[mirror_zfs.git] / module / zfs / bpobj.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 */

#include <sys/bpobj.h>
#include <sys/zfs_context.h>
#include <sys/refcount.h>

uint64_t
bpobj_alloc(objset_t *os, int blocksize, dmu_tx_t *tx)
{
	int size;

	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
		size = BPOBJ_SIZE_V0;
	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
		size = BPOBJ_SIZE_V1;
	else
		size = sizeof (bpobj_phys_t);

	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
	    DMU_OT_BPOBJ_HDR, size, tx));
}

void
bpobj_free(objset_t *os, uint64_t obj, dmu_tx_t *tx)
{
	int64_t i;
	bpobj_t bpo;
	dmu_object_info_t doi;
	int epb;
	dmu_buf_t *dbuf = NULL;

	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));

	mutex_enter(&bpo.bpo_lock);

	if (!bpo.bpo_havesubobj || bpo.bpo_phys->bpo_subobjs == 0)
		goto out;

	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
	epb = doi.doi_data_block_size / sizeof (uint64_t);

	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
		uint64_t *objarray;
		uint64_t offset, blkoff;

		offset = i * sizeof (uint64_t);
		blkoff = P2PHASE(i, epb);

		if (dbuf == NULL || dbuf->db_offset > offset) {
			if (dbuf)
				dmu_buf_rele(dbuf, FTAG);
			VERIFY3U(0, ==, dmu_buf_hold(os,
			    bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
		}

		ASSERT3U(offset, >=, dbuf->db_offset);
		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);

		objarray = dbuf->db_data;
		bpobj_free(os, objarray[blkoff], tx);
	}
	if (dbuf) {
		dmu_buf_rele(dbuf, FTAG);
		dbuf = NULL;
	}
	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));

out:
	mutex_exit(&bpo.bpo_lock);
	bpobj_close(&bpo);

	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
}

int
bpobj_open(bpobj_t *bpo, objset_t *os, uint64_t object)
{
	dmu_object_info_t doi;
	int err;

	err = dmu_object_info(os, object, &doi);
	if (err)
		return (err);

	bzero(bpo, sizeof (*bpo));
	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);

	ASSERT(bpo->bpo_dbuf == NULL);
	ASSERT(bpo->bpo_phys == NULL);
	ASSERT(object != 0);
	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);

	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
	if (err)
		return (err);

	bpo->bpo_os = os;
	bpo->bpo_object = object;
	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
	return (0);
}

void
bpobj_close(bpobj_t *bpo)
{
	/* Lame workaround for closing a bpobj that was never opened. */
	if (bpo->bpo_object == 0)
		return;

	dmu_buf_rele(bpo->bpo_dbuf, bpo);
	if (bpo->bpo_cached_dbuf != NULL)
		dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	bpo->bpo_dbuf = NULL;
	bpo->bpo_phys = NULL;
	bpo->bpo_cached_dbuf = NULL;
	bpo->bpo_object = 0;

	mutex_destroy(&bpo->bpo_lock);
}

static int
bpobj_iterate_impl(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx,
    boolean_t free)
{
	dmu_object_info_t doi;
	int epb;
	int64_t i;
	int err = 0;
	dmu_buf_t *dbuf = NULL;

	mutex_enter(&bpo->bpo_lock);

	if (free)
		dmu_buf_will_dirty(bpo->bpo_dbuf, tx);

	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
		blkptr_t *bparray;
		blkptr_t *bp;
		uint64_t offset, blkoff;

		offset = i * sizeof (blkptr_t);
		blkoff = P2PHASE(i, bpo->bpo_epb);

		if (dbuf == NULL || dbuf->db_offset > offset) {
			if (dbuf)
				dmu_buf_rele(dbuf, FTAG);
			err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
			    FTAG, &dbuf, 0);
			if (err)
				break;
		}

