Print this page
5056 ZFS deadlock on db_mtx and dn_holds
Reviewed by: Will Andrews <willa@spectralogic.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Approved by: Dan McDonald <danmcd@omniti.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
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
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) 2011, 2014 by Delphix. All rights reserved.
24 24 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25 25 * Copyright (c) 2014 RackTop Systems.
26 + * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
26 27 */
27 28
28 29 #include <sys/dmu_objset.h>
29 30 #include <sys/dsl_dataset.h>
30 31 #include <sys/dsl_dir.h>
31 32 #include <sys/dsl_prop.h>
32 33 #include <sys/dsl_synctask.h>
33 34 #include <sys/dmu_traverse.h>
34 35 #include <sys/dmu_impl.h>
35 36 #include <sys/dmu_tx.h>
36 37 #include <sys/arc.h>
37 38 #include <sys/zio.h>
38 39 #include <sys/zap.h>
39 40 #include <sys/zfeature.h>
40 41 #include <sys/unique.h>
41 42 #include <sys/zfs_context.h>
42 43 #include <sys/zfs_ioctl.h>
43 44 #include <sys/spa.h>
44 45 #include <sys/zfs_znode.h>
45 46 #include <sys/zfs_onexit.h>
46 47 #include <sys/zvol.h>
47 48 #include <sys/dsl_scan.h>
48 49 #include <sys/dsl_deadlist.h>
49 50 #include <sys/dsl_destroy.h>
50 51 #include <sys/dsl_userhold.h>
51 52 #include <sys/dsl_bookmark.h>
52 53
53 54 /*
54 55 * The SPA supports block sizes up to 16MB. However, very large blocks
55 56 * can have an impact on i/o latency (e.g. tying up a spinning disk for
56 57 * ~300ms), and also potentially on the memory allocator. Therefore,
57 58 * we do not allow the recordsize to be set larger than zfs_max_recordsize
58 59 * (default 1MB). Larger blocks can be created by changing this tunable,
59 60 * and pools with larger blocks can always be imported and used, regardless
60 61 * of this setting.
61 62 */
62 63 int zfs_max_recordsize = 1 * 1024 * 1024;
63 64
|
↓ open down ↓ |
28 lines elided |
↑ open up ↑ |
64 65 #define SWITCH64(x, y) \
65 66 { \
66 67 uint64_t __tmp = (x); \
67 68 (x) = (y); \
68 69 (y) = __tmp; \
69 70 }
70 71
71 72 #define DS_REF_MAX (1ULL << 62)
72 73
73 74 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
74 -extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
75 75
76 76 /*
77 77 * Figure out how much of this delta should be propogated to the dsl_dir
78 78 * layer. If there's a refreservation, that space has already been
79 79 * partially accounted for in our ancestors.
80 80 */
81 81 static int64_t
82 82 parent_delta(dsl_dataset_t *ds, int64_t delta)
83 83 {
84 84 dsl_dataset_phys_t *ds_phys;
85 85 uint64_t old_bytes, new_bytes;
86 86
87 87 if (ds->ds_reserved == 0)
88 88 return (delta);
89 89
90 90 ds_phys = dsl_dataset_phys(ds);
91 91 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
92 92 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
93 93
94 94 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
95 95 return (new_bytes - old_bytes);
96 96 }
97 97
98 98 void
99 99 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
100 100 {
101 101 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
102 102 int compressed = BP_GET_PSIZE(bp);
103 103 int uncompressed = BP_GET_UCSIZE(bp);
104 104 int64_t delta;
105 105
106 106 dprintf_bp(bp, "ds=%p", ds);
107 107
108 108 ASSERT(dmu_tx_is_syncing(tx));
109 109 /* It could have been compressed away to nothing */
110 110 if (BP_IS_HOLE(bp))
111 111 return;
112 112 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
113 113 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
114 114 if (ds == NULL) {
115 115 dsl_pool_mos_diduse_space(tx->tx_pool,
116 116 used, compressed, uncompressed);
117 117 return;
118 118 }
119 119
120 120 dmu_buf_will_dirty(ds->ds_dbuf, tx);
121 121 mutex_enter(&ds->ds_lock);
122 122 delta = parent_delta(ds, used);
123 123 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
124 124 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
125 125 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
126 126 dsl_dataset_phys(ds)->ds_unique_bytes += used;
127 127 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
128 128 ds->ds_need_large_blocks = B_TRUE;
129 129 mutex_exit(&ds->ds_lock);
130 130 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
131 131 compressed, uncompressed, tx);
132 132 dsl_dir_transfer_space(ds->ds_dir, used - delta,
133 133 DD_USED_REFRSRV, DD_USED_HEAD, tx);
134 134 }
135 135
136 136 int
137 137 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
138 138 boolean_t async)
139 139 {
140 140 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
141 141 int compressed = BP_GET_PSIZE(bp);
142 142 int uncompressed = BP_GET_UCSIZE(bp);
143 143
144 144 if (BP_IS_HOLE(bp))
145 145 return (0);
146 146
147 147 ASSERT(dmu_tx_is_syncing(tx));
|
↓ open down ↓ |
63 lines elided |
↑ open up ↑ |
148 148 ASSERT(bp->blk_birth <= tx->tx_txg);
149 149
150 150 if (ds == NULL) {
151 151 dsl_free(tx->tx_pool, tx->tx_txg, bp);
152 152 dsl_pool_mos_diduse_space(tx->tx_pool,
153 153 -used, -compressed, -uncompressed);
154 154 return (used);
155 155 }
156 156 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
157 157
158 - ASSERT(!dsl_dataset_is_snapshot(ds));
158 + ASSERT(!ds->ds_is_snapshot);
159 159 dmu_buf_will_dirty(ds->ds_dbuf, tx);
160 160
161 161 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
162 162 int64_t delta;
163 163
164 164 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
165 165 dsl_free(tx->tx_pool, tx->tx_txg, bp);
166 166
167 167 mutex_enter(&ds->ds_lock);
168 168 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
169 169 !DS_UNIQUE_IS_ACCURATE(ds));
170 170 delta = parent_delta(ds, -used);
171 171 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
172 172 mutex_exit(&ds->ds_lock);
173 173 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
174 174 delta, -compressed, -uncompressed, tx);
175 175 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
176 176 DD_USED_REFRSRV, DD_USED_HEAD, tx);
177 177 } else {
178 178 dprintf_bp(bp, "putting on dead list: %s", "");
179 179 if (async) {
180 180 /*
181 181 * We are here as part of zio's write done callback,
182 182 * which means we're a zio interrupt thread. We can't
183 183 * call dsl_deadlist_insert() now because it may block
184 184 * waiting for I/O. Instead, put bp on the deferred
185 185 * queue and let dsl_pool_sync() finish the job.
186 186 */
187 187 bplist_append(&ds->ds_pending_deadlist, bp);
188 188 } else {
189 189 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
190 190 }
191 191 ASSERT3U(ds->ds_prev->ds_object, ==,
192 192 dsl_dataset_phys(ds)->ds_prev_snap_obj);
193 193 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
194 194 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
195 195 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
196 196 ds->ds_object && bp->blk_birth >
197 197 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
198 198 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
199 199 mutex_enter(&ds->ds_prev->ds_lock);
200 200 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
201 201 mutex_exit(&ds->ds_prev->ds_lock);
202 202 }
203 203 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
204 204 dsl_dir_transfer_space(ds->ds_dir, used,
205 205 DD_USED_HEAD, DD_USED_SNAP, tx);
206 206 }
207 207 }
208 208 mutex_enter(&ds->ds_lock);
209 209 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
210 210 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
211 211 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
212 212 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
213 213 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
214 214 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
215 215 mutex_exit(&ds->ds_lock);
216 216
217 217 return (used);
218 218 }
219 219
220 220 uint64_t
221 221 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
222 222 {
223 223 uint64_t trysnap = 0;
224 224
225 225 if (ds == NULL)
226 226 return (0);
227 227 /*
228 228 * The snapshot creation could fail, but that would cause an
229 229 * incorrect FALSE return, which would only result in an
230 230 * overestimation of the amount of space that an operation would
231 231 * consume, which is OK.
232 232 *
233 233 * There's also a small window where we could miss a pending
234 234 * snapshot, because we could set the sync task in the quiescing
235 235 * phase. So this should only be used as a guess.
236 236 */
237 237 if (ds->ds_trysnap_txg >
238 238 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
239 239 trysnap = ds->ds_trysnap_txg;
240 240 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
241 241 }
242 242
243 243 boolean_t
244 244 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
245 245 uint64_t blk_birth)
|
↓ open down ↓ |
77 lines elided |
↑ open up ↑ |
246 246 {
247 247 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
248 248 (bp != NULL && BP_IS_HOLE(bp)))
249 249 return (B_FALSE);
250 250
251 251 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
252 252
253 253 return (B_TRUE);
254 254 }
255 255
256 -/* ARGSUSED */
257 256 static void
258 -dsl_dataset_evict(dmu_buf_t *db, void *dsv)
257 +dsl_dataset_evict(void *dbu)
259 258 {
260 - dsl_dataset_t *ds = dsv;
259 + dsl_dataset_t *ds = dbu;
261 260
262 261 ASSERT(ds->ds_owner == NULL);
263 262
263 + ds->ds_dbuf = NULL;
264 +
264 265 unique_remove(ds->ds_fsid_guid);
265 266
266 267 if (ds->ds_objset != NULL)
267 268 dmu_objset_evict(ds->ds_objset);
268 269
269 270 if (ds->ds_prev) {
270 271 dsl_dataset_rele(ds->ds_prev, ds);
271 272 ds->ds_prev = NULL;
272 273 }
273 274
274 275 bplist_destroy(&ds->ds_pending_deadlist);
275 - if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
276 + if (ds->ds_deadlist.dl_os != NULL)
276 277 dsl_deadlist_close(&ds->ds_deadlist);
277 278 if (ds->ds_dir)
278 - dsl_dir_rele(ds->ds_dir, ds);
279 + dsl_dir_async_rele(ds->ds_dir, ds);
279 280
280 281 ASSERT(!list_link_active(&ds->ds_synced_link));
281 282
282 283 mutex_destroy(&ds->ds_lock);
283 284 mutex_destroy(&ds->ds_opening_lock);
284 285 mutex_destroy(&ds->ds_sendstream_lock);
285 286 refcount_destroy(&ds->ds_longholds);
286 287
287 288 kmem_free(ds, sizeof (dsl_dataset_t));
288 289 }
289 290
290 291 int
291 292 dsl_dataset_get_snapname(dsl_dataset_t *ds)
292 293 {
293 294 dsl_dataset_phys_t *headphys;
294 295 int err;
295 296 dmu_buf_t *headdbuf;
296 297 dsl_pool_t *dp = ds->ds_dir->dd_pool;
297 298 objset_t *mos = dp->dp_meta_objset;
298 299
299 300 if (ds->ds_snapname[0])
300 301 return (0);
301 302 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
302 303 return (0);
303 304
304 305 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
305 306 FTAG, &headdbuf);
306 307 if (err != 0)
307 308 return (err);
308 309 headphys = headdbuf->db_data;
309 310 err = zap_value_search(dp->dp_meta_objset,
310 311 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
311 312 dmu_buf_rele(headdbuf, FTAG);
312 313 return (err);
313 314 }
314 315
315 316 int
316 317 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
317 318 {
318 319 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
319 320 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
320 321 matchtype_t mt;
321 322 int err;
322 323
323 324 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
324 325 mt = MT_FIRST;
325 326 else
326 327 mt = MT_EXACT;
327 328
328 329 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
329 330 value, mt, NULL, 0, NULL);
330 331 if (err == ENOTSUP && mt == MT_FIRST)
331 332 err = zap_lookup(mos, snapobj, name, 8, 1, value);
332 333 return (err);
333 334 }
334 335
335 336 int
336 337 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
337 338 boolean_t adj_cnt)
338 339 {
339 340 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
340 341 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
341 342 matchtype_t mt;
342 343 int err;
343 344
344 345 dsl_dir_snap_cmtime_update(ds->ds_dir);
345 346
346 347 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
347 348 mt = MT_FIRST;
348 349 else
349 350 mt = MT_EXACT;
350 351
351 352 err = zap_remove_norm(mos, snapobj, name, mt, tx);
352 353 if (err == ENOTSUP && mt == MT_FIRST)
353 354 err = zap_remove(mos, snapobj, name, tx);
354 355
355 356 if (err == 0 && adj_cnt)
356 357 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
357 358 DD_FIELD_SNAPSHOT_COUNT, tx);
358 359
359 360 return (err);
360 361 }
361 362
362 363 int
363 364 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
364 365 dsl_dataset_t **dsp)
365 366 {
366 367 objset_t *mos = dp->dp_meta_objset;
367 368 dmu_buf_t *dbuf;
368 369 dsl_dataset_t *ds;
369 370 int err;
370 371 dmu_object_info_t doi;
371 372
372 373 ASSERT(dsl_pool_config_held(dp));
373 374
374 375 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
375 376 if (err != 0)
376 377 return (err);
377 378
378 379 /* Make sure dsobj has the correct object type. */
379 380 dmu_object_info_from_db(dbuf, &doi);
380 381 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
381 382 dmu_buf_rele(dbuf, tag);
|
↓ open down ↓ |
93 lines elided |
↑ open up ↑ |
382 383 return (SET_ERROR(EINVAL));
383 384 }
384 385
385 386 ds = dmu_buf_get_user(dbuf);
386 387 if (ds == NULL) {
387 388 dsl_dataset_t *winner = NULL;
388 389
389 390 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
390 391 ds->ds_dbuf = dbuf;
391 392 ds->ds_object = dsobj;
393 + ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
392 394
393 395 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
394 396 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
395 397 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
396 398 refcount_create(&ds->ds_longholds);
397 399
398 400 bplist_create(&ds->ds_pending_deadlist);
399 401 dsl_deadlist_open(&ds->ds_deadlist,
400 402 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
401 403
402 404 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
403 405 offsetof(dmu_sendarg_t, dsa_link));
404 406
405 407 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
406 408 err = zap_contains(mos, dsobj, DS_FIELD_LARGE_BLOCKS);
407 409 if (err == 0)
408 410 ds->ds_large_blocks = B_TRUE;
409 411 else
410 412 ASSERT3U(err, ==, ENOENT);
411 413 }
412 414
413 415 if (err == 0) {
414 416 err = dsl_dir_hold_obj(dp,
415 417 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
416 418 &ds->ds_dir);
417 419 }
418 420 if (err != 0) {
419 421 mutex_destroy(&ds->ds_lock);
|
↓ open down ↓ |
18 lines elided |
↑ open up ↑ |
420 422 mutex_destroy(&ds->ds_opening_lock);
421 423 mutex_destroy(&ds->ds_sendstream_lock);
