]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
f5fc4aca | 23 | * Copyright (c) 2011 by Delphix. All rights reserved. |
34dc7c2f BB |
24 | */ |
25 | ||
34dc7c2f BB |
26 | #include <sys/dmu_objset.h> |
27 | #include <sys/dsl_dataset.h> | |
28 | #include <sys/dsl_dir.h> | |
29 | #include <sys/dsl_prop.h> | |
30 | #include <sys/dsl_synctask.h> | |
31 | #include <sys/dmu_traverse.h> | |
32 | #include <sys/dmu_tx.h> | |
33 | #include <sys/arc.h> | |
34 | #include <sys/zio.h> | |
35 | #include <sys/zap.h> | |
36 | #include <sys/unique.h> | |
37 | #include <sys/zfs_context.h> | |
38 | #include <sys/zfs_ioctl.h> | |
39 | #include <sys/spa.h> | |
b128c09f | 40 | #include <sys/zfs_znode.h> |
572e2857 | 41 | #include <sys/zfs_onexit.h> |
45d1cae3 | 42 | #include <sys/zvol.h> |
428870ff BB |
43 | #include <sys/dsl_scan.h> |
44 | #include <sys/dsl_deadlist.h> | |
45 | ||
b128c09f BB |
46 | static char *dsl_reaper = "the grim reaper"; |
47 | ||
34dc7c2f BB |
48 | static dsl_checkfunc_t dsl_dataset_destroy_begin_check; |
49 | static dsl_syncfunc_t dsl_dataset_destroy_begin_sync; | |
34dc7c2f BB |
50 | static dsl_syncfunc_t dsl_dataset_set_reservation_sync; |
51 | ||
428870ff BB |
52 | #define SWITCH64(x, y) \ |
53 | { \ | |
54 | uint64_t __tmp = (x); \ | |
55 | (x) = (y); \ | |
56 | (y) = __tmp; \ | |
57 | } | |
58 | ||
34dc7c2f BB |
59 | #define DS_REF_MAX (1ULL << 62) |
60 | ||
61 | #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE | |
62 | ||
b128c09f BB |
63 | #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper) |
64 | ||
34dc7c2f BB |
65 | |
66 | /* | |
67 | * Figure out how much of this delta should be propogated to the dsl_dir | |
68 | * layer. If there's a refreservation, that space has already been | |
69 | * partially accounted for in our ancestors. | |
70 | */ | |
71 | static int64_t | |
72 | parent_delta(dsl_dataset_t *ds, int64_t delta) | |
73 | { | |
74 | uint64_t old_bytes, new_bytes; | |
75 | ||
76 | if (ds->ds_reserved == 0) | |
77 | return (delta); | |
78 | ||
79 | old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); | |
80 | new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved); | |
81 | ||
82 | ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta)); | |
83 | return (new_bytes - old_bytes); | |
84 | } | |
85 | ||
86 | void | |
428870ff | 87 | dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx) |
34dc7c2f | 88 | { |
d6320ddb | 89 | int used, compressed, uncompressed; |
34dc7c2f BB |
90 | int64_t delta; |
91 | ||
d6320ddb BB |
92 | used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); |
93 | compressed = BP_GET_PSIZE(bp); | |
94 | uncompressed = BP_GET_UCSIZE(bp); | |
95 | ||
428870ff | 96 | dprintf_bp(bp, "ds=%p", ds); |
34dc7c2f BB |
97 | |
98 | ASSERT(dmu_tx_is_syncing(tx)); | |
99 | /* It could have been compressed away to nothing */ | |
100 | if (BP_IS_HOLE(bp)) | |
101 | return; | |
102 | ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE); | |
103 | ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES); | |
104 | if (ds == NULL) { | |
105 | /* | |
106 | * Account for the meta-objset space in its placeholder | |
107 | * dsl_dir. | |
108 | */ | |
109 | ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */ | |
b128c09f | 110 | dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, |
34dc7c2f BB |
111 | used, compressed, uncompressed, tx); |
112 | dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
113 | return; | |
114 | } | |
115 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
428870ff | 116 | |
b128c09f | 117 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f BB |
118 | mutex_enter(&ds->ds_lock); |
119 | delta = parent_delta(ds, used); | |
120 | ds->ds_phys->ds_used_bytes += used; | |
121 | ds->ds_phys->ds_compressed_bytes += compressed; | |
122 | ds->ds_phys->ds_uncompressed_bytes += uncompressed; | |
123 | ds->ds_phys->ds_unique_bytes += used; | |
124 | mutex_exit(&ds->ds_lock); | |
b128c09f BB |
125 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta, |
126 | compressed, uncompressed, tx); | |
127 | dsl_dir_transfer_space(ds->ds_dir, used - delta, | |
128 | DD_USED_REFRSRV, DD_USED_HEAD, tx); | |
129 | mutex_exit(&ds->ds_dir->dd_lock); | |
34dc7c2f BB |
130 | } |
131 | ||
b128c09f | 132 | int |
428870ff BB |
133 | dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx, |
134 | boolean_t async) | |
34dc7c2f | 135 | { |
d6320ddb BB |
136 | int used, compressed, uncompressed; |
137 | ||
34dc7c2f | 138 | if (BP_IS_HOLE(bp)) |
b128c09f | 139 | return (0); |
34dc7c2f | 140 | |
428870ff BB |
141 | ASSERT(dmu_tx_is_syncing(tx)); |
142 | ASSERT(bp->blk_birth <= tx->tx_txg); | |
143 | ||
d6320ddb BB |
144 | used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp); |
145 | compressed = BP_GET_PSIZE(bp); | |
146 | uncompressed = BP_GET_UCSIZE(bp); | |
428870ff | 147 | |
34dc7c2f BB |
148 | ASSERT(used > 0); |
149 | if (ds == NULL) { | |
34dc7c2f BB |
150 | /* |
151 | * Account for the meta-objset space in its placeholder | |
152 | * dataset. | |
153 | */ | |
428870ff | 154 | dsl_free(tx->tx_pool, tx->tx_txg, bp); |
34dc7c2f | 155 | |
b128c09f | 156 | dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD, |
34dc7c2f BB |
157 | -used, -compressed, -uncompressed, tx); |
158 | dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx); | |
b128c09f | 159 | return (used); |
34dc7c2f BB |
160 | } |
161 | ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool); | |
162 | ||
b128c09f | 163 | ASSERT(!dsl_dataset_is_snapshot(ds)); |
34dc7c2f BB |
164 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
165 | ||
166 | if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) { | |
34dc7c2f BB |
167 | int64_t delta; |
168 | ||
428870ff BB |
169 | dprintf_bp(bp, "freeing ds=%llu", ds->ds_object); |
170 | dsl_free(tx->tx_pool, tx->tx_txg, bp); | |
34dc7c2f | 171 | |
b128c09f | 172 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f BB |
173 | mutex_enter(&ds->ds_lock); |
174 | ASSERT(ds->ds_phys->ds_unique_bytes >= used || | |
175 | !DS_UNIQUE_IS_ACCURATE(ds)); | |
176 | delta = parent_delta(ds, -used); | |
177 | ds->ds_phys->ds_unique_bytes -= used; | |
178 | mutex_exit(&ds->ds_lock); | |
b128c09f | 179 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, |
34dc7c2f | 180 | delta, -compressed, -uncompressed, tx); |
b128c09f BB |
181 | dsl_dir_transfer_space(ds->ds_dir, -used - delta, |
182 | DD_USED_REFRSRV, DD_USED_HEAD, tx); | |
183 | mutex_exit(&ds->ds_dir->dd_lock); | |
34dc7c2f BB |
184 | } else { |
185 | dprintf_bp(bp, "putting on dead list: %s", ""); | |
428870ff BB |
186 | if (async) { |
187 | /* | |
188 | * We are here as part of zio's write done callback, | |
189 | * which means we're a zio interrupt thread. We can't | |
190 | * call dsl_deadlist_insert() now because it may block | |
191 | * waiting for I/O. Instead, put bp on the deferred | |
192 | * queue and let dsl_pool_sync() finish the job. | |
193 | */ | |
194 | bplist_append(&ds->ds_pending_deadlist, bp); | |
195 | } else { | |
196 | dsl_deadlist_insert(&ds->ds_deadlist, bp, tx); | |
197 | } | |
34dc7c2f BB |
198 | ASSERT3U(ds->ds_prev->ds_object, ==, |
199 | ds->ds_phys->ds_prev_snap_obj); | |
200 | ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0); | |
201 | /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */ | |
202 | if (ds->ds_prev->ds_phys->ds_next_snap_obj == | |
203 | ds->ds_object && bp->blk_birth > | |
204 | ds->ds_prev->ds_phys->ds_prev_snap_txg) { | |
205 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
206 | mutex_enter(&ds->ds_prev->ds_lock); | |
207 | ds->ds_prev->ds_phys->ds_unique_bytes += used; | |
208 | mutex_exit(&ds->ds_prev->ds_lock); | |
209 | } | |
428870ff | 210 | if (bp->blk_birth > ds->ds_dir->dd_origin_txg) { |
b128c09f BB |
211 | dsl_dir_transfer_space(ds->ds_dir, used, |
212 | DD_USED_HEAD, DD_USED_SNAP, tx); | |
213 | } | |
34dc7c2f BB |
214 | } |
215 | mutex_enter(&ds->ds_lock); | |
216 | ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used); | |
217 | ds->ds_phys->ds_used_bytes -= used; | |
218 | ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed); | |
219 | ds->ds_phys->ds_compressed_bytes -= compressed; | |
220 | ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed); | |
221 | ds->ds_phys->ds_uncompressed_bytes -= uncompressed; | |
222 | mutex_exit(&ds->ds_lock); | |
b128c09f BB |
223 | |
224 | return (used); | |
34dc7c2f BB |
225 | } |
226 | ||
227 | uint64_t | |
228 | dsl_dataset_prev_snap_txg(dsl_dataset_t *ds) | |
229 | { | |
230 | uint64_t trysnap = 0; | |
231 | ||
232 | if (ds == NULL) | |
233 | return (0); | |
234 | /* | |
235 | * The snapshot creation could fail, but that would cause an | |
236 | * incorrect FALSE return, which would only result in an | |
237 | * overestimation of the amount of space that an operation would | |
238 | * consume, which is OK. | |
239 | * | |
240 | * There's also a small window where we could miss a pending | |
241 | * snapshot, because we could set the sync task in the quiescing | |
242 | * phase. So this should only be used as a guess. | |
243 | */ | |
244 | if (ds->ds_trysnap_txg > | |
245 | spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa)) | |
246 | trysnap = ds->ds_trysnap_txg; | |
247 | return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap)); | |
248 | } | |
249 | ||
9babb374 | 250 | boolean_t |
428870ff BB |
251 | dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp, |
252 | uint64_t blk_birth) | |
34dc7c2f | 253 | { |
428870ff BB |
254 | if (blk_birth <= dsl_dataset_prev_snap_txg(ds)) |
255 | return (B_FALSE); | |
256 | ||
572e2857 | 257 | ddt_prefetch(dsl_dataset_get_spa(ds), bp); |
428870ff BB |
258 | |
259 | return (B_TRUE); | |
34dc7c2f BB |
260 | } |
261 | ||
262 | /* ARGSUSED */ | |
263 | static void | |
264 | dsl_dataset_evict(dmu_buf_t *db, void *dsv) | |
265 | { | |
266 | dsl_dataset_t *ds = dsv; | |
267 | ||
b128c09f | 268 | ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds)); |
34dc7c2f | 269 | |
34dc7c2f BB |
270 | unique_remove(ds->ds_fsid_guid); |
271 | ||
428870ff BB |
272 | if (ds->ds_objset != NULL) |
273 | dmu_objset_evict(ds->ds_objset); | |
34dc7c2f BB |
274 | |
275 | if (ds->ds_prev) { | |
b128c09f | 276 | dsl_dataset_drop_ref(ds->ds_prev, ds); |
34dc7c2f BB |
277 | ds->ds_prev = NULL; |
278 | } | |
279 | ||
428870ff BB |
280 | bplist_destroy(&ds->ds_pending_deadlist); |
281 | if (db != NULL) { | |
282 | dsl_deadlist_close(&ds->ds_deadlist); | |
283 | } else { | |
284 | ASSERT(ds->ds_deadlist.dl_dbuf == NULL); | |
285 | ASSERT(!ds->ds_deadlist.dl_oldfmt); | |
286 | } | |
b128c09f BB |
287 | if (ds->ds_dir) |
288 | dsl_dir_close(ds->ds_dir, ds); | |
34dc7c2f BB |
289 | |
290 | ASSERT(!list_link_active(&ds->ds_synced_link)); | |
291 | ||
292 | mutex_destroy(&ds->ds_lock); | |
45d1cae3 | 293 | mutex_destroy(&ds->ds_recvlock); |
34dc7c2f | 294 | mutex_destroy(&ds->ds_opening_lock); |
b128c09f BB |
295 | rw_destroy(&ds->ds_rwlock); |
296 | cv_destroy(&ds->ds_exclusive_cv); | |
34dc7c2f BB |
297 | |
298 | kmem_free(ds, sizeof (dsl_dataset_t)); | |
299 | } | |
300 | ||
301 | static int | |
302 | dsl_dataset_get_snapname(dsl_dataset_t *ds) | |
303 | { | |
304 | dsl_dataset_phys_t *headphys; | |
305 | int err; | |
306 | dmu_buf_t *headdbuf; | |
307 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
308 | objset_t *mos = dp->dp_meta_objset; | |
309 | ||
310 | if (ds->ds_snapname[0]) | |
311 | return (0); | |
312 | if (ds->ds_phys->ds_next_snap_obj == 0) | |
313 | return (0); | |
314 | ||
315 | err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj, | |
316 | FTAG, &headdbuf); | |
317 | if (err) | |
318 | return (err); | |
319 | headphys = headdbuf->db_data; | |
320 | err = zap_value_search(dp->dp_meta_objset, | |
321 | headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname); | |
322 | dmu_buf_rele(headdbuf, FTAG); | |
323 | return (err); | |
324 | } | |
325 | ||
326 | static int | |
b128c09f | 327 | dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value) |
34dc7c2f | 328 | { |
b128c09f BB |
329 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; |
330 | uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; | |
34dc7c2f BB |
331 | matchtype_t mt; |
332 | int err; | |
333 | ||
b128c09f | 334 | if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) |
34dc7c2f BB |
335 | mt = MT_FIRST; |
336 | else | |
337 | mt = MT_EXACT; | |
338 | ||
b128c09f | 339 | err = zap_lookup_norm(mos, snapobj, name, 8, 1, |
34dc7c2f BB |
340 | value, mt, NULL, 0, NULL); |
341 | if (err == ENOTSUP && mt == MT_FIRST) | |
b128c09f | 342 | err = zap_lookup(mos, snapobj, name, 8, 1, value); |
34dc7c2f BB |
343 | return (err); |
344 | } | |
345 | ||
346 | static int | |
b128c09f | 347 | dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx) |
34dc7c2f | 348 | { |
b128c09f BB |
349 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; |
350 | uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj; | |
34dc7c2f BB |
351 | matchtype_t mt; |
352 | int err; | |
353 | ||
428870ff BB |
354 | dsl_dir_snap_cmtime_update(ds->ds_dir); |
355 | ||
b128c09f | 356 | if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET) |
34dc7c2f BB |
357 | mt = MT_FIRST; |
358 | else | |
359 | mt = MT_EXACT; | |
360 | ||
b128c09f | 361 | err = zap_remove_norm(mos, snapobj, name, mt, tx); |
34dc7c2f | 362 | if (err == ENOTSUP && mt == MT_FIRST) |
b128c09f | 363 | err = zap_remove(mos, snapobj, name, tx); |
34dc7c2f BB |
364 | return (err); |
365 | } | |
366 | ||
b128c09f BB |
367 | static int |
368 | dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag, | |
369 | dsl_dataset_t **dsp) | |
34dc7c2f | 370 | { |
34dc7c2f BB |
371 | objset_t *mos = dp->dp_meta_objset; |
372 | dmu_buf_t *dbuf; | |
373 | dsl_dataset_t *ds; | |
374 | int err; | |
572e2857 | 375 | dmu_object_info_t doi; |
34dc7c2f BB |
376 | |
377 | ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
378 | dsl_pool_sync_context(dp)); | |
379 | ||
380 | err = dmu_bonus_hold(mos, dsobj, tag, &dbuf); | |
381 | if (err) | |
382 | return (err); | |
572e2857 BB |
383 | |
384 | /* Make sure dsobj has the correct object type. */ | |
385 | dmu_object_info_from_db(dbuf, &doi); | |
386 | if (doi.doi_type != DMU_OT_DSL_DATASET) | |
387 | return (EINVAL); | |
388 | ||
34dc7c2f BB |
389 | ds = dmu_buf_get_user(dbuf); |
390 | if (ds == NULL) { | |
d4ed6673 | 391 | dsl_dataset_t *winner = NULL; |
34dc7c2f BB |
392 | |
393 | ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); | |
394 | ds->ds_dbuf = dbuf; | |
395 | ds->ds_object = dsobj; | |
396 | ds->ds_phys = dbuf->db_data; | |
98f72a53 | 397 | list_link_init(&ds->ds_synced_link); |
34dc7c2f BB |
398 | |
399 | mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL); | |
45d1cae3 | 400 | mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL); |
34dc7c2f | 401 | mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL); |
ef5319df | 402 | rw_init(&ds->ds_rwlock, NULL, RW_DEFAULT, NULL); |
b128c09f | 403 | cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL); |
34dc7c2f | 404 | |
428870ff BB |
405 | bplist_create(&ds->ds_pending_deadlist); |
406 | dsl_deadlist_open(&ds->ds_deadlist, | |
34dc7c2f | 407 | mos, ds->ds_phys->ds_deadlist_obj); |
428870ff | 408 | |
34dc7c2f BB |
409 | if (err == 0) { |
410 | err = dsl_dir_open_obj(dp, | |
411 | ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir); | |
412 | } | |
413 | if (err) { | |
34dc7c2f | 414 | mutex_destroy(&ds->ds_lock); |
45d1cae3 | 415 | mutex_destroy(&ds->ds_recvlock); |
34dc7c2f | 416 | mutex_destroy(&ds->ds_opening_lock); |
b128c09f BB |
417 | rw_destroy(&ds->ds_rwlock); |
418 | cv_destroy(&ds->ds_exclusive_cv); | |
428870ff BB |
419 | bplist_destroy(&ds->ds_pending_deadlist); |
420 | dsl_deadlist_close(&ds->ds_deadlist); | |
34dc7c2f BB |
421 | kmem_free(ds, sizeof (dsl_dataset_t)); |
422 | dmu_buf_rele(dbuf, tag); | |
423 | return (err); | |
424 | } | |
425 | ||
b128c09f | 426 | if (!dsl_dataset_is_snapshot(ds)) { |
34dc7c2f BB |
427 | ds->ds_snapname[0] = '\0'; |
428 | if (ds->ds_phys->ds_prev_snap_obj) { | |
b128c09f BB |
429 | err = dsl_dataset_get_ref(dp, |
430 | ds->ds_phys->ds_prev_snap_obj, | |
431 | ds, &ds->ds_prev); | |
34dc7c2f | 432 | } |
45d1cae3 BB |
433 | } else { |
434 | if (zfs_flags & ZFS_DEBUG_SNAPNAMES) | |
435 | err = dsl_dataset_get_snapname(ds); | |
436 | if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) { | |
437 | err = zap_count( | |
438 | ds->ds_dir->dd_pool->dp_meta_objset, | |
439 | ds->ds_phys->ds_userrefs_obj, | |
440 | &ds->ds_userrefs); | |
441 | } | |
34dc7c2f BB |
442 | } |
443 | ||
b128c09f | 444 | if (err == 0 && !dsl_dataset_is_snapshot(ds)) { |
34dc7c2f BB |
445 | /* |
446 | * In sync context, we're called with either no lock | |
447 | * or with the write lock. If we're not syncing, | |
448 | * we're always called with the read lock held. | |
449 | */ | |
450 | boolean_t need_lock = | |
451 | !RW_WRITE_HELD(&dp->dp_config_rwlock) && | |
452 | dsl_pool_sync_context(dp); | |
453 | ||
454 | if (need_lock) | |
455 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
456 | ||
b128c09f | 457 | err = dsl_prop_get_ds(ds, |
34dc7c2f BB |
458 | "refreservation", sizeof (uint64_t), 1, |
459 | &ds->ds_reserved, NULL); | |
460 | if (err == 0) { | |
b128c09f | 461 | err = dsl_prop_get_ds(ds, |
34dc7c2f BB |
462 | "refquota", sizeof (uint64_t), 1, |
463 | &ds->ds_quota, NULL); | |
464 | } | |
465 | ||
466 | if (need_lock) | |
467 | rw_exit(&dp->dp_config_rwlock); | |
468 | } else { | |
469 | ds->ds_reserved = ds->ds_quota = 0; | |
470 | } | |
471 | ||
472 | if (err == 0) { | |
473 | winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys, | |
474 | dsl_dataset_evict); | |
475 | } | |
476 | if (err || winner) { | |
428870ff BB |
477 | bplist_destroy(&ds->ds_pending_deadlist); |
478 | dsl_deadlist_close(&ds->ds_deadlist); | |
b128c09f BB |
479 | if (ds->ds_prev) |
480 | dsl_dataset_drop_ref(ds->ds_prev, ds); | |
34dc7c2f BB |
481 | dsl_dir_close(ds->ds_dir, ds); |
482 | mutex_destroy(&ds->ds_lock); | |
45d1cae3 | 483 | mutex_destroy(&ds->ds_recvlock); |
34dc7c2f | 484 | mutex_destroy(&ds->ds_opening_lock); |
b128c09f BB |
485 | rw_destroy(&ds->ds_rwlock); |
486 | cv_destroy(&ds->ds_exclusive_cv); | |
34dc7c2f BB |
487 | kmem_free(ds, sizeof (dsl_dataset_t)); |
488 | if (err) { | |
489 | dmu_buf_rele(dbuf, tag); | |
490 | return (err); | |
491 | } | |
492 | ds = winner; | |
493 | } else { | |
494 | ds->ds_fsid_guid = | |
495 | unique_insert(ds->ds_phys->ds_fsid_guid); | |
496 | } | |
497 | } | |
498 | ASSERT3P(ds->ds_dbuf, ==, dbuf); | |
499 | ASSERT3P(ds->ds_phys, ==, dbuf->db_data); | |
b128c09f BB |
500 | ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 || |
501 | spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN || | |
502 | dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap); | |
34dc7c2f | 503 | mutex_enter(&ds->ds_lock); |
b128c09f | 504 | if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) { |
34dc7c2f | 505 | mutex_exit(&ds->ds_lock); |
b128c09f BB |
506 | dmu_buf_rele(ds->ds_dbuf, tag); |
507 | return (ENOENT); | |
34dc7c2f | 508 | } |
34dc7c2f | 509 | mutex_exit(&ds->ds_lock); |
34dc7c2f BB |
510 | *dsp = ds; |
511 | return (0); | |
512 | } | |
513 | ||
b128c09f BB |
514 | static int |
515 | dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag) | |
516 | { | |
517 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
518 | ||
519 | /* | |
520 | * In syncing context we don't want the rwlock lock: there | |
521 | * may be an existing writer waiting for sync phase to | |
522 | * finish. We don't need to worry about such writers, since | |
523 | * sync phase is single-threaded, so the writer can't be | |
524 | * doing anything while we are active. | |
525 | */ | |