		ASSERT3U(offset, >=, dbuf->db_offset);
		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);

		bparray = dbuf->db_data;
		bp = &bparray[blkoff];
		err = func(arg, bp, tx);
		if (err)
			break;
		if (free) {
			bpo->bpo_phys->bpo_bytes -=
			    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
			if (bpo->bpo_havecomp) {
				bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
				bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
			}
			bpo->bpo_phys->bpo_num_blkptrs--;
			ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
		}
	}
	if (dbuf) {
		dmu_buf_rele(dbuf, FTAG);
		dbuf = NULL;
	}
	if (free) {
		i++;
		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
		    i * sizeof (blkptr_t), -1ULL, tx));
	}
	if (err || !bpo->bpo_havesubobj || bpo->bpo_phys->bpo_subobjs == 0)
		goto out;

	ASSERT(bpo->bpo_havecomp);
	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
	if (err) {
		mutex_exit(&bpo->bpo_lock);
		return (err);
	}
	epb = doi.doi_data_block_size / sizeof (uint64_t);

	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
		uint64_t *objarray;
		uint64_t offset, blkoff;
		bpobj_t sublist;
		uint64_t used_before, comp_before, uncomp_before;
		uint64_t used_after, comp_after, uncomp_after;

		offset = i * sizeof (uint64_t);
		blkoff = P2PHASE(i, epb);

		if (dbuf == NULL || dbuf->db_offset > offset) {
			if (dbuf)
				dmu_buf_rele(dbuf, FTAG);
			err = dmu_buf_hold(bpo->bpo_os,
			    bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
			if (err)
				break;
		}

		ASSERT3U(offset, >=, dbuf->db_offset);
		ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);

		objarray = dbuf->db_data;
		err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
		if (err)
			break;
		if (free) {
			err = bpobj_space(&sublist,
			    &used_before, &comp_before, &uncomp_before);
			if (err)
				break;
		}
		err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
		if (free) {
			VERIFY3U(0, ==, bpobj_space(&sublist,
			    &used_after, &comp_after, &uncomp_after));
			bpo->bpo_phys->bpo_bytes -= used_before - used_after;
			ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
			bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
			bpo->bpo_phys->bpo_uncomp -=
			    uncomp_before - uncomp_after;
		}

		bpobj_close(&sublist);
		if (err)
			break;
		if (free) {
			err = dmu_object_free(bpo->bpo_os,
			    objarray[blkoff], tx);
			if (err)
				break;
			bpo->bpo_phys->bpo_num_subobjs--;
			ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
		}
	}
	if (dbuf) {
		dmu_buf_rele(dbuf, FTAG);
		dbuf = NULL;
	}
	if (free) {
		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
		    bpo->bpo_phys->bpo_subobjs,
		    (i + 1) * sizeof (uint64_t), -1ULL, tx));
	}

out:
	/* If there are no entries, there should be no bytes. */
	ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 ||
	    (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) ||
	    bpo->bpo_phys->bpo_bytes == 0);

	mutex_exit(&bpo->bpo_lock);
	return (err);
}

/*
 * Iterate and remove the entries.  If func returns nonzero, iteration
 * will stop and that entry will not be removed.
 */
int
bpobj_iterate(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
{
	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
}

/*
 * Iterate the entries.  If func returns nonzero, iteration will stop.
 */
int
bpobj_iterate_nofree(bpobj_t *bpo, bpobj_itor_t func, void *arg, dmu_tx_t *tx)
{
	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
}

void
bpobj_enqueue_subobj(bpobj_t *bpo, uint64_t subobj, dmu_tx_t *tx)
{
	bpobj_t subbpo;
	uint64_t used, comp, uncomp, subsubobjs;

	ASSERT(bpo->bpo_havesubobj);
	ASSERT(bpo->bpo_havecomp);

	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));

	if (used == 0) {
		/* No point in having an empty subobj. */
		bpobj_close(&subbpo);
		bpobj_free(bpo->bpo_os, subobj, tx);
		return;
	}

