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

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

#include <sys/zfs_context.h>
#include <sys/dmu_objset.h>
#include <sys/dmu_traverse.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_pool.h>
#include <sys/dnode.h>
#include <sys/spa.h>
#include <sys/zio.h>
#include <sys/dmu_impl.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
#include <sys/callb.h>

int zfs_pd_blks_max = 100;

typedef struct prefetch_data {
	kmutex_t pd_mtx;
	kcondvar_t pd_cv;
	int pd_blks_max;
	int pd_blks_fetched;
	int pd_flags;
	boolean_t pd_cancel;
	boolean_t pd_exited;
} prefetch_data_t;

typedef struct traverse_data {
	spa_t *td_spa;
	uint64_t td_objset;
	blkptr_t *td_rootbp;
	uint64_t td_min_txg;
	int td_flags;
	prefetch_data_t *td_pfd;
	blkptr_cb_t *td_func;
	void *td_arg;
} traverse_data_t;

static int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
    arc_buf_t *buf, uint64_t objset, uint64_t object);

static int
traverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
{
	traverse_data_t *td = arg;
	zbookmark_t zb;

	if (bp->blk_birth == 0)
		return (0);

	if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
		return (0);

	SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
	    bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);

	(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, td->td_arg);

	return (0);
}

static int
traverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
{
	traverse_data_t *td = arg;

	if (lrc->lrc_txtype == TX_WRITE) {
		lr_write_t *lr = (lr_write_t *)lrc;
		blkptr_t *bp = &lr->lr_blkptr;
		zbookmark_t zb;

		if (bp->blk_birth == 0)
			return (0);

		if (claim_txg == 0 || bp->blk_birth < claim_txg)
			return (0);

		SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
		    ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));

		(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL,
		    td->td_arg);
	}
	return (0);
}

static void
traverse_zil(traverse_data_t *td, zil_header_t *zh)
{
	uint64_t claim_txg = zh->zh_claim_txg;
	zilog_t *zilog;

	/*
	 * We only want to visit blocks that have been claimed but not yet
	 * replayed; plus, in read-only mode, blocks that are already stable.
	 */
	if (claim_txg == 0 && spa_writeable(td->td_spa))
		return;

	zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);

	(void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
	    claim_txg);

	zil_free(zilog);
}

static int
traverse_visitbp(traverse_data_t *td, const dnode_phys_t *dnp,
    arc_buf_t *pbuf, blkptr_t *bp, const zbookmark_t *zb)
{
	zbookmark_t czb;
	int err = 0, lasterr = 0;
	arc_buf_t *buf = NULL;
	prefetch_data_t *pd = td->td_pfd;
	boolean_t hard = td->td_flags & TRAVERSE_HARD;

	if (bp->blk_birth == 0) {
		err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
		    td->td_arg);
		return (err);
	}

	if (bp->blk_birth <= td->td_min_txg)
		return (0);

	if (pd && !pd->pd_exited &&
	    ((pd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0)) {
		mutex_enter(&pd->pd_mtx);
		ASSERT(pd->pd_blks_fetched >= 0);
		while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
			cv_wait(&pd->pd_cv, &pd->pd_mtx);
		pd->pd_blks_fetched--;
		cv_broadcast(&pd->pd_cv);
		mutex_exit(&pd->pd_mtx);
	}

	if (td->td_flags & TRAVERSE_PRE) {
		err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
		    td->td_arg);
		if (err == TRAVERSE_VISIT_NO_CHILDREN)
			return (0);
		if (err)
			return (err);
	}

	if (BP_GET_LEVEL(bp) > 0) {
		uint32_t flags = ARC_WAIT;
		int i;
		blkptr_t *cbp;
		int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;

		err = dsl_read(NULL, td->td_spa, bp, pbuf,
		    arc_getbuf_func, &buf,
		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
		if (err)
			return (err);

		/* recursively visitbp() blocks below this */
		cbp = buf->b_data;
		for (i = 0; i < epb; i++, cbp++) {
			SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
			    zb->zb_level - 1,
			    zb->zb_blkid * epb + i);
			err = traverse_visitbp(td, dnp, buf, cbp, &czb);
			if (err) {
				if (!hard)
					break;
				lasterr = err;
			}
		}
	} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
		uint32_t flags = ARC_WAIT;
		int i;
		int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;

		err = dsl_read(NULL, td->td_spa, bp, pbuf,
		    arc_getbuf_func, &buf,
		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
		if (err)
			return (err);

