has no other clones, no
1720 1722 * user holds, and has been marked for deferred
1721 1723 * deletion, then we should have done the necessary
1722 1724 * destroy setup for it.
1723 1725 */
1724 1726 if (ds_prev->ds_phys->ds_num_children == 1 &&
1725 1727 ds_prev->ds_userrefs == 0 &&
1726 1728 DS_IS_DEFER_DESTROY(ds_prev)) {
1727 1729 ASSERT3P(dsda->rm_origin, !=, NULL);
1728 1730 } else {
1729 1731 ASSERT3P(dsda->rm_origin, ==, NULL);
1730 1732 }
1731 1733 } else if (!after_branch_point) {
1732 1734 ds_prev->ds_phys->ds_next_snap_obj =
1733 1735 ds->ds_phys->ds_next_snap_obj;
1734 1736 }
1735 1737 }
1736 1738
1737 1739 if (dsl_dataset_is_snapshot(ds)) {
1738 1740 dsl_dataset_t *ds_next;
1739 1741 uint64_t old_unique;
1740 1742 uint64_t used = 0, comp = 0, uncomp = 0;
1741 1743
1742 1744 VERIFY(0 == dsl_dataset_hold_obj(dp,
1743 1745 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1744 1746 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1745 1747
1746 1748 old_unique = ds_next->ds_phys->ds_unique_bytes;
1747 1749
1748 1750 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1749 1751 ds_next->ds_phys->ds_prev_snap_obj =
1750 1752 ds->ds_phys->ds_prev_snap_obj;
1751 1753 ds_next->ds_phys->ds_prev_snap_txg =
1752 1754 ds->ds_phys->ds_prev_snap_txg;
1753 1755 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1754 1756 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1755 1757
1756 1758
1757 1759 if (ds_next->ds_deadlist.dl_oldfmt) {
1758 1760 process_old_deadlist(ds, ds_prev, ds_next,
1759 1761 after_branch_point, tx);
1760 1762 } else {
1761 1763 /* Adjust prev's unique space. */
1762 1764 if (ds_prev && !after_branch_point) {
1763 1765 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1764 1766 ds_prev->ds_phys->ds_prev_snap_txg,
1765 1767 ds->ds_phys->ds_prev_snap_txg,
1766 1768 &used, &comp, &uncomp);
1767 1769 ds_prev->ds_phys->ds_unique_bytes += used;
1768 1770 }
1769 1771
1770 1772 /* Adjust snapused. */
1771 1773 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1772 1774 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1773 1775 &used, &comp, &uncomp);
1774 1776 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1775 1777 -used, -comp, -uncomp, tx);
1776 1778
1777 1779 /* Move blocks to be freed to pool's free list. */
1778 1780 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1779 1781 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1780 1782 tx);
1781 1783 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1782 1784 DD_USED_HEAD, used, comp, uncomp, tx);
1783 1785
1784 1786 /* Merge our deadlist into next's and free it. */
1785 1787 dsl_deadlist_merge(&ds_next->ds_deadlist,
1786 1788 ds->ds_phys->ds_deadlist_obj, tx);
1787 1789 }
1788 1790 dsl_deadlist_close(&ds->ds_deadlist);
1789 1791 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1790 1792
1791 1793 /* Collapse range in clone heads */
1792 1794 dsl_dataset_remove_clones_key(ds,
1793 1795 ds->ds_phys->ds_creation_txg, tx);
1794 1796
1795 1797 if (dsl_dataset_is_snapshot(ds_next)) {
1796 1798 dsl_dataset_t *ds_nextnext;
1797 1799
1798 1800 /*
1799 1801 * Update next's unique to include blocks which
1800 1802 * were previously shared by only this snapshot
1801 1803 * and it. Those blocks will be born after the
1802 1804 * prev snap and before this snap, and will have
1803 1805 * died after the next snap and before the one
1804 1806 * after that (ie. be on the snap after next's
1805 1807 * deadlist).
1806 1808 */
1807 1809 VERIFY(0 == dsl_dataset_hold_obj(dp,
1808 1810 ds_next->ds_phys->ds_next_snap_obj,
1809 1811 FTAG, &ds_nextnext));
1810 1812 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1811 1813 ds->ds_phys->ds_prev_snap_txg,
1812 1814 ds->ds_phys->ds_creation_txg,
1813 1815 &used, &comp, &uncomp);
1814 1816 ds_next->ds_phys->ds_unique_bytes += used;
1815 1817 dsl_dataset_rele(ds_nextnext, FTAG);
1816 1818 ASSERT3P(ds_next->ds_prev, ==, NULL);
1817 1819
1818 1820 /* Collapse range in this head. */
1819 1821 dsl_dataset_t *hds;
1820 1822 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1821 1823 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1822 1824 FTAG, &hds));
1823 1825 dsl_deadlist_remove_key(&hds->ds_deadlist,
1824 1826 ds->ds_phys->ds_creation_txg, tx);
1825 1827 dsl_dataset_rele(hds, FTAG);
1826 1828
1827 1829 } else {
1828 1830 ASSERT3P(ds_next->ds_prev, ==, ds);
1829 1831 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1830 1832 ds_next->ds_prev = NULL;
1831 1833 if (ds_prev) {
1832 1834 VERIFY(0 == dsl_dataset_get_ref(dp,
1833 1835 ds->ds_phys->ds_prev_snap_obj,
1834 1836 ds_next, &ds_next->ds_prev));
1835 1837 }
1836 1838
1837 1839 dsl_dataset_recalc_head_uniq(ds_next);
1838 1840
1839 1841 /*
1840 1842 * Reduce the amount of our unconsmed refreservation
1841 1843 * being charged to our parent by the amount of
1842 1844 * new unique data we have gained.
1843 1845 */
1844 1846 if (old_unique < ds_next->ds_reserved) {
1845 1847 int64_t mrsdelta;
1846 1848 uint64_t new_unique =
1847 1849 ds_next->ds_phys->ds_unique_bytes;
1848 1850
1849 1851 ASSERT(old_unique <= new_unique);
1850 1852 mrsdelta = MIN(new_unique - old_unique,
1851 1853 ds_next->ds_reserved - old_unique);
1852 1854 dsl_dir_diduse_space(ds->ds_dir,
1853 1855 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1854 1856 }
1855 1857 }
1856 1858 dsl_dataset_rele(ds_next, FTAG);
1857 1859 } else {
1858 1860 zfeature_info_t *async_destroy =
1859 1861 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1860 1862
1861 1863 /*
1862 1864 * There's no next snapshot, so this is a head dataset.
1863 1865 * Destroy the deadlist. Unless it's a clone, the
1864 1866 * deadlist should be empty. (If it's a clone, it's
1865 1867 * safe to ignore the deadlist contents.)
1866 1868 */
1867 1869 dsl_deadlist_close(&ds->ds_deadlist);
1868 1870 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1869 1871 ds->ds_phys->ds_deadlist_obj = 0;
1870 1872
1871 1873 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1872 1874 err = old_synchronous_dataset_destroy(ds, tx);
1873 1875 } else {
1874 1876 /*
1875 1877 * Move the bptree into the pool's list of trees to
1876 1878 * clean up and update space accounting information.
1877 1879 */
1878 1880 uint64_t used, comp, uncomp;
1879 1881
1880 1882 ASSERT(err == 0 || err == EBUSY);
1881 1883 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1882 1884 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1883 1885 dp->dp_bptree_obj = bptree_alloc(
1884 1886 dp->dp_meta_objset, tx);
1885 1887 VERIFY(zap_add(dp->dp_meta_objset,
1886 1888 DMU_POOL_DIRECTORY_OBJECT,
1887 1889 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1888 1890 &dp->dp_bptree_obj, tx) == 0);
1889 1891 }
1890 1892
1891 1893 used = ds->ds_dir->dd_phys->dd_used_bytes;
1892 1894 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1893 1895 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1894 1896
1895 1897 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1896 1898 ds->ds_phys->ds_unique_bytes == used);
1897 1899
1898 1900 bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
1899 1901 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1900 1902 used, comp, uncomp, tx);
1901 1903 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1902 1904 -used, -comp, -uncomp, tx);
1903 1905 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1904 1906 used, comp, uncomp, tx);
1905 1907 }
1906 1908
1907 1909 if (ds->ds_prev != NULL) {
1908 1910 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1909 1911 VERIFY3U(0, ==, zap_remove_int(mos,
1910 1912 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1911 1913 ds->ds_object, tx));
1912 1914 }
1913 1915 dsl_dataset_rele(ds->ds_prev, ds);
1914 1916 ds->ds_prev = ds_prev = NULL;
1915 1917 }
1916 1918 }
1917 1919
1918 1920 /*
1919 1921 * This must be done after the dsl_traverse(), because it will
1920 1922 * re-open the objset.
1921 1923 */
1922 1924 if (ds->ds_objset) {
1923 1925 dmu_objset_evict(ds->ds_objset);
1924 1926 ds->ds_objset = NULL;
1925 1927 }
1926 1928
1927 1929 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1928 1930 /* Erase the link in the dir */
1929 1931 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1930 1932 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1931 1933 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1932 1934 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1933 1935 ASSERT(err == 0);
1934 1936 } else {
1935 1937 /* remove from snapshot namespace */
1936 1938 dsl_dataset_t *ds_head;
1937 1939 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1938 1940 VERIFY(0 == dsl_dataset_hold_obj(dp,
1939 1941 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1940 1942 VERIFY(0 == dsl_dataset_get_snapname(ds));
1941 1943 #ifdef ZFS_DEBUG
1942 1944 {
1943 1945 uint64_t val;
1944 1946
1945 1947 err = dsl_dataset_snap_lookup(ds_head,
1946 1948 ds->ds_snapname, &val);
1947 1949 ASSERT3U(err, ==, 0);
1948 1950 ASSERT3U(val, ==, obj);
1949 1951 }
↓ open down ↓ |
274 lines elided |
↑ open up ↑ |
1950 1952 #endif
1951 1953 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1952 1954 ASSERT(err == 0);
1953 1955 dsl_dataset_rele(ds_head, FTAG);
1954 1956 }
1955 1957
1956 1958 if (ds_prev && ds->ds_prev != ds_prev)
1957 1959 dsl_dataset_rele(ds_prev, FTAG);
1958 1960
1959 1961 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1960 - spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1961 - "dataset = %llu", ds->ds_object);
1962 1962
1963 1963 if (ds->ds_phys->ds_next_clones_obj != 0) {
1964 1964 uint64_t count;
1965 1965 ASSERT(0 == zap_count(mos,
1966 1966 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1967 1967 VERIFY(0 == dmu_object_free(mos,
1968 1968 ds->ds_phys->ds_next_clones_obj, tx));
1969 1969 }
1970 1970 if (ds->ds_phys->ds_props_obj != 0)
1971 1971 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1972 1972 if (ds->ds_phys->ds_userrefs_obj != 0)
1973 1973 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1974 1974 dsl_dir_close(ds->ds_dir, ds);
1975 1975 ds->ds_dir = NULL;
1976 1976 dsl_dataset_drain_refs(ds, tag);
1977 1977 VERIFY(0 == dmu_object_free(mos, obj, tx));
1978 1978
1979 1979 if (dsda->rm_origin) {
1980 1980 /*
1981 1981 * Remove the origin of the clone we just destroyed.
1982 1982 */
1983 1983 struct dsl_ds_destroyarg ndsda = {0};
1984 1984
1985 1985 ndsda.ds = dsda->rm_origin;
1986 1986 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1987 1987 }
1988 1988 }
1989 1989
1990 1990 static int
1991 1991 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1992 1992 {
1993 1993 uint64_t asize;
1994 1994
1995 1995 if (!dmu_tx_is_syncing(tx))
1996 1996 return (0);
1997 1997
1998 1998 /*
↓ open down ↓ |
27 lines elided |
↑ open up ↑ |
1999 1999 * If there's an fs-only reservation, any blocks that might become
2000 2000 * owned by the snapshot dataset must be accommodated by space
2001 2001 * outside of the reservation.
2002 2002 */
2003 2003 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2004 2004 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2005 2005 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2006 2006 return (ENOSPC);
2007 2007
2008 2008 /*
2009 - * Propogate any reserved space for this snapshot to other
2009 + * Propagate any reserved space for this snapshot to other
2010 2010 * snapshot checks in this sync group.
