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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
26 #include <sys/bpobj.h>
27 #include <sys/zfs_context.h>
28 #include <sys/refcount.h>
29 #include <sys/dsl_pool.h>
30 #include <sys/zfeature.h>
34 * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
37 bpobj_alloc_empty(objset_t
*os
, int blocksize
, dmu_tx_t
*tx
)
39 spa_t
*spa
= dmu_objset_spa(os
);
40 dsl_pool_t
*dp
= dmu_objset_pool(os
);
42 if (spa_feature_is_enabled(spa
, SPA_FEATURE_EMPTY_BPOBJ
)) {
43 if (!spa_feature_is_active(spa
, SPA_FEATURE_EMPTY_BPOBJ
)) {
44 ASSERT0(dp
->dp_empty_bpobj
);
46 bpobj_alloc(os
, SPA_OLD_MAXBLOCKSIZE
, tx
);
48 DMU_POOL_DIRECTORY_OBJECT
,
49 DMU_POOL_EMPTY_BPOBJ
, sizeof (uint64_t), 1,
50 &dp
->dp_empty_bpobj
, tx
) == 0);
52 spa_feature_incr(spa
, SPA_FEATURE_EMPTY_BPOBJ
, tx
);
53 ASSERT(dp
->dp_empty_bpobj
!= 0);
54 return (dp
->dp_empty_bpobj
);
56 return (bpobj_alloc(os
, blocksize
, tx
));
61 bpobj_decr_empty(objset_t
*os
, dmu_tx_t
*tx
)
63 dsl_pool_t
*dp
= dmu_objset_pool(os
);
65 spa_feature_decr(dmu_objset_spa(os
), SPA_FEATURE_EMPTY_BPOBJ
, tx
);
66 if (!spa_feature_is_active(dmu_objset_spa(os
),
67 SPA_FEATURE_EMPTY_BPOBJ
)) {
68 VERIFY3U(0, ==, zap_remove(dp
->dp_meta_objset
,
69 DMU_POOL_DIRECTORY_OBJECT
,
70 DMU_POOL_EMPTY_BPOBJ
, tx
));
71 VERIFY3U(0, ==, dmu_object_free(os
, dp
->dp_empty_bpobj
, tx
));
72 dp
->dp_empty_bpobj
= 0;
77 bpobj_alloc(objset_t
*os
, int blocksize
, dmu_tx_t
*tx
)
81 if (spa_version(dmu_objset_spa(os
)) < SPA_VERSION_BPOBJ_ACCOUNT
)
83 else if (spa_version(dmu_objset_spa(os
)) < SPA_VERSION_DEADLISTS
)
86 size
= sizeof (bpobj_phys_t
);
88 return (dmu_object_alloc(os
, DMU_OT_BPOBJ
, blocksize
,
89 DMU_OT_BPOBJ_HDR
, size
, tx
));
93 bpobj_free(objset_t
*os
, uint64_t obj
, dmu_tx_t
*tx
)
97 dmu_object_info_t doi
;
99 dmu_buf_t
*dbuf
= NULL
;
101 ASSERT(obj
!= dmu_objset_pool(os
)->dp_empty_bpobj
);
102 VERIFY3U(0, ==, bpobj_open(&bpo
, os
, obj
));
104 mutex_enter(&bpo
.bpo_lock
);
106 if (!bpo
.bpo_havesubobj
|| bpo
.bpo_phys
->bpo_subobjs
== 0)
109 VERIFY3U(0, ==, dmu_object_info(os
, bpo
.bpo_phys
->bpo_subobjs
, &doi
));
110 epb
= doi
.doi_data_block_size
/ sizeof (uint64_t);
112 for (i
= bpo
.bpo_phys
->bpo_num_subobjs
- 1; i
>= 0; i
--) {
114 uint64_t offset
, blkoff
;
116 offset
= i
* sizeof (uint64_t);
117 blkoff
= P2PHASE(i
, epb
);
119 if (dbuf
== NULL
|| dbuf