422 424 refcount_destroy(&ds->ds_longholds);
423 425 bplist_destroy(&ds->ds_pending_deadlist);
424 426 dsl_deadlist_close(&ds->ds_deadlist);
425 427 kmem_free(ds, sizeof (dsl_dataset_t));
426 428 dmu_buf_rele(dbuf, tag);
427 429 return (err);
428 430 }
429 431
430 - if (!dsl_dataset_is_snapshot(ds)) {
432 + if (!ds->ds_is_snapshot) {
431 433 ds->ds_snapname[0] = '\0';
432 434 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
433 435 err = dsl_dataset_hold_obj(dp,
434 436 dsl_dataset_phys(ds)->ds_prev_snap_obj,
435 437 ds, &ds->ds_prev);
436 438 }
437 439 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
438 440 int zaperr = zap_lookup(mos, ds->ds_object,
439 441 DS_FIELD_BOOKMARK_NAMES,
440 442 sizeof (ds->ds_bookmarks), 1,
441 443 &ds->ds_bookmarks);
442 444 if (zaperr != ENOENT)
443 445 VERIFY0(zaperr);
444 446 }
445 447 } else {
446 448 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
|
↓ open down ↓ |
6 lines elided |
↑ open up ↑ |
447 449 err = dsl_dataset_get_snapname(ds);
448 450 if (err == 0 &&
449 451 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
450 452 err = zap_count(
451 453 ds->ds_dir->dd_pool->dp_meta_objset,
452 454 dsl_dataset_phys(ds)->ds_userrefs_obj,
453 455 &ds->ds_userrefs);
454 456 }
455 457 }
456 458
457 - if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
459 + if (err == 0 && !ds->ds_is_snapshot) {
458 460 err = dsl_prop_get_int_ds(ds,
459 461 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
460 462 &ds->ds_reserved);
461 463 if (err == 0) {
462 464 err = dsl_prop_get_int_ds(ds,
463 465 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
464 466 &ds->ds_quota);
465 467 }
466 468 } else {
467 469 ds->ds_reserved = ds->ds_quota = 0;
468 470 }
469 471
470 - if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
471 - dsl_dataset_evict)) != NULL) {
472 + dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
473 + if (err == 0)
474 + winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
475 +
476 + if (err != 0 || winner != NULL) {
472 477 bplist_destroy(&ds->ds_pending_deadlist);
473 478 dsl_deadlist_close(&ds->ds_deadlist);
474 479 if (ds->ds_prev)
475 480 dsl_dataset_rele(ds->ds_prev, ds);
476 481 dsl_dir_rele(ds->ds_dir, ds);
477 482 mutex_destroy(&ds->ds_lock);
478 483 mutex_destroy(&ds->ds_opening_lock);
479 484 mutex_destroy(&ds->ds_sendstream_lock);
480 485 refcount_destroy(&ds->ds_longholds);
481 486 kmem_free(ds, sizeof (dsl_dataset_t));
482 487 if (err != 0) {
483 488 dmu_buf_rele(dbuf, tag);
484 489 return (err);
485 490 }
486 491 ds = winner;
487 492 } else {
488 493 ds->ds_fsid_guid =
489 494 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
490 495 }
491 496 }
492 497 ASSERT3P(ds->ds_dbuf, ==, dbuf);
493 498 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
494 499 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
495 500 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
496 501 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
497 502 *dsp = ds;
498 503 return (0);
499 504 }
500 505
501 506 int
502 507 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
503 508 void *tag, dsl_dataset_t **dsp)
504 509 {
505 510 dsl_dir_t *dd;
506 511 const char *snapname;
507 512 uint64_t obj;
508 513 int err = 0;
509 514
510 515 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
511 516 if (err != 0)
512 517 return (err);
513 518
514 519 ASSERT(dsl_pool_config_held(dp));
515 520 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
516 521 if (obj != 0)
517 522 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
518 523 else
519 524 err = SET_ERROR(ENOENT);
520 525
521 526 /* we may be looking for a snapshot */
522 527 if (err == 0 && snapname != NULL) {
523 528 dsl_dataset_t *ds;
524 529
525 530 if (*snapname++ != '@') {
526 531 dsl_dataset_rele(*dsp, tag);
527 532 dsl_dir_rele(dd, FTAG);
528 533 return (SET_ERROR(ENOENT));
529 534 }
530 535
531 536 dprintf("looking for snapshot '%s'\n", snapname);
532 537 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
533 538 if (err == 0)
534 539 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
535 540 dsl_dataset_rele(*dsp, tag);
536 541
537 542 if (err == 0) {
538 543 mutex_enter(&ds->ds_lock);
539 544 if (ds->ds_snapname[0] == 0)
540 545 (void) strlcpy(ds->ds_snapname, snapname,
541 546 sizeof (ds->ds_snapname));
542 547 mutex_exit(&ds->ds_lock);
543 548 *dsp = ds;
544 549 }
545 550 }
546 551
547 552 dsl_dir_rele(dd, FTAG);
548 553 return (err);
549 554 }
550 555
551 556 int
552 557 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
553 558 void *tag, dsl_dataset_t **dsp)
554 559 {
555 560 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
556 561 if (err != 0)
557 562 return (err);
558 563 if (!dsl_dataset_tryown(*dsp, tag)) {
559 564 dsl_dataset_rele(*dsp, tag);
560 565 *dsp = NULL;
561 566 return (SET_ERROR(EBUSY));
562 567 }
563 568 return (0);
564 569 }
565 570
566 571 int
567 572 dsl_dataset_own(dsl_pool_t *dp, const char *name,
568 573 void *tag, dsl_dataset_t **dsp)
569 574 {
570 575 int err = dsl_dataset_hold(dp, name, tag, dsp);
571 576 if (err != 0)
572 577 return (err);
573 578 if (!dsl_dataset_tryown(*dsp, tag)) {
574 579 dsl_dataset_rele(*dsp, tag);
575 580 return (SET_ERROR(EBUSY));
576 581 }
577 582 return (0);
578 583 }
579 584
580 585 /*
581 586 * See the comment above dsl_pool_hold() for details. In summary, a long
582 587 * hold is used to prevent destruction of a dataset while the pool hold
583 588 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
584 589 *
585 590 * The dataset and pool must be held when this function is called. After it
586 591 * is called, the pool hold may be released while the dataset is still held
587 592 * and accessed.
588 593 */
589 594 void
590 595 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
591 596 {
592 597 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
593 598 (void) refcount_add(&ds->ds_longholds, tag);
594 599 }
595 600
596 601 void
597 602 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
598 603 {
599 604 (void) refcount_remove(&ds->ds_longholds, tag);
600 605 }
601 606
602 607 /* Return B_TRUE if there are any long holds on this dataset. */
603 608 boolean_t
604 609 dsl_dataset_long_held(dsl_dataset_t *ds)
605 610 {
606 611 return (!refcount_is_zero(&ds->ds_longholds));
607 612 }
608 613
609 614 void
610 615 dsl_dataset_name(dsl_dataset_t *ds, char *name)
611 616 {
612 617 if (ds == NULL) {
613 618 (void) strcpy(name, "mos");
614 619 } else {
615 620 dsl_dir_name(ds->ds_dir, name);
616 621 VERIFY0(dsl_dataset_get_snapname(ds));
617 622 if (ds->ds_snapname[0]) {
618 623 (void) strcat(name, "@");
619 624 /*
620 625 * We use a "recursive" mutex so that we
621 626 * can call dprintf_ds() with ds_lock held.
622 627 */
623 628 if (!MUTEX_HELD(&ds->ds_lock)) {
624 629 mutex_enter(&ds->ds_lock);
625 630 (void) strcat(name, ds->ds_snapname);
626 631 mutex_exit(&ds->ds_lock);
627 632 } else {
628 633 (void) strcat(name, ds->ds_snapname);
629 634 }
630 635 }
631 636 }
632 637 }
633 638
634 639 void
635 640 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
636 641 {
637 642 dmu_buf_rele(ds->ds_dbuf, tag);
638 643 }
639 644
640 645 void
641 646 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
642 647 {
643 648 ASSERT3P(ds->ds_owner, ==, tag);
644 649 ASSERT(ds->ds_dbuf != NULL);
645 650
646 651 mutex_enter(&ds->ds_lock);
647 652 ds->ds_owner = NULL;
648 653 mutex_exit(&ds->ds_lock);
649 654 dsl_dataset_long_rele(ds, tag);
650 655 dsl_dataset_rele(ds, tag);
651 656 }
652 657
653 658 boolean_t
654 659 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
655 660 {
656 661 boolean_t gotit = FALSE;
657 662
658 663 mutex_enter(&ds->ds_lock);
659 664 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
660 665 ds->ds_owner = tag;
661 666 dsl_dataset_long_hold(ds, tag);
662 667 gotit = TRUE;
663 668 }
664 669 mutex_exit(&ds->ds_lock);
665 670 return (gotit);
666 671 }
667 672
668 673 uint64_t
669 674 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
670 675 uint64_t flags, dmu_tx_t *tx)
671 676 {
672 677 dsl_pool_t *dp = dd->dd_pool;
673 678 dmu_buf_t *dbuf;
674 679 dsl_dataset_phys_t *dsphys;
675 680 uint64_t dsobj;
676 681 objset_t *mos = dp->dp_meta_objset;
677 682
678 683 if (origin == NULL)
679 684 origin = dp->dp_origin_snap;
680 685
681 686 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
682 687 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
683 688 ASSERT(dmu_tx_is_syncing(tx));
684 689 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
685 690
686 691 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
687 692 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
688 693 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
689 694 dmu_buf_will_dirty(dbuf, tx);
690 695 dsphys = dbuf->db_data;
691 696 bzero(dsphys, sizeof (dsl_dataset_phys_t));
692 697 dsphys->ds_dir_obj = dd->dd_object;
693 698 dsphys->ds_flags = flags;
694 699 dsphys->ds_fsid_guid = unique_create();
695 700 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
696 701 sizeof (dsphys->ds_guid));
697 702 dsphys->ds_snapnames_zapobj =
698 703 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
699 704 DMU_OT_NONE, 0, tx);
700 705 dsphys->ds_creation_time = gethrestime_sec();
701 706 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
702 707
703 708 if (origin == NULL) {
704 709 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
705 710 } else {
706 711 dsl_dataset_t *ohds; /* head of the origin snapshot */
707 712
708 713 dsphys->ds_prev_snap_obj = origin->ds_object;
709 714 dsphys->ds_prev_snap_txg =
710 715 dsl_dataset_phys(origin)->ds_creation_txg;
711 716 dsphys->ds_referenced_bytes =
712 717 dsl_dataset_phys(origin)->ds_referenced_bytes;
713 718 dsphys->ds_compressed_bytes =
714 719 dsl_dataset_phys(origin)->ds_compressed_bytes;
715 720 dsphys->ds_uncompressed_bytes =
716 721 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
717 722 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
718 723
719 724 /*
720 725 * Inherit flags that describe the dataset's contents
721 726 * (INCONSISTENT) or properties (Case Insensitive).
722 727 */
723 728 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
724 729 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
725 730
726 731 if (origin->ds_large_blocks)
727 732 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
728 733
729 734 dmu_buf_will_dirty(origin->ds_dbuf, tx);
730 735 dsl_dataset_phys(origin)->ds_num_children++;
731 736
732 737 VERIFY0(dsl_dataset_hold_obj(dp,
733 738 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
734 739 FTAG, &ohds));
735 740 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
736 741 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
737 742 dsl_dataset_rele(ohds, FTAG);
738 743
739 744 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
740 745 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
741 746 dsl_dataset_phys(origin)->ds_next_clones_obj =
742 747 zap_create(mos,
743 748 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
744 749 }
745 750 VERIFY0(zap_add_int(mos,
746 751 dsl_dataset_phys(origin)->ds_next_clones_obj,
747 752 dsobj, tx));
748 753 }
749 754
750 755 dmu_buf_will_dirty(dd->dd_dbuf, tx);
751 756 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
752 757 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
753 758 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
754 759 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
755 760 dsl_dir_phys(origin->ds_dir)->dd_clones =
756 761 zap_create(mos,
757 762 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
758 763 }
759 764 VERIFY0(zap_add_int(mos,
760 765 dsl_dir_phys(origin->ds_dir)->dd_clones,
761 766 dsobj, tx));
762 767 }
763 768 }
764 769
765 770 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
766 771 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
767 772
768 773 dmu_buf_rele(dbuf, FTAG);
769 774
770 775 dmu_buf_will_dirty(dd->dd_dbuf, tx);
771 776 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
772 777
773 778 return (dsobj);
774 779 }
775 780
776 781 static void
777 782 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
778 783 {
779 784 objset_t *os;
780 785
781 786 VERIFY0(dmu_objset_from_ds(ds, &os));
782 787 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
783 788 dsl_dataset_dirty(ds, tx);
784 789 }
785 790
786 791 uint64_t
787 792 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
788 793 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
789 794 {
790 795 dsl_pool_t *dp = pdd->dd_pool;
791 796 uint64_t dsobj, ddobj;
792 797 dsl_dir_t *dd;
793 798
794 799 ASSERT(dmu_tx_is_syncing(tx));
795 800 ASSERT(lastname[0] != '@');
796 801
797 802 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
798 803 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
799 804
800 805 dsobj = dsl_dataset_create_sync_dd(dd, origin,
801 806 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
802 807
803 808 dsl_deleg_set_create_perms(dd, tx, cr);
804 809
805 810 /*
806 811 * Since we're creating a new node we know it's a leaf, so we can
807 812 * initialize the counts if the limit feature is active.
808 813 */
809 814 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
810 815 uint64_t cnt = 0;
811 816 objset_t *os = dd->dd_pool->dp_meta_objset;
812 817
813 818 dsl_dir_zapify(dd, tx);
814 819 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
815 820 sizeof (cnt), 1, &cnt, tx));
816 821 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
817 822 sizeof (cnt), 1, &cnt, tx));
818 823 }
819 824
820 825 dsl_dir_rele(dd, FTAG);
821 826
822 827 /*
823 828 * If we are creating a clone, make sure we zero out any stale
824 829 * data from the origin snapshots zil header.
825 830 */
826 831 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
827 832 dsl_dataset_t *ds;
828 833
829 834 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
830 835 dsl_dataset_zero_zil(ds, tx);
831 836 dsl_dataset_rele(ds, FTAG);
832 837 }
833 838
834 839 return (dsobj);
835 840 }
836 841
837 842 /*
838 843 * The unique space in the head dataset can be calculated by subtracting
839 844 * the space used in the most recent snapshot, that is still being used
840 845 * in this file system, from the space currently in use. To figure out
|
↓ open down ↓ |
359 lines elided |
↑ open up ↑ |
841 846 * the space in the most recent snapshot still in use, we need to take
842 847 * the total space used in the snapshot and subtract out the space that
843 848 * has been freed up since the snapshot was taken.
