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 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
28 #include <sys/dsl_dataset.h>
30 #include <sys/refcount.h>
32 #include <sys/zfs_context.h>
33 #include <sys/dsl_pool.h>
36 * Deadlist concurrency:
38 * Deadlists can only be modified from the syncing thread.
40 * Except for dsl_deadlist_insert(), it can only be modified with the
41 * dp_config_rwlock held with RW_WRITER.
43 * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
44 * be called concurrently, from open context, with the dl_config_rwlock held
47 * Therefore, we only need to provide locking between dsl_deadlist_insert() and
48 * the accessors, protecting:
49 * dl_phys->dl_used,comp,uncomp
50 * and protecting the dl_tree from being loaded.
51 * The locking is provided by dl_lock. Note that locking on the bpobj_t
52 * provides its own locking, and dl_oldfmt is immutable.
56 dsl_deadlist_compare(const void *arg1
, const void *arg2
)
58 const dsl_deadlist_entry_t
*dle1
= arg1
;
59 const dsl_deadlist_entry_t
*dle2
= arg2
;
61 if (dle1
->dle_mintxg
< dle2
->dle_mintxg
)
63 else if (dle1
->dle_mintxg
> dle2
->dle_mintxg
)
70 dsl_deadlist_load_tree(dsl_deadlist_t
*dl
)
75 ASSERT(!dl
->dl_oldfmt
);
79 avl_create(&dl
->dl_tree
, dsl_deadlist_compare
,
80 sizeof (dsl_deadlist_entry_t
),
81 offsetof(dsl_deadlist_entry_t
, dle_node
));
82 for (zap_cursor_init(&zc
, dl
->dl_os
, dl
->dl_object
);
83 zap_cursor_retrieve(&zc
, &za
) == 0;
84 zap_cursor_advance(&zc
)) {
85 dsl_deadlist_entry_t
*dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
86 dle
->dle_mintxg
= strtonum(za
.za_name
, NULL
);
87 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
,
88 za
.za_first_integer
));
89 avl_add(&dl
->dl_tree
, dle
);
92 dl
->dl_havetree
= B_TRUE
;
96 dsl_deadlist_open(dsl_deadlist_t
*dl
, objset_t
*os
, uint64_t object
)
98 dmu_object_info_t doi
;
100 mutex_init(&dl
->dl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
102 dl
->dl_object
= object
;
103 VERIFY3U(0, ==, dmu_bonus_hold(os
, object
, dl
, &dl
->dl_dbuf
));
104 dmu_object_info_from_db(dl
->dl_dbuf
, &doi
);
105 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
106 dmu_buf_rele(dl
->dl_dbuf
, dl
);
108 dl
->dl_oldfmt
= B_TRUE
;
109 VERIFY3U(0, ==, bpobj_open(&dl
->dl_bpobj
, os
, object
));
113 dl
->dl_oldfmt
= B_FALSE
;
114 dl
->dl_phys
= dl
->dl_dbuf
->db_data
;
115 dl
->dl_havetree
= B_FALSE
;
119 dsl_deadlist_close(dsl_deadlist_t
*dl
)
122 dsl_deadlist_entry_t
*dle
;
127 dl
->dl_oldfmt
= B_FALSE
;
128 bpobj_close(&dl
->dl_bpobj
);
132 if (dl
->dl_havetree
) {
133 while ((dle
= avl_destroy_nodes(&dl
->dl_tree
, &cookie
))
135 bpobj_close(&dle
->dle_bpobj
);
136 kmem_free(dle
, sizeof (*dle
));
138 avl_destroy(&dl
->dl_tree
);
140 dmu_buf_rele(dl
->dl_dbuf
, dl
);
141 mutex_destroy(&dl
->dl_lock
);
147 dsl_deadlist_alloc(objset_t
*os
, dmu_tx_t
*tx
)
149 if (spa_version(dmu_objset_spa(os
)) < SPA_VERSION_DEADLISTS
)
150 return (bpobj_alloc(os
, SPA_OLD_MAXBLOCKSIZE
, tx
));
151 return (zap_create(os
, DMU_OT_DEADLIST
, DMU_OT_DEADLIST_HDR
,
152 sizeof (dsl_deadlist_phys_t
), tx
));
156 dsl_deadlist_free(objset_t
*os
, uint64_t dlobj
, dmu_tx_t
*tx
)
158 dmu_object_info_t doi
;
162 VERIFY3U(0, ==, dmu_object_info(os
, dlobj
, &doi
));
163 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
164 bpobj_free(os
, dlobj
, tx
);
168 for (zap_cursor_init(&zc
, os
, dlobj
);
169 zap_cursor_retrieve(&zc
, &za
) == 0;
170 zap_cursor_advance(&zc
)) {
171 uint64_t obj
= za
.za_first_integer
;
172 if (obj
== dmu_objset_pool(os
)->dp_empty_bpobj
)
173 bpobj_decr_empty(os
, tx
);
175 bpobj_free(os
, obj
, tx
);
177 zap_cursor_fini(&zc
);
178 VERIFY3U(0, ==, dmu_object_free(os
, dlobj
, tx
));
182 dle_enqueue(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
183 const blkptr_t
*bp
, dmu_tx_t
*tx
)
185 if (dle
->dle_bpobj
.bpo_object
==
186 dmu_objset_pool(dl
->dl_os
)->dp_empty_bpobj
) {
187 uint64_t obj
= bpobj_alloc(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
188 bpobj_close(&dle
->dle_bpobj
);
189 bpobj_decr_empty(dl
->dl_os
, tx
);
190 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
191 VERIFY3U(0, ==, zap_update_int_key(dl
->dl_os
, dl
->dl_object
,
192 dle
->dle_mintxg
, obj
, tx
));
194 bpobj_enqueue(&dle
->dle_bpobj
, bp
, tx
);
198 dle_enqueue_subobj(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
199 uint64_t obj
, dmu_tx_t
*tx
)
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 dsl_deadlist_load_tree(dl
);
227 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
228 mutex_enter(&dl
->dl_lock
);
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
);
233 mutex_exit(&dl
->dl_lock
);
235 dle_tofind
.dle_mintxg
= bp
->blk_birth
;
236 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
238 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
240 dle
= AVL_PREV(&dl
->dl_tree
, dle
);
241 dle_enqueue(dl
, dle
, bp
, tx
);
245 * Insert new key in deadlist, which must be > all current entries.