2011 2011 */
2012 2012 if (asize > 0)
2013 2013 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2014 2014
2015 2015 return (0);
2016 2016 }
2017 2017
2018 2018 int
2019 -dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
2019 +dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname,
2020 + dmu_tx_t *tx)
2020 2021 {
2021 - dsl_dataset_t *ds = arg1;
2022 - const char *snapname = arg2;
2023 2022 int err;
2024 2023 uint64_t value;
2025 2024
2026 2025 /*
2027 2026 * We don't allow multiple snapshots of the same txg. If there
2028 2027 * is already one, try again.
2029 2028 */
2030 2029 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2031 2030 return (EAGAIN);
2032 2031
2033 2032 /*
2034 - * Check for conflicting name snapshot name.
2033 + * Check for conflicting snapshot name.
2035 2034 */
2036 2035 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2037 2036 if (err == 0)
2038 2037 return (EEXIST);
2039 2038 if (err != ENOENT)
2040 2039 return (err);
2041 2040
2042 2041 /*
2043 2042 * Check that the dataset's name is not too long. Name consists
2044 2043 * of the dataset's length + 1 for the @-sign + snapshot name's length
2045 2044 */
2046 2045 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2047 2046 return (ENAMETOOLONG);
↓ open down ↓ |
3 lines elided |
↑ open up ↑ |
2048 2047
2049 2048 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2050 2049 if (err)
2051 2050 return (err);
2052 2051
2053 2052 ds->ds_trysnap_txg = tx->tx_txg;
2054 2053 return (0);
2055 2054 }
2056 2055
2057 2056 void
2058 -dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2057 +dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname,
2058 + dmu_tx_t *tx)
2059 2059 {
2060 - dsl_dataset_t *ds = arg1;
2061 - const char *snapname = arg2;
2062 2060 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2063 2061 dmu_buf_t *dbuf;
2064 2062 dsl_dataset_phys_t *dsphys;
2065 2063 uint64_t dsobj, crtxg;
2066 2064 objset_t *mos = dp->dp_meta_objset;
2067 2065 int err;
2068 2066
2069 2067 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2070 2068
2071 2069 /*
2072 2070 * The origin's ds_creation_txg has to be < TXG_INITIAL
2073 2071 */
2074 2072 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2075 2073 crtxg = 1;
2076 2074 else
2077 2075 crtxg = tx->tx_txg;
2078 2076
2079 2077 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2080 2078 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2081 2079 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2082 2080 dmu_buf_will_dirty(dbuf, tx);
2083 2081 dsphys = dbuf->db_data;
2084 2082 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2085 2083 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2086 2084 dsphys->ds_fsid_guid = unique_create();
2087 2085 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2088 2086 sizeof (dsphys->ds_guid));
2089 2087 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2090 2088 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2091 2089 dsphys->ds_next_snap_obj = ds->ds_object;
2092 2090 dsphys->ds_num_children = 1;
2093 2091 dsphys->ds_creation_time = gethrestime_sec();
2094 2092 dsphys->ds_creation_txg = crtxg;
2095 2093 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2096 2094 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2097 2095 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2098 2096 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2099 2097 dsphys->ds_flags = ds->ds_phys->ds_flags;
2100 2098 dsphys->ds_bp = ds->ds_phys->ds_bp;
2101 2099 dmu_buf_rele(dbuf, FTAG);
2102 2100
2103 2101 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2104 2102 if (ds->ds_prev) {
2105 2103 uint64_t next_clones_obj =
2106 2104 ds->ds_prev->ds_phys->ds_next_clones_obj;
2107 2105 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2108 2106 ds->ds_object ||
2109 2107 ds->ds_prev->ds_phys->ds_num_children > 1);
2110 2108 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2111 2109 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2112 2110 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2113 2111 ds->ds_prev->ds_phys->ds_creation_txg);
2114 2112 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2115 2113 } else if (next_clones_obj != 0) {
2116 2114 remove_from_next_clones(ds->ds_prev,
2117 2115 dsphys->ds_next_snap_obj, tx);
2118 2116 VERIFY3U(0, ==, zap_add_int(mos,
2119 2117 next_clones_obj, dsobj, tx));
2120 2118 }
2121 2119 }
2122 2120
2123 2121 /*
2124 2122 * If we have a reference-reservation on this dataset, we will
2125 2123 * need to increase the amount of refreservation being charged
2126 2124 * since our unique space is going to zero.
2127 2125 */
2128 2126 if (ds->ds_reserved) {
2129 2127 int64_t delta;
2130 2128 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2131 2129 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2132 2130 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2133 2131 delta, 0, 0, tx);
2134 2132 }
2135 2133
2136 2134 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2137 2135 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2138 2136 ds->ds_dir->dd_myname, snapname, dsobj,
2139 2137 ds->ds_phys->ds_prev_snap_txg);
2140 2138 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2141 2139 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2142 2140 dsl_deadlist_close(&ds->ds_deadlist);
2143 2141 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2144 2142 dsl_deadlist_add_key(&ds->ds_deadlist,
2145 2143 ds->ds_phys->ds_prev_snap_txg, tx);
2146 2144
2147 2145 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2148 2146 ds->ds_phys->ds_prev_snap_obj = dsobj;
2149 2147 ds->ds_phys->ds_prev_snap_txg = crtxg;
2150 2148 ds->ds_phys->ds_unique_bytes = 0;
2151 2149 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2152 2150 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2153 2151
2154 2152 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2155 2153 snapname, 8, 1, &dsobj, tx);
2156 2154 ASSERT(err == 0);
↓ open down ↓ |
85 lines elided |
↑ open up ↑ |
2157 2155
2158 2156 if (ds->ds_prev)
2159 2157 dsl_dataset_drop_ref(ds->ds_prev, ds);
2160 2158 VERIFY(0 == dsl_dataset_get_ref(dp,
2161 2159 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2162 2160
2163 2161 dsl_scan_ds_snapshotted(ds, tx);
2164 2162
2165 2163 dsl_dir_snap_cmtime_update(ds->ds_dir);
2166 2164
2167 - spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2168 - "dataset = %llu", dsobj);
2165 + spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
2169 2166 }
2170 2167
2171 2168 void
2172 2169 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2173 2170 {
2174 2171 ASSERT(dmu_tx_is_syncing(tx));
2175 2172 ASSERT(ds->ds_objset != NULL);
2176 2173 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2177 2174
2178 2175 /*
2179 2176 * in case we had to change ds_fsid_guid when we opened it,
2180 2177 * sync it out now.
2181 2178 */
2182 2179 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2183 2180 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2184 2181
2185 2182 dsl_dir_dirty(ds->ds_dir, tx);
2186 2183 dmu_objset_sync(ds->ds_objset, zio, tx);
2187 2184 }
2188 2185
2189 2186 static void
2190 2187 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2191 2188 {
2192 2189 uint64_t count = 0;
2193 2190 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2194 2191 zap_cursor_t zc;
2195 2192 zap_attribute_t za;
2196 2193 nvlist_t *propval;
2197 2194 nvlist_t *val;
2198 2195
2199 2196 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2200 2197 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2201 2198 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2202 2199
2203 2200 /*
2204 2201 * There may me missing entries in ds_next_clones_obj
2205 2202 * due to a bug in a previous version of the code.
2206 2203 * Only trust it if it has the right number of entries.
2207 2204 */
2208 2205 if (ds->ds_phys->ds_next_clones_obj != 0) {
2209 2206 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2210 2207 &count));
2211 2208 }
2212 2209 if (count != ds->ds_phys->ds_num_children - 1) {
2213 2210 goto fail;
2214 2211 }
2215 2212 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2216 2213 zap_cursor_retrieve(&zc, &za) == 0;
2217 2214 zap_cursor_advance(&zc)) {
2218 2215 dsl_dataset_t *clone;
2219 2216 char buf[ZFS_MAXNAMELEN];
2220 2217 /*
2221 2218 * Even though we hold the dp_config_rwlock, the dataset
2222 2219 * may fail to open, returning ENOENT. If there is a
2223 2220 * thread concurrently attempting to destroy this
2224 2221 * dataset, it will have the ds_rwlock held for
2225 2222 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2226 2223 * dsl_dataset_hold_ref() will fail its
2227 2224 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2228 2225 * dp_config_rwlock, and wait for the destroy progress
2229 2226 * and signal ds_exclusive_cv. If the destroy was
2230 2227 * successful, we will see that
2231 2228 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2232 2229 */
2233 2230 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2234 2231 za.za_first_integer, FTAG, &clone) != 0)
2235 2232 continue;
2236 2233 dsl_dir_name(clone->ds_dir, buf);
2237 2234 VERIFY(nvlist_add_boolean(val, buf) == 0);
2238 2235 dsl_dataset_rele(clone, FTAG);
2239 2236 }
2240 2237 zap_cursor_fini(&zc);
2241 2238 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2242 2239 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2243 2240 propval) == 0);
2244 2241 fail:
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
2245 2242 nvlist_free(val);
2246 2243 nvlist_free(propval);
2247 2244 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2248 2245 }
2249 2246
2250 2247 void
2251 2248 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2252 2249 {
2253 2250 uint64_t refd, avail, uobjs, aobjs, ratio;
2254 2251
2255 - dsl_dir_stats(ds->ds_dir, nv);
2252 + ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2253 + (ds->ds_phys->ds_uncompressed_bytes * 100 /
2254 + ds->ds_phys->ds_compressed_bytes);
2255 +
2256 + dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2257 +
2258 + if (dsl_dataset_is_snapshot(ds)) {
2259 + dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2260 + dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2261 + ds->ds_phys->ds_unique_bytes);
2262 + get_clones_stat(ds, nv);
2263 + } else {
2264 + dsl_dir_stats(ds->ds_dir, nv);
2265 + }
2256 2266
2257 2267 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2258 2268 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2259 2269 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2260 2270
2261 2271 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2262 2272 ds->ds_phys->ds_creation_time);
2263 2273 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2264 2274 ds->ds_phys->ds_creation_txg);
2265 2275 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2266 2276 ds->ds_quota);
2267 2277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2268 2278 ds->ds_reserved);
2269 2279 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2270 2280 ds->ds_phys->ds_guid);
2271 2281 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2272 2282 ds->ds_phys->ds_unique_bytes);
2273 2283 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2274 2284 ds->ds_object);
2275 2285 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2276 2286 ds->ds_userrefs);
2277 2287 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2278 2288 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2279 2289
2280 2290 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2281 2291 uint64_t written, comp, uncomp;
2282 2292 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2283 2293 dsl_dataset_t *prev;
2284 2294
2285 2295 rw_enter(&dp->dp_config_rwlock, RW_READER);
2286 2296 int err = dsl_dataset_hold_obj(dp,
2287 2297 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2288 2298 rw_exit(&dp->dp_config_rwlock);
2289 2299 if (err == 0) {
↓ open down ↓ |
24 lines elided |
↑ open up ↑ |
2290 2300 err = dsl_dataset_space_written(prev, ds, &written,
2291 2301 &comp, &uncomp);
2292 2302 dsl_dataset_rele(prev, FTAG);
2293 2303 if (err == 0) {
2294 2304 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2295 2305 written);
2296 2306 }
2297 2307 }
2298 2308 }
2299 2309
2300 - ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2301 - (ds->ds_phys->ds_uncompressed_bytes * 100 /
2302 - ds->ds_phys->ds_compressed_bytes);
2303 - dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2304 -
2305 - if (ds->ds_phys->ds_next_snap_obj) {
2306 - /*
2307 - * This is a snapshot; override the dd's space used with
2308 - * our unique space and compression ratio.
