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(MUTEX_HELD(&dl
->dl_lock
));
77 ASSERT(!dl
->dl_oldfmt
);
81 avl_create(&dl
->dl_tree
, dsl_deadlist_compare
,
82 sizeof (dsl_deadlist_entry_t
),
83 offsetof(dsl_deadlist_entry_t
, dle_node
));
84 for (zap_cursor_init(&zc
, dl
->dl_os
, dl
->dl_object
);
85 zap_cursor_retrieve(&zc
, &za
) == 0;
86 zap_cursor_advance(&zc
)) {
87 dsl_deadlist_entry_t
*dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
88 dle
->dle_mintxg
= zfs_strtonum(za
.za_name
, NULL
);
89 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
,
90 za
.za_first_integer
));
91 avl_add(&dl
->dl_tree
, dle
);
94 dl
->dl_havetree
= B_TRUE
;
98 dsl_deadlist_open(dsl_deadlist_t
*dl
, objset_t
*os
, uint64_t object
)
100 dmu_object_info_t doi
;
102 mutex_init(&dl
->dl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
104 dl
->dl_object
= object
;
105 VERIFY3U(0, ==, dmu_bonus_hold(os
, object
, dl
, &dl
->dl_dbuf
));
106 dmu_object_info_from_db(dl
->dl_dbuf
, &doi
);
107 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
108 dmu_buf_rele(dl
->dl_dbuf
, dl
);
110 dl
->dl_oldfmt
= B_TRUE
;
111 VERIFY3U(0, ==, bpobj_open(&dl
->dl_bpobj
, os
, object
));
115 dl
->dl_oldfmt
= B_FALSE
;
116 dl
->dl_phys
= dl
->dl_dbuf
->db_data
;
117 dl
->dl_havetree
= B_FALSE
;
121 dsl_deadlist_close(dsl_deadlist_t
*dl
)
124 dsl_deadlist_entry_t
*dle
;
129 dl
->dl_oldfmt
= B_FALSE
;
130 bpobj_close(&dl
->dl_bpobj
);
134 if (dl
->dl_havetree
) {
135 while ((dle
= avl_destroy_nodes(&dl
->dl_tree
, &cookie
))
137 bpobj_close(&dle
->dle_bpobj
);
138 kmem_free(dle
, sizeof (*dle
));
140 avl_destroy(&dl
->dl_tree
);
142 dmu_buf_rele(dl
->dl_dbuf
, dl
);
143 mutex_destroy(&dl
->dl_lock
);
149 dsl_deadlist_alloc(objset_t
*os
, dmu_tx_t
*tx
)
151 if (spa_version(dmu_objset_spa(os
)) < SPA_VERSION_DEADLISTS
)
152 return (bpobj_alloc(os
, SPA_OLD_MAXBLOCKSIZE
, tx
));
153 return (zap_create(os
, DMU_OT_DEADLIST
, DMU_OT_DEADLIST_HDR
,
154 sizeof (dsl_deadlist_phys_t
), tx
));
158 dsl_deadlist_free(objset_t
*os
, uint64_t dlobj
, dmu_tx_t
*tx
)
160 dmu_object_info_t doi
;
164 VERIFY3U(0, ==, dmu_object_info(os
, dlobj
, &doi
));
165 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
166 bpobj_free(os
, dlobj
, tx
);
170 for (zap_cursor_init(&zc
, os
, dlobj
);
171 zap_cursor_retrieve(&zc
, &za
) == 0;
172 zap_cursor_advance(&zc
)) {
173 uint64_t obj
= za
.za_first_integer
;
174 if (obj
== dmu_objset_pool(os
)->dp_empty_bpobj
)
175 bpobj_decr_empty(os
, tx
);
177 bpobj_free(os
, obj
, tx
);
179 zap_cursor_fini(&zc
);
180 VERIFY3U(0, ==, dmu_object_free(os
, dlobj
, tx
));
184 dle_enqueue(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
185 const blkptr_t
*bp
, dmu_tx_t
*tx
)
187 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
188 if (dle
->dle_bpobj
.bpo_object
==
189 dmu_objset_pool(dl
->dl_os
)->dp_empty_bpobj
) {
190 uint64_t obj
= bpobj_alloc(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
191 bpobj_close(&dle
->dle_bpobj
);
192 bpobj_decr_empty(dl
