[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) 2013 by Delphix. All rights reserved.
 */

#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_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);
}

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);
	}
	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)
				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;
	ASSERTV(dmu_object_info_t doi);

	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 (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);
	}

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

	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));
			/*
			 * 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(!BP_IS_HOLE(bp));
	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);

	/* 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) {
		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)
{
	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.
d1fada1e	23	* Copyright (c) 2013 by Delphix. All rights reserved.
428870ff BB	24	*/
	25
	26	#include <sys/bpobj.h>
	27	#include <sys/zfs_context.h>
	28	#include <sys/refcount.h>
330d06f9	29	#include <sys/dsl_pool.h>
753c3839 MA	30	#include <sys/zfeature.h>
	31	#include <sys/zap.h>
	32
	33	/*
	34	* Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
	35	*/
	36	uint64_t
	37	bpobj_alloc_empty(objset_t os, int blocksize, dmu_tx_t tx)
	38	{
753c3839 MA	39	spa_t *spa = dmu_objset_spa(os);
	40	dsl_pool_t *dp = dmu_objset_pool(os);
	41
fa86b5db MA	42	if (spa_feature_is_enabled(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
fa86b5db MA	43	if (!spa_feature_is_active(spa, SPA_FEATURE_EMPTY_BPOBJ)) {
c99c9001	44	ASSERT0(dp->dp_empty_bpobj);
753c3839 MA	45	dp->dp_empty_bpobj =
	46	bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx);
	47	VERIFY(zap_add(os,
	48	DMU_POOL_DIRECTORY_OBJECT,
	49	DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
	50	&dp->dp_empty_bpobj, tx) == 0);
	51	}
fa86b5db	52	spa_feature_incr(spa, SPA_FEATURE_EMPTY_BPOBJ, tx);
753c3839 MA	53	ASSERT(dp->dp_empty_bpobj != 0);
	54	return (dp->dp_empty_bpobj);
	55	} else {
	56	return (bpobj_alloc(os, blocksize, tx));
	57	}
	58	}
	59
	60	void
	61	bpobj_decr_empty(objset_t os, dmu_tx_t tx)
	62	{
753c3839 MA	63	dsl_pool_t *dp = dmu_objset_pool(os);
753c3839 MA	64
fa86b5db MA	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)) {
753c3839 MA	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;
	73	}
	74	}
428870ff BB	75
	76	uint64_t
	77	bpobj_alloc(objset_t os, int blocksize, dmu_tx_t tx)
	78	{
	79	int size;
	80
	81	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_BPOBJ_ACCOUNT)
	82	size = BPOBJ_SIZE_V0;
	83	else if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
	84	size = BPOBJ_SIZE_V1;
	85	else
	86	size = sizeof (bpobj_phys_t);
	87
	88	return (dmu_object_alloc(os, DMU_OT_BPOBJ, blocksize,
	89	DMU_OT_BPOBJ_HDR, size, tx));
	90	}
	91
	92	void
	93	bpobj_free(objset_t os, uint64_t obj, dmu_tx_t tx)
	94	{
	95	int64_t i;
	96	bpobj_t bpo;
	97	dmu_object_info_t doi;
	98	int epb;
	99	dmu_buf_t *dbuf = NULL;
	100
753c3839	101	ASSERT(obj != dmu_objset_pool(os)->dp_empty_bpobj);
428870ff BB	102	VERIFY3U(0, ==, bpobj_open(&bpo, os, obj));
	103
	104	mutex_enter(&bpo.bpo_lock);
	105
	106	if (!bpo.bpo_havesubobj \|\| bpo.bpo_phys->bpo_subobjs == 0)
	107	goto out;
	108
	109	VERIFY3U(0, ==, dmu_object_info(os, bpo.bpo_phys->bpo_subobjs, &doi));