844 849 */
845 850 void
846 851 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
847 852 {
848 853 uint64_t mrs_used;
849 854 uint64_t dlused, dlcomp, dluncomp;
850 855
851 - ASSERT(!dsl_dataset_is_snapshot(ds));
856 + ASSERT(!ds->ds_is_snapshot);
852 857
853 858 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
854 859 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
855 860 else
856 861 mrs_used = 0;
857 862
858 863 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
859 864
860 865 ASSERT3U(dlused, <=, mrs_used);
861 866 dsl_dataset_phys(ds)->ds_unique_bytes =
862 867 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
863 868
864 869 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
865 870 SPA_VERSION_UNIQUE_ACCURATE)
866 871 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867 872 }
868 873
869 874 void
870 875 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
871 876 dmu_tx_t *tx)
872 877 {
873 878 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
874 879 uint64_t count;
875 880 int err;
876 881
877 882 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
878 883 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
879 884 obj, tx);
880 885 /*
881 886 * The err should not be ENOENT, but a bug in a previous version
882 887 * of the code could cause upgrade_clones_cb() to not set
883 888 * ds_next_snap_obj when it should, leading to a missing entry.
884 889 * If we knew that the pool was created after
885 890 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
886 891 * ENOENT. However, at least we can check that we don't have
887 892 * too many entries in the next_clones_obj even after failing to
888 893 * remove this one.
889 894 */
890 895 if (err != ENOENT)
891 896 VERIFY0(err);
892 897 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
893 898 &count));
894 899 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
895 900 }
896 901
897 902
898 903 blkptr_t *
899 904 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
900 905 {
901 906 return (&dsl_dataset_phys(ds)->ds_bp);
902 907 }
903 908
904 909 void
905 910 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
906 911 {
907 912 ASSERT(dmu_tx_is_syncing(tx));
908 913 /* If it's the meta-objset, set dp_meta_rootbp */
909 914 if (ds == NULL) {
910 915 tx->tx_pool->dp_meta_rootbp = *bp;
911 916 } else {
912 917 dmu_buf_will_dirty(ds->ds_dbuf, tx);
913 918 dsl_dataset_phys(ds)->ds_bp = *bp;
914 919 }
915 920 }
916 921
917 922 spa_t *
918 923 dsl_dataset_get_spa(dsl_dataset_t *ds)
919 924 {
920 925 return (ds->ds_dir->dd_pool->dp_spa);
921 926 }
922 927
923 928 void
924 929 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
925 930 {
926 931 dsl_pool_t *dp;
927 932
928 933 if (ds == NULL) /* this is the meta-objset */
929 934 return;
930 935
931 936 ASSERT(ds->ds_objset != NULL);
932 937
933 938 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
934 939 panic("dirtying snapshot!");
935 940
936 941 dp = ds->ds_dir->dd_pool;
937 942
938 943 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
939 944 /* up the hold count until we can be written out */
940 945 dmu_buf_add_ref(ds->ds_dbuf, ds);
941 946 }
942 947 }
943 948
944 949 boolean_t
945 950 dsl_dataset_is_dirty(dsl_dataset_t *ds)
946 951 {
947 952 for (int t = 0; t < TXG_SIZE; t++) {
948 953 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
949 954 ds, t))
950 955 return (B_TRUE);
951 956 }
952 957 return (B_FALSE);
953 958 }
954 959
955 960 static int
956 961 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
957 962 {
958 963 uint64_t asize;
959 964
960 965 if (!dmu_tx_is_syncing(tx))
961 966 return (0);
962 967
963 968 /*
964 969 * If there's an fs-only reservation, any blocks that might become
965 970 * owned by the snapshot dataset must be accommodated by space
966 971 * outside of the reservation.
967 972 */
968 973 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
969 974 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
970 975 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
971 976 return (SET_ERROR(ENOSPC));
972 977
973 978 /*
974 979 * Propagate any reserved space for this snapshot to other
975 980 * snapshot checks in this sync group.
976 981 */
977 982 if (asize > 0)
978 983 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
979 984
980 985 return (0);
981 986 }
982 987
983 988 typedef struct dsl_dataset_snapshot_arg {
984 989 nvlist_t *ddsa_snaps;
985 990 nvlist_t *ddsa_props;
986 991 nvlist_t *ddsa_errors;
987 992 cred_t *ddsa_cr;
988 993 } dsl_dataset_snapshot_arg_t;
989 994
990 995 int
991 996 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
992 997 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
993 998 {
994 999 int error;
995 1000 uint64_t value;
996 1001
997 1002 ds->ds_trysnap_txg = tx->tx_txg;
998 1003
999 1004 if (!dmu_tx_is_syncing(tx))
1000 1005 return (0);
1001 1006
1002 1007 /*
1003 1008 * We don't allow multiple snapshots of the same txg. If there
1004 1009 * is already one, try again.
1005 1010 */
1006 1011 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1007 1012 return (SET_ERROR(EAGAIN));
1008 1013
1009 1014 /*
1010 1015 * Check for conflicting snapshot name.
1011 1016 */
1012 1017 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1013 1018 if (error == 0)
1014 1019 return (SET_ERROR(EEXIST));
1015 1020 if (error != ENOENT)
1016 1021 return (error);
1017 1022
1018 1023 /*
1019 1024 * We don't allow taking snapshots of inconsistent datasets, such as
1020 1025 * those into which we are currently receiving. However, if we are
1021 1026 * creating this snapshot as part of a receive, this check will be
1022 1027 * executed atomically with respect to the completion of the receive
1023 1028 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1024 1029 * case we ignore this, knowing it will be fixed up for us shortly in
1025 1030 * dmu_recv_end_sync().
1026 1031 */
1027 1032 if (!recv && DS_IS_INCONSISTENT(ds))
1028 1033 return (SET_ERROR(EBUSY));
1029 1034
1030 1035 /*
1031 1036 * Skip the check for temporary snapshots or if we have already checked
1032 1037 * the counts in dsl_dataset_snapshot_check. This means we really only
1033 1038 * check the count here when we're receiving a stream.
1034 1039 */
1035 1040 if (cnt != 0 && cr != NULL) {
1036 1041 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1037 1042 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1038 1043 if (error != 0)
1039 1044 return (error);
1040 1045 }
1041 1046
1042 1047 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1043 1048 if (error != 0)
1044 1049 return (error);
1045 1050
1046 1051 return (0);
1047 1052 }
1048 1053
1049 1054 static int
1050 1055 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1051 1056 {
1052 1057 dsl_dataset_snapshot_arg_t *ddsa = arg;
1053 1058 dsl_pool_t *dp = dmu_tx_pool(tx);
1054 1059 nvpair_t *pair;
1055 1060 int rv = 0;
1056 1061
1057 1062 /*
1058 1063 * Pre-compute how many total new snapshots will be created for each
1059 1064 * level in the tree and below. This is needed for validating the
1060 1065 * snapshot limit when either taking a recursive snapshot or when
1061 1066 * taking multiple snapshots.
1062 1067 *
1063 1068 * The problem is that the counts are not actually adjusted when
1064 1069 * we are checking, only when we finally sync. For a single snapshot,
1065 1070 * this is easy, the count will increase by 1 at each node up the tree,
1066 1071 * but its more complicated for the recursive/multiple snapshot case.
1067 1072 *
1068 1073 * The dsl_fs_ss_limit_check function does recursively check the count
1069 1074 * at each level up the tree but since it is validating each snapshot
1070 1075 * independently we need to be sure that we are validating the complete
1071 1076 * count for the entire set of snapshots. We do this by rolling up the
1072 1077 * counts for each component of the name into an nvlist and then
1073 1078 * checking each of those cases with the aggregated count.
1074 1079 *
1075 1080 * This approach properly handles not only the recursive snapshot
1076 1081 * case (where we get all of those on the ddsa_snaps list) but also
1077 1082 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1078 1083 * validate the limit on 'a' using a count of 2).
1079 1084 *
1080 1085 * We validate the snapshot names in the third loop and only report
1081 1086 * name errors once.
1082 1087 */
1083 1088 if (dmu_tx_is_syncing(tx)) {
1084 1089 nvlist_t *cnt_track = NULL;
1085 1090 cnt_track = fnvlist_alloc();
1086 1091
1087 1092 /* Rollup aggregated counts into the cnt_track list */
1088 1093 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1089 1094 pair != NULL;
1090 1095 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1091 1096 char *pdelim;
1092 1097 uint64_t val;
1093 1098 char nm[MAXPATHLEN];
1094 1099
1095 1100 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1096 1101 pdelim = strchr(nm, '@');
1097 1102 if (pdelim == NULL)
1098 1103 continue;
1099 1104 *pdelim = '\0';
1100 1105
1101 1106 do {
1102 1107 if (nvlist_lookup_uint64(cnt_track, nm,
1103 1108 &val) == 0) {
1104 1109 /* update existing entry */
1105 1110 fnvlist_add_uint64(cnt_track, nm,
1106 1111 val + 1);
1107 1112 } else {
1108 1113 /* add to list */
1109 1114 fnvlist_add_uint64(cnt_track, nm, 1);
1110 1115 }
1111 1116
1112 1117 pdelim = strrchr(nm, '/');
1113 1118 if (pdelim != NULL)
1114 1119 *pdelim = '\0';
1115 1120 } while (pdelim != NULL);
1116 1121 }
1117 1122
1118 1123 /* Check aggregated counts at each level */
1119 1124 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1120 1125 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1121 1126 int error = 0;
1122 1127 char *name;
1123 1128 uint64_t cnt = 0;
1124 1129 dsl_dataset_t *ds;
1125 1130
1126 1131 name = nvpair_name(pair);
1127 1132 cnt = fnvpair_value_uint64(pair);
1128 1133 ASSERT(cnt > 0);
1129 1134
1130 1135 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1131 1136 if (error == 0) {
1132 1137 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1133 1138 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1134 1139 ddsa->ddsa_cr);
1135 1140 dsl_dataset_rele(ds, FTAG);
1136 1141 }
1137 1142
1138 1143 if (error != 0) {
1139 1144 if (ddsa->ddsa_errors != NULL)
1140 1145 fnvlist_add_int32(ddsa->ddsa_errors,
1141 1146 name, error);
1142 1147 rv = error;
1143 1148 /* only report one error for this check */
1144 1149 break;
1145 1150 }
1146 1151 }
1147 1152 nvlist_free(cnt_track);
1148 1153 }
1149 1154
1150 1155 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1151 1156 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1152 1157 int error = 0;
1153 1158 dsl_dataset_t *ds;
1154 1159 char *name, *atp;
1155 1160 char dsname[MAXNAMELEN];
1156 1161
1157 1162 name = nvpair_name(pair);
1158 1163 if (strlen(name) >= MAXNAMELEN)
1159 1164 error = SET_ERROR(ENAMETOOLONG);
1160 1165 if (error == 0) {
1161 1166 atp = strchr(name, '@');
1162 1167 if (atp == NULL)
1163 1168 error = SET_ERROR(EINVAL);
1164 1169 if (error == 0)
1165 1170 (void) strlcpy(dsname, name, atp - name + 1);
1166 1171 }
1167 1172 if (error == 0)
1168 1173 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1169 1174 if (error == 0) {
1170 1175 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1171 1176 error = dsl_dataset_snapshot_check_impl(ds,
1172 1177 atp + 1, tx, B_FALSE, 0, NULL);
1173 1178 dsl_dataset_rele(ds, FTAG);
1174 1179 }
1175 1180
1176 1181 if (error != 0) {
1177 1182 if (ddsa->ddsa_errors != NULL) {
1178 1183 fnvlist_add_int32(ddsa->ddsa_errors,
1179 1184 name, error);
1180 1185 }
1181 1186 rv = error;
1182 1187 }
1183 1188 }
1184 1189
1185 1190 return (rv);
1186 1191 }
1187 1192
1188 1193 void
1189 1194 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1190 1195 dmu_tx_t *tx)
1191 1196 {
1192 1197 static zil_header_t zero_zil;
1193 1198
1194 1199 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1195 1200 dmu_buf_t *dbuf;
1196 1201 dsl_dataset_phys_t *dsphys;
1197 1202 uint64_t dsobj, crtxg;
1198 1203 objset_t *mos = dp->dp_meta_objset;
1199 1204 objset_t *os;
1200 1205
1201 1206 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1202 1207
1203 1208 /*
1204 1209 * If we are on an old pool, the zil must not be active, in which
1205 1210 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1206 1211 */
1207 1212 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1208 1213 dmu_objset_from_ds(ds, &os) != 0 ||
1209 1214 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1210 1215 sizeof (zero_zil)) == 0);
1211 1216
1212 1217 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1213 1218
1214 1219 /*
1215 1220 * The origin's ds_creation_txg has to be < TXG_INITIAL
1216 1221 */
1217 1222 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1218 1223 crtxg = 1;
1219 1224 else
1220 1225 crtxg = tx->tx_txg;
1221 1226
1222 1227 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1223 1228 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1224 1229 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1225 1230 dmu_buf_will_dirty(dbuf, tx);
1226 1231 dsphys = dbuf->db_data;
1227 1232 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1228 1233 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1229 1234 dsphys->ds_fsid_guid = unique_create();
1230 1235 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1231 1236 sizeof (dsphys->ds_guid));
1232 1237 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1233 1238 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1234 1239 dsphys->ds_next_snap_obj = ds->ds_object;
1235 1240 dsphys->ds_num_children = 1;
1236 1241 dsphys->ds_creation_time = gethrestime_sec();
1237 1242 dsphys->ds_creation_txg = crtxg;
1238 1243 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1239 1244 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1240 1245 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1241 1246 dsphys->ds_uncompressed_bytes =
1242 1247 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1243 1248 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1244 1249 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1245 1250 dmu_buf_rele(dbuf, FTAG);
1246 1251
1247 1252 if (ds->ds_large_blocks)
1248 1253 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1249 1254
1250 1255 ASSERT3U(ds->ds_prev != 0, ==,
1251 1256 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1252 1257 if (ds->ds_prev) {
1253 1258 uint64_t next_clones_obj =
1254 1259 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1255 1260 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1256 1261 ds->ds_object ||
1257 1262 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1258 1263 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1259 1264 ds->ds_object) {
1260 1265 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1261 1266 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1262 1267 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1263 1268 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1264 1269 } else if (next_clones_obj != 0) {
1265 1270 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1266 1271 dsphys->ds_next_snap_obj, tx);
1267 1272 VERIFY0(zap_add_int(mos,
1268 1273 next_clones_obj, dsobj, tx));
1269 1274 }
1270 1275 }
1271 1276
1272 1277 /*
1273 1278 * If we have a reference-reservation on this dataset, we will
1274 1279 * need to increase the amount of refreservation being charged
1275 1280 * since our unique space is going to zero.
