[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.
 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
 * Copyright (c) 2017 Datto Inc.
 */

#include <sys/bpobj.h>
#include <sys/zfs_context.h>
#include <sys/refcount.h>
#include <sys/dsl_pool.h>
#include <sys/zfeature.h>
#include <sys/zap.h>

/*
 * Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
 */
uint64_t
bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
{
	spa_t *spa = dmu_objset_spa(os);
	dsl_pool_t *dp = dmu_objset_pool(os);

	if (spa_feature_is_enabled(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
		if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
			ASSERT0(dp->dp_empty_bpobj);
			dp->dp_empty_bpobj =
			    bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx);
			VERIFY(zap_add(os,
			    DMU_POOL_DIRECTORY_OBJECT,
			    DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
			    &dp->dp_empty_bpobj, tx) == 0);
		}
		spa_feature_incr(spa, SPA_FEATURE_EMPTY_BPOBJ, tx);
		ASSERT(dp->dp_empty_bpobj != 0);
		return (dp->dp_empty_bpobj);
	} else {
		return (bpobj_alloc(os, blocksize, tx));
	}
}

void
bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
{
	dsl_pool_t *dp = dmu_objset_pool(os);

	spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
	if (!spa_feature_is_active(dmu_objset_spa(os),
	    SPA_FEATURE_EMPTY_BPOBJ)) {
		VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
		    DMU_POOL_DIRECTORY_OBJECT,
		    DMU_POOL_EMPTY_BPOBJ, tx));
		VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
		dp->dp_empty_bpobj = 0;
	}
}

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;

	ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
	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);
}

boolean_t
bpobj_is_open(const bpobj_t *bpo)
{
	return (bpo->bpo_object != 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);
}

boolean_t
bpobj_is_empty(bpobj_t *bpo)
{
	return (bpo->bpo_phys->bpo_num_blkptrs == 0 &&
	    (!bpo->bpo_havesubobj || bpo->bpo_phys->bpo_num_subobjs == 0));
}

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;

	ASSERT(bpobj_is_open(bpo));
	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) {
		VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
		    (i + 1) * sizeof (blkptr_t), DMU_OBJECT_END, 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);
	}
	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
	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 != 0) {
				bpobj_close(&sublist);
				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), DMU_OBJECT_END, tx));
	}

out:
	/* If there are no entries, there should be no bytes. */
	if (bpobj_is_empty(bpo)) {
		ASSERT0(bpo->bpo_phys->bpo_bytes);
		ASSERT0(bpo->bpo_phys->bpo_comp);
		ASSERT0(bpo->bpo_phys->bpo_uncomp);
	}

	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(bpobj_is_open(bpo));
	ASSERT(subobj != 0);
	ASSERT(bpo->bpo_havesubobj);
	ASSERT(bpo->bpo_havecomp);
	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);

	if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
		bpobj_decr_empty(bpo->bpo_os, tx);
		return;
	}

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

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

	mutex_enter(&bpo->bpo_lock);
	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_OLD_MAXBLOCKSIZE,
		    DMU_OT_NONE, 0, tx);
	}

	ASSERTV(dmu_object_info_t doi);
	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);

	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));
			/*
			 * Make sure that we are not asking dmu_write()
			 * to write more data than we have in our buffer.
			 */
			VERIFY3U(subdb->db_size, >=,
			    numsubsub * sizeof (subobj));
			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(bpobj_is_open(bpo));
	ASSERT(!BP_IS_HOLE(bp));
	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);

	if (BP_IS_EMBEDDED(bp)) {
		/*
		 * The bpobj will compress better without the payload.
		 *
		 * Note that we store EMBEDDED bp's because they have an
		 * uncompressed size, which must be accounted for.  An
		 * alternative would be to add their size to bpo_uncomp
		 * without storing the bp, but that would create additional
		 * complications: bpo_uncomp would be inconsistent with the
		 * set of BP's stored, and bpobj_iterate() wouldn't visit
		 * all the space accounted for in the bpobj.
		 */
		bzero(&stored_bp, sizeof (stored_bp));
		stored_bp.blk_prop = bp->blk_prop;
		stored_bp.blk_birth = bp->blk_birth;
	} else if (!BP_GET_DEDUP(bp)) {
		/* The bpobj will compress better without the checksum */
		bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
	}