	110	epb = doi.doi_data_block_size / sizeof (uint64_t);
	111
	112	for (i = bpo.bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
	113	uint64_t *objarray;
	114	uint64_t offset, blkoff;
	115
	116	offset = i * sizeof (uint64_t);
	117	blkoff = P2PHASE(i, epb);
	118
	119	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	120	if (dbuf)
	121	dmu_buf_rele(dbuf, FTAG);
	122	VERIFY3U(0, ==, dmu_buf_hold(os,
	123	bpo.bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0));
	124	}
	125
	126	ASSERT3U(offset, >=, dbuf->db_offset);
	127	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	128
	129	objarray = dbuf->db_data;
	130	bpobj_free(os, objarray[blkoff], tx);
	131	}
	132	if (dbuf) {
	133	dmu_buf_rele(dbuf, FTAG);
	134	dbuf = NULL;
	135	}
	136	VERIFY3U(0, ==, dmu_object_free(os, bpo.bpo_phys->bpo_subobjs, tx));
	137
	138	out:
	139	mutex_exit(&bpo.bpo_lock);
	140	bpobj_close(&bpo);
	141
	142	VERIFY3U(0, ==, dmu_object_free(os, obj, tx));
	143	}
	144
	145	int
	146	bpobj_open(bpobj_t bpo, objset_t os, uint64_t object)
	147	{
	148	dmu_object_info_t doi;
	149	int err;
	150
	151	err = dmu_object_info(os, object, &doi);
	152	if (err)
	153	return (err);
	154
	155	bzero(bpo, sizeof (*bpo));
	156	mutex_init(&bpo->bpo_lock, NULL, MUTEX_DEFAULT, NULL);
	157
	158	ASSERT(bpo->bpo_dbuf == NULL);
	159	ASSERT(bpo->bpo_phys == NULL);
	160	ASSERT(object != 0);
	161	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ);
	162	ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_BPOBJ_HDR);
	163
572e2857 BB	164	err = dmu_bonus_hold(os, object, bpo, &bpo->bpo_dbuf);
	165	if (err)
	166	return (err);
	167
428870ff BB	168	bpo->bpo_os = os;
	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);
428870ff BB	173	bpo->bpo_phys = bpo->bpo_dbuf->db_data;
	174	return (0);
	175	}
	176
	177	void
	178	bpobj_close(bpobj_t *bpo)
	179	{
	180	/* Lame workaround for closing a bpobj that was never opened. */
	181	if (bpo->bpo_object == 0)
	182	return;
	183
	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;
572e2857	190	bpo->bpo_object = 0;
428870ff BB	191
	192	mutex_destroy(&bpo->bpo_lock);
	193	}
	194
	195	static int
	196	bpobj_iterate_impl(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx,
	197	boolean_t free)
	198	{
	199	dmu_object_info_t doi;
	200	int epb;
	201	int64_t i;
	202	int err = 0;
	203	dmu_buf_t *dbuf = NULL;
	204
	205	mutex_enter(&bpo->bpo_lock);
	206
	207	if (free)
	208	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	209
	210	for (i = bpo->bpo_phys->bpo_num_blkptrs - 1; i >= 0; i--) {
	211	blkptr_t *bparray;
	212	blkptr_t *bp;
	213	uint64_t offset, blkoff;
	214
	215	offset = i * sizeof (blkptr_t);
	216	blkoff = P2PHASE(i, bpo->bpo_epb);
	217
	218	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	219	if (dbuf)
	220	dmu_buf_rele(dbuf, FTAG);
	221	err = dmu_buf_hold(bpo->bpo_os, bpo->bpo_object, offset,
	222	FTAG, &dbuf, 0);
	223	if (err)
	224	break;
	225	}
	226
	227	ASSERT3U(offset, >=, dbuf->db_offset);
	228	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	229
	230	bparray = dbuf->db_data;
	231	bp = &bparray[blkoff];
	232	err = func(arg, bp, tx);
	233	if (err)
	234	break;
	235	if (free) {
	236	bpo->bpo_phys->bpo_bytes -=
	237	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	238	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
	239	if (bpo->bpo_havecomp) {
	240	bpo->bpo_phys->bpo_comp -= BP_GET_PSIZE(bp);