1276 1281 */
1277 1282 if (ds->ds_reserved) {
1278 1283 int64_t delta;
1279 1284 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1280 1285 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1281 1286 ds->ds_reserved);
1282 1287 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1283 1288 delta, 0, 0, tx);
1284 1289 }
1285 1290
1286 1291 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1287 1292 dsl_dataset_phys(ds)->ds_deadlist_obj =
1288 1293 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1289 1294 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1290 1295 dsl_deadlist_close(&ds->ds_deadlist);
1291 1296 dsl_deadlist_open(&ds->ds_deadlist, mos,
1292 1297 dsl_dataset_phys(ds)->ds_deadlist_obj);
1293 1298 dsl_deadlist_add_key(&ds->ds_deadlist,
1294 1299 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1295 1300
1296 1301 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1297 1302 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1298 1303 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1299 1304 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1300 1305 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1301 1306 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1302 1307
1303 1308 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1304 1309 snapname, 8, 1, &dsobj, tx));
1305 1310
1306 1311 if (ds->ds_prev)
1307 1312 dsl_dataset_rele(ds->ds_prev, ds);
1308 1313 VERIFY0(dsl_dataset_hold_obj(dp,
1309 1314 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1310 1315
1311 1316 dsl_scan_ds_snapshotted(ds, tx);
1312 1317
1313 1318 dsl_dir_snap_cmtime_update(ds->ds_dir);
1314 1319
1315 1320 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1316 1321 }
1317 1322
1318 1323 static void
1319 1324 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1320 1325 {
1321 1326 dsl_dataset_snapshot_arg_t *ddsa = arg;
1322 1327 dsl_pool_t *dp = dmu_tx_pool(tx);
1323 1328 nvpair_t *pair;
1324 1329
1325 1330 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1326 1331 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1327 1332 dsl_dataset_t *ds;
1328 1333 char *name, *atp;
1329 1334 char dsname[MAXNAMELEN];
1330 1335
1331 1336 name = nvpair_name(pair);
1332 1337 atp = strchr(name, '@');
1333 1338 (void) strlcpy(dsname, name, atp - name + 1);
1334 1339 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1335 1340
1336 1341 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1337 1342 if (ddsa->ddsa_props != NULL) {
1338 1343 dsl_props_set_sync_impl(ds->ds_prev,
1339 1344 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1340 1345 }
1341 1346 dsl_dataset_rele(ds, FTAG);
1342 1347 }
1343 1348 }
1344 1349
1345 1350 /*
1346 1351 * The snapshots must all be in the same pool.
1347 1352 * All-or-nothing: if there are any failures, nothing will be modified.
1348 1353 */
1349 1354 int
1350 1355 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1351 1356 {
1352 1357 dsl_dataset_snapshot_arg_t ddsa;
1353 1358 nvpair_t *pair;
1354 1359 boolean_t needsuspend;
1355 1360 int error;
1356 1361 spa_t *spa;
1357 1362 char *firstname;
1358 1363 nvlist_t *suspended = NULL;
1359 1364
1360 1365 pair = nvlist_next_nvpair(snaps, NULL);
1361 1366 if (pair == NULL)
1362 1367 return (0);
1363 1368 firstname = nvpair_name(pair);
1364 1369
1365 1370 error = spa_open(firstname, &spa, FTAG);
1366 1371 if (error != 0)
1367 1372 return (error);
1368 1373 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1369 1374 spa_close(spa, FTAG);
1370 1375
1371 1376 if (needsuspend) {
1372 1377 suspended = fnvlist_alloc();
1373 1378 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1374 1379 pair = nvlist_next_nvpair(snaps, pair)) {
1375 1380 char fsname[MAXNAMELEN];
1376 1381 char *snapname = nvpair_name(pair);
1377 1382 char *atp;
1378 1383 void *cookie;
1379 1384
1380 1385 atp = strchr(snapname, '@');
1381 1386 if (atp == NULL) {
1382 1387 error = SET_ERROR(EINVAL);
1383 1388 break;
1384 1389 }
1385 1390 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1386 1391
1387 1392 error = zil_suspend(fsname, &cookie);
1388 1393 if (error != 0)
1389 1394 break;
1390 1395 fnvlist_add_uint64(suspended, fsname,
1391 1396 (uintptr_t)cookie);
1392 1397 }
1393 1398 }
1394 1399
1395 1400 ddsa.ddsa_snaps = snaps;
1396 1401 ddsa.ddsa_props = props;
1397 1402 ddsa.ddsa_errors = errors;
1398 1403 ddsa.ddsa_cr = CRED();
1399 1404
1400 1405 if (error == 0) {
1401 1406 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1402 1407 dsl_dataset_snapshot_sync, &ddsa,
1403 1408 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1404 1409 }
1405 1410
1406 1411 if (suspended != NULL) {
1407 1412 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1408 1413 pair = nvlist_next_nvpair(suspended, pair)) {
1409 1414 zil_resume((void *)(uintptr_t)
1410 1415 fnvpair_value_uint64(pair));
1411 1416 }
1412 1417 fnvlist_free(suspended);
1413 1418 }
1414 1419
1415 1420 return (error);
1416 1421 }
1417 1422
1418 1423 typedef struct dsl_dataset_snapshot_tmp_arg {
1419 1424 const char *ddsta_fsname;
1420 1425 const char *ddsta_snapname;
1421 1426 minor_t ddsta_cleanup_minor;
1422 1427 const char *ddsta_htag;
1423 1428 } dsl_dataset_snapshot_tmp_arg_t;
1424 1429
1425 1430 static int
1426 1431 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1427 1432 {
1428 1433 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1429 1434 dsl_pool_t *dp = dmu_tx_pool(tx);
1430 1435 dsl_dataset_t *ds;
1431 1436 int error;
1432 1437
1433 1438 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1434 1439 if (error != 0)
1435 1440 return (error);
1436 1441
1437 1442 /* NULL cred means no limit check for tmp snapshot */
1438 1443 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1439 1444 tx, B_FALSE, 0, NULL);
1440 1445 if (error != 0) {
1441 1446 dsl_dataset_rele(ds, FTAG);
1442 1447 return (error);
1443 1448 }
1444 1449
1445 1450 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1446 1451 dsl_dataset_rele(ds, FTAG);
1447 1452 return (SET_ERROR(ENOTSUP));
1448 1453 }
1449 1454 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1450 1455 B_TRUE, tx);
1451 1456 if (error != 0) {
1452 1457 dsl_dataset_rele(ds, FTAG);
1453 1458 return (error);
1454 1459 }
1455 1460
1456 1461 dsl_dataset_rele(ds, FTAG);
1457 1462 return (0);
1458 1463 }
1459 1464
1460 1465 static void
1461 1466 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1462 1467 {
1463 1468 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1464 1469 dsl_pool_t *dp = dmu_tx_pool(tx);
1465 1470 dsl_dataset_t *ds;
1466 1471
1467 1472 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1468 1473
1469 1474 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1470 1475 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1471 1476 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1472 1477 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1473 1478
1474 1479 dsl_dataset_rele(ds, FTAG);
1475 1480 }
1476 1481
1477 1482 int
1478 1483 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1479 1484 minor_t cleanup_minor, const char *htag)
1480 1485 {
1481 1486 dsl_dataset_snapshot_tmp_arg_t ddsta;
1482 1487 int error;
1483 1488 spa_t *spa;
1484 1489 boolean_t needsuspend;
1485 1490 void *cookie;
1486 1491
1487 1492 ddsta.ddsta_fsname = fsname;
1488 1493 ddsta.ddsta_snapname = snapname;
1489 1494 ddsta.ddsta_cleanup_minor = cleanup_minor;
1490 1495 ddsta.ddsta_htag = htag;
1491 1496
1492 1497 error = spa_open(fsname, &spa, FTAG);
1493 1498 if (error != 0)
1494 1499 return (error);
1495 1500 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1496 1501 spa_close(spa, FTAG);
1497 1502
1498 1503 if (needsuspend) {
1499 1504 error = zil_suspend(fsname, &cookie);
1500 1505 if (error != 0)
1501 1506 return (error);
1502 1507 }
1503 1508
1504 1509 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1505 1510 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1506 1511
1507 1512 if (needsuspend)
1508 1513 zil_resume(cookie);
1509 1514 return (error);
1510 1515 }
1511 1516
1512 1517
1513 1518 void
1514 1519 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1515 1520 {
1516 1521 ASSERT(dmu_tx_is_syncing(tx));
1517 1522 ASSERT(ds->ds_objset != NULL);
1518 1523 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1519 1524
1520 1525 /*
1521 1526 * in case we had to change ds_fsid_guid when we opened it,
1522 1527 * sync it out now.
1523 1528 */
1524 1529 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1525 1530 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1526 1531
1527 1532 dmu_objset_sync(ds->ds_objset, zio, tx);
1528 1533
1529 1534 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1530 1535 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1531 1536 ds->ds_large_blocks = B_TRUE;
1532 1537 }
1533 1538 }
1534 1539
1535 1540 static void
1536 1541 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1537 1542 {
1538 1543 uint64_t count = 0;
1539 1544 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1540 1545 zap_cursor_t zc;
1541 1546 zap_attribute_t za;
1542 1547 nvlist_t *propval = fnvlist_alloc();
1543 1548 nvlist_t *val = fnvlist_alloc();
1544 1549
1545 1550 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1546 1551
1547 1552 /*
1548 1553 * There may be missing entries in ds_next_clones_obj
1549 1554 * due to a bug in a previous version of the code.
1550 1555 * Only trust it if it has the right number of entries.
1551 1556 */
1552 1557 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1553 1558 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1554 1559 &count));
1555 1560 }
1556 1561 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1557 1562 goto fail;
1558 1563 for (zap_cursor_init(&zc, mos,
1559 1564 dsl_dataset_phys(ds)->ds_next_clones_obj);
1560 1565 zap_cursor_retrieve(&zc, &za) == 0;
1561 1566 zap_cursor_advance(&zc)) {
1562 1567 dsl_dataset_t *clone;
1563 1568 char buf[ZFS_MAXNAMELEN];
1564 1569 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1565 1570 za.za_first_integer, FTAG, &clone));
1566 1571 dsl_dir_name(clone->ds_dir, buf);
1567 1572 fnvlist_add_boolean(val, buf);
1568 1573 dsl_dataset_rele(clone, FTAG);
1569 1574 }
1570 1575 zap_cursor_fini(&zc);
1571 1576 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1572 1577 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1573 1578 fail:
1574 1579 nvlist_free(val);
1575 1580 nvlist_free(propval);
1576 1581 }
1577 1582
1578 1583 void
1579 1584 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1580 1585 {
1581 1586 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1582 1587 uint64_t refd, avail, uobjs, aobjs, ratio;
1583 1588
|
↓ open down ↓ |
722 lines elided |
↑ open up ↑ |
1584 1589 ASSERT(dsl_pool_config_held(dp));
1585 1590
1586 1591 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1587 1592 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1588 1593 dsl_dataset_phys(ds)->ds_compressed_bytes);
1589 1594
1590 1595 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1591 1596 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1592 1597 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1593 1598
1594 - if (dsl_dataset_is_snapshot(ds)) {
1599 + if (ds->ds_is_snapshot) {
1595 1600 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1596 1601 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1597 1602 dsl_dataset_phys(ds)->ds_unique_bytes);
1598 1603 get_clones_stat(ds, nv);
1599 1604 } else {
1600 1605 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1601 1606 char buf[MAXNAMELEN];
1602 1607 dsl_dataset_name(ds->ds_prev, buf);
1603 1608 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1604 1609 }
1605 1610
1606 1611 dsl_dir_stats(ds->ds_dir, nv);
1607 1612 }
1608 1613
1609 1614 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1610 1615 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1611 1616 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1612 1617
1613 1618 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1614 1619 dsl_dataset_phys(ds)->ds_creation_time);
1615 1620 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1616 1621 dsl_dataset_phys(ds)->ds_creation_txg);
1617 1622 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1618 1623 ds->ds_quota);
1619 1624 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1620 1625 ds->ds_reserved);
1621 1626 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1622 1627 dsl_dataset_phys(ds)->ds_guid);
1623 1628 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1624 1629 dsl_dataset_phys(ds)->ds_unique_bytes);
1625 1630 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1626 1631 ds->ds_object);
1627 1632 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1628 1633 ds->ds_userrefs);
1629 1634 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1630 1635 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1631 1636
1632 1637 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1633 1638 uint64_t written, comp, uncomp;
1634 1639 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1635 1640 dsl_dataset_t *prev;
1636 1641
1637 1642 int err = dsl_dataset_hold_obj(dp,
1638 1643 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1639 1644 if (err == 0) {
1640 1645 err = dsl_dataset_space_written(prev, ds, &written,
1641 1646 &comp, &uncomp);
1642 1647 dsl_dataset_rele(prev, FTAG);
1643 1648 if (err == 0) {
1644 1649 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1645 1650 written);
1646 1651 }
1647 1652 }
1648 1653 }
1649 1654 }
1650 1655
1651 1656 void
|
↓ open down ↓ |
47 lines elided |
↑ open up ↑ |
1652 1657 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1653 1658 {
1654 1659 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1655 1660 ASSERT(dsl_pool_config_held(dp));
1656 1661
1657 1662 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1658 1663 stat->dds_inconsistent =
1659 1664 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1660 1665 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1661 1666 stat->dds_origin[0] = '\0';
1662 - if (dsl_dataset_is_snapshot(ds)) {
1667 + if (ds->ds_is_snapshot) {
1663 1668 stat->dds_is_snapshot = B_TRUE;
1664 1669 stat->dds_num_clones =
1665 1670 dsl_dataset_phys(ds)->ds_num_children - 1;
1666 1671 } else {
1667 1672 stat->dds_is_snapshot = B_FALSE;
1668 1673 stat->dds_num_clones = 0;
1669 1674
1670 1675 if (dsl_dir_is_clone(ds->ds_dir)) {
1671 1676 dsl_dataset_t *ods;
1672 1677
1673 1678 VERIFY0(dsl_dataset_hold_obj(dp,
1674 1679 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1675 1680 FTAG, &ods));
1676 1681 dsl_dataset_name(ods, stat->dds_origin);
1677 1682 dsl_dataset_rele(ods, FTAG);
1678 1683 }
1679 1684 }
1680 1685 }
1681 1686
1682 1687 uint64_t
1683 1688 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1684 1689 {
1685 1690 return (ds->ds_fsid_guid);
1686 1691 }
1687 1692
1688 1693 void
1689 1694 dsl_dataset_space(dsl_dataset_t *ds,
1690 1695 uint64_t *refdbytesp, uint64_t *availbytesp,
1691 1696 uint64_t *usedobjsp, uint64_t *availobjsp)
1692 1697 {
1693 1698 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1694 1699 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1695 1700 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1696 1701 *availbytesp +=
1697 1702 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1698 1703 if (ds->ds_quota != 0) {
1699 1704 /*
1700 1705 * Adjust available bytes according to refquota
1701 1706 */
1702 1707 if (*refdbytesp < ds->ds_quota)
1703 1708 *availbytesp = MIN(*availbytesp,
1704 1709 ds->ds_quota - *refdbytesp);
1705 1710 else
1706 1711 *availbytesp = 0;
1707 1712 }
1708 1713 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1709 1714 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1710 1715 }
1711 1716
1712 1717 boolean_t
1713 1718 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1714 1719 {
1715 1720 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1716 1721
1717 1722 ASSERT(dsl_pool_config_held(dp));
1718 1723 if (snap == NULL)
1719 1724 return (B_FALSE);
1720 1725 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1721 1726 dsl_dataset_phys(snap)->ds_creation_txg) {
1722 1727 objset_t *os, *os_snap;
1723 1728 /*
1724 1729 * It may be that only the ZIL differs, because it was
1725 1730 * reset in the head. Don't count that as being
1726 1731 * modified.