526 | if (dsl_pool_sync_context(dp)) { | |
527 | ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); | |
528 | return (0); | |
529 | } | |
530 | ||
531 | /* | |
532 | * Normal users will hold the ds_rwlock as a READER until they | |
533 | * are finished (i.e., call dsl_dataset_rele()). "Owners" will | |
534 | * drop their READER lock after they set the ds_owner field. | |
535 | * | |
536 | * If the dataset is being destroyed, the destroy thread will | |
537 | * obtain a WRITER lock for exclusive access after it's done its | |
538 | * open-context work and then change the ds_owner to | |
539 | * dsl_reaper once destruction is assured. So threads | |
540 | * may block here temporarily, until the "destructability" of | |
541 | * the dataset is determined. | |
542 | */ | |
543 | ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock)); | |
544 | mutex_enter(&ds->ds_lock); | |
545 | while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) { | |
546 | rw_exit(&dp->dp_config_rwlock); | |
547 | cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock); | |
548 | if (DSL_DATASET_IS_DESTROYED(ds)) { | |
549 | mutex_exit(&ds->ds_lock); | |
550 | dsl_dataset_drop_ref(ds, tag); | |
551 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
552 | return (ENOENT); | |
553 | } | |
428870ff BB |
554 | /* |
555 | * The dp_config_rwlock lives above the ds_lock. And | |
556 | * we need to check DSL_DATASET_IS_DESTROYED() while | |
557 | * holding the ds_lock, so we have to drop and reacquire | |
558 | * the ds_lock here. | |
559 | */ | |
560 | mutex_exit(&ds->ds_lock); | |
b128c09f | 561 | rw_enter(&dp->dp_config_rwlock, RW_READER); |
428870ff | 562 | mutex_enter(&ds->ds_lock); |
b128c09f BB |
563 | } |
564 | mutex_exit(&ds->ds_lock); | |
565 | return (0); | |
566 | } | |
567 | ||
34dc7c2f | 568 | int |
b128c09f BB |
569 | dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag, |
570 | dsl_dataset_t **dsp) | |
571 | { | |
572 | int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp); | |
573 | ||
574 | if (err) | |
575 | return (err); | |
576 | return (dsl_dataset_hold_ref(*dsp, tag)); | |
577 | } | |
578 | ||
579 | int | |
428870ff BB |
580 | dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok, |
581 | void *tag, dsl_dataset_t **dsp) | |
b128c09f | 582 | { |
428870ff | 583 | int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp); |
b128c09f BB |
584 | if (err) |
585 | return (err); | |
428870ff BB |
586 | if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { |
587 | dsl_dataset_rele(*dsp, tag); | |
9babb374 | 588 | *dsp = NULL; |
b128c09f BB |
589 | return (EBUSY); |
590 | } | |
591 | return (0); | |
592 | } | |
593 | ||
594 | int | |
595 | dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp) | |
34dc7c2f BB |
596 | { |
597 | dsl_dir_t *dd; | |
598 | dsl_pool_t *dp; | |
b128c09f | 599 | const char *snapname; |
34dc7c2f | 600 | uint64_t obj; |
34dc7c2f BB |
601 | int err = 0; |
602 | ||
b128c09f | 603 | err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname); |
34dc7c2f BB |
604 | if (err) |
605 | return (err); | |
606 | ||
607 | dp = dd->dd_pool; | |
608 | obj = dd->dd_phys->dd_head_dataset_obj; | |
609 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
b128c09f BB |
610 | if (obj) |
611 | err = dsl_dataset_get_ref(dp, obj, tag, dsp); | |
612 | else | |
34dc7c2f | 613 | err = ENOENT; |
b128c09f | 614 | if (err) |
34dc7c2f | 615 | goto out; |
34dc7c2f | 616 | |
b128c09f | 617 | err = dsl_dataset_hold_ref(*dsp, tag); |
34dc7c2f | 618 | |
b128c09f BB |
619 | /* we may be looking for a snapshot */ |
620 | if (err == 0 && snapname != NULL) { | |
621 | dsl_dataset_t *ds = NULL; | |
34dc7c2f | 622 | |
b128c09f BB |
623 | if (*snapname++ != '@') { |
624 | dsl_dataset_rele(*dsp, tag); | |
34dc7c2f BB |
625 | err = ENOENT; |
626 | goto out; | |
627 | } | |
34dc7c2f | 628 | |
b128c09f BB |
629 | dprintf("looking for snapshot '%s'\n", snapname); |
630 | err = dsl_dataset_snap_lookup(*dsp, snapname, &obj); | |
631 | if (err == 0) | |
632 | err = dsl_dataset_get_ref(dp, obj, tag, &ds); | |
633 | dsl_dataset_rele(*dsp, tag); | |
634 | ||
635 | ASSERT3U((err == 0), ==, (ds != NULL)); | |
636 | ||
637 | if (ds) { | |
638 | mutex_enter(&ds->ds_lock); | |
639 | if (ds->ds_snapname[0] == 0) | |
640 | (void) strlcpy(ds->ds_snapname, snapname, | |
641 | sizeof (ds->ds_snapname)); | |
642 | mutex_exit(&ds->ds_lock); | |
643 | err = dsl_dataset_hold_ref(ds, tag); | |
644 | *dsp = err ? NULL : ds; | |
34dc7c2f | 645 | } |
34dc7c2f | 646 | } |
34dc7c2f BB |
647 | out: |
648 | rw_exit(&dp->dp_config_rwlock); | |
649 | dsl_dir_close(dd, FTAG); | |
34dc7c2f BB |
650 | return (err); |
651 | } | |
652 | ||
653 | int | |
428870ff BB |
654 | dsl_dataset_own(const char *name, boolean_t inconsistentok, |
655 | void *tag, dsl_dataset_t **dsp) | |
34dc7c2f | 656 | { |
428870ff | 657 | int err = dsl_dataset_hold(name, tag, dsp); |
b128c09f BB |
658 | if (err) |
659 | return (err); | |
428870ff BB |
660 | if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) { |
661 | dsl_dataset_rele(*dsp, tag); | |
b128c09f BB |
662 | return (EBUSY); |
663 | } | |
664 | return (0); | |
34dc7c2f BB |
665 | } |
666 | ||
667 | void | |
668 | dsl_dataset_name(dsl_dataset_t *ds, char *name) | |
669 | { | |
670 | if (ds == NULL) { | |
671 | (void) strcpy(name, "mos"); | |
672 | } else { | |
673 | dsl_dir_name(ds->ds_dir, name); | |
674 | VERIFY(0 == dsl_dataset_get_snapname(ds)); | |
675 | if (ds->ds_snapname[0]) { | |
676 | (void) strcat(name, "@"); | |
b128c09f BB |
677 | /* |
678 | * We use a "recursive" mutex so that we | |
679 | * can call dprintf_ds() with ds_lock held. | |
680 | */ | |
34dc7c2f | 681 | if (!MUTEX_HELD(&ds->ds_lock)) { |
34dc7c2f BB |
682 | mutex_enter(&ds->ds_lock); |
683 | (void) strcat(name, ds->ds_snapname); | |
684 | mutex_exit(&ds->ds_lock); | |
685 | } else { | |
686 | (void) strcat(name, ds->ds_snapname); | |
687 | } | |
688 | } | |
689 | } | |
690 | } | |
691 | ||
692 | static int | |
693 | dsl_dataset_namelen(dsl_dataset_t *ds) | |
694 | { | |
695 | int result; | |
696 | ||
697 | if (ds == NULL) { | |
698 | result = 3; /* "mos" */ | |
699 | } else { | |
700 | result = dsl_dir_namelen(ds->ds_dir); | |
701 | VERIFY(0 == dsl_dataset_get_snapname(ds)); | |
702 | if (ds->ds_snapname[0]) { | |
703 | ++result; /* adding one for the @-sign */ | |
704 | if (!MUTEX_HELD(&ds->ds_lock)) { | |
34dc7c2f BB |
705 | mutex_enter(&ds->ds_lock); |
706 | result += strlen(ds->ds_snapname); | |
707 | mutex_exit(&ds->ds_lock); | |
708 | } else { | |
709 | result += strlen(ds->ds_snapname); | |
710 | } | |
711 | } | |
712 | } | |
713 | ||
714 | return (result); | |
715 | } | |
716 | ||
717 | void | |
b128c09f | 718 | dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag) |
34dc7c2f | 719 | { |
34dc7c2f BB |
720 | dmu_buf_rele(ds->ds_dbuf, tag); |
721 | } | |
722 | ||
723 | void | |
b128c09f BB |
724 | dsl_dataset_rele(dsl_dataset_t *ds, void *tag) |
725 | { | |
726 | if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) { | |
727 | rw_exit(&ds->ds_rwlock); | |
728 | } | |
729 | dsl_dataset_drop_ref(ds, tag); | |
730 | } | |
731 | ||
732 | void | |
428870ff | 733 | dsl_dataset_disown(dsl_dataset_t *ds, void *tag) |
34dc7c2f | 734 | { |
428870ff | 735 | ASSERT((ds->ds_owner == tag && ds->ds_dbuf) || |
b128c09f BB |
736 | (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL)); |
737 | ||
34dc7c2f | 738 | mutex_enter(&ds->ds_lock); |
b128c09f BB |
739 | ds->ds_owner = NULL; |
740 | if (RW_WRITE_HELD(&ds->ds_rwlock)) { | |
741 | rw_exit(&ds->ds_rwlock); | |
742 | cv_broadcast(&ds->ds_exclusive_cv); | |
743 | } | |
34dc7c2f | 744 | mutex_exit(&ds->ds_lock); |
b128c09f | 745 | if (ds->ds_dbuf) |
428870ff | 746 | dsl_dataset_drop_ref(ds, tag); |
b128c09f | 747 | else |
428870ff | 748 | dsl_dataset_evict(NULL, ds); |
34dc7c2f BB |
749 | } |
750 | ||
751 | boolean_t | |
428870ff | 752 | dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag) |
34dc7c2f | 753 | { |
b128c09f BB |
754 | boolean_t gotit = FALSE; |
755 | ||
34dc7c2f | 756 | mutex_enter(&ds->ds_lock); |
b128c09f BB |
757 | if (ds->ds_owner == NULL && |
758 | (!DS_IS_INCONSISTENT(ds) || inconsistentok)) { | |
428870ff | 759 | ds->ds_owner = tag; |
b128c09f BB |
760 | if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) |
761 | rw_exit(&ds->ds_rwlock); | |
762 | gotit = TRUE; | |
34dc7c2f BB |
763 | } |
764 | mutex_exit(&ds->ds_lock); | |
b128c09f | 765 | return (gotit); |
34dc7c2f BB |
766 | } |
767 | ||
768 | void | |
b128c09f | 769 | dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner) |
34dc7c2f | 770 | { |
b128c09f BB |
771 | ASSERT3P(owner, ==, ds->ds_owner); |
772 | if (!RW_WRITE_HELD(&ds->ds_rwlock)) | |
773 | rw_enter(&ds->ds_rwlock, RW_WRITER); | |
34dc7c2f BB |
774 | } |
775 | ||
776 | uint64_t | |
b128c09f | 777 | dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin, |
34dc7c2f BB |
778 | uint64_t flags, dmu_tx_t *tx) |
779 | { | |
780 | dsl_pool_t *dp = dd->dd_pool; | |
781 | dmu_buf_t *dbuf; | |
782 | dsl_dataset_phys_t *dsphys; | |
783 | uint64_t dsobj; | |
784 | objset_t *mos = dp->dp_meta_objset; | |
785 | ||
b128c09f BB |
786 | if (origin == NULL) |
787 | origin = dp->dp_origin_snap; | |
788 | ||
34dc7c2f BB |
789 | ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp); |
790 | ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0); | |
791 | ASSERT(dmu_tx_is_syncing(tx)); | |
792 | ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); | |
793 | ||
794 | dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, | |
795 | DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); | |
796 | VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); | |
797 | dmu_buf_will_dirty(dbuf, tx); | |
798 | dsphys = dbuf->db_data; | |
b128c09f | 799 | bzero(dsphys, sizeof (dsl_dataset_phys_t)); |
34dc7c2f BB |
800 | dsphys->ds_dir_obj = dd->dd_object; |
801 | dsphys->ds_flags = flags; | |
802 | dsphys->ds_fsid_guid = unique_create(); | |
803 | (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
804 | sizeof (dsphys->ds_guid)); | |
805 | dsphys->ds_snapnames_zapobj = | |
806 | zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP, | |
807 | DMU_OT_NONE, 0, tx); | |
808 | dsphys->ds_creation_time = gethrestime_sec(); | |
b128c09f | 809 | dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg; |
34dc7c2f | 810 | |
428870ff BB |
811 | if (origin == NULL) { |
812 | dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx); | |
813 | } else { | |
814 | dsl_dataset_t *ohds; | |
815 | ||
34dc7c2f BB |
816 | dsphys->ds_prev_snap_obj = origin->ds_object; |
817 | dsphys->ds_prev_snap_txg = | |
818 | origin->ds_phys->ds_creation_txg; | |
819 | dsphys->ds_used_bytes = | |
820 | origin->ds_phys->ds_used_bytes; | |
821 | dsphys->ds_compressed_bytes = | |
822 | origin->ds_phys->ds_compressed_bytes; | |
823 | dsphys->ds_uncompressed_bytes = | |
824 | origin->ds_phys->ds_uncompressed_bytes; | |
825 | dsphys->ds_bp = origin->ds_phys->ds_bp; | |
826 | dsphys->ds_flags |= origin->ds_phys->ds_flags; | |
827 | ||
828 | dmu_buf_will_dirty(origin->ds_dbuf, tx); | |
829 | origin->ds_phys->ds_num_children++; | |
830 | ||
428870ff BB |
831 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, |
832 | origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds)); | |
833 | dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist, | |
834 | dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx); | |
835 | dsl_dataset_rele(ohds, FTAG); | |
836 | ||
b128c09f BB |
837 | if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) { |
838 | if (origin->ds_phys->ds_next_clones_obj == 0) { | |
839 | origin->ds_phys->ds_next_clones_obj = | |
840 | zap_create(mos, | |
841 | DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx); | |
842 | } | |
843 | VERIFY(0 == zap_add_int(mos, | |
844 | origin->ds_phys->ds_next_clones_obj, | |
845 | dsobj, tx)); | |
846 | } | |
847 | ||
34dc7c2f BB |
848 | dmu_buf_will_dirty(dd->dd_dbuf, tx); |
849 | dd->dd_phys->dd_origin_obj = origin->ds_object; | |
428870ff BB |
850 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { |
851 | if (origin->ds_dir->dd_phys->dd_clones == 0) { | |
852 | dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx); | |
853 | origin->ds_dir->dd_phys->dd_clones = | |
854 | zap_create(mos, | |
855 | DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); | |
856 | } | |
857 | VERIFY3U(0, ==, zap_add_int(mos, | |
858 | origin->ds_dir->dd_phys->dd_clones, dsobj, tx)); | |
859 | } | |
34dc7c2f BB |
860 | } |
861 | ||
862 | if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) | |
863 | dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; | |
864 | ||
865 | dmu_buf_rele(dbuf, FTAG); | |
866 | ||
867 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
868 | dd->dd_phys->dd_head_dataset_obj = dsobj; | |
869 | ||
870 | return (dsobj); | |
871 | } | |
872 | ||
873 | uint64_t | |
874 | dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname, | |
875 | dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx) | |
876 | { | |
877 | dsl_pool_t *dp = pdd->dd_pool; | |
878 | uint64_t dsobj, ddobj; | |
879 | dsl_dir_t *dd; | |
880 | ||
881 | ASSERT(lastname[0] != '@'); | |
882 | ||
b128c09f | 883 | ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx); |
34dc7c2f BB |
884 | VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd)); |
885 | ||
b128c09f | 886 | dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx); |
34dc7c2f BB |
887 | |
888 | dsl_deleg_set_create_perms(dd, tx, cr); | |
889 | ||
890 | dsl_dir_close(dd, FTAG); | |
891 | ||
572e2857 BB |
892 | /* |
893 | * If we are creating a clone, make sure we zero out any stale | |
894 | * data from the origin snapshots zil header. | |
895 | */ | |
896 | if (origin != NULL) { | |
897 | dsl_dataset_t *ds; | |
898 | objset_t *os; | |
899 | ||
900 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds)); | |
901 | VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os)); | |
902 | bzero(&os->os_zil_header, sizeof (os->os_zil_header)); | |
903 | dsl_dataset_dirty(ds, tx); | |
904 | dsl_dataset_rele(ds, FTAG); | |
905 | } | |
906 | ||
34dc7c2f BB |
907 | return (dsobj); |
908 | } | |
909 | ||
910 | struct destroyarg { | |
911 | dsl_sync_task_group_t *dstg; | |
912 | char *snapname; | |
913 | char *failed; | |
45d1cae3 | 914 | boolean_t defer; |
34dc7c2f BB |
915 | }; |
916 | ||
917 | static int | |
428870ff | 918 | dsl_snapshot_destroy_one(const char *name, void *arg) |
34dc7c2f BB |
919 | { |
920 | struct destroyarg *da = arg; | |
921 | dsl_dataset_t *ds; | |
34dc7c2f | 922 | int err; |
45d1cae3 | 923 | char *dsname; |
428870ff BB |
924 | |
925 | dsname = kmem_asprintf("%s@%s", name, da->snapname); | |
926 | err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds); | |
927 | strfree(dsname); | |
b128c09f | 928 | if (err == 0) { |
45d1cae3 BB |
929 | struct dsl_ds_destroyarg *dsda; |
930 | ||
b128c09f | 931 | dsl_dataset_make_exclusive(ds, da->dstg); |
45d1cae3 BB |
932 | dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP); |
933 | dsda->ds = ds; | |
934 | dsda->defer = da->defer; | |
b128c09f | 935 | dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check, |
45d1cae3 | 936 | dsl_dataset_destroy_sync, dsda, da->dstg, 0); |
b128c09f BB |
937 | } else if (err == ENOENT) { |
938 | err = 0; | |
939 | } else { | |
34dc7c2f | 940 | (void) strcpy(da->failed, name); |
34dc7c2f | 941 | } |
b128c09f | 942 | return (err); |
34dc7c2f BB |
943 | } |
944 | ||
945 | /* | |
946 | * Destroy 'snapname' in all descendants of 'fsname'. | |
947 | */ | |
948 | #pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy | |
949 | int | |
45d1cae3 | 950 | dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer) |
34dc7c2f BB |
951 | { |
952 | int err; | |
953 | struct destroyarg da; | |
954 | dsl_sync_task_t *dst; | |
955 | spa_t *spa; | |
956 | ||
957 | err = spa_open(fsname, &spa, FTAG); | |
958 | if (err) | |
959 | return (err); | |
960 | da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
961 | da.snapname = snapname; | |
962 | da.failed = fsname; | |
45d1cae3 | 963 | da.defer = defer; |
34dc7c2f BB |
964 | |
965 | err = dmu_objset_find(fsname, | |
966 | dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN); | |
967 | ||
968 | if (err == 0) | |
969 | err = dsl_sync_task_group_wait(da.dstg); | |
970 | ||
971 | for (dst = list_head(&da.dstg->dstg_tasks); dst; | |
972 | dst = list_next(&da.dstg->dstg_tasks, dst)) { | |
45d1cae3 BB |
973 | struct dsl_ds_destroyarg *dsda = dst->dst_arg1; |
974 | dsl_dataset_t *ds = dsda->ds; | |
975 | ||
b128c09f BB |
976 | /* |
977 | * Return the file system name that triggered the error | |
978 | */ | |
34dc7c2f BB |
979 | if (dst->dst_err) { |
980 | dsl_dataset_name(ds, fsname); | |
981 | *strchr(fsname, '@') = '\0'; | |
982 | } | |
45d1cae3 | 983 | ASSERT3P(dsda->rm_origin, ==, NULL); |
b128c09f | 984 | dsl_dataset_disown(ds, da.dstg); |
45d1cae3 | 985 | kmem_free(dsda, sizeof (struct dsl_ds_destroyarg)); |
34dc7c2f BB |
986 | } |
987 | ||
988 | dsl_sync_task_group_destroy(da.dstg); | |
989 | spa_close(spa, FTAG); | |
990 | return (err); | |
991 | } | |
992 | ||
45d1cae3 BB |
993 | static boolean_t |
994 | dsl_dataset_might_destroy_origin(dsl_dataset_t *ds) | |
995 | { | |
996 | boolean_t might_destroy = B_FALSE; | |
997 | ||
998 | mutex_enter(&ds->ds_lock); | |
999 | if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 && | |
1000 | DS_IS_DEFER_DESTROY(ds)) | |
1001 | might_destroy = B_TRUE; | |
1002 | mutex_exit(&ds->ds_lock); | |
1003 | ||
1004 | return (might_destroy); | |
1005 | } | |
1006 | ||
45d1cae3 BB |
1007 | /* |
1008 | * If we're removing a clone, and these three conditions are true: | |
1009 | * 1) the clone's origin has no other children | |
1010 | * 2) the clone's origin has no user references | |
1011 | * 3) the clone's origin has been marked for deferred destruction | |
1012 | * Then, prepare to remove the origin as part of this sync task group. | |
1013 | */ | |
1014 | static int | |
1015 | dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag) | |
1016 | { | |
1017 | dsl_dataset_t *ds = dsda->ds; | |
1018 | dsl_dataset_t *origin = ds->ds_prev; | |
1019 | ||
1020 | if (dsl_dataset_might_destroy_origin(origin)) { | |
1021 | char *name; | |
1022 | int namelen; | |
1023 | int error; | |
1024 | ||
1025 | namelen = dsl_dataset_namelen(origin) + 1; | |
1026 | name = kmem_alloc(namelen, KM_SLEEP); | |
1027 | dsl_dataset_name(origin, name); | |
1028 | #ifdef _KERNEL | |
1029 | error = zfs_unmount_snap(name, NULL); | |
1030 | if (error) { | |
1031 | kmem_free(name, namelen); | |
1032 | return (error); | |
1033 | } | |
45d1cae3 | 1034 | #endif |
428870ff | 1035 | error = dsl_dataset_own(name, B_TRUE, tag, &origin); |
45d1cae3 BB |
1036 | kmem_free(name, namelen); |
1037 | if (error) | |
1038 | return (error); | |
1039 | dsda->rm_origin = origin; | |
1040 | dsl_dataset_make_exclusive(origin, tag); | |
1041 | } | |
1042 | ||
1043 | return (0); | |
1044 | } | |
1045 | ||
34dc7c2f | 1046 | /* |
b128c09f BB |
1047 | * ds must be opened as OWNER. On return (whether successful or not), |
1048 | * ds will be closed and caller can no longer dereference it. | |
34dc7c2f BB |
1049 | */ |
1050 | int | |
45d1cae3 | 1051 | dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer) |
34dc7c2f BB |
1052 | { |
1053 | int err; | |
1054 | dsl_sync_task_group_t *dstg; | |
1055 | objset_t *os; | |
1056 | dsl_dir_t *dd; | |
1057 | uint64_t obj; | |
428870ff | 1058 | struct dsl_ds_destroyarg dsda = { 0 }; |
81a49663 | 1059 | dsl_dataset_t *dummy_ds; |
45d1cae3 BB |
1060 | |
1061 | dsda.ds = ds; | |
34dc7c2f | 1062 | |
34dc7c2f BB |
1063 | if (dsl_dataset_is_snapshot(ds)) { |