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

	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) {
		if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
			sra->used += bp_get_dsize_sync(sra->spa, bp);
		else
			sra->used += bp_get_dsize(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)
{
	ASSERT(bpobj_is_open(bpo));
	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;

	ASSERT(bpobj_is_open(bpo));

	/*
	 * 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.
df7eeccc	23	* Copyright (c) 2011, 2016 by Delphix. All rights reserved.
9210e43a	24	* Copyright (c) 2017 Datto Inc.
428870ff BB	25	*/
	26
	27	#include <sys/bpobj.h>
	28	#include <sys/zfs_context.h>
	29	#include <sys/refcount.h>
330d06f9	30	#include <sys/dsl_pool.h>
753c3839 MA	31	#include <sys/zfeature.h>
	32	#include <sys/zap.h>
	33
	34	/*
	35	* Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
	36	*/
	37	uint64_t
	38	bpobj_alloc_empty(objset_t os, int blocksize, dmu_tx_t tx)
	39	{
753c3839 MA	40	spa_t *spa = dmu_objset_spa(os);
	41	dsl_pool_t *dp = dmu_objset_pool(os);
	42
fa86b5db MA	43	if (spa_feature_is_enabled(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
fa86b5db MA	44	if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
c99c9001	45	ASSERT0(dp->dp_empty_bpobj);
753c3839	46	dp->dp_empty_bpobj =
f1512ee6	47	bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx);
753c3839 MA	48	VERIFY(zap_add(os,
	49	DMU_POOL_DIRECTORY_OBJECT,
	50	DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
	51	&dp->dp_empty_bpobj, tx) == 0);
	52	}
fa86b5db	53	spa_feature_incr(spa, SPA_FEATURE_EMPTY_BPOBJ, tx);
753c3839 MA	54	ASSERT(dp->dp_empty_bpobj != 0);
	55	return (dp->dp_empty_bpobj);
	56	} else {
	57	return (bpobj_alloc(os, blocksize, tx));
	58	}
	59	}
	60
	61	void
	62	bpobj_decr_empty(objset_t os, dmu_tx_t tx)
	63	{
753c3839 MA	64	dsl_pool_t *dp = dmu_objset_pool(os);
753c3839 MA	65
fa86b5db MA	66	spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
	67	if (!spa_feature_is_active(dmu_objset_spa(os),
	68	SPA_FEATURE_EMPTY_BPOBJ)) {
753c3839 MA	69	VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
	70	DMU_POOL_DIRECTORY_OBJECT,
	71	DMU_POOL_EMPTY_BPOBJ, tx));
	72	VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
	73	dp->dp_empty_bpobj = 0;
	74	}
	75	}
428870ff BB	76
	77	uint64_t
	78	bpobj_alloc(objset_t os, int blocksize, dmu_tx_t tx)
	79	{
	80	int size;
	81
	82	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
	83	size = BPOBJ_SIZE_V0;
	84	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
	85	size = BPOBJ_SIZE_V1;
	86	else
	87	size = sizeof (bpobj_phys_t);
	88
	89	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
	90	DMU_OT_BPOBJ_HDR, size, tx));
	91	}
	92
	93	void
	94	bpobj_free(objset_t os, uint64_t obj, dmu_tx_t tx)
	95	{
	96	int64_t i;
	97	bpobj_t bpo;
	98	dmu_object_info_t doi;
	99	int epb;
	100	dmu_buf_t *dbuf = NULL;
	101
753c3839	102	ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
428870ff BB	103	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
	104
	105	mutex_enter(&bpo.bpo_lock);
	106
	107	if (!bpo.bpo_havesubobj \|\| bpo.bpo_phys->bpo_subobjs == 0)
	108	goto out;
	109
	110	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
	111	epb = doi.doi_data_block_size / sizeof (uint64_t);
	112
	113	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
	114	uint64_t *objarray;
	115	uint64_t offset, blkoff;
	116
	117	offset = i * sizeof (uint64_t);
	118	blkoff = P2PHASE(i, epb);
	119
	120	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	121	if (dbuf)
	122	dmu_buf_rele(dbuf, FTAG);
	123	VERIFY3U(0, ==, dmu_buf_hold(os,
	124	bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
	125	}
	126
	127	ASSERT3U(offset, >=, dbuf->db_offset);
