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, 2014 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);
306 err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
308 VERIFY3U(0, ==, bpobj_space(&sublist,
309 &used_after, &comp_after, &uncomp_after));
310 bpo->bpo_phys->bpo_bytes -= used_before - used_after;
311 ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
312 bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
313 bpo->bpo_phys->bpo_uncomp -=
314 uncomp_before - uncomp_after;
317 bpobj_close(&sublist);
321 err = dmu_object_free(bpo->bpo_os,
322 objarray[blkoff], tx);
325 bpo->bpo_phys->bpo_num_subobjs--;
326 ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
330 dmu_buf_rele(dbuf, FTAG);
334 VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
335 bpo->bpo_phys->bpo_subobjs,
336 (i + 1) * sizeof (uint64_t), -1ULL, tx));
340 /* If there are no entries, there should be no bytes. */
341 if (!bpobj_hasentries(bpo)) {
342 ASSERT0(bpo->bpo_phys->bpo_bytes);
343 ASSERT0(bpo->bpo_phys->bpo_comp);
344 ASSERT0(bpo->bpo_phys->bpo_uncomp);
347 mutex_exit(&bpo->bpo_lock);
352 * Iterate and remove the entries. If func returns nonzero, iteration
353 * will stop and that entry will not be removed.
356 bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
358 return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
362 * Iterate the entries. If func returns nonzero, iteration will stop.
365 bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
367 return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
371 bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
374 uint64_t used, comp, uncomp, subsubobjs;
376 ASSERT(bpo->bpo_havesubobj);
377 ASSERT(bpo->bpo_havecomp);
378 ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
380 if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
381 bpobj_decr_empty(bpo->bpo_os, tx);
385 VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
386 VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
388 if (!bpobj_hasentries(&subbpo)) {
389 /* No point in having an empty subobj. */
390 bpobj_close(&subbpo);
391 bpobj_free(bpo->bpo_os, subobj, tx);
395 dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
396 if (bpo->bpo_phys->bpo_subobjs == 0) {
397 bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
398 DMU_OT_BPOBJ_SUBOBJ, SPA_OLD_MAXBLOCKSIZE,
402 dmu_object_info_t doi;
403 ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
404 ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
406 mutex_enter(&bpo->bpo_lock);
407 dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
408 bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
409 sizeof (subobj), &subobj, tx);
410 bpo->bpo_phys->bpo_num_subobjs++;
413 * If subobj has only one block of subobjs, then move subobj's
414 * subobjs to bpo's subobj list directly. This reduces
415 * recursion in bpobj_iterate due to nested subobjs.
417 subsubobjs = subbpo.bpo_phys->bpo_subobjs;
418 if (subsubobjs != 0) {
419 dmu_object_info_t doi;
421 VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
422 if (doi.doi_max_offset == doi.doi_data_block_size) {
424 uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
426 VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
427 0, FTAG, &subdb, 0));
429 * Make sure that we are not asking dmu_write()
430 * to write more data than we have in our buffer.
432 VERIFY3U(subdb->db_size, >=,
433 numsubsub * sizeof (subobj));
434 dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
435 bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
436 numsubsub * sizeof (subobj), subdb->db_data, tx);
437 dmu_buf_rele(subdb, FTAG);
438 bpo->bpo_phys->bpo_num_subobjs += numsubsub;
440 dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
441 subbpo.bpo_phys->bpo_subobjs = 0;
442 VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
446 bpo->bpo_phys->bpo_bytes += used;
447 bpo->bpo_phys->bpo_comp += comp;
448 bpo->bpo_phys->bpo_uncomp += uncomp;
449 mutex_exit(&bpo->bpo_lock);
451 bpobj_close(&subbpo);
455 bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
457 blkptr_t stored_bp = *bp;
462 ASSERT(!BP_IS_HOLE(bp));
463 ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
465 if (BP_IS_EMBEDDED(bp)) {
467 * The bpobj will compress better without the payload.
469 * Note that we store EMBEDDED bp's because they have an
470 * uncompressed size, which must be accounted for. An
471 * alternative would be to add their size to bpo_uncomp
472 * without storing the bp, but that would create additional
473 * complications: bpo_uncomp would be inconsistent with the
474 * set of BP's stored, and bpobj_iterate() wouldn't visit
475 * all the space accounted for in the bpobj.
477 bzero(&stored_bp, sizeof (stored_bp));
478 stored_bp.blk_prop = bp->blk_prop;
479 stored_bp.blk_birth = bp->blk_birth;
480 } else if (!BP_GET_DEDUP(bp)) {
481 /* The bpobj will compress better without the checksum */
482 bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
485 /* We never need the fill count. */
486 stored_bp.blk_fill = 0;
488 mutex_enter(&bpo->bpo_lock);
490 offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
491 blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
493 if (bpo->bpo_cached_dbuf == NULL ||
494 offset < bpo->bpo_cached_dbuf->db_offset ||
495 offset >= bpo->bpo_cached_dbuf->db_offset +
496 bpo->bpo_cached_dbuf->db_size) {
497 if (bpo->bpo_cached_dbuf)
498 dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
499 VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
500 offset, bpo, &bpo->bpo_cached_dbuf, 0));
503 dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
504 bparray = bpo->bpo_cached_dbuf->db_data;
505 bparray[blkoff] = stored_bp;
507 dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
508 bpo->bpo_phys->bpo_num_blkptrs++;
509 bpo->bpo_phys->bpo_bytes +=
510 bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
511 if (bpo->bpo_havecomp) {
512 bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
513 bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
515 mutex_exit(&bpo->bpo_lock);
518 struct space_range_arg {
529 space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
531 struct space_range_arg *sra = arg;
533 if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
534 if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
535 sra->used += bp_get_dsize_sync(sra->spa, bp);
537 sra->used += bp_get_dsize(sra->spa, bp);
538 sra->comp += BP_GET_PSIZE(bp);
539 sra->uncomp += BP_GET_UCSIZE(bp);
545 bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
547 mutex_enter(&bpo->bpo_lock);
549 *usedp = bpo->bpo_phys->bpo_bytes;
550 if (bpo->bpo_havecomp) {
551 *compp = bpo->bpo_phys->bpo_comp;
552 *uncompp = bpo->bpo_phys->bpo_uncomp;
553 mutex_exit(&bpo->bpo_lock);
556 mutex_exit(&bpo->bpo_lock);
557 return (bpobj_space_range(bpo, 0, UINT64_MAX,
558 usedp, compp, uncompp));
563 * Return the amount of space in the bpobj which is:
564 * mintxg < blk_birth <= maxtxg
567 bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
568 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
570 struct space_range_arg sra = { 0 };
574 * As an optimization, if they want the whole txg range, just
575 * get bpo_bytes rather than iterating over the bps.
577 if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
578 return (bpobj_space(bpo, usedp, compp, uncompp));
580 sra.spa = dmu_objset_spa(bpo->bpo_os);
584 err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
587 *uncompp = sra.uncomp;