1064 | /* Destroying a snapshot is simpler */ | |
b128c09f BB |
1065 | dsl_dataset_make_exclusive(ds, tag); |
1066 | ||
45d1cae3 | 1067 | dsda.defer = defer; |
34dc7c2f BB |
1068 | err = dsl_sync_task_do(ds->ds_dir->dd_pool, |
1069 | dsl_dataset_destroy_check, dsl_dataset_destroy_sync, | |
45d1cae3 BB |
1070 | &dsda, tag, 0); |
1071 | ASSERT3P(dsda.rm_origin, ==, NULL); | |
34dc7c2f | 1072 | goto out; |
428870ff BB |
1073 | } else if (defer) { |
1074 | err = EINVAL; | |
1075 | goto out; | |
34dc7c2f BB |
1076 | } |
1077 | ||
1078 | dd = ds->ds_dir; | |
81a49663 BB |
1079 | dummy_ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP); |
1080 | dummy_ds->ds_dir = dd; | |
1081 | dummy_ds->ds_object = ds->ds_object; | |
34dc7c2f BB |
1082 | |
1083 | /* | |
1084 | * Check for errors and mark this ds as inconsistent, in | |
1085 | * case we crash while freeing the objects. | |
1086 | */ | |
1087 | err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check, | |
1088 | dsl_dataset_destroy_begin_sync, ds, NULL, 0); | |
1089 | if (err) | |
81a49663 | 1090 | goto out_free; |
34dc7c2f | 1091 | |
428870ff | 1092 | err = dmu_objset_from_ds(ds, &os); |
34dc7c2f | 1093 | if (err) |
81a49663 | 1094 | goto out_free; |
34dc7c2f BB |
1095 | |
1096 | /* | |
1097 | * remove the objects in open context, so that we won't | |
1098 | * have too much to do in syncing context. | |
1099 | */ | |
1100 | for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, | |
1101 | ds->ds_phys->ds_prev_snap_txg)) { | |
b128c09f BB |
1102 | /* |
1103 | * Ignore errors, if there is not enough disk space | |
1104 | * we will deal with it in dsl_dataset_destroy_sync(). | |
1105 | */ | |
1106 | (void) dmu_free_object(os, obj); | |
34dc7c2f | 1107 | } |
572e2857 | 1108 | if (err != ESRCH) |
81a49663 | 1109 | goto out_free; |
34dc7c2f | 1110 | |
9babb374 | 1111 | /* |
572e2857 BB |
1112 | * Only the ZIL knows how to free log blocks. |
1113 | */ | |
1114 | zil_destroy(dmu_objset_zil(os), B_FALSE); | |
1115 | ||
1116 | /* | |
1117 | * Sync out all in-flight IO. | |
9babb374 BB |
1118 | */ |
1119 | txg_wait_synced(dd->dd_pool, 0); | |
1120 | ||
1121 | /* | |
1122 | * If we managed to free all the objects in open | |
1123 | * context, the user space accounting should be zero. | |
1124 | */ | |
1125 | if (ds->ds_phys->ds_bp.blk_fill == 0 && | |
428870ff | 1126 | dmu_objset_userused_enabled(os)) { |
1fde1e37 | 1127 | ASSERTV(uint64_t count); |
9babb374 BB |
1128 | ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 || |
1129 | count == 0); | |
1130 | ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 || | |
1131 | count == 0); | |
1132 | } | |
1133 | ||
34dc7c2f BB |
1134 | rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); |
1135 | err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd); | |
1136 | rw_exit(&dd->dd_pool->dp_config_rwlock); | |
1137 | ||
1138 | if (err) | |
81a49663 | 1139 | goto out_free; |
34dc7c2f BB |
1140 | |
1141 | /* | |
1142 | * Blow away the dsl_dir + head dataset. | |
1143 | */ | |
b128c09f | 1144 | dsl_dataset_make_exclusive(ds, tag); |
45d1cae3 BB |
1145 | /* |
1146 | * If we're removing a clone, we might also need to remove its | |
1147 | * origin. | |
1148 | */ | |
1149 | do { | |
1150 | dsda.need_prep = B_FALSE; | |
1151 | if (dsl_dir_is_clone(dd)) { | |
1152 | err = dsl_dataset_origin_rm_prep(&dsda, tag); | |
1153 | if (err) { | |
1154 | dsl_dir_close(dd, FTAG); | |
81a49663 | 1155 | goto out_free; |
45d1cae3 BB |
1156 | } |
1157 | } | |
1158 | ||
1159 | dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool); | |
1160 | dsl_sync_task_create(dstg, dsl_dataset_destroy_check, | |
1161 | dsl_dataset_destroy_sync, &dsda, tag, 0); | |
1162 | dsl_sync_task_create(dstg, dsl_dir_destroy_check, | |
81a49663 | 1163 | dsl_dir_destroy_sync, dummy_ds, FTAG, 0); |
45d1cae3 BB |
1164 | err = dsl_sync_task_group_wait(dstg); |
1165 | dsl_sync_task_group_destroy(dstg); | |
1166 | ||
1167 | /* | |
1168 | * We could be racing against 'zfs release' or 'zfs destroy -d' | |
1169 | * on the origin snap, in which case we can get EBUSY if we | |
1170 | * needed to destroy the origin snap but were not ready to | |
1171 | * do so. | |
1172 | */ | |
1173 | if (dsda.need_prep) { | |
1174 | ASSERT(err == EBUSY); | |
1175 | ASSERT(dsl_dir_is_clone(dd)); | |
1176 | ASSERT(dsda.rm_origin == NULL); | |
1177 | } | |
1178 | } while (dsda.need_prep); | |
1179 | ||
1180 | if (dsda.rm_origin != NULL) | |
1181 | dsl_dataset_disown(dsda.rm_origin, tag); | |
1182 | ||
b128c09f | 1183 | /* if it is successful, dsl_dir_destroy_sync will close the dd */ |
34dc7c2f BB |
1184 | if (err) |
1185 | dsl_dir_close(dd, FTAG); | |
81a49663 BB |
1186 | |
1187 | out_free: | |
1188 | kmem_free(dummy_ds, sizeof (dsl_dataset_t)); | |
34dc7c2f | 1189 | out: |
b128c09f | 1190 | dsl_dataset_disown(ds, tag); |
34dc7c2f BB |
1191 | return (err); |
1192 | } | |
1193 | ||
34dc7c2f BB |
1194 | blkptr_t * |
1195 | dsl_dataset_get_blkptr(dsl_dataset_t *ds) | |
1196 | { | |
1197 | return (&ds->ds_phys->ds_bp); | |
1198 | } | |
1199 | ||
1200 | void | |
1201 | dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx) | |
1202 | { | |
1203 | ASSERT(dmu_tx_is_syncing(tx)); | |
1204 | /* If it's the meta-objset, set dp_meta_rootbp */ | |
1205 | if (ds == NULL) { | |
1206 | tx->tx_pool->dp_meta_rootbp = *bp; | |
1207 | } else { | |
1208 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
1209 | ds->ds_phys->ds_bp = *bp; | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | spa_t * | |
1214 | dsl_dataset_get_spa(dsl_dataset_t *ds) | |
1215 | { | |
1216 | return (ds->ds_dir->dd_pool->dp_spa); | |
1217 | } | |
1218 | ||
1219 | void | |
1220 | dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1221 | { | |
1222 | dsl_pool_t *dp; | |
1223 | ||
1224 | if (ds == NULL) /* this is the meta-objset */ | |
1225 | return; | |
1226 | ||
428870ff | 1227 | ASSERT(ds->ds_objset != NULL); |
34dc7c2f BB |
1228 | |
1229 | if (ds->ds_phys->ds_next_snap_obj != 0) | |
1230 | panic("dirtying snapshot!"); | |
1231 | ||
1232 | dp = ds->ds_dir->dd_pool; | |
1233 | ||
1234 | if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) { | |
1235 | /* up the hold count until we can be written out */ | |
1236 | dmu_buf_add_ref(ds->ds_dbuf, ds); | |
1237 | } | |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * The unique space in the head dataset can be calculated by subtracting | |
1242 | * the space used in the most recent snapshot, that is still being used | |
1243 | * in this file system, from the space currently in use. To figure out | |
1244 | * the space in the most recent snapshot still in use, we need to take | |
1245 | * the total space used in the snapshot and subtract out the space that | |
1246 | * has been freed up since the snapshot was taken. | |
1247 | */ | |
1248 | static void | |
1249 | dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds) | |
1250 | { | |
1251 | uint64_t mrs_used; | |
1252 | uint64_t dlused, dlcomp, dluncomp; | |
1253 | ||
428870ff | 1254 | ASSERT(!dsl_dataset_is_snapshot(ds)); |
34dc7c2f BB |
1255 | |
1256 | if (ds->ds_phys->ds_prev_snap_obj != 0) | |
1257 | mrs_used = ds->ds_prev->ds_phys->ds_used_bytes; | |
1258 | else | |
1259 | mrs_used = 0; | |
1260 | ||
428870ff | 1261 | dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp); |
34dc7c2f BB |
1262 | |
1263 | ASSERT3U(dlused, <=, mrs_used); | |
1264 | ds->ds_phys->ds_unique_bytes = | |
1265 | ds->ds_phys->ds_used_bytes - (mrs_used - dlused); | |
1266 | ||
428870ff | 1267 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) >= |
34dc7c2f BB |
1268 | SPA_VERSION_UNIQUE_ACCURATE) |
1269 | ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; | |
1270 | } | |
1271 | ||
34dc7c2f | 1272 | struct killarg { |
b128c09f | 1273 | dsl_dataset_t *ds; |
34dc7c2f BB |
1274 | dmu_tx_t *tx; |
1275 | }; | |
1276 | ||
b128c09f | 1277 | /* ARGSUSED */ |
34dc7c2f | 1278 | static int |
428870ff BB |
1279 | kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf, |
1280 | const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg) | |
34dc7c2f BB |
1281 | { |
1282 | struct killarg *ka = arg; | |
428870ff | 1283 | dmu_tx_t *tx = ka->tx; |
34dc7c2f | 1284 | |
b128c09f BB |
1285 | if (bp == NULL) |
1286 | return (0); | |
1287 | ||
428870ff BB |
1288 | if (zb->zb_level == ZB_ZIL_LEVEL) { |
1289 | ASSERT(zilog != NULL); | |
9babb374 BB |
1290 | /* |
1291 | * It's a block in the intent log. It has no | |
1292 | * accounting, so just free it. | |
1293 | */ | |
428870ff | 1294 | dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp); |
9babb374 | 1295 | } else { |
428870ff | 1296 | ASSERT(zilog == NULL); |
9babb374 | 1297 | ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg); |
428870ff | 1298 | (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE); |
9babb374 | 1299 | } |
34dc7c2f | 1300 | |
34dc7c2f BB |
1301 | return (0); |
1302 | } | |
1303 | ||
34dc7c2f BB |
1304 | /* ARGSUSED */ |
1305 | static int | |
1306 | dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
1307 | { | |
1308 | dsl_dataset_t *ds = arg1; | |
1309 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
1310 | uint64_t count; | |
1311 | int err; | |
1312 | ||
1313 | /* | |
1314 | * Can't delete a head dataset if there are snapshots of it. | |
1315 | * (Except if the only snapshots are from the branch we cloned | |
1316 | * from.) | |
1317 | */ | |
1318 | if (ds->ds_prev != NULL && | |
1319 | ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) | |
428870ff | 1320 | return (EBUSY); |
34dc7c2f BB |
1321 | |
1322 | /* | |
1323 | * This is really a dsl_dir thing, but check it here so that | |
1324 | * we'll be less likely to leave this dataset inconsistent & | |
1325 | * nearly destroyed. | |
1326 | */ | |
1327 | err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count); | |
1328 | if (err) | |
1329 | return (err); | |
1330 | if (count != 0) | |
1331 | return (EEXIST); | |
1332 | ||
1333 | return (0); | |
1334 | } | |
1335 | ||
1336 | /* ARGSUSED */ | |
1337 | static void | |
428870ff | 1338 | dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
1339 | { |
1340 | dsl_dataset_t *ds = arg1; | |
1341 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
1342 | ||
1343 | /* Mark it as inconsistent on-disk, in case we crash */ | |
1344 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
1345 | ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT; | |
1346 | ||
428870ff BB |
1347 | spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx, |
1348 | "dataset = %llu", ds->ds_object); | |
34dc7c2f BB |
1349 | } |
1350 | ||
45d1cae3 BB |
1351 | static int |
1352 | dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag, | |
1353 | dmu_tx_t *tx) | |
1354 | { | |
1355 | dsl_dataset_t *ds = dsda->ds; | |
1356 | dsl_dataset_t *ds_prev = ds->ds_prev; | |
1357 | ||
1358 | if (dsl_dataset_might_destroy_origin(ds_prev)) { | |
1359 | struct dsl_ds_destroyarg ndsda = {0}; | |
1360 | ||
1361 | /* | |
1362 | * If we're not prepared to remove the origin, don't remove | |
1363 | * the clone either. | |
1364 | */ | |
1365 | if (dsda->rm_origin == NULL) { | |
1366 | dsda->need_prep = B_TRUE; | |
1367 | return (EBUSY); | |
1368 | } | |
1369 | ||
1370 | ndsda.ds = ds_prev; | |
1371 | ndsda.is_origin_rm = B_TRUE; | |
1372 | return (dsl_dataset_destroy_check(&ndsda, tag, tx)); | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * If we're not going to remove the origin after all, | |
1377 | * undo the open context setup. | |
1378 | */ | |
1379 | if (dsda->rm_origin != NULL) { | |
1380 | dsl_dataset_disown(dsda->rm_origin, tag); | |
1381 | dsda->rm_origin = NULL; | |
1382 | } | |
1383 | ||
1384 | return (0); | |
1385 | } | |
1386 | ||
572e2857 BB |
1387 | /* |
1388 | * If you add new checks here, you may need to add | |
1389 | * additional checks to the "temporary" case in | |
1390 | * snapshot_check() in dmu_objset.c. | |
1391 | */ | |
34dc7c2f BB |
1392 | /* ARGSUSED */ |
1393 | int | |
1394 | dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
1395 | { | |
45d1cae3 BB |
1396 | struct dsl_ds_destroyarg *dsda = arg1; |
1397 | dsl_dataset_t *ds = dsda->ds; | |
34dc7c2f | 1398 | |
b128c09f BB |
1399 | /* we have an owner hold, so noone else can destroy us */ |
1400 | ASSERT(!DSL_DATASET_IS_DESTROYED(ds)); | |
1401 | ||
45d1cae3 BB |
1402 | /* |
1403 | * Only allow deferred destroy on pools that support it. | |
1404 | * NOTE: deferred destroy is only supported on snapshots. | |
1405 | */ | |
1406 | if (dsda->defer) { | |
1407 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) < | |
1408 | SPA_VERSION_USERREFS) | |
1409 | return (ENOTSUP); | |
1410 | ASSERT(dsl_dataset_is_snapshot(ds)); | |
1411 | return (0); | |
1412 | } | |
34dc7c2f BB |
1413 | |
1414 | /* | |
1415 | * Can't delete a head dataset if there are snapshots of it. | |
1416 | * (Except if the only snapshots are from the branch we cloned | |
1417 | * from.) | |
1418 | */ | |
1419 | if (ds->ds_prev != NULL && | |
1420 | ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) | |
428870ff | 1421 | return (EBUSY); |
34dc7c2f BB |
1422 | |
1423 | /* | |
1424 | * If we made changes this txg, traverse_dsl_dataset won't find | |
1425 | * them. Try again. | |
1426 | */ | |
1427 | if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg) | |
1428 | return (EAGAIN); | |
1429 | ||
45d1cae3 BB |
1430 | if (dsl_dataset_is_snapshot(ds)) { |
1431 | /* | |
1432 | * If this snapshot has an elevated user reference count, | |
1433 | * we can't destroy it yet. | |
1434 | */ | |
1435 | if (ds->ds_userrefs > 0 && !dsda->releasing) | |
1436 | return (EBUSY); | |
1437 | ||
1438 | mutex_enter(&ds->ds_lock); | |
1439 | /* | |
1440 | * Can't delete a branch point. However, if we're destroying | |
1441 | * a clone and removing its origin due to it having a user | |
1442 | * hold count of 0 and having been marked for deferred destroy, | |
1443 | * it's OK for the origin to have a single clone. | |
1444 | */ | |
1445 | if (ds->ds_phys->ds_num_children > | |
1446 | (dsda->is_origin_rm ? 2 : 1)) { | |
1447 | mutex_exit(&ds->ds_lock); | |
1448 | return (EEXIST); | |
1449 | } | |
1450 | mutex_exit(&ds->ds_lock); | |
1451 | } else if (dsl_dir_is_clone(ds->ds_dir)) { | |
1452 | return (dsl_dataset_origin_check(dsda, arg2, tx)); | |
1453 | } | |
1454 | ||
34dc7c2f BB |
1455 | /* XXX we should do some i/o error checking... */ |
1456 | return (0); | |
1457 | } | |
1458 | ||
b128c09f BB |
1459 | struct refsarg { |
1460 | kmutex_t lock; | |
1461 | boolean_t gone; | |
1462 | kcondvar_t cv; | |
1463 | }; | |
1464 | ||
1465 | /* ARGSUSED */ | |
1466 | static void | |
1467 | dsl_dataset_refs_gone(dmu_buf_t *db, void *argv) | |
1468 | { | |
1469 | struct refsarg *arg = argv; | |
1470 | ||
1471 | mutex_enter(&arg->lock); | |
1472 | arg->gone = TRUE; | |
1473 | cv_signal(&arg->cv); | |
1474 | mutex_exit(&arg->lock); | |
1475 | } | |
1476 | ||
1477 | static void | |
1478 | dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag) | |
1479 | { | |
1480 | struct refsarg arg; | |
1481 | ||
1482 | mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL); | |
1483 | cv_init(&arg.cv, NULL, CV_DEFAULT, NULL); | |
1484 | arg.gone = FALSE; | |
1485 | (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys, | |
1486 | dsl_dataset_refs_gone); | |
1487 | dmu_buf_rele(ds->ds_dbuf, tag); | |
1488 | mutex_enter(&arg.lock); | |
1489 | while (!arg.gone) | |
1490 | cv_wait(&arg.cv, &arg.lock); | |
1491 | ASSERT(arg.gone); | |
1492 | mutex_exit(&arg.lock); | |
1493 | ds->ds_dbuf = NULL; | |
1494 | ds->ds_phys = NULL; | |
1495 | mutex_destroy(&arg.lock); | |
1496 | cv_destroy(&arg.cv); | |
1497 | } | |
1498 | ||
428870ff BB |
1499 | static void |
1500 | remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx) | |
1501 | { | |
1502 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
428870ff | 1503 | int err; |
1fde1e37 | 1504 | ASSERTV(uint64_t count); |
428870ff BB |
1505 | |
1506 | ASSERT(ds->ds_phys->ds_num_children >= 2); | |
1507 | err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx); | |
1508 | /* | |
1509 | * The err should not be ENOENT, but a bug in a previous version | |
1510 | * of the code could cause upgrade_clones_cb() to not set | |
1511 | * ds_next_snap_obj when it should, leading to a missing entry. | |
1512 | * If we knew that the pool was created after | |
1513 | * SPA_VERSION_NEXT_CLONES, we could assert that it isn't | |
1514 | * ENOENT. However, at least we can check that we don't have | |
1515 | * too many entries in the next_clones_obj even after failing to | |
1516 | * remove this one. | |
1517 | */ | |
1518 | if (err != ENOENT) { | |
1519 | VERIFY3U(err, ==, 0); | |
1520 | } | |
1521 | ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj, | |
1522 | &count)); | |
1523 | ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2); | |
1524 | } | |
1525 | ||
1526 | static void | |
1527 | dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx) | |
1528 | { | |
1529 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
1530 | zap_cursor_t zc; | |
1531 | zap_attribute_t za; | |
1532 | ||
1533 | /* | |
1534 | * If it is the old version, dd_clones doesn't exist so we can't | |
1535 | * find the clones, but deadlist_remove_key() is a no-op so it | |
1536 | * doesn't matter. | |
1537 | */ | |
1538 | if (ds->ds_dir->dd_phys->dd_clones == 0) | |
1539 | return; | |
1540 | ||
1541 | for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones); | |
1542 | zap_cursor_retrieve(&zc, &za) == 0; | |
1543 | zap_cursor_advance(&zc)) { | |
1544 | dsl_dataset_t *clone; | |
1545 | ||
1546 | VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool, | |
1547 | za.za_first_integer, FTAG, &clone)); | |
1548 | if (clone->ds_dir->dd_origin_txg > mintxg) { | |
1549 | dsl_deadlist_remove_key(&clone->ds_deadlist, | |
1550 | mintxg, tx); | |
1551 | dsl_dataset_remove_clones_key(clone, mintxg, tx); | |
1552 | } | |
1553 | dsl_dataset_rele(clone, FTAG); | |
1554 | } | |
1555 | zap_cursor_fini(&zc); | |
1556 | } | |
1557 | ||
1558 | struct process_old_arg { | |
1559 | dsl_dataset_t *ds; | |
1560 | dsl_dataset_t *ds_prev; | |
1561 | boolean_t after_branch_point; | |
1562 | zio_t *pio; | |
1563 | uint64_t used, comp, uncomp; | |
1564 | }; | |
1565 | ||
1566 | static int | |
1567 | process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx) | |
1568 | { | |
1569 | struct process_old_arg *poa = arg; | |
1570 | dsl_pool_t *dp = poa->ds->ds_dir->dd_pool; | |
1571 | ||
1572 | if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) { | |
1573 | dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx); | |
1574 | if (poa->ds_prev && !poa->after_branch_point && | |
1575 | bp->blk_birth > | |
1576 | poa->ds_prev->ds_phys->ds_prev_snap_txg) { | |
1577 | poa->ds_prev->ds_phys->ds_unique_bytes += | |
1578 | bp_get_dsize_sync(dp->dp_spa, bp); | |
1579 | } | |
1580 | } else { | |
1581 | poa->used += bp_get_dsize_sync(dp->dp_spa, bp); | |
1582 | poa->comp += BP_GET_PSIZE(bp); | |
1583 | poa->uncomp += BP_GET_UCSIZE(bp); | |
1584 | dsl_free_sync(poa->pio, dp, tx->tx_txg, bp); | |
1585 | } | |
1586 | return (0); | |
1587 | } | |
1588 | ||
1589 | static void | |
1590 | process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev, | |
1591 | dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx) | |
1592 | { | |
1593 | struct process_old_arg poa = { 0 }; | |
1594 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
1595 | objset_t *mos = dp->dp_meta_objset; | |
1596 | ||
1597 | ASSERT(ds->ds_deadlist.dl_oldfmt); | |
1598 | ASSERT(ds_next->ds_deadlist.dl_oldfmt); | |
1599 | ||
1600 | poa.ds = ds; | |
1601 | poa.ds_prev = ds_prev; | |
1602 | poa.after_branch_point = after_branch_point; | |
1603 | poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED); | |