	241	bpo->bpo_phys->bpo_uncomp -= BP_GET_UCSIZE(bp);
	242	}
	243	bpo->bpo_phys->bpo_num_blkptrs--;
	244	ASSERT3S(bpo->bpo_phys->bpo_num_blkptrs, >=, 0);
	245	}
	246	}
	247	if (dbuf) {
	248	dmu_buf_rele(dbuf, FTAG);
	249	dbuf = NULL;
	250	}
	251	if (free) {
	252	i++;
	253	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os, bpo->bpo_object,
	254	i * sizeof (blkptr_t), -1ULL, tx));
255	}
256	if (err \|\| !bpo->bpo_havesubobj \|\| bpo->bpo_phys->bpo_subobjs == 0)
257	goto out;
258
259	ASSERT(bpo->bpo_havecomp);
260	err = dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi);
572e2857 BB	261	if (err) {
572e2857 BB	262	mutex_exit(&bpo->bpo_lock);
428870ff	263	return (err);
572e2857	264	}
fa86b5db	265	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
428870ff BB	266	epb = doi.doi_data_block_size / sizeof (uint64_t);
	267
	268	for (i = bpo->bpo_phys->bpo_num_subobjs - 1; i >= 0; i--) {
	269	uint64_t *objarray;
	270	uint64_t offset, blkoff;
	271	bpobj_t sublist;
	272	uint64_t used_before, comp_before, uncomp_before;
	273	uint64_t used_after, comp_after, uncomp_after;
	274
	275	offset = i * sizeof (uint64_t);
	276	blkoff = P2PHASE(i, epb);
	277
	278	if (dbuf == NULL \|\| dbuf->db_offset > offset) {
	279	if (dbuf)
	280	dmu_buf_rele(dbuf, FTAG);
	281	err = dmu_buf_hold(bpo->bpo_os,
	282	bpo->bpo_phys->bpo_subobjs, offset, FTAG, &dbuf, 0);
	283	if (err)
	284	break;
	285	}
	286
	287	ASSERT3U(offset, >=, dbuf->db_offset);
	288	ASSERT3U(offset, <, dbuf->db_offset + dbuf->db_size);
	289
	290	objarray = dbuf->db_data;
	291	err = bpobj_open(&sublist, bpo->bpo_os, objarray[blkoff]);
	292	if (err)
	293	break;
	294	if (free) {
	295	err = bpobj_space(&sublist,
	296	&used_before, &comp_before, &uncomp_before);
	297	if (err)
	298	break;
	299	}
	300	err = bpobj_iterate_impl(&sublist, func, arg, tx, free);
	301	if (free) {
	302	VERIFY3U(0, ==, bpobj_space(&sublist,
	303	&used_after, &comp_after, &uncomp_after));
	304	bpo->bpo_phys->bpo_bytes -= used_before - used_after;
	305	ASSERT3S(bpo->bpo_phys->bpo_bytes, >=, 0);
572e2857	306	bpo->bpo_phys->bpo_comp -= comp_before - comp_after;
428870ff BB	307	bpo->bpo_phys->bpo_uncomp -=
	308	uncomp_before - uncomp_after;
	309	}
	310
	311	bpobj_close(&sublist);
	312	if (err)
	313	break;
	314	if (free) {
	315	err = dmu_object_free(bpo->bpo_os,
	316	objarray[blkoff], tx);
	317	if (err)
	318	break;
	319	bpo->bpo_phys->bpo_num_subobjs--;
	320	ASSERT3S(bpo->bpo_phys->bpo_num_subobjs, >=, 0);
	321	}
	322	}
	323	if (dbuf) {
	324	dmu_buf_rele(dbuf, FTAG);
	325	dbuf = NULL;
	326	}
	327	if (free) {
	328	VERIFY3U(0, ==, dmu_free_range(bpo->bpo_os,
	329	bpo->bpo_phys->bpo_subobjs,
	330	(i + 1) * sizeof (uint64_t), -1ULL, tx));
	331	}
	332
	333	out:
	334	/* If there are no entries, there should be no bytes. */
	335	ASSERT(bpo->bpo_phys->bpo_num_blkptrs > 0 \|\|
	336	(bpo->bpo_havesubobj && bpo->bpo_phys->bpo_num_subobjs > 0) \|\|
	337	bpo->bpo_phys->bpo_bytes == 0);
	338
	339	mutex_exit(&bpo->bpo_lock);
	340	return (err);
	341	}
	342
	343	/*
	344	* Iterate and remove the entries. If func returns nonzero, iteration
	345	* will stop and that entry will not be removed.
	346	*/
	347	int
	348	bpobj_iterate(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	349	{
	350	return (bpobj_iterate_impl(bpo, func, arg, tx, B_TRUE));
	351	}
	352
	353	/*
	354	* Iterate the entries. If func returns nonzero, iteration will stop.
	355	*/
	356	int
	357	bpobj_iterate_nofree(bpobj_t bpo, bpobj_itor_t func, void arg, dmu_tx_t *tx)
	358	{
	359	return (bpobj_iterate_impl(bpo, func, arg, tx, B_FALSE));
	360	}
	361
	362	void
	363	bpobj_enqueue_subobj(bpobj_t bpo, uint64_t subobj, dmu_tx_t tx)
	364	{
	365	bpobj_t subbpo;
572e2857	366	uint64_t used, comp, uncomp, subsubobjs;
13fe0198	367	ASSERTV(dmu_object_info_t doi);
428870ff BB	368
	369	ASSERT(bpo->bpo_havesubobj);
	370	ASSERT(bpo->bpo_havecomp);
753c3839 MA	371	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
	372
	373	if (subobj == dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj) {
	374	bpobj_decr_empty(bpo->bpo_os, tx);
	375	return;
	376	}
428870ff BB	377
	378	VERIFY3U(0, ==, bpobj_open(&subbpo, bpo->bpo_os, subobj));
	379	VERIFY3U(0, ==, bpobj_space(&subbpo, &used, &comp, &uncomp));
428870ff BB	380
	381	if (used == 0) {
	382	/* No point in having an empty subobj. */
572e2857	383	bpobj_close(&subbpo);
428870ff BB	384	bpobj_free(bpo->bpo_os, subobj, tx);
	385	return;
	386	}
	387
	388	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	389	if (bpo->bpo_phys->bpo_subobjs == 0) {
	390	bpo->bpo_phys->bpo_subobjs = dmu_object_alloc(bpo->bpo_os,
	391	DMU_OT_BPOBJ_SUBOBJ, SPA_MAXBLOCKSIZE, DMU_OT_NONE, 0, tx);
	392	}
	393
13fe0198 MA	394	ASSERT0(dmu_object_info(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs, &doi));
	395	ASSERT3U(doi.doi_type, ==, DMU_OT_BPOBJ_SUBOBJ);
	396
428870ff BB	397	mutex_enter(&bpo->bpo_lock);
	398	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	399	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	400	sizeof (subobj), &subobj, tx);
	401	bpo->bpo_phys->bpo_num_subobjs++;
572e2857 BB	402
	403	/*
	404	* If subobj has only one block of subobjs, then move subobj's
	405	* subobjs to bpo's subobj list directly. This reduces
	406	* recursion in bpobj_iterate due to nested subobjs.
	407	*/
	408	subsubobjs = subbpo.bpo_phys->bpo_subobjs;
	409	if (subsubobjs != 0) {
	410	dmu_object_info_t doi;
	411
	412	VERIFY3U(0, ==, dmu_object_info(bpo->bpo_os, subsubobjs, &doi));
	413	if (doi.doi_max_offset == doi.doi_data_block_size) {
	414	dmu_buf_t *subdb;
	415	uint64_t numsubsub = subbpo.bpo_phys->bpo_num_subobjs;
	416
	417	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, subsubobjs,
	418	0, FTAG, &subdb, 0));
d1fada1e MA	419	/*
	420	* Make sure that we are not asking dmu_write()
	421	* to write more data than we have in our buffer.
	422	*/
	423	VERIFY3U(subdb->db_size, >=,
	424	numsubsub * sizeof (subobj));
572e2857 BB	425	dmu_write(bpo->bpo_os, bpo->bpo_phys->bpo_subobjs,
	426	bpo->bpo_phys->bpo_num_subobjs * sizeof (subobj),
	427	numsubsub * sizeof (subobj), subdb->db_data, tx);
	428	dmu_buf_rele(subdb, FTAG);
	429	bpo->bpo_phys->bpo_num_subobjs += numsubsub;
	430
	431	dmu_buf_will_dirty(subbpo.bpo_dbuf, tx);
	432	subbpo.bpo_phys->bpo_subobjs = 0;
	433	VERIFY3U(0, ==, dmu_object_free(bpo->bpo_os,
	434	subsubobjs, tx));
	435	}
	436	}
428870ff BB	437	bpo->bpo_phys->bpo_bytes += used;
	438	bpo->bpo_phys->bpo_comp += comp;
	439	bpo->bpo_phys->bpo_uncomp += uncomp;
	440	mutex_exit(&bpo->bpo_lock);
572e2857 BB	441
572e2857 BB	442	bpobj_close(&subbpo);
428870ff BB	443	}
	444
	445	void
	446	bpobj_enqueue(bpobj_t bpo, const blkptr_t bp, dmu_tx_t *tx)
	447	{
	448	blkptr_t stored_bp = *bp;
	449	uint64_t offset;
	450	int blkoff;
	451	blkptr_t *bparray;
	452
	453	ASSERT(!BP_IS_HOLE(bp));