1727 1732 */
1728 1733 if (dmu_objset_from_ds(ds, &os) != 0)
1729 1734 return (B_TRUE);
1730 1735 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1731 1736 return (B_TRUE);
1732 1737 return (bcmp(&os->os_phys->os_meta_dnode,
1733 1738 &os_snap->os_phys->os_meta_dnode,
1734 1739 sizeof (os->os_phys->os_meta_dnode)) != 0);
1735 1740 }
1736 1741 return (B_FALSE);
1737 1742 }
1738 1743
1739 1744 typedef struct dsl_dataset_rename_snapshot_arg {
1740 1745 const char *ddrsa_fsname;
1741 1746 const char *ddrsa_oldsnapname;
1742 1747 const char *ddrsa_newsnapname;
1743 1748 boolean_t ddrsa_recursive;
1744 1749 dmu_tx_t *ddrsa_tx;
1745 1750 } dsl_dataset_rename_snapshot_arg_t;
1746 1751
1747 1752 /* ARGSUSED */
1748 1753 static int
1749 1754 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1750 1755 dsl_dataset_t *hds, void *arg)
1751 1756 {
1752 1757 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1753 1758 int error;
1754 1759 uint64_t val;
1755 1760
1756 1761 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1757 1762 if (error != 0) {
1758 1763 /* ignore nonexistent snapshots */
1759 1764 return (error == ENOENT ? 0 : error);
1760 1765 }
1761 1766
1762 1767 /* new name should not exist */
1763 1768 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1764 1769 if (error == 0)
1765 1770 error = SET_ERROR(EEXIST);
1766 1771 else if (error == ENOENT)
1767 1772 error = 0;
1768 1773
1769 1774 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1770 1775 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1771 1776 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1772 1777 error = SET_ERROR(ENAMETOOLONG);
1773 1778
1774 1779 return (error);
1775 1780 }
1776 1781
1777 1782 static int
1778 1783 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1779 1784 {
1780 1785 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1781 1786 dsl_pool_t *dp = dmu_tx_pool(tx);
1782 1787 dsl_dataset_t *hds;
1783 1788 int error;
1784 1789
1785 1790 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1786 1791 if (error != 0)
1787 1792 return (error);
1788 1793
1789 1794 if (ddrsa->ddrsa_recursive) {
1790 1795 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1791 1796 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1792 1797 DS_FIND_CHILDREN);
1793 1798 } else {
1794 1799 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1795 1800 }
1796 1801 dsl_dataset_rele(hds, FTAG);
1797 1802 return (error);
1798 1803 }
1799 1804
1800 1805 static int
1801 1806 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1802 1807 dsl_dataset_t *hds, void *arg)
1803 1808 {
1804 1809 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1805 1810 dsl_dataset_t *ds;
1806 1811 uint64_t val;
1807 1812 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1808 1813 int error;
1809 1814
1810 1815 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1811 1816 ASSERT(error == 0 || error == ENOENT);
1812 1817 if (error == ENOENT) {
1813 1818 /* ignore nonexistent snapshots */
1814 1819 return (0);
1815 1820 }
1816 1821
1817 1822 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1818 1823
1819 1824 /* log before we change the name */
1820 1825 spa_history_log_internal_ds(ds, "rename", tx,
1821 1826 "-> @%s", ddrsa->ddrsa_newsnapname);
1822 1827
1823 1828 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1824 1829 B_FALSE));
1825 1830 mutex_enter(&ds->ds_lock);
1826 1831 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1827 1832 mutex_exit(&ds->ds_lock);
1828 1833 VERIFY0(zap_add(dp->dp_meta_objset,
1829 1834 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1830 1835 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1831 1836
1832 1837 dsl_dataset_rele(ds, FTAG);
1833 1838 return (0);
1834 1839 }
1835 1840
1836 1841 static void
1837 1842 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1838 1843 {
1839 1844 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1840 1845 dsl_pool_t *dp = dmu_tx_pool(tx);
1841 1846 dsl_dataset_t *hds;
1842 1847
1843 1848 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1844 1849 ddrsa->ddrsa_tx = tx;
1845 1850 if (ddrsa->ddrsa_recursive) {
1846 1851 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1847 1852 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1848 1853 DS_FIND_CHILDREN));
1849 1854 } else {
1850 1855 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1851 1856 }
1852 1857 dsl_dataset_rele(hds, FTAG);
1853 1858 }
1854 1859
1855 1860 int
1856 1861 dsl_dataset_rename_snapshot(const char *fsname,
1857 1862 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1858 1863 {
1859 1864 dsl_dataset_rename_snapshot_arg_t ddrsa;
1860 1865
1861 1866 ddrsa.ddrsa_fsname = fsname;
1862 1867 ddrsa.ddrsa_oldsnapname = oldsnapname;
1863 1868 ddrsa.ddrsa_newsnapname = newsnapname;
1864 1869 ddrsa.ddrsa_recursive = recursive;
1865 1870
1866 1871 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1867 1872 dsl_dataset_rename_snapshot_sync, &ddrsa,
1868 1873 1, ZFS_SPACE_CHECK_RESERVED));
1869 1874 }
1870 1875
1871 1876 /*
1872 1877 * If we're doing an ownership handoff, we need to make sure that there is
1873 1878 * only one long hold on the dataset. We're not allowed to change anything here
1874 1879 * so we don't permanently release the long hold or regular hold here. We want
1875 1880 * to do this only when syncing to avoid the dataset unexpectedly going away
1876 1881 * when we release the long hold.
1877 1882 */
1878 1883 static int
1879 1884 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1880 1885 {
1881 1886 boolean_t held;
1882 1887
1883 1888 if (!dmu_tx_is_syncing(tx))
1884 1889 return (0);
1885 1890
1886 1891 if (owner != NULL) {
1887 1892 VERIFY3P(ds->ds_owner, ==, owner);
1888 1893 dsl_dataset_long_rele(ds, owner);
1889 1894 }
1890 1895
1891 1896 held = dsl_dataset_long_held(ds);
1892 1897
1893 1898 if (owner != NULL)
1894 1899 dsl_dataset_long_hold(ds, owner);
1895 1900
1896 1901 if (held)
1897 1902 return (SET_ERROR(EBUSY));
1898 1903
1899 1904 return (0);
1900 1905 }
1901 1906
1902 1907 typedef struct dsl_dataset_rollback_arg {
1903 1908 const char *ddra_fsname;
1904 1909 void *ddra_owner;
1905 1910 nvlist_t *ddra_result;
1906 1911 } dsl_dataset_rollback_arg_t;
1907 1912
1908 1913 static int
1909 1914 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1910 1915 {
1911 1916 dsl_dataset_rollback_arg_t *ddra = arg;
|
↓ open down ↓ |
239 lines elided |
↑ open up ↑ |
1912 1917 dsl_pool_t *dp = dmu_tx_pool(tx);
1913 1918 dsl_dataset_t *ds;
1914 1919 int64_t unused_refres_delta;
1915 1920 int error;
1916 1921
1917 1922 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
1918 1923 if (error != 0)
1919 1924 return (error);
1920 1925
1921 1926 /* must not be a snapshot */
1922 - if (dsl_dataset_is_snapshot(ds)) {
1927 + if (ds->ds_is_snapshot) {
1923 1928 dsl_dataset_rele(ds, FTAG);
1924 1929 return (SET_ERROR(EINVAL));
1925 1930 }
1926 1931
1927 1932 /* must have a most recent snapshot */
1928 1933 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
1929 1934 dsl_dataset_rele(ds, FTAG);
1930 1935 return (SET_ERROR(EINVAL));
1931 1936 }
1932 1937
1933 1938 /* must not have any bookmarks after the most recent snapshot */
1934 1939 nvlist_t *proprequest = fnvlist_alloc();
1935 1940 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
1936 1941 nvlist_t *bookmarks = fnvlist_alloc();
1937 1942 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
1938 1943 fnvlist_free(proprequest);
1939 1944 if (error != 0)
1940 1945 return (error);
1941 1946 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
1942 1947 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
1943 1948 nvlist_t *valuenv =
1944 1949 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
1945 1950 zfs_prop_to_name(ZFS_PROP_CREATETXG));
1946 1951 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
1947 1952 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
1948 1953 fnvlist_free(bookmarks);
1949 1954 dsl_dataset_rele(ds, FTAG);
1950 1955 return (SET_ERROR(EEXIST));
1951 1956 }
1952 1957 }
1953 1958 fnvlist_free(bookmarks);
1954 1959
1955 1960 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
1956 1961 if (error != 0) {
1957 1962 dsl_dataset_rele(ds, FTAG);
1958 1963 return (error);
1959 1964 }
1960 1965
1961 1966 /*
1962 1967 * Check if the snap we are rolling back to uses more than
1963 1968 * the refquota.
1964 1969 */
1965 1970 if (ds->ds_quota != 0 &&
1966 1971 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
1967 1972 dsl_dataset_rele(ds, FTAG);
1968 1973 return (SET_ERROR(EDQUOT));
1969 1974 }
1970 1975
1971 1976 /*
1972 1977 * When we do the clone swap, we will temporarily use more space
1973 1978 * due to the refreservation (the head will no longer have any
1974 1979 * unique space, so the entire amount of the refreservation will need
1975 1980 * to be free). We will immediately destroy the clone, freeing
1976 1981 * this space, but the freeing happens over many txg's.
1977 1982 */
1978 1983 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1979 1984 dsl_dataset_phys(ds)->ds_unique_bytes);
1980 1985
1981 1986 if (unused_refres_delta > 0 &&
1982 1987 unused_refres_delta >
1983 1988 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1984 1989 dsl_dataset_rele(ds, FTAG);
1985 1990 return (SET_ERROR(ENOSPC));
1986 1991 }
1987 1992
1988 1993 dsl_dataset_rele(ds, FTAG);
1989 1994 return (0);
1990 1995 }
1991 1996
1992 1997 static void
1993 1998 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1994 1999 {
1995 2000 dsl_dataset_rollback_arg_t *ddra = arg;
1996 2001 dsl_pool_t *dp = dmu_tx_pool(tx);
1997 2002 dsl_dataset_t *ds, *clone;
1998 2003 uint64_t cloneobj;
1999 2004 char namebuf[ZFS_MAXNAMELEN];
2000 2005
2001 2006 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2002 2007
2003 2008 dsl_dataset_name(ds->ds_prev, namebuf);
2004 2009 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2005 2010
2006 2011 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2007 2012 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2008 2013
2009 2014 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2010 2015
2011 2016 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2012 2017 dsl_dataset_zero_zil(ds, tx);
2013 2018
2014 2019 dsl_destroy_head_sync_impl(clone, tx);
2015 2020
2016 2021 dsl_dataset_rele(clone, FTAG);
2017 2022 dsl_dataset_rele(ds, FTAG);
2018 2023 }
2019 2024
2020 2025 /*
2021 2026 * Rolls back the given filesystem or volume to the most recent snapshot.
2022 2027 * The name of the most recent snapshot will be returned under key "target"
2023 2028 * in the result nvlist.
2024 2029 *
2025 2030 * If owner != NULL:
2026 2031 * - The existing dataset MUST be owned by the specified owner at entry
2027 2032 * - Upon return, dataset will still be held by the same owner, whether we
2028 2033 * succeed or not.
2029 2034 *
2030 2035 * This mode is required any time the existing filesystem is mounted. See
2031 2036 * notes above zfs_suspend_fs() for further details.
2032 2037 */
2033 2038 int
2034 2039 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2035 2040 {
2036 2041 dsl_dataset_rollback_arg_t ddra;
2037 2042
2038 2043 ddra.ddra_fsname = fsname;
2039 2044 ddra.ddra_owner = owner;
2040 2045 ddra.ddra_result = result;
2041 2046
2042 2047 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2043 2048 dsl_dataset_rollback_sync, &ddra,
2044 2049 1, ZFS_SPACE_CHECK_RESERVED));
2045 2050 }
2046 2051
2047 2052 struct promotenode {
2048 2053 list_node_t link;
2049 2054 dsl_dataset_t *ds;
2050 2055 };
2051 2056
2052 2057 typedef struct dsl_dataset_promote_arg {
2053 2058 const char *ddpa_clonename;
2054 2059 dsl_dataset_t *ddpa_clone;
2055 2060 list_t shared_snaps, origin_snaps, clone_snaps;
2056 2061 dsl_dataset_t *origin_origin; /* origin of the origin */
2057 2062 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2058 2063 char *err_ds;
2059 2064 cred_t *cr;
2060 2065 } dsl_dataset_promote_arg_t;
2061 2066
2062 2067 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2063 2068 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2064 2069 void *tag);
2065 2070 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2066 2071
2067 2072 static int
2068 2073 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2069 2074 {
2070 2075 dsl_dataset_promote_arg_t *ddpa = arg;
2071 2076 dsl_pool_t *dp = dmu_tx_pool(tx);
2072 2077 dsl_dataset_t *hds;
2073 2078 struct promotenode *snap;
2074 2079 dsl_dataset_t *origin_ds;
2075 2080 int err;
2076 2081 uint64_t unused;
2077 2082 uint64_t ss_mv_cnt;
2078 2083
2079 2084 err = promote_hold(ddpa, dp, FTAG);
2080 2085 if (err != 0)
2081 2086 return (err);
2082 2087
2083 2088 hds = ddpa->ddpa_clone;
2084 2089
2085 2090 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2086 2091 promote_rele(ddpa, FTAG);
2087 2092 return (SET_ERROR(EXDEV));
2088 2093 }
2089 2094
2090 2095 /*
2091 2096 * Compute and check the amount of space to transfer. Since this is
2092 2097 * so expensive, don't do the preliminary check.
2093 2098 */
2094 2099 if (!dmu_tx_is_syncing(tx)) {
2095 2100 promote_rele(ddpa, FTAG);
2096 2101 return (0);
2097 2102 }
2098 2103
2099 2104 snap = list_head(&ddpa->shared_snaps);
2100 2105 origin_ds = snap->ds;
2101 2106
2102 2107 /* compute origin's new unique space */
2103 2108 snap = list_tail(&ddpa->clone_snaps);
2104 2109 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2105 2110 origin_ds->ds_object);
2106 2111 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2107 2112 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2108 2113 &ddpa->unique, &unused, &unused);
2109 2114
2110 2115 /*
2111 2116 * Walk the snapshots that we are moving
2112 2117 *
2113 2118 * Compute space to transfer. Consider the incremental changes
2114 2119 * to used by each snapshot:
2115 2120 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2116 2121 * So each snapshot gave birth to:
2117 2122 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2118 2123 * So a sequence would look like:
2119 2124 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2120 2125 * Which simplifies to:
2121 2126 * uN + kN + kN-1 + ... + k1 + k0
2122 2127 * Note however, if we stop before we reach the ORIGIN we get:
2123 2128 * uN + kN + kN-1 + ... + kM - uM-1
2124 2129 */
2125 2130 ss_mv_cnt = 0;
2126 2131 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2127 2132 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2128 2133 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2129 2134 for (snap = list_head(&ddpa->shared_snaps); snap;
2130 2135 snap = list_next(&ddpa->shared_snaps, snap)) {
2131 2136 uint64_t val, dlused, dlcomp, dluncomp;
2132 2137 dsl_dataset_t *ds = snap->ds;
2133 2138
2134 2139 ss_mv_cnt++;
2135 2140
2136 2141 /*
2137 2142 * If there are long holds, we won't be able to evict
2138 2143 * the objset.
2139 2144 */
2140 2145 if (dsl_dataset_long_held(ds)) {
2141 2146 err = SET_ERROR(EBUSY);
2142 2147 goto out;
2143 2148 }
2144 2149
2145 2150 /* Check that the snapshot name does not conflict */
2146 2151 VERIFY0(dsl_dataset_get_snapname(ds));
2147 2152 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2148 2153 if (err == 0) {
2149 2154 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2150 2155 err = SET_ERROR(EEXIST);
2151 2156 goto out;
2152 2157 }
2153 2158 if (err != ENOENT)
2154 2159 goto out;
2155 2160
2156 2161 /* The very first snapshot does not have a deadlist */
2157 2162 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2158 2163 continue;
2159 2164
2160 2165 dsl_deadlist_space(&ds->ds_deadlist,
2161 2166 &dlused, &dlcomp, &dluncomp);
2162 2167 ddpa->used += dlused;
2163 2168 ddpa->comp += dlcomp;
2164 2169 ddpa->uncomp += dluncomp;
2165 2170 }
2166 2171
2167 2172 /*
2168 2173 * If we are a clone of a clone then we never reached ORIGIN,
2169 2174 * so we need to subtract out the clone origin's used space.
2170 2175 */
2171 2176 if (ddpa->origin_origin) {
2172 2177 ddpa->used -=
2173 2178 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2174 2179 ddpa->comp -=
2175 2180 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2176 2181 ddpa->uncomp -=
2177 2182 dsl_dataset_phys(ddpa->origin_origin)->
2178 2183 ds_uncompressed_bytes;
2179 2184 }
2180 2185
2181 2186 /* Check that there is enough space and limit headroom here */
2182 2187 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2183 2188 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2184 2189 if (err != 0)
2185 2190 goto out;
2186 2191
2187 2192 /*
2188 2193 * Compute the amounts of space that will be used by snapshots
2189 2194 * after the promotion (for both origin and clone). For each,
2190 2195 * it is the amount of space that will be on all of their
2191 2196 * deadlists (that was not born before their new origin).
2192 2197 */
2193 2198 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2194 2199 uint64_t space;
2195 2200
2196 2201 /*
2197 2202 * Note, typically this will not be a clone of a clone,
2198 2203 * so dd_origin_txg will be < TXG_INITIAL, so
2199 2204 * these snaplist_space() -> dsl_deadlist_space_range()
2200 2205 * calls will be fast because they do not have to
2201 2206 * iterate over all bps.
2202 2207 */
2203 2208 snap = list_head(&ddpa->origin_snaps);
2204 2209 err = snaplist_space(&ddpa->shared_snaps,
2205 2210 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2206 2211 if (err != 0)
2207 2212 goto out;
2208 2213
2209 2214 err = snaplist_space(&ddpa->clone_snaps,
2210 2215 snap->ds->ds_dir->dd_origin_txg, &space);
2211 2216 if (err != 0)
2212 2217 goto out;
2213 2218 ddpa->cloneusedsnap += space;
2214 2219 }
2215 2220 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2216 2221 DD_FLAG_USED_BREAKDOWN) {
2217 2222 err = snaplist_space(&ddpa->origin_snaps,
2218 2223 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2219 2224 &ddpa->originusedsnap);
2220 2225 if (err != 0)
2221 2226 goto out;
2222 2227 }
2223 2228
2224 2229 out:
2225 2230 promote_rele(ddpa, FTAG);
2226 2231 return (err);
2227 2232 }
2228 2233
2229 2234 static void
2230 2235 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2231 2236 {
2232 2237 dsl_dataset_promote_arg_t *ddpa = arg;
2233 2238 dsl_pool_t *dp = dmu_tx_pool(tx);
2234 2239 dsl_dataset_t *hds;
2235 2240 struct promotenode *snap;
2236 2241 dsl_dataset_t *origin_ds;
2237 2242 dsl_dataset_t *origin_head;
2238 2243 dsl_dir_t *dd;
2239 2244 dsl_dir_t *odd = NULL;
2240 2245 uint64_t oldnext_obj;
2241 2246 int64_t delta;
2242 2247
2243 2248 VERIFY0(promote_hold(ddpa, dp, FTAG));
2244 2249 hds = ddpa->ddpa_clone;
2245 2250
2246 2251 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2247 2252
2248 2253 snap = list_head(&ddpa->shared_snaps);
2249 2254 origin_ds = snap->ds;
2250 2255 dd = hds->ds_dir;
2251 2256
2252 2257 snap = list_head(&ddpa->origin_snaps);
2253 2258 origin_head = snap->ds;
2254 2259
2255 2260 /*
2256 2261 * We need to explicitly open odd, since origin_ds's dd will be
2257 2262 * changing.
2258 2263 */
2259 2264 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2260 2265 NULL, FTAG, &odd));
2261 2266
2262 2267 /* change origin's next snap */
2263 2268 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2264 2269 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2265 2270 snap = list_tail(&ddpa->clone_snaps);
2266 2271 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2267 2272 origin_ds->ds_object);
2268 2273 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2269 2274
2270 2275 /* change the origin's next clone */
2271 2276 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2272 2277 dsl_dataset_remove_from_next_clones(origin_ds,
2273 2278 snap->ds->ds_object, tx);
2274 2279 VERIFY0(zap_add_int(dp->dp_meta_objset,
2275 2280 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2276 2281 oldnext_obj, tx));
2277 2282 }
2278 2283
2279 2284 /* change origin */
2280 2285 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2281 2286 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2282 2287 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2283 2288 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2284 2289 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2285 2290 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2286 2291 origin_head->ds_dir->dd_origin_txg =
2287 2292 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2288 2293
2289 2294 /* change dd_clone entries */
2290 2295 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2291 2296 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2292 2297 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2293 2298 VERIFY0(zap_add_int(dp->dp_meta_objset,
2294 2299 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2295 2300 hds->ds_object, tx));
2296 2301
2297 2302 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2298 2303 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2299 2304 origin_head->ds_object, tx));
2300 2305 if (dsl_dir_phys(dd)->dd_clones == 0) {
2301 2306 dsl_dir_phys(dd)->dd_clones =
2302 2307 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2303 2308 DMU_OT_NONE, 0, tx);
2304 2309 }
2305 2310 VERIFY0(zap_add_int(dp->dp_meta_objset,
2306 2311 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2307 2312 }
2308 2313
2309 2314 /* move snapshots to this dir */
2310 2315 for (snap = list_head(&ddpa->shared_snaps); snap;
2311 2316 snap = list_next(&ddpa->shared_snaps, snap)) {
2312 2317 dsl_dataset_t *ds = snap->ds;
2313 2318
2314 2319 /*
2315 2320 * Property callbacks are registered to a particular
2316 2321 * dsl_dir. Since ours is changing, evict the objset
2317 2322 * so that they will be unregistered from the old dsl_dir.
2318 2323 */
2319 2324 if (ds->ds_objset) {
2320 2325 dmu_objset_evict(ds->ds_objset);
2321 2326 ds->ds_objset = NULL;
2322 2327 }
2323 2328
2324 2329 /* move snap name entry */
2325 2330 VERIFY0(dsl_dataset_get_snapname(ds));
2326 2331 VERIFY0(dsl_dataset_snap_remove(origin_head,
2327 2332 ds->ds_snapname, tx, B_TRUE));
2328 2333 VERIFY0(zap_add(dp->dp_meta_objset,
2329 2334 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2330 2335 8, 1, &ds->ds_object, tx));
2331 2336 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2332 2337 DD_FIELD_SNAPSHOT_COUNT, tx);
2333 2338
2334 2339 /* change containing dsl_dir */
2335 2340 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2336 2341 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2337 2342 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2338 2343 ASSERT3P(ds->ds_dir, ==, odd);
2339 2344 dsl_dir_rele(ds->ds_dir, ds);
2340 2345 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2341 2346 NULL, ds, &ds->ds_dir));
2342 2347
2343 2348 /* move any clone references */
2344 2349 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2345 2350 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2346 2351 zap_cursor_t zc;
2347 2352 zap_attribute_t za;
2348 2353
2349 2354 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2350 2355 dsl_dataset_phys(ds)->ds_next_clones_obj);
2351 2356 zap_cursor_retrieve(&zc, &za) == 0;
2352 2357 zap_cursor_advance(&zc)) {
2353 2358 dsl_dataset_t *cnds;
2354 2359 uint64_t o;
2355 2360
2356 2361 if (za.za_first_integer == oldnext_obj) {
2357 2362 /*
2358 2363 * We've already moved the
2359 2364 * origin's reference.
2360 2365 */
2361 2366 continue;
2362 2367 }
2363 2368
2364 2369 VERIFY0(dsl_dataset_hold_obj(dp,
2365 2370 za.za_first_integer, FTAG, &cnds));
2366 2371 o = dsl_dir_phys(cnds->ds_dir)->
2367 2372 dd_head_dataset_obj;
2368 2373
2369 2374 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2370 2375 dsl_dir_phys(odd)->dd_clones, o, tx));
2371 2376 VERIFY0(zap_add_int(dp->dp_meta_objset,
2372 2377 dsl_dir_phys(dd)->dd_clones, o, tx));
2373 2378 dsl_dataset_rele(cnds, FTAG);
2374 2379 }
2375 2380 zap_cursor_fini(&zc);
2376 2381 }
2377 2382
2378 2383 ASSERT(!dsl_prop_hascb(ds));
2379 2384 }
2380 2385
2381 2386 /*
2382 2387 * Change space accounting.
2383 2388 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2384 2389 * both be valid, or both be 0 (resulting in delta == 0). This
2385 2390 * is true for each of {clone,origin} independently.
2386 2391 */
2387 2392
2388 2393 delta = ddpa->cloneusedsnap -
2389 2394 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2390 2395 ASSERT3S(delta, >=, 0);
2391 2396 ASSERT3U(ddpa->used, >=, delta);
2392 2397 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2393 2398 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2394 2399 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2395 2400
2396 2401 delta = ddpa->originusedsnap -
2397 2402 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2398 2403 ASSERT3S(delta, <=, 0);
2399 2404 ASSERT3U(ddpa->used, >=, -delta);
2400 2405 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2401 2406 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2402 2407 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2403 2408
2404 2409 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2405 2410
2406 2411 /* log history record */
2407 2412 spa_history_log_internal_ds(hds, "promote", tx, "");
2408 2413
2409 2414 dsl_dir_rele(odd, FTAG);
2410 2415 promote_rele(ddpa, FTAG);
2411 2416 }
2412 2417
2413 2418 /*
2414 2419 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2415 2420 * (exclusive) and last_obj (inclusive). The list will be in reverse
2416 2421 * order (last_obj will be the list_head()). If first_obj == 0, do all
2417 2422 * snapshots back to this dataset's origin.
2418 2423 */
2419 2424 static int
2420 2425 snaplist_make(dsl_pool_t *dp,
2421 2426 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2422 2427 {
2423 2428 uint64_t obj = last_obj;
2424 2429
2425 2430 list_create(l, sizeof (struct promotenode),
2426 2431 offsetof(struct promotenode, link));
2427 2432
2428 2433 while (obj != first_obj) {
2429 2434 dsl_dataset_t *ds;
2430 2435 struct promotenode *snap;
2431 2436 int err;
2432 2437
2433 2438 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2434 2439 ASSERT(err != ENOENT);
2435 2440 if (err != 0)
2436 2441 return (err);
2437 2442
2438 2443 if (first_obj == 0)
2439 2444 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2440 2445
2441 2446 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2442 2447 snap->ds = ds;
2443 2448 list_insert_tail(l, snap);
2444 2449 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2445 2450 }
2446 2451
2447 2452 return (0);
2448 2453 }
2449 2454
2450 2455 static int
2451 2456 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2452 2457 {
2453 2458 struct promotenode *snap;
2454 2459
2455 2460 *spacep = 0;
2456 2461 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2457 2462 uint64_t used, comp, uncomp;
2458 2463 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2459 2464 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2460 2465 *spacep += used;
2461 2466 }
2462 2467 return (0);
2463 2468 }
2464 2469
2465 2470 static void
2466 2471 snaplist_destroy(list_t *l, void *tag)
2467 2472 {
2468 2473 struct promotenode *snap;
2469 2474
2470 2475 if (l == NULL || !list_link_active(&l->list_head))
2471 2476 return;
2472 2477
2473 2478 while ((snap = list_tail(l)) != NULL) {
2474 2479 list_remove(l, snap);
2475 2480 dsl_dataset_rele(snap->ds, tag);
2476 2481 kmem_free(snap, sizeof (*snap));
2477 2482 }
2478 2483 list_destroy(l);
2479 2484 }
2480 2485
2481 2486 static int
2482 2487 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2483 2488 {
|
↓ open down ↓ |
551 lines elided |
↑ open up ↑ |
2484 2489 int error;
2485 2490 dsl_dir_t *dd;
2486 2491 struct promotenode *snap;
2487 2492
2488 2493 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2489 2494 &ddpa->ddpa_clone);
2490 2495 if (error != 0)
2491 2496 return (error);
2492 2497 dd = ddpa->ddpa_clone->ds_dir;
2493 2498
2494 - if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2499 + if (ddpa->ddpa_clone->ds_is_snapshot ||
2495 2500 !dsl_dir_is_clone(dd)) {
2496 2501 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2497 2502 return (SET_ERROR(EINVAL));
2498 2503 }
2499 2504
2500 2505 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2501 2506 &ddpa->shared_snaps, tag);
2502 2507 if (error != 0)
2503 2508 goto out;
2504 2509
2505 2510 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2506 2511 &ddpa->clone_snaps, tag);
2507 2512 if (error != 0)
2508 2513 goto out;
2509 2514
2510 2515 snap = list_head(&ddpa->shared_snaps);
2511 2516 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2512 2517 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2513 2518 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2514 2519 &ddpa->origin_snaps, tag);
2515 2520 if (error != 0)
2516 2521 goto out;
2517 2522
2518 2523 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2519 2524 error = dsl_dataset_hold_obj(dp,
2520 2525 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2521 2526 tag, &ddpa->origin_origin);
2522 2527 if (error != 0)
2523 2528 goto out;
2524 2529 }
2525 2530 out:
2526 2531 if (error != 0)
2527 2532 promote_rele(ddpa, tag);
2528 2533 return (error);
2529 2534 }
2530 2535
2531 2536 static void
2532 2537 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2533 2538 {
2534 2539 snaplist_destroy(&ddpa->shared_snaps, tag);
2535 2540 snaplist_destroy(&ddpa->clone_snaps, tag);
2536 2541 snaplist_destroy(&ddpa->origin_snaps, tag);
2537 2542 if (ddpa->origin_origin != NULL)
2538 2543 dsl_dataset_rele(ddpa->origin_origin, tag);
2539 2544 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2540 2545 }
2541 2546
2542 2547 /*
2543 2548 * Promote a clone.
2544 2549 *
2545 2550 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2546 2551 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2547 2552 */
2548 2553 int
2549 2554 dsl_dataset_promote(const char *name, char *conflsnap)
2550 2555 {
2551 2556 dsl_dataset_promote_arg_t ddpa = { 0 };
2552 2557 uint64_t numsnaps;
2553 2558 int error;
2554 2559 objset_t *os;
2555 2560
2556 2561 /*
2557 2562 * We will modify space proportional to the number of
2558 2563 * snapshots. Compute numsnaps.
2559 2564 */
2560 2565 error = dmu_objset_hold(name, FTAG, &os);
2561 2566 if (error != 0)
2562 2567 return (error);
2563 2568 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2564 2569 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2565 2570 &numsnaps);
2566 2571 dmu_objset_rele(os, FTAG);
2567 2572 if (error != 0)
2568 2573 return (error);
2569 2574
2570 2575 ddpa.ddpa_clonename = name;
2571 2576 ddpa.err_ds = conflsnap;
2572 2577 ddpa.cr = CRED();
2573 2578
2574 2579 return (dsl_sync_task(name, dsl_dataset_promote_check,
2575 2580 dsl_dataset_promote_sync, &ddpa,
|
↓ open down ↓ |
71 lines elided |
↑ open up ↑ |
2576 2581 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2577 2582 }
2578 2583
2579 2584 int
2580 2585 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2581 2586 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2582 2587 {
2583 2588 int64_t unused_refres_delta;
2584 2589
2585 2590 /* they should both be heads */
2586 - if (dsl_dataset_is_snapshot(clone) ||
2587 - dsl_dataset_is_snapshot(origin_head))
2591 + if (clone->ds_is_snapshot ||
2592 + origin_head->ds_is_snapshot)
2588 2593 return (SET_ERROR(EINVAL));
2589 2594
2590 2595 /* if we are not forcing, the branch point should be just before them */
2591 2596 if (!force && clone->ds_prev != origin_head->ds_prev)
2592 2597 return (SET_ERROR(EINVAL));
2593 2598
2594 2599 /* clone should be the clone (unless they are unrelated) */
2595 2600 if (clone->ds_prev != NULL &&
2596 2601 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2597 2602 origin_head->ds_dir != clone->ds_prev->ds_dir)
2598 2603 return (SET_ERROR(EINVAL));
2599 2604
2600 2605 /* the clone should be a child of the origin */
2601 2606 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2602 2607 return (SET_ERROR(EINVAL));
2603 2608
2604 2609 /* origin_head shouldn't be modified unless 'force' */
2605 2610 if (!force &&
2606 2611 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2607 2612 return (SET_ERROR(ETXTBSY));
2608 2613
2609 2614 /* origin_head should have no long holds (e.g. is not mounted) */
2610 2615 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2611 2616 return (SET_ERROR(EBUSY));
2612 2617
2613 2618 /* check amount of any unconsumed refreservation */
2614 2619 unused_refres_delta =
2615 2620 (int64_t)MIN(origin_head->ds_reserved,
2616 2621 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2617 2622 (int64_t)MIN(origin_head->ds_reserved,
2618 2623 dsl_dataset_phys(clone)->ds_unique_bytes);
2619 2624
2620 2625 if (unused_refres_delta > 0 &&
2621 2626 unused_refres_delta >
2622 2627 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2623 2628 return (SET_ERROR(ENOSPC));
2624 2629
2625 2630 /* clone can't be over the head's refquota */
2626 2631 if (origin_head->ds_quota != 0 &&
2627 2632 dsl_dataset_phys(clone)->ds_referenced_bytes >
2628 2633 origin_head->ds_quota)
2629 2634 return (SET_ERROR(EDQUOT));
2630 2635
2631 2636 return (0);
2632 2637 }
2633 2638
2634 2639 void
2635 2640 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2636 2641 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2637 2642 {
2638 2643 dsl_pool_t *dp = dmu_tx_pool(tx);
2639 2644 int64_t unused_refres_delta;
2640 2645
2641 2646 ASSERT(clone->ds_reserved == 0);
2642 2647 ASSERT(origin_head->ds_quota == 0 ||
2643 2648 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2644 2649 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2645 2650
2646 2651 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2647 2652 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2648 2653
2649 2654 if (clone->ds_objset != NULL) {
2650 2655 dmu_objset_evict(clone->ds_objset);
2651 2656 clone->ds_objset = NULL;
2652 2657 }
2653 2658
2654 2659 if (origin_head->ds_objset != NULL) {
2655 2660 dmu_objset_evict(origin_head->ds_objset);
2656 2661 origin_head->ds_objset = NULL;
2657 2662 }
2658 2663
2659 2664 unused_refres_delta =
2660 2665 (int64_t)MIN(origin_head->ds_reserved,
2661 2666 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2662 2667 (int64_t)MIN(origin_head->ds_reserved,
2663 2668 dsl_dataset_phys(clone)->ds_unique_bytes);
2664 2669
2665 2670 /*
2666 2671 * Reset origin's unique bytes, if it exists.
2667 2672 */
2668 2673 if (clone->ds_prev) {
2669 2674 dsl_dataset_t *origin = clone->ds_prev;
2670 2675 uint64_t comp, uncomp;
2671 2676
2672 2677 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2673 2678 dsl_deadlist_space_range(&clone->ds_deadlist,
2674 2679 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2675 2680 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2676 2681 }
2677 2682
2678 2683 /* swap blkptrs */
2679 2684 {
2680 2685 blkptr_t tmp;
2681 2686 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2682 2687 dsl_dataset_phys(origin_head)->ds_bp =
2683 2688 dsl_dataset_phys(clone)->ds_bp;
2684 2689 dsl_dataset_phys(clone)->ds_bp = tmp;
2685 2690 }
2686 2691
2687 2692 /* set dd_*_bytes */
2688 2693 {
2689 2694 int64_t dused, dcomp, duncomp;
2690 2695 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2691 2696 uint64_t odl_used, odl_comp, odl_uncomp;
2692 2697
2693 2698 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2694 2699 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2695 2700
2696 2701 dsl_deadlist_space(&clone->ds_deadlist,
2697 2702 &cdl_used, &cdl_comp, &cdl_uncomp);
2698 2703 dsl_deadlist_space(&origin_head->ds_deadlist,
2699 2704 &odl_used, &odl_comp, &odl_uncomp);
2700 2705
2701 2706 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2702 2707 cdl_used -
2703 2708 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2704 2709 odl_used);
2705 2710 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2706 2711 cdl_comp -
2707 2712 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2708 2713 odl_comp);
2709 2714 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2710 2715 cdl_uncomp -
2711 2716 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2712 2717 odl_uncomp);
2713 2718
2714 2719 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2715 2720 dused, dcomp, duncomp, tx);
2716 2721 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2717 2722 -dused, -dcomp, -duncomp, tx);
2718 2723
2719 2724 /*
2720 2725 * The difference in the space used by snapshots is the
2721 2726 * difference in snapshot space due to the head's
2722 2727 * deadlist (since that's the only thing that's
2723 2728 * changing that affects the snapused).
2724 2729 */
2725 2730 dsl_deadlist_space_range(&clone->ds_deadlist,
2726 2731 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2727 2732 &cdl_used, &cdl_comp, &cdl_uncomp);
2728 2733 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2729 2734 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2730 2735 &odl_used, &odl_comp, &odl_uncomp);
2731 2736 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2732 2737 DD_USED_HEAD, DD_USED_SNAP, tx);
2733 2738 }
2734 2739
2735 2740 /* swap ds_*_bytes */
2736 2741 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2737 2742 dsl_dataset_phys(clone)->ds_referenced_bytes);
2738 2743 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2739 2744 dsl_dataset_phys(clone)->ds_compressed_bytes);
2740 2745 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2741 2746 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2742 2747 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2743 2748 dsl_dataset_phys(clone)->ds_unique_bytes);
2744 2749
2745 2750 /* apply any parent delta for change in unconsumed refreservation */
2746 2751 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2747 2752 unused_refres_delta, 0, 0, tx);
2748 2753
2749 2754 /*
2750 2755 * Swap deadlists.
2751 2756 */
2752 2757 dsl_deadlist_close(&clone->ds_deadlist);
2753 2758 dsl_deadlist_close(&origin_head->ds_deadlist);
2754 2759 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2755 2760 dsl_dataset_phys(clone)->ds_deadlist_obj);
2756 2761 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2757 2762 dsl_dataset_phys(clone)->ds_deadlist_obj);
2758 2763 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2759 2764 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2760 2765
2761 2766 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2762 2767
2763 2768 spa_history_log_internal_ds(clone, "clone swap", tx,
2764 2769 "parent=%s", origin_head->ds_dir->dd_myname);
2765 2770 }
2766 2771
2767 2772 /*
2768 2773 * Given a pool name and a dataset object number in that pool,
2769 2774 * return the name of that dataset.
2770 2775 */
2771 2776 int
2772 2777 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2773 2778 {
2774 2779 dsl_pool_t *dp;
2775 2780 dsl_dataset_t *ds;
2776 2781 int error;
2777 2782
2778 2783 error = dsl_pool_hold(pname, FTAG, &dp);
2779 2784 if (error != 0)
2780 2785 return (error);
2781 2786
2782 2787 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2783 2788 if (error == 0) {
2784 2789 dsl_dataset_name(ds, buf);
2785 2790 dsl_dataset_rele(ds, FTAG);
2786 2791 }
2787 2792 dsl_pool_rele(dp, FTAG);
2788 2793
2789 2794 return (error);
2790 2795 }
2791 2796
2792 2797 int
2793 2798 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2794 2799 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2795 2800 {
2796 2801 int error = 0;
2797 2802
2798 2803 ASSERT3S(asize, >, 0);
2799 2804
2800 2805 /*
2801 2806 * *ref_rsrv is the portion of asize that will come from any
2802 2807 * unconsumed refreservation space.
2803 2808 */
2804 2809 *ref_rsrv = 0;
2805 2810
2806 2811 mutex_enter(&ds->ds_lock);
2807 2812 /*
2808 2813 * Make a space adjustment for reserved bytes.
2809 2814 */
2810 2815 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2811 2816 ASSERT3U(*used, >=,
2812 2817 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2813 2818 *used -=
2814 2819 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2815 2820 *ref_rsrv =
2816 2821 asize - MIN(asize, parent_delta(ds, asize + inflight));
2817 2822 }
2818 2823
2819 2824 if (!check_quota || ds->ds_quota == 0) {
2820 2825 mutex_exit(&ds->ds_lock);
2821 2826 return (0);
2822 2827 }
2823 2828 /*
2824 2829 * If they are requesting more space, and our current estimate
2825 2830 * is over quota, they get to try again unless the actual
2826 2831 * on-disk is over quota and there are no pending changes (which
2827 2832 * may free up space for us).
2828 2833 */
2829 2834 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2830 2835 ds->ds_quota) {
2831 2836 if (inflight > 0 ||
2832 2837 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2833 2838 error = SET_ERROR(ERESTART);
2834 2839 else
2835 2840 error = SET_ERROR(EDQUOT);
2836 2841 }
2837 2842 mutex_exit(&ds->ds_lock);
2838 2843
2839 2844 return (error);
2840 2845 }
2841 2846
2842 2847 typedef struct dsl_dataset_set_qr_arg {
2843 2848 const char *ddsqra_name;
2844 2849 zprop_source_t ddsqra_source;
2845 2850 uint64_t ddsqra_value;
2846 2851 } dsl_dataset_set_qr_arg_t;
2847 2852
2848 2853
2849 2854 /* ARGSUSED */
2850 2855 static int
2851 2856 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2852 2857 {
2853 2858 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2854 2859 dsl_pool_t *dp = dmu_tx_pool(tx);
2855 2860 dsl_dataset_t *ds;
|
↓ open down ↓ |
258 lines elided |
↑ open up ↑ |
2856 2861 int error;
2857 2862 uint64_t newval;
2858 2863
2859 2864 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2860 2865 return (SET_ERROR(ENOTSUP));
2861 2866
2862 2867 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2863 2868 if (error != 0)
2864 2869 return (error);
2865 2870
2866 - if (dsl_dataset_is_snapshot(ds)) {
2871 + if (ds->ds_is_snapshot) {
2867 2872 dsl_dataset_rele(ds, FTAG);
2868 2873 return (SET_ERROR(EINVAL));
2869 2874 }
2870 2875
2871 2876 error = dsl_prop_predict(ds->ds_dir,
2872 2877 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2873 2878 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2874 2879 if (error != 0) {
2875 2880 dsl_dataset_rele(ds, FTAG);
2876 2881 return (error);
2877 2882 }
2878 2883
2879 2884 if (newval == 0) {
2880 2885 dsl_dataset_rele(ds, FTAG);
2881 2886 return (0);
2882 2887 }
2883 2888
2884 2889 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
2885 2890 newval < ds->ds_reserved) {
2886 2891 dsl_dataset_rele(ds, FTAG);
2887 2892 return (SET_ERROR(ENOSPC));
2888 2893 }
2889 2894
2890 2895 dsl_dataset_rele(ds, FTAG);
2891 2896 return (0);
2892 2897 }
2893 2898
2894 2899 static void
2895 2900 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2896 2901 {
2897 2902 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2898 2903 dsl_pool_t *dp = dmu_tx_pool(tx);
2899 2904 dsl_dataset_t *ds;
2900 2905 uint64_t newval;
2901 2906
2902 2907 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2903 2908
2904 2909 dsl_prop_set_sync_impl(ds,
2905 2910 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2906 2911 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2907 2912 &ddsqra->ddsqra_value, tx);
2908 2913
2909 2914 VERIFY0(dsl_prop_get_int_ds(ds,
2910 2915 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2911 2916
2912 2917 if (ds->ds_quota != newval) {
2913 2918 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2914 2919 ds->ds_quota = newval;
2915 2920 }
2916 2921 dsl_dataset_rele(ds, FTAG);
2917 2922 }
2918 2923
2919 2924 int
2920 2925 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2921 2926 uint64_t refquota)
2922 2927 {
2923 2928 dsl_dataset_set_qr_arg_t ddsqra;
2924 2929
2925 2930 ddsqra.ddsqra_name = dsname;
2926 2931 ddsqra.ddsqra_source = source;
2927 2932 ddsqra.ddsqra_value = refquota;
2928 2933
2929 2934 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2930 2935 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
2931 2936 }
2932 2937
2933 2938 static int
2934 2939 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2935 2940 {
2936 2941 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2937 2942 dsl_pool_t *dp = dmu_tx_pool(tx);
2938 2943 dsl_dataset_t *ds;
|
↓ open down ↓ |
62 lines elided |
↑ open up ↑ |
2939 2944 int error;
2940 2945 uint64_t newval, unique;
2941 2946
2942 2947 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2943 2948 return (SET_ERROR(ENOTSUP));
2944 2949
2945 2950 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2946 2951 if (error != 0)
2947 2952 return (error);
2948 2953
2949 - if (dsl_dataset_is_snapshot(ds)) {
2954 + if (ds->ds_is_snapshot) {
2950 2955 dsl_dataset_rele(ds, FTAG);
2951 2956 return (SET_ERROR(EINVAL));
2952 2957 }
2953 2958
2954 2959 error = dsl_prop_predict(ds->ds_dir,
2955 2960 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2956 2961 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2957 2962 if (error != 0) {
2958 2963 dsl_dataset_rele(ds, FTAG);
2959 2964 return (error);
2960 2965 }
2961 2966
2962 2967 /*
2963 2968 * If we are doing the preliminary check in open context, the
2964 2969 * space estimates may be inaccurate.