246 * mintxg is not inclusive.
249 dsl_deadlist_add_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
252 dsl_deadlist_entry_t
*dle
;
257 dsl_deadlist_load_tree(dl
);
259 dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
260 dle
->dle_mintxg
= mintxg
;
261 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
262 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
263 avl_add(&dl
->dl_tree
, dle
);
265 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, dl
->dl_object
,
270 * Remove this key, merging its entries into the previous key.
273 dsl_deadlist_remove_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
275 dsl_deadlist_entry_t dle_tofind
;
276 dsl_deadlist_entry_t
*dle
, *dle_prev
;
281 dsl_deadlist_load_tree(dl
);
283 dle_tofind
.dle_mintxg
= mintxg
;
284 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, NULL
);
285 dle_prev
= AVL_PREV(&dl
->dl_tree
, dle
);
287 dle_enqueue_subobj(dl
, dle_prev
, dle
->dle_bpobj
.bpo_object
, tx
);
289 avl_remove(&dl
->dl_tree
, dle
);
290 bpobj_close(&dle
->dle_bpobj
);
291 kmem_free(dle
, sizeof (*dle
));
293 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
, mintxg
, tx
));
297 * Walk ds's snapshots to regenerate generate ZAP & AVL.
300 dsl_deadlist_regenerate(objset_t
*os
, uint64_t dlobj
,
301 uint64_t mrs_obj
, dmu_tx_t
*tx
)
304 dsl_pool_t
*dp
= dmu_objset_pool(os
);
306 dsl_deadlist_open(&dl
, os
, dlobj
);
308 dsl_deadlist_close(&dl
);
312 while (mrs_obj
!= 0) {
314 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, mrs_obj
, FTAG
, &ds
));
315 dsl_deadlist_add_key(&dl
,
316 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
317 mrs_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
318 dsl_dataset_rele(ds
, FTAG
);
320 dsl_deadlist_close(&dl
);
324 dsl_deadlist_clone(dsl_deadlist_t
*dl
, uint64_t maxtxg
,
325 uint64_t mrs_obj
, dmu_tx_t
*tx
)
327 dsl_deadlist_entry_t
*dle
;
330 newobj
= dsl_deadlist_alloc(dl
->dl_os
, tx
);
333 dsl_deadlist_regenerate(dl
->dl_os
, newobj
, mrs_obj
, tx
);
337 dsl_deadlist_load_tree(dl
);
339 for (dle
= avl_first(&dl
->dl_tree
); dle
;
340 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
343 if (dle
->dle_mintxg
>= maxtxg
)
346 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
347 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, newobj
,
348 dle
->dle_mintxg
, obj
, tx
));
354 dsl_deadlist_space(dsl_deadlist_t
*dl
,
355 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
358 VERIFY3U(0, ==, bpobj_space(&dl
->dl_bpobj
,
359 usedp
, compp
, uncompp
));
363 mutex_enter(&dl
->dl_lock
);
364 *usedp
= dl
->dl_phys
->dl_used
;
365 *compp
= dl
->dl_phys
->dl_comp
;
366 *uncompp
= dl
->dl_phys
->dl_uncomp
;
367 mutex_exit(&dl
->dl_lock
);
371 * return space used in the range (mintxg, maxtxg].
372 * Includes maxtxg, does not include mintxg.
373 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
374 * larger than any bp in the deadlist (eg. UINT64_MAX)).