1604 | VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj, | |
1605 | process_old_cb, &poa, tx)); | |
1606 | VERIFY3U(zio_wait(poa.pio), ==, 0); | |
1607 | ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes); | |
1608 | ||
1609 | /* change snapused */ | |
1610 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, | |
1611 | -poa.used, -poa.comp, -poa.uncomp, tx); | |
1612 | ||
1613 | /* swap next's deadlist to our deadlist */ | |
1614 | dsl_deadlist_close(&ds->ds_deadlist); | |
1615 | dsl_deadlist_close(&ds_next->ds_deadlist); | |
1616 | SWITCH64(ds_next->ds_phys->ds_deadlist_obj, | |
1617 | ds->ds_phys->ds_deadlist_obj); | |
1618 | dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); | |
1619 | dsl_deadlist_open(&ds_next->ds_deadlist, mos, | |
1620 | ds_next->ds_phys->ds_deadlist_obj); | |
1621 | } | |
1622 | ||
34dc7c2f | 1623 | void |
428870ff | 1624 | dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) |
34dc7c2f | 1625 | { |
45d1cae3 BB |
1626 | struct dsl_ds_destroyarg *dsda = arg1; |
1627 | dsl_dataset_t *ds = dsda->ds; | |
34dc7c2f BB |
1628 | int err; |
1629 | int after_branch_point = FALSE; | |
1630 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
1631 | objset_t *mos = dp->dp_meta_objset; | |
1632 | dsl_dataset_t *ds_prev = NULL; | |
572e2857 | 1633 | boolean_t wont_destroy; |
34dc7c2f BB |
1634 | uint64_t obj; |
1635 | ||
572e2857 BB |
1636 | wont_destroy = (dsda->defer && |
1637 | (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)); | |
1638 | ||
1639 | ASSERT(ds->ds_owner || wont_destroy); | |
45d1cae3 | 1640 | ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1); |
34dc7c2f BB |
1641 | ASSERT(ds->ds_prev == NULL || |
1642 | ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object); | |
1643 | ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg); | |
1644 | ||
572e2857 | 1645 | if (wont_destroy) { |
45d1cae3 | 1646 | ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS); |
572e2857 BB |
1647 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
1648 | ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY; | |
1649 | return; | |
45d1cae3 BB |
1650 | } |
1651 | ||
b128c09f BB |
1652 | /* signal any waiters that this dataset is going away */ |
1653 | mutex_enter(&ds->ds_lock); | |
1654 | ds->ds_owner = dsl_reaper; | |
1655 | cv_broadcast(&ds->ds_exclusive_cv); | |
1656 | mutex_exit(&ds->ds_lock); | |
1657 | ||
34dc7c2f BB |
1658 | /* Remove our reservation */ |
1659 | if (ds->ds_reserved != 0) { | |
428870ff BB |
1660 | dsl_prop_setarg_t psa; |
1661 | uint64_t value = 0; | |
1662 | ||
1663 | dsl_prop_setarg_init_uint64(&psa, "refreservation", | |
1664 | (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED), | |
1665 | &value); | |
1666 | psa.psa_effective_value = 0; /* predict default value */ | |
1667 | ||
1668 | dsl_dataset_set_reservation_sync(ds, &psa, tx); | |
34dc7c2f BB |
1669 | ASSERT3U(ds->ds_reserved, ==, 0); |
1670 | } | |
1671 | ||
1672 | ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); | |
1673 | ||
428870ff | 1674 | dsl_scan_ds_destroyed(ds, tx); |
b128c09f | 1675 | |
34dc7c2f BB |
1676 | obj = ds->ds_object; |
1677 | ||
1678 | if (ds->ds_phys->ds_prev_snap_obj != 0) { | |
1679 | if (ds->ds_prev) { | |
1680 | ds_prev = ds->ds_prev; | |
1681 | } else { | |
b128c09f BB |
1682 | VERIFY(0 == dsl_dataset_hold_obj(dp, |
1683 | ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev)); | |
34dc7c2f BB |
1684 | } |
1685 | after_branch_point = | |
1686 | (ds_prev->ds_phys->ds_next_snap_obj != obj); | |
1687 | ||
1688 | dmu_buf_will_dirty(ds_prev->ds_dbuf, tx); | |
b128c09f BB |
1689 | if (after_branch_point && |
1690 | ds_prev->ds_phys->ds_next_clones_obj != 0) { | |
428870ff | 1691 | remove_from_next_clones(ds_prev, obj, tx); |
b128c09f BB |
1692 | if (ds->ds_phys->ds_next_snap_obj != 0) { |
1693 | VERIFY(0 == zap_add_int(mos, | |
1694 | ds_prev->ds_phys->ds_next_clones_obj, | |
1695 | ds->ds_phys->ds_next_snap_obj, tx)); | |
1696 | } | |
1697 | } | |
34dc7c2f BB |
1698 | if (after_branch_point && |
1699 | ds->ds_phys->ds_next_snap_obj == 0) { | |
1700 | /* This clone is toast. */ | |
1701 | ASSERT(ds_prev->ds_phys->ds_num_children > 1); | |
1702 | ds_prev->ds_phys->ds_num_children--; | |
45d1cae3 BB |
1703 | |
1704 | /* | |
1705 | * If the clone's origin has no other clones, no | |
1706 | * user holds, and has been marked for deferred | |
1707 | * deletion, then we should have done the necessary | |
1708 | * destroy setup for it. | |
1709 | */ | |
1710 | if (ds_prev->ds_phys->ds_num_children == 1 && | |
1711 | ds_prev->ds_userrefs == 0 && | |
1712 | DS_IS_DEFER_DESTROY(ds_prev)) { | |
1713 | ASSERT3P(dsda->rm_origin, !=, NULL); | |
1714 | } else { | |
1715 | ASSERT3P(dsda->rm_origin, ==, NULL); | |
1716 | } | |
34dc7c2f BB |
1717 | } else if (!after_branch_point) { |
1718 | ds_prev->ds_phys->ds_next_snap_obj = | |
1719 | ds->ds_phys->ds_next_snap_obj; | |
1720 | } | |
1721 | } | |
1722 | ||
428870ff | 1723 | if (dsl_dataset_is_snapshot(ds)) { |
34dc7c2f | 1724 | dsl_dataset_t *ds_next; |
34dc7c2f | 1725 | uint64_t old_unique; |
428870ff | 1726 | uint64_t used = 0, comp = 0, uncomp = 0; |
34dc7c2f | 1727 | |
b128c09f BB |
1728 | VERIFY(0 == dsl_dataset_hold_obj(dp, |
1729 | ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next)); | |
34dc7c2f BB |
1730 | ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj); |
1731 | ||
428870ff | 1732 | old_unique = ds_next->ds_phys->ds_unique_bytes; |
34dc7c2f BB |
1733 | |
1734 | dmu_buf_will_dirty(ds_next->ds_dbuf, tx); | |
1735 | ds_next->ds_phys->ds_prev_snap_obj = | |
1736 | ds->ds_phys->ds_prev_snap_obj; | |
1737 | ds_next->ds_phys->ds_prev_snap_txg = | |
1738 | ds->ds_phys->ds_prev_snap_txg; | |
1739 | ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
1740 | ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0); | |
1741 | ||
34dc7c2f | 1742 | |
428870ff BB |
1743 | if (ds_next->ds_deadlist.dl_oldfmt) { |
1744 | process_old_deadlist(ds, ds_prev, ds_next, | |
1745 | after_branch_point, tx); | |
1746 | } else { | |
1747 | /* Adjust prev's unique space. */ | |
1748 | if (ds_prev && !after_branch_point) { | |
1749 | dsl_deadlist_space_range(&ds_next->ds_deadlist, | |
1750 | ds_prev->ds_phys->ds_prev_snap_txg, | |
1751 | ds->ds_phys->ds_prev_snap_txg, | |
1752 | &used, &comp, &uncomp); | |
1753 | ds_prev->ds_phys->ds_unique_bytes += used; | |
1754 | } | |
b128c09f | 1755 | |
428870ff BB |
1756 | /* Adjust snapused. */ |
1757 | dsl_deadlist_space_range(&ds_next->ds_deadlist, | |
1758 | ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, | |
1759 | &used, &comp, &uncomp); | |
1760 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP, | |
1761 | -used, -comp, -uncomp, tx); | |
1762 | ||
1763 | /* Move blocks to be freed to pool's free list. */ | |
1764 | dsl_deadlist_move_bpobj(&ds_next->ds_deadlist, | |
1765 | &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg, | |
1766 | tx); | |
1767 | dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, | |
1768 | DD_USED_HEAD, used, comp, uncomp, tx); | |
1769 | dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx); | |
1770 | ||
1771 | /* Merge our deadlist into next's and free it. */ | |
1772 | dsl_deadlist_merge(&ds_next->ds_deadlist, | |
1773 | ds->ds_phys->ds_deadlist_obj, tx); | |
1774 | } | |
1775 | dsl_deadlist_close(&ds->ds_deadlist); | |
1776 | dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
b128c09f | 1777 | |
428870ff BB |
1778 | /* Collapse range in clone heads */ |
1779 | dsl_dataset_remove_clones_key(ds, | |
1780 | ds->ds_phys->ds_creation_txg, tx); | |
34dc7c2f | 1781 | |
428870ff BB |
1782 | if (dsl_dataset_is_snapshot(ds_next)) { |
1783 | dsl_dataset_t *ds_nextnext; | |
d6320ddb | 1784 | dsl_dataset_t *hds; |
34dc7c2f | 1785 | |
34dc7c2f BB |
1786 | /* |
1787 | * Update next's unique to include blocks which | |
1788 | * were previously shared by only this snapshot | |
1789 | * and it. Those blocks will be born after the | |
1790 | * prev snap and before this snap, and will have | |
1791 | * died after the next snap and before the one | |
1792 | * after that (ie. be on the snap after next's | |
1793 | * deadlist). | |
34dc7c2f | 1794 | */ |
b128c09f BB |
1795 | VERIFY(0 == dsl_dataset_hold_obj(dp, |
1796 | ds_next->ds_phys->ds_next_snap_obj, | |
428870ff BB |
1797 | FTAG, &ds_nextnext)); |
1798 | dsl_deadlist_space_range(&ds_nextnext->ds_deadlist, | |
b128c09f | 1799 | ds->ds_phys->ds_prev_snap_txg, |
428870ff BB |
1800 | ds->ds_phys->ds_creation_txg, |
1801 | &used, &comp, &uncomp); | |
1802 | ds_next->ds_phys->ds_unique_bytes += used; | |
1803 | dsl_dataset_rele(ds_nextnext, FTAG); | |
34dc7c2f | 1804 | ASSERT3P(ds_next->ds_prev, ==, NULL); |
428870ff BB |
1805 | |
1806 | /* Collapse range in this head. */ | |
428870ff BB |
1807 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, |
1808 | ds->ds_dir->dd_phys->dd_head_dataset_obj, | |
1809 | FTAG, &hds)); | |
1810 | dsl_deadlist_remove_key(&hds->ds_deadlist, | |
1811 | ds->ds_phys->ds_creation_txg, tx); | |
1812 | dsl_dataset_rele(hds, FTAG); | |
1813 | ||
34dc7c2f BB |
1814 | } else { |
1815 | ASSERT3P(ds_next->ds_prev, ==, ds); | |
b128c09f BB |
1816 | dsl_dataset_drop_ref(ds_next->ds_prev, ds_next); |
1817 | ds_next->ds_prev = NULL; | |
34dc7c2f | 1818 | if (ds_prev) { |
b128c09f BB |
1819 | VERIFY(0 == dsl_dataset_get_ref(dp, |
1820 | ds->ds_phys->ds_prev_snap_obj, | |
1821 | ds_next, &ds_next->ds_prev)); | |
34dc7c2f BB |
1822 | } |
1823 | ||
1824 | dsl_dataset_recalc_head_uniq(ds_next); | |
1825 | ||
1826 | /* | |
1827 | * Reduce the amount of our unconsmed refreservation | |
1828 | * being charged to our parent by the amount of | |
1829 | * new unique data we have gained. | |
1830 | */ | |
1831 | if (old_unique < ds_next->ds_reserved) { | |
1832 | int64_t mrsdelta; | |
1833 | uint64_t new_unique = | |
1834 | ds_next->ds_phys->ds_unique_bytes; | |
1835 | ||
1836 | ASSERT(old_unique <= new_unique); | |
1837 | mrsdelta = MIN(new_unique - old_unique, | |
1838 | ds_next->ds_reserved - old_unique); | |
b128c09f BB |
1839 | dsl_dir_diduse_space(ds->ds_dir, |
1840 | DD_USED_REFRSRV, -mrsdelta, 0, 0, tx); | |
34dc7c2f BB |
1841 | } |
1842 | } | |
b128c09f | 1843 | dsl_dataset_rele(ds_next, FTAG); |
34dc7c2f BB |
1844 | } else { |
1845 | /* | |
1846 | * There's no next snapshot, so this is a head dataset. | |
1847 | * Destroy the deadlist. Unless it's a clone, the | |
1848 | * deadlist should be empty. (If it's a clone, it's | |
1849 | * safe to ignore the deadlist contents.) | |
1850 | */ | |
1851 | struct killarg ka; | |
1852 | ||
428870ff BB |
1853 | dsl_deadlist_close(&ds->ds_deadlist); |
1854 | dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx); | |
34dc7c2f BB |
1855 | ds->ds_phys->ds_deadlist_obj = 0; |
1856 | ||
1857 | /* | |
1858 | * Free everything that we point to (that's born after | |
1859 | * the previous snapshot, if we are a clone) | |
1860 | * | |
b128c09f BB |
1861 | * NB: this should be very quick, because we already |
1862 | * freed all the objects in open context. | |
34dc7c2f | 1863 | */ |
b128c09f | 1864 | ka.ds = ds; |
34dc7c2f | 1865 | ka.tx = tx; |
b128c09f BB |
1866 | err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg, |
1867 | TRAVERSE_POST, kill_blkptr, &ka); | |
34dc7c2f | 1868 | ASSERT3U(err, ==, 0); |
9babb374 | 1869 | ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || |
b128c09f | 1870 | ds->ds_phys->ds_unique_bytes == 0); |
34dc7c2f | 1871 | |
428870ff BB |
1872 | if (ds->ds_prev != NULL) { |
1873 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { | |
1874 | VERIFY3U(0, ==, zap_remove_int(mos, | |
1875 | ds->ds_prev->ds_dir->dd_phys->dd_clones, | |
1876 | ds->ds_object, tx)); | |
1877 | } | |
1878 | dsl_dataset_rele(ds->ds_prev, ds); | |
1879 | ds->ds_prev = ds_prev = NULL; | |
1880 | } | |
1881 | } | |
34dc7c2f | 1882 | |
572e2857 BB |
1883 | /* |
1884 | * This must be done after the dsl_traverse(), because it will | |
1885 | * re-open the objset. | |
1886 | */ | |
1887 | if (ds->ds_objset) { | |
1888 | dmu_objset_evict(ds->ds_objset); | |
1889 | ds->ds_objset = NULL; | |
1890 | } | |
1891 | ||
34dc7c2f | 1892 | if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) { |
b128c09f | 1893 | /* Erase the link in the dir */ |
34dc7c2f BB |
1894 | dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx); |
1895 | ds->ds_dir->dd_phys->dd_head_dataset_obj = 0; | |
b128c09f BB |
1896 | ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0); |
1897 | err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx); | |
1898 | ASSERT(err == 0); | |
34dc7c2f BB |
1899 | } else { |
1900 | /* remove from snapshot namespace */ | |
1901 | dsl_dataset_t *ds_head; | |
b128c09f BB |
1902 | ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0); |
1903 | VERIFY(0 == dsl_dataset_hold_obj(dp, | |
1904 | ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head)); | |
34dc7c2f BB |
1905 | VERIFY(0 == dsl_dataset_get_snapname(ds)); |
1906 | #ifdef ZFS_DEBUG | |
1907 | { | |
1908 | uint64_t val; | |
1909 | ||
b128c09f | 1910 | err = dsl_dataset_snap_lookup(ds_head, |
34dc7c2f BB |
1911 | ds->ds_snapname, &val); |
1912 | ASSERT3U(err, ==, 0); | |
1913 | ASSERT3U(val, ==, obj); | |
1914 | } | |
1915 | #endif | |
b128c09f | 1916 | err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx); |
34dc7c2f | 1917 | ASSERT(err == 0); |
b128c09f | 1918 | dsl_dataset_rele(ds_head, FTAG); |
34dc7c2f BB |
1919 | } |
1920 | ||
1921 | if (ds_prev && ds->ds_prev != ds_prev) | |
b128c09f | 1922 | dsl_dataset_rele(ds_prev, FTAG); |
34dc7c2f BB |
1923 | |
1924 | spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx); | |
428870ff BB |
1925 | spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx, |
1926 | "dataset = %llu", ds->ds_object); | |
34dc7c2f | 1927 | |
b128c09f | 1928 | if (ds->ds_phys->ds_next_clones_obj != 0) { |
1fde1e37 | 1929 | ASSERTV(uint64_t count); |
b128c09f BB |
1930 | ASSERT(0 == zap_count(mos, |
1931 | ds->ds_phys->ds_next_clones_obj, &count) && count == 0); | |
1932 | VERIFY(0 == dmu_object_free(mos, | |
1933 | ds->ds_phys->ds_next_clones_obj, tx)); | |
1934 | } | |
1935 | if (ds->ds_phys->ds_props_obj != 0) | |
1936 | VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx)); | |
45d1cae3 BB |
1937 | if (ds->ds_phys->ds_userrefs_obj != 0) |
1938 | VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx)); | |
b128c09f BB |
1939 | dsl_dir_close(ds->ds_dir, ds); |
1940 | ds->ds_dir = NULL; | |
1941 | dsl_dataset_drain_refs(ds, tag); | |
34dc7c2f | 1942 | VERIFY(0 == dmu_object_free(mos, obj, tx)); |
45d1cae3 BB |
1943 | |
1944 | if (dsda->rm_origin) { | |
1945 | /* | |
1946 | * Remove the origin of the clone we just destroyed. | |
1947 | */ | |
45d1cae3 BB |
1948 | struct dsl_ds_destroyarg ndsda = {0}; |
1949 | ||
428870ff BB |
1950 | ndsda.ds = dsda->rm_origin; |
1951 | dsl_dataset_destroy_sync(&ndsda, tag, tx); | |
45d1cae3 | 1952 | } |
34dc7c2f BB |
1953 | } |
1954 | ||
1955 | static int | |
1956 | dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx) | |
1957 | { | |
1958 | uint64_t asize; | |
1959 | ||
1960 | if (!dmu_tx_is_syncing(tx)) | |
1961 | return (0); | |
1962 | ||
1963 | /* | |
1964 | * If there's an fs-only reservation, any blocks that might become | |
1965 | * owned by the snapshot dataset must be accommodated by space | |
1966 | * outside of the reservation. | |
1967 | */ | |
428870ff BB |
1968 | ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds)); |
1969 | asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); | |
572e2857 | 1970 | if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) |
34dc7c2f BB |
1971 | return (ENOSPC); |
1972 | ||
1973 | /* | |
1974 | * Propogate any reserved space for this snapshot to other | |
1975 | * snapshot checks in this sync group. | |
1976 | */ | |
1977 | if (asize > 0) | |
1978 | dsl_dir_willuse_space(ds->ds_dir, asize, tx); | |
1979 | ||
1980 | return (0); | |
1981 | } | |
1982 | ||
34dc7c2f BB |
1983 | int |
1984 | dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
1985 | { | |
1986 | dsl_dataset_t *ds = arg1; | |
1987 | const char *snapname = arg2; | |
34dc7c2f BB |
1988 | int err; |
1989 | uint64_t value; | |
1990 | ||
1991 | /* | |
1992 | * We don't allow multiple snapshots of the same txg. If there | |
1993 | * is already one, try again. | |
1994 | */ | |
1995 | if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg) | |
1996 | return (EAGAIN); | |
1997 | ||
1998 | /* | |
1999 | * Check for conflicting name snapshot name. | |
2000 | */ | |
b128c09f | 2001 | err = dsl_dataset_snap_lookup(ds, snapname, &value); |
34dc7c2f BB |
2002 | if (err == 0) |
2003 | return (EEXIST); | |
2004 | if (err != ENOENT) | |
2005 | return (err); | |
2006 | ||
2007 | /* | |
2008 | * Check that the dataset's name is not too long. Name consists | |
2009 | * of the dataset's length + 1 for the @-sign + snapshot name's length | |
2010 | */ | |
2011 | if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN) | |
2012 | return (ENAMETOOLONG); | |
2013 | ||
2014 | err = dsl_dataset_snapshot_reserve_space(ds, tx); | |
2015 | if (err) | |
2016 | return (err); | |
2017 | ||
2018 | ds->ds_trysnap_txg = tx->tx_txg; | |
2019 | return (0); | |
2020 | } | |
2021 | ||
2022 | void | |
428870ff | 2023 | dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
2024 | { |
2025 | dsl_dataset_t *ds = arg1; | |
2026 | const char *snapname = arg2; | |
2027 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
2028 | dmu_buf_t *dbuf; | |
2029 | dsl_dataset_phys_t *dsphys; | |
b128c09f | 2030 | uint64_t dsobj, crtxg; |
34dc7c2f BB |
2031 | objset_t *mos = dp->dp_meta_objset; |
2032 | int err; | |
2033 | ||
34dc7c2f BB |
2034 | ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock)); |
2035 | ||
b128c09f BB |
2036 | /* |
2037 | * The origin's ds_creation_txg has to be < TXG_INITIAL | |
2038 | */ | |
2039 | if (strcmp(snapname, ORIGIN_DIR_NAME) == 0) | |
2040 | crtxg = 1; | |
2041 | else | |
2042 | crtxg = tx->tx_txg; | |
2043 | ||
34dc7c2f BB |
2044 | dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0, |
2045 | DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx); | |
2046 | VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf)); | |
2047 | dmu_buf_will_dirty(dbuf, tx); | |
2048 | dsphys = dbuf->db_data; | |
b128c09f | 2049 | bzero(dsphys, sizeof (dsl_dataset_phys_t)); |
34dc7c2f BB |
2050 | dsphys->ds_dir_obj = ds->ds_dir->dd_object; |
2051 | dsphys->ds_fsid_guid = unique_create(); | |
2052 | (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid, | |
2053 | sizeof (dsphys->ds_guid)); | |
2054 | dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj; | |
2055 | dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg; | |
2056 | dsphys->ds_next_snap_obj = ds->ds_object; | |
2057 | dsphys->ds_num_children = 1; | |
2058 | dsphys->ds_creation_time = gethrestime_sec(); | |
b128c09f | 2059 | dsphys->ds_creation_txg = crtxg; |
34dc7c2f BB |
2060 | dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj; |
2061 | dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes; | |
2062 | dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes; | |
2063 | dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes; | |
2064 | dsphys->ds_flags = ds->ds_phys->ds_flags; | |
2065 | dsphys->ds_bp = ds->ds_phys->ds_bp; | |
2066 | dmu_buf_rele(dbuf, FTAG); | |
2067 | ||
2068 | ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0); | |
2069 | if (ds->ds_prev) { | |
b128c09f BB |
2070 | uint64_t next_clones_obj = |
2071 | ds->ds_prev->ds_phys->ds_next_clones_obj; | |
34dc7c2f BB |
2072 | ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj == |
2073 | ds->ds_object || | |
2074 | ds->ds_prev->ds_phys->ds_num_children > 1); | |