2309 - */
2310 - dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2311 - ds->ds_phys->ds_unique_bytes);
2312 - dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2313 -
2314 - get_clones_stat(ds, nv);
2315 - }
2316 2310 }
2317 2311
2318 2312 void
2319 2313 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2320 2314 {
2321 2315 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2322 2316 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2323 2317 stat->dds_guid = ds->ds_phys->ds_guid;
2324 - if (ds->ds_phys->ds_next_snap_obj) {
2318 + stat->dds_origin[0] = '\0';
2319 + if (dsl_dataset_is_snapshot(ds)) {
2325 2320 stat->dds_is_snapshot = B_TRUE;
2326 2321 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2327 2322 } else {
2328 2323 stat->dds_is_snapshot = B_FALSE;
2329 2324 stat->dds_num_clones = 0;
2330 - }
2331 -
2332 - /* clone origin is really a dsl_dir thing... */
2333 - rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2334 - if (dsl_dir_is_clone(ds->ds_dir)) {
2335 - dsl_dataset_t *ods;
2336 2325
2337 - VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2338 - ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2339 - dsl_dataset_name(ods, stat->dds_origin);
2340 - dsl_dataset_drop_ref(ods, FTAG);
2341 - } else {
2342 - stat->dds_origin[0] = '\0';
2326 + rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2327 + if (dsl_dir_is_clone(ds->ds_dir)) {
2328 + dsl_dataset_t *ods;
2329 +
2330 + VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2331 + ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2332 + dsl_dataset_name(ods, stat->dds_origin);
2333 + dsl_dataset_drop_ref(ods, FTAG);
2334 + }
2335 + rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2343 2336 }
2344 - rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2345 2337 }
2346 2338
2347 2339 uint64_t
2348 2340 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2349 2341 {
2350 2342 return (ds->ds_fsid_guid);
2351 2343 }
2352 2344
2353 2345 void
2354 2346 dsl_dataset_space(dsl_dataset_t *ds,
2355 2347 uint64_t *refdbytesp, uint64_t *availbytesp,
2356 2348 uint64_t *usedobjsp, uint64_t *availobjsp)
2357 2349 {
2358 2350 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2359 2351 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2360 2352 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2361 2353 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2362 2354 if (ds->ds_quota != 0) {
2363 2355 /*
2364 2356 * Adjust available bytes according to refquota
2365 2357 */
2366 2358 if (*refdbytesp < ds->ds_quota)
2367 2359 *availbytesp = MIN(*availbytesp,
2368 2360 ds->ds_quota - *refdbytesp);
2369 2361 else
2370 2362 *availbytesp = 0;
2371 2363 }
2372 2364 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2373 2365 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2374 2366 }
2375 2367
2376 2368 boolean_t
2377 2369 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2378 2370 {
2379 2371 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2380 2372
2381 2373 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2382 2374 dsl_pool_sync_context(dp));
2383 2375 if (ds->ds_prev == NULL)
2384 2376 return (B_FALSE);
2385 2377 if (ds->ds_phys->ds_bp.blk_birth >
2386 2378 ds->ds_prev->ds_phys->ds_creation_txg) {
2387 2379 objset_t *os, *os_prev;
2388 2380 /*
2389 2381 * It may be that only the ZIL differs, because it was
2390 2382 * reset in the head. Don't count that as being
2391 2383 * modified.
2392 2384 */
2393 2385 if (dmu_objset_from_ds(ds, &os) != 0)
2394 2386 return (B_TRUE);
2395 2387 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2396 2388 return (B_TRUE);
2397 2389 return (bcmp(&os->os_phys->os_meta_dnode,
2398 2390 &os_prev->os_phys->os_meta_dnode,
2399 2391 sizeof (os->os_phys->os_meta_dnode)) != 0);
2400 2392 }
2401 2393 return (B_FALSE);
2402 2394 }
2403 2395
2404 2396 /* ARGSUSED */
2405 2397 static int
2406 2398 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2407 2399 {
2408 2400 dsl_dataset_t *ds = arg1;
2409 2401 char *newsnapname = arg2;
2410 2402 dsl_dir_t *dd = ds->ds_dir;
2411 2403 dsl_dataset_t *hds;
2412 2404 uint64_t val;
2413 2405 int err;
2414 2406
2415 2407 err = dsl_dataset_hold_obj(dd->dd_pool,
2416 2408 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2417 2409 if (err)
2418 2410 return (err);
2419 2411
2420 2412 /* new name better not be in use */
2421 2413 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2422 2414 dsl_dataset_rele(hds, FTAG);
2423 2415
2424 2416 if (err == 0)
2425 2417 err = EEXIST;
2426 2418 else if (err == ENOENT)
2427 2419 err = 0;
2428 2420
2429 2421 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2430 2422 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2431 2423 err = ENAMETOOLONG;
2432 2424
2433 2425 return (err);
2434 2426 }
2435 2427
2436 2428 static void
2437 2429 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2438 2430 {
2439 2431 dsl_dataset_t *ds = arg1;
2440 2432 const char *newsnapname = arg2;
2441 2433 dsl_dir_t *dd = ds->ds_dir;
2442 2434 objset_t *mos = dd->dd_pool->dp_meta_objset;
2443 2435 dsl_dataset_t *hds;
2444 2436 int err;
2445 2437
2446 2438 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2447 2439
2448 2440 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2449 2441 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2450 2442
↓ open down ↓ |
96 lines elided |
↑ open up ↑ |
2451 2443 VERIFY(0 == dsl_dataset_get_snapname(ds));
2452 2444 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2453 2445 ASSERT3U(err, ==, 0);
2454 2446 mutex_enter(&ds->ds_lock);
2455 2447 (void) strcpy(ds->ds_snapname, newsnapname);
2456 2448 mutex_exit(&ds->ds_lock);
2457 2449 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2458 2450 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2459 2451 ASSERT3U(err, ==, 0);
2460 2452
2461 - spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2462 - "dataset = %llu", ds->ds_object);
2453 + spa_history_log_internal_ds(ds, "rename", tx,
2454 + "-> @%s", newsnapname);
2463 2455 dsl_dataset_rele(hds, FTAG);
2464 2456 }
2465 2457
2466 2458 struct renamesnaparg {
2467 2459 dsl_sync_task_group_t *dstg;
2468 2460 char failed[MAXPATHLEN];
2469 2461 char *oldsnap;
2470 2462 char *newsnap;
2471 2463 };
2472 2464
2473 2465 static int
2474 2466 dsl_snapshot_rename_one(const char *name, void *arg)
2475 2467 {
2476 2468 struct renamesnaparg *ra = arg;
2477 2469 dsl_dataset_t *ds = NULL;
2478 2470 char *snapname;
2479 2471 int err;
2480 2472
2481 2473 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2482 2474 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2483 2475
2484 2476 /*
2485 2477 * For recursive snapshot renames the parent won't be changing
2486 2478 * so we just pass name for both the to/from argument.
2487 2479 */
2488 2480 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2489 2481 if (err != 0) {
2490 2482 strfree(snapname);
2491 2483 return (err == ENOENT ? 0 : err);
2492 2484 }
2493 2485
2494 2486 #ifdef _KERNEL
2495 2487 /*
2496 2488 * For all filesystems undergoing rename, we'll need to unmount it.
2497 2489 */
2498 2490 (void) zfs_unmount_snap(snapname, NULL);
2499 2491 #endif
2500 2492 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2501 2493 strfree(snapname);
2502 2494 if (err != 0)
2503 2495 return (err == ENOENT ? 0 : err);
2504 2496
2505 2497 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2506 2498 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2507 2499
2508 2500 return (0);
2509 2501 }
2510 2502
2511 2503 static int
2512 2504 dsl_recursive_rename(char *oldname, const char *newname)
2513 2505 {
2514 2506 int err;
2515 2507 struct renamesnaparg *ra;
2516 2508 dsl_sync_task_t *dst;
2517 2509 spa_t *spa;
2518 2510 char *cp, *fsname = spa_strdup(oldname);
2519 2511 int len = strlen(oldname) + 1;
2520 2512
2521 2513 /* truncate the snapshot name to get the fsname */
2522 2514 cp = strchr(fsname, '@');
2523 2515 *cp = '\0';
2524 2516
2525 2517 err = spa_open(fsname, &spa, FTAG);
2526 2518 if (err) {
2527 2519 kmem_free(fsname, len);
2528 2520 return (err);
2529 2521 }
2530 2522 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2531 2523 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2532 2524
2533 2525 ra->oldsnap = strchr(oldname, '@') + 1;
2534 2526 ra->newsnap = strchr(newname, '@') + 1;
2535 2527 *ra->failed = '\0';
2536 2528
2537 2529 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2538 2530 DS_FIND_CHILDREN);
2539 2531 kmem_free(fsname, len);
2540 2532
2541 2533 if (err == 0) {
2542 2534 err = dsl_sync_task_group_wait(ra->dstg);
2543 2535 }
2544 2536
2545 2537 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2546 2538 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2547 2539 dsl_dataset_t *ds = dst->dst_arg1;
2548 2540 if (dst->dst_err) {
2549 2541 dsl_dir_name(ds->ds_dir, ra->failed);
2550 2542 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2551 2543 (void) strlcat(ra->failed, ra->newsnap,
2552 2544 sizeof (ra->failed));
2553 2545 }
2554 2546 dsl_dataset_rele(ds, ra->dstg);
2555 2547 }
2556 2548
2557 2549 if (err)
2558 2550 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2559 2551
2560 2552 dsl_sync_task_group_destroy(ra->dstg);
2561 2553 kmem_free(ra, sizeof (struct renamesnaparg));
2562 2554 spa_close(spa, FTAG);
2563 2555 return (err);
2564 2556 }
2565 2557
2566 2558 static int
2567 2559 dsl_valid_rename(const char *oldname, void *arg)
2568 2560 {
2569 2561 int delta = *(int *)arg;
2570 2562
2571 2563 if (strlen(oldname) + delta >= MAXNAMELEN)
2572 2564 return (ENAMETOOLONG);
2573 2565
2574 2566 return (0);
2575 2567 }
2576 2568
2577 2569 #pragma weak dmu_objset_rename = dsl_dataset_rename
2578 2570 int
2579 2571 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2580 2572 {
2581 2573 dsl_dir_t *dd;
2582 2574 dsl_dataset_t *ds;
2583 2575 const char *tail;
2584 2576 int err;
2585 2577
2586 2578 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2587 2579 if (err)
2588 2580 return (err);
2589 2581
2590 2582 if (tail == NULL) {
2591 2583 int delta = strlen(newname) - strlen(oldname);
2592 2584
2593 2585 /* if we're growing, validate child name lengths */
2594 2586 if (delta > 0)
2595 2587 err = dmu_objset_find(oldname, dsl_valid_rename,
2596 2588 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2597 2589
2598 2590 if (err == 0)
2599 2591 err = dsl_dir_rename(dd, newname);
2600 2592 dsl_dir_close(dd, FTAG);
2601 2593 return (err);
2602 2594 }
2603 2595
2604 2596 if (tail[0] != '@') {
2605 2597 /* the name ended in a nonexistent component */
2606 2598 dsl_dir_close(dd, FTAG);
2607 2599 return (ENOENT);
2608 2600 }
2609 2601
2610 2602 dsl_dir_close(dd, FTAG);
2611 2603
2612 2604 /* new name must be snapshot in same filesystem */
2613 2605 tail = strchr(newname, '@');
2614 2606 if (tail == NULL)
2615 2607 return (EINVAL);
2616 2608 tail++;
2617 2609 if (strncmp(oldname, newname, tail - newname) != 0)
2618 2610 return (EXDEV);
2619 2611
2620 2612 if (recursive) {
2621 2613 err = dsl_recursive_rename(oldname, newname);
2622 2614 } else {
2623 2615 err = dsl_dataset_hold(oldname, FTAG, &ds);
2624 2616 if (err)
2625 2617 return (err);
2626 2618
2627 2619 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2628 2620 dsl_dataset_snapshot_rename_check,
2629 2621 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2630 2622
2631 2623 dsl_dataset_rele(ds, FTAG);
2632 2624 }
2633 2625
2634 2626 return (err);
2635 2627 }
2636 2628
2637 2629 struct promotenode {
2638 2630 list_node_t link;
2639 2631 dsl_dataset_t *ds;
2640 2632 };
2641 2633
2642 2634 struct promotearg {
2643 2635 list_t shared_snaps, origin_snaps, clone_snaps;
2644 2636 dsl_dataset_t *origin_origin;
2645 2637 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2646 2638 char *err_ds;
2647 2639 };
2648 2640
2649 2641 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2650 2642 static boolean_t snaplist_unstable(list_t *l);
2651 2643
2652 2644 static int
2653 2645 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2654 2646 {
2655 2647 dsl_dataset_t *hds = arg1;
2656 2648 struct promotearg *pa = arg2;
2657 2649 struct promotenode *snap = list_head(&pa->shared_snaps);
2658 2650 dsl_dataset_t *origin_ds = snap->ds;
2659 2651 int err;
2660 2652 uint64_t unused;
2661 2653
2662 2654 /* Check that it is a real clone */
2663 2655 if (!dsl_dir_is_clone(hds->ds_dir))
2664 2656 return (EINVAL);
2665 2657
2666 2658 /* Since this is so expensive, don't do the preliminary check */
2667 2659 if (!dmu_tx_is_syncing(tx))
2668 2660 return (0);
2669 2661
2670 2662 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2671 2663 return (EXDEV);
2672 2664
2673 2665 /* compute origin's new unique space */
2674 2666 snap = list_tail(&pa->clone_snaps);
2675 2667 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2676 2668 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2677 2669 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2678 2670 &pa->unique, &unused, &unused);
2679 2671
2680 2672 /*
2681 2673 * Walk the snapshots that we are moving
2682 2674 *
2683 2675 * Compute space to transfer. Consider the incremental changes
2684 2676 * to used for each snapshot:
2685 2677 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2686 2678 * So each snapshot gave birth to:
2687 2679 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2688 2680 * So a sequence would look like:
2689 2681 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2690 2682 * Which simplifies to:
2691 2683 * uN + kN + kN-1 + ... + k1 + k0
2692 2684 * Note however, if we stop before we reach the ORIGIN we get:
2693 2685 * uN + kN + kN-1 + ... + kM - uM-1
2694 2686 */
2695 2687 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2696 2688 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2697 2689 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2698 2690 for (snap = list_head(&pa->shared_snaps); snap;
2699 2691 snap = list_next(&pa->shared_snaps, snap)) {
2700 2692 uint64_t val, dlused, dlcomp, dluncomp;
2701 2693 dsl_dataset_t *ds = snap->ds;
2702 2694
2703 2695 /* Check that the snapshot name does not conflict */
2704 2696 VERIFY(0 == dsl_dataset_get_snapname(ds));
2705 2697 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2706 2698 if (err == 0) {
2707 2699 err = EEXIST;
2708 2700 goto out;
2709 2701 }
2710 2702 if (err != ENOENT)
2711 2703 goto out;
2712 2704
2713 2705 /* The very first snapshot does not have a deadlist */
2714 2706 if (ds->ds_phys->ds_prev_snap_obj == 0)
2715 2707 continue;
2716 2708
2717 2709 dsl_deadlist_space(&ds->ds_deadlist,
2718 2710 &dlused, &dlcomp, &dluncomp);
2719 2711 pa->used += dlused;
2720 2712 pa->comp += dlcomp;
2721 2713 pa->uncomp += dluncomp;
2722 2714 }
2723 2715
2724 2716 /*
2725 2717 * If we are a clone of a clone then we never reached ORIGIN,
2726 2718 * so we need to subtract out the clone origin's used space.