->db_offset
> offset
) {
121 dmu_buf_rele(dbuf
, FTAG
);
122 VERIFY3U(0, ==, dmu_buf_hold(os
,
123 bpo
.bpo_phys
->bpo_subobjs
, offset
, FTAG
, &dbuf
, 0));
126 ASSERT3U(offset
, >=, dbuf
->db_offset
);
127 ASSERT3U(offset
, <, dbuf
->db_offset
+ dbuf
->db_size
);
129 objarray
= dbuf
->db_data
;
130 bpobj_free(os
, objarray
[blkoff
], tx
);
133 dmu_buf_rele(dbuf
, FTAG
);
136 VERIFY3U(0, ==, dmu_object_free(os
, bpo
.bpo_phys
->bpo_subobjs
, tx
));
139 mutex_exit(&bpo
.bpo_lock
);
142 VERIFY3U(0, ==, dmu_object_free(os
, obj
, tx
));
146 bpobj_open(bpobj_t
*bpo
, objset_t
*os
, uint64_t object
)
148 dmu_object_info_t doi
;
151 err
= dmu_object_info(os
, object
, &doi
);
155 bzero(bpo
, sizeof (*bpo
));
156 mutex_init(&bpo
->bpo_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
158 ASSERT(bpo
->bpo_dbuf
== NULL
);
159 ASSERT(bpo
->bpo_phys
== NULL
);
161 ASSERT3U(doi
.doi_type
, ==, DMU_OT_BPOBJ
);
162 ASSERT3U(doi
.doi_bonus_type
, ==, DMU_OT_BPOBJ_HDR
);
164 err
= dmu_bonus_hold(os
, object
, bpo
, &bpo
->bpo_dbuf
);
169 bpo
->bpo_object
= object
;
170 bpo
->bpo_epb
= doi
.doi_data_block_size
>> SPA_BLKPTRSHIFT
;
171 bpo
->bpo_havecomp
= (doi
.doi_bonus_size
> BPOBJ_SIZE_V0
);
172 bpo
->bpo_havesubobj
= (doi
.doi_bonus_size
> BPOBJ_SIZE_V1
);
173 bpo
->bpo_phys
= bpo
->bpo_dbuf
->db_data
;
178 bpobj_close(bpobj_t
*bpo
)
180 /* Lame workaround for closing a bpobj that was never opened. */
181 if (bpo
->bpo_object
== 0)
184 dmu_buf_rele(bpo
->bpo_dbuf
, bpo
);
185 if (bpo
->bpo_cached_dbuf
!= NULL
)
186 dmu_buf_rele(bpo
->bpo_cached_dbuf
, bpo
);
187 bpo
->bpo_dbuf
= NULL
;
188 bpo
->bpo_phys
= NULL
;
189 bpo
->bpo_cached_dbuf
= NULL
;
192 mutex_destroy(&bpo
->bpo_lock
);
196 bpobj_hasentries(bpobj_t
*bpo
)
198 return (bpo
->bpo_phys
->bpo_num_blkptrs
!= 0 ||
199 (bpo
->bpo_havesubobj
&& bpo
->bpo_phys
->bpo_num_subobjs
!= 0));
203 bpobj_iterate_impl(bpobj_t
*bpo
, bpobj_itor_t func
, void *arg
, dmu_tx_t
*tx
,
206 dmu_object_info_t doi
;
210 dmu_buf_t
*dbuf
= NULL
;
212 mutex_enter(&bpo
->bpo_lock
);
215 dmu_buf_will_dirty(bpo
->bpo_dbuf
, tx
);
217 for (i
= bpo
->bpo_phys
->bpo_num_blkptrs
- 1; i
>= 0; i
--) {
220 uint64_t offset
, blkoff
;
222 offset
= i
* sizeof (blkptr_t
);
223 blkoff
= P2PHASE(i
, bpo
->bpo_epb
);
225 if (dbuf
== NULL
|| dbuf
->db_offset
> offset
) {
227 dmu_buf_rele(dbuf
, FTAG
);
228 err
= dmu_buf_hold(bpo
->bpo_os
, bpo
->bpo_object
, offset
,
234 ASSERT3U(offset
, >=, dbuf
->db_offset
);
235 ASSERT3U(offset
, <, dbuf
->db_offset
+ dbuf
->db_size
);
237 bparray
= dbuf
->db_data
;
238 bp
= &bparray
[blkoff
];
239 err
= func(arg
, bp
, tx
);
243 bpo
->bpo_phys
->bpo_bytes
-=
244 bp_get_dsize_sync(dmu_objset_spa(bpo
->bpo_os
), bp
);
245 ASSERT3S(bpo
->bpo_phys
->bpo_bytes
, >=, 0);
246 if (bpo
->bpo_havecomp
) {
247 bpo
->bpo_phys
->bpo_comp
-= BP_GET_PSIZE(bp
);
248 bpo
->bpo_phys
->bpo_uncomp
-= BP_GET_UCSIZE(bp
);
250 bpo
->bpo_phys
->bpo_num_blkptrs
--;
251 ASSERT3S(bpo
->bpo_phys
->bpo_num_blkptrs
, >=, 0);
255 dmu_buf_rele(dbuf
, FTAG
);
259 VERIFY3U(0, ==, dmu_free_range(bpo
->bpo_os
, bpo
->bpo_object
,
260 (i
+ 1) * sizeof (blkptr_t
), -1ULL, tx
));
262 if (err
|| !bpo
->bpo_havesubobj
|| bpo
->bpo_phys
->bpo_subobjs
== 0)
265 ASSERT(bpo
->bpo_havecomp
);
266 err
= dmu_object_info(bpo
->bpo_os
, bpo
->bpo_phys
->bpo_subobjs
, &doi
);
268 mutex_exit(&bpo
->bpo_lock
);
271 ASSERT3U(doi
.doi_type
, ==, DMU_OT_BPOBJ_SUBOBJ
);
272 epb
= doi
.doi_data_block_size
/ sizeof (uint64_t);
274 for (i
= bpo
->bpo_phys
->bpo_num_subobjs
- 1; i
>= 0; i
--) {
276 uint64_t offset
, blkoff
;
278 uint64_t used_before
, comp_before
, uncomp_before
;
279 uint64_t used_after
, comp_after
, uncomp_after
;
281 offset
= i
* sizeof (uint64_t);
282 blkoff
= P2PHASE(i
, epb
);
284 if (dbuf
== NULL
|| dbuf
->db_offset
> offset
) {
286 dmu_buf_rele(dbuf
, FTAG
);
287 err
= dmu_buf_hold(bpo
->bpo_os
,
288 bpo
->bpo_phys
->bpo_subobjs
, offset
, FTAG
, &dbuf
, 0);
293 ASSERT3U(offset
, >=, dbuf
->db_offset
);
294 ASSERT3U(offset
, <, dbuf
->db_offset
+ dbuf
->db_size
);
296 objarray
= dbuf
->db_data
;
297 err
= bpobj_open(&sublist
, bpo
->bpo_os
, objarray
[blkoff
]);
301 err
= bpobj_space(&sublist
,
302 &used_before
, &comp_before
, &uncomp_before
);
304 bpobj_close(&sublist
);
308 err
= bpobj_iterate_impl(&sublist
, func
, arg
, tx
, free
);
310 VERIFY3U(0, ==, bpobj_space(&sublist
,
311 &used_after
, &comp_after
, &uncomp_after
));
312 bpo
->bpo_phys
->bpo_bytes
-= used_before
- used_after
;
313 ASSERT3S(bpo
->bpo_phys
->bpo_bytes
, >=, 0);
314 bpo
->bpo_phys
->bpo_comp
-= comp_before
- comp_after
;
315 bpo
->bpo_phys
->bpo_uncomp
-=
316 uncomp_before
- uncomp_after
;
319 bpobj_close(&sublist
);
323 err
= dmu_object_free(bpo
->bpo_os
,
324 objarray
[blkoff
], tx
);
327 bpo
->bpo_phys
->bpo_num_subobjs
--;
328 ASSERT3S(bpo
->bpo_phys
->bpo_num_subobjs
, >=, 0);
332 dmu_buf_rele(dbuf
, FTAG
);
336 VERIFY3U(0, ==, dmu_free_range(bpo
->bpo_os
,
337 bpo
->bpo_phys
->bpo_subobjs
,
338 (i
+ 1) * sizeof (uint64_t), -1ULL, tx
));
342 /* If there are no entries, there should be no bytes. */
343 if (!bpobj_hasentries(bpo
)) {
344 ASSERT0(bpo
->bpo_phys
->bpo_bytes
);
345 ASSERT0(bpo
->bpo_phys
->bpo_comp
);
346 ASSERT0(bpo
->bpo_phys
->bpo_uncomp
);
349 mutex_exit(&bpo
->bpo_lock
);
354 * Iterate and remove the entries. If func returns nonzero, iteration
355 * will stop and that entry will not be removed.
358 bpobj_iterate(bpobj_t
*bpo
, bpobj_itor_t func
, void *arg
, dmu_tx_t
*tx
)
360 return (bpobj_iterate_impl(bpo
, func
, arg
, tx
, B_TRUE
));
364 * Iterate the entries. If func returns nonzero, iteration will stop.
367 bpobj_iterate_nofree(bpobj_t
*bpo
, bpobj_itor_t func
, void *arg
, dmu_tx_t
*tx
)
369 return (bpobj_iterate_impl(bpo
, func
, arg
, tx
, B_FALSE
));
373 bpobj_enqueue_subobj(bpobj_t
*bpo
, uint64_t subobj
, dmu_tx_t
*tx
)
376 uint64_t used
, comp
, uncomp
, subsubobjs
;
377 ASSERTV(dmu_object_info_t doi
);
379 ASSERT(bpo
->bpo_havesubobj
);
380 ASSERT(bpo
->bpo_havecomp
);
381 ASSERT(bpo
->bpo_object
!= dmu_objset_pool(bpo
->bpo_os
)->dp_empty_bpobj
);
383 if (subobj
== dmu_objset_pool(bpo
->bpo_os
)->dp_empty_bpobj
) {
384 bpobj_decr_empty(bpo
->bpo_os
, tx
);
388 VERIFY3U(0, ==, bpobj_open(&subbpo
, bpo
->bpo_os
, subobj
));
389 VERIFY3U(0, ==, bpobj_space(&subbpo
, &used
, &comp
, &uncomp
));
391 if (!bpobj_hasentries(&subbpo
)) {
392 /* No point in having an empty subobj. */
393 bpobj_close(&subbpo
);
394 bpobj_free(bpo
->bpo_os
, subobj
, tx
);
398 mutex_enter(&bpo
->bpo_lock
);
399 dmu_buf_will_dirty(bpo
->bpo_dbuf
, tx
);
400 if (bpo
->bpo_phys
->bpo_subobjs
== 0) {
401 bpo
->bpo_phys
->bpo_subobjs
= dmu_object_alloc(bpo
->bpo_os
,
402 DMU_OT_BPOBJ_SUBOBJ
, SPA_OLD_MAXBLOCKSIZE
,
406 ASSERT0(dmu_object_info(bpo
->bpo_os
, bpo
->bpo_phys
->bpo_subobjs
, &doi
));
407 ASSERT3U(doi
.doi_type
, ==, DMU_OT_BPOBJ_SUBOBJ
);
409 dmu_write(bpo
->bpo_os
, bpo
->bpo_phys
->bpo_subobjs
,
410 bpo
->bpo_phys
->bpo_num_subobjs
* sizeof (subobj
),
411 sizeof (subobj
), &subobj
, tx
);
412 bpo
->bpo_phys
->bpo_num_subobjs
++;
415 * If subobj has only one block of subobjs, then move subobj's
416 * subobjs to bpo's subobj list directly. This reduces
417 * recursion in bpobj_iterate due to nested subobjs.
419 subsubobjs
= subbpo
.bpo_phys
->bpo_subobjs
;
420 if (subsubobjs
!= 0) {
421 dmu_object_info_t doi
;
423 VERIFY3U(0, ==, dmu_object_info(bpo
->bpo_os
, subsubobjs
, &doi
));
424 if (doi
.doi_max_offset
== doi
.doi_data_block_size
) {
426 uint64_t numsubsub
= subbpo
.bpo_phys
->bpo_num_subobjs
;
428 VERIFY3U(0, ==, dmu_buf_hold(bpo
->bpo_os
, subsubobjs
,
429 0, FTAG
, &subdb
, 0));
431 * Make sure that we are not asking dmu_write()
432 * to write more data than we have in our buffer.
434 VERIFY3U(subdb
->db_size
, >=,
435 numsubsub
* sizeof (subobj
));
436 dmu_write(bpo
->bpo_os
, bpo
->bpo_phys
->bpo_subobjs
,
437 bpo
->bpo_phys
->bpo_num_subobjs
* sizeof (subobj
),
438 numsubsub
* sizeof (subobj
), subdb
->db_data
, tx
);
439 dmu_buf_rele(subdb
, FTAG
);
440 bpo
->bpo_phys
->bpo_num_subobjs
+= numsubsub
;
442 dmu_buf_will_dirty(subbpo
.bpo_dbuf
, tx
);
443 subbpo
.bpo_phys
->bpo_subobjs
= 0;
444 VERIFY3U(0, ==, dmu_object_free(bpo
->bpo_os
,
448 bpo
->bpo_phys
->bpo_bytes
+= used
;
449 bpo
->bpo_phys
->bpo_comp
+= comp
;
450 bpo
->bpo_phys
->bpo_uncomp
+= uncomp
;
451 mutex_exit(&bpo
->bpo_lock
);
453 bpobj_close(&subbpo
);
457 bpobj_enqueue(bpobj_t
*bpo
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
459 blkptr_t stored_bp
= *bp
;
464 ASSERT(!BP_IS_HOLE(bp
));
465 ASSERT(bpo
->bpo_object
!= dmu_objset_pool(bpo
->bpo_os
)->dp_empty_bpobj
);
467 if (BP_IS_EMBEDDED(bp
)) {
469 * The bpobj will compress better without the payload.
471 * Note that we store EMBEDDED bp's because they have an
472 * uncompressed size, which must be accounted for. An
473 * alternative would be to add their size to bpo_uncomp
474 * without storing the bp, but that would create additional
475 * complications: bpo_uncomp would be inconsistent with the
476 * set of BP's stored, and bpobj_iterate() wouldn't visit
477 * all the space accounted for in the bpobj.
479 bzero(&stored_bp
, sizeof (stored_bp
));
480 stored_bp
.blk_prop
= bp
->blk_prop
;
481 stored_bp
.blk_birth
= bp
->blk_birth
;
482 } else if (!BP_GET_DEDUP(bp
)) {
483 /* The bpobj will compress better without the checksum */
484 bzero(&stored_bp
.blk_cksum
, sizeof (stored_bp
.blk_cksum
));
487 /* We never need the fill count. */
488 stored_bp
.blk_fill
= 0;
490 mutex_enter(&bpo
->bpo_lock
);
492 offset
= bpo
->bpo_phys
->bpo_num_blkptrs
* sizeof (stored_bp
);
493 blkoff
= P2PHASE(bpo
->bpo_phys
->bpo_num_blkptrs
, bpo
->bpo_epb
);
495 if (bpo
->bpo_cached_dbuf
== NULL
||
496 offset
< bpo
->bpo_cached_dbuf
->db_offset
||
497 offset
>= bpo
->bpo_cached_dbuf
->db_offset
+
498 bpo
->bpo_cached_dbuf
->db_size
) {
499 if (bpo
->bpo_cached_dbuf
)
500 dmu_buf_rele(bpo
->bpo_cached_dbuf
, bpo
);
501 VERIFY3U(0, ==, dmu_buf_hold(bpo
->bpo_os
, bpo
->bpo_object
,
502 offset
, bpo
, &bpo
->bpo_cached_dbuf
, 0));
505 dmu_buf_will_dirty(bpo
->bpo_cached_dbuf
, tx
);
506 bparray
= bpo
->bpo_cached_dbuf
->db_data
;
507 bparray
[blkoff
] = stored_bp
;
509 dmu_buf_will_dirty(bpo
->bpo_dbuf
, tx
);
510 bpo
->bpo_phys
->bpo_num_blkptrs
++;
511 bpo
->bpo_phys
->bpo_bytes
+=
512 bp_get_dsize_sync(dmu_objset_spa(bpo
->bpo_os
), bp
);
513 if (bpo
->bpo_havecomp
) {
514 bpo
->bpo_phys
->bpo_comp
+= BP_GET_PSIZE(bp
);
515 bpo
->bpo_phys
->bpo_uncomp
+= BP_GET_UCSIZE(bp
);
517 mutex_exit(&bpo
->bpo_lock
);
520 struct space_range_arg
{
531 space_range_cb(void *arg
, const blkptr_t
*bp
, dmu_tx_t
*tx
)
533 struct space_range_arg
*sra
= arg
;
535 if (bp
->blk_birth
> sra
->mintxg
&& bp
->blk_birth
<= sra
->maxtxg
) {
536 if (dsl_pool_sync_context(spa_get_dsl(sra
->spa
)))
537 sra
->used
+= bp_get_dsize_sync(sra
->spa
, bp
);
539 sra
->used
+= bp_get_dsize(sra
->spa
, bp
);
540 sra
->comp
+= BP_GET_PSIZE(bp
);
541 sra
->uncomp
+= BP_GET_UCSIZE(bp
);
547 bpobj_space(bpobj_t
*bpo
, uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
549 mutex_enter(&bpo
->bpo_lock
);
551 *usedp
= bpo
->bpo_phys
->bpo_bytes
;
552 if (bpo
->bpo_havecomp
) {
553 *compp
= bpo
->bpo_phys
->bpo_comp
;
554 *uncompp
= bpo
->bpo_phys
->bpo_uncomp
;
555 mutex_exit(&bpo
->bpo_lock
);
558 mutex_exit(&bpo
->bpo_lock
);
559 return (bpobj_space_range(bpo
, 0, UINT64_MAX
,
560 usedp
, compp
, uncompp
));
565 * Return the amount of space in the bpobj which is:
566 * mintxg < blk_birth <= maxtxg
569 bpobj_space_range(bpobj_t
*bpo
, uint64_t mintxg
, uint64_t maxtxg
,
570 uint64_t *usedp
, uint64_t *compp
, uint64_t *uncompp
)
572 struct space_range_arg sra
= { 0 };
576 * As an optimization, if they want the whole txg range, just
577 * get bpo_bytes rather than iterating over the bps.
579 if (mintxg
< TXG_INITIAL
&& maxtxg
== UINT64_MAX
&& bpo
->bpo_havecomp
)
580 return (bpobj_space(bpo
, usedp
, compp
, uncompp
));
582 sra
.spa
= dmu_objset_spa(bpo
->bpo_os
);
586 err
= bpobj_iterate_nofree(bpo
, space_range_cb
, &sra
, NULL
);
589 *uncompp
= sra
.uncomp
;