	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	if (bpo->bpo_phys->bpo_subobjs == 0) {
		bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
		    DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
	}

	mutex_enter(&bpo->bpo_lock);
	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	    sizeof (subobj), &subobj, tx);
	bpo->bpo_phys->bpo_num_subobjs++;

	/*
	 * If subobj has only one block of subobjs, then move subobj's
	 * subobjs to bpo's subobj list directly.  This reduces
	 * recursion in bpobj_iterate due to nested subobjs.
	 */
	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
	if (subsubobjs != 0) {
		dmu_object_info_t doi;

		VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
		if (doi.doi_max_offset == doi.doi_data_block_size) {
			dmu_buf_t *subdb;
			uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;

			VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
			    0, FTAG, &subdb, 0));
			dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
			    bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
			    numsubsub * sizeof (subobj), subdb->db_data, tx);
			dmu_buf_rele(subdb, FTAG);
			bpo->bpo_phys->bpo_num_subobjs += numsubsub;

			dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
			subbpo.bpo_phys->bpo_subobjs = 0;
			VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
			    subsubobjs, tx));
		}
	}
	bpo->bpo_phys->bpo_bytes += used;
	bpo->bpo_phys->bpo_comp += comp;
	bpo->bpo_phys->bpo_uncomp += uncomp;
	mutex_exit(&bpo->bpo_lock);

	bpobj_close(&subbpo);
}

void
bpobj_enqueue(bpobj_t *bpo, const blkptr_t *bp, dmu_tx_t *tx)
{
	blkptr_t stored_bp = *bp;
	uint64_t offset;
	int blkoff;
	blkptr_t *bparray;

	ASSERT(!BP_IS_HOLE(bp));

	/* We never need the fill count. */
	stored_bp.blk_fill = 0;

	/* The bpobj will compress better if we can leave off the checksum */
	if (!BP_GET_DEDUP(bp))
		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));

	mutex_enter(&bpo->bpo_lock);

	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);

	if (bpo->bpo_cached_dbuf == NULL ||
	    offset < bpo->bpo_cached_dbuf->db_offset ||
	    offset >= bpo->bpo_cached_dbuf->db_offset +
	    bpo->bpo_cached_dbuf->db_size) {
		if (bpo->bpo_cached_dbuf)
			dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
		VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
		    offset, bpo, &bpo->bpo_cached_dbuf, 0));
	}

	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
	bparray = bpo->bpo_cached_dbuf->db_data;
	bparray[blkoff] = stored_bp;

	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	bpo->bpo_phys->bpo_num_blkptrs++;
	bpo->bpo_phys->bpo_bytes +=
	    bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	if (bpo->bpo_havecomp) {
		bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
		bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
	}
	mutex_exit(&bpo->bpo_lock);
}

struct space_range_arg {
	spa_t *spa;
	uint64_t mintxg;
	uint64_t maxtxg;
	uint64_t used;
	uint64_t comp;
	uint64_t uncomp;
};

/* ARGSUSED */
static int
space_range_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
	struct space_range_arg *sra = arg;

	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
		sra->used += bp_get_dsize_sync(sra->spa, bp);
		sra->comp += BP_GET_PSIZE(bp);
		sra->uncomp += BP_GET_UCSIZE(bp);
	}
	return (0);
}

int
bpobj_space(bpobj_t *bpo, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
{
	mutex_enter(&bpo->bpo_lock);

	*usedp = bpo->bpo_phys->bpo_bytes;
	if (bpo->bpo_havecomp) {
		*compp = bpo->bpo_phys->bpo_comp;
		*uncompp = bpo->bpo_phys->bpo_uncomp;
		mutex_exit(&bpo->bpo_lock);
		return (0);
	} else {
		mutex_exit(&bpo->bpo_lock);
		return (bpobj_space_range(bpo, 0, UINT64_MAX,
		    usedp, compp, uncompp));
	}
}