->dl_os
, tx
);
193 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
194 VERIFY3U(0, ==, zap_update_int_key(dl
->dl_os
, dl
->dl_object
,
195 dle
->dle_mintxg
, obj
, tx
));
197 bpobj_enqueue(&dle
->dle_bpobj
, bp
, tx
);
201 dle_enqueue_subobj(dsl_deadlist_t
*dl
, dsl_deadlist_entry_t
*dle
,
202 uint64_t obj
, dmu_tx_t
*tx
)
204 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
205 if (dle
->dle_bpobj
.bpo_object
!=
206 dmu_objset_pool(dl
->dl_os
)->dp_empty_bpobj
) {
207 bpobj_enqueue_subobj(&dle
->dle_bpobj
, obj
, tx
);
209 bpobj_close(&dle
->dle_bpobj
);
210 bpobj_decr_empty(dl
->dl_os
, tx
);
211 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
212 VERIFY3U(0, ==, zap_update_int_key(dl
->dl_os
, dl
->dl_object
,
213 dle
->dle_mintxg
, obj
, tx
));
218 dsl_deadlist_insert(dsl_deadlist_t
*dl
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
220 dsl_deadlist_entry_t dle_tofind
;
221 dsl_deadlist_entry_t
*dle
;
225 bpobj_enqueue(&dl
->dl_bpobj
, bp
, tx
);
229 mutex_enter(&dl
->dl_lock
);
230 dsl_deadlist_load_tree(dl
);
232 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
233 dl
->dl_phys
->dl_used
+=
234 bp_get_dsize_sync(dmu_objset_spa(dl
->dl_os
), bp
);
235 dl
->dl_phys
->dl_comp
+= BP_GET_PSIZE(bp
);
236 dl
->dl_phys
->dl_uncomp
+= BP_GET_UCSIZE(bp
);
238 dle_tofind
.dle_mintxg
= bp
->blk_birth
;
239 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
241 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
243 dle
= AVL_PREV(&dl
->dl_tree
, dle
);
244 dle_enqueue(dl
, dle
, bp
, tx
);
245 mutex_exit(&dl
->dl_lock
);
249 * Insert new key in deadlist, which must be > all current entries.
250 * mintxg is not inclusive.
253 dsl_deadlist_add_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
256 dsl_deadlist_entry_t
*dle
;
261 dle
= kmem_alloc(sizeof (*dle
), KM_SLEEP
);
262 dle
->dle_mintxg
= mintxg
;
264 mutex_enter(&dl
->dl_lock
);
265 dsl_deadlist_load_tree(dl
);
267 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
268 VERIFY3U(0, ==, bpobj_open(&dle
->dle_bpobj
, dl
->dl_os
, obj
));
269 avl_add(&dl
->dl_tree
, dle
);
271 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, dl
->dl_object
,
273 mutex_exit(&dl
->dl_lock
);
277 * Remove this key, merging its entries into the previous key.
280 dsl_deadlist_remove_key(dsl_deadlist_t
*dl
, uint64_t mintxg
, dmu_tx_t
*tx
)
282 dsl_deadlist_entry_t dle_tofind
;
283 dsl_deadlist_entry_t
*dle
, *dle_prev
;
288 mutex_enter(&dl
->dl_lock
);
289 dsl_deadlist_load_tree(dl
);
291 dle_tofind
.dle_mintxg
= mintxg
;
292 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, NULL
);
293 dle_prev
= AVL_PREV(&dl
->dl_tree
, dle
);
295 dle_enqueue_subobj(dl
, dle_prev
, dle
->dle_bpobj
.bpo_object
, tx
);
297 avl_remove(&dl
->dl_tree
, dle
);
298 bpobj_close(&dle
->dle_bpobj
);
299 kmem_free(dle
, sizeof (*dle
));
301 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
, mintxg
, tx
));
302 mutex_exit(&dl
->dl_lock
);
306 * Walk ds's snapshots to regenerate generate ZAP & AVL.