2727 2719 */
2728 2720 if (pa->origin_origin) {
2729 2721 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2730 2722 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2731 2723 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2732 2724 }
2733 2725
2734 2726 /* Check that there is enough space here */
2735 2727 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2736 2728 pa->used);
2737 2729 if (err)
2738 2730 return (err);
2739 2731
2740 2732 /*
2741 2733 * Compute the amounts of space that will be used by snapshots
2742 2734 * after the promotion (for both origin and clone). For each,
2743 2735 * it is the amount of space that will be on all of their
2744 2736 * deadlists (that was not born before their new origin).
2745 2737 */
2746 2738 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2747 2739 uint64_t space;
2748 2740
2749 2741 /*
2750 2742 * Note, typically this will not be a clone of a clone,
2751 2743 * so dd_origin_txg will be < TXG_INITIAL, so
2752 2744 * these snaplist_space() -> dsl_deadlist_space_range()
2753 2745 * calls will be fast because they do not have to
2754 2746 * iterate over all bps.
2755 2747 */
2756 2748 snap = list_head(&pa->origin_snaps);
2757 2749 err = snaplist_space(&pa->shared_snaps,
2758 2750 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2759 2751 if (err)
2760 2752 return (err);
2761 2753
2762 2754 err = snaplist_space(&pa->clone_snaps,
2763 2755 snap->ds->ds_dir->dd_origin_txg, &space);
2764 2756 if (err)
2765 2757 return (err);
2766 2758 pa->cloneusedsnap += space;
2767 2759 }
2768 2760 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2769 2761 err = snaplist_space(&pa->origin_snaps,
2770 2762 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2771 2763 if (err)
2772 2764 return (err);
2773 2765 }
2774 2766
2775 2767 return (0);
2776 2768 out:
2777 2769 pa->err_ds = snap->ds->ds_snapname;
2778 2770 return (err);
2779 2771 }
2780 2772
2781 2773 static void
2782 2774 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2783 2775 {
2784 2776 dsl_dataset_t *hds = arg1;
2785 2777 struct promotearg *pa = arg2;
2786 2778 struct promotenode *snap = list_head(&pa->shared_snaps);
2787 2779 dsl_dataset_t *origin_ds = snap->ds;
2788 2780 dsl_dataset_t *origin_head;
2789 2781 dsl_dir_t *dd = hds->ds_dir;
2790 2782 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2791 2783 dsl_dir_t *odd = NULL;
2792 2784 uint64_t oldnext_obj;
2793 2785 int64_t delta;
2794 2786
2795 2787 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2796 2788
2797 2789 snap = list_head(&pa->origin_snaps);
2798 2790 origin_head = snap->ds;
2799 2791
2800 2792 /*
2801 2793 * We need to explicitly open odd, since origin_ds's dd will be
2802 2794 * changing.
2803 2795 */
2804 2796 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2805 2797 NULL, FTAG, &odd));
2806 2798
2807 2799 /* change origin's next snap */
2808 2800 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2809 2801 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2810 2802 snap = list_tail(&pa->clone_snaps);
2811 2803 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2812 2804 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2813 2805
2814 2806 /* change the origin's next clone */
2815 2807 if (origin_ds->ds_phys->ds_next_clones_obj) {
2816 2808 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2817 2809 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2818 2810 origin_ds->ds_phys->ds_next_clones_obj,
2819 2811 oldnext_obj, tx));
2820 2812 }
2821 2813
2822 2814 /* change origin */
2823 2815 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2824 2816 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2825 2817 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2826 2818 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2827 2819 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2828 2820 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2829 2821 origin_head->ds_dir->dd_origin_txg =
2830 2822 origin_ds->ds_phys->ds_creation_txg;
2831 2823
2832 2824 /* change dd_clone entries */
2833 2825 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2834 2826 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2835 2827 odd->dd_phys->dd_clones, hds->ds_object, tx));
2836 2828 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2837 2829 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2838 2830 hds->ds_object, tx));
2839 2831
2840 2832 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2841 2833 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2842 2834 origin_head->ds_object, tx));
2843 2835 if (dd->dd_phys->dd_clones == 0) {
2844 2836 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2845 2837 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2846 2838 }
2847 2839 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2848 2840 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2849 2841
2850 2842 }
2851 2843
2852 2844 /* move snapshots to this dir */
2853 2845 for (snap = list_head(&pa->shared_snaps); snap;
2854 2846 snap = list_next(&pa->shared_snaps, snap)) {
2855 2847 dsl_dataset_t *ds = snap->ds;
2856 2848
2857 2849 /* unregister props as dsl_dir is changing */
2858 2850 if (ds->ds_objset) {
2859 2851 dmu_objset_evict(ds->ds_objset);
2860 2852 ds->ds_objset = NULL;
2861 2853 }
2862 2854 /* move snap name entry */
2863 2855 VERIFY(0 == dsl_dataset_get_snapname(ds));
2864 2856 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2865 2857 ds->ds_snapname, tx));
2866 2858 VERIFY(0 == zap_add(dp->dp_meta_objset,
2867 2859 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2868 2860 8, 1, &ds->ds_object, tx));
2869 2861
2870 2862 /* change containing dsl_dir */
2871 2863 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2872 2864 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2873 2865 ds->ds_phys->ds_dir_obj = dd->dd_object;
2874 2866 ASSERT3P(ds->ds_dir, ==, odd);
2875 2867 dsl_dir_close(ds->ds_dir, ds);
2876 2868 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2877 2869 NULL, ds, &ds->ds_dir));
2878 2870
2879 2871 /* move any clone references */
2880 2872 if (ds->ds_phys->ds_next_clones_obj &&
2881 2873 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2882 2874 zap_cursor_t zc;
2883 2875 zap_attribute_t za;
2884 2876
2885 2877 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2886 2878 ds->ds_phys->ds_next_clones_obj);
2887 2879 zap_cursor_retrieve(&zc, &za) == 0;
2888 2880 zap_cursor_advance(&zc)) {
2889 2881 dsl_dataset_t *cnds;
2890 2882 uint64_t o;
2891 2883
2892 2884 if (za.za_first_integer == oldnext_obj) {
2893 2885 /*
2894 2886 * We've already moved the
2895 2887 * origin's reference.
2896 2888 */
2897 2889 continue;
2898 2890 }
2899 2891
2900 2892 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2901 2893 za.za_first_integer, FTAG, &cnds));
2902 2894 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2903 2895
2904 2896 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2905 2897 odd->dd_phys->dd_clones, o, tx), ==, 0);
2906 2898 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2907 2899 dd->dd_phys->dd_clones, o, tx), ==, 0);
2908 2900 dsl_dataset_rele(cnds, FTAG);
2909 2901 }
2910 2902 zap_cursor_fini(&zc);
2911 2903 }
2912 2904
2913 2905 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2914 2906 }
2915 2907
2916 2908 /*
2917 2909 * Change space accounting.
2918 2910 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2919 2911 * both be valid, or both be 0 (resulting in delta == 0). This
2920 2912 * is true for each of {clone,origin} independently.
2921 2913 */
2922 2914
2923 2915 delta = pa->cloneusedsnap -
2924 2916 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2925 2917 ASSERT3S(delta, >=, 0);
2926 2918 ASSERT3U(pa->used, >=, delta);
2927 2919 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2928 2920 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2929 2921 pa->used - delta, pa->comp, pa->uncomp, tx);
2930 2922
2931 2923 delta = pa->originusedsnap -
↓ open down ↓ |
459 lines elided |
↑ open up ↑ |
2932 2924 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2933 2925 ASSERT3S(delta, <=, 0);
2934 2926 ASSERT3U(pa->used, >=, -delta);
2935 2927 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2936 2928 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2937 2929 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2938 2930
2939 2931 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2940 2932
2941 2933 /* log history record */
2942 - spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2943 - "dataset = %llu", hds->ds_object);
2934 + spa_history_log_internal_ds(hds, "promote", tx, "");
2944 2935
2945 2936 dsl_dir_close(odd, FTAG);
2946 2937 }
2947 2938
2948 2939 static char *snaplist_tag = "snaplist";
2949 2940 /*
2950 2941 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2951 2942 * (exclusive) and last_obj (inclusive). The list will be in reverse
2952 2943 * order (last_obj will be the list_head()). If first_obj == 0, do all
2953 2944 * snapshots back to this dataset's origin.