377 dsl_deadlist_space_range(dsl_deadlist_t
*dl
, uint64_t mintxg
, uint64_t maxtxg
,
378 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
380 dsl_deadlist_entry_t
*dle
;
381 dsl_deadlist_entry_t dle_tofind
;
385 VERIFY3U(0, ==, bpobj_space_range(&dl
->dl_bpobj
,
386 mintxg
, maxtxg
, usedp
, compp
, uncompp
));
390 *usedp
= *compp
= *uncompp
= 0;
392 mutex_enter(&dl
->dl_lock
);
393 dsl_deadlist_load_tree(dl
);
394 dle_tofind
.dle_mintxg
= mintxg
;
395 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
397 * If we don't find this mintxg, there shouldn't be anything
400 ASSERT(dle
!= NULL
||
401 avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
) == NULL
);
403 for (; dle
&& dle
->dle_mintxg
< maxtxg
;
404 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
405 uint64_t used
, comp
, uncomp
;
407 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
408 &used
, &comp
, &uncomp
));
414 mutex_exit(&dl
->dl_lock
);
418 dsl_deadlist_insert_bpobj(dsl_deadlist_t
*dl
, uint64_t obj
, uint64_t birth
,
421 dsl_deadlist_entry_t dle_tofind
;
422 dsl_deadlist_entry_t
*dle
;
424 uint64_t used
, comp
, uncomp
;
427 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
428 VERIFY3U(0, ==, bpobj_space(&bpo
, &used
, &comp
, &uncomp
));
431 dsl_deadlist_load_tree(dl
);
433 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
434 mutex_enter(&dl
->dl_lock
);
435 dl
->dl_phys
->dl_used
+= used
;
436 dl
->dl_phys
->dl_comp
+= comp
;
437 dl
->dl_phys
->dl_uncomp
+= uncomp
;
438 mutex_exit(&dl
->dl_lock
);
440 dle_tofind
.dle_mintxg
= birth
;
441 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
443 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
444 dle_enqueue_subobj(dl
, dle
, obj
, tx
);
448 dsl_deadlist_insert_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
450 dsl_deadlist_t
*dl
= arg
;
451 dsl_deadlist_insert(dl
, bp
, tx
);
456 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
460 dsl_deadlist_merge(dsl_deadlist_t
*dl
, uint64_t obj
, dmu_tx_t
*tx
)
465 dsl_deadlist_phys_t
*dlp
;
466 dmu_object_info_t doi
;
468 VERIFY3U(0, ==, dmu_object_info(dl
->dl_os
, obj
, &doi
));
469 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
471 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
472 VERIFY3U(0, ==, bpobj_iterate(&bpo
,
473 dsl_deadlist_insert_cb
, dl
, tx
));
478 for (zap_cursor_init(&zc
, dl
->dl_os
, obj
);
479 zap_cursor_retrieve(&zc
, &za
) == 0;
480 zap_cursor_advance(&zc
)) {
481 uint64_t mintxg
= strtonum(za
.za_name
, NULL
);
482 dsl_deadlist_insert_bpobj(dl
, za
.za_first_integer
, mintxg
, tx
);
483 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, obj
, mintxg
, tx
));
485 zap_cursor_fini(&zc
);
487 VERIFY3U(0, ==, dmu_bonus_hold(dl
->dl_os
, obj
, FTAG
, &bonus
));
488 dlp
= bonus
->db_data
;
489 dmu_buf_will_dirty(bonus
, tx
);
490 bzero(dlp
, sizeof (*dlp
));
491 dmu_buf_rele(bonus
, FTAG
);
495 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
498 dsl_deadlist_move_bpobj(dsl_deadlist_t
*dl
, bpobj_t
*bpo
, uint64_t mintxg
,
501 dsl_deadlist_entry_t dle_tofind
;
502 dsl_deadlist_entry_t
*dle
;
505 ASSERT(!dl
->dl_oldfmt
);
506 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
507 dsl_deadlist_load_tree(dl
);
509 dle_tofind
.dle_mintxg
= mintxg
;
510 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
512 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
);
514 uint64_t used
, comp
, uncomp
;
515 dsl_deadlist_entry_t
*dle_next
;
517 bpobj_enqueue_subobj(bpo
, dle
->dle_bpobj
.bpo_object
, tx
);
519 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
520 &used
, &comp
, &uncomp
));
521 mutex_enter(&dl
->dl_lock
);
522 ASSERT3U(dl
->dl_phys
->dl_used
, >=, used
);
523 ASSERT3U(dl
->dl_phys
->dl_comp
, >=, comp
);
524 ASSERT3U(dl
->dl_phys
->dl_uncomp
, >=, uncomp
);
525 dl
->dl_phys
->dl_used
-= used
;
526 dl
->dl_phys
->dl_comp
-= comp
;
527 dl
->dl_phys
->dl_uncomp
-= uncomp
;
528 mutex_exit(&dl
->dl_lock
);
530 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
,
531 dle
->dle_mintxg
, tx
));
533 dle_next
= AVL_NEXT(&dl
->dl_tree
, dle
);
534 avl_remove(&dl
->dl_tree
, dle
);
535 bpobj_close(&dle
->dle_bpobj
);
536 kmem_free(dle
, sizeof (*dle
));