753c3839	454	ASSERT(bpo->bpo_object != dmu_objset_pool(bpo->bpo_os)->dp_empty_bpobj);
428870ff BB	455
	456	/* We never need the fill count. */
	457	stored_bp.blk_fill = 0;
	458
	459	/* The bpobj will compress better if we can leave off the checksum */
	460	if (!BP_GET_DEDUP(bp))
	461	bzero(&stored_bp.blk_cksum, sizeof (stored_bp.blk_cksum));
	462
	463	mutex_enter(&bpo->bpo_lock);
	464
	465	offset = bpo->bpo_phys->bpo_num_blkptrs * sizeof (stored_bp);
	466	blkoff = P2PHASE(bpo->bpo_phys->bpo_num_blkptrs, bpo->bpo_epb);
	467
	468	if (bpo->bpo_cached_dbuf == NULL \|\|
	469	offset < bpo->bpo_cached_dbuf->db_offset \|\|
	470	offset >= bpo->bpo_cached_dbuf->db_offset +
	471	bpo->bpo_cached_dbuf->db_size) {
	472	if (bpo->bpo_cached_dbuf)
	473	dmu_buf_rele(bpo->bpo_cached_dbuf, bpo);
	474	VERIFY3U(0, ==, dmu_buf_hold(bpo->bpo_os, bpo->bpo_object,
	475	offset, bpo, &bpo->bpo_cached_dbuf, 0));
	476	}
	477
	478	dmu_buf_will_dirty(bpo->bpo_cached_dbuf, tx);
	479	bparray = bpo->bpo_cached_dbuf->db_data;
	480	bparray[blkoff] = stored_bp;
	481
	482	dmu_buf_will_dirty(bpo->bpo_dbuf, tx);
	483	bpo->bpo_phys->bpo_num_blkptrs++;
	484	bpo->bpo_phys->bpo_bytes +=
	485	bp_get_dsize_sync(dmu_objset_spa(bpo->bpo_os), bp);
	486	if (bpo->bpo_havecomp) {
	487	bpo->bpo_phys->bpo_comp += BP_GET_PSIZE(bp);
	488	bpo->bpo_phys->bpo_uncomp += BP_GET_UCSIZE(bp);
	489	}
	490	mutex_exit(&bpo->bpo_lock);
	491	}
	492
	493	struct space_range_arg {
	494	spa_t *spa;
	495	uint64_t mintxg;
	496	uint64_t maxtxg;
	497	uint64_t used;
	498	uint64_t comp;
	499	uint64_t uncomp;
	500	};
	501
	502	/* ARGSUSED */
	503	static int
	504	space_range_cb(void arg, const blkptr_t bp, dmu_tx_t *tx)
	505	{
	506	struct space_range_arg *sra = arg;
	507
	508	if (bp->blk_birth > sra->mintxg && bp->blk_birth <= sra->maxtxg) {
330d06f9 MA	509	if (dsl_pool_sync_context(spa_get_dsl(sra->spa)))
	510	sra->used += bp_get_dsize_sync(sra->spa, bp);
	511	else
	512	sra->used += bp_get_dsize(sra->spa, bp);
428870ff BB	513	sra->comp += BP_GET_PSIZE(bp);
	514	sra->uncomp += BP_GET_UCSIZE(bp);
	515	}
	516	return (0);
	517	}
	518
	519	int
	520	bpobj_space(bpobj_t bpo, uint64_t usedp, uint64_t compp, uint64_t uncompp)
	521	{
	522	mutex_enter(&bpo->bpo_lock);
	523
	524	*usedp = bpo->bpo_phys->bpo_bytes;
	525	if (bpo->bpo_havecomp) {
	526	*compp = bpo->bpo_phys->bpo_comp;
	527	*uncompp = bpo->bpo_phys->bpo_uncomp;
	528	mutex_exit(&bpo->bpo_lock);
	529	return (0);
	530	} else {
	531	mutex_exit(&bpo->bpo_lock);
	532	return (bpobj_space_range(bpo, 0, UINT64_MAX,
	533	usedp, compp, uncompp));
	534	}
	535	}
	536
	537	/*
	538	* Return the amount of space in the bpobj which is:
	539	* mintxg < blk_birth <= maxtxg
	540	*/
	541	int
	542	bpobj_space_range(bpobj_t *bpo, uint64_t mintxg, uint64_t maxtxg,
	543	uint64_t usedp, uint64_t compp, uint64_t *uncompp)
	544	{
	545	struct space_range_arg sra = { 0 };
	546	int err;
	547
	548	/*
	549	* As an optimization, if they want the whole txg range, just
	550	* get bpo_bytes rather than iterating over the bps.
	551	*/
	552	if (mintxg < TXG_INITIAL && maxtxg == UINT64_MAX && bpo->bpo_havecomp)
	553	return (bpobj_space(bpo, usedp, compp, uncompp));
	554
	555	sra.spa = dmu_objset_spa(bpo->bpo_os);
	556	sra.mintxg = mintxg;
	557	sra.maxtxg = maxtxg;
	558
	559	err = bpobj_iterate_nofree(bpo, space_range_cb, &sra, NULL);
	560	*usedp = sra.used;
	561	*compp = sra.comp;
	562	*uncompp = sra.uncomp;
	563	return (err);
	564	}