		/* recursively visitbp() blocks below this */
		dnp = buf->b_data;
		for (i = 0; i < epb; i++, dnp++) {
			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
			    zb->zb_blkid * epb + i);
			if (err) {
				if (!hard)
					break;
				lasterr = err;
			}
		}
	} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
		uint32_t flags = ARC_WAIT;
		objset_phys_t *osp;
		dnode_phys_t *dnp;

		err = dsl_read_nolock(NULL, td->td_spa, bp,
		    arc_getbuf_func, &buf,
		    ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
		if (err)
			return (err);

		osp = buf->b_data;
		dnp = &osp->os_meta_dnode;
		err = traverse_dnode(td, dnp, buf, zb->zb_objset,
		    DMU_META_DNODE_OBJECT);
		if (err && hard) {
			lasterr = err;
			err = 0;
		}
		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
			dnp = &osp->os_userused_dnode;
			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
			    DMU_USERUSED_OBJECT);
		}
		if (err && hard) {
			lasterr = err;
			err = 0;
		}
		if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
			dnp = &osp->os_groupused_dnode;
			err = traverse_dnode(td, dnp, buf, zb->zb_objset,
			    DMU_GROUPUSED_OBJECT);
		}
	}

	if (buf)
		(void) arc_buf_remove_ref(buf, &buf);

	if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) {
		err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
		    td->td_arg);
	}

	return (err != 0 ? err : lasterr);
}

static int
traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
    arc_buf_t *buf, uint64_t objset, uint64_t object)
{
	int j, err = 0, lasterr = 0;
	zbookmark_t czb;
	boolean_t hard = (td->td_flags & TRAVERSE_HARD);

	for (j = 0; j < dnp->dn_nblkptr; j++) {
		SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
		err = traverse_visitbp(td, dnp, buf,
		    (blkptr_t *)&dnp->dn_blkptr[j], &czb);
		if (err) {
			if (!hard)
				break;
			lasterr = err;
		}
	}

	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
		SET_BOOKMARK(&czb, objset,
		    object, 0, DMU_SPILL_BLKID);
		err = traverse_visitbp(td, dnp, buf,
		    (blkptr_t *)&dnp->dn_spill, &czb);
		if (err) {
			if (!hard)
				return (err);
			lasterr = err;
		}
	}
	return (err != 0 ? err : lasterr);
}

/* ARGSUSED */
static int
traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
    arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp,
    void *arg)
{
	prefetch_data_t *pfd = arg;
	uint32_t aflags = ARC_NOWAIT | ARC_PREFETCH;

	ASSERT(pfd->pd_blks_fetched >= 0);
	if (pfd->pd_cancel)
		return (EINTR);

	if (bp == NULL || !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) ||
	    BP_GET_TYPE(bp) == DMU_OT_DNODE || BP_GET_LEVEL(bp) > 0) ||
	    BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
		return (0);

	mutex_enter(&pfd->pd_mtx);
	while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
		cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
	pfd->pd_blks_fetched++;
	cv_broadcast(&pfd->pd_cv);
	mutex_exit(&pfd->pd_mtx);

	(void) dsl_read(NULL, spa, bp, pbuf, NULL, NULL,
	    ZIO_PRIORITY_ASYNC_READ,
	    ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
	    &aflags, zb);

	return (0);
}

static void
traverse_prefetch_thread(void *arg)
{
	traverse_data_t *td_main = arg;
	traverse_data_t td = *td_main;
	zbookmark_t czb;

	td.td_func = traverse_prefetcher;
	td.td_arg = td_main->td_pfd;
	td.td_pfd = NULL;

	SET_BOOKMARK(&czb, td.td_objset,
	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
	(void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb);

	mutex_enter(&td_main->td_pfd->pd_mtx);
	td_main->td_pfd->pd_exited = B_TRUE;
	cv_broadcast(&td_main->td_pfd->pd_cv);
	mutex_exit(&td_main->td_pfd->pd_mtx);
}

/*
 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 * in syncing context).
 */
static int
traverse_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *rootbp,
    uint64_t txg_start, int flags, blkptr_cb_t func, void *arg)
{
	traverse_data_t td;
	prefetch_data_t pd = { 0 };
	zbookmark_t czb;
	int err;

	td.td_spa = spa;
	td.td_objset = ds ? ds->ds_object : 0;
	td.td_rootbp = rootbp;
	td.td_min_txg = txg_start;
	td.td_func = func;
	td.td_arg = arg;
	td.td_pfd = &pd;
	td.td_flags = flags;

	pd.pd_blks_max = zfs_pd_blks_max;
	pd.pd_flags = flags;
	mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
	cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);

	/* See comment on ZIL traversal in dsl_scan_visitds. */
	if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
		objset_t *os;

		err = dmu_objset_from_ds(ds, &os);
		if (err)
			return (err);

		traverse_zil(&td, &os->os_zil_header);
	}

	if (!(flags & TRAVERSE_PREFETCH) ||
	    0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
	    &td, TQ_NOQUEUE))
		pd.pd_exited = B_TRUE;

	SET_BOOKMARK(&czb, td.td_objset,
	    ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
	err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);

	mutex_enter(&pd.pd_mtx);
	pd.pd_cancel = B_TRUE;
	cv_broadcast(&pd.pd_cv);
	while (!pd.pd_exited)
		cv_wait(&pd.pd_cv, &pd.pd_mtx);
	mutex_exit(&pd.pd_mtx);

	mutex_destroy(&pd.pd_mtx);
	cv_destroy(&pd.pd_cv);

	return (err);
}

/*
 * NB: dataset must not be changing on-disk (eg, is a snapshot or we are
 * in syncing context).
 */
int
traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
    blkptr_cb_t func, void *arg)
{
	return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds,
	    &ds->ds_phys->ds_bp, txg_start, flags, func, arg));
}

/*
 * NB: pool must not be changing on-disk (eg, from zdb or sync context).
 */
int
traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
    blkptr_cb_t func, void *arg)
{
	int err, lasterr = 0;
	uint64_t obj;
	dsl_pool_t *dp = spa_get_dsl(spa);
	objset_t *mos = dp->dp_meta_objset;
	boolean_t hard = (flags & TRAVERSE_HARD);

	/* visit the MOS */
	err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa),
	    txg_start, flags, func, arg);
	if (err)
		return (err);

	/* visit each dataset */
	for (obj = 1; err == 0 || (err != ESRCH && hard);
	    err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
		dmu_object_info_t doi;

		err = dmu_object_info(mos, obj, &doi);
		if (err) {
			if (!hard)
				return (err);
			lasterr = err;
			continue;
		}

		if (doi.doi_type == DMU_OT_DSL_DATASET) {
			dsl_dataset_t *ds;
			uint64_t txg = txg_start;

			rw_enter(&dp->dp_config_rwlock, RW_READER);
			err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
			rw_exit(&dp->dp_config_rwlock);
			if (err) {
				if (!hard)
					return (err);
				lasterr = err;
				continue;
			}
			if (ds->ds_phys->ds_prev_snap_txg > txg)
				txg = ds->ds_phys->ds_prev_snap_txg;
			err = traverse_dataset(ds, txg, flags, func, arg);
			dsl_dataset_rele(ds, FTAG);
			if (err) {
				if (!hard)
					return (err);
				lasterr = err;
			}
		}
	}
	if (err == ESRCH)
		err = 0;
	return (err != 0 ? err : lasterr);
}
Commit	Line	Data
34dc7c2f 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	/*
428870ff	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
34dc7c2f BB	23	*/
34dc7c2f BB	24
34dc7c2f BB	25	#include <sys/zfs_context.h>
	26	#include <sys/dmu_objset.h>
	27	#include <sys/dmu_traverse.h>
	28	#include <sys/dsl_dataset.h>
	29	#include <sys/dsl_dir.h>
	30	#include <sys/dsl_pool.h>
	31	#include <sys/dnode.h>
	32	#include <sys/spa.h>
	33	#include <sys/zio.h>
	34	#include <sys/dmu_impl.h>
428870ff BB	35	#include <sys/sa.h>
428870ff BB	36	#include <sys/sa_impl.h>
b128c09f BB	37	#include <sys/callb.h>
b128c09f BB	38
572e2857 BB	39	int zfs_pd_blks_max = 100;
	40
	41	typedef struct prefetch_data {
b128c09f BB	42	kmutex_t pd_mtx;
	43	kcondvar_t pd_cv;
	44	int pd_blks_max;
	45	int pd_blks_fetched;
	46	int pd_flags;
	47	boolean_t pd_cancel;
	48	boolean_t pd_exited;
572e2857	49	} prefetch_data_t;
b128c09f	50
572e2857	51	typedef struct traverse_data {
b128c09f BB	52	spa_t *td_spa;
	53	uint64_t td_objset;
	54	blkptr_t *td_rootbp;
	55	uint64_t td_min_txg;
	56	int td_flags;
572e2857	57	prefetch_data_t *td_pfd;
b128c09f BB	58	blkptr_cb_t *td_func;
b128c09f BB	59	void *td_arg;
572e2857	60	} traverse_data_t;
34dc7c2f	61
572e2857	62	static int traverse_dnode(traverse_data_t td, const dnode_phys_t dnp,
9babb374 BB	63	arc_buf_t *buf, uint64_t objset, uint64_t object);
9babb374 BB	64
428870ff	65	static int
34dc7c2f BB	66	traverse_zil_block(zilog_t zilog, blkptr_t bp, void *arg, uint64_t claim_txg)
34dc7c2f BB	67	{
572e2857	68	traverse_data_t *td = arg;
b128c09f	69	zbookmark_t zb;
34dc7c2f	70
b128c09f	71	if (bp->blk_birth == 0)
428870ff	72	return (0);
34dc7c2f	73
b128c09f	74	if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
428870ff BB	75	return (0);
	76
	77	SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
	78	bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
	79
	80	(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL, td->td_arg);
b128c09f	81
428870ff	82	return (0);
34dc7c2f BB	83	}
34dc7c2f BB	84
428870ff	85	static int
34dc7c2f BB	86	traverse_zil_record(zilog_t zilog, lr_t lrc, void *arg, uint64_t claim_txg)
34dc7c2f BB	87	{
572e2857	88	traverse_data_t *td = arg;
34dc7c2f BB	89
	90	if (lrc->lrc_txtype == TX_WRITE) {
	91	lr_write_t lr = (lr_write_t )lrc;
	92	blkptr_t *bp = &lr->lr_blkptr;
b128c09f	93	zbookmark_t zb;
34dc7c2f	94
b128c09f	95	if (bp->blk_birth == 0)
428870ff	96	return (0);
34dc7c2f	97
b128c09f	98	if (claim_txg == 0 \|\| bp->blk_birth < claim_txg)
428870ff	99	return (0);
b128c09f	100
572e2857 BB	101	SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
572e2857 BB	102	ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
428870ff BB	103
	104	(void) td->td_func(td->td_spa, zilog, bp, NULL, &zb, NULL,
	105	td->td_arg);
34dc7c2f	106	}
428870ff	107	return (0);
34dc7c2f BB	108	}
	109
	110	static void
572e2857	111	traverse_zil(traverse_data_t td, zil_header_t zh)
34dc7c2f	112	{
34dc7c2f BB	113	uint64_t claim_txg = zh->zh_claim_txg;
	114	zilog_t *zilog;
	115
34dc7c2f BB	116	/*
34dc7c2f BB	117	* We only want to visit blocks that have been claimed but not yet
428870ff	118	* replayed; plus, in read-only mode, blocks that are already stable.
34dc7c2f	119	*/
fb5f0bc8	120	if (claim_txg == 0 && spa_writeable(td->td_spa))
34dc7c2f BB	121	return;
34dc7c2f BB	122
b128c09f	123	zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
34dc7c2f	124
b128c09f	125	(void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
34dc7c2f BB	126	claim_txg);
	127
	128	zil_free(zilog);
	129	}
	130
	131	static int
572e2857	132	traverse_visitbp(traverse_data_t td, const dnode_phys_t dnp,
b128c09f	133	arc_buf_t pbuf, blkptr_t bp, const zbookmark_t *zb)
34dc7c2f	134	{
b128c09f	135	zbookmark_t czb;
428870ff	136	int err = 0, lasterr = 0;
b128c09f	137	arc_buf_t *buf = NULL;
572e2857	138	prefetch_data_t *pd = td->td_pfd;
428870ff	139	boolean_t hard = td->td_flags & TRAVERSE_HARD;
b128c09f BB	140
b128c09f BB	141	if (bp->blk_birth == 0) {
428870ff BB	142	err = td->td_func(td->td_spa, NULL, NULL, pbuf, zb, dnp,
428870ff BB	143	td->td_arg);
b128c09f	144	return (err);
34dc7c2f BB	145	}
34dc7c2f BB	146
b128c09f BB	147	if (bp->blk_birth <= td->td_min_txg)
b128c09f BB	148	return (0);
34dc7c2f	149
b128c09f BB	150	if (pd && !pd->pd_exited &&
	151	((pd->pd_flags & TRAVERSE_PREFETCH_DATA) \|\|
	152	BP_GET_TYPE(bp) == DMU_OT_DNODE \|\| BP_GET_LEVEL(bp) > 0)) {
	153	mutex_enter(&pd->pd_mtx);
	154	ASSERT(pd->pd_blks_fetched >= 0);
	155	while (pd->pd_blks_fetched == 0 && !pd->pd_exited)
	156	cv_wait(&pd->pd_cv, &pd->pd_mtx);
	157	pd->pd_blks_fetched--;
	158	cv_broadcast(&pd->pd_cv);
	159	mutex_exit(&pd->pd_mtx);
34dc7c2f BB	160	}
34dc7c2f BB	161
b128c09f	162	if (td->td_flags & TRAVERSE_PRE) {
428870ff BB	163	err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
428870ff BB	164	td->td_arg);
572e2857 BB	165	if (err == TRAVERSE_VISIT_NO_CHILDREN)
572e2857 BB	166	return (0);
b128c09f BB	167	if (err)
b128c09f BB	168	return (err);
34dc7c2f BB	169	}
34dc7c2f BB	170
b128c09f BB	171	if (BP_GET_LEVEL(bp) > 0) {
	172	uint32_t flags = ARC_WAIT;
	173	int i;
	174	blkptr_t *cbp;
	175	int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
	176
428870ff	177	err = dsl_read(NULL, td->td_spa, bp, pbuf,
b128c09f BB	178	arc_getbuf_func, &buf,
	179	ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
	180	if (err)
	181	return (err);
	182
	183	/* recursively visitbp() blocks below this */
	184	cbp = buf->b_data;
	185	for (i = 0; i < epb; i++, cbp++) {
	186	SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
	187	zb->zb_level - 1,
	188	zb->zb_blkid * epb + i);
	189	err = traverse_visitbp(td, dnp, buf, cbp, &czb);
428870ff BB	190	if (err) {
	191	if (!hard)
	192	break;
	193	lasterr = err;
	194	}
b128c09f BB	195	}
	196	} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
	197	uint32_t flags = ARC_WAIT;
9babb374	198	int i;
b128c09f BB	199	int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
b128c09f BB	200
428870ff	201	err = dsl_read(NULL, td->td_spa, bp, pbuf,
b128c09f BB	202	arc_getbuf_func, &buf,
	203	ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
	204	if (err)
	205	return (err);
	206
	207	/* recursively visitbp() blocks below this */
	208	dnp = buf->b_data;
428870ff	209	for (i = 0; i < epb; i++, dnp++) {
9babb374 BB	210	err = traverse_dnode(td, dnp, buf, zb->zb_objset,
9babb374 BB	211	zb->zb_blkid * epb + i);
428870ff BB	212	if (err) {
	213	if (!hard)
	214	break;
	215	lasterr = err;
	216	}
34dc7c2f	217	}
b128c09f BB	218	} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
	219	uint32_t flags = ARC_WAIT;
	220	objset_phys_t *osp;
9babb374	221	dnode_phys_t *dnp;
b128c09f	222
428870ff	223	err = dsl_read_nolock(NULL, td->td_spa, bp,
b128c09f BB	224	arc_getbuf_func, &buf,
	225	ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
	226	if (err)
	227	return (err);
	228
	229	osp = buf->b_data;
9babb374	230	dnp = &osp->os_meta_dnode;
428870ff BB	231	err = traverse_dnode(td, dnp, buf, zb->zb_objset,
	232	DMU_META_DNODE_OBJECT);
	233	if (err && hard) {
	234	lasterr = err;
	235	err = 0;
	236	}
9babb374 BB	237	if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
	238	dnp = &osp->os_userused_dnode;
	239	err = traverse_dnode(td, dnp, buf, zb->zb_objset,
	240	DMU_USERUSED_OBJECT);
	241	}
428870ff BB	242	if (err && hard) {
	243	lasterr = err;
	244	err = 0;
	245	}
9babb374 BB	246	if (err == 0 && arc_buf_size(buf) >= sizeof (objset_phys_t)) {
	247	dnp = &osp->os_groupused_dnode;
	248	err = traverse_dnode(td, dnp, buf, zb->zb_objset,
	249	DMU_GROUPUSED_OBJECT);
34dc7c2f	250	}
34dc7c2f BB	251	}
34dc7c2f BB	252
b128c09f BB	253	if (buf)
b128c09f BB	254	(void) arc_buf_remove_ref(buf, &buf);
34dc7c2f	255
428870ff BB	256	if (err == 0 && lasterr == 0 && (td->td_flags & TRAVERSE_POST)) {
	257	err = td->td_func(td->td_spa, NULL, bp, pbuf, zb, dnp,
	258	td->td_arg);
	259	}
34dc7c2f	260
428870ff	261	return (err != 0 ? err : lasterr);
34dc7c2f BB	262	}
34dc7c2f BB	263
9babb374	264	static int
572e2857	265	traverse_dnode(traverse_data_t td, const dnode_phys_t dnp,
9babb374 BB	266	arc_buf_t *buf, uint64_t objset, uint64_t object)
9babb374 BB	267	{
428870ff	268	int j, err = 0, lasterr = 0;
9babb374	269	zbookmark_t czb;
428870ff	270	boolean_t hard = (td->td_flags & TRAVERSE_HARD);
9babb374 BB	271
	272	for (j = 0; j < dnp->dn_nblkptr; j++) {
	273	SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
	274	err = traverse_visitbp(td, dnp, buf,
	275	(blkptr_t *)&dnp->dn_blkptr[j], &czb);
428870ff BB	276	if (err) {
	277	if (!hard)
	278	break;
	279	lasterr = err;
	280	}
9babb374	281	}
428870ff BB	282
	283	if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
	284	SET_BOOKMARK(&czb, objset,
	285	object, 0, DMU_SPILL_BLKID);
	286	err = traverse_visitbp(td, dnp, buf,
	287	(blkptr_t *)&dnp->dn_spill, &czb);
	288	if (err) {
	289	if (!hard)
	290	return (err);
	291	lasterr = err;
	292	}
	293	}
	294	return (err != 0 ? err : lasterr);
9babb374 BB	295	}
9babb374 BB	296
b128c09f BB	297	/* ARGSUSED */
b128c09f BB	298	static int
428870ff BB	299	traverse_prefetcher(spa_t spa, zilog_t zilog, const blkptr_t *bp,
	300	arc_buf_t pbuf, const zbookmark_t zb, const dnode_phys_t *dnp,
	301	void *arg)
34dc7c2f	302	{
572e2857	303	prefetch_data_t *pfd = arg;
b128c09f	304	uint32_t aflags = ARC_NOWAIT \| ARC_PREFETCH;
34dc7c2f	305
b128c09f BB	306	ASSERT(pfd->pd_blks_fetched >= 0);
	307	if (pfd->pd_cancel)
	308	return (EINTR);
34dc7c2f	309
b128c09f	310	if (bp == NULL \|\| !((pfd->pd_flags & TRAVERSE_PREFETCH_DATA) \|\|
428870ff BB	311	BP_GET_TYPE(bp) == DMU_OT_DNODE \|\| BP_GET_LEVEL(bp) > 0) \|\|
428870ff BB	312	BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
34dc7c2f BB	313	return (0);
34dc7c2f BB	314
b128c09f BB	315	mutex_enter(&pfd->pd_mtx);
	316	while (!pfd->pd_cancel && pfd->pd_blks_fetched >= pfd->pd_blks_max)
	317	cv_wait(&pfd->pd_cv, &pfd->pd_mtx);
	318	pfd->pd_blks_fetched++;
	319	cv_broadcast(&pfd->pd_cv);
	320	mutex_exit(&pfd->pd_mtx);
34dc7c2f	321
428870ff	322	(void) dsl_read(NULL, spa, bp, pbuf, NULL, NULL,
b128c09f BB	323	ZIO_PRIORITY_ASYNC_READ,
	324	ZIO_FLAG_CANFAIL \| ZIO_FLAG_SPECULATIVE,
	325	&aflags, zb);
34dc7c2f	326
b128c09f	327	return (0);
34dc7c2f BB	328	}
34dc7c2f BB	329
34dc7c2f	330	static void
b128c09f	331	traverse_prefetch_thread(void *arg)
34dc7c2f	332	{
572e2857 BB	333	traverse_data_t *td_main = arg;
572e2857 BB	334	traverse_data_t td = *td_main;
b128c09f	335	zbookmark_t czb;
34dc7c2f	336
b128c09f BB	337	td.td_func = traverse_prefetcher;
	338	td.td_arg = td_main->td_pfd;
	339	td.td_pfd = NULL;
34dc7c2f	340
428870ff BB	341	SET_BOOKMARK(&czb, td.td_objset,
428870ff BB	342	ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
b128c09f	343	(void) traverse_visitbp(&td, NULL, NULL, td.td_rootbp, &czb);
34dc7c2f	344
b128c09f BB	345	mutex_enter(&td_main->td_pfd->pd_mtx);
	346	td_main->td_pfd->pd_exited = B_TRUE;
	347	cv_broadcast(&td_main->td_pfd->pd_cv);
	348	mutex_exit(&td_main->td_pfd->pd_mtx);
34dc7c2f BB	349	}
34dc7c2f BB	350
b128c09f BB	351	/*
	352	* NB: dataset must not be changing on-disk (eg, is a snapshot or we are
	353	* in syncing context).
	354	*/
	355	static int
572e2857	356	traverse_impl(spa_t spa, dsl_dataset_t ds, blkptr_t *rootbp,
b128c09f	357	uint64_t txg_start, int flags, blkptr_cb_t func, void *arg)
34dc7c2f	358	{
572e2857 BB	359	traverse_data_t td;
572e2857 BB	360	prefetch_data_t pd = { 0 };
b128c09f BB	361	zbookmark_t czb;
b128c09f BB	362	int err;
34dc7c2f	363
b128c09f	364	td.td_spa = spa;
572e2857	365	td.td_objset = ds ? ds->ds_object : 0;
b128c09f BB	366	td.td_rootbp = rootbp;
	367	td.td_min_txg = txg_start;
	368	td.td_func = func;
	369	td.td_arg = arg;
	370	td.td_pfd = &pd;
	371	td.td_flags = flags;
	372
572e2857	373	pd.pd_blks_max = zfs_pd_blks_max;
b128c09f BB	374	pd.pd_flags = flags;
	375	mutex_init(&pd.pd_mtx, NULL, MUTEX_DEFAULT, NULL);
	376	cv_init(&pd.pd_cv, NULL, CV_DEFAULT, NULL);
	377
572e2857 BB	378	/* See comment on ZIL traversal in dsl_scan_visitds. */
	379	if (ds != NULL && !dsl_dataset_is_snapshot(ds)) {
	380	objset_t *os;
	381
	382	err = dmu_objset_from_ds(ds, &os);
	383	if (err)
	384	return (err);
	385
	386	traverse_zil(&td, &os->os_zil_header);
	387	}
	388
b128c09f BB	389	if (!(flags & TRAVERSE_PREFETCH) \|\|
	390	0 == taskq_dispatch(system_taskq, traverse_prefetch_thread,
	391	&td, TQ_NOQUEUE))
	392	pd.pd_exited = B_TRUE;
	393
572e2857	394	SET_BOOKMARK(&czb, td.td_objset,
428870ff	395	ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
b128c09f BB	396	err = traverse_visitbp(&td, NULL, NULL, rootbp, &czb);
	397
	398	mutex_enter(&pd.pd_mtx);
	399	pd.pd_cancel = B_TRUE;
	400	cv_broadcast(&pd.pd_cv);
	401	while (!pd.pd_exited)
	402	cv_wait(&pd.pd_cv, &pd.pd_mtx);
	403	mutex_exit(&pd.pd_mtx);
	404
	405	mutex_destroy(&pd.pd_mtx);
	406	cv_destroy(&pd.pd_cv);
34dc7c2f	407
b128c09f	408	return (err);
34dc7c2f BB	409	}
34dc7c2f BB	410
b128c09f BB	411	/*
	412	* NB: dataset must not be changing on-disk (eg, is a snapshot or we are
	413	* in syncing context).
	414	*/
	415	int
	416	traverse_dataset(dsl_dataset_t *ds, uint64_t txg_start, int flags,
	417	blkptr_cb_t func, void *arg)
34dc7c2f	418	{
572e2857	419	return (traverse_impl(ds->ds_dir->dd_pool->dp_spa, ds,
b128c09f	420	&ds->ds_phys->ds_bp, txg_start, flags, func, arg));
34dc7c2f BB	421	}
34dc7c2f BB	422
b128c09f BB	423	/*
	424	* NB: pool must not be changing on-disk (eg, from zdb or sync context).
	425	*/
	426	int
428870ff BB	427	traverse_pool(spa_t *spa, uint64_t txg_start, int flags,
428870ff BB	428	blkptr_cb_t func, void *arg)
34dc7c2f	429	{
428870ff	430	int err, lasterr = 0;
b128c09f BB	431	uint64_t obj;
	432	dsl_pool_t *dp = spa_get_dsl(spa);
	433	objset_t *mos = dp->dp_meta_objset;
428870ff	434	boolean_t hard = (flags & TRAVERSE_HARD);
b128c09f BB	435
b128c09f BB	436	/* visit the MOS */
572e2857	437	err = traverse_impl(spa, NULL, spa_get_rootblkptr(spa),
428870ff	438	txg_start, flags, func, arg);
b128c09f BB	439	if (err)
	440	return (err);
	441
	442	/* visit each dataset */
428870ff BB	443	for (obj = 1; err == 0 \|\| (err != ESRCH && hard);
428870ff BB	444	err = dmu_object_next(mos, &obj, FALSE, txg_start)) {
b128c09f BB	445	dmu_object_info_t doi;
	446
	447	err = dmu_object_info(mos, obj, &doi);
428870ff BB	448	if (err) {
	449	if (!hard)
	450	return (err);
	451	lasterr = err;
	452	continue;
	453	}
b128c09f BB	454
	455	if (doi.doi_type == DMU_OT_DSL_DATASET) {
	456	dsl_dataset_t *ds;
428870ff BB	457	uint64_t txg = txg_start;
428870ff BB	458
b128c09f BB	459	rw_enter(&dp->dp_config_rwlock, RW_READER);
	460	err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
	461	rw_exit(&dp->dp_config_rwlock);
428870ff BB	462	if (err) {
	463	if (!hard)
	464	return (err);
	465	lasterr = err;
	466	continue;
	467	}
	468	if (ds->ds_phys->ds_prev_snap_txg > txg)
	469	txg = ds->ds_phys->ds_prev_snap_txg;
	470	err = traverse_dataset(ds, txg, flags, func, arg);
b128c09f	471	dsl_dataset_rele(ds, FTAG);
428870ff BB	472	if (err) {
	473	if (!hard)
	474	return (err);
	475	lasterr = err;
	476	}
b128c09f	477	}
34dc7c2f	478	}
b128c09f BB	479	if (err == ESRCH)
b128c09f BB	480	err = 0;
428870ff	481	return (err != 0 ? err : lasterr);
34dc7c2f	482	}