fs/zfs/dmu_traverse.c

	dmu_traverse.c revision bc9014e6a81272073b9854d9f65dd59e18d18c35
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott/*
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * CDDL HEADER START
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott *
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * The contents of this file are subject to the terms of the
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * Common Development and Distribution License (the "License").
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * You may not use this file except in compliance with the License.
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott *
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * or http://www.opensolaris.org/os/licensing.
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * See the License for the specific language governing permissions
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * and limitations under the License.
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott *
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * When distributing Covered Code, include this CDDL HEADER in each
d78764efc954da87cd81023cc846a6a5af360d95Robert Wapshott * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * If applicable, add the following below this CDDL HEADER, with the
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * fields enclosed by brackets "[]" replaced with your own identifying
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * information: Portions Copyright [yyyy] [name of copyright owner]
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden *
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden * CDDL HEADER END
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden */
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts/*
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts */
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/zfs_context.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/dmu_objset.h>
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden#include <sys/dmu_traverse.h>
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden#include <sys/dsl_dataset.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/dsl_dir.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/dsl_pool.h>
de9054f2fe27b20f2c12a9fac5b1a542a838ef9fNeil Madden#include <sys/dnode.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/spa.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/zio.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/dmu_impl.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/sa.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/sa_impl.h>
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden#include <sys/callb.h>
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott#include <sys/zfeature.h>
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenint zfs_pd_blks_max = 100;
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Maddentypedef struct prefetch_data {
0fb4093da08d574d3d1b661d4425dfbac8e02aabJames Phillpotts    kmutex_t pd_mtx;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    kcondvar_t pd_cv;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    int pd_blks_max;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    int pd_blks_fetched;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    int pd_flags;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    boolean_t pd_cancel;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    boolean_t pd_exited;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott} prefetch_data_t;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshotttypedef struct traverse_data {
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    spa_t *td_spa;
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden    uint64_t td_objset;
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden    blkptr_t *td_rootbp;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    uint64_t td_min_txg;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    zbookmark_phys_t *td_resume;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    int td_flags;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    prefetch_data_t *td_pfd;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    boolean_t td_paused;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    uint64_t td_hole_birth_enabled_txg;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    blkptr_cb_t *td_func;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    void *td_arg;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden} traverse_data_t;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenstatic int traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    uint64_t objset, uint64_t object);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenstatic void prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    uint64_t objset, uint64_t object);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenstatic int
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddentraverse_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden{
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    traverse_data_t *td = arg;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    zbookmark_phys_t zb;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    if (BP_IS_HOLE(bp))
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        return (0);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(td->td_spa))
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        return (0);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    SET_BOOKMARK(&zb, td->td_objset, ZB_ZIL_OBJECT, ZB_ZIL_LEVEL,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL, td->td_arg);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    return (0);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden}
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshottstatic int
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshotttraverse_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott{
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    traverse_data_t *td = arg;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    if (lrc->lrc_txtype == TX_WRITE) {
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        lr_write_t *lr = (lr_write_t *)lrc;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        blkptr_t *bp = &lr->lr_blkptr;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        zbookmark_phys_t zb;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        if (BP_IS_HOLE(bp))
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott            return (0);
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        if (claim_txg == 0 || bp->blk_birth < claim_txg)
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden            return (0);
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        SET_BOOKMARK(&zb, td->td_objset, lr->lr_foid,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden            ZB_ZIL_LEVEL, lr->lr_offset / BP_GET_LSIZE(bp));
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        (void) td->td_func(td->td_spa, zilog, bp, &zb, NULL,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden            td->td_arg);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    }
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    return (0);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden}
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenstatic void
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddentraverse_zil(traverse_data_t *td, zil_header_t *zh)
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden{
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    uint64_t claim_txg = zh->zh_claim_txg;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    zilog_t *zilog;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    /*
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott     * We only want to visit blocks that have been claimed but not yet
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott     * replayed; plus, in read-only mode, blocks that are already stable.
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden     */
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    if (claim_txg == 0 && spa_writeable(td->td_spa))
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        return;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    zilog = zil_alloc(spa_get_dsl(td->td_spa)->dp_meta_objset, zh);
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    (void) zil_parse(zilog, traverse_zil_block, traverse_zil_record, td,
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        claim_txg);
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    zil_free(zilog);
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott}
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshotttypedef enum resume_skip {
d78764efc954da87cd81023cc846a6a5af360d95Robert Wapshott    RESUME_SKIP_ALL,
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    RESUME_SKIP_NONE,
d78764efc954da87cd81023cc846a6a5af360d95Robert Wapshott    RESUME_SKIP_CHILDREN
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott} resume_skip_t;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott/*
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington * Returns RESUME_SKIP_ALL if td indicates that we are resuming a traversal and
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * the block indicated by zb does not need to be visited at all. Returns
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * RESUME_SKIP_CHILDREN if we are resuming a post traversal and we reach the
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott * resume point. This indicates that this block should be visited but not its
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden * children (since they must have been visited in a previous traversal).
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden * Otherwise returns RESUME_SKIP_NONE.
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden */
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenstatic resume_skip_t
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Maddenresume_skip_check(traverse_data_t *td, const dnode_phys_t *dnp,
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    const zbookmark_phys_t *zb)
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden{
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume)) {
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        /*
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden         * If we already visited this bp & everything below,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden         * don't bother doing it again.
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden         */
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        if (zbookmark_is_before(dnp, zb, td->td_resume))
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden            return (RESUME_SKIP_ALL);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        /*
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden         * If we found the block we're trying to resume from, zero
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott         * the bookmark out to indicate that we have resumed.
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott         */
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        if (bcmp(zb, td->td_resume, sizeof (*zb)) == 0) {
d78764efc954da87cd81023cc846a6a5af360d95Robert Wapshott            bzero(td->td_resume, sizeof (*zb));
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott            if (td->td_flags & TRAVERSE_POST)
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott                return (RESUME_SKIP_CHILDREN);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        }
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    }
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    return (RESUME_SKIP_NONE);
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott}
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshottstatic void
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshotttraverse_prefetch_metadata(traverse_data_t *td,
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    const blkptr_t *bp, const zbookmark_phys_t *zb)
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden{
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    arc_flags_t flags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    if (!(td->td_flags & TRAVERSE_PREFETCH_METADATA))
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        return;
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott    /*
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington     * If we are in the process of resuming, don't prefetch, because
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington     * some children will not be needed (and in fact may have already
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington     * been freed).
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington     */
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington    if (td->td_resume != NULL && !ZB_IS_ZERO(td->td_resume))
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington        return;
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington    if (BP_IS_HOLE(bp) || bp->blk_birth <= td->td_min_txg)
2778561141de5c5f382c2ef5a123adfd90abf466Phill Cunnington        return;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden    if (BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE)
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden        return;
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden
997d6667b8c483bf582a231b1b24f84fbe6c8390Neil Madden    (void) arc_read(NULL, td->td_spa, bp, NULL, NULL,
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott        ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
8d75f7fba11e555d87ff0f6f2a7504681c482e0bNeil Madden}
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshott
d0da70ccbba38b773e7a7cc71bc124b06206d201Robert Wapshottstatic boolean_t
prefetch_needed(prefetch_data_t *pfd, const blkptr_t *bp)
{
    ASSERT(pfd->pd_flags & TRAVERSE_PREFETCH_DATA);
    if (BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp) ||
        BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG)
        return (B_FALSE);
    return (B_TRUE);
}

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

    switch (resume_skip_check(td, dnp, zb)) {
    case RESUME_SKIP_ALL:
        return (0);
    case RESUME_SKIP_CHILDREN:
        goto post;
    case RESUME_SKIP_NONE:
        break;
    default:
        ASSERT(0);
    }

    if (bp->blk_birth == 0) {
        /*
         * Since this block has a birth time of 0 it must be a
         * hole created before the SPA_FEATURE_HOLE_BIRTH
         * feature was enabled.  If SPA_FEATURE_HOLE_BIRTH
         * was enabled before the min_txg for this traveral we
         * know the hole must have been created before the
         * min_txg for this traveral, so we can skip it. If
         * SPA_FEATURE_HOLE_BIRTH was enabled after the min_txg
         * for this traveral we cannot tell if the hole was
         * created before or after the min_txg for this
         * traversal, so we cannot skip it.
         */
        if (td->td_hole_birth_enabled_txg < td->td_min_txg)
            return (0);
    } else if (bp->blk_birth <= td->td_min_txg) {
        return (0);
    }

    if (pd != NULL && !pd->pd_exited && prefetch_needed(pd, bp)) {
        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 (BP_IS_HOLE(bp)) {
        err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);
        if (err != 0)
            goto post;
        return (0);
    }

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

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

        err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
            ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
        if (err != 0)
            goto post;
        cbp = buf->b_data;

        for (i = 0; i < epb; i++) {
            SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
                zb->zb_level - 1,
                zb->zb_blkid * epb + i);
            traverse_prefetch_metadata(td, &cbp[i], &czb);
        }

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

        err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
            ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
        if (err != 0)
            goto post;
        dnp = buf->b_data;

        for (i = 0; i < epb; i++) {
            prefetch_dnode_metadata(td, &dnp[i], zb->zb_objset,
                zb->zb_blkid * epb + i);
        }

        /* recursively visitbp() blocks below this */
        for (i = 0; i < epb; i++) {
            err = traverse_dnode(td, &dnp[i], zb->zb_objset,
                zb->zb_blkid * epb + i);
            if (err != 0)
                break;
        }
    } else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
        arc_flags_t flags = ARC_FLAG_WAIT;
        objset_phys_t *osp;
        dnode_phys_t *dnp;

        err = arc_read(NULL, td->td_spa, bp, arc_getbuf_func, &buf,
            ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
        if (err != 0)
            goto post;