2954 2945 */
2955 2946 static int
2956 2947 snaplist_make(dsl_pool_t *dp, boolean_t own,
2957 2948 uint64_t first_obj, uint64_t last_obj, list_t *l)
2958 2949 {
2959 2950 uint64_t obj = last_obj;
2960 2951
2961 2952 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2962 2953
2963 2954 list_create(l, sizeof (struct promotenode),
2964 2955 offsetof(struct promotenode, link));
2965 2956
2966 2957 while (obj != first_obj) {
2967 2958 dsl_dataset_t *ds;
2968 2959 struct promotenode *snap;
2969 2960 int err;
2970 2961
2971 2962 if (own) {
2972 2963 err = dsl_dataset_own_obj(dp, obj,
2973 2964 0, snaplist_tag, &ds);
2974 2965 if (err == 0)
2975 2966 dsl_dataset_make_exclusive(ds, snaplist_tag);
2976 2967 } else {
2977 2968 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2978 2969 }
2979 2970 if (err == ENOENT) {
2980 2971 /* lost race with snapshot destroy */
2981 2972 struct promotenode *last = list_tail(l);
2982 2973 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2983 2974 obj = last->ds->ds_phys->ds_prev_snap_obj;
2984 2975 continue;
2985 2976 } else if (err) {
2986 2977 return (err);
2987 2978 }
2988 2979
2989 2980 if (first_obj == 0)
2990 2981 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2991 2982
2992 2983 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2993 2984 snap->ds = ds;
2994 2985 list_insert_tail(l, snap);
2995 2986 obj = ds->ds_phys->ds_prev_snap_obj;
2996 2987 }
2997 2988
2998 2989 return (0);
2999 2990 }
3000 2991
3001 2992 static int
3002 2993 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3003 2994 {
3004 2995 struct promotenode *snap;
3005 2996
3006 2997 *spacep = 0;
3007 2998 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3008 2999 uint64_t used, comp, uncomp;
3009 3000 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3010 3001 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3011 3002 *spacep += used;
3012 3003 }
3013 3004 return (0);
3014 3005 }
3015 3006
3016 3007 static void
3017 3008 snaplist_destroy(list_t *l, boolean_t own)
3018 3009 {
3019 3010 struct promotenode *snap;
3020 3011
3021 3012 if (!l || !list_link_active(&l->list_head))
3022 3013 return;
3023 3014
3024 3015 while ((snap = list_tail(l)) != NULL) {
3025 3016 list_remove(l, snap);
3026 3017 if (own)
3027 3018 dsl_dataset_disown(snap->ds, snaplist_tag);
3028 3019 else
3029 3020 dsl_dataset_rele(snap->ds, snaplist_tag);
3030 3021 kmem_free(snap, sizeof (struct promotenode));
3031 3022 }
3032 3023 list_destroy(l);
3033 3024 }
3034 3025
3035 3026 /*
3036 3027 * Promote a clone. Nomenclature note:
3037 3028 * "clone" or "cds": the original clone which is being promoted
3038 3029 * "origin" or "ods": the snapshot which is originally clone's origin
3039 3030 * "origin head" or "ohds": the dataset which is the head
3040 3031 * (filesystem/volume) for the origin
3041 3032 * "origin origin": the origin of the origin's filesystem (typically
3042 3033 * NULL, indicating that the clone is not a clone of a clone).
3043 3034 */
3044 3035 int
3045 3036 dsl_dataset_promote(const char *name, char *conflsnap)
3046 3037 {
3047 3038 dsl_dataset_t *ds;
3048 3039 dsl_dir_t *dd;
3049 3040 dsl_pool_t *dp;
3050 3041 dmu_object_info_t doi;
3051 3042 struct promotearg pa = { 0 };
3052 3043 struct promotenode *snap;
3053 3044 int err;
3054 3045
3055 3046 err = dsl_dataset_hold(name, FTAG, &ds);
3056 3047 if (err)
3057 3048 return (err);
3058 3049 dd = ds->ds_dir;
3059 3050 dp = dd->dd_pool;
3060 3051
3061 3052 err = dmu_object_info(dp->dp_meta_objset,
3062 3053 ds->ds_phys->ds_snapnames_zapobj, &doi);
3063 3054 if (err) {
3064 3055 dsl_dataset_rele(ds, FTAG);
3065 3056 return (err);
3066 3057 }
3067 3058
3068 3059 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3069 3060 dsl_dataset_rele(ds, FTAG);
3070 3061 return (EINVAL);
3071 3062 }
3072 3063
3073 3064 /*
3074 3065 * We are going to inherit all the snapshots taken before our
3075 3066 * origin (i.e., our new origin will be our parent's origin).
3076 3067 * Take ownership of them so that we can rename them into our
3077 3068 * namespace.
3078 3069 */
3079 3070 rw_enter(&dp->dp_config_rwlock, RW_READER);
3080 3071
3081 3072 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3082 3073 &pa.shared_snaps);
3083 3074 if (err != 0)
3084 3075 goto out;
3085 3076
3086 3077 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3087 3078 if (err != 0)
3088 3079 goto out;
3089 3080
3090 3081 snap = list_head(&pa.shared_snaps);
3091 3082 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3092 3083 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3093 3084 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3094 3085 if (err != 0)
3095 3086 goto out;
3096 3087
3097 3088 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3098 3089 err = dsl_dataset_hold_obj(dp,
3099 3090 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3100 3091 FTAG, &pa.origin_origin);
3101 3092 if (err != 0)
3102 3093 goto out;
3103 3094 }
3104 3095
3105 3096 out:
3106 3097 rw_exit(&dp->dp_config_rwlock);
3107 3098
3108 3099 /*
3109 3100 * Add in 128x the snapnames zapobj size, since we will be moving
3110 3101 * a bunch of snapnames to the promoted ds, and dirtying their
3111 3102 * bonus buffers.
3112 3103 */
3113 3104 if (err == 0) {
3114 3105 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3115 3106 dsl_dataset_promote_sync, ds, &pa,
3116 3107 2 + 2 * doi.doi_physical_blocks_512);
3117 3108 if (err && pa.err_ds && conflsnap)
3118 3109 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3119 3110 }
3120 3111
3121 3112 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3122 3113 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3123 3114 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3124 3115 if (pa.origin_origin)
3125 3116 dsl_dataset_rele(pa.origin_origin, FTAG);
3126 3117 dsl_dataset_rele(ds, FTAG);
3127 3118 return (err);
3128 3119 }
3129 3120
3130 3121 struct cloneswaparg {
3131 3122 dsl_dataset_t *cds; /* clone dataset */
3132 3123 dsl_dataset_t *ohds; /* origin's head dataset */
3133 3124 boolean_t force;
3134 3125 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3135 3126 };
3136 3127
3137 3128 /* ARGSUSED */
3138 3129 static int
3139 3130 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3140 3131 {
3141 3132 struct cloneswaparg *csa = arg1;
3142 3133
3143 3134 /* they should both be heads */
3144 3135 if (dsl_dataset_is_snapshot(csa->cds) ||
3145 3136 dsl_dataset_is_snapshot(csa->ohds))
3146 3137 return (EINVAL);
3147 3138
3148 3139 /* the branch point should be just before them */
3149 3140 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3150 3141 return (EINVAL);
3151 3142
3152 3143 /* cds should be the clone (unless they are unrelated) */
3153 3144 if (csa->cds->ds_prev != NULL &&
3154 3145 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3155 3146 csa->ohds->ds_object !=
3156 3147 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3157 3148 return (EINVAL);
3158 3149
3159 3150 /* the clone should be a child of the origin */
3160 3151 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3161 3152 return (EINVAL);
3162 3153
3163 3154 /* ohds shouldn't be modified unless 'force' */
3164 3155 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3165 3156 return (ETXTBSY);
3166 3157
3167 3158 /* adjust amount of any unconsumed refreservation */
3168 3159 csa->unused_refres_delta =
3169 3160 (int64_t)MIN(csa->ohds->ds_reserved,
3170 3161 csa->ohds->ds_phys->ds_unique_bytes) -
3171 3162 (int64_t)MIN(csa->ohds->ds_reserved,
3172 3163 csa->cds->ds_phys->ds_unique_bytes);
3173 3164
3174 3165 if (csa->unused_refres_delta > 0 &&
3175 3166 csa->unused_refres_delta >
3176 3167 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3177 3168 return (ENOSPC);
3178 3169
3179 3170 if (csa->ohds->ds_quota != 0 &&
3180 3171 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3181 3172 return (EDQUOT);
3182 3173
3183 3174 return (0);
3184 3175 }
3185 3176
3186 3177 /* ARGSUSED */
3187 3178 static void
3188 3179 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3189 3180 {
3190 3181 struct cloneswaparg *csa = arg1;
3191 3182 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3192 3183
3193 3184 ASSERT(csa->cds->ds_reserved == 0);
3194 3185 ASSERT(csa->ohds->ds_quota == 0 ||
3195 3186 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3196 3187
3197 3188 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3198 3189 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3199 3190
3200 3191 if (csa->cds->ds_objset != NULL) {
3201 3192 dmu_objset_evict(csa->cds->ds_objset);
3202 3193 csa->cds->ds_objset = NULL;
3203 3194 }
3204 3195
3205 3196 if (csa->ohds->ds_objset != NULL) {
3206 3197 dmu_objset_evict(csa->ohds->ds_objset);
3207 3198 csa->ohds->ds_objset = NULL;
3208 3199 }
3209 3200
3210 3201 /*
3211 3202 * Reset origin's unique bytes, if it exists.
3212 3203 */
3213 3204 if (csa->cds->ds_prev) {
3214 3205 dsl_dataset_t *origin = csa->cds->ds_prev;
3215 3206 uint64_t comp, uncomp;
3216 3207
3217 3208 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3218 3209 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3219 3210 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3220 3211 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3221 3212 }
3222 3213
3223 3214 /* swap blkptrs */
3224 3215 {
3225 3216 blkptr_t tmp;
3226 3217 tmp = csa->ohds->ds_phys->ds_bp;
3227 3218 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3228 3219 csa->cds->ds_phys->ds_bp = tmp;
3229 3220 }
3230 3221
3231 3222 /* set dd_*_bytes */
3232 3223 {
3233 3224 int64_t dused, dcomp, duncomp;
3234 3225 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3235 3226 uint64_t odl_used, odl_comp, odl_uncomp;
3236 3227
3237 3228 ASSERT3U(csa->cds->ds_dir->dd_phys->
3238 3229 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3239 3230
3240 3231 dsl_deadlist_space(&csa->cds->ds_deadlist,
3241 3232 &cdl_used, &cdl_comp, &cdl_uncomp);
3242 3233 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3243 3234 &odl_used, &odl_comp, &odl_uncomp);
3244 3235
3245 3236 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3246 3237 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3247 3238 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3248 3239 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3249 3240 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3250 3241 cdl_uncomp -
3251 3242 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3252 3243
3253 3244 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3254 3245 dused, dcomp, duncomp, tx);
3255 3246 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3256 3247 -dused, -dcomp, -duncomp, tx);
3257 3248
3258 3249 /*
3259 3250 * The difference in the space used by snapshots is the
3260 3251 * difference in snapshot space due to the head's
3261 3252 * deadlist (since that's the only thing that's
3262 3253 * changing that affects the snapused).
3263 3254 */
3264 3255 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3265 3256 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3266 3257 &cdl_used, &cdl_comp, &cdl_uncomp);
3267 3258 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3268 3259 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3269 3260 &odl_used, &odl_comp, &odl_uncomp);
3270 3261 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3271 3262 DD_USED_HEAD, DD_USED_SNAP, tx);
3272 3263 }
3273 3264
3274 3265 /* swap ds_*_bytes */
3275 3266 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3276 3267 csa->cds->ds_phys->ds_referenced_bytes);
3277 3268 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3278 3269 csa->cds->ds_phys->ds_compressed_bytes);
3279 3270 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3280 3271 csa->cds->ds_phys->ds_uncompressed_bytes);
3281 3272 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3282 3273 csa->cds->ds_phys->ds_unique_bytes);
3283 3274
3284 3275 /* apply any parent delta for change in unconsumed refreservation */
3285 3276 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3286 3277 csa->unused_refres_delta, 0, 0, tx);
3287 3278
3288 3279 /*
3289 3280 * Swap deadlists.
3290 3281 */
↓ open down ↓ |
337 lines elided |
↑ open up ↑ |
3291 3282 dsl_deadlist_close(&csa->cds->ds_deadlist);
3292 3283 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3293 3284 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3294 3285 csa->cds->ds_phys->ds_deadlist_obj);
3295 3286 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3296 3287 csa->cds->ds_phys->ds_deadlist_obj);
3297 3288 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3298 3289 csa->ohds->ds_phys->ds_deadlist_obj);
3299 3290
3300 3291 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3292 +
3293 + spa_history_log_internal_ds(csa->cds, "clone swap", tx,
3294 + "parent=%s", csa->ohds->ds_dir->dd_myname);
3301 3295 }
3302 3296
3303 3297 /*
3304 3298 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3305 3299 * recv" into an existing fs to swizzle the file system to the new
3306 3300 * version, and by "zfs rollback". Can also be used to swap two
3307 3301 * independent head datasets if neither has any snapshots.