2965 2970 */
2966 2971 if (!dmu_tx_is_syncing(tx)) {
2967 2972 dsl_dataset_rele(ds, FTAG);
2968 2973 return (0);
2969 2974 }
2970 2975
2971 2976 mutex_enter(&ds->ds_lock);
2972 2977 if (!DS_UNIQUE_IS_ACCURATE(ds))
2973 2978 dsl_dataset_recalc_head_uniq(ds);
2974 2979 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
2975 2980 mutex_exit(&ds->ds_lock);
2976 2981
2977 2982 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2978 2983 uint64_t delta = MAX(unique, newval) -
2979 2984 MAX(unique, ds->ds_reserved);
2980 2985
2981 2986 if (delta >
2982 2987 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2983 2988 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2984 2989 dsl_dataset_rele(ds, FTAG);
2985 2990 return (SET_ERROR(ENOSPC));
2986 2991 }
2987 2992 }
2988 2993
2989 2994 dsl_dataset_rele(ds, FTAG);
2990 2995 return (0);
2991 2996 }
2992 2997
2993 2998 void
2994 2999 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2995 3000 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2996 3001 {
2997 3002 uint64_t newval;
2998 3003 uint64_t unique;
2999 3004 int64_t delta;
3000 3005
3001 3006 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3002 3007 source, sizeof (value), 1, &value, tx);
3003 3008
3004 3009 VERIFY0(dsl_prop_get_int_ds(ds,
3005 3010 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3006 3011
3007 3012 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3008 3013 mutex_enter(&ds->ds_dir->dd_lock);
3009 3014 mutex_enter(&ds->ds_lock);
3010 3015 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3011 3016 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3012 3017 delta = MAX(0, (int64_t)(newval - unique)) -
3013 3018 MAX(0, (int64_t)(ds->ds_reserved - unique));
3014 3019 ds->ds_reserved = newval;
3015 3020 mutex_exit(&ds->ds_lock);
3016 3021
3017 3022 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3018 3023 mutex_exit(&ds->ds_dir->dd_lock);
3019 3024 }
3020 3025
3021 3026 static void
3022 3027 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3023 3028 {
3024 3029 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3025 3030 dsl_pool_t *dp = dmu_tx_pool(tx);
3026 3031 dsl_dataset_t *ds;
3027 3032
3028 3033 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3029 3034 dsl_dataset_set_refreservation_sync_impl(ds,
3030 3035 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3031 3036 dsl_dataset_rele(ds, FTAG);
3032 3037 }
3033 3038
3034 3039 int
3035 3040 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3036 3041 uint64_t refreservation)
3037 3042 {
3038 3043 dsl_dataset_set_qr_arg_t ddsqra;
3039 3044
3040 3045 ddsqra.ddsqra_name = dsname;
3041 3046 ddsqra.ddsqra_source = source;
3042 3047 ddsqra.ddsqra_value = refreservation;
3043 3048
3044 3049 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3045 3050 dsl_dataset_set_refreservation_sync, &ddsqra,
3046 3051 0, ZFS_SPACE_CHECK_NONE));
3047 3052 }
3048 3053
3049 3054 /*
3050 3055 * Return (in *usedp) the amount of space written in new that is not
3051 3056 * present in oldsnap. New may be a snapshot or the head. Old must be
3052 3057 * a snapshot before new, in new's filesystem (or its origin). If not then
3053 3058 * fail and return EINVAL.
3054 3059 *
3055 3060 * The written space is calculated by considering two components: First, we
3056 3061 * ignore any freed space, and calculate the written as new's used space
3057 3062 * minus old's used space. Next, we add in the amount of space that was freed
3058 3063 * between the two snapshots, thus reducing new's used space relative to old's.
3059 3064 * Specifically, this is the space that was born before old->ds_creation_txg,
3060 3065 * and freed before new (ie. on new's deadlist or a previous deadlist).
3061 3066 *
3062 3067 * space freed [---------------------]
3063 3068 * snapshots ---O-------O--------O-------O------
3064 3069 * oldsnap new
3065 3070 */
3066 3071 int
3067 3072 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3068 3073 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3069 3074 {
3070 3075 int err = 0;
3071 3076 uint64_t snapobj;
3072 3077 dsl_pool_t *dp = new->ds_dir->dd_pool;
3073 3078
3074 3079 ASSERT(dsl_pool_config_held(dp));
3075 3080
3076 3081 *usedp = 0;
3077 3082 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3078 3083 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3079 3084
3080 3085 *compp = 0;
3081 3086 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3082 3087 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3083 3088
3084 3089 *uncompp = 0;
3085 3090 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3086 3091 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3087 3092
3088 3093 snapobj = new->ds_object;
3089 3094 while (snapobj != oldsnap->ds_object) {
3090 3095 dsl_dataset_t *snap;
3091 3096 uint64_t used, comp, uncomp;
3092 3097
3093 3098 if (snapobj == new->ds_object) {
3094 3099 snap = new;
3095 3100 } else {
3096 3101 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3097 3102 if (err != 0)
3098 3103 break;
3099 3104 }
3100 3105
3101 3106 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3102 3107 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3103 3108 /*
3104 3109 * The blocks in the deadlist can not be born after
3105 3110 * ds_prev_snap_txg, so get the whole deadlist space,
3106 3111 * which is more efficient (especially for old-format
3107 3112 * deadlists). Unfortunately the deadlist code
3108 3113 * doesn't have enough information to make this
3109 3114 * optimization itself.
3110 3115 */
3111 3116 dsl_deadlist_space(&snap->ds_deadlist,
3112 3117 &used, &comp, &uncomp);
3113 3118 } else {
3114 3119 dsl_deadlist_space_range(&snap->ds_deadlist,
3115 3120 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3116 3121 &used, &comp, &uncomp);
3117 3122 }
3118 3123 *usedp += used;
3119 3124 *compp += comp;
3120 3125 *uncompp += uncomp;
3121 3126
3122 3127 /*
3123 3128 * If we get to the beginning of the chain of snapshots
3124 3129 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3125 3130 * was not a snapshot of/before new.
3126 3131 */
3127 3132 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3128 3133 if (snap != new)
3129 3134 dsl_dataset_rele(snap, FTAG);
3130 3135 if (snapobj == 0) {
3131 3136 err = SET_ERROR(EINVAL);
3132 3137 break;
3133 3138 }
3134 3139
3135 3140 }
3136 3141 return (err);
3137 3142 }
3138 3143
3139 3144 /*
3140 3145 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3141 3146 * lastsnap, and all snapshots in between are deleted.
3142 3147 *
3143 3148 * blocks that would be freed [---------------------------]
3144 3149 * snapshots ---O-------O--------O-------O--------O
3145 3150 * firstsnap lastsnap
3146 3151 *
3147 3152 * This is the set of blocks that were born after the snap before firstsnap,
3148 3153 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3149 3154 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3150 3155 * We calculate this by iterating over the relevant deadlists (from the snap
3151 3156 * after lastsnap, backward to the snap after firstsnap), summing up the
3152 3157 * space on the deadlist that was born after the snap before firstsnap.
|
↓ open down ↓ |
193 lines elided |
↑ open up ↑ |
3153 3158 */
3154 3159 int
3155 3160 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3156 3161 dsl_dataset_t *lastsnap,
3157 3162 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3158 3163 {
3159 3164 int err = 0;
3160 3165 uint64_t snapobj;
3161 3166 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3162 3167
3163 - ASSERT(dsl_dataset_is_snapshot(firstsnap));
3164 - ASSERT(dsl_dataset_is_snapshot(lastsnap));
3168 + ASSERT(firstsnap->ds_is_snapshot);
3169 + ASSERT(lastsnap->ds_is_snapshot);
3165 3170
3166 3171 /*
3167 3172 * Check that the snapshots are in the same dsl_dir, and firstsnap
3168 3173 * is before lastsnap.
3169 3174 */
3170 3175 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3171 3176 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3172 3177 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3173 3178 return (SET_ERROR(EINVAL));
3174 3179
3175 3180 *usedp = *compp = *uncompp = 0;
3176 3181
3177 3182 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3178 3183 while (snapobj != firstsnap->ds_object) {
3179 3184 dsl_dataset_t *ds;
3180 3185 uint64_t used, comp, uncomp;
3181 3186
3182 3187 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3183 3188 if (err != 0)
3184 3189 break;
3185 3190
3186 3191 dsl_deadlist_space_range(&ds->ds_deadlist,
3187 3192 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3188 3193 &used, &comp, &uncomp);
3189 3194 *usedp += used;
3190 3195 *compp += comp;
3191 3196 *uncompp += uncomp;
3192 3197
3193 3198 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3194 3199 ASSERT3U(snapobj, !=, 0);
3195 3200 dsl_dataset_rele(ds, FTAG);
3196 3201 }
3197 3202 return (err);
3198 3203 }
3199 3204
3200 3205 static int
3201 3206 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3202 3207 {
3203 3208 const char *dsname = arg;
3204 3209 dsl_dataset_t *ds;
3205 3210 dsl_pool_t *dp = dmu_tx_pool(tx);
3206 3211 int error = 0;
3207 3212
3208 3213 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3209 3214 return (SET_ERROR(ENOTSUP));
3210 3215
3211 3216 ASSERT(spa_feature_is_enabled(dp->dp_spa,
3212 3217 SPA_FEATURE_EXTENSIBLE_DATASET));
3213 3218
3214 3219 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3215 3220 if (error != 0)
3216 3221 return (error);
3217 3222
3218 3223 if (ds->ds_large_blocks)
3219 3224 error = EALREADY;
3220 3225 dsl_dataset_rele(ds, FTAG);
3221 3226
3222 3227 return (error);
3223 3228 }
3224 3229
3225 3230 void
3226 3231 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3227 3232 {
3228 3233 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3229 3234 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3230 3235 uint64_t zero = 0;
3231 3236
3232 3237 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3233 3238 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3234 3239
3235 3240 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3236 3241 sizeof (zero), 1, &zero, tx));
3237 3242 }
3238 3243
3239 3244 static void
3240 3245 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3241 3246 {
3242 3247 const char *dsname = arg;
3243 3248 dsl_dataset_t *ds;
3244 3249
3245 3250 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3246 3251
3247 3252 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3248 3253 ASSERT(!ds->ds_large_blocks);
3249 3254 ds->ds_large_blocks = B_TRUE;
3250 3255 dsl_dataset_rele(ds, FTAG);
3251 3256 }
3252 3257
3253 3258 int
3254 3259 dsl_dataset_activate_large_blocks(const char *dsname)
3255 3260 {
3256 3261 int error;
3257 3262
3258 3263 error = dsl_sync_task(dsname,
3259 3264 dsl_dataset_activate_large_blocks_check,
3260 3265 dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3261 3266 1, ZFS_SPACE_CHECK_RESERVED);
3262 3267
3263 3268 /*
3264 3269 * EALREADY indicates that this dataset already supports large blocks.
3265 3270 */
3266 3271 if (error == EALREADY)
3267 3272 error = 0;
3268 3273 return (error);
3269 3274 }
3270 3275
3271 3276 /*
3272 3277 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3273 3278 * For example, they could both be snapshots of the same filesystem, and
3274 3279 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3275 3280 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3276 3281 * filesystem. Or 'earlier' could be the origin's origin.
3277 3282 *
3278 3283 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
|
↓ open down ↓ |
104 lines elided |
↑ open up ↑ |
3279 3284 */
3280 3285 boolean_t
3281 3286 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3282 3287 uint64_t earlier_txg)
3283 3288 {
3284 3289 dsl_pool_t *dp = later->ds_dir->dd_pool;
3285 3290 int error;
3286 3291 boolean_t ret;
3287 3292
3288 3293 ASSERT(dsl_pool_config_held(dp));
3289 - ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3294 + ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3290 3295
3291 3296 if (earlier_txg == 0)
3292 3297 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3293 3298
3294 - if (dsl_dataset_is_snapshot(later) &&
3299 + if (later->ds_is_snapshot &&
3295 3300 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3296 3301 return (B_FALSE);
3297 3302
3298 3303 if (later->ds_dir == earlier->ds_dir)
3299 3304 return (B_TRUE);
3300 3305 if (!dsl_dir_is_clone(later->ds_dir))
3301 3306 return (B_FALSE);
3302 3307
3303 3308 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3304 3309 return (B_TRUE);
3305 3310 dsl_dataset_t *origin;
3306 3311 error = dsl_dataset_hold_obj(dp,
3307 3312 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3308 3313 if (error != 0)
3309 3314 return (B_FALSE);
3310 3315 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3311 3316 dsl_dataset_rele(origin, FTAG);
3312 3317 return (ret);
3313 3318 }
3314 3319
3315 3320
3316 3321 void
3317 3322 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3318 3323 {
3319 3324 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3320 3325 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3321 3326 }
|
↓ open down ↓ |
17 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX