[mirror_zfs-debian.git] / module / zfs / dsl_deadlist.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) 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2012, 2016 by Delphix. All rights reserved.
 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
 */

#include <sys/dsl_dataset.h>
#include <sys/dmu.h>
#include <sys/refcount.h>
#include <sys/zap.h>
#include <sys/zfs_context.h>
#include <sys/dsl_pool.h>

/*
 * Deadlist concurrency:
 *
 * Deadlists can only be modified from the syncing thread.
 *
 * Except for dsl_deadlist_insert(), it can only be modified with the
 * dp_config_rwlock held with RW_WRITER.
 *
 * The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
 * be called concurrently, from open context, with the dl_config_rwlock held
 * with RW_READER.
 *
 * Therefore, we only need to provide locking between dsl_deadlist_insert() and
 * the accessors, protecting:
 *     dl_phys->dl_used,comp,uncomp
 *     and protecting the dl_tree from being loaded.
 * The locking is provided by dl_lock.  Note that locking on the bpobj_t
 * provides its own locking, and dl_oldfmt is immutable.
 */

static int
dsl_deadlist_compare(const void *arg1, const void *arg2)
{
	const dsl_deadlist_entry_t *dle1 = (const dsl_deadlist_entry_t *)arg1;
	const dsl_deadlist_entry_t *dle2 = (const dsl_deadlist_entry_t *)arg2;

	return (AVL_CMP(dle1->dle_mintxg, dle2->dle_mintxg));
}

static void
dsl_deadlist_load_tree(dsl_deadlist_t *dl)
{
	zap_cursor_t zc;
	zap_attribute_t za;

	ASSERT(MUTEX_HELD(&dl->dl_lock));

	ASSERT(!dl->dl_oldfmt);
	if (dl->dl_havetree)
		return;

	avl_create(&dl->dl_tree, dsl_deadlist_compare,
	    sizeof (dsl_deadlist_entry_t),
	    offsetof(dsl_deadlist_entry_t, dle_node));
	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
	    zap_cursor_retrieve(&zc, &za) == 0;
	    zap_cursor_advance(&zc)) {
		dsl_deadlist_entry_t *dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
		dle->dle_mintxg = zfs_strtonum(za.za_name, NULL);
		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os,
		    za.za_first_integer));
		avl_add(&dl->dl_tree, dle);
	}
	zap_cursor_fini(&zc);
	dl->dl_havetree = B_TRUE;
}

void
dsl_deadlist_open(dsl_deadlist_t *dl, objset_t *os, uint64_t object)
{
	dmu_object_info_t doi;

	mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
	dl->dl_os = os;
	dl->dl_object = object;
	VERIFY3U(0, ==, dmu_bonus_hold(os, object, dl, &dl->dl_dbuf));
	dmu_object_info_from_db(dl->dl_dbuf, &doi);
	if (doi.doi_type == DMU_OT_BPOBJ) {
		dmu_buf_rele(dl->dl_dbuf, dl);
		dl->dl_dbuf = NULL;
		dl->dl_oldfmt = B_TRUE;
		VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
		return;
	}

	dl->dl_oldfmt = B_FALSE;
	dl->dl_phys = dl->dl_dbuf->db_data;
	dl->dl_havetree = B_FALSE;
}

void
dsl_deadlist_close(dsl_deadlist_t *dl)
{
	void *cookie = NULL;
	dsl_deadlist_entry_t *dle;

	dl->dl_os = NULL;
	mutex_destroy(&dl->dl_lock);

	if (dl->dl_oldfmt) {
		dl->dl_oldfmt = B_FALSE;
		bpobj_close(&dl->dl_bpobj);
		return;
	}

	if (dl->dl_havetree) {
		while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
		    != NULL) {
			bpobj_close(&dle->dle_bpobj);
			kmem_free(dle, sizeof (*dle));
		}
		avl_destroy(&dl->dl_tree);
	}
	dmu_buf_rele(dl->dl_dbuf, dl);
	dl->dl_dbuf = NULL;
	dl->dl_phys = NULL;
}

uint64_t
dsl_deadlist_alloc(objset_t *os, dmu_tx_t *tx)
{
	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
		return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx));
	return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
	    sizeof (dsl_deadlist_phys_t), tx));
}

void
dsl_deadlist_free(objset_t *os, uint64_t dlobj, dmu_tx_t *tx)
{
	dmu_object_info_t doi;
	zap_cursor_t zc;
	zap_attribute_t za;

	VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
	if (doi.doi_type == DMU_OT_BPOBJ) {
		bpobj_free(os, dlobj, tx);
		return;
	}

	for (zap_cursor_init(&zc, os, dlobj);
	    zap_cursor_retrieve(&zc, &za) == 0;
	    zap_cursor_advance(&zc)) {
		uint64_t obj = za.za_first_integer;
		if (obj == dmu_objset_pool(os)->dp_empty_bpobj)
			bpobj_decr_empty(os, tx);
		else
			bpobj_free(os, obj, tx);
	}
	zap_cursor_fini(&zc);
	VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
}

static void
dle_enqueue(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
    const blkptr_t *bp, dmu_tx_t *tx)
{
	ASSERT(MUTEX_HELD(&dl->dl_lock));
	if (dle->dle_bpobj.bpo_object ==
	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
		uint64_t obj = bpobj_alloc(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
		bpobj_close(&dle->dle_bpobj);
		bpobj_decr_empty(dl->dl_os, tx);
		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
		VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
		    dle->dle_mintxg, obj, tx));
	}
	bpobj_enqueue(&dle->dle_bpobj, bp, tx);
}

static void
dle_enqueue_subobj(dsl_deadlist_t *dl, dsl_deadlist_entry_t *dle,
    uint64_t obj, dmu_tx_t *tx)
{
	ASSERT(MUTEX_HELD(&dl->dl_lock));
	if (dle->dle_bpobj.bpo_object !=
	    dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
		bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
	} else {
		bpobj_close(&dle->dle_bpobj);
		bpobj_decr_empty(dl->dl_os, tx);
		VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
		VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
		    dle->dle_mintxg, obj, tx));
	}
}

void
dsl_deadlist_insert(dsl_deadlist_t *dl, const blkptr_t *bp, dmu_tx_t *tx)
{
	dsl_deadlist_entry_t dle_tofind;
	dsl_deadlist_entry_t *dle;
	avl_index_t where;

	if (dl->dl_oldfmt) {
		bpobj_enqueue(&dl->dl_bpobj, bp, tx);
		return;
	}

	mutex_enter(&dl->dl_lock);
	dsl_deadlist_load_tree(dl);

	dmu_buf_will_dirty(dl->dl_dbuf, tx);
	dl->dl_phys->dl_used +=
	    bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
	dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
	dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);

	dle_tofind.dle_mintxg = bp->blk_birth;
	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	if (dle == NULL)
		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
	else
		dle = AVL_PREV(&dl->dl_tree, dle);

	if (dle == NULL) {
		zfs_panic_recover("blkptr at %p has invalid BLK_BIRTH %llu",
		    bp, (longlong_t)bp->blk_birth);
		dle = avl_first(&dl->dl_tree);
	}

	ASSERT3P(dle, !=, NULL);
	dle_enqueue(dl, dle, bp, tx);
	mutex_exit(&dl->dl_lock);
}

/*
 * Insert new key in deadlist, which must be > all current entries.
 * mintxg is not inclusive.
 */
void
dsl_deadlist_add_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
{
	uint64_t obj;
	dsl_deadlist_entry_t *dle;

	if (dl->dl_oldfmt)
		return;

	dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
	dle->dle_mintxg = mintxg;

	mutex_enter(&dl->dl_lock);
	dsl_deadlist_load_tree(dl);

	obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
	avl_add(&dl->dl_tree, dle);

	VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
	    mintxg, obj, tx));
	mutex_exit(&dl->dl_lock);
}

/*
 * Remove this key, merging its entries into the previous key.
 */
void
dsl_deadlist_remove_key(dsl_deadlist_t *dl, uint64_t mintxg, dmu_tx_t *tx)
{
	dsl_deadlist_entry_t dle_tofind;
	dsl_deadlist_entry_t *dle, *dle_prev;

	if (dl->dl_oldfmt)
		return;

	mutex_enter(&dl->dl_lock);
	dsl_deadlist_load_tree(dl);

	dle_tofind.dle_mintxg = mintxg;
	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
	dle_prev = AVL_PREV(&dl->dl_tree, dle);

	dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);

	avl_remove(&dl->dl_tree, dle);
	bpobj_close(&dle->dle_bpobj);
	kmem_free(dle, sizeof (*dle));

	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
	mutex_exit(&dl->dl_lock);
}

/*
 * Walk ds's snapshots to regenerate generate ZAP & AVL.
 */
static void
dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
    uint64_t mrs_obj, dmu_tx_t *tx)
{
	dsl_deadlist_t dl;
	dsl_pool_t *dp = dmu_objset_pool(os);

	dsl_deadlist_open(&dl, os, dlobj);
	if (dl.dl_oldfmt) {
		dsl_deadlist_close(&dl);
		return;
	}

	while (mrs_obj != 0) {
		dsl_dataset_t *ds;
		VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
		dsl_deadlist_add_key(&dl,
		    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
		mrs_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
		dsl_dataset_rele(ds, FTAG);
	}
	dsl_deadlist_close(&dl);
}

uint64_t
dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
    uint64_t mrs_obj, dmu_tx_t *tx)
{
	dsl_deadlist_entry_t *dle;
	uint64_t newobj;

	newobj = dsl_deadlist_alloc(dl->dl_os, tx);

	if (dl->dl_oldfmt) {
		dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
		return (newobj);
	}

	mutex_enter(&dl->dl_lock);
	dsl_deadlist_load_tree(dl);

	for (dle = avl_first(&dl->dl_tree); dle;
	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
		uint64_t obj;

		if (dle->dle_mintxg >= maxtxg)
			break;

		obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
		VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
		    dle->dle_mintxg, obj, tx));
	}
	mutex_exit(&dl->dl_lock);
	return (newobj);
}

void
dsl_deadlist_space(dsl_deadlist_t *dl,
    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
{
	if (dl->dl_oldfmt) {
		VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
		    usedp, compp, uncompp));
		return;
	}

	mutex_enter(&dl->dl_lock);
	*usedp = dl->dl_phys->dl_used;
	*compp = dl->dl_phys->dl_comp;
	*uncompp = dl->dl_phys->dl_uncomp;
	mutex_exit(&dl->dl_lock);
}

/*
 * return space used in the range (mintxg, maxtxg].
 * Includes maxtxg, does not include mintxg.
 * mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
 * larger than any bp in the deadlist (eg. UINT64_MAX)).
 */
void
dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
{
	dsl_deadlist_entry_t *dle;
	dsl_deadlist_entry_t dle_tofind;
	avl_index_t where;

	if (dl->dl_oldfmt) {
		VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
		    mintxg, maxtxg, usedp, compp, uncompp));
		return;
	}

	*usedp = *compp = *uncompp = 0;

	mutex_enter(&dl->dl_lock);
	dsl_deadlist_load_tree(dl);
	dle_tofind.dle_mintxg = mintxg;
	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	/*
	 * If we don't find this mintxg, there shouldn't be anything
	 * after it either.
	 */
	ASSERT(dle != NULL ||
	    avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);

	for (; dle && dle->dle_mintxg < maxtxg;
	    dle = AVL_NEXT(&dl->dl_tree, dle)) {
		uint64_t used, comp, uncomp;

		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
		    &used, &comp, &uncomp));

		*usedp += used;
		*compp += comp;
		*uncompp += uncomp;
	}
	mutex_exit(&dl->dl_lock);
}

static void
dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
    dmu_tx_t *tx)
{
	dsl_deadlist_entry_t dle_tofind;
	dsl_deadlist_entry_t *dle;
	avl_index_t where;
	uint64_t used, comp, uncomp;
	bpobj_t bpo;

	ASSERT(MUTEX_HELD(&dl->dl_lock));

	VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
	VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
	bpobj_close(&bpo);

	dsl_deadlist_load_tree(dl);

	dmu_buf_will_dirty(dl->dl_dbuf, tx);
	dl->dl_phys->dl_used += used;
	dl->dl_phys->dl_comp += comp;
	dl->dl_phys->dl_uncomp += uncomp;

	dle_tofind.dle_mintxg = birth;
	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	if (dle == NULL)
		dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
	dle_enqueue_subobj(dl, dle, obj, tx);
}

static int
dsl_deadlist_insert_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
	dsl_deadlist_t *dl = arg;
	dsl_deadlist_insert(dl, bp, tx);
	return (0);
}

/*
 * Merge the deadlist pointed to by 'obj' into dl.  obj will be left as
 * an empty deadlist.
 */
void
dsl_deadlist_merge(dsl_deadlist_t *dl, uint64_t obj, dmu_tx_t *tx)
{
	zap_cursor_t zc;
	zap_attribute_t za;
	dmu_buf_t *bonus;
	dsl_deadlist_phys_t *dlp;
	dmu_object_info_t doi;

	VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
	if (doi.doi_type == DMU_OT_BPOBJ) {
		bpobj_t bpo;
		VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
		VERIFY3U(0, ==, bpobj_iterate(&bpo,
		    dsl_deadlist_insert_cb, dl, tx));
		bpobj_close(&bpo);
		return;
	}

	mutex_enter(&dl->dl_lock);
	for (zap_cursor_init(&zc, dl->dl_os, obj);
	    zap_cursor_retrieve(&zc, &za) == 0;
	    zap_cursor_advance(&zc)) {
		uint64_t mintxg = zfs_strtonum(za.za_name, NULL);
		dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
	}
	zap_cursor_fini(&zc);

	VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
	dlp = bonus->db_data;
	dmu_buf_will_dirty(bonus, tx);
	bzero(dlp, sizeof (*dlp));
	dmu_buf_rele(bonus, FTAG);
	mutex_exit(&dl->dl_lock);
}

/*
 * Remove entries on dl that are >= mintxg, and put them on the bpobj.
 */
void
dsl_deadlist_move_bpobj(dsl_deadlist_t *dl, bpobj_t *bpo, uint64_t mintxg,
    dmu_tx_t *tx)
{
	dsl_deadlist_entry_t dle_tofind;
	dsl_deadlist_entry_t *dle;
	avl_index_t where;

	ASSERT(!dl->dl_oldfmt);

	mutex_enter(&dl->dl_lock);
	dmu_buf_will_dirty(dl->dl_dbuf, tx);
	dsl_deadlist_load_tree(dl);

	dle_tofind.dle_mintxg = mintxg;
	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	if (dle == NULL)
		dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
	while (dle) {
		uint64_t used, comp, uncomp;
		dsl_deadlist_entry_t *dle_next;

		bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);

		VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
		    &used, &comp, &uncomp));
		ASSERT3U(dl->dl_phys->dl_used, >=, used);
		ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
		ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
		dl->dl_phys->dl_used -= used;
		dl->dl_phys->dl_comp -= comp;
		dl->dl_phys->dl_uncomp -= uncomp;

		VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
		    dle->dle_mintxg, tx));

		dle_next = AVL_NEXT(&dl->dl_tree, dle);
		avl_remove(&dl->dl_tree, dle);
		bpobj_close(&dle->dle_bpobj);
		kmem_free(dle, sizeof (*dle));
		dle = dle_next;
	}
	mutex_exit(&dl->dl_lock);
}
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) 2010, Oracle and/or its affiliates. All rights reserved.
cae5b340	23	* Copyright (c) 2012, 2016 by Delphix. All rights reserved.
e10b0808	24	* Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
428870ff BB	25	*/
	26
	27	#include <sys/dsl_dataset.h>
	28	#include <sys/dmu.h>
	29	#include <sys/refcount.h>
	30	#include <sys/zap.h>
	31	#include <sys/zfs_context.h>
	32	#include <sys/dsl_pool.h>
	33
330d06f9 MA	34	/*
	35	* Deadlist concurrency:
	36	*
	37	* Deadlists can only be modified from the syncing thread.
	38	*
	39	* Except for dsl_deadlist_insert(), it can only be modified with the
	40	* dp_config_rwlock held with RW_WRITER.
	41	*
	42	* The accessors (dsl_deadlist_space() and dsl_deadlist_space_range()) can
	43	* be called concurrently, from open context, with the dl_config_rwlock held
	44	* with RW_READER.
	45	*
	46	* Therefore, we only need to provide locking between dsl_deadlist_insert() and
	47	* the accessors, protecting:
	48	* dl_phys->dl_used,comp,uncomp
	49	* and protecting the dl_tree from being loaded.
	50	* The locking is provided by dl_lock. Note that locking on the bpobj_t
	51	* provides its own locking, and dl_oldfmt is immutable.
	52	*/
	53
428870ff BB	54	static int
	55	dsl_deadlist_compare(const void arg1, const void arg2)
	56	{
cae5b340 AX	57	const dsl_deadlist_entry_t dle1 = (const dsl_deadlist_entry_t )arg1;
cae5b340 AX	58	const dsl_deadlist_entry_t dle2 = (const dsl_deadlist_entry_t )arg2;
428870ff	59
cae5b340	60	return (AVL_CMP(dle1->dle_mintxg, dle2->dle_mintxg));
428870ff BB	61	}
	62
	63	static void
	64	dsl_deadlist_load_tree(dsl_deadlist_t *dl)
	65	{
	66	zap_cursor_t zc;
	67	zap_attribute_t za;
	68
cae5b340 AX	69	ASSERT(MUTEX_HELD(&dl->dl_lock));
cae5b340 AX	70
428870ff BB	71	ASSERT(!dl->dl_oldfmt);
	72	if (dl->dl_havetree)
	73	return;
	74
	75	avl_create(&dl->dl_tree, dsl_deadlist_compare,
	76	sizeof (dsl_deadlist_entry_t),
	77	offsetof(dsl_deadlist_entry_t, dle_node));
	78	for (zap_cursor_init(&zc, dl->dl_os, dl->dl_object);
	79	zap_cursor_retrieve(&zc, &za) == 0;
	80	zap_cursor_advance(&zc)) {
cae5b340 AX	81	dsl_deadlist_entry_t dle = kmem_alloc(sizeof (dle), KM_SLEEP);
cae5b340 AX	82	dle->dle_mintxg = zfs_strtonum(za.za_name, NULL);
428870ff BB	83	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os,
	84	za.za_first_integer));
	85	avl_add(&dl->dl_tree, dle);
	86	}
	87	zap_cursor_fini(&zc);
	88	dl->dl_havetree = B_TRUE;
	89	}
	90
	91	void
	92	dsl_deadlist_open(dsl_deadlist_t dl, objset_t os, uint64_t object)
	93	{
	94	dmu_object_info_t doi;
	95
	96	mutex_init(&dl->dl_lock, NULL, MUTEX_DEFAULT, NULL);
	97	dl->dl_os = os;
	98	dl->dl_object = object;
	99	VERIFY3U(0, ==, dmu_bonus_hold(os, object, dl, &dl->dl_dbuf));
	100	dmu_object_info_from_db(dl->dl_dbuf, &doi);
	101	if (doi.doi_type == DMU_OT_BPOBJ) {
	102	dmu_buf_rele(dl->dl_dbuf, dl);
	103	dl->dl_dbuf = NULL;
	104	dl->dl_oldfmt = B_TRUE;
	105	VERIFY3U(0, ==, bpobj_open(&dl->dl_bpobj, os, object));
	106	return;
	107	}
	108
	109	dl->dl_oldfmt = B_FALSE;
	110	dl->dl_phys = dl->dl_dbuf->db_data;
	111	dl->dl_havetree = B_FALSE;
	112	}
	113
	114	void
	115	dsl_deadlist_close(dsl_deadlist_t *dl)
	116	{
	117	void *cookie = NULL;
	118	dsl_deadlist_entry_t *dle;
	119
e10b0808	120	dl->dl_os = NULL;
cae5b340	121	mutex_destroy(&dl->dl_lock);
e10b0808	122
428870ff BB	123	if (dl->dl_oldfmt) {
	124	dl->dl_oldfmt = B_FALSE;
	125	bpobj_close(&dl->dl_bpobj);
	126	return;
	127	}
	128
	129	if (dl->dl_havetree) {
	130	while ((dle = avl_destroy_nodes(&dl->dl_tree, &cookie))
	131	!= NULL) {
	132	bpobj_close(&dle->dle_bpobj);
	133	kmem_free(dle, sizeof (*dle));
	134	}
	135	avl_destroy(&dl->dl_tree);
	136	}
	137	dmu_buf_rele(dl->dl_dbuf, dl);
428870ff BB	138	dl->dl_dbuf = NULL;
	139	dl->dl_phys = NULL;
	140	}
	141
	142	uint64_t
	143	dsl_deadlist_alloc(objset_t os, dmu_tx_t tx)
	144	{
	145	if (spa_version(dmu_objset_spa(os)) < SPA_VERSION_DEADLISTS)
e10b0808	146	return (bpobj_alloc(os, SPA_OLD_MAXBLOCKSIZE, tx));
428870ff BB	147	return (zap_create(os, DMU_OT_DEADLIST, DMU_OT_DEADLIST_HDR,
	148	sizeof (dsl_deadlist_phys_t), tx));
	149	}
	150
	151	void
	152	dsl_deadlist_free(objset_t os, uint64_t dlobj, dmu_tx_t tx)
	153	{
	154	dmu_object_info_t doi;
	155	zap_cursor_t zc;
	156	zap_attribute_t za;
	157
	158	VERIFY3U(0, ==, dmu_object_info(os, dlobj, &doi));
	159	if (doi.doi_type == DMU_OT_BPOBJ) {
	160	bpobj_free(os, dlobj, tx);
	161	return;
	162	}
	163
	164	for (zap_cursor_init(&zc, os, dlobj);
	165	zap_cursor_retrieve(&zc, &za) == 0;
753c3839 MA	166	zap_cursor_advance(&zc)) {
	167	uint64_t obj = za.za_first_integer;
	168	if (obj == dmu_objset_pool(os)->dp_empty_bpobj)
	169	bpobj_decr_empty(os, tx);
	170	else
	171	bpobj_free(os, obj, tx);
	172	}
428870ff BB	173	zap_cursor_fini(&zc);
	174	VERIFY3U(0, ==, dmu_object_free(os, dlobj, tx));
	175	}
	176
753c3839 MA	177	static void
	178	dle_enqueue(dsl_deadlist_t dl, dsl_deadlist_entry_t dle,
	179	const blkptr_t bp, dmu_tx_t tx)
	180	{
cae5b340	181	ASSERT(MUTEX_HELD(&dl->dl_lock));
753c3839 MA	182	if (dle->dle_bpobj.bpo_object ==
753c3839 MA	183	dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
e10b0808	184	uint64_t obj = bpobj_alloc(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
753c3839 MA	185	bpobj_close(&dle->dle_bpobj);
	186	bpobj_decr_empty(dl->dl_os, tx);
	187	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
	188	VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
	189	dle->dle_mintxg, obj, tx));
	190	}
	191	bpobj_enqueue(&dle->dle_bpobj, bp, tx);
	192	}
	193
	194	static void
	195	dle_enqueue_subobj(dsl_deadlist_t dl, dsl_deadlist_entry_t dle,
	196	uint64_t obj, dmu_tx_t *tx)
	197	{
cae5b340	198	ASSERT(MUTEX_HELD(&dl->dl_lock));
753c3839 MA	199	if (dle->dle_bpobj.bpo_object !=
	200	dmu_objset_pool(dl->dl_os)->dp_empty_bpobj) {
	201	bpobj_enqueue_subobj(&dle->dle_bpobj, obj, tx);
	202	} else {
	203	bpobj_close(&dle->dle_bpobj);
	204	bpobj_decr_empty(dl->dl_os, tx);
	205	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
	206	VERIFY3U(0, ==, zap_update_int_key(dl->dl_os, dl->dl_object,
	207	dle->dle_mintxg, obj, tx));
	208	}
	209	}
	210
428870ff BB	211	void
	212	dsl_deadlist_insert(dsl_deadlist_t dl, const blkptr_t bp, dmu_tx_t *tx)
	213	{
	214	dsl_deadlist_entry_t dle_tofind;
	215	dsl_deadlist_entry_t *dle;
	216	avl_index_t where;
	217
	218	if (dl->dl_oldfmt) {
	219	bpobj_enqueue(&dl->dl_bpobj, bp, tx);
	220	return;
	221	}
	222
cae5b340	223	mutex_enter(&dl->dl_lock);
428870ff BB	224	dsl_deadlist_load_tree(dl);
	225
	226	dmu_buf_will_dirty(dl->dl_dbuf, tx);
428870ff BB	227	dl->dl_phys->dl_used +=
	228	bp_get_dsize_sync(dmu_objset_spa(dl->dl_os), bp);
	229	dl->dl_phys->dl_comp += BP_GET_PSIZE(bp);
	230	dl->dl_phys->dl_uncomp += BP_GET_UCSIZE(bp);
428870ff BB	231
	232	dle_tofind.dle_mintxg = bp->blk_birth;
	233	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	234	if (dle == NULL)
	235	dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
	236	else
	237	dle = AVL_PREV(&dl->dl_tree, dle);
cae5b340 AX	238
	239	if (dle == NULL) {
	240	zfs_panic_recover("blkptr at %p has invalid BLK_BIRTH %llu",
	241	bp, (longlong_t)bp->blk_birth);
	242	dle = avl_first(&dl->dl_tree);
	243	}
	244
	245	ASSERT3P(dle, !=, NULL);
753c3839	246	dle_enqueue(dl, dle, bp, tx);
cae5b340	247	mutex_exit(&dl->dl_lock);
428870ff BB	248	}
	249
	250	/*
	251	* Insert new key in deadlist, which must be > all current entries.
	252	* mintxg is not inclusive.
	253	*/
	254	void
	255	dsl_deadlist_add_key(dsl_deadlist_t dl, uint64_t mintxg, dmu_tx_t tx)
	256	{
	257	uint64_t obj;
	258	dsl_deadlist_entry_t *dle;
	259
	260	if (dl->dl_oldfmt)
	261	return;
	262
ea04106b	263	dle = kmem_alloc(sizeof (*dle), KM_SLEEP);
428870ff	264	dle->dle_mintxg = mintxg;
cae5b340 AX	265
	266	mutex_enter(&dl->dl_lock);
	267	dsl_deadlist_load_tree(dl);
	268
e10b0808	269	obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
428870ff BB	270	VERIFY3U(0, ==, bpobj_open(&dle->dle_bpobj, dl->dl_os, obj));
	271	avl_add(&dl->dl_tree, dle);
	272
	273	VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, dl->dl_object,
	274	mintxg, obj, tx));
cae5b340	275	mutex_exit(&dl->dl_lock);
428870ff BB	276	}
	277
	278	/*
	279	* Remove this key, merging its entries into the previous key.
	280	*/
	281	void
	282	dsl_deadlist_remove_key(dsl_deadlist_t dl, uint64_t mintxg, dmu_tx_t tx)
	283	{
	284	dsl_deadlist_entry_t dle_tofind;
	285	dsl_deadlist_entry_t dle, dle_prev;
	286
	287	if (dl->dl_oldfmt)
	288	return;
	289
cae5b340	290	mutex_enter(&dl->dl_lock);
428870ff BB	291	dsl_deadlist_load_tree(dl);
	292
	293	dle_tofind.dle_mintxg = mintxg;
	294	dle = avl_find(&dl->dl_tree, &dle_tofind, NULL);
	295	dle_prev = AVL_PREV(&dl->dl_tree, dle);
	296
753c3839	297	dle_enqueue_subobj(dl, dle_prev, dle->dle_bpobj.bpo_object, tx);
428870ff BB	298
	299	avl_remove(&dl->dl_tree, dle);
	300	bpobj_close(&dle->dle_bpobj);
	301	kmem_free(dle, sizeof (*dle));
	302
	303	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object, mintxg, tx));
cae5b340	304	mutex_exit(&dl->dl_lock);
428870ff BB	305	}
	306
	307	/*
	308	* Walk ds's snapshots to regenerate generate ZAP & AVL.
	309	*/
	310	static void
	311	dsl_deadlist_regenerate(objset_t *os, uint64_t dlobj,
	312	uint64_t mrs_obj, dmu_tx_t *tx)
	313	{
	314	dsl_deadlist_t dl;
	315	dsl_pool_t *dp = dmu_objset_pool(os);
	316
	317	dsl_deadlist_open(&dl, os, dlobj);
	318	if (dl.dl_oldfmt) {
	319	dsl_deadlist_close(&dl);
	320	return;
	321	}
	322
	323	while (mrs_obj != 0) {
	324	dsl_dataset_t *ds;
	325	VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, mrs_obj, FTAG, &ds));
e10b0808 AX	326	dsl_deadlist_add_key(&dl,
	327	dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
	328	mrs_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