2075 | if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) { | |
2076 | dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx); | |
2077 | ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==, | |
2078 | ds->ds_prev->ds_phys->ds_creation_txg); | |
2079 | ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj; | |
b128c09f | 2080 | } else if (next_clones_obj != 0) { |
428870ff BB |
2081 | remove_from_next_clones(ds->ds_prev, |
2082 | dsphys->ds_next_snap_obj, tx); | |
b128c09f BB |
2083 | VERIFY3U(0, ==, zap_add_int(mos, |
2084 | next_clones_obj, dsobj, tx)); | |
34dc7c2f BB |
2085 | } |
2086 | } | |
2087 | ||
2088 | /* | |
2089 | * If we have a reference-reservation on this dataset, we will | |
2090 | * need to increase the amount of refreservation being charged | |
2091 | * since our unique space is going to zero. | |
2092 | */ | |
2093 | if (ds->ds_reserved) { | |
428870ff BB |
2094 | int64_t delta; |
2095 | ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); | |
2096 | delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved); | |
b128c09f | 2097 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, |
428870ff | 2098 | delta, 0, 0, tx); |
34dc7c2f BB |
2099 | } |
2100 | ||
34dc7c2f | 2101 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
428870ff BB |
2102 | zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu", |
2103 | ds->ds_dir->dd_myname, snapname, dsobj, | |
2104 | ds->ds_phys->ds_prev_snap_txg); | |
2105 | ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist, | |
2106 | UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx); | |
2107 | dsl_deadlist_close(&ds->ds_deadlist); | |
2108 | dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj); | |
2109 | dsl_deadlist_add_key(&ds->ds_deadlist, | |
2110 | ds->ds_phys->ds_prev_snap_txg, tx); | |
2111 | ||
34dc7c2f BB |
2112 | ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg); |
2113 | ds->ds_phys->ds_prev_snap_obj = dsobj; | |
b128c09f | 2114 | ds->ds_phys->ds_prev_snap_txg = crtxg; |
34dc7c2f BB |
2115 | ds->ds_phys->ds_unique_bytes = 0; |
2116 | if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE) | |
2117 | ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE; | |
34dc7c2f | 2118 | |
34dc7c2f BB |
2119 | err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj, |
2120 | snapname, 8, 1, &dsobj, tx); | |
2121 | ASSERT(err == 0); | |
2122 | ||
2123 | if (ds->ds_prev) | |
b128c09f BB |
2124 | dsl_dataset_drop_ref(ds->ds_prev, ds); |
2125 | VERIFY(0 == dsl_dataset_get_ref(dp, | |
2126 | ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev)); | |
2127 | ||
428870ff BB |
2128 | dsl_scan_ds_snapshotted(ds, tx); |
2129 | ||
2130 | dsl_dir_snap_cmtime_update(ds->ds_dir); | |
34dc7c2f | 2131 | |
428870ff | 2132 | spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx, |
34dc7c2f BB |
2133 | "dataset = %llu", dsobj); |
2134 | } | |
2135 | ||
2136 | void | |
2137 | dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx) | |
2138 | { | |
2139 | ASSERT(dmu_tx_is_syncing(tx)); | |
428870ff | 2140 | ASSERT(ds->ds_objset != NULL); |
34dc7c2f BB |
2141 | ASSERT(ds->ds_phys->ds_next_snap_obj == 0); |
2142 | ||
2143 | /* | |
2144 | * in case we had to change ds_fsid_guid when we opened it, | |
2145 | * sync it out now. | |
2146 | */ | |
2147 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2148 | ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid; | |
2149 | ||
2150 | dsl_dir_dirty(ds->ds_dir, tx); | |
428870ff | 2151 | dmu_objset_sync(ds->ds_objset, zio, tx); |
34dc7c2f BB |
2152 | } |
2153 | ||
2154 | void | |
2155 | dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv) | |
2156 | { | |
f5fc4aca | 2157 | uint64_t refd, avail, uobjs, aobjs, ratio; |
34dc7c2f BB |
2158 | |
2159 | dsl_dir_stats(ds->ds_dir, nv); | |
2160 | ||
2161 | dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs); | |
2162 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail); | |
2163 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd); | |
2164 | ||
2165 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION, | |
2166 | ds->ds_phys->ds_creation_time); | |
2167 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG, | |
2168 | ds->ds_phys->ds_creation_txg); | |
2169 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA, | |
2170 | ds->ds_quota); | |
2171 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION, | |
2172 | ds->ds_reserved); | |
b128c09f BB |
2173 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID, |
2174 | ds->ds_phys->ds_guid); | |
428870ff BB |
2175 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE, |
2176 | ds->ds_phys->ds_unique_bytes); | |
2177 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID, | |
2178 | ds->ds_object); | |
2179 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS, | |
2180 | ds->ds_userrefs); | |
45d1cae3 BB |
2181 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY, |
2182 | DS_IS_DEFER_DESTROY(ds) ? 1 : 0); | |
34dc7c2f | 2183 | |
f5fc4aca MA |
2184 | ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 : |
2185 | (ds->ds_phys->ds_uncompressed_bytes * 100 / | |
2186 | ds->ds_phys->ds_compressed_bytes); | |
2187 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio); | |
2188 | ||
34dc7c2f BB |
2189 | if (ds->ds_phys->ds_next_snap_obj) { |
2190 | /* | |
2191 | * This is a snapshot; override the dd's space used with | |
2192 | * our unique space and compression ratio. | |
2193 | */ | |
2194 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, | |
2195 | ds->ds_phys->ds_unique_bytes); | |
f5fc4aca | 2196 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio); |
34dc7c2f BB |
2197 | } |
2198 | } | |
2199 | ||
2200 | void | |
2201 | dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat) | |
2202 | { | |
2203 | stat->dds_creation_txg = ds->ds_phys->ds_creation_txg; | |
2204 | stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT; | |
2205 | stat->dds_guid = ds->ds_phys->ds_guid; | |
2206 | if (ds->ds_phys->ds_next_snap_obj) { | |
2207 | stat->dds_is_snapshot = B_TRUE; | |
2208 | stat->dds_num_clones = ds->ds_phys->ds_num_children - 1; | |
fb5f0bc8 BB |
2209 | } else { |
2210 | stat->dds_is_snapshot = B_FALSE; | |
2211 | stat->dds_num_clones = 0; | |
34dc7c2f BB |
2212 | } |
2213 | ||
2214 | /* clone origin is really a dsl_dir thing... */ | |
2215 | rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER); | |
b128c09f | 2216 | if (dsl_dir_is_clone(ds->ds_dir)) { |
34dc7c2f BB |
2217 | dsl_dataset_t *ods; |
2218 | ||
b128c09f BB |
2219 | VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool, |
2220 | ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods)); | |
34dc7c2f | 2221 | dsl_dataset_name(ods, stat->dds_origin); |
b128c09f | 2222 | dsl_dataset_drop_ref(ods, FTAG); |
fb5f0bc8 BB |
2223 | } else { |
2224 | stat->dds_origin[0] = '\0'; | |
34dc7c2f BB |
2225 | } |
2226 | rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock); | |
2227 | } | |
2228 | ||
2229 | uint64_t | |
2230 | dsl_dataset_fsid_guid(dsl_dataset_t *ds) | |
2231 | { | |
2232 | return (ds->ds_fsid_guid); | |
2233 | } | |
2234 | ||
2235 | void | |
2236 | dsl_dataset_space(dsl_dataset_t *ds, | |
2237 | uint64_t *refdbytesp, uint64_t *availbytesp, | |
2238 | uint64_t *usedobjsp, uint64_t *availobjsp) | |
2239 | { | |
2240 | *refdbytesp = ds->ds_phys->ds_used_bytes; | |
2241 | *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE); | |
2242 | if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) | |
2243 | *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes; | |
2244 | if (ds->ds_quota != 0) { | |
2245 | /* | |
2246 | * Adjust available bytes according to refquota | |
2247 | */ | |
2248 | if (*refdbytesp < ds->ds_quota) | |
2249 | *availbytesp = MIN(*availbytesp, | |
2250 | ds->ds_quota - *refdbytesp); | |
2251 | else | |
2252 | *availbytesp = 0; | |
2253 | } | |
2254 | *usedobjsp = ds->ds_phys->ds_bp.blk_fill; | |
2255 | *availobjsp = DN_MAX_OBJECT - *usedobjsp; | |
2256 | } | |
2257 | ||
2258 | boolean_t | |
2259 | dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds) | |
2260 | { | |
1fde1e37 | 2261 | ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool); |
34dc7c2f BB |
2262 | |
2263 | ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) || | |
2264 | dsl_pool_sync_context(dp)); | |
2265 | if (ds->ds_prev == NULL) | |
2266 | return (B_FALSE); | |
2267 | if (ds->ds_phys->ds_bp.blk_birth > | |
572e2857 BB |
2268 | ds->ds_prev->ds_phys->ds_creation_txg) { |
2269 | objset_t *os, *os_prev; | |
2270 | /* | |
2271 | * It may be that only the ZIL differs, because it was | |
2272 | * reset in the head. Don't count that as being | |
2273 | * modified. | |
2274 | */ | |
2275 | if (dmu_objset_from_ds(ds, &os) != 0) | |
2276 | return (B_TRUE); | |
2277 | if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0) | |
2278 | return (B_TRUE); | |
2279 | return (bcmp(&os->os_phys->os_meta_dnode, | |
2280 | &os_prev->os_phys->os_meta_dnode, | |
2281 | sizeof (os->os_phys->os_meta_dnode)) != 0); | |
2282 | } | |
34dc7c2f BB |
2283 | return (B_FALSE); |
2284 | } | |
2285 | ||
2286 | /* ARGSUSED */ | |
2287 | static int | |
2288 | dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
2289 | { | |
2290 | dsl_dataset_t *ds = arg1; | |
2291 | char *newsnapname = arg2; | |
2292 | dsl_dir_t *dd = ds->ds_dir; | |
34dc7c2f BB |
2293 | dsl_dataset_t *hds; |
2294 | uint64_t val; | |
2295 | int err; | |
2296 | ||
b128c09f BB |
2297 | err = dsl_dataset_hold_obj(dd->dd_pool, |
2298 | dd->dd_phys->dd_head_dataset_obj, FTAG, &hds); | |
34dc7c2f BB |
2299 | if (err) |
2300 | return (err); | |
2301 | ||
2302 | /* new name better not be in use */ | |
b128c09f BB |
2303 | err = dsl_dataset_snap_lookup(hds, newsnapname, &val); |
2304 | dsl_dataset_rele(hds, FTAG); | |
34dc7c2f BB |
2305 | |
2306 | if (err == 0) | |
2307 | err = EEXIST; | |
2308 | else if (err == ENOENT) | |
2309 | err = 0; | |
2310 | ||
2311 | /* dataset name + 1 for the "@" + the new snapshot name must fit */ | |
2312 | if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN) | |
2313 | err = ENAMETOOLONG; | |
2314 | ||
2315 | return (err); | |
2316 | } | |
2317 | ||
2318 | static void | |
428870ff | 2319 | dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
2320 | { |
2321 | dsl_dataset_t *ds = arg1; | |
2322 | const char *newsnapname = arg2; | |
2323 | dsl_dir_t *dd = ds->ds_dir; | |
2324 | objset_t *mos = dd->dd_pool->dp_meta_objset; | |
2325 | dsl_dataset_t *hds; | |
2326 | int err; | |
2327 | ||
2328 | ASSERT(ds->ds_phys->ds_next_snap_obj != 0); | |
2329 | ||
b128c09f BB |
2330 | VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, |
2331 | dd->dd_phys->dd_head_dataset_obj, FTAG, &hds)); | |
34dc7c2f BB |
2332 | |
2333 | VERIFY(0 == dsl_dataset_get_snapname(ds)); | |
b128c09f | 2334 | err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx); |
34dc7c2f BB |
2335 | ASSERT3U(err, ==, 0); |
2336 | mutex_enter(&ds->ds_lock); | |
2337 | (void) strcpy(ds->ds_snapname, newsnapname); | |
2338 | mutex_exit(&ds->ds_lock); | |
2339 | err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj, | |
2340 | ds->ds_snapname, 8, 1, &ds->ds_object, tx); | |
2341 | ASSERT3U(err, ==, 0); | |
2342 | ||
428870ff BB |
2343 | spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx, |
2344 | "dataset = %llu", ds->ds_object); | |
b128c09f | 2345 | dsl_dataset_rele(hds, FTAG); |
34dc7c2f BB |
2346 | } |
2347 | ||
2348 | struct renamesnaparg { | |
2349 | dsl_sync_task_group_t *dstg; | |
2350 | char failed[MAXPATHLEN]; | |
2351 | char *oldsnap; | |
2352 | char *newsnap; | |
2353 | }; | |
2354 | ||
2355 | static int | |
428870ff | 2356 | dsl_snapshot_rename_one(const char *name, void *arg) |
34dc7c2f BB |
2357 | { |
2358 | struct renamesnaparg *ra = arg; | |
2359 | dsl_dataset_t *ds = NULL; | |
428870ff | 2360 | char *snapname; |
34dc7c2f BB |
2361 | int err; |
2362 | ||
428870ff BB |
2363 | snapname = kmem_asprintf("%s@%s", name, ra->oldsnap); |
2364 | (void) strlcpy(ra->failed, snapname, sizeof (ra->failed)); | |
34dc7c2f BB |
2365 | |
2366 | /* | |
2367 | * For recursive snapshot renames the parent won't be changing | |
2368 | * so we just pass name for both the to/from argument. | |
2369 | */ | |
428870ff BB |
2370 | err = zfs_secpolicy_rename_perms(snapname, snapname, CRED()); |
2371 | if (err != 0) { | |
2372 | strfree(snapname); | |
2373 | return (err == ENOENT ? 0 : err); | |
34dc7c2f BB |
2374 | } |
2375 | ||
ebe7e575 | 2376 | #ifdef _KERNEL |
b128c09f BB |
2377 | /* |
2378 | * For all filesystems undergoing rename, we'll need to unmount it. | |
2379 | */ | |
428870ff | 2380 | (void) zfs_unmount_snap(snapname, NULL); |
34dc7c2f | 2381 | #endif |
428870ff BB |
2382 | err = dsl_dataset_hold(snapname, ra->dstg, &ds); |
2383 | strfree(snapname); | |
2384 | if (err != 0) | |
2385 | return (err == ENOENT ? 0 : err); | |
34dc7c2f BB |
2386 | |
2387 | dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check, | |
2388 | dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0); | |
2389 | ||
2390 | return (0); | |
2391 | } | |
2392 | ||
2393 | static int | |
2394 | dsl_recursive_rename(char *oldname, const char *newname) | |
2395 | { | |
2396 | int err; | |
2397 | struct renamesnaparg *ra; | |
2398 | dsl_sync_task_t *dst; | |
2399 | spa_t *spa; | |
2400 | char *cp, *fsname = spa_strdup(oldname); | |
428870ff | 2401 | int len = strlen(oldname) + 1; |
34dc7c2f BB |
2402 | |
2403 | /* truncate the snapshot name to get the fsname */ | |
2404 | cp = strchr(fsname, '@'); | |
2405 | *cp = '\0'; | |
2406 | ||
2407 | err = spa_open(fsname, &spa, FTAG); | |
2408 | if (err) { | |
428870ff | 2409 | kmem_free(fsname, len); |
34dc7c2f BB |
2410 | return (err); |
2411 | } | |
2412 | ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP); | |
2413 | ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
2414 | ||
2415 | ra->oldsnap = strchr(oldname, '@') + 1; | |
2416 | ra->newsnap = strchr(newname, '@') + 1; | |
2417 | *ra->failed = '\0'; | |
2418 | ||
2419 | err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra, | |
2420 | DS_FIND_CHILDREN); | |
428870ff | 2421 | kmem_free(fsname, len); |
34dc7c2f BB |
2422 | |
2423 | if (err == 0) { | |
2424 | err = dsl_sync_task_group_wait(ra->dstg); | |
2425 | } | |
2426 | ||
2427 | for (dst = list_head(&ra->dstg->dstg_tasks); dst; | |
2428 | dst = list_next(&ra->dstg->dstg_tasks, dst)) { | |
2429 | dsl_dataset_t *ds = dst->dst_arg1; | |
2430 | if (dst->dst_err) { | |
2431 | dsl_dir_name(ds->ds_dir, ra->failed); | |
428870ff BB |
2432 | (void) strlcat(ra->failed, "@", sizeof (ra->failed)); |
2433 | (void) strlcat(ra->failed, ra->newsnap, | |
2434 | sizeof (ra->failed)); | |
34dc7c2f | 2435 | } |
b128c09f | 2436 | dsl_dataset_rele(ds, ra->dstg); |
34dc7c2f BB |
2437 | } |
2438 | ||
2439 | if (err) | |
428870ff | 2440 | (void) strlcpy(oldname, ra->failed, sizeof (ra->failed)); |
34dc7c2f BB |
2441 | |
2442 | dsl_sync_task_group_destroy(ra->dstg); | |
2443 | kmem_free(ra, sizeof (struct renamesnaparg)); | |
2444 | spa_close(spa, FTAG); | |
2445 | return (err); | |
2446 | } | |
2447 | ||
2448 | static int | |
428870ff | 2449 | dsl_valid_rename(const char *oldname, void *arg) |
34dc7c2f BB |
2450 | { |
2451 | int delta = *(int *)arg; | |
2452 | ||
2453 | if (strlen(oldname) + delta >= MAXNAMELEN) | |
2454 | return (ENAMETOOLONG); | |
2455 | ||
2456 | return (0); | |
2457 | } | |
2458 | ||
2459 | #pragma weak dmu_objset_rename = dsl_dataset_rename | |
2460 | int | |
b128c09f | 2461 | dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive) |
34dc7c2f BB |
2462 | { |
2463 | dsl_dir_t *dd; | |
2464 | dsl_dataset_t *ds; | |
2465 | const char *tail; | |
2466 | int err; | |
2467 | ||
2468 | err = dsl_dir_open(oldname, FTAG, &dd, &tail); | |
2469 | if (err) | |
2470 | return (err); | |
428870ff | 2471 | |
34dc7c2f BB |
2472 | if (tail == NULL) { |
2473 | int delta = strlen(newname) - strlen(oldname); | |
2474 | ||
b128c09f | 2475 | /* if we're growing, validate child name lengths */ |
34dc7c2f BB |
2476 | if (delta > 0) |
2477 | err = dmu_objset_find(oldname, dsl_valid_rename, | |
2478 | &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS); | |
2479 | ||
428870ff | 2480 | if (err == 0) |
34dc7c2f BB |
2481 | err = dsl_dir_rename(dd, newname); |
2482 | dsl_dir_close(dd, FTAG); | |
2483 | return (err); | |
2484 | } | |
428870ff | 2485 | |
34dc7c2f | 2486 | if (tail[0] != '@') { |
428870ff | 2487 | /* the name ended in a nonexistent component */ |
34dc7c2f BB |
2488 | dsl_dir_close(dd, FTAG); |
2489 | return (ENOENT); | |
2490 | } | |
2491 | ||
2492 | dsl_dir_close(dd, FTAG); | |
2493 | ||
2494 | /* new name must be snapshot in same filesystem */ | |
2495 | tail = strchr(newname, '@'); | |
2496 | if (tail == NULL) | |
2497 | return (EINVAL); | |
2498 | tail++; | |
2499 | if (strncmp(oldname, newname, tail - newname) != 0) | |
2500 | return (EXDEV); | |
2501 | ||
2502 | if (recursive) { | |
2503 | err = dsl_recursive_rename(oldname, newname); | |
2504 | } else { | |
b128c09f | 2505 | err = dsl_dataset_hold(oldname, FTAG, &ds); |
34dc7c2f BB |
2506 | if (err) |
2507 | return (err); | |
2508 | ||
2509 | err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
2510 | dsl_dataset_snapshot_rename_check, | |
2511 | dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1); | |
2512 | ||
b128c09f | 2513 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
2514 | } |
2515 | ||
2516 | return (err); | |
2517 | } | |
2518 | ||
b128c09f BB |
2519 | struct promotenode { |
2520 | list_node_t link; | |
2521 | dsl_dataset_t *ds; | |
2522 | }; | |
2523 | ||
34dc7c2f | 2524 | struct promotearg { |
b128c09f | 2525 | list_t shared_snaps, origin_snaps, clone_snaps; |
428870ff | 2526 | dsl_dataset_t *origin_origin; |
b128c09f | 2527 | uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap; |
428870ff | 2528 | char *err_ds; |
34dc7c2f BB |
2529 | }; |
2530 | ||
b128c09f BB |
2531 | static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep); |
2532 | ||
34dc7c2f BB |
2533 | static int |
2534 | dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
2535 | { | |
2536 | dsl_dataset_t *hds = arg1; | |
2537 | struct promotearg *pa = arg2; | |
b128c09f BB |
2538 | struct promotenode *snap = list_head(&pa->shared_snaps); |
2539 | dsl_dataset_t *origin_ds = snap->ds; | |
34dc7c2f | 2540 | int err; |
428870ff | 2541 | uint64_t unused; |
34dc7c2f | 2542 | |
b128c09f BB |
2543 | /* Check that it is a real clone */ |
2544 | if (!dsl_dir_is_clone(hds->ds_dir)) | |
34dc7c2f BB |
2545 | return (EINVAL); |
2546 | ||
2547 | /* Since this is so expensive, don't do the preliminary check */ | |
2548 | if (!dmu_tx_is_syncing(tx)) | |
2549 | return (0); | |
2550 | ||
b128c09f BB |
2551 | if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) |
2552 | return (EXDEV); | |
34dc7c2f BB |
2553 | |
2554 | /* compute origin's new unique space */ | |
b128c09f BB |
2555 | snap = list_tail(&pa->clone_snaps); |
2556 | ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); | |
428870ff BB |
2557 | dsl_deadlist_space_range(&snap->ds->ds_deadlist, |
2558 | origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX, | |
2559 | &pa->unique, &unused, &unused); | |
34dc7c2f | 2560 | |
b128c09f BB |
2561 | /* |
2562 | * Walk the snapshots that we are moving | |
2563 | * | |
2564 | * Compute space to transfer. Consider the incremental changes | |
2565 | * to used for each snapshot: | |
2566 | * (my used) = (prev's used) + (blocks born) - (blocks killed) | |
2567 | * So each snapshot gave birth to: | |
2568 | * (blocks born) = (my used) - (prev's used) + (blocks killed) | |
2569 | * So a sequence would look like: | |
2570 | * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0) | |
2571 | * Which simplifies to: | |
2572 | * uN + kN + kN-1 + ... + k1 + k0 | |
2573 | * Note however, if we stop before we reach the ORIGIN we get: | |
2574 | * uN + kN + kN-1 + ... + kM - uM-1 | |
2575 | */ | |
2576 | pa->used = origin_ds->ds_phys->ds_used_bytes; | |
2577 | pa->comp = origin_ds->ds_phys->ds_compressed_bytes; | |
2578 | pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes; | |
2579 | for (snap = list_head(&pa->shared_snaps); snap; | |
2580 | snap = list_next(&pa->shared_snaps, snap)) { | |
34dc7c2f | 2581 | uint64_t val, dlused, dlcomp, dluncomp; |
b128c09f | 2582 | dsl_dataset_t *ds = snap->ds; |
34dc7c2f BB |
2583 | |
2584 | /* Check that the snapshot name does not conflict */ | |
b128c09f BB |
2585 | VERIFY(0 == dsl_dataset_get_snapname(ds)); |
2586 | err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val); | |
428870ff BB |
2587 | if (err == 0) { |
2588 | err = EEXIST; | |
2589 | goto out; | |
2590 | } | |
b128c09f | 2591 | if (err != ENOENT) |
428870ff | 2592 | goto out; |
34dc7c2f | 2593 | |
b128c09f | 2594 | /* The very first snapshot does not have a deadlist */ |
34dc7c2f | 2595 | if (ds->ds_phys->ds_prev_snap_obj == 0) |
b128c09f | 2596 | continue; |
34dc7c2f | 2597 | |
428870ff BB |
2598 | dsl_deadlist_space(&ds->ds_deadlist, |
2599 | &dlused, &dlcomp, &dluncomp); | |
b128c09f BB |
2600 | pa->used += dlused; |
2601 | pa->comp += dlcomp; | |
2602 | pa->uncomp += dluncomp; | |
2603 | } | |
34dc7c2f | 2604 | |
b128c09f BB |
2605 | /* |
2606 | * If we are a clone of a clone then we never reached ORIGIN, | |
2607 | * so we need to subtract out the clone origin's used space. | |
2608 | */ | |
2609 | if (pa->origin_origin) { | |
2610 | pa->used -= pa->origin_origin->ds_phys->ds_used_bytes; | |
2611 | pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes; | |
2612 | pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes; | |
34dc7c2f BB |
2613 | } |
2614 | ||
2615 | /* Check that there is enough space here */ | |
b128c09f BB |
2616 | err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir, |
2617 | pa->used); | |
2618 | if (err) | |
2619 | return (err); | |
34dc7c2f | 2620 | |
b128c09f BB |
2621 | /* |
2622 | * Compute the amounts of space that will be used by snapshots | |
2623 | * after the promotion (for both origin and clone). For each, | |
2624 | * it is the amount of space that will be on all of their | |
2625 | * deadlists (that was not born before their new origin). | |
2626 | */ | |
2627 | if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { | |
2628 | uint64_t space; | |
2629 | ||
2630 | /* | |
2631 | * Note, typically this will not be a clone of a clone, | |
428870ff BB |
2632 | * so dd_origin_txg will be < TXG_INITIAL, so |
2633 | * these snaplist_space() -> dsl_deadlist_space_range() | |
b128c09f BB |
2634 | * calls will be fast because they do not have to |
2635 | * iterate over all bps. | |
2636 | */ | |
2637 | snap = list_head(&pa->origin_snaps); | |
2638 | err = snaplist_space(&pa->shared_snaps, | |
428870ff | 2639 | snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap); |
b128c09f BB |
2640 | if (err) |
2641 | return (err); | |
2642 | ||
2643 | err = snaplist_space(&pa->clone_snaps, | |
428870ff | 2644 | snap->ds->ds_dir->dd_origin_txg, &space); |
b128c09f BB |
2645 | if (err) |
2646 | return (err); | |
2647 | pa->cloneusedsnap += space; | |
2648 | } | |
2649 | if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { | |
2650 | err = snaplist_space(&pa->origin_snaps, | |
2651 | origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap); | |
2652 | if (err) | |
2653 | return (err); | |
2654 | } | |
2655 | ||
2656 | return (0); | |
428870ff BB |
2657 | out: |
2658 | pa->err_ds = snap->ds->ds_snapname; | |
2659 | return (err); | |
34dc7c2f BB |
2660 | } |
2661 | ||
2662 | static void | |
428870ff | 2663 | dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
2664 | { |
2665 | dsl_dataset_t *hds = arg1; | |
2666 | struct promotearg *pa = arg2; | |
b128c09f BB |
2667 | struct promotenode *snap = list_head(&pa->shared_snaps); |
2668 | dsl_dataset_t *origin_ds = snap->ds; | |
2669 | dsl_dataset_t *origin_head; | |
34dc7c2f BB |
2670 | dsl_dir_t *dd = hds->ds_dir; |
2671 | dsl_pool_t *dp = hds->ds_dir->dd_pool; | |
2672 | dsl_dir_t *odd = NULL; | |
b128c09f BB |
2673 | uint64_t oldnext_obj; |
2674 | int64_t delta; | |
34dc7c2f | 2675 | |
34dc7c2f BB |
2676 | ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)); |
2677 | ||
b128c09f BB |
2678 | snap = list_head(&pa->origin_snaps); |
2679 | origin_head = snap->ds; | |
2680 | ||
34dc7c2f BB |
2681 | /* |
2682 | * We need to explicitly open odd, since origin_ds's dd will be | |
2683 | * changing. | |
2684 | */ | |
2685 | VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object, | |
2686 | NULL, FTAG, &odd)); | |
2687 | ||
b128c09f BB |
2688 | /* change origin's next snap */ |
2689 | dmu_buf_will_dirty(origin_ds->ds_dbuf, tx); | |
2690 | oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj; | |
2691 | snap = list_tail(&pa->clone_snaps); | |
2692 | ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object); | |
2693 | origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object; | |
2694 | ||
2695 | /* change the origin's next clone */ | |
2696 | if (origin_ds->ds_phys->ds_next_clones_obj) { | |
428870ff | 2697 | remove_from_next_clones(origin_ds, snap->ds->ds_object, tx); |
b128c09f BB |
2698 | VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, |
2699 | origin_ds->ds_phys->ds_next_clones_obj, | |
2700 | oldnext_obj, tx)); | |
2701 | } | |
2702 | ||
2703 | /* change origin */ | |
2704 | dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
2705 | ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object); | |
2706 | dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj; | |
428870ff | 2707 | dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg; |
b128c09f BB |
2708 | dmu_buf_will_dirty(odd->dd_dbuf, tx); |
2709 | odd->dd_phys->dd_origin_obj = origin_ds->ds_object; | |
428870ff BB |
2710 | origin_head->ds_dir->dd_origin_txg = |
2711 | origin_ds->ds_phys->ds_creation_txg; | |
2712 | ||
2713 | /* change dd_clone entries */ | |
2714 | if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { | |
2715 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, | |
2716 | odd->dd_phys->dd_clones, hds->ds_object, tx)); | |
2717 | VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, | |
2718 | pa->origin_origin->ds_dir->dd_phys->dd_clones, | |
2719 | hds->ds_object, tx)); | |
2720 | ||
2721 | VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset, | |
2722 | pa->origin_origin->ds_dir->dd_phys->dd_clones, | |
2723 | origin_head->ds_object, tx)); | |
2724 | if (dd->dd_phys->dd_clones == 0) { | |
2725 | dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset, | |
2726 | DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx); | |
2727 | } | |
2728 | VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset, | |
2729 | dd->dd_phys->dd_clones, origin_head->ds_object, tx)); | |
2730 | ||
2731 | } | |
34dc7c2f | 2732 | |
b128c09f BB |
2733 | /* move snapshots to this dir */ |
2734 | for (snap = list_head(&pa->shared_snaps); snap; | |
2735 | snap = list_next(&pa->shared_snaps, snap)) { | |
2736 | dsl_dataset_t *ds = snap->ds; | |
2737 | ||
2738 | /* unregister props as dsl_dir is changing */ | |
428870ff BB |
2739 | if (ds->ds_objset) { |
2740 | dmu_objset_evict(ds->ds_objset); | |
2741 | ds->ds_objset = NULL; | |
b128c09f | 2742 | } |
34dc7c2f | 2743 | /* move snap name entry */ |
b128c09f BB |
2744 | VERIFY(0 == dsl_dataset_get_snapname(ds)); |
2745 | VERIFY(0 == dsl_dataset_snap_remove(origin_head, | |
2746 | ds->ds_snapname, tx)); | |
34dc7c2f BB |
2747 | VERIFY(0 == zap_add(dp->dp_meta_objset, |
2748 | hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname, | |
2749 | 8, 1, &ds->ds_object, tx)); | |
428870ff | 2750 | |
34dc7c2f BB |
2751 | /* change containing dsl_dir */ |
2752 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
2753 | ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object); | |
2754 | ds->ds_phys->ds_dir_obj = dd->dd_object; | |
2755 | ASSERT3P(ds->ds_dir, ==, odd); | |
2756 | dsl_dir_close(ds->ds_dir, ds); | |
2757 | VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object, | |
2758 | NULL, ds, &ds->ds_dir)); | |
2759 | ||
428870ff BB |
2760 | /* move any clone references */ |
2761 | if (ds->ds_phys->ds_next_clones_obj && | |
2762 | spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) { | |
2763 | zap_cursor_t zc; | |
2764 | zap_attribute_t za; | |
2765 | ||
2766 | for (zap_cursor_init(&zc, dp->dp_meta_objset, | |
2767 | ds->ds_phys->ds_next_clones_obj); | |
2768 | zap_cursor_retrieve(&zc, &za) == 0; | |
2769 | zap_cursor_advance(&zc)) { | |
2770 | dsl_dataset_t *cnds; | |
2771 | uint64_t o; | |
2772 | ||
2773 | if (za.za_first_integer == oldnext_obj) { | |
2774 | /* | |
2775 | * We've already moved the | |
2776 | * origin's reference. | |
2777 | */ | |
2778 | continue; | |
2779 | } | |
2780 | ||
2781 | VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, | |
2782 | za.za_first_integer, FTAG, &cnds)); | |
2783 | o = cnds->ds_dir->dd_phys->dd_head_dataset_obj; | |
2784 | ||
2785 | VERIFY3U(zap_remove_int(dp->dp_meta_objset, | |
2786 | odd->dd_phys->dd_clones, o, tx), ==, 0); | |
2787 | VERIFY3U(zap_add_int(dp->dp_meta_objset, | |
2788 | dd->dd_phys->dd_clones, o, tx), ==, 0); | |
2789 | dsl_dataset_rele(cnds, FTAG); | |
2790 | } | |
2791 | zap_cursor_fini(&zc); | |
2792 | } | |
2793 | ||
34dc7c2f | 2794 | ASSERT3U(dsl_prop_numcb(ds), ==, 0); |
34dc7c2f | 2795 | } |
34dc7c2f | 2796 | |
b128c09f BB |
2797 | /* |
2798 | * Change space accounting. | |
2799 | * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either | |
2800 | * both be valid, or both be 0 (resulting in delta == 0). This | |
2801 | * is true for each of {clone,origin} independently. | |
2802 | */ | |
34dc7c2f | 2803 | |
b128c09f BB |
2804 | delta = pa->cloneusedsnap - |
2805 | dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; | |
2806 | ASSERT3S(delta, >=, 0); | |
2807 | ASSERT3U(pa->used, >=, delta); | |
2808 | dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx); | |
2809 | dsl_dir_diduse_space(dd, DD_USED_HEAD, | |
2810 | pa->used - delta, pa->comp, pa->uncomp, tx); | |
2811 | ||
2812 | delta = pa->originusedsnap - | |
2813 | odd->dd_phys->dd_used_breakdown[DD_USED_SNAP]; | |
2814 | ASSERT3S(delta, <=, 0); | |
2815 | ASSERT3U(pa->used, >=, -delta); | |
2816 | dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx); | |
2817 | dsl_dir_diduse_space(odd, DD_USED_HEAD, | |
2818 | -pa->used - delta, -pa->comp, -pa->uncomp, tx); | |
34dc7c2f | 2819 | |
34dc7c2f BB |
2820 | origin_ds->ds_phys->ds_unique_bytes = pa->unique; |
2821 | ||
2822 | /* log history record */ | |
428870ff BB |
2823 | spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx, |
2824 | "dataset = %llu", hds->ds_object); | |
34dc7c2f BB |
2825 | |
2826 | dsl_dir_close(odd, FTAG); | |
34dc7c2f BB |
2827 | } |
2828 | ||
b128c09f BB |
2829 | static char *snaplist_tag = "snaplist"; |
2830 | /* | |
2831 | * Make a list of dsl_dataset_t's for the snapshots between first_obj | |
2832 | * (exclusive) and last_obj (inclusive). The list will be in reverse | |
2833 | * order (last_obj will be the list_head()). If first_obj == 0, do all | |
2834 | * snapshots back to this dataset's origin. | |
2835 | */ | |
2836 | static int | |
2837 | snaplist_make(dsl_pool_t *dp, boolean_t own, | |
2838 | uint64_t first_obj, uint64_t last_obj, list_t *l) | |
2839 | { | |
2840 | uint64_t obj = last_obj; | |
2841 | ||
2842 | ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock)); | |
2843 | ||
2844 | list_create(l, sizeof (struct promotenode), | |
2845 | offsetof(struct promotenode, link)); | |
2846 | ||
2847 | while (obj != first_obj) { | |
2848 | dsl_dataset_t *ds; | |
2849 | struct promotenode *snap; | |
2850 | int err; | |
2851 | ||
2852 | if (own) { | |
2853 | err = dsl_dataset_own_obj(dp, obj, | |
2854 | 0, snaplist_tag, &ds); | |
2855 | if (err == 0) | |
2856 | dsl_dataset_make_exclusive(ds, snaplist_tag); | |
2857 | } else { | |
2858 | err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds); | |
2859 | } | |
2860 | if (err == ENOENT) { | |
2861 | /* lost race with snapshot destroy */ | |
2862 | struct promotenode *last = list_tail(l); | |
2863 | ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj); | |
2864 | obj = last->ds->ds_phys->ds_prev_snap_obj; | |
2865 | continue; | |
2866 | } else if (err) { | |
2867 | return (err); | |
2868 | } | |
2869 | ||
2870 | if (first_obj == 0) | |
2871 | first_obj = ds->ds_dir->dd_phys->dd_origin_obj; | |
2872 | ||
2873 | snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP); | |
2874 | snap->ds = ds; | |
2875 | list_insert_tail(l, snap); | |
2876 | obj = ds->ds_phys->ds_prev_snap_obj; | |
2877 | } | |
2878 | ||
2879 | return (0); | |
2880 | } | |
2881 | ||
2882 | static int | |
2883 | snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep) | |
2884 | { | |
2885 | struct promotenode *snap; | |
2886 | ||
2887 | *spacep = 0; | |
2888 | for (snap = list_head(l); snap; snap = list_next(l, snap)) { | |
428870ff BB |
2889 | uint64_t used, comp, uncomp; |
2890 | dsl_deadlist_space_range(&snap->ds->ds_deadlist, | |
2891 | mintxg, UINT64_MAX, &used, &comp, &uncomp); | |
b128c09f BB |
2892 | *spacep += used; |
2893 | } | |
2894 | return (0); | |
2895 | } | |
2896 | ||
2897 | static void | |
2898 | snaplist_destroy(list_t *l, boolean_t own) | |
2899 | { | |
2900 | struct promotenode *snap; | |
2901 | ||
9babb374 | 2902 | if (!l || !list_link_active(&l->list_head)) |
b128c09f BB |
2903 | return; |
2904 | ||
2905 | while ((snap = list_tail(l)) != NULL) { | |
2906 | list_remove(l, snap); | |
2907 | if (own) | |
2908 | dsl_dataset_disown(snap->ds, snaplist_tag); | |
2909 | else | |
2910 | dsl_dataset_rele(snap->ds, snaplist_tag); | |
2911 | kmem_free(snap, sizeof (struct promotenode)); | |
2912 | } | |
2913 | list_destroy(l); | |
2914 | } | |
2915 | ||
2916 | /* | |
2917 | * Promote a clone. Nomenclature note: | |
2918 | * "clone" or "cds": the original clone which is being promoted | |
2919 | * "origin" or "ods": the snapshot which is originally clone's origin | |
2920 | * "origin head" or "ohds": the dataset which is the head | |
2921 | * (filesystem/volume) for the origin | |
2922 | * "origin origin": the origin of the origin's filesystem (typically | |
2923 | * NULL, indicating that the clone is not a clone of a clone). | |
2924 | */ | |
34dc7c2f | 2925 | int |
428870ff | 2926 | dsl_dataset_promote(const char *name, char *conflsnap) |
34dc7c2f BB |
2927 | { |
2928 | dsl_dataset_t *ds; | |
b128c09f BB |
2929 | dsl_dir_t *dd; |
2930 | dsl_pool_t *dp; | |
34dc7c2f | 2931 | dmu_object_info_t doi; |
2598c001 | 2932 | struct promotearg pa; |
b128c09f BB |
2933 | struct promotenode *snap; |
2934 | int err; | |
34dc7c2f | 2935 | |
2598c001 | 2936 | bzero(&pa, sizeof(struct promotearg)); |
b128c09f | 2937 | err = dsl_dataset_hold(name, FTAG, &ds); |
34dc7c2f BB |
2938 | if (err) |
2939 | return (err); | |
b128c09f BB |
2940 | dd = ds->ds_dir; |
2941 | dp = dd->dd_pool; | |
34dc7c2f | 2942 | |
b128c09f | 2943 | err = dmu_object_info(dp->dp_meta_objset, |
34dc7c2f BB |
2944 | ds->ds_phys->ds_snapnames_zapobj, &doi); |
2945 | if (err) { | |
b128c09f | 2946 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
2947 | return (err); |
2948 | } | |
2949 | ||
b128c09f BB |
2950 | if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) { |
2951 | dsl_dataset_rele(ds, FTAG); | |
2952 | return (EINVAL); | |
2953 | } | |
2954 | ||
2955 | /* | |
2956 | * We are going to inherit all the snapshots taken before our | |
2957 | * origin (i.e., our new origin will be our parent's origin). | |
2958 | * Take ownership of them so that we can rename them into our | |
2959 | * namespace. | |
2960 | */ | |
2961 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
2962 | ||
2963 | err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj, | |
2964 | &pa.shared_snaps); | |
2965 | if (err != 0) | |
2966 | goto out; | |
2967 | ||
2968 | err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps); | |
2969 | if (err != 0) | |
2970 | goto out; | |
2971 | ||
2972 | snap = list_head(&pa.shared_snaps); | |
2973 | ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj); | |
2974 | err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj, | |
2975 | snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps); | |
2976 | if (err != 0) | |
2977 | goto out; | |
2978 | ||
428870ff BB |
2979 | if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) { |
2980 | err = dsl_dataset_hold_obj(dp, | |
b128c09f | 2981 | snap->ds->ds_dir->dd_phys->dd_origin_obj, |
428870ff | 2982 | FTAG, &pa.origin_origin); |
b128c09f BB |
2983 | if (err != 0) |
2984 | goto out; | |
2985 | } | |
2986 | ||
2987 | out: | |
2988 | rw_exit(&dp->dp_config_rwlock); | |
2989 | ||
34dc7c2f BB |
2990 | /* |
2991 | * Add in 128x the snapnames zapobj size, since we will be moving | |
2992 | * a bunch of snapnames to the promoted ds, and dirtying their | |
2993 | * bonus buffers. | |
2994 | */ | |
b128c09f BB |
2995 | if (err == 0) { |
2996 | err = dsl_sync_task_do(dp, dsl_dataset_promote_check, | |
2997 | dsl_dataset_promote_sync, ds, &pa, | |
428870ff BB |
2998 | 2 + 2 * doi.doi_physical_blocks_512); |
2999 | if (err && pa.err_ds && conflsnap) | |
3000 | (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN); | |
b128c09f BB |
3001 | } |
3002 | ||
3003 | snaplist_destroy(&pa.shared_snaps, B_TRUE); | |
3004 | snaplist_destroy(&pa.clone_snaps, B_FALSE); | |
3005 | snaplist_destroy(&pa.origin_snaps, B_FALSE); | |
3006 | if (pa.origin_origin) | |
428870ff | 3007 | dsl_dataset_rele(pa.origin_origin, FTAG); |
b128c09f | 3008 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
3009 | return (err); |
3010 | } | |
3011 | ||
3012 | struct cloneswaparg { | |
3013 | dsl_dataset_t *cds; /* clone dataset */ | |
3014 | dsl_dataset_t *ohds; /* origin's head dataset */ | |
3015 | boolean_t force; | |
3016 | int64_t unused_refres_delta; /* change in unconsumed refreservation */ | |
3017 | }; | |
3018 | ||
3019 | /* ARGSUSED */ | |
3020 | static int | |
3021 | dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
3022 | { | |
3023 | struct cloneswaparg *csa = arg1; | |
3024 | ||
3025 | /* they should both be heads */ | |
3026 | if (dsl_dataset_is_snapshot(csa->cds) || | |
3027 | dsl_dataset_is_snapshot(csa->ohds)) | |
3028 | return (EINVAL); | |
3029 | ||
3030 | /* the branch point should be just before them */ | |
3031 | if (csa->cds->ds_prev != csa->ohds->ds_prev) | |
3032 | return (EINVAL); | |
3033 | ||
428870ff BB |
3034 | /* cds should be the clone (unless they are unrelated) */ |
3035 | if (csa->cds->ds_prev != NULL && | |
3036 | csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap && | |
3037 | csa->ohds->ds_object != | |
3038 | csa->cds->ds_prev->ds_phys->ds_next_snap_obj) | |
34dc7c2f BB |
3039 | return (EINVAL); |
3040 | ||
3041 | /* the clone should be a child of the origin */ | |
3042 | if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir) | |
3043 | return (EINVAL); | |
3044 | ||
3045 | /* ohds shouldn't be modified unless 'force' */ | |
3046 | if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds)) | |
3047 | return (ETXTBSY); | |
3048 | ||
3049 | /* adjust amount of any unconsumed refreservation */ | |
3050 | csa->unused_refres_delta = | |
3051 | (int64_t)MIN(csa->ohds->ds_reserved, | |
3052 | csa->ohds->ds_phys->ds_unique_bytes) - | |
3053 | (int64_t)MIN(csa->ohds->ds_reserved, | |
3054 | csa->cds->ds_phys->ds_unique_bytes); | |
3055 | ||
3056 | if (csa->unused_refres_delta > 0 && | |
3057 | csa->unused_refres_delta > | |
3058 | dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE)) | |
3059 | return (ENOSPC); | |
3060 | ||
428870ff BB |
3061 | if (csa->ohds->ds_quota != 0 && |
3062 | csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota) | |
3063 | return (EDQUOT); | |
3064 | ||
34dc7c2f BB |
3065 | return (0); |
3066 | } | |
3067 | ||
3068 | /* ARGSUSED */ | |
3069 | static void | |
428870ff | 3070 | dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
3071 | { |
3072 | struct cloneswaparg *csa = arg1; | |
3073 | dsl_pool_t *dp = csa->cds->ds_dir->dd_pool; | |
34dc7c2f BB |
3074 | |
3075 | ASSERT(csa->cds->ds_reserved == 0); | |
428870ff BB |
3076 | ASSERT(csa->ohds->ds_quota == 0 || |
3077 | csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota); | |
34dc7c2f BB |
3078 | |
3079 | dmu_buf_will_dirty(csa->cds->ds_dbuf, tx); | |
3080 | dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx); | |
34dc7c2f | 3081 | |
428870ff BB |
3082 | if (csa->cds->ds_objset != NULL) { |
3083 | dmu_objset_evict(csa->cds->ds_objset); | |
3084 | csa->cds->ds_objset = NULL; | |
34dc7c2f BB |
3085 | } |
3086 | ||
428870ff BB |
3087 | if (csa->ohds->ds_objset != NULL) { |
3088 | dmu_objset_evict(csa->ohds->ds_objset); | |
3089 | csa->ohds->ds_objset = NULL; | |
34dc7c2f BB |
3090 | } |
3091 | ||
428870ff BB |
3092 | /* |
3093 | * Reset origin's unique bytes, if it exists. | |
3094 | */ | |
3095 | if (csa->cds->ds_prev) { | |
3096 | dsl_dataset_t *origin = csa->cds->ds_prev; | |
3097 | uint64_t comp, uncomp; | |
3098 | ||
3099 | dmu_buf_will_dirty(origin->ds_dbuf, tx); | |
3100 | dsl_deadlist_space_range(&csa->cds->ds_deadlist, | |
3101 | origin->ds_phys->ds_prev_snap_txg, UINT64_MAX, | |
3102 | &origin->ds_phys->ds_unique_bytes, &comp, &uncomp); | |
3103 | } | |
34dc7c2f BB |
3104 | |
3105 | /* swap blkptrs */ | |
3106 | { | |
3107 | blkptr_t tmp; | |
3108 | tmp = csa->ohds->ds_phys->ds_bp; | |
3109 | csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp; | |
3110 | csa->cds->ds_phys->ds_bp = tmp; | |
3111 | } | |
3112 | ||
3113 | /* set dd_*_bytes */ | |
3114 | { | |
3115 | int64_t dused, dcomp, duncomp; | |
3116 | uint64_t cdl_used, cdl_comp, cdl_uncomp; | |
3117 | uint64_t odl_used, odl_comp, odl_uncomp; | |
3118 | ||
b128c09f BB |
3119 | ASSERT3U(csa->cds->ds_dir->dd_phys-> |
3120 | dd_used_breakdown[DD_USED_SNAP], ==, 0); | |
3121 | ||
428870ff BB |
3122 | dsl_deadlist_space(&csa->cds->ds_deadlist, |
3123 | &cdl_used, &cdl_comp, &cdl_uncomp); | |
3124 | dsl_deadlist_space(&csa->ohds->ds_deadlist, | |
3125 | &odl_used, &odl_comp, &odl_uncomp); | |
b128c09f | 3126 | |
34dc7c2f BB |
3127 | dused = csa->cds->ds_phys->ds_used_bytes + cdl_used - |
3128 | (csa->ohds->ds_phys->ds_used_bytes + odl_used); | |
3129 | dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp - | |
3130 | (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp); | |
3131 | duncomp = csa->cds->ds_phys->ds_uncompressed_bytes + | |
3132 | cdl_uncomp - | |
3133 | (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp); | |
3134 | ||
b128c09f | 3135 | dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD, |
34dc7c2f | 3136 | dused, dcomp, duncomp, tx); |
b128c09f | 3137 | dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD, |
34dc7c2f | 3138 | -dused, -dcomp, -duncomp, tx); |
b128c09f BB |
3139 | |
3140 | /* | |
3141 | * The difference in the space used by snapshots is the | |
3142 | * difference in snapshot space due to the head's | |
3143 | * deadlist (since that's the only thing that's | |
3144 | * changing that affects the snapused). | |
3145 | */ | |
428870ff BB |
3146 | dsl_deadlist_space_range(&csa->cds->ds_deadlist, |
3147 | csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, | |
3148 | &cdl_used, &cdl_comp, &cdl_uncomp); | |
3149 | dsl_deadlist_space_range(&csa->ohds->ds_deadlist, | |
3150 | csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX, | |
3151 | &odl_used, &odl_comp, &odl_uncomp); | |
b128c09f BB |
3152 | dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used, |
3153 | DD_USED_HEAD, DD_USED_SNAP, tx); | |
34dc7c2f BB |
3154 | } |
3155 | ||
34dc7c2f BB |
3156 | /* swap ds_*_bytes */ |
3157 | SWITCH64(csa->ohds->ds_phys->ds_used_bytes, | |
3158 | csa->cds->ds_phys->ds_used_bytes); | |
3159 | SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes, | |
3160 | csa->cds->ds_phys->ds_compressed_bytes); | |
3161 | SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes, | |
3162 | csa->cds->ds_phys->ds_uncompressed_bytes); | |
3163 | SWITCH64(csa->ohds->ds_phys->ds_unique_bytes, | |
3164 | csa->cds->ds_phys->ds_unique_bytes); | |
3165 | ||
3166 | /* apply any parent delta for change in unconsumed refreservation */ | |
b128c09f BB |
3167 | dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV, |
3168 | csa->unused_refres_delta, 0, 0, tx); | |
34dc7c2f | 3169 | |
428870ff BB |
3170 | /* |
3171 | * Swap deadlists. | |
3172 | */ | |
3173 | dsl_deadlist_close(&csa->cds->ds_deadlist); | |
3174 | dsl_deadlist_close(&csa->ohds->ds_deadlist); | |
34dc7c2f BB |
3175 | SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj, |
3176 | csa->cds->ds_phys->ds_deadlist_obj); | |
428870ff BB |
3177 | dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset, |
3178 | csa->cds->ds_phys->ds_deadlist_obj); | |
3179 | dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset, | |
3180 | csa->ohds->ds_phys->ds_deadlist_obj); | |
b128c09f | 3181 | |
428870ff | 3182 | dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx); |
34dc7c2f BB |
3183 | } |
3184 | ||
3185 | /* | |
428870ff BB |
3186 | * Swap 'clone' with its origin head datasets. Used at the end of "zfs |
3187 | * recv" into an existing fs to swizzle the file system to the new | |
3188 | * version, and by "zfs rollback". Can also be used to swap two | |
3189 | * independent head datasets if neither has any snapshots. | |
34dc7c2f BB |
3190 | */ |
3191 | int | |
3192 | dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head, | |
3193 | boolean_t force) | |
3194 | { | |
3195 | struct cloneswaparg csa; | |
b128c09f | 3196 | int error; |
34dc7c2f | 3197 | |
b128c09f BB |
3198 | ASSERT(clone->ds_owner); |
3199 | ASSERT(origin_head->ds_owner); | |
3200 | retry: | |
572e2857 BB |
3201 | /* |
3202 | * Need exclusive access for the swap. If we're swapping these | |
3203 | * datasets back after an error, we already hold the locks. | |
3204 | */ | |
3205 | if (!RW_WRITE_HELD(&clone->ds_rwlock)) | |
3206 | rw_enter(&clone->ds_rwlock, RW_WRITER); | |
3207 | if (!RW_WRITE_HELD(&origin_head->ds_rwlock) && | |
3208 | !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) { | |
b128c09f BB |
3209 | rw_exit(&clone->ds_rwlock); |
3210 | rw_enter(&origin_head->ds_rwlock, RW_WRITER); | |
3211 | if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) { | |
3212 | rw_exit(&origin_head->ds_rwlock); | |
3213 | goto retry; | |
3214 | } | |
3215 | } | |
34dc7c2f BB |
3216 | csa.cds = clone; |
3217 | csa.ohds = origin_head; | |
3218 | csa.force = force; | |
b128c09f | 3219 | error = dsl_sync_task_do(clone->ds_dir->dd_pool, |
34dc7c2f | 3220 | dsl_dataset_clone_swap_check, |
b128c09f BB |
3221 | dsl_dataset_clone_swap_sync, &csa, NULL, 9); |
3222 | return (error); | |
34dc7c2f BB |
3223 | } |
3224 | ||
3225 | /* | |
3226 | * Given a pool name and a dataset object number in that pool, | |
3227 | * return the name of that dataset. | |
3228 | */ | |
3229 | int | |
3230 | dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf) | |
3231 | { | |
3232 | spa_t *spa; | |
3233 | dsl_pool_t *dp; | |
b128c09f | 3234 | dsl_dataset_t *ds; |
34dc7c2f BB |
3235 | int error; |
3236 | ||
3237 | if ((error = spa_open(pname, &spa, FTAG)) != 0) | |
3238 | return (error); | |
3239 | dp = spa_get_dsl(spa); | |
3240 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
b128c09f BB |
3241 | if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) { |
3242 | dsl_dataset_name(ds, buf); | |
3243 | dsl_dataset_rele(ds, FTAG); | |
34dc7c2f | 3244 | } |
34dc7c2f BB |
3245 | rw_exit(&dp->dp_config_rwlock); |
3246 | spa_close(spa, FTAG); | |
3247 | ||
b128c09f | 3248 | return (error); |
34dc7c2f BB |
3249 | } |
3250 | ||
3251 | int | |
3252 | dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota, | |
b128c09f | 3253 | uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv) |
34dc7c2f BB |
3254 | { |
3255 | int error = 0; | |
3256 | ||
3257 | ASSERT3S(asize, >, 0); | |
3258 | ||
3259 | /* | |
3260 | * *ref_rsrv is the portion of asize that will come from any | |
3261 | * unconsumed refreservation space. | |
3262 | */ | |
3263 | *ref_rsrv = 0; | |
3264 | ||
3265 | mutex_enter(&ds->ds_lock); | |
3266 | /* | |
3267 | * Make a space adjustment for reserved bytes. | |
3268 | */ | |
3269 | if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) { | |
3270 | ASSERT3U(*used, >=, | |
3271 | ds->ds_reserved - ds->ds_phys->ds_unique_bytes); | |
3272 | *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes); | |
3273 | *ref_rsrv = | |
3274 | asize - MIN(asize, parent_delta(ds, asize + inflight)); | |
3275 | } | |
3276 | ||
3277 | if (!check_quota || ds->ds_quota == 0) { | |
3278 | mutex_exit(&ds->ds_lock); | |
3279 | return (0); | |
3280 | } | |
3281 | /* | |
3282 | * If they are requesting more space, and our current estimate | |
3283 | * is over quota, they get to try again unless the actual | |
3284 | * on-disk is over quota and there are no pending changes (which | |
3285 | * may free up space for us). | |
3286 | */ | |
3287 | if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) { | |
3288 | if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota) | |
3289 | error = ERESTART; | |
3290 | else | |
3291 | error = EDQUOT; | |
570827e1 BB |
3292 | |
3293 | DMU_TX_STAT_BUMP(dmu_tx_quota); | |
34dc7c2f BB |
3294 | } |
3295 | mutex_exit(&ds->ds_lock); | |
3296 | ||
3297 | return (error); | |
3298 | } | |
3299 | ||
3300 | /* ARGSUSED */ | |
3301 | static int | |
3302 | dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
3303 | { | |
3304 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3305 | dsl_prop_setarg_t *psa = arg2; |
3306 | int err; | |
34dc7c2f BB |
3307 | |
3308 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA) | |
3309 | return (ENOTSUP); | |
3310 | ||
428870ff BB |
3311 | if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) |
3312 | return (err); | |
3313 | ||
3314 | if (psa->psa_effective_value == 0) | |
34dc7c2f BB |
3315 | return (0); |
3316 | ||
428870ff BB |
3317 | if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes || |
3318 | psa->psa_effective_value < ds->ds_reserved) | |
34dc7c2f BB |
3319 | return (ENOSPC); |
3320 | ||
3321 | return (0); | |
3322 | } | |
3323 | ||
428870ff BB |
3324 | extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *); |
3325 | ||
34dc7c2f | 3326 | void |
428870ff | 3327 | dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
3328 | { |
3329 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3330 | dsl_prop_setarg_t *psa = arg2; |
3331 | uint64_t effective_value = psa->psa_effective_value; | |
34dc7c2f | 3332 | |
428870ff BB |
3333 | dsl_prop_set_sync(ds, psa, tx); |
3334 | DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); | |
34dc7c2f | 3335 | |
428870ff BB |
3336 | if (ds->ds_quota != effective_value) { |
3337 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
3338 | ds->ds_quota = effective_value; | |
34dc7c2f | 3339 | |
428870ff BB |
3340 | spa_history_log_internal(LOG_DS_REFQUOTA, |
3341 | ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ", | |
3342 | (longlong_t)ds->ds_quota, ds->ds_object); | |
3343 | } | |
34dc7c2f BB |
3344 | } |
3345 | ||
3346 | int | |
428870ff | 3347 | dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota) |
34dc7c2f BB |
3348 | { |
3349 | dsl_dataset_t *ds; | |
428870ff | 3350 | dsl_prop_setarg_t psa; |
34dc7c2f BB |
3351 | int err; |
3352 | ||
428870ff BB |
3353 | dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a); |
3354 | ||
b128c09f | 3355 | err = dsl_dataset_hold(dsname, FTAG, &ds); |
34dc7c2f BB |
3356 | if (err) |
3357 | return (err); | |
3358 | ||
428870ff BB |
3359 | /* |
3360 | * If someone removes a file, then tries to set the quota, we | |
3361 | * want to make sure the file freeing takes effect. | |
3362 | */ | |
3363 | txg_wait_open(ds->ds_dir->dd_pool, 0); | |
3364 | ||
3365 | err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
3366 | dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync, | |
3367 | ds, &psa, 0); | |
34dc7c2f | 3368 | |
b128c09f | 3369 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
3370 | return (err); |
3371 | } | |
3372 | ||
3373 | static int | |
3374 | dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
3375 | { | |
3376 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3377 | dsl_prop_setarg_t *psa = arg2; |
3378 | uint64_t effective_value; | |
34dc7c2f | 3379 | uint64_t unique; |
428870ff | 3380 | int err; |
34dc7c2f | 3381 | |
34dc7c2f BB |
3382 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) < |
3383 | SPA_VERSION_REFRESERVATION) | |
3384 | return (ENOTSUP); | |
3385 | ||
3386 | if (dsl_dataset_is_snapshot(ds)) | |
3387 | return (EINVAL); | |
3388 | ||
428870ff BB |
3389 | if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0) |
3390 | return (err); | |
3391 | ||
3392 | effective_value = psa->psa_effective_value; | |
3393 | ||
34dc7c2f BB |
3394 | /* |
3395 | * If we are doing the preliminary check in open context, the | |
3396 | * space estimates may be inaccurate. | |
3397 | */ | |
3398 | if (!dmu_tx_is_syncing(tx)) | |
3399 | return (0); | |
3400 | ||
3401 | mutex_enter(&ds->ds_lock); | |
428870ff BB |
3402 | if (!DS_UNIQUE_IS_ACCURATE(ds)) |
3403 | dsl_dataset_recalc_head_uniq(ds); | |
3404 | unique = ds->ds_phys->ds_unique_bytes; | |
34dc7c2f BB |
3405 | mutex_exit(&ds->ds_lock); |
3406 | ||
428870ff BB |
3407 | if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) { |
3408 | uint64_t delta = MAX(unique, effective_value) - | |
d164b209 BB |
3409 | MAX(unique, ds->ds_reserved); |
3410 | ||
3411 | if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) | |
3412 | return (ENOSPC); | |
3413 | if (ds->ds_quota > 0 && | |
428870ff | 3414 | effective_value > ds->ds_quota) |
d164b209 BB |
3415 | return (ENOSPC); |
3416 | } | |
34dc7c2f BB |
3417 | |
3418 | return (0); | |
3419 | } | |
3420 | ||
34dc7c2f | 3421 | static void |
428870ff | 3422 | dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
34dc7c2f BB |
3423 | { |
3424 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3425 | dsl_prop_setarg_t *psa = arg2; |
3426 | uint64_t effective_value = psa->psa_effective_value; | |
34dc7c2f BB |
3427 | uint64_t unique; |
3428 | int64_t delta; | |
3429 | ||
428870ff BB |
3430 | dsl_prop_set_sync(ds, psa, tx); |
3431 | DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa); | |
3432 | ||
34dc7c2f BB |
3433 | dmu_buf_will_dirty(ds->ds_dbuf, tx); |
3434 | ||
b128c09f | 3435 | mutex_enter(&ds->ds_dir->dd_lock); |
34dc7c2f | 3436 | mutex_enter(&ds->ds_lock); |
428870ff BB |
3437 | ASSERT(DS_UNIQUE_IS_ACCURATE(ds)); |
3438 | unique = ds->ds_phys->ds_unique_bytes; | |
3439 | delta = MAX(0, (int64_t)(effective_value - unique)) - | |
34dc7c2f | 3440 | MAX(0, (int64_t)(ds->ds_reserved - unique)); |
428870ff | 3441 | ds->ds_reserved = effective_value; |
34dc7c2f BB |
3442 | mutex_exit(&ds->ds_lock); |
3443 | ||
b128c09f BB |
3444 | dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx); |
3445 | mutex_exit(&ds->ds_dir->dd_lock); | |
34dc7c2f | 3446 | |
428870ff BB |
3447 | spa_history_log_internal(LOG_DS_REFRESERV, |
3448 | ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu", | |
3449 | (longlong_t)effective_value, ds->ds_object); | |
34dc7c2f BB |
3450 | } |
3451 | ||
3452 | int | |
428870ff BB |
3453 | dsl_dataset_set_reservation(const char *dsname, zprop_source_t source, |
3454 | uint64_t reservation) | |
34dc7c2f BB |
3455 | { |
3456 | dsl_dataset_t *ds; | |
428870ff | 3457 | dsl_prop_setarg_t psa; |
34dc7c2f BB |
3458 | int err; |
3459 | ||
428870ff BB |
3460 | dsl_prop_setarg_init_uint64(&psa, "refreservation", source, |
3461 | &reservation); | |
3462 | ||
b128c09f | 3463 | err = dsl_dataset_hold(dsname, FTAG, &ds); |
34dc7c2f BB |
3464 | if (err) |
3465 | return (err); | |
3466 | ||
3467 | err = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
3468 | dsl_dataset_set_reservation_check, | |
428870ff BB |
3469 | dsl_dataset_set_reservation_sync, ds, &psa, 0); |
3470 | ||
b128c09f | 3471 | dsl_dataset_rele(ds, FTAG); |
34dc7c2f BB |
3472 | return (err); |
3473 | } | |
45d1cae3 | 3474 | |
572e2857 BB |
3475 | typedef struct zfs_hold_cleanup_arg { |
3476 | dsl_pool_t *dp; | |
3477 | uint64_t dsobj; | |
3478 | char htag[MAXNAMELEN]; | |
3479 | } zfs_hold_cleanup_arg_t; | |
3480 | ||
3481 | static void | |
3482 | dsl_dataset_user_release_onexit(void *arg) | |
3483 | { | |
3484 | zfs_hold_cleanup_arg_t *ca = arg; | |
3485 | ||
3486 | (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag, | |
3487 | B_TRUE); | |
3488 | kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t)); | |
3489 | } | |
3490 | ||
3491 | void | |
3492 | dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag, | |
3493 | minor_t minor) | |
3494 | { | |
3495 | zfs_hold_cleanup_arg_t *ca; | |
3496 | ||
3497 | ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP); | |
3498 | ca->dp = ds->ds_dir->dd_pool; | |
3499 | ca->dsobj = ds->ds_object; | |
3500 | (void) strlcpy(ca->htag, htag, sizeof (ca->htag)); | |
3501 | VERIFY3U(0, ==, zfs_onexit_add_cb(minor, | |
3502 | dsl_dataset_user_release_onexit, ca, NULL)); | |
3503 | } | |
428870ff BB |
3504 | |
3505 | /* | |
572e2857 BB |
3506 | * If you add new checks here, you may need to add |
3507 | * additional checks to the "temporary" case in | |
3508 | * snapshot_check() in dmu_objset.c. | |
428870ff | 3509 | */ |
45d1cae3 BB |
3510 | static int |
3511 | dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx) | |
3512 | { | |
3513 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3514 | struct dsl_ds_holdarg *ha = arg2; |
3515 | char *htag = ha->htag; | |
45d1cae3 BB |
3516 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; |
3517 | int error = 0; | |
3518 | ||
3519 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) | |
3520 | return (ENOTSUP); | |
3521 | ||
3522 | if (!dsl_dataset_is_snapshot(ds)) | |
3523 | return (EINVAL); | |
3524 | ||
45d1cae3 BB |
3525 | /* tags must be unique */ |
3526 | mutex_enter(&ds->ds_lock); | |
3527 | if (ds->ds_phys->ds_userrefs_obj) { | |
3528 | error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag, | |
3529 | 8, 1, tx); | |
3530 | if (error == 0) | |
3531 | error = EEXIST; | |
3532 | else if (error == ENOENT) | |
3533 | error = 0; | |
3534 | } | |
3535 | mutex_exit(&ds->ds_lock); | |
3536 | ||
428870ff BB |
3537 | if (error == 0 && ha->temphold && |
3538 | strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN) | |
3539 | error = E2BIG; | |
3540 | ||
45d1cae3 BB |
3541 | return (error); |
3542 | } | |
3543 | ||
572e2857 | 3544 | void |
428870ff | 3545 | dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx) |
45d1cae3 BB |
3546 | { |
3547 | dsl_dataset_t *ds = arg1; | |
428870ff BB |
3548 | struct dsl_ds_holdarg *ha = arg2; |
3549 | char *htag = ha->htag; | |
3550 | dsl_pool_t *dp = ds->ds_dir->dd_pool; | |
3551 | objset_t *mos = dp->dp_meta_objset; | |
3552 | uint64_t now = gethrestime_sec(); | |
45d1cae3 BB |
3553 | uint64_t zapobj; |
3554 | ||
3555 | mutex_enter(&ds->ds_lock); | |
3556 | if (ds->ds_phys->ds_userrefs_obj == 0) { | |
3557 | /* | |
3558 | * This is the first user hold for this dataset. Create | |
3559 | * the userrefs zap object. | |
3560 | */ | |
3561 | dmu_buf_will_dirty(ds->ds_dbuf, tx); | |
3562 | zapobj = ds->ds_phys->ds_userrefs_obj = | |
3563 | zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx); | |
3564 | } else { | |
3565 | zapobj = ds->ds_phys->ds_userrefs_obj; | |
3566 | } | |
3567 | ds->ds_userrefs++; | |
3568 | mutex_exit(&ds->ds_lock); | |
3569 | ||
3570 | VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx)); | |
3571 | ||
428870ff BB |
3572 | if (ha->temphold) { |
3573 | VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object, | |
3574 | htag, &now, tx)); | |
3575 | } | |
45d1cae3 | 3576 | |
428870ff BB |
3577 | spa_history_log_internal(LOG_DS_USER_HOLD, |
3578 | dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag, | |
3579 | (int)ha->temphold, ds->ds_object); | |
3580 | } | |
45d1cae3 BB |
3581 | |
3582 | static int | |
428870ff | 3583 | dsl_dataset_user_hold_one(const char *dsname, void *arg) |
45d1cae3 BB |
3584 | { |
3585 | struct dsl_ds_holdarg *ha = arg; | |
3586 | dsl_dataset_t *ds; | |
3587 | int error; | |
3588 | char *name; | |
45d1cae3 BB |
3589 | |
3590 | /* alloc a buffer to hold dsname@snapname plus terminating NULL */ | |
428870ff | 3591 | name = kmem_asprintf("%s@%s", dsname, ha->snapname); |
45d1cae3 | 3592 | error = dsl_dataset_hold(name, ha->dstg, &ds); |
428870ff | 3593 | strfree(name); |
45d1cae3 | 3594 | if (error == 0) { |
428870ff | 3595 | ha->gotone = B_TRUE; |
45d1cae3 | 3596 | dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check, |
428870ff | 3597 | dsl_dataset_user_hold_sync, ds, ha, 0); |
45d1cae3 BB |
3598 | } else if (error == ENOENT && ha->recursive) { |
3599 | error = 0; | |
3600 | } else { | |
428870ff | 3601 | (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); |
45d1cae3 BB |
3602 | } |
3603 | return (error); | |
3604 | } | |
3605 | ||
572e2857 BB |
3606 | int |
3607 | dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag, | |
3608 | boolean_t temphold) | |
3609 | { | |
3610 | struct dsl_ds_holdarg *ha; | |
3611 | int error; | |
3612 | ||
3613 | ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); | |
3614 | ha->htag = htag; | |
3615 | ha->temphold = temphold; | |
3616 | error = dsl_sync_task_do(ds->ds_dir->dd_pool, | |
3617 | dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync, | |
3618 | ds, ha, 0); | |
3619 | kmem_free(ha, sizeof (struct dsl_ds_holdarg)); | |
3620 | ||
3621 | return (error); | |
3622 | } | |
3623 | ||
45d1cae3 BB |
3624 | int |
3625 | dsl_dataset_user_hold(char *dsname, char *snapname, char *htag, | |
572e2857 | 3626 | boolean_t recursive, boolean_t temphold, int cleanup_fd) |
45d1cae3 BB |
3627 | { |
3628 | struct dsl_ds_holdarg *ha; | |
3629 | dsl_sync_task_t *dst; | |
3630 | spa_t *spa; | |
3631 | int error; | |
572e2857 BB |
3632 | minor_t minor = 0; |
3633 | ||
3634 | if (cleanup_fd != -1) { | |
3635 | /* Currently we only support cleanup-on-exit of tempholds. */ | |
3636 | if (!temphold) | |
3637 | return (EINVAL); | |
3638 | error = zfs_onexit_fd_hold(cleanup_fd, &minor); | |
3639 | if (error) | |
3640 | return (error); | |
3641 | } | |
45d1cae3 BB |
3642 | |
3643 | ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); | |
3644 | ||
3645 | (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); | |
3646 | ||
3647 | error = spa_open(dsname, &spa, FTAG); | |
3648 | if (error) { | |
3649 | kmem_free(ha, sizeof (struct dsl_ds_holdarg)); | |
572e2857 BB |
3650 | if (cleanup_fd != -1) |
3651 | zfs_onexit_fd_rele(cleanup_fd); | |
45d1cae3 BB |
3652 | return (error); |
3653 | } | |
3654 | ||
3655 | ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
3656 | ha->htag = htag; | |
3657 | ha->snapname = snapname; | |
3658 | ha->recursive = recursive; | |
428870ff | 3659 | ha->temphold = temphold; |
572e2857 | 3660 | |
45d1cae3 BB |
3661 | if (recursive) { |
3662 | error = dmu_objset_find(dsname, dsl_dataset_user_hold_one, | |
3663 | ha, DS_FIND_CHILDREN); | |
3664 | } else { | |
3665 | error = dsl_dataset_user_hold_one(dsname, ha); | |
3666 | } | |
3667 | if (error == 0) | |
3668 | error = dsl_sync_task_group_wait(ha->dstg); | |
3669 | ||
3670 | for (dst = list_head(&ha->dstg->dstg_tasks); dst; | |
3671 | dst = list_next(&ha->dstg->dstg_tasks, dst)) { | |
3672 | dsl_dataset_t *ds = dst->dst_arg1; | |
3673 | ||
3674 | if (dst->dst_err) { | |
3675 | dsl_dataset_name(ds, ha->failed); | |
3676 | *strchr(ha->failed, '@') = '\0'; | |
572e2857 BB |
3677 | } else if (error == 0 && minor != 0 && temphold) { |
3678 | /* | |
3679 | * If this hold is to be released upon process exit, | |
3680 | * register that action now. | |
3681 | */ | |
3682 | dsl_register_onexit_hold_cleanup(ds, htag, minor); | |
45d1cae3 BB |
3683 | } |
3684 | dsl_dataset_rele(ds, ha->dstg); | |
3685 | } | |
3686 | ||
428870ff BB |
3687 | if (error == 0 && recursive && !ha->gotone) |
3688 | error = ENOENT; | |
3689 | ||
45d1cae3 | 3690 | if (error) |
428870ff | 3691 | (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); |
45d1cae3 BB |
3692 | |
3693 | dsl_sync_task_group_destroy(ha->dstg); | |
572e2857 | 3694 | |
45d1cae3 BB |
3695 | kmem_free(ha, sizeof (struct dsl_ds_holdarg)); |
3696 | spa_close(spa, FTAG); | |
572e2857 BB |
3697 | if (cleanup_fd != -1) |
3698 | zfs_onexit_fd_rele(cleanup_fd); | |
45d1cae3 BB |
3699 | return (error); |
3700 | } | |
3701 | ||
3702 | struct dsl_ds_releasearg { | |
3703 | dsl_dataset_t *ds; | |
3704 | const char *htag; | |
3705 | boolean_t own; /* do we own or just hold ds? */ | |
3706 | }; | |
3707 | ||
3708 | static int | |
3709 | dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag, | |
3710 | boolean_t *might_destroy) | |
3711 | { | |
3712 | objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset; | |
3713 | uint64_t zapobj; | |
3714 | uint64_t tmp; | |
3715 | int error; | |
3716 | ||
3717 | *might_destroy = B_FALSE; | |
3718 | ||
3719 | mutex_enter(&ds->ds_lock); | |
3720 | zapobj = ds->ds_phys->ds_userrefs_obj; | |
3721 | if (zapobj == 0) { | |
3722 | /* The tag can't possibly exist */ | |
3723 | mutex_exit(&ds->ds_lock); | |
3724 | return (ESRCH); | |
3725 | } | |
3726 | ||
3727 | /* Make sure the tag exists */ | |
3728 | error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp); | |
3729 | if (error) { | |
3730 | mutex_exit(&ds->ds_lock); | |
3731 | if (error == ENOENT) | |
3732 | error = ESRCH; | |
3733 | return (error); | |
3734 | } | |
3735 | ||
3736 | if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 && | |
3737 | DS_IS_DEFER_DESTROY(ds)) | |
3738 | *might_destroy = B_TRUE; | |
3739 | ||
3740 | mutex_exit(&ds->ds_lock); | |
3741 | return (0); | |
3742 | } | |
3743 | ||
3744 | static int | |
3745 | dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx) | |
3746 | { | |
3747 | struct dsl_ds_releasearg *ra = arg1; | |
3748 | dsl_dataset_t *ds = ra->ds; | |
3749 | boolean_t might_destroy; | |
3750 | int error; | |
3751 | ||
3752 | if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS) | |
3753 | return (ENOTSUP); | |
3754 | ||
3755 | error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy); | |
3756 | if (error) | |
3757 | return (error); | |
3758 | ||
3759 | if (might_destroy) { | |
3760 | struct dsl_ds_destroyarg dsda = {0}; | |
3761 | ||
3762 | if (dmu_tx_is_syncing(tx)) { | |
3763 | /* | |
3764 | * If we're not prepared to remove the snapshot, | |
3765 | * we can't allow the release to happen right now. | |
3766 | */ | |
3767 | if (!ra->own) | |
3768 | return (EBUSY); | |
45d1cae3 BB |
3769 | } |
3770 | dsda.ds = ds; | |
3771 | dsda.releasing = B_TRUE; | |
3772 | return (dsl_dataset_destroy_check(&dsda, tag, tx)); | |
3773 | } | |
3774 | ||
3775 | return (0); | |
3776 | } | |
3777 | ||
3778 | static void | |
428870ff | 3779 | dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx) |
45d1cae3 BB |
3780 | { |
3781 | struct dsl_ds_releasearg *ra = arg1; | |
3782 | dsl_dataset_t *ds = ra->ds; | |
428870ff BB |
3783 | dsl_pool_t *dp = ds->ds_dir->dd_pool; |
3784 | objset_t *mos = dp->dp_meta_objset; | |
45d1cae3 BB |
3785 | uint64_t zapobj; |
3786 | uint64_t dsobj = ds->ds_object; | |
3787 | uint64_t refs; | |
428870ff BB |
3788 | int error; |
3789 | ||
45d1cae3 BB |
3790 | mutex_enter(&ds->ds_lock); |
3791 | ds->ds_userrefs--; | |
3792 | refs = ds->ds_userrefs; | |
3793 | mutex_exit(&ds->ds_lock); | |
428870ff BB |
3794 | error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx); |
3795 | VERIFY(error == 0 || error == ENOENT); | |
45d1cae3 BB |
3796 | zapobj = ds->ds_phys->ds_userrefs_obj; |
3797 | VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx)); | |
3798 | if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 && | |
3799 | DS_IS_DEFER_DESTROY(ds)) { | |
3800 | struct dsl_ds_destroyarg dsda = {0}; | |
3801 | ||
3802 | ASSERT(ra->own); | |
3803 | dsda.ds = ds; | |
3804 | dsda.releasing = B_TRUE; | |
3805 | /* We already did the destroy_check */ | |
428870ff | 3806 | dsl_dataset_destroy_sync(&dsda, tag, tx); |
45d1cae3 BB |
3807 | } |
3808 | ||
428870ff BB |
3809 | spa_history_log_internal(LOG_DS_USER_RELEASE, |
3810 | dp->dp_spa, tx, "<%s> %lld dataset = %llu", | |
45d1cae3 BB |
3811 | ra->htag, (longlong_t)refs, dsobj); |
3812 | } | |
3813 | ||
3814 | static int | |
428870ff | 3815 | dsl_dataset_user_release_one(const char *dsname, void *arg) |
45d1cae3 BB |
3816 | { |
3817 | struct dsl_ds_holdarg *ha = arg; | |
3818 | struct dsl_ds_releasearg *ra; | |
3819 | dsl_dataset_t *ds; | |
3820 | int error; | |
3821 | void *dtag = ha->dstg; | |
3822 | char *name; | |
45d1cae3 BB |
3823 | boolean_t own = B_FALSE; |
3824 | boolean_t might_destroy; | |
3825 | ||
45d1cae3 | 3826 | /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */ |
428870ff | 3827 | name = kmem_asprintf("%s@%s", dsname, ha->snapname); |
45d1cae3 | 3828 | error = dsl_dataset_hold(name, dtag, &ds); |
428870ff | 3829 | strfree(name); |
45d1cae3 BB |
3830 | if (error == ENOENT && ha->recursive) |
3831 | return (0); | |
428870ff | 3832 | (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); |
45d1cae3 BB |
3833 | if (error) |
3834 | return (error); | |
3835 | ||
428870ff BB |
3836 | ha->gotone = B_TRUE; |
3837 | ||
45d1cae3 BB |
3838 | ASSERT(dsl_dataset_is_snapshot(ds)); |
3839 | ||
3840 | error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy); | |
3841 | if (error) { | |
3842 | dsl_dataset_rele(ds, dtag); | |
3843 | return (error); | |
3844 | } | |
3845 | ||
3846 | if (might_destroy) { | |
3847 | #ifdef _KERNEL | |
428870ff | 3848 | name = kmem_asprintf("%s@%s", dsname, ha->snapname); |
45d1cae3 | 3849 | error = zfs_unmount_snap(name, NULL); |
428870ff | 3850 | strfree(name); |
45d1cae3 BB |
3851 | if (error) { |
3852 | dsl_dataset_rele(ds, dtag); | |
3853 | return (error); | |
3854 | } | |
3855 | #endif | |
428870ff | 3856 | if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) { |
45d1cae3 BB |
3857 | dsl_dataset_rele(ds, dtag); |
3858 | return (EBUSY); | |
3859 | } else { | |
3860 | own = B_TRUE; | |
3861 | dsl_dataset_make_exclusive(ds, dtag); | |
3862 | } | |
3863 | } | |
3864 | ||
3865 | ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP); | |
3866 | ra->ds = ds; | |
3867 | ra->htag = ha->htag; | |
3868 | ra->own = own; | |
3869 | dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check, | |
3870 | dsl_dataset_user_release_sync, ra, dtag, 0); | |
3871 | ||
3872 | return (0); | |
3873 | } | |
3874 | ||
3875 | int | |
3876 | dsl_dataset_user_release(char *dsname, char *snapname, char *htag, | |
3877 | boolean_t recursive) | |
3878 | { | |
3879 | struct dsl_ds_holdarg *ha; | |
3880 | dsl_sync_task_t *dst; | |
3881 | spa_t *spa; | |
3882 | int error; | |
3883 | ||
428870ff | 3884 | top: |
45d1cae3 BB |
3885 | ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP); |
3886 | ||
3887 | (void) strlcpy(ha->failed, dsname, sizeof (ha->failed)); | |
3888 | ||
3889 | error = spa_open(dsname, &spa, FTAG); | |
3890 | if (error) { | |
3891 | kmem_free(ha, sizeof (struct dsl_ds_holdarg)); | |
3892 | return (error); | |
3893 | } | |
3894 | ||
3895 | ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa)); | |
3896 | ha->htag = htag; | |
3897 | ha->snapname = snapname; | |
3898 | ha->recursive = recursive; | |
3899 | if (recursive) { | |
3900 | error = dmu_objset_find(dsname, dsl_dataset_user_release_one, | |
3901 | ha, DS_FIND_CHILDREN); | |
3902 | } else { | |
3903 | error = dsl_dataset_user_release_one(dsname, ha); | |
3904 | } | |
3905 | if (error == 0) | |
3906 | error = dsl_sync_task_group_wait(ha->dstg); | |
3907 | ||
3908 | for (dst = list_head(&ha->dstg->dstg_tasks); dst; | |
3909 | dst = list_next(&ha->dstg->dstg_tasks, dst)) { | |
3910 | struct dsl_ds_releasearg *ra = dst->dst_arg1; | |
3911 | dsl_dataset_t *ds = ra->ds; | |
3912 | ||
3913 | if (dst->dst_err) | |
3914 | dsl_dataset_name(ds, ha->failed); | |
3915 | ||
3916 | if (ra->own) | |
3917 | dsl_dataset_disown(ds, ha->dstg); | |
3918 | else | |
3919 | dsl_dataset_rele(ds, ha->dstg); | |
3920 | ||
3921 | kmem_free(ra, sizeof (struct dsl_ds_releasearg)); | |
3922 | } | |
3923 | ||
428870ff BB |
3924 | if (error == 0 && recursive && !ha->gotone) |
3925 | error = ENOENT; | |
3926 | ||
3927 | if (error && error != EBUSY) | |
3928 | (void) strlcpy(dsname, ha->failed, sizeof (ha->failed)); | |
45d1cae3 BB |
3929 | |
3930 | dsl_sync_task_group_destroy(ha->dstg); | |
3931 | kmem_free(ha, sizeof (struct dsl_ds_holdarg)); | |
3932 | spa_close(spa, FTAG); | |
428870ff BB |
3933 | |
3934 | /* | |
3935 | * We can get EBUSY if we were racing with deferred destroy and | |
3936 | * dsl_dataset_user_release_check() hadn't done the necessary | |
3937 | * open context setup. We can also get EBUSY if we're racing | |
3938 | * with destroy and that thread is the ds_owner. Either way | |
3939 | * the busy condition should be transient, and we should retry | |
3940 | * the release operation. | |
3941 | */ | |
3942 | if (error == EBUSY) | |
3943 | goto top; | |
3944 | ||
45d1cae3 BB |
3945 | return (error); |
3946 | } | |
3947 | ||
428870ff | 3948 | /* |
572e2857 BB |
3949 | * Called at spa_load time (with retry == B_FALSE) to release a stale |
3950 | * temporary user hold. Also called by the onexit code (with retry == B_TRUE). | |
428870ff BB |
3951 | */ |
3952 | int | |
572e2857 BB |
3953 | dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag, |
3954 | boolean_t retry) | |
428870ff BB |
3955 | { |
3956 | dsl_dataset_t *ds; | |
3957 | char *snap; | |
3958 | char *name; | |
3959 | int namelen; | |
3960 | int error; | |
3961 | ||
572e2857 BB |
3962 | do { |
3963 | rw_enter(&dp->dp_config_rwlock, RW_READER); | |
3964 | error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds); | |
3965 | rw_exit(&dp->dp_config_rwlock); | |
3966 | if (error) | |
3967 | return (error); | |
3968 | namelen = dsl_dataset_namelen(ds)+1; | |
3969 | name = kmem_alloc(namelen, KM_SLEEP); | |
3970 | dsl_dataset_name(ds, name); | |
3971 | dsl_dataset_rele(ds, FTAG); | |
428870ff | 3972 | |
572e2857 BB |
3973 | snap = strchr(name, '@'); |
3974 | *snap = '\0'; | |
3975 | ++snap; | |
3976 | error = dsl_dataset_user_release(name, snap, htag, B_FALSE); | |
3977 | kmem_free(name, namelen); | |
3978 | ||
3979 | /* | |
3980 | * The object can't have been destroyed because we have a hold, | |
3981 | * but it might have been renamed, resulting in ENOENT. Retry | |
3982 | * if we've been requested to do so. | |
3983 | * | |
3984 | * It would be nice if we could use the dsobj all the way | |
3985 | * through and avoid ENOENT entirely. But we might need to | |
3986 | * unmount the snapshot, and there's currently no way to lookup | |
3987 | * a vfsp using a ZFS object id. | |
3988 | */ | |
3989 | } while ((error == ENOENT) && retry); | |
3990 | ||
3991 | return (error); | |
428870ff BB |
3992 | } |
3993 | ||
45d1cae3 BB |
3994 | int |
3995 | dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp) | |
3996 | { | |
3997 | dsl_dataset_t *ds; | |
3998 | int err; | |
3999 | ||
4000 | err = dsl_dataset_hold(dsname, FTAG, &ds); | |
4001 | if (err) | |
4002 | return (err); | |
4003 | ||
4004 | VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP)); | |
4005 | if (ds->ds_phys->ds_userrefs_obj != 0) { | |
4006 | zap_attribute_t *za; | |
4007 | zap_cursor_t zc; | |
4008 | ||
4009 | za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP); | |
4010 | for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset, | |
4011 | ds->ds_phys->ds_userrefs_obj); | |
4012 | zap_cursor_retrieve(&zc, za) == 0; | |
4013 | zap_cursor_advance(&zc)) { | |
4014 | VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name, | |
4015 | za->za_first_integer)); | |
4016 | } | |
4017 | zap_cursor_fini(&zc); | |
4018 | kmem_free(za, sizeof (zap_attribute_t)); | |
4019 | } | |
4020 | dsl_dataset_rele(ds, FTAG); | |
4021 | return (0); | |
4022 | } | |
428870ff BB |
4023 | |
4024 | /* | |
4025 | * Note, this fuction is used as the callback for dmu_objset_find(). We | |
4026 | * always return 0 so that we will continue to find and process | |
4027 | * inconsistent datasets, even if we encounter an error trying to | |
4028 | * process one of them. | |
4029 | */ | |
4030 | /* ARGSUSED */ | |
4031 | int | |
4032 | dsl_destroy_inconsistent(const char *dsname, void *arg) | |
4033 | { | |
4034 | dsl_dataset_t *ds; | |
4035 | ||
4036 | if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) { | |
4037 | if (DS_IS_INCONSISTENT(ds)) | |
4038 | (void) dsl_dataset_destroy(ds, FTAG, B_FALSE); | |
4039 | else | |
4040 | dsl_dataset_disown(ds, FTAG); | |
4041 | } | |
4042 | return (0); | |
4043 | } | |
c28b2279 BB |
4044 | |
4045 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
4046 | EXPORT_SYMBOL(dsl_dataset_hold); | |
4047 | EXPORT_SYMBOL(dsl_dataset_hold_obj); | |
4048 | EXPORT_SYMBOL(dsl_dataset_own); | |
4049 | EXPORT_SYMBOL(dsl_dataset_own_obj); | |
4050 | EXPORT_SYMBOL(dsl_dataset_name); | |
4051 | EXPORT_SYMBOL(dsl_dataset_rele); | |
4052 | EXPORT_SYMBOL(dsl_dataset_disown); | |
4053 | EXPORT_SYMBOL(dsl_dataset_drop_ref); | |
4054 | EXPORT_SYMBOL(dsl_dataset_tryown); | |
4055 | EXPORT_SYMBOL(dsl_dataset_make_exclusive); | |
4056 | EXPORT_SYMBOL(dsl_dataset_create_sync); | |
4057 | EXPORT_SYMBOL(dsl_dataset_create_sync_dd); | |
4058 | EXPORT_SYMBOL(dsl_dataset_destroy); | |
4059 | EXPORT_SYMBOL(dsl_snapshots_destroy); | |
4060 | EXPORT_SYMBOL(dsl_dataset_destroy_check); | |
4061 | EXPORT_SYMBOL(dsl_dataset_destroy_sync); | |
4062 | EXPORT_SYMBOL(dsl_dataset_snapshot_check); | |
4063 | EXPORT_SYMBOL(dsl_dataset_snapshot_sync); | |
4064 | EXPORT_SYMBOL(dsl_dataset_rename); | |
4065 | EXPORT_SYMBOL(dsl_dataset_promote); | |
4066 | EXPORT_SYMBOL(dsl_dataset_clone_swap); | |
4067 | EXPORT_SYMBOL(dsl_dataset_user_hold); | |
4068 | EXPORT_SYMBOL(dsl_dataset_user_release); | |
4069 | EXPORT_SYMBOL(dsl_dataset_user_release_tmp); | |
4070 | EXPORT_SYMBOL(dsl_dataset_get_holds); | |
4071 | EXPORT_SYMBOL(dsl_dataset_get_blkptr); | |
4072 | EXPORT_SYMBOL(dsl_dataset_set_blkptr); | |
4073 | EXPORT_SYMBOL(dsl_dataset_get_spa); | |
4074 | EXPORT_SYMBOL(dsl_dataset_modified_since_lastsnap); | |
4075 | EXPORT_SYMBOL(dsl_dataset_sync); | |
4076 | EXPORT_SYMBOL(dsl_dataset_block_born); | |
4077 | EXPORT_SYMBOL(dsl_dataset_block_kill); | |
4078 | EXPORT_SYMBOL(dsl_dataset_block_freeable); | |
4079 | EXPORT_SYMBOL(dsl_dataset_prev_snap_txg); | |
4080 | EXPORT_SYMBOL(dsl_dataset_dirty); | |
4081 | EXPORT_SYMBOL(dsl_dataset_stats); | |
4082 | EXPORT_SYMBOL(dsl_dataset_fast_stat); | |
4083 | EXPORT_SYMBOL(dsl_dataset_space); | |
4084 | EXPORT_SYMBOL(dsl_dataset_fsid_guid); | |
4085 | EXPORT_SYMBOL(dsl_dsobj_to_dsname); | |
4086 | EXPORT_SYMBOL(dsl_dataset_check_quota); | |
4087 | EXPORT_SYMBOL(dsl_dataset_set_quota); | |
4088 | EXPORT_SYMBOL(dsl_dataset_set_quota_sync); | |
4089 | EXPORT_SYMBOL(dsl_dataset_set_reservation); | |
4090 | EXPORT_SYMBOL(dsl_destroy_inconsistent); | |
4091 | #endif |