3308 3302 */
3309 3303 int
3310 3304 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3311 3305 boolean_t force)
3312 3306 {
3313 3307 struct cloneswaparg csa;
3314 3308 int error;
3315 3309
3316 3310 ASSERT(clone->ds_owner);
3317 3311 ASSERT(origin_head->ds_owner);
3318 3312 retry:
3319 3313 /*
3320 3314 * Need exclusive access for the swap. If we're swapping these
3321 3315 * datasets back after an error, we already hold the locks.
3322 3316 */
3323 3317 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3324 3318 rw_enter(&clone->ds_rwlock, RW_WRITER);
3325 3319 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3326 3320 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3327 3321 rw_exit(&clone->ds_rwlock);
3328 3322 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3329 3323 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3330 3324 rw_exit(&origin_head->ds_rwlock);
3331 3325 goto retry;
3332 3326 }
3333 3327 }
3334 3328 csa.cds = clone;
3335 3329 csa.ohds = origin_head;
3336 3330 csa.force = force;
3337 3331 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3338 3332 dsl_dataset_clone_swap_check,
3339 3333 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3340 3334 return (error);
3341 3335 }
3342 3336
3343 3337 /*
3344 3338 * Given a pool name and a dataset object number in that pool,
3345 3339 * return the name of that dataset.
3346 3340 */
3347 3341 int
3348 3342 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3349 3343 {
3350 3344 spa_t *spa;
3351 3345 dsl_pool_t *dp;
3352 3346 dsl_dataset_t *ds;
3353 3347 int error;
3354 3348
3355 3349 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3356 3350 return (error);
3357 3351 dp = spa_get_dsl(spa);
3358 3352 rw_enter(&dp->dp_config_rwlock, RW_READER);
3359 3353 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3360 3354 dsl_dataset_name(ds, buf);
3361 3355 dsl_dataset_rele(ds, FTAG);
3362 3356 }
3363 3357 rw_exit(&dp->dp_config_rwlock);
3364 3358 spa_close(spa, FTAG);
3365 3359
3366 3360 return (error);
3367 3361 }
3368 3362
3369 3363 int
3370 3364 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3371 3365 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3372 3366 {
3373 3367 int error = 0;
3374 3368
3375 3369 ASSERT3S(asize, >, 0);
3376 3370
3377 3371 /*
3378 3372 * *ref_rsrv is the portion of asize that will come from any
3379 3373 * unconsumed refreservation space.
3380 3374 */
3381 3375 *ref_rsrv = 0;
3382 3376
3383 3377 mutex_enter(&ds->ds_lock);
3384 3378 /*
3385 3379 * Make a space adjustment for reserved bytes.
3386 3380 */
3387 3381 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3388 3382 ASSERT3U(*used, >=,
3389 3383 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3390 3384 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3391 3385 *ref_rsrv =
3392 3386 asize - MIN(asize, parent_delta(ds, asize + inflight));
3393 3387 }
3394 3388
3395 3389 if (!check_quota || ds->ds_quota == 0) {
3396 3390 mutex_exit(&ds->ds_lock);
3397 3391 return (0);
3398 3392 }
3399 3393 /*
3400 3394 * If they are requesting more space, and our current estimate
3401 3395 * is over quota, they get to try again unless the actual
3402 3396 * on-disk is over quota and there are no pending changes (which
3403 3397 * may free up space for us).
3404 3398 */
3405 3399 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3406 3400 if (inflight > 0 ||
3407 3401 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3408 3402 error = ERESTART;
3409 3403 else
3410 3404 error = EDQUOT;
3411 3405 }
3412 3406 mutex_exit(&ds->ds_lock);
3413 3407
3414 3408 return (error);
3415 3409 }
3416 3410
3417 3411 /* ARGSUSED */
3418 3412 static int
3419 3413 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3420 3414 {
3421 3415 dsl_dataset_t *ds = arg1;
3422 3416 dsl_prop_setarg_t *psa = arg2;
3423 3417 int err;
3424 3418
3425 3419 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3426 3420 return (ENOTSUP);
3427 3421
3428 3422 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3429 3423 return (err);
3430 3424
3431 3425 if (psa->psa_effective_value == 0)
3432 3426 return (0);
3433 3427
3434 3428 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3435 3429 psa->psa_effective_value < ds->ds_reserved)
3436 3430 return (ENOSPC);
3437 3431
3438 3432 return (0);
3439 3433 }
3440 3434
3441 3435 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3442 3436
3443 3437 void
3444 3438 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3445 3439 {
3446 3440 dsl_dataset_t *ds = arg1;
↓ open down ↓ |
136 lines elided |
↑ open up ↑ |
3447 3441 dsl_prop_setarg_t *psa = arg2;
3448 3442 uint64_t effective_value = psa->psa_effective_value;
3449 3443
3450 3444 dsl_prop_set_sync(ds, psa, tx);
3451 3445 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3452 3446
3453 3447 if (ds->ds_quota != effective_value) {
3454 3448 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3455 3449 ds->ds_quota = effective_value;
3456 3450
3457 - spa_history_log_internal(LOG_DS_REFQUOTA,
3458 - ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3459 - (longlong_t)ds->ds_quota, ds->ds_object);
3451 + spa_history_log_internal_ds(ds, "set refquota", tx,
3452 + "refquota=%lld", (longlong_t)ds->ds_quota);
3460 3453 }
3461 3454 }
3462 3455
3463 3456 int
3464 3457 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3465 3458 {
3466 3459 dsl_dataset_t *ds;
3467 3460 dsl_prop_setarg_t psa;
3468 3461 int err;
3469 3462
3470 3463 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3471 3464
3472 3465 err = dsl_dataset_hold(dsname, FTAG, &ds);
3473 3466 if (err)
3474 3467 return (err);
3475 3468
3476 3469 /*
3477 3470 * If someone removes a file, then tries to set the quota, we
3478 3471 * want to make sure the file freeing takes effect.
3479 3472 */
3480 3473 txg_wait_open(ds->ds_dir->dd_pool, 0);
3481 3474
3482 3475 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3483 3476 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3484 3477 ds, &psa, 0);
3485 3478
3486 3479 dsl_dataset_rele(ds, FTAG);
3487 3480 return (err);
3488 3481 }
3489 3482
3490 3483 static int
3491 3484 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3492 3485 {
3493 3486 dsl_dataset_t *ds = arg1;
3494 3487 dsl_prop_setarg_t *psa = arg2;
3495 3488 uint64_t effective_value;
3496 3489 uint64_t unique;
3497 3490 int err;
3498 3491
3499 3492 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3500 3493 SPA_VERSION_REFRESERVATION)
3501 3494 return (ENOTSUP);
3502 3495
3503 3496 if (dsl_dataset_is_snapshot(ds))
3504 3497 return (EINVAL);
3505 3498
3506 3499 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3507 3500 return (err);
3508 3501
3509 3502 effective_value = psa->psa_effective_value;
3510 3503
3511 3504 /*
3512 3505 * If we are doing the preliminary check in open context, the
3513 3506 * space estimates may be inaccurate.
3514 3507 */
3515 3508 if (!dmu_tx_is_syncing(tx))
3516 3509 return (0);
3517 3510
3518 3511 mutex_enter(&ds->ds_lock);
3519 3512 if (!DS_UNIQUE_IS_ACCURATE(ds))
3520 3513 dsl_dataset_recalc_head_uniq(ds);
3521 3514 unique = ds->ds_phys->ds_unique_bytes;
3522 3515 mutex_exit(&ds->ds_lock);
3523 3516
3524 3517 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3525 3518 uint64_t delta = MAX(unique, effective_value) -
3526 3519 MAX(unique, ds->ds_reserved);
3527 3520
3528 3521 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3529 3522 return (ENOSPC);
3530 3523 if (ds->ds_quota > 0 &&
3531 3524 effective_value > ds->ds_quota)
3532 3525 return (ENOSPC);
3533 3526 }
3534 3527
3535 3528 return (0);
3536 3529 }
3537 3530
3538 3531 static void
3539 3532 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3540 3533 {
3541 3534 dsl_dataset_t *ds = arg1;
3542 3535 dsl_prop_setarg_t *psa = arg2;
3543 3536 uint64_t effective_value = psa->psa_effective_value;
3544 3537 uint64_t unique;
3545 3538 int64_t delta;
3546 3539
3547 3540 dsl_prop_set_sync(ds, psa, tx);
3548 3541 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3549 3542
3550 3543 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3551 3544
3552 3545 mutex_enter(&ds->ds_dir->dd_lock);
3553 3546 mutex_enter(&ds->ds_lock);
↓ open down ↓ |
84 lines elided |
↑ open up ↑ |
3554 3547 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3555 3548 unique = ds->ds_phys->ds_unique_bytes;
3556 3549 delta = MAX(0, (int64_t)(effective_value - unique)) -
3557 3550 MAX(0, (int64_t)(ds->ds_reserved - unique));
3558 3551 ds->ds_reserved = effective_value;
3559 3552 mutex_exit(&ds->ds_lock);
3560 3553
3561 3554 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3562 3555 mutex_exit(&ds->ds_dir->dd_lock);
3563 3556
3564 - spa_history_log_internal(LOG_DS_REFRESERV,
3565 - ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3566 - (longlong_t)effective_value, ds->ds_object);
3557 + spa_history_log_internal_ds(ds, "set refreservation", tx,
3558 + "refreservation=%lld", (longlong_t)effective_value);
3567 3559 }
3568 3560
3569 3561 int
3570 3562 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3571 3563 uint64_t reservation)
3572 3564 {
3573 3565 dsl_dataset_t *ds;
3574 3566 dsl_prop_setarg_t psa;
3575 3567 int err;
3576 3568
3577 3569 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3578 3570 &reservation);
3579 3571
3580 3572 err = dsl_dataset_hold(dsname, FTAG, &ds);
3581 3573 if (err)
3582 3574 return (err);
3583 3575
3584 3576 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3585 3577 dsl_dataset_set_reservation_check,
3586 3578 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3587 3579
3588 3580 dsl_dataset_rele(ds, FTAG);
3589 3581 return (err);
3590 3582 }
3591 3583
3592 3584 typedef struct zfs_hold_cleanup_arg {
3593 3585 dsl_pool_t *dp;
3594 3586 uint64_t dsobj;
3595 3587 char htag[MAXNAMELEN];
3596 3588 } zfs_hold_cleanup_arg_t;
3597 3589
3598 3590 static void
3599 3591 dsl_dataset_user_release_onexit(void *arg)
3600 3592 {
3601 3593 zfs_hold_cleanup_arg_t *ca = arg;
3602 3594
3603 3595 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3604 3596 B_TRUE);
3605 3597 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3606 3598 }
3607 3599
3608 3600 void
3609 3601 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3610 3602 minor_t minor)
3611 3603 {
3612 3604 zfs_hold_cleanup_arg_t *ca;
3613 3605
3614 3606 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3615 3607 ca->dp = ds->ds_dir->dd_pool;
3616 3608 ca->dsobj = ds->ds_object;
3617 3609 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3618 3610 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3619 3611 dsl_dataset_user_release_onexit, ca, NULL));
3620 3612 }
3621 3613
↓ open down ↓ |
45 lines elided |
↑ open up ↑ |
3622 3614 /*
3623 3615 * If you add new checks here, you may need to add
3624 3616 * additional checks to the "temporary" case in
3625 3617 * snapshot_check() in dmu_objset.c.
3626 3618 */
3627 3619 static int
3628 3620 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3629 3621 {
3630 3622 dsl_dataset_t *ds = arg1;
3631 3623 struct dsl_ds_holdarg *ha = arg2;
3632 - char *htag = ha->htag;
3624 + const char *htag = ha->htag;
3633 3625 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3634 3626 int error = 0;
3635 3627
3636 3628 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3637 3629 return (ENOTSUP);
3638 3630
3639 3631 if (!dsl_dataset_is_snapshot(ds))
3640 3632 return (EINVAL);
3641 3633
3642 3634 /* tags must be unique */
3643 3635 mutex_enter(&ds->ds_lock);
3644 3636 if (ds->ds_phys->ds_userrefs_obj) {
3645 3637 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3646 3638 8, 1, tx);
3647 3639 if (error == 0)
3648 3640 error = EEXIST;
3649 3641 else if (error == ENOENT)
3650 3642 error = 0;
3651 3643 }
3652 3644 mutex_exit(&ds->ds_lock);
3653 3645
3654 3646 if (error == 0 && ha->temphold &&
3655 3647 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
3656 3648 error = E2BIG;
3657 3649
3658 3650 return (error);
3659 3651 }
3660 3652
3661 3653 void
3662 3654 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3663 3655 {
3664 3656 dsl_dataset_t *ds = arg1;
3665 3657 struct dsl_ds_holdarg *ha = arg2;
3666 - char *htag = ha->htag;
3658 + const char *htag = ha->htag;
3667 3659 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3668 3660 objset_t *mos = dp->dp_meta_objset;
3669 3661 uint64_t now = gethrestime_sec();
3670 3662 uint64_t zapobj;
3671 3663
3672 3664 mutex_enter(&ds->ds_lock);
3673 3665 if (ds->ds_phys->ds_userrefs_obj == 0) {
3674 3666 /*
3675 3667 * This is the first user hold for this dataset. Create
3676 3668 * the userrefs zap object.