	128	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	129
	130	objarray = dbuf->db_data;
	131	bpobj_free(os, objarray[blkoff], tx);
	132	}
	133	if (dbuf) {
	134	dmu_buf_rele(dbuf, FTAG);
	135	dbuf = NULL;
	136	}
	137	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
	138
	139	out:
	140	mutex_exit(&bpo.bpo_lock);
	141	bpobj_close(&bpo);
	142
	143	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
	144	}
	145
	146	int
	147	bpobj_open(bpobj_t bpo, objset_t os, uint64_t object)
	148	{
	149	dmu_object_info_t doi;
	150	int err;
	151
	152	err = dmu_object_info(os, object, &doi);
	153	if (err)
	154	return (err);
	155
	156	bzero(bpo, sizeof (*bpo));
	157	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
	158
	159	ASSERT(bpo->bpo_dbuf == NULL);
	160	ASSERT(bpo->bpo_phys == NULL);
	161	ASSERT(object != 0);
	162	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
	163	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
	164
572e2857 BB	165	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
	166	if (err)
	167	return (err);
	168
428870ff BB	169	bpo->bpo_os = os;
	170	bpo->bpo_object = object;
	171	bpo->bpo_epb = doi.doi_data_block_size >> SPA_BLKPTRSHIFT;
	172	bpo->bpo_havecomp = (doi.doi_bonus_size > BPOBJ_SIZE_V0);
	173	bpo->bpo_havesubobj = (doi.doi_bonus_size > BPOBJ_SIZE_V1);
428870ff BB	174	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
	175	return (0);
	176	}
	177
a1d477c2 MA	178	boolean_t
	179	bpobj_is_open(const bpobj_t *bpo)
	180	{
	181	return (bpo->bpo_object != 0);
	182	}
	183
428870ff BB	184	void
	185	bpobj_close(bpobj_t *bpo)
	186	{
	187	/* Lame workaround for closing a bpobj that was never opened. */
	188	if (bpo->bpo_object == 0)
	189	return;
	190
	191	dmu_buf_rele(bpo->bpo_dbuf, bpo);
	192	if (bpo->bpo_cached_dbuf != NULL)
	193	dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	194	bpo->bpo_dbuf = NULL;
	195	bpo->bpo_phys = NULL;
	196	bpo->bpo_cached_dbuf = NULL;
572e2857	197	bpo->bpo_object = 0;
428870ff BB	198
	199	mutex_destroy(&bpo->bpo_lock);
	200	}
	201
a1d477c2 MA	202	boolean_t
a1d477c2 MA	203	bpobj_is_empty(bpobj_t *bpo)
9b67f605	204	{
a1d477c2 MA	205	return (bpo->bpo_phys->bpo_num_blkptrs == 0 &&
a1d477c2 MA	206	(!bpo->bpo_havesubobj \|\| bpo->bpo_phys->bpo_num_subobjs == 0));
9b67f605 MA	207	}
9b67f605 MA	208
428870ff BB	209	static int
	210	bpobj_iterate_impl(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx,
	211	boolean_t free)
	212	{
	213	dmu_object_info_t doi;
	214	int epb;
	215	int64_t i;
	216	int err = 0;
	217	dmu_buf_t *dbuf = NULL;
	218
a1d477c2	219	ASSERT(bpobj_is_open(bpo));
428870ff BB	220	mutex_enter(&bpo->bpo_lock);
	221
	222	if (free)
	223	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	224
	225	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
	226	blkptr_t *bparray;
	227	blkptr_t *bp;
	228	uint64_t offset, blkoff;
	229
	230	offset = i * sizeof (blkptr_t);
	231	blkoff = P2PHASE(i, bpo->bpo_epb);
	232
	233	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	234	if (dbuf)
	235	dmu_buf_rele(dbuf, FTAG);
	236	err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
	237	FTAG, &dbuf, 0);
	238	if (err)
	239	break;
	240	}
	241
	242	ASSERT3U(offset, >=, dbuf->db_offset);
	243	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	244
	245	bparray = dbuf->db_data;
	246	bp = &bparray[blkoff];
	247	err = func(arg, bp, tx);
	248	if (err)
	249	break;
	250	if (free) {
	251	bpo->bpo_phys->bpo_bytes -=
	252	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	253	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
	254	if (bpo->bpo_havecomp) {
	255	bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
	256	bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
	257	}
	258	bpo->bpo_phys->bpo_num_blkptrs--;
	259	ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
	260	}
	261	}
	262	if (dbuf) {
	263	dmu_buf_rele(dbuf, FTAG);
	264	dbuf = NULL;
	265	}
	266	if (free) {
428870ff	267	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
ee45fbd8	268	(i + 1) * sizeof (blkptr_t), DMU_OBJECT_END, tx));
428870ff BB	269	}
	270	if (err \|\| !bpo->bpo_havesubobj \|\| bpo->bpo_phys->bpo_subobjs == 0)
	271	goto out;
	272
	273	ASSERT(bpo->bpo_havecomp);
	274	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
572e2857 BB	275	if (err) {
572e2857 BB	276	mutex_exit(&bpo->bpo_lock);
428870ff	277	return (err);
572e2857	278	}
fa86b5db	279	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
428870ff BB	280	epb = doi.doi_data_block_size / sizeof (uint64_t);
	281
	282	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
	283	uint64_t *objarray;
	284	uint64_t offset, blkoff;
	285	bpobj_t sublist;
	286	uint64_t used_before, comp_before, uncomp_before;
	287	uint64_t used_after, comp_after, uncomp_after;
	288
	289	offset = i * sizeof (uint64_t);
	290	blkoff = P2PHASE(i, epb);
	291
	292	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	293	if (dbuf)
	294	dmu_buf_rele(dbuf, FTAG);
	295	err = dmu_buf_hold(bpo->bpo_os,
	296	bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
	297	if (err)
	298	break;
	299	}
	300
	301	ASSERT3U(offset, >=, dbuf->db_offset);
	302	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	303
	304	objarray = dbuf->db_data;
	305	err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
	306	if (err)
	307	break;
	308	if (free) {
	309	err = bpobj_space(&sublist,
	310	&used_before, &comp_before, &uncomp_before);
50f9ea01 WA	311	if (err != 0) {
50f9ea01 WA	312	bpobj_close(&sublist);
428870ff	313	break;
50f9ea01	314	}
428870ff BB	315	}
	316	err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
	317	if (free) {
	318	VERIFY3U(0, ==, bpobj_space(&sublist,
	319	&used_after, &comp_after, &uncomp_after));
	320	bpo->bpo_phys->bpo_bytes -= used_before - used_after;
	321	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
572e2857	322	bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
428870ff BB	323	bpo->bpo_phys->bpo_uncomp -=
	324	uncomp_before - uncomp_after;
	325	}
	326
	327	bpobj_close(&sublist);
	328	if (err)
	329	break;
	330	if (free) {
	331	err = dmu_object_free(bpo->bpo_os,
	332	objarray[blkoff], tx);
	333	if (err)
	334	break;
	335	bpo->bpo_phys->bpo_num_subobjs--;
	336	ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
	337	}
	338	}
	339	if (dbuf) {
	340	dmu_buf_rele(dbuf, FTAG);
	341	dbuf = NULL;
	342	}
	343	if (free) {
	344	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
	345	bpo->bpo_phys->bpo_subobjs,
ee45fbd8	346	(i + 1) * sizeof (uint64_t), DMU_OBJECT_END, tx));
428870ff BB	347	}
	348
	349	out:
	350	/* If there are no entries, there should be no bytes. */
a1d477c2	351	if (bpobj_is_empty(bpo)) {
9b67f605 MA	352	ASSERT0(bpo->bpo_phys->bpo_bytes);
	353	ASSERT0(bpo->bpo_phys->bpo_comp);
	354	ASSERT0(bpo->bpo_phys->bpo_uncomp);
	355	}
428870ff BB	356
	357	mutex_exit(&bpo->bpo_lock);
	358	return (err);
	359	}
	360
	361	/*
	362	* Iterate and remove the entries. If func returns nonzero, iteration
	363	* will stop and that entry will not be removed.
	364	*/
	365	int
	366	bpobj_iterate(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	367	{
	368	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
	369	}
	370
	371	/*
	372	* Iterate the entries. If func returns nonzero, iteration will stop.
	373	*/
	374	int
	375	bpobj_iterate_nofree(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	376	{
	377	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
	378	}
	379
	380	void
	381	bpobj_enqueue_subobj(bpobj_t bpo, uint64_t subobj, dmu_tx_t tx)
	382	{
	383	bpobj_t subbpo;
572e2857	384	uint64_t used, comp, uncomp, subsubobjs;
428870ff	385
a1d477c2 MA	386	ASSERT(bpobj_is_open(bpo));
a1d477c2 MA	387	ASSERT(subobj != 0);
428870ff BB	388	ASSERT(bpo->bpo_havesubobj);
428870ff BB	389	ASSERT(bpo->bpo_havecomp);
753c3839 MA	390	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
	391
	392	if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
	393	bpobj_decr_empty(bpo->bpo_os, tx);
	394	return;
	395	}
428870ff BB	396
	397	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
	398	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
428870ff	399
a1d477c2	400	if (bpobj_is_empty(&subbpo)) {
428870ff	401	/* No point in having an empty subobj. */
572e2857	402	bpobj_close(&subbpo);
428870ff BB	403	bpobj_free(bpo->bpo_os, subobj, tx);
	404	return;
	405	}
	406
df7eeccc	407	mutex_enter(&bpo->bpo_lock);
428870ff BB	408	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	409	if (bpo->bpo_phys->bpo_subobjs == 0) {
	410	bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
f1512ee6 MA	411	DMU_OT_BPOBJ_SUBOBJ, SPA_OLD_MAXBLOCKSIZE,
f1512ee6 MA	412	DMU_OT_NONE, 0, tx);
428870ff BB	413	}
428870ff BB	414
1c27024e	415	ASSERTV(dmu_object_info_t doi);
13fe0198 MA	416	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
	417	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
	418
428870ff BB	419	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	420	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	421	sizeof (subobj), &subobj, tx);
	422	bpo->bpo_phys->bpo_num_subobjs++;
572e2857 BB	423
	424	/*
	425	* If subobj has only one block of subobjs, then move subobj's
	426	* subobjs to bpo's subobj list directly. This reduces
	427	* recursion in bpobj_iterate due to nested subobjs.
	428	*/
	429	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
	430	if (subsubobjs != 0) {
	431	dmu_object_info_t doi;
	432
	433	VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
	434	if (doi.doi_max_offset == doi.doi_data_block_size) {
	435	dmu_buf_t *subdb;
	436	uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
	437
	438	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
	439	0, FTAG, &subdb, 0));
d1fada1e MA	440	/*
	441	* Make sure that we are not asking dmu_write()
	442	* to write more data than we have in our buffer.
	443	*/
	444	VERIFY3U(subdb->db_size, >=,
	445	numsubsub * sizeof (subobj));
572e2857 BB	446	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	447	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	448	numsubsub * sizeof (subobj), subdb->db_data, tx);
	449	dmu_buf_rele(subdb, FTAG);
	450	bpo->bpo_phys->bpo_num_subobjs += numsubsub;
	451
	452	dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
	453	subbpo.bpo_phys->bpo_subobjs = 0;
	454	VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
	455	subsubobjs, tx));
	456	}
	457	}
428870ff BB	458	bpo->bpo_phys->bpo_bytes += used;
	459	bpo->bpo_phys->bpo_comp += comp;
	460	bpo->bpo_phys->bpo_uncomp += uncomp;
	461	mutex_exit(&bpo->bpo_lock);
572e2857 BB	462
572e2857 BB	463	bpobj_close(&subbpo);
428870ff BB	464	}
	465
	466	void
	467	bpobj_enqueue(bpobj_t bpo, const blkptr_t bp, dmu_tx_t *tx)
	468	{
	469	blkptr_t stored_bp = *bp;
	470	uint64_t offset;
	471	int blkoff;
	472	blkptr_t *bparray;
	473
a1d477c2	474	ASSERT(bpobj_is_open(bpo));
428870ff	475	ASSERT(!BP_IS_HOLE(bp));
753c3839	476	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
428870ff	477
9b67f605 MA	478	if (BP_IS_EMBEDDED(bp)) {
	479	/*
	480	* The bpobj will compress better without the payload.
	481	*
	482	* Note that we store EMBEDDED bp's because they have an
	483	* uncompressed size, which must be accounted for. An
	484	* alternative would be to add their size to bpo_uncomp
	485	* without storing the bp, but that would create additional
	486	* complications: bpo_uncomp would be inconsistent with the
	487	* set of BP's stored, and bpobj_iterate() wouldn't visit
	488	* all the space accounted for in the bpobj.
	489	*/
	490	bzero(&stored_bp, sizeof (stored_bp));
	491	stored_bp.blk_prop = bp->blk_prop;
	492	stored_bp.blk_birth = bp->blk_birth;
	493	} else if (!BP_GET_DEDUP(bp)) {
	494	/* The bpobj will compress better without the checksum */
	495	bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
	496	}
	497
428870ff BB	498	/* We never need the fill count. */
	499	stored_bp.blk_fill = 0;
	500
428870ff BB	501	mutex_enter(&bpo->bpo_lock);
	502
	503	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
	504	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
	505
	506	if (bpo->bpo_cached_dbuf == NULL \|\|
	507	offset < bpo->bpo_cached_dbuf->db_offset \|\|
	508	offset >= bpo->bpo_cached_dbuf->db_offset +
	509	bpo->bpo_cached_dbuf->db_size) {
	510	if (bpo->bpo_cached_dbuf)
	511	dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	512	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
	513	offset, bpo, &bpo->bpo_cached_dbuf, 0));
	514	}
	515
	516	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
	517	bparray = bpo->bpo_cached_dbuf->db_data;
	518	bparray[blkoff] = stored_bp;
	519
	520	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	521	bpo->bpo_phys->bpo_num_blkptrs++;
	522	bpo->bpo_phys->bpo_bytes +=
	523	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	524	if (bpo->bpo_havecomp) {
	525	bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
	526	bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
	527	}
	528	mutex_exit(&bpo->bpo_lock);
	529	}
	530
	531	struct space_range_arg {
	532	spa_t *spa;
	533	uint64_t mintxg;
	534	uint64_t maxtxg;
	535	uint64_t used;
	536	uint64_t comp;
	537	uint64_t uncomp;
	538	};
	539
	540	/* ARGSUSED */
	541	static int
	542	space_range_cb(void arg, const blkptr_t bp, dmu_tx_t *tx)
	543	{
	544	struct space_range_arg *sra = arg;
	545
	546	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
330d06f9 MA	547	if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
	548	sra->used += bp_get_dsize_sync(sra->spa, bp);
	549	else
	550	sra->used += bp_get_dsize(sra->spa, bp);
428870ff BB	551	sra->comp += BP_GET_PSIZE(bp);
	552	sra->uncomp += BP_GET_UCSIZE(bp);
	553	}
	554	return (0);
	555	}
	556
	557	int
	558	bpobj_space(bpobj_t bpo, uint64_t usedp, uint64_t compp, uint64_t uncompp)
	559	{
a1d477c2	560	ASSERT(bpobj_is_open(bpo));
428870ff BB	561	mutex_enter(&bpo->bpo_lock);
	562
	563	*usedp = bpo->bpo_phys->bpo_bytes;
	564	if (bpo->bpo_havecomp) {
	565	*compp = bpo->bpo_phys->bpo_comp;
	566	*uncompp = bpo->bpo_phys->bpo_uncomp;
	567	mutex_exit(&bpo->bpo_lock);
	568	return (0);
	569	} else {
	570	mutex_exit(&bpo->bpo_lock);
	571	return (bpobj_space_range(bpo, 0, UINT64_MAX,
	572	usedp, compp, uncompp));
	573	}
	574	}
	575
	576	/*
	577	* Return the amount of space in the bpobj which is:
	578	* mintxg < blk_birth <= maxtxg
	579	*/
	580	int
	581	bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
	582	uint64_t usedp, uint64_t compp, uint64_t *uncompp)
	583	{
	584	struct space_range_arg sra = { 0 };
	585	int err;
	586
a1d477c2 MA	587	ASSERT(bpobj_is_open(bpo));
a1d477c2 MA	588
428870ff BB	589	/*
	590	* As an optimization, if they want the whole txg range, just
	591	* get bpo_bytes rather than iterating over the bps.
	592	*/
	593	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
	594	return (bpobj_space(bpo, usedp, compp, uncompp));
	595
	596	sra.spa = dmu_objset_spa(bpo->bpo_os);
	597	sra.mintxg = mintxg;
	598	sra.maxtxg = maxtxg;
	599
	600	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
	601	*usedp = sra.used;
	602	*compp = sra.comp;
	603	*uncompp = sra.uncomp;
	604	return (err);
	605	}