/*
 * Return the amount of space in the bpobj which is:
 * mintxg < blk_birth <= maxtxg
 */
int
bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
{
	struct space_range_arg sra = { 0 };
	int err;

	/*
	 * As an optimization, if they want the whole txg range, just
	 * get bpo_bytes rather than iterating over the bps.
	 */
	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
		return (bpobj_space(bpo, usedp, compp, uncompp));

	sra.spa = dmu_objset_spa(bpo->bpo_os);
	sra.mintxg = mintxg;
	sra.maxtxg = maxtxg;

	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
	*usedp = sra.used;
	*compp = sra.comp;
	*uncompp = sra.uncomp;
	return (err);
}
Commit	Line	Data
428870ff BB	1	/*
	2	* CDDL HEADER START
	3	*
	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.
	7	*
	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.
	12	*
	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]
	18	*
	19	* CDDL HEADER END
	20	*/
	21	/*
	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
	23	*/
	24
	25	#include <sys/bpobj.h>
	26	#include <sys/zfs_context.h>
	27	#include <sys/refcount.h>
	28
	29	uint64_t
	30	bpobj_alloc(objset_t os, int blocksize, dmu_tx_t tx)
	31	{
	32	int size;
	33
	34	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
	35	size = BPOBJ_SIZE_V0;
	36	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
	37	size = BPOBJ_SIZE_V1;
	38	else
	39	size = sizeof (bpobj_phys_t);
	40
	41	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
	42	DMU_OT_BPOBJ_HDR, size, tx));
	43	}
	44
	45	void
	46	bpobj_free(objset_t os, uint64_t obj, dmu_tx_t tx)
	47	{
	48	int64_t i;
	49	bpobj_t bpo;
	50	dmu_object_info_t doi;
	51	int epb;
	52	dmu_buf_t *dbuf = NULL;
	53
	54	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
	55
	56	mutex_enter(&bpo.bpo_lock);
	57
	58	if (!bpo.bpo_havesubobj \|\| bpo.bpo_phys->bpo_subobjs == 0)
	59	goto out;
	60
	61	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
	62	epb = doi.doi_data_block_size / sizeof (uint64_t);
	63
	64	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
65	uint64_t *objarray;
66	uint64_t offset, blkoff;
67
68	offset = i * sizeof (uint64_t);
69	blkoff = P2PHASE(i, epb);
70
71	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
72	if (dbuf)
73	dmu_buf_rele(dbuf, FTAG);
74	VERIFY3U(0, ==, dmu_buf_hold(os,
75	bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
76	}
77
78	ASSERT3U(offset, >=, dbuf->db_offset);
79	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
80
81	objarray = dbuf->db_data;
82	bpobj_free(os, objarray[blkoff], tx);
83	}
84	if (dbuf) {
85	dmu_buf_rele(dbuf, FTAG);
86	dbuf = NULL;
87	}
88	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
89
90	out:
91	mutex_exit(&bpo.bpo_lock);
92	bpobj_close(&bpo);
93
94	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
95	}
96
97	int
98	bpobj_open(bpobj_t bpo, objset_t os, uint64_t object)
99	{
100	dmu_object_info_t doi;
101	int err;
102
103	err = dmu_object_info(os, object, &doi);
104	if (err)
105	return (err);
106
107	bzero(bpo, sizeof (*bpo));
108	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
109
110	ASSERT(bpo->bpo_dbuf == NULL);
111	ASSERT(bpo->bpo_phys == NULL);
112	ASSERT(object != 0);
113	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
114	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
115
572e2857 BB	116	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
	117	if (err)
	118	return (err);
	119
428870ff BB	120	bpo->bpo_os = os;
	121	bpo->bpo_object = object;
	122	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
	123	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
	124	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
428870ff BB	125	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
	126	return (0);
	127	}
	128
	129	void
	130	bpobj_close(bpobj_t *bpo)
	131	{
	132	/* Lame workaround for closing a bpobj that was never opened. */
	133	if (bpo->bpo_object == 0)
	134	return;
	135
	136	dmu_buf_rele(bpo->bpo_dbuf, bpo);
	137	if (bpo->bpo_cached_dbuf != NULL)
	138	dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	139	bpo->bpo_dbuf = NULL;
	140	bpo->bpo_phys = NULL;
	141	bpo->bpo_cached_dbuf = NULL;
572e2857	142	bpo->bpo_object = 0;
428870ff BB	143
	144	mutex_destroy(&bpo->bpo_lock);
	145	}
	146
	147	static int
	148	bpobj_iterate_impl(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx,
	149	boolean_t free)
	150	{
	151	dmu_object_info_t doi;
	152	int epb;
	153	int64_t i;
	154	int err = 0;
	155	dmu_buf_t *dbuf = NULL;
	156
	157	mutex_enter(&bpo->bpo_lock);
	158
	159	if (free)
	160	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	161
	162	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
	163	blkptr_t *bparray;
	164	blkptr_t *bp;
	165	uint64_t offset, blkoff;
	166
	167	offset = i * sizeof (blkptr_t);
	168	blkoff = P2PHASE(i, bpo->bpo_epb);
	169
	170	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	171	if (dbuf)
	172	dmu_buf_rele(dbuf, FTAG);
	173	err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
	174	FTAG, &dbuf, 0);
	175	if (err)
	176	break;
	177	}
	178
	179	ASSERT3U(offset, >=, dbuf->db_offset);
	180	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	181
	182	bparray = dbuf->db_data;
	183	bp = &bparray[blkoff];
	184	err = func(arg, bp, tx);
	185	if (err)
	186	break;
	187	if (free) {
	188	bpo->bpo_phys->bpo_bytes -=
	189	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	190	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
	191	if (bpo->bpo_havecomp) {
	192	bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
	193	bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
	194	}
	195	bpo->bpo_phys->bpo_num_blkptrs--;
	196	ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
	197	}
	198	}
	199	if (dbuf) {
	200	dmu_buf_rele(dbuf, FTAG);
	201	dbuf = NULL;
	202	}
	203	if (free) {
	204	i++;
	205	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
	206	i * sizeof (blkptr_t), -1ULL, tx));
207	}
208	if (err \|\| !bpo->bpo_havesubobj \|\| bpo->bpo_phys->bpo_subobjs == 0)
209	goto out;
210
211	ASSERT(bpo->bpo_havecomp);
212	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
572e2857 BB	213	if (err) {
572e2857 BB	214	mutex_exit(&bpo->bpo_lock);
428870ff	215	return (err);
572e2857	216	}
428870ff BB	217	epb = doi.doi_data_block_size / sizeof (uint64_t);
	218
	219	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
	220	uint64_t *objarray;
	221	uint64_t offset, blkoff;
	222	bpobj_t sublist;
	223	uint64_t used_before, comp_before, uncomp_before;
	224	uint64_t used_after, comp_after, uncomp_after;
	225
	226	offset = i * sizeof (uint64_t);
	227	blkoff = P2PHASE(i, epb);
	228
	229	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	230	if (dbuf)
	231	dmu_buf_rele(dbuf, FTAG);
	232	err = dmu_buf_hold(bpo->bpo_os,
	233	bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
	234	if (err)
	235	break;
	236	}
	237
	238	ASSERT3U(offset, >=, dbuf->db_offset);
	239	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	240
	241	objarray = dbuf->db_data;
	242	err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
	243	if (err)
	244	break;
	245	if (free) {
	246	err = bpobj_space(&sublist,
	247	&used_before, &comp_before, &uncomp_before);
	248	if (err)
	249	break;
	250	}
	251	err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
	252	if (free) {
	253	VERIFY3U(0, ==, bpobj_space(&sublist,
	254	&used_after, &comp_after, &uncomp_after));