3677 3669 */
3678 3670 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3679 3671 zapobj = ds->ds_phys->ds_userrefs_obj =
3680 3672 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3681 3673 } else {
3682 3674 zapobj = ds->ds_phys->ds_userrefs_obj;
3683 3675 }
↓ open down ↓ |
7 lines elided |
↑ open up ↑ |
3684 3676 ds->ds_userrefs++;
3685 3677 mutex_exit(&ds->ds_lock);
3686 3678
3687 3679 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3688 3680
3689 3681 if (ha->temphold) {
3690 3682 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3691 3683 htag, &now, tx));
3692 3684 }
3693 3685
3694 - spa_history_log_internal(LOG_DS_USER_HOLD,
3695 - dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3696 - (int)ha->temphold, ds->ds_object);
3686 + spa_history_log_internal_ds(ds, "hold", tx,
3687 + "tag = %s temp = %d holds now = %llu",
3688 + htag, (int)ha->temphold, ds->ds_userrefs);
3697 3689 }
3698 3690
3699 3691 static int
3700 3692 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3701 3693 {
3702 3694 struct dsl_ds_holdarg *ha = arg;
3703 3695 dsl_dataset_t *ds;
3704 3696 int error;
3705 3697 char *name;
3706 3698
3707 3699 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3708 3700 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3709 3701 error = dsl_dataset_hold(name, ha->dstg, &ds);
3710 3702 strfree(name);
3711 3703 if (error == 0) {
3712 3704 ha->gotone = B_TRUE;
3713 3705 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3714 3706 dsl_dataset_user_hold_sync, ds, ha, 0);
3715 3707 } else if (error == ENOENT && ha->recursive) {
3716 3708 error = 0;
3717 3709 } else {
3718 3710 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3719 3711 }
3720 3712 return (error);
3721 3713 }
3722 3714
3723 3715 int
3724 3716 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3725 3717 boolean_t temphold)
3726 3718 {
3727 3719 struct dsl_ds_holdarg *ha;
3728 3720 int error;
3729 3721
3730 3722 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3731 3723 ha->htag = htag;
3732 3724 ha->temphold = temphold;
3733 3725 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3734 3726 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3735 3727 ds, ha, 0);
3736 3728 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3737 3729
3738 3730 return (error);
3739 3731 }
3740 3732
3741 3733 int
3742 3734 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3743 3735 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3744 3736 {
3745 3737 struct dsl_ds_holdarg *ha;
3746 3738 dsl_sync_task_t *dst;
3747 3739 spa_t *spa;
3748 3740 int error;
3749 3741 minor_t minor = 0;
3750 3742
3751 3743 if (cleanup_fd != -1) {
3752 3744 /* Currently we only support cleanup-on-exit of tempholds. */
3753 3745 if (!temphold)
3754 3746 return (EINVAL);
3755 3747 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3756 3748 if (error)
3757 3749 return (error);
3758 3750 }
3759 3751
3760 3752 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3761 3753
3762 3754 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3763 3755
3764 3756 error = spa_open(dsname, &spa, FTAG);
3765 3757 if (error) {
3766 3758 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3767 3759 if (cleanup_fd != -1)
3768 3760 zfs_onexit_fd_rele(cleanup_fd);
3769 3761 return (error);
3770 3762 }
3771 3763
3772 3764 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3773 3765 ha->htag = htag;
3774 3766 ha->snapname = snapname;
3775 3767 ha->recursive = recursive;
3776 3768 ha->temphold = temphold;
3777 3769
3778 3770 if (recursive) {
3779 3771 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3780 3772 ha, DS_FIND_CHILDREN);
3781 3773 } else {
3782 3774 error = dsl_dataset_user_hold_one(dsname, ha);
3783 3775 }
3784 3776 if (error == 0)
3785 3777 error = dsl_sync_task_group_wait(ha->dstg);
3786 3778
3787 3779 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3788 3780 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3789 3781 dsl_dataset_t *ds = dst->dst_arg1;
3790 3782
3791 3783 if (dst->dst_err) {
3792 3784 dsl_dataset_name(ds, ha->failed);
3793 3785 *strchr(ha->failed, '@') = '\0';
3794 3786 } else if (error == 0 && minor != 0 && temphold) {
3795 3787 /*
3796 3788 * If this hold is to be released upon process exit,
3797 3789 * register that action now.
3798 3790 */
3799 3791 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3800 3792 }
3801 3793 dsl_dataset_rele(ds, ha->dstg);
3802 3794 }
3803 3795
3804 3796 if (error == 0 && recursive && !ha->gotone)
3805 3797 error = ENOENT;
3806 3798
3807 3799 if (error)
3808 3800 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3809 3801
3810 3802 dsl_sync_task_group_destroy(ha->dstg);
3811 3803
3812 3804 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3813 3805 spa_close(spa, FTAG);
3814 3806 if (cleanup_fd != -1)
3815 3807 zfs_onexit_fd_rele(cleanup_fd);
3816 3808 return (error);
3817 3809 }
3818 3810
3819 3811 struct dsl_ds_releasearg {
3820 3812 dsl_dataset_t *ds;
3821 3813 const char *htag;
3822 3814 boolean_t own; /* do we own or just hold ds? */
3823 3815 };
3824 3816
3825 3817 static int
3826 3818 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3827 3819 boolean_t *might_destroy)
3828 3820 {
3829 3821 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3830 3822 uint64_t zapobj;
3831 3823 uint64_t tmp;
3832 3824 int error;
3833 3825
3834 3826 *might_destroy = B_FALSE;
3835 3827
3836 3828 mutex_enter(&ds->ds_lock);
3837 3829 zapobj = ds->ds_phys->ds_userrefs_obj;
3838 3830 if (zapobj == 0) {
3839 3831 /* The tag can't possibly exist */
3840 3832 mutex_exit(&ds->ds_lock);
3841 3833 return (ESRCH);
3842 3834 }
3843 3835
3844 3836 /* Make sure the tag exists */
3845 3837 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3846 3838 if (error) {
3847 3839 mutex_exit(&ds->ds_lock);
3848 3840 if (error == ENOENT)
3849 3841 error = ESRCH;
3850 3842 return (error);
3851 3843 }
3852 3844
3853 3845 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3854 3846 DS_IS_DEFER_DESTROY(ds))
3855 3847 *might_destroy = B_TRUE;
3856 3848
3857 3849 mutex_exit(&ds->ds_lock);
3858 3850 return (0);
3859 3851 }
3860 3852
3861 3853 static int
3862 3854 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3863 3855 {
3864 3856 struct dsl_ds_releasearg *ra = arg1;
3865 3857 dsl_dataset_t *ds = ra->ds;
3866 3858 boolean_t might_destroy;
3867 3859 int error;
3868 3860
3869 3861 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3870 3862 return (ENOTSUP);
3871 3863
3872 3864 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3873 3865 if (error)
3874 3866 return (error);
3875 3867
3876 3868 if (might_destroy) {
3877 3869 struct dsl_ds_destroyarg dsda = {0};
3878 3870
3879 3871 if (dmu_tx_is_syncing(tx)) {
3880 3872 /*
3881 3873 * If we're not prepared to remove the snapshot,
3882 3874 * we can't allow the release to happen right now.
3883 3875 */
3884 3876 if (!ra->own)
3885 3877 return (EBUSY);
3886 3878 }
3887 3879 dsda.ds = ds;
3888 3880 dsda.releasing = B_TRUE;
3889 3881 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3890 3882 }
3891 3883
3892 3884 return (0);
↓ open down ↓ |
186 lines elided |
↑ open up ↑ |
3893 3885 }
3894 3886
3895 3887 static void
3896 3888 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3897 3889 {
3898 3890 struct dsl_ds_releasearg *ra = arg1;
3899 3891 dsl_dataset_t *ds = ra->ds;
3900 3892 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3901 3893 objset_t *mos = dp->dp_meta_objset;
3902 3894 uint64_t zapobj;
3903 - uint64_t dsobj = ds->ds_object;
3904 3895 uint64_t refs;
3905 3896 int error;
3906 3897
3907 3898 mutex_enter(&ds->ds_lock);
3908 3899 ds->ds_userrefs--;
3909 3900 refs = ds->ds_userrefs;
3910 3901 mutex_exit(&ds->ds_lock);
3911 3902 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3912 3903 VERIFY(error == 0 || error == ENOENT);
3913 3904 zapobj = ds->ds_phys->ds_userrefs_obj;
3914 3905 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3915 3906 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
↓ open down ↓ |
2 lines elided |
↑ open up ↑ |
3916 3907 DS_IS_DEFER_DESTROY(ds)) {
3917 3908 struct dsl_ds_destroyarg dsda = {0};
3918 3909
3919 3910 ASSERT(ra->own);
3920 3911 dsda.ds = ds;
3921 3912 dsda.releasing = B_TRUE;
3922 3913 /* We already did the destroy_check */
3923 3914 dsl_dataset_destroy_sync(&dsda, tag, tx);
3924 3915 }
3925 3916
3926 - spa_history_log_internal(LOG_DS_USER_RELEASE,
3927 - dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3928 - ra->htag, (longlong_t)refs, dsobj);
3917 + spa_history_log_internal_ds(ds, "release", tx,
3918 + "tag = %s refs now = %lld", ra->htag, (longlong_t)refs);
3929 3919 }
3930 3920
3931 3921 static int
3932 3922 dsl_dataset_user_release_one(const char *dsname, void *arg)
3933 3923 {
3934 3924 struct dsl_ds_holdarg *ha = arg;
3935 3925 struct dsl_ds_releasearg *ra;
3936 3926 dsl_dataset_t *ds;
3937 3927 int error;
3938 3928 void *dtag = ha->dstg;
3939 3929 char *name;
3940 3930 boolean_t own = B_FALSE;
3941 3931 boolean_t might_destroy;
3942 3932
3943 3933 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3944 3934 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3945 3935 error = dsl_dataset_hold(name, dtag, &ds);
3946 3936 strfree(name);
3947 3937 if (error == ENOENT && ha->recursive)
3948 3938 return (0);
3949 3939 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3950 3940 if (error)
3951 3941 return (error);
3952 3942
3953 3943 ha->gotone = B_TRUE;
3954 3944
3955 3945 ASSERT(dsl_dataset_is_snapshot(ds));
3956 3946
3957 3947 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3958 3948 if (error) {
3959 3949 dsl_dataset_rele(ds, dtag);
3960 3950 return (error);
3961 3951 }
3962 3952
3963 3953 if (might_destroy) {
3964 3954 #ifdef _KERNEL
3965 3955 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3966 3956 error = zfs_unmount_snap(name, NULL);
3967 3957 strfree(name);
3968 3958 if (error) {
3969 3959 dsl_dataset_rele(ds, dtag);
3970 3960 return (error);
3971 3961 }
3972 3962 #endif
3973 3963 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3974 3964 dsl_dataset_rele(ds, dtag);
3975 3965 return (EBUSY);
3976 3966 } else {
3977 3967 own = B_TRUE;
3978 3968 dsl_dataset_make_exclusive(ds, dtag);
3979 3969 }
3980 3970 }
3981 3971
3982 3972 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3983 3973 ra->ds = ds;
3984 3974 ra->htag = ha->htag;
3985 3975 ra->own = own;
3986 3976 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3987 3977 dsl_dataset_user_release_sync, ra, dtag, 0);
3988 3978
3989 3979 return (0);
3990 3980 }
3991 3981
3992 3982 int
3993 3983 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3994 3984 boolean_t recursive)
3995 3985 {
3996 3986 struct dsl_ds_holdarg *ha;
3997 3987 dsl_sync_task_t *dst;
3998 3988 spa_t *spa;
3999 3989 int error;
4000 3990
4001 3991 top:
4002 3992 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
4003 3993
4004 3994 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4005 3995
4006 3996 error = spa_open(dsname, &spa, FTAG);
4007 3997 if (error) {
4008 3998 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4009 3999 return (error);
4010 4000 }
4011 4001
4012 4002 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4013 4003 ha->htag = htag;
4014 4004 ha->snapname = snapname;
4015 4005 ha->recursive = recursive;
4016 4006 if (recursive) {
4017 4007 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4018 4008 ha, DS_FIND_CHILDREN);
4019 4009 } else {
4020 4010 error = dsl_dataset_user_release_one(dsname, ha);
4021 4011 }
4022 4012 if (error == 0)
4023 4013 error = dsl_sync_task_group_wait(ha->dstg);
4024 4014
4025 4015 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4026 4016 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4027 4017 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4028 4018 dsl_dataset_t *ds = ra->ds;
4029 4019
4030 4020 if (dst->dst_err)
4031 4021 dsl_dataset_name(ds, ha->failed);
4032 4022
4033 4023 if (ra->own)
4034 4024 dsl_dataset_disown(ds, ha->dstg);
4035 4025 else
4036 4026 dsl_dataset_rele(ds, ha->dstg);
4037 4027
4038 4028 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4039 4029 }
4040 4030
4041 4031 if (error == 0 && recursive && !ha->gotone)
4042 4032 error = ENOENT;
4043 4033
4044 4034 if (error && error != EBUSY)
4045 4035 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4046 4036
4047 4037 dsl_sync_task_group_destroy(ha->dstg);
4048 4038 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4049 4039 spa_close(spa, FTAG);
4050 4040
4051 4041 /*
4052 4042 * We can get EBUSY if we were racing with deferred destroy and
4053 4043 * dsl_dataset_user_release_check() hadn't done the necessary
4054 4044 * open context setup. We can also get EBUSY if we're racing
4055 4045 * with destroy and that thread is the ds_owner. Either way
4056 4046 * the busy condition should be transient, and we should retry
4057 4047 * the release operation.