309 dsl_deadlist_regenerate(objset_t
*os
, uint64_t dlobj
,
310 uint64_t mrs_obj
, dmu_tx_t
*tx
)
313 dsl_pool_t
*dp
= dmu_objset_pool(os
);
315 dsl_deadlist_open(&dl
, os
, dlobj
);
317 dsl_deadlist_close(&dl
);
321 while (mrs_obj
!= 0) {
323 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp
, mrs_obj
, FTAG
, &ds
));
324 dsl_deadlist_add_key(&dl
,
325 dsl_dataset_phys(ds
)->ds_prev_snap_txg
, tx
);
326 mrs_obj
= dsl_dataset_phys(ds
)->ds_prev_snap_obj
;
327 dsl_dataset_rele(ds
, FTAG
);
329 dsl_deadlist_close(&dl
);
333 dsl_deadlist_clone(dsl_deadlist_t
*dl
, uint64_t maxtxg
,
334 uint64_t mrs_obj
, dmu_tx_t
*tx
)
336 dsl_deadlist_entry_t
*dle
;
339 newobj
= dsl_deadlist_alloc(dl
->dl_os
, tx
);
342 dsl_deadlist_regenerate(dl
->dl_os
, newobj
, mrs_obj
, tx
);
346 mutex_enter(&dl
->dl_lock
);
347 dsl_deadlist_load_tree(dl
);
349 for (dle
= avl_first(&dl
->dl_tree
); dle
;
350 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
353 if (dle
->dle_mintxg
>= maxtxg
)
356 obj
= bpobj_alloc_empty(dl
->dl_os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
357 VERIFY3U(0, ==, zap_add_int_key(dl
->dl_os
, newobj
,
358 dle
->dle_mintxg
, obj
, tx
));
360 mutex_exit(&dl
->dl_lock
);
365 dsl_deadlist_space(dsl_deadlist_t
*dl
,
366 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
369 VERIFY3U(0, ==, bpobj_space(&dl
->dl_bpobj
,
370 usedp
, compp
, uncompp
));
374 mutex_enter(&dl
->dl_lock
);
375 *usedp
= dl
->dl_phys
->dl_used
;
376 *compp
= dl
->dl_phys
->dl_comp
;
377 *uncompp
= dl
->dl_phys
->dl_uncomp
;
378 mutex_exit(&dl
->dl_lock
);
382 * return space used in the range (mintxg, maxtxg].
383 * Includes maxtxg, does not include mintxg.
384 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
385 * larger than any bp in the deadlist (eg. UINT64_MAX)).
388 dsl_deadlist_space_range(dsl_deadlist_t
*dl
, uint64_t mintxg
, uint64_t maxtxg
,
389 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
391 dsl_deadlist_entry_t
*dle
;
392 dsl_deadlist_entry_t dle_tofind
;
396 VERIFY3U(0, ==, bpobj_space_range(&dl
->dl_bpobj
,
397 mintxg
, maxtxg
, usedp
, compp
, uncompp
));
401 *usedp
= *compp
= *uncompp
= 0;
403 mutex_enter(&dl
->dl_lock
);
404 dsl_deadlist_load_tree(dl
);
405 dle_tofind
.dle_mintxg
= mintxg
;
406 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
408 * If we don't find this mintxg, there shouldn't be anything
411 ASSERT(dle
!= NULL
||
412 avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
) == NULL
);
414 for (; dle
&& dle
->dle_mintxg
< maxtxg
;
415 dle
= AVL_NEXT(&dl
->dl_tree
, dle
)) {
416 uint64_t used
, comp
, uncomp
;
418 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
419 &used
, &comp
, &uncomp
));
425 mutex_exit(&dl
->dl_lock
);
429 dsl_deadlist_insert_bpobj(dsl_deadlist_t
*dl
, uint64_t obj
, uint64_t birth
,
432 dsl_deadlist_entry_t dle_tofind
;
433 dsl_deadlist_entry_t
*dle
;
435 uint64_t used
, comp
, uncomp
;
438 ASSERT(MUTEX_HELD(&dl
->dl_lock
));
440 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