	255	bpo->bpo_phys->bpo_bytes -= used_before - used_after;
	256	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
572e2857	257	bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
428870ff BB	258	bpo->bpo_phys->bpo_uncomp -=
	259	uncomp_before - uncomp_after;
	260	}
	261
	262	bpobj_close(&sublist);
	263	if (err)
	264	break;
	265	if (free) {
	266	err = dmu_object_free(bpo->bpo_os,
	267	objarray[blkoff], tx);
	268	if (err)
	269	break;
	270	bpo->bpo_phys->bpo_num_subobjs--;
	271	ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
	272	}
	273	}
	274	if (dbuf) {
	275	dmu_buf_rele(dbuf, FTAG);
	276	dbuf = NULL;
	277	}
	278	if (free) {
	279	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
	280	bpo->bpo_phys->bpo_subobjs,
	281	(i + 1) * sizeof (uint64_t), -1ULL, tx));
	282	}
	283
	284	out:
	285	/* If there are no entries, there should be no bytes. */
	286	ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 \|\|
	287	(bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) \|\|
	288	bpo->bpo_phys->bpo_bytes == 0);
	289
	290	mutex_exit(&bpo->bpo_lock);
	291	return (err);
	292	}
	293
	294	/*
	295	* Iterate and remove the entries. If func returns nonzero, iteration
	296	* will stop and that entry will not be removed.
	297	*/
	298	int
	299	bpobj_iterate(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	300	{
	301	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
	302	}
	303
	304	/*
	305	* Iterate the entries. If func returns nonzero, iteration will stop.
	306	*/
	307	int
	308	bpobj_iterate_nofree(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	309	{
	310	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
	311	}
	312
	313	void
	314	bpobj_enqueue_subobj(bpobj_t bpo, uint64_t subobj, dmu_tx_t tx)
	315	{
	316	bpobj_t subbpo;
572e2857	317	uint64_t used, comp, uncomp, subsubobjs;
428870ff BB	318
	319	ASSERT(bpo->bpo_havesubobj);
	320	ASSERT(bpo->bpo_havecomp);
	321
	322	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
	323	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
428870ff BB	324
	325	if (used == 0) {
	326	/* No point in having an empty subobj. */
572e2857	327	bpobj_close(&subbpo);
428870ff BB	328	bpobj_free(bpo->bpo_os, subobj, tx);
	329	return;
	330	}
	331
	332	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	333	if (bpo->bpo_phys->bpo_subobjs == 0) {
	334	bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
	335	DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
	336	}
	337
	338	mutex_enter(&bpo->bpo_lock);
	339	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	340	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	341	sizeof (subobj), &subobj, tx);
	342	bpo->bpo_phys->bpo_num_subobjs++;
572e2857 BB	343
	344	/*
	345	* If subobj has only one block of subobjs, then move subobj's
	346	* subobjs to bpo's subobj list directly. This reduces
	347	* recursion in bpobj_iterate due to nested subobjs.
	348	*/
	349	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
	350	if (subsubobjs != 0) {
	351	dmu_object_info_t doi;
	352
	353	VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
	354	if (doi.doi_max_offset == doi.doi_data_block_size) {
	355	dmu_buf_t *subdb;
	356	uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
	357
	358	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
	359	0, FTAG, &subdb, 0));
	360	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	361	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	362	numsubsub * sizeof (subobj), subdb->db_data, tx);
	363	dmu_buf_rele(subdb, FTAG);
	364	bpo->bpo_phys->bpo_num_subobjs += numsubsub;
	365
	366	dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
	367	subbpo.bpo_phys->bpo_subobjs = 0;