4058 4048 */
4059 4049 if (error == EBUSY)
4060 4050 goto top;
4061 4051
4062 4052 return (error);
4063 4053 }
4064 4054
4065 4055 /*
4066 4056 * Called at spa_load time (with retry == B_FALSE) to release a stale
4067 4057 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4068 4058 */
4069 4059 int
4070 4060 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4071 4061 boolean_t retry)
4072 4062 {
4073 4063 dsl_dataset_t *ds;
4074 4064 char *snap;
4075 4065 char *name;
4076 4066 int namelen;
4077 4067 int error;
4078 4068
4079 4069 do {
4080 4070 rw_enter(&dp->dp_config_rwlock, RW_READER);
4081 4071 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4082 4072 rw_exit(&dp->dp_config_rwlock);
4083 4073 if (error)
4084 4074 return (error);
4085 4075 namelen = dsl_dataset_namelen(ds)+1;
4086 4076 name = kmem_alloc(namelen, KM_SLEEP);
4087 4077 dsl_dataset_name(ds, name);
4088 4078 dsl_dataset_rele(ds, FTAG);
4089 4079
4090 4080 snap = strchr(name, '@');
4091 4081 *snap = '\0';
4092 4082 ++snap;
4093 4083 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4094 4084 kmem_free(name, namelen);
4095 4085
4096 4086 /*
4097 4087 * The object can't have been destroyed because we have a hold,
4098 4088 * but it might have been renamed, resulting in ENOENT. Retry
4099 4089 * if we've been requested to do so.
4100 4090 *
4101 4091 * It would be nice if we could use the dsobj all the way
4102 4092 * through and avoid ENOENT entirely. But we might need to
4103 4093 * unmount the snapshot, and there's currently no way to lookup
4104 4094 * a vfsp using a ZFS object id.
4105 4095 */
4106 4096 } while ((error == ENOENT) && retry);
4107 4097
4108 4098 return (error);
4109 4099 }
4110 4100
4111 4101 int
4112 4102 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4113 4103 {
4114 4104 dsl_dataset_t *ds;
4115 4105 int err;
4116 4106
4117 4107 err = dsl_dataset_hold(dsname, FTAG, &ds);
4118 4108 if (err)
4119 4109 return (err);
4120 4110
4121 4111 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4122 4112 if (ds->ds_phys->ds_userrefs_obj != 0) {
4123 4113 zap_attribute_t *za;
4124 4114 zap_cursor_t zc;
4125 4115
4126 4116 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4127 4117 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4128 4118 ds->ds_phys->ds_userrefs_obj);
4129 4119 zap_cursor_retrieve(&zc, za) == 0;
4130 4120 zap_cursor_advance(&zc)) {
4131 4121 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4132 4122 za->za_first_integer));
4133 4123 }
4134 4124 zap_cursor_fini(&zc);
4135 4125 kmem_free(za, sizeof (zap_attribute_t));
4136 4126 }
4137 4127 dsl_dataset_rele(ds, FTAG);
4138 4128 return (0);
4139 4129 }
4140 4130
4141 4131 /*
4142 4132 * Note, this function is used as the callback for dmu_objset_find(). We
4143 4133 * always return 0 so that we will continue to find and process
4144 4134 * inconsistent datasets, even if we encounter an error trying to
4145 4135 * process one of them.
4146 4136 */
4147 4137 /* ARGSUSED */
4148 4138 int
4149 4139 dsl_destroy_inconsistent(const char *dsname, void *arg)
4150 4140 {
4151 4141 dsl_dataset_t *ds;
4152 4142
4153 4143 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4154 4144 if (DS_IS_INCONSISTENT(ds))
4155 4145 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4156 4146 else
4157 4147 dsl_dataset_disown(ds, FTAG);
4158 4148 }
4159 4149 return (0);
4160 4150 }
4161 4151
4162 4152 /*
4163 4153 * Return (in *usedp) the amount of space written in new that is not
4164 4154 * present in oldsnap. New may be a snapshot or the head. Old must be
4165 4155 * a snapshot before new, in new's filesystem (or its origin). If not then
4166 4156 * fail and return EINVAL.
4167 4157 *
4168 4158 * The written space is calculated by considering two components: First, we
4169 4159 * ignore any freed space, and calculate the written as new's used space
4170 4160 * minus old's used space. Next, we add in the amount of space that was freed
4171 4161 * between the two snapshots, thus reducing new's used space relative to old's.
4172 4162 * Specifically, this is the space that was born before old->ds_creation_txg,
4173 4163 * and freed before new (ie. on new's deadlist or a previous deadlist).
4174 4164 *
4175 4165 * space freed [---------------------]
4176 4166 * snapshots ---O-------O--------O-------O------
4177 4167 * oldsnap new
4178 4168 */
4179 4169 int
4180 4170 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4181 4171 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4182 4172 {
4183 4173 int err = 0;
4184 4174 uint64_t snapobj;
4185 4175 dsl_pool_t *dp = new->ds_dir->dd_pool;
4186 4176
4187 4177 *usedp = 0;
4188 4178 *usedp += new->ds_phys->ds_referenced_bytes;
4189 4179 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4190 4180
4191 4181 *compp = 0;
4192 4182 *compp += new->ds_phys->ds_compressed_bytes;
4193 4183 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4194 4184
4195 4185 *uncompp = 0;
4196 4186 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4197 4187 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4198 4188
4199 4189 rw_enter(&dp->dp_config_rwlock, RW_READER);
4200 4190 snapobj = new->ds_object;
4201 4191 while (snapobj != oldsnap->ds_object) {
4202 4192 dsl_dataset_t *snap;
4203 4193 uint64_t used, comp, uncomp;
4204 4194
4205 4195 if (snapobj == new->ds_object) {
4206 4196 snap = new;
4207 4197 } else {
4208 4198 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4209 4199 if (err != 0)
4210 4200 break;
4211 4201 }
4212 4202
4213 4203 if (snap->ds_phys->ds_prev_snap_txg ==
4214 4204 oldsnap->ds_phys->ds_creation_txg) {
4215 4205 /*
4216 4206 * The blocks in the deadlist can not be born after
4217 4207 * ds_prev_snap_txg, so get the whole deadlist space,
4218 4208 * which is more efficient (especially for old-format
4219 4209 * deadlists). Unfortunately the deadlist code
4220 4210 * doesn't have enough information to make this
4221 4211 * optimization itself.
4222 4212 */
4223 4213 dsl_deadlist_space(&snap->ds_deadlist,
4224 4214 &used, &comp, &uncomp);
4225 4215 } else {
4226 4216 dsl_deadlist_space_range(&snap->ds_deadlist,
4227 4217 0, oldsnap->ds_phys->ds_creation_txg,
4228 4218 &used, &comp, &uncomp);
4229 4219 }
4230 4220 *usedp += used;
4231 4221 *compp += comp;
4232 4222 *uncompp += uncomp;
4233 4223
4234 4224 /*
4235 4225 * If we get to the beginning of the chain of snapshots
4236 4226 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4237 4227 * was not a snapshot of/before new.
4238 4228 */
4239 4229 snapobj = snap->ds_phys->ds_prev_snap_obj;
4240 4230 if (snap != new)
4241 4231 dsl_dataset_rele(snap, FTAG);
4242 4232 if (snapobj == 0) {
4243 4233 err = EINVAL;
4244 4234 break;
4245 4235 }
4246 4236
4247 4237 }
4248 4238 rw_exit(&dp->dp_config_rwlock);
4249 4239 return (err);
4250 4240 }
4251 4241
4252 4242 /*
4253 4243 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4254 4244 * lastsnap, and all snapshots in between are deleted.
4255 4245 *
4256 4246 * blocks that would be freed [---------------------------]
4257 4247 * snapshots ---O-------O--------O-------O--------O
4258 4248 * firstsnap lastsnap
4259 4249 *
4260 4250 * This is the set of blocks that were born after the snap before firstsnap,
4261 4251 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4262 4252 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4263 4253 * We calculate this by iterating over the relevant deadlists (from the snap
4264 4254 * after lastsnap, backward to the snap after firstsnap), summing up the
4265 4255 * space on the deadlist that was born after the snap before firstsnap.
4266 4256 */
4267 4257 int
4268 4258 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4269 4259 dsl_dataset_t *lastsnap,
4270 4260 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4271 4261 {
4272 4262 int err = 0;
4273 4263 uint64_t snapobj;
4274 4264 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4275 4265
4276 4266 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4277 4267 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4278 4268
4279 4269 /*
4280 4270 * Check that the snapshots are in the same dsl_dir, and firstsnap
4281 4271 * is before lastsnap.
4282 4272 */
4283 4273 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4284 4274 firstsnap->ds_phys->ds_creation_txg >
4285 4275 lastsnap->ds_phys->ds_creation_txg)
4286 4276 return (EINVAL);
4287 4277
4288 4278 *usedp = *compp = *uncompp = 0;
4289 4279
4290 4280 rw_enter(&dp->dp_config_rwlock, RW_READER);
4291 4281 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4292 4282 while (snapobj != firstsnap->ds_object) {
4293 4283 dsl_dataset_t *ds;
4294 4284 uint64_t used, comp, uncomp;
4295 4285
4296 4286 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4297 4287 if (err != 0)
4298 4288 break;
4299 4289
4300 4290 dsl_deadlist_space_range(&ds->ds_deadlist,
4301 4291 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4302 4292 &used, &comp, &uncomp);
4303 4293 *usedp += used;
4304 4294 *compp += comp;
4305 4295 *uncompp += uncomp;
4306 4296
4307 4297 snapobj = ds->ds_phys->ds_prev_snap_obj;
4308 4298 ASSERT3U(snapobj, !=, 0);
4309 4299 dsl_dataset_rele(ds, FTAG);
4310 4300 }
4311 4301 rw_exit(&dp->dp_config_rwlock);
4312 4302 return (err);
4313 4303 }
↓ open down ↓ |
375 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX