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.
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.
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]
22 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 #include <sys/dsl_dataset.h>
29 #include <sys/refcount.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dsl_pool.h>
35 * Deadlist concurrency:
37 * Deadlists can only be modified from the syncing thread.
39 * Except for dsl_deadlist_insert(), it can only be modified with the
40 * dp_config_rwlock held with RW_WRITER.
42 * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
43 * be called concurrently, from open context, with the dl_config_rwlock held
46 * Therefore, we only need to provide locking between dsl_deadlist_insert() and
47 * the accessors, protecting:
48 * dl_phys->dl_used,comp,uncomp
49 * and protecting the dl_tree from being loaded.
50 * The locking is provided by dl_lock. Note that locking on the bpobj_t
51 * provides its own locking, and dl_oldfmt is immutable.
55 dsl_deadlist_compare(const void *arg1
, const void *arg2
)
57 const dsl_deadlist_entry_t
*dle1
= (const dsl_deadlist_entry_t
*)arg1
;
58 const dsl_deadlist_entry_t
*dle2
= (const dsl_deadlist_entry_t
*)arg2
;
60 return (AVL_CMP(dle1
->dle_mintxg
, dle2
->dle_mintxg
));
64 dsl_deadlist_load_tree(dsl_deadlist_t
*dl
)
69 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
71 ASSERT(!dl
->dl_oldfmt
);
75 avl_create(&dl
->dl_tree
, dsl_deadlist_compare
,
76 sizeof (dsl_deadlist_entry_t
),
77 offsetof(dsl_deadlist_entry_t
, dle_node
));
78 for (zap_cursor_init(&zc
, dl
->dl_os
, dl
->dl_object
);
79 zap_cursor_retrieve(&zc
, &za
) == 0;
80 zap_cursor_advance(&zc
)) {
81 dsl_deadlist_entry_t
*dle
;
83 dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
84 dle
->dle_mintxg
= strtonum(za
.za_name
, NULL
);
85 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
,
86 za
.za_first_integer
));
87 avl_add(&dl
->dl_tree
, dle
);
90 dl
->dl_havetree
= B_TRUE
;
94 dsl_deadlist_open(dsl_deadlist_t
*dl
, objset_t
*os
, uint64_t object
)
96 dmu_object_info_t doi
;
98 mutex_init(&dl
->dl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
100 dl
->dl_object
= object
;
101 VERIFY3U(0, ==, dmu_bonus_hold(os
, object
, dl
, &dl
->dl_dbuf
));
102 dmu_object_info_from_db(dl
->dl_dbuf
, &doi
);
103 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
104 dmu_buf_rele(dl
->dl_dbuf
, dl
);
106 dl
->dl_oldfmt
= B_TRUE
;
107 VERIFY3U(0, ==, bpobj_open(&dl
->dl_bpobj
, os
, object
));
111 dl
->dl_oldfmt
= B_FALSE
;
112 dl
->dl_phys
= dl
->dl_dbuf
->db_data
;
113 dl
->dl_havetree
= B_FALSE
;
117 dsl_deadlist_close(dsl_deadlist_t
*dl
)
120 dsl_deadlist_entry_t
*dle
;
125 dl
->dl_oldfmt
= B_FALSE
;
126 bpobj_close(&dl
->dl_bpobj
);
130 if (dl
->dl_havetree
) {
131 while ((dle
= avl_destroy_nodes(&dl
->dl_tree
, &cookie
))
133 bpobj_close(&dle
->dle_bpobj
);
134 kmem_free(dle
, sizeof (*dle
));
136 avl_destroy(&dl
->dl_tree
);
138 dmu_buf_rele(dl
->dl_dbuf
, dl
);
139 mutex_destroy(&dl
->dl_lock
);
145 dsl_deadlist_alloc(objset_t
*os
, dmu_tx_t
*tx
)
147 if (spa_version(dmu_objset_spa(os
)) < SPA_VERSION_DEADLISTS
)
148 return (bpobj_alloc(os
, SPA_OLD_MAXBLOCKSIZE
, tx
));
149 return (zap_create(os
, DMU_OT_DEADLIST
, DMU_OT_DEADLIST_HDR
,
150 sizeof (dsl_deadlist_phys_t
), tx
));
154 dsl_deadlist_free(objset_t
*os
, uint64_t dlobj
, dmu_tx_t
*tx
)
156 dmu_object_info_t doi
;
160 VERIFY3U(0, ==, dmu_object_info(os
, dlobj
, &doi
));
161 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
162 bpobj_free(os
, dlobj
, tx
);
166 for (zap_cursor_init(&zc
, os
, dlobj
);
167 zap_cursor_retrieve(&zc
, &za
) == 0;
168 zap_cursor_advance(&zc
)) {
169 uint64_t obj
= za
.za_first_integer
;
170 if (obj
== dmu_objset_pool(os
)->dp_empty_bpobj
)
171 bpobj_decr_empty(os
, tx
);
173 bpobj_free(os
, obj
, tx
);
175 zap_cursor_fini(&zc
);
176 VERIFY3U(0, ==, dmu_object_free(os
, dlobj
, tx
));
180 dle_enqueue(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
181 const blkptr_t
*bp
, dmu_tx_t
*tx
)
183 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
184 if (dle
->dle_bpobj
.bpo_object
==
185 dmu_objset_pool(dl
->dl_os
)->dp_empty_bpobj
) {
186 uint64_t obj
= bpobj_alloc(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
187 bpobj_close(&dle
->dle_bpobj
);
188 bpobj_decr_empty(dl
->dl_os
, tx
);
189 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
190 VERIFY3U(0, ==, zap_update_int_key(dl
->dl_os
, dl
->dl_object
,
191 dle
->dle_mintxg
, obj
, tx
));
193 bpobj_enqueue(&dle
->dle_bpobj
, bp
, tx
);
197 dle_enqueue_subobj(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
198 uint64_t obj
, dmu_tx_t
*tx
)
200 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
201 if (dle
->dle_bpobj
.bpo_object
!=
202 dmu_objset_pool(dl
->dl_os
)->dp_empty_bpobj
) {
203 bpobj_enqueue_subobj(&dle
->dle_bpobj
, obj
, tx
);
205 bpobj_close(&dle
->dle_bpobj
);
206 bpobj_decr_empty(dl
->dl_os
, tx
);
207 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
208 VERIFY3U(0, ==, zap_update_int_key(dl
->dl_os
, dl
->dl_object
,
209 dle
->dle_mintxg
, obj
, tx
));
214 dsl_deadlist_insert(dsl_deadlist_t
*dl
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
216 dsl_deadlist_entry_t dle_tofind
;
217 dsl_deadlist_entry_t
*dle
;
221 bpobj_enqueue(&dl
->dl_bpobj
, bp
, tx
);
225 mutex_enter(&dl
->dl_lock
);
226 dsl_deadlist_load_tree(dl
);
228 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
229 dl
->dl_phys
->dl_used
+=
230 bp_get_dsize_sync(dmu_objset_spa(dl
->dl_os
), bp
);
231 dl
->dl_phys
->dl_comp
+= BP_GET_PSIZE(bp
);
232 dl
->dl_phys
->dl_uncomp
+= BP_GET_UCSIZE(bp
);
234 dle_tofind
.dle_mintxg
= bp
->blk_birth
;
235 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
237 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
239 dle
= AVL_PREV(&dl
->dl_tree
, dle
);
242 zfs_panic_recover("blkptr at %p has invalid BLK_BIRTH %llu",
243 bp
, (longlong_t
)bp
->blk_birth
);
244 dle
= avl_first(&dl
->dl_tree
);
247 ASSERT3P(dle
, !=, NULL
);
248 dle_enqueue(dl
, dle
, bp
, tx
);
249 mutex_exit(&dl
->dl_lock
);
253 * Insert new key in deadlist, which must be > all current entries.
254 * mintxg is not inclusive.
257 dsl_deadlist_add_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
260 dsl_deadlist_entry_t
*dle
;
265 dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
266 dle
->dle_mintxg
= mintxg
;
268 mutex_enter(&dl
->dl_lock
);
269 dsl_deadlist_load_tree(dl
);
271 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
272 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
273 avl_add(&dl
->dl_tree
, dle
);
275 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, dl
->dl_object
,
277 mutex_exit(&dl
->dl_lock
);
281 * Remove this key, merging its entries into the previous key.
284 dsl_deadlist_remove_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
286 dsl_deadlist_entry_t dle_tofind
;
287 dsl_deadlist_entry_t
*dle
, *dle_prev
;
292 mutex_enter(&dl
->dl_lock
);
293 dsl_deadlist_load_tree(dl
);
295 dle_tofind
.dle_mintxg
= mintxg
;
296 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, NULL
);
297 dle_prev
= AVL_PREV(&dl
->dl_tree
, dle
);
299 dle_enqueue_subobj(dl
, dle_prev
, dle
->dle_bpobj
.bpo_object
, tx
);
301 avl_remove(&dl
->dl_tree
, dle
);
302 bpobj_close(&dle
->dle_bpobj
);
303 kmem_free(dle
, sizeof (*dle
));
305 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
, mintxg
, tx
));
306 mutex_exit(&dl
->dl_lock
);
310 * Walk ds's snapshots to regenerate generate ZAP & AVL.
313 dsl_deadlist_regenerate(objset_t
*os
, uint64_t dlobj
,
314 uint64_t mrs_obj
, dmu_tx_t
*tx
)
317 dsl_pool_t
*dp
= dmu_objset_pool(os
);
319 dsl_deadlist_open(&dl
, os
, dlobj
);
321 dsl_deadlist_close(&dl
);
325 while (mrs_obj
!= 0) {
327 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, mrs_obj
, FTAG
, &ds
));
328 dsl_deadlist_add_key(&dl
,
329 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
330 mrs_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
331 dsl_dataset_rele(ds
, FTAG
);
333 dsl_deadlist_close(&dl
);
337 dsl_deadlist_clone(dsl_deadlist_t
*dl
, uint64_t maxtxg
,
338 uint64_t mrs_obj
, dmu_tx_t
*tx
)
340 dsl_deadlist_entry_t
*dle
;
343 newobj
= dsl_deadlist_alloc(dl
->dl_os
, tx
);
346 dsl_deadlist_regenerate(dl
->dl_os
, newobj
, mrs_obj
, tx
);
350 mutex_enter(&dl
->dl_lock
);
351 dsl_deadlist_load_tree(dl
);
353 for (dle
= avl_first(&dl
->dl_tree
); dle
;
354 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
357 if (dle
->dle_mintxg
>= maxtxg
)
360 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
361 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, newobj
,
362 dle
->dle_mintxg
, obj
, tx
));
364 mutex_exit(&dl
->dl_lock
);
369 dsl_deadlist_space(dsl_deadlist_t
*dl
,
370 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
373 VERIFY3U(0, ==, bpobj_space(&dl
->dl_bpobj
,
374 usedp
, compp
, uncompp
));
378 mutex_enter(&dl
->dl_lock
);
379 *usedp
= dl
->dl_phys
->dl_used
;
380 *compp
= dl
->dl_phys
->dl_comp
;
381 *uncompp
= dl
->dl_phys
->dl_uncomp
;
382 mutex_exit(&dl
->dl_lock
);
386 * return space used in the range (mintxg, maxtxg].
387 * Includes maxtxg, does not include mintxg.
388 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
389 * larger than any bp in the deadlist (eg. UINT64_MAX)).
392 dsl_deadlist_space_range(dsl_deadlist_t
*dl
, uint64_t mintxg
, uint64_t maxtxg
,
393 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
395 dsl_deadlist_entry_t
*dle
;
396 dsl_deadlist_entry_t dle_tofind
;
400 VERIFY3U(0, ==, bpobj_space_range(&dl
->dl_bpobj
,
401 mintxg
, maxtxg
, usedp
, compp
, uncompp
));
405 *usedp
= *compp
= *uncompp
= 0;
407 mutex_enter(&dl
->dl_lock
);
408 dsl_deadlist_load_tree(dl
);
409 dle_tofind
.dle_mintxg
= mintxg
;
410 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
412 * If we don't find this mintxg, there shouldn't be anything
415 ASSERT(dle
!= NULL
||
416 avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
) == NULL
);
418 for (; dle
&& dle
->dle_mintxg
< maxtxg
;
419 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
420 uint64_t used
, comp
, uncomp
;
422 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
423 &used
, &comp
, &uncomp
));
429 mutex_exit(&dl
->dl_lock
);
433 dsl_deadlist_insert_bpobj(dsl_deadlist_t
*dl
, uint64_t obj
, uint64_t birth
,
436 dsl_deadlist_entry_t dle_tofind
;
437 dsl_deadlist_entry_t
*dle
;
439 uint64_t used
, comp
, uncomp
;
442 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
444 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
445 VERIFY3U(0, ==, bpobj_space(&bpo
, &used
, &comp
, &uncomp
));
448 dsl_deadlist_load_tree(dl
);
450 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
451 dl
->dl_phys
->dl_used
+= used
;
452 dl
->dl_phys
->dl_comp
+= comp
;
453 dl
->dl_phys
->dl_uncomp
+= uncomp
;
455 dle_tofind
.dle_mintxg
= birth
;
456 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
458 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
459 dle_enqueue_subobj(dl
, dle
, obj
, tx
);
463 dsl_deadlist_insert_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
465 dsl_deadlist_t
*dl
= arg
;
466 dsl_deadlist_insert(dl
, bp
, tx
);
471 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
475 dsl_deadlist_merge(dsl_deadlist_t
*dl
, uint64_t obj
, dmu_tx_t
*tx
)
480 dsl_deadlist_phys_t
*dlp
;
481 dmu_object_info_t doi
;
483 VERIFY3U(0, ==, dmu_object_info(dl
->dl_os
, obj
, &doi
));
484 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
486 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
487 VERIFY3U(0, ==, bpobj_iterate(&bpo
,
488 dsl_deadlist_insert_cb
, dl
, tx
));
493 mutex_enter(&dl
->dl_lock
);
494 for (zap_cursor_init(&zc
, dl
->dl_os
, obj
);
495 zap_cursor_retrieve(&zc
, &za
) == 0;
496 zap_cursor_advance(&zc
)) {
497 uint64_t mintxg
= strtonum(za
.za_name
, NULL
);
498 dsl_deadlist_insert_bpobj(dl
, za
.za_first_integer
, mintxg
, tx
);
499 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, obj
, mintxg
, tx
));
501 zap_cursor_fini(&zc
);
503 VERIFY3U(0, ==, dmu_bonus_hold(dl
->dl_os
, obj
, FTAG
, &bonus
));
504 dlp
= bonus
->db_data
;
505 dmu_buf_will_dirty(bonus
, tx
);
506 bzero(dlp
, sizeof (*dlp
));
507 dmu_buf_rele(bonus
, FTAG
);
508 mutex_exit(&dl
->dl_lock
);
512 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
515 dsl_deadlist_move_bpobj(dsl_deadlist_t
*dl
, bpobj_t
*bpo
, uint64_t mintxg
,
518 dsl_deadlist_entry_t dle_tofind
;
519 dsl_deadlist_entry_t
*dle
;
522 ASSERT(!dl
->dl_oldfmt
);
524 mutex_enter(&dl
->dl_lock
);
525 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
526 dsl_deadlist_load_tree(dl
);
528 dle_tofind
.dle_mintxg
= mintxg
;
529 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
531 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
);
533 uint64_t used
, comp
, uncomp
;
534 dsl_deadlist_entry_t
*dle_next
;
536 bpobj_enqueue_subobj(bpo
, dle
->dle_bpobj
.bpo_object
, tx
);
538 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
539 &used
, &comp
, &uncomp
));
540 ASSERT3U(dl
->dl_phys
->dl_used
, >=, used
);
541 ASSERT3U(dl
->dl_phys
->dl_comp
, >=, comp
);
542 ASSERT3U(dl
->dl_phys
->dl_uncomp
, >=, uncomp
);
543 dl
->dl_phys
->dl_used
-= used
;
544 dl
->dl_phys
->dl_comp
-= comp
;
545 dl
->dl_phys
->dl_uncomp
-= uncomp
;
547 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
,
548 dle
->dle_mintxg
, tx
));
550 dle_next
= AVL_NEXT(&dl
->dl_tree
, dle
);
551 avl_remove(&dl
->dl_tree
, dle
);
552 bpobj_close(&dle
->dle_bpobj
);
553 kmem_free(dle
, sizeof (*dle
));
556 mutex_exit(&dl
->dl_lock
);