441 VERIFY3U(0, ==, bpobj_space(&bpo
, &used
, &comp
, &uncomp
));
444 dsl_deadlist_load_tree(dl
);
446 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
447 dl
->dl_phys
->dl_used
+= used
;
448 dl
->dl_phys
->dl_comp
+= comp
;
449 dl
->dl_phys
->dl_uncomp
+= uncomp
;
451 dle_tofind
.dle_mintxg
= birth
;
452 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
454 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_BEFORE
);
455 dle_enqueue_subobj(dl
, dle
, obj
, tx
);
459 dsl_deadlist_insert_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
461 dsl_deadlist_t
*dl
= arg
;
462 dsl_deadlist_insert(dl
, bp
, tx
);
467 * Merge the deadlist pointed to by 'obj' into dl. obj will be left as
471 dsl_deadlist_merge(dsl_deadlist_t
*dl
, uint64_t obj
, dmu_tx_t
*tx
)
476 dsl_deadlist_phys_t
*dlp
;
477 dmu_object_info_t doi
;
479 VERIFY3U(0, ==, dmu_object_info(dl
->dl_os
, obj
, &doi
));
480 if (doi
.doi_type
== DMU_OT_BPOBJ
) {
482 VERIFY3U(0, ==, bpobj_open(&bpo
, dl
->dl_os
, obj
));
483 VERIFY3U(0, ==, bpobj_iterate(&bpo
,
484 dsl_deadlist_insert_cb
, dl
, tx
));
489 mutex_enter(&dl
->dl_lock
);
490 for (zap_cursor_init(&zc
, dl
->dl_os
, obj
);
491 zap_cursor_retrieve(&zc
, &za
) == 0;
492 zap_cursor_advance(&zc
)) {
493 uint64_t mintxg
= zfs_strtonum(za
.za_name
, NULL
);
494 dsl_deadlist_insert_bpobj(dl
, za
.za_first_integer
, mintxg
, tx
);
495 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, obj
, mintxg
, tx
));
497 zap_cursor_fini(&zc
);
499 VERIFY3U(0, ==, dmu_bonus_hold(dl
->dl_os
, obj
, FTAG
, &bonus
));
500 dlp
= bonus
->db_data
;
501 dmu_buf_will_dirty(bonus
, tx
);
502 bzero(dlp
, sizeof (*dlp
));
503 dmu_buf_rele(bonus
, FTAG
);
504 mutex_exit(&dl
->dl_lock
);
508 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
511 dsl_deadlist_move_bpobj(dsl_deadlist_t
*dl
, bpobj_t
*bpo
, uint64_t mintxg
,
514 dsl_deadlist_entry_t dle_tofind
;
515 dsl_deadlist_entry_t
*dle
;
518 ASSERT(!dl
->dl_oldfmt
);
520 mutex_enter(&dl
->dl_lock
);
521 dmu_buf_will_dirty(dl
->dl_dbuf
, tx
);
522 dsl_deadlist_load_tree(dl
);
524 dle_tofind
.dle_mintxg
= mintxg
;
525 dle
= avl_find(&dl
->dl_tree
, &dle_tofind
, &where
);
527 dle
= avl_nearest(&dl
->dl_tree
, where
, AVL_AFTER
);
529 uint64_t used
, comp
, uncomp
;
530 dsl_deadlist_entry_t
*dle_next
;
532 bpobj_enqueue_subobj(bpo
, dle
->dle_bpobj
.bpo_object
, tx
);
534 VERIFY3U(0, ==, bpobj_space(&dle
->dle_bpobj
,
535 &used
, &comp
, &uncomp
));
536 ASSERT3U(dl
->dl_phys
->dl_used
, >=, used
);
537 ASSERT3U(dl
->dl_phys
->dl_comp
, >=, comp
);
538 ASSERT3U(dl
->dl_phys
->dl_uncomp
, >=, uncomp
);
539 dl
->dl_phys
->dl_used
-= used
;
540 dl
->dl_phys
->dl_comp
-= comp
;
541 dl
->dl_phys
->dl_uncomp
-= uncomp
;
543 VERIFY3U(0, ==, zap_remove_int(dl
->dl_os
, dl
->dl_object
,
544 dle
->dle_mintxg
, tx
));
546 dle_next
= AVL_NEXT(&dl
->dl_tree
, dle
);
547 avl_remove(&dl
->dl_tree
, dle
);
548 bpobj_close(&dle
->dle_bpobj
);
549 kmem_free(dle
, sizeof (*dle
));
552 mutex_exit(&dl
->dl_lock
);