428870ff BB	329	dsl_dataset_rele(ds, FTAG);
	330	}
	331	dsl_deadlist_close(&dl);
	332	}
	333
	334	uint64_t
	335	dsl_deadlist_clone(dsl_deadlist_t *dl, uint64_t maxtxg,
	336	uint64_t mrs_obj, dmu_tx_t *tx)
	337	{
	338	dsl_deadlist_entry_t *dle;
	339	uint64_t newobj;
	340
	341	newobj = dsl_deadlist_alloc(dl->dl_os, tx);
	342
	343	if (dl->dl_oldfmt) {
	344	dsl_deadlist_regenerate(dl->dl_os, newobj, mrs_obj, tx);
	345	return (newobj);
	346	}
	347
cae5b340	348	mutex_enter(&dl->dl_lock);
428870ff BB	349	dsl_deadlist_load_tree(dl);
	350
	351	for (dle = avl_first(&dl->dl_tree); dle;
	352	dle = AVL_NEXT(&dl->dl_tree, dle)) {
	353	uint64_t obj;
	354
	355	if (dle->dle_mintxg >= maxtxg)
	356	break;
	357
e10b0808	358	obj = bpobj_alloc_empty(dl->dl_os, SPA_OLD_MAXBLOCKSIZE, tx);
428870ff BB	359	VERIFY3U(0, ==, zap_add_int_key(dl->dl_os, newobj,
	360	dle->dle_mintxg, obj, tx));
	361	}
cae5b340	362	mutex_exit(&dl->dl_lock);
428870ff BB	363	return (newobj);
	364	}
	365
	366	void
	367	dsl_deadlist_space(dsl_deadlist_t *dl,
	368	uint64_t usedp, uint64_t compp, uint64_t *uncompp)
	369	{
	370	if (dl->dl_oldfmt) {
	371	VERIFY3U(0, ==, bpobj_space(&dl->dl_bpobj,
	372	usedp, compp, uncompp));
	373	return;
	374	}
	375
	376	mutex_enter(&dl->dl_lock);
	377	*usedp = dl->dl_phys->dl_used;
	378	*compp = dl->dl_phys->dl_comp;
	379	*uncompp = dl->dl_phys->dl_uncomp;
	380	mutex_exit(&dl->dl_lock);
	381	}
	382
	383	/*
	384	* return space used in the range (mintxg, maxtxg].
	385	* Includes maxtxg, does not include mintxg.
	386	* mintxg and maxtxg must both be keys in the deadlist (unless maxtxg is
330d06f9	387	* larger than any bp in the deadlist (eg. UINT64_MAX)).
428870ff BB	388	*/
	389	void
	390	dsl_deadlist_space_range(dsl_deadlist_t *dl, uint64_t mintxg, uint64_t maxtxg,
	391	uint64_t usedp, uint64_t compp, uint64_t *uncompp)
	392	{
428870ff	393	dsl_deadlist_entry_t *dle;
330d06f9	394	dsl_deadlist_entry_t dle_tofind;
428870ff BB	395	avl_index_t where;
	396
	397	if (dl->dl_oldfmt) {
	398	VERIFY3U(0, ==, bpobj_space_range(&dl->dl_bpobj,
	399	mintxg, maxtxg, usedp, compp, uncompp));
	400	return;
	401	}
	402
428870ff BB	403	usedp = compp = *uncompp = 0;
428870ff BB	404
330d06f9 MA	405	mutex_enter(&dl->dl_lock);
330d06f9 MA	406	dsl_deadlist_load_tree(dl);
428870ff BB	407	dle_tofind.dle_mintxg = mintxg;
	408	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	409	/*
	410	* If we don't find this mintxg, there shouldn't be anything
	411	* after it either.
	412	*/
	413	ASSERT(dle != NULL \|\|
	414	avl_nearest(&dl->dl_tree, where, AVL_AFTER) == NULL);
330d06f9	415
428870ff BB	416	for (; dle && dle->dle_mintxg < maxtxg;
	417	dle = AVL_NEXT(&dl->dl_tree, dle)) {
	418	uint64_t used, comp, uncomp;
	419
	420	VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
	421	&used, &comp, &uncomp));
	422
	423	*usedp += used;
	424	*compp += comp;
	425	*uncompp += uncomp;
	426	}
330d06f9	427	mutex_exit(&dl->dl_lock);
428870ff BB	428	}
	429
	430	static void
	431	dsl_deadlist_insert_bpobj(dsl_deadlist_t *dl, uint64_t obj, uint64_t birth,
	432	dmu_tx_t *tx)
	433	{
	434	dsl_deadlist_entry_t dle_tofind;
	435	dsl_deadlist_entry_t *dle;
	436	avl_index_t where;
	437	uint64_t used, comp, uncomp;
	438	bpobj_t bpo;
	439
cae5b340 AX	440	ASSERT(MUTEX_HELD(&dl->dl_lock));
cae5b340 AX	441
428870ff BB	442	VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
	443	VERIFY3U(0, ==, bpobj_space(&bpo, &used, &comp, &uncomp));
	444	bpobj_close(&bpo);
	445
	446	dsl_deadlist_load_tree(dl);
	447
	448	dmu_buf_will_dirty(dl->dl_dbuf, tx);
428870ff BB	449	dl->dl_phys->dl_used += used;
	450	dl->dl_phys->dl_comp += comp;
	451	dl->dl_phys->dl_uncomp += uncomp;
428870ff BB	452
	453	dle_tofind.dle_mintxg = birth;
	454	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	455	if (dle == NULL)
	456	dle = avl_nearest(&dl->dl_tree, where, AVL_BEFORE);
753c3839	457	dle_enqueue_subobj(dl, dle, obj, tx);
428870ff BB	458	}
	459
	460	static int
	461	dsl_deadlist_insert_cb(void arg, const blkptr_t bp, dmu_tx_t *tx)
	462	{
	463	dsl_deadlist_t *dl = arg;
	464	dsl_deadlist_insert(dl, bp, tx);
	465	return (0);
	466	}
	467
	468	/*
	469	* Merge the deadlist pointed to by 'obj' into dl. obj will be left as
	470	* an empty deadlist.
	471	*/
	472	void
	473	dsl_deadlist_merge(dsl_deadlist_t dl, uint64_t obj, dmu_tx_t tx)
	474	{
	475	zap_cursor_t zc;
	476	zap_attribute_t za;
	477	dmu_buf_t *bonus;
	478	dsl_deadlist_phys_t *dlp;
	479	dmu_object_info_t doi;
	480
	481	VERIFY3U(0, ==, dmu_object_info(dl->dl_os, obj, &doi));
	482	if (doi.doi_type == DMU_OT_BPOBJ) {
	483	bpobj_t bpo;
	484	VERIFY3U(0, ==, bpobj_open(&bpo, dl->dl_os, obj));
	485	VERIFY3U(0, ==, bpobj_iterate(&bpo,
	486	dsl_deadlist_insert_cb, dl, tx));
	487	bpobj_close(&bpo);
	488	return;
	489	}
	490
cae5b340	491	mutex_enter(&dl->dl_lock);
428870ff BB	492	for (zap_cursor_init(&zc, dl->dl_os, obj);
	493	zap_cursor_retrieve(&zc, &za) == 0;
	494	zap_cursor_advance(&zc)) {
cae5b340	495	uint64_t mintxg = zfs_strtonum(za.za_name, NULL);
428870ff BB	496	dsl_deadlist_insert_bpobj(dl, za.za_first_integer, mintxg, tx);
	497	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, obj, mintxg, tx));
	498	}
	499	zap_cursor_fini(&zc);
	500
	501	VERIFY3U(0, ==, dmu_bonus_hold(dl->dl_os, obj, FTAG, &bonus));
	502	dlp = bonus->db_data;
	503	dmu_buf_will_dirty(bonus, tx);
	504	bzero(dlp, sizeof (*dlp));
	505	dmu_buf_rele(bonus, FTAG);
cae5b340	506	mutex_exit(&dl->dl_lock);
428870ff BB	507	}
	508
	509	/*
	510	* Remove entries on dl that are >= mintxg, and put them on the bpobj.
	511	*/
	512	void
	513	dsl_deadlist_move_bpobj(dsl_deadlist_t dl, bpobj_t bpo, uint64_t mintxg,
	514	dmu_tx_t *tx)
	515	{
	516	dsl_deadlist_entry_t dle_tofind;
	517	dsl_deadlist_entry_t *dle;
	518	avl_index_t where;
	519
	520	ASSERT(!dl->dl_oldfmt);
cae5b340 AX	521
cae5b340 AX	522	mutex_enter(&dl->dl_lock);
428870ff BB	523	dmu_buf_will_dirty(dl->dl_dbuf, tx);
	524	dsl_deadlist_load_tree(dl);
	525
	526	dle_tofind.dle_mintxg = mintxg;
	527	dle = avl_find(&dl->dl_tree, &dle_tofind, &where);
	528	if (dle == NULL)
	529	dle = avl_nearest(&dl->dl_tree, where, AVL_AFTER);
	530	while (dle) {
	531	uint64_t used, comp, uncomp;
	532	dsl_deadlist_entry_t *dle_next;
	533
	534	bpobj_enqueue_subobj(bpo, dle->dle_bpobj.bpo_object, tx);
	535
	536	VERIFY3U(0, ==, bpobj_space(&dle->dle_bpobj,
	537	&used, &comp, &uncomp));
428870ff BB	538	ASSERT3U(dl->dl_phys->dl_used, >=, used);
	539	ASSERT3U(dl->dl_phys->dl_comp, >=, comp);
	540	ASSERT3U(dl->dl_phys->dl_uncomp, >=, uncomp);
	541	dl->dl_phys->dl_used -= used;
	542	dl->dl_phys->dl_comp -= comp;
	543	dl->dl_phys->dl_uncomp -= uncomp;
428870ff BB	544
	545	VERIFY3U(0, ==, zap_remove_int(dl->dl_os, dl->dl_object,
	546	dle->dle_mintxg, tx));
	547
	548	dle_next = AVL_NEXT(&dl->dl_tree, dle);
	549	avl_remove(&dl->dl_tree, dle);
	550	bpobj_close(&dle->dle_bpobj);
	551	kmem_free(dle, sizeof (*dle));
	552	dle = dle_next;
	553	}
cae5b340	554	mutex_exit(&dl->dl_lock);
428870ff	555	}