	368	VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
	369	subsubobjs, tx));
	370	}
	371	}
428870ff BB	372	bpo->bpo_phys->bpo_bytes += used;
	373	bpo->bpo_phys->bpo_comp += comp;
	374	bpo->bpo_phys->bpo_uncomp += uncomp;
	375	mutex_exit(&bpo->bpo_lock);
572e2857 BB	376
572e2857 BB	377	bpobj_close(&subbpo);
428870ff BB	378	}
	379
	380	void
	381	bpobj_enqueue(bpobj_t bpo, const blkptr_t bp, dmu_tx_t *tx)
	382	{
	383	blkptr_t stored_bp = *bp;
	384	uint64_t offset;
	385	int blkoff;
	386	blkptr_t *bparray;
	387
	388	ASSERT(!BP_IS_HOLE(bp));
	389
	390	/* We never need the fill count. */
	391	stored_bp.blk_fill = 0;
	392
	393	/* The bpobj will compress better if we can leave off the checksum */
	394	if (!BP_GET_DEDUP(bp))
	395	bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
	396
	397	mutex_enter(&bpo->bpo_lock);
	398
	399	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
	400	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
	401
	402	if (bpo->bpo_cached_dbuf == NULL \|\|
	403	offset < bpo->bpo_cached_dbuf->db_offset \|\|
	404	offset >= bpo->bpo_cached_dbuf->db_offset +
	405	bpo->bpo_cached_dbuf->db_size) {
	406	if (bpo->bpo_cached_dbuf)
	407	dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	408	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
	409	offset, bpo, &bpo->bpo_cached_dbuf, 0));
	410	}
	411
	412	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
	413	bparray = bpo->bpo_cached_dbuf->db_data;
	414	bparray[blkoff] = stored_bp;
	415
	416	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	417	bpo->bpo_phys->bpo_num_blkptrs++;
	418	bpo->bpo_phys->bpo_bytes +=
	419	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	420	if (bpo->bpo_havecomp) {
	421	bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
	422	bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
	423	}
	424	mutex_exit(&bpo->bpo_lock);
	425	}
	426
	427	struct space_range_arg {
	428	spa_t *spa;
	429	uint64_t mintxg;
	430	uint64_t maxtxg;
	431	uint64_t used;
	432	uint64_t comp;
	433	uint64_t uncomp;
	434	};
	435
	436	/* ARGSUSED */
	437	static int
	438	space_range_cb(void arg, const blkptr_t bp, dmu_tx_t *tx)
	439	{
	440	struct space_range_arg *sra = arg;
	441
442	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
443	sra->used += bp_get_dsize_sync(sra->spa, bp);
444	sra->comp += BP_GET_PSIZE(bp);
445	sra->uncomp += BP_GET_UCSIZE(bp);
446	}
447	return (0);
448	}
449
450	int
451	bpobj_space(bpobj_t bpo, uint64_t usedp, uint64_t compp, uint64_t uncompp)
452	{
453	mutex_enter(&bpo->bpo_lock);
454
455	*usedp = bpo->bpo_phys->bpo_bytes;
456	if (bpo->bpo_havecomp) {
457	*compp = bpo->bpo_phys->bpo_comp;
458	*uncompp = bpo->bpo_phys->bpo_uncomp;
459	mutex_exit(&bpo->bpo_lock);
460	return (0);
461	} else {
462	mutex_exit(&bpo->bpo_lock);
463	return (bpobj_space_range(bpo, 0, UINT64_MAX,
464	usedp, compp, uncompp));
465	}
466	}
467
468	/*
469	* Return the amount of space in the bpobj which is:
470	* mintxg < blk_birth <= maxtxg
471	*/
472	int
473	bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
474	uint64_t usedp, uint64_t compp, uint64_t *uncompp)
475	{
476	struct space_range_arg sra = { 0 };
477	int err;
478
479	/*
480	* As an optimization, if they want the whole txg range, just
481	* get bpo_bytes rather than iterating over the bps.
482	*/
483	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
484	return (bpobj_space(bpo, usedp, compp, uncompp));
485
486	sra.spa = dmu_objset_spa(bpo->bpo_os);
487	sra.mintxg = mintxg;
488	sra.maxtxg = maxtxg;
489
490	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
491	*usedp = sra.used;
492	*compp = sra.comp;
493	*uncompp = sra.uncomp;
494	return (err);
495	}