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

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

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

post:
    if (err == 0 && (td->td_flags & TRAVERSE_POST))
        err = td->td_func(td->td_spa, NULL, bp, zb, dnp, td->td_arg);

    if (hard && (err == EIO || err == ECKSUM)) {
        /*
         * Ignore this disk error as requested by the HARD flag,
         * and continue traversal.
         */
        err = 0;
    }

    /*
     * If we are stopping here, set td_resume.
     */
    if (td->td_resume != NULL && err != 0 && !td->td_paused) {
        td->td_resume->zb_objset = zb->zb_objset;
        td->td_resume->zb_object = zb->zb_object;
        td->td_resume->zb_level = 0;
        /*
         * If we have stopped on an indirect block (e.g. due to
         * i/o error), we have not visited anything below it.
         * Set the bookmark to the first level-0 block that we need
         * to visit.  This way, the resuming code does not need to
         * deal with resuming from indirect blocks.
         */
        td->td_resume->zb_blkid = zb->zb_blkid <<
            (zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT));
        td->td_paused = B_TRUE;
    }

    return (err);
}

static void
prefetch_dnode_metadata(traverse_data_t *td, const dnode_phys_t *dnp,
    uint64_t objset, uint64_t object)
{
    int j;
    zbookmark_phys_t czb;

    for (j = 0; j < dnp->dn_nblkptr; j++) {
        SET_BOOKMARK(&czb, objset, object, dnp->dn_nlevels - 1, j);
        traverse_prefetch_metadata(td, &dnp->dn_blkptr[j], &czb);
    }

    if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
        SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
        traverse_prefetch_metadata(td, &dnp->dn_spill, &czb);
    }
}

static int
traverse_dnode(traverse_data_t *td, const dnode_phys_t *dnp,
    uint64_t objset, uint64_t object)
{
    int j, err = 0;
    zbookmark_phys_t czb;

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

    if (err == 0 && dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
        SET_BOOKMARK(&czb, objset, object, 0, DMU_SPILL_BLKID);
        err = traverse_visitbp(td, dnp, &dnp->dn_spill, &czb);
    }
    return (err);
}

/* ARGSUSED */
static int
traverse_prefetcher(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
    const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
{
    prefetch_data_t *pfd = arg;
    arc_flags_t aflags = ARC_FLAG_NOWAIT | ARC_FLAG_PREFETCH;

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

    if (!prefetch_needed(pfd, bp))
        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) arc_read(NULL, spa, bp, 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_phys_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, 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, uint64_t objset, blkptr_t *rootbp,
    uint64_t txg_start, zbookmark_phys_t *resume, int flags,
    blkptr_cb_t func, void *arg)
{
    traverse_data_t td;
    prefetch_data_t pd = { 0 };
    zbookmark_phys_t czb;
    int err;

    ASSERT(ds == NULL || objset == ds->ds_object);
    ASSERT(!(flags & TRAVERSE_PRE) || !(flags & TRAVERSE_POST));

    /*
     * The data prefetching mechanism (the prefetch thread) is incompatible
     * with resuming from a bookmark.
     */
    ASSERT(resume == NULL || !(flags & TRAVERSE_PREFETCH_DATA));

    td.td_spa = spa;
    td.td_objset = objset;
    td.td_rootbp = rootbp;
    td.td_min_txg = txg_start;
    td.td_resume = resume;
    td.td_func = func;
    td.td_arg = arg;
    td.td_pfd = &pd;
    td.td_flags = flags;
    td.td_paused = B_FALSE;

    if (spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
        VERIFY(spa_feature_enabled_txg(spa,
            SPA_FEATURE_HOLE_BIRTH, &td.td_hole_birth_enabled_txg));
    } else {
        td.td_hole_birth_enabled_txg = 0;
    }

    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 && !ds->ds_is_snapshot && !BP_IS_HOLE(rootbp)) {
        arc_flags_t flags = ARC_FLAG_WAIT;
        objset_phys_t *osp;
        arc_buf_t *buf;

        err = arc_read(NULL, td.td_spa, rootbp,
            arc_getbuf_func, &buf,
            ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, NULL);
        if (err != 0)
            return (err);

        osp = buf->b_data;
        traverse_zil(&td, &osp->os_zil_header);
        (void) arc_buf_remove_ref(buf, &buf);
    }

    if (!(flags & TRAVERSE_PREFETCH_DATA) ||
        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, 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_object,
        &dsl_dataset_phys(ds)->ds_bp, txg_start, NULL, flags, func, arg));
}

int
traverse_dataset_destroyed(spa_t *spa, blkptr_t *blkptr,
    uint64_t txg_start, zbookmark_phys_t *resume, int flags,
    blkptr_cb_t func, void *arg)
{
    return (traverse_impl(spa, NULL, ZB_DESTROYED_OBJSET,
        blkptr, txg_start, resume, 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;
    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, 0, spa_get_rootblkptr(spa),
        txg_start, NULL, flags, func, arg);
    if (err != 0)
        return (err);

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

        err = dmu_object_info(mos, obj, &doi);
        if (err != 0) {
            if (hard)
                continue;
            break;
        }

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

            dsl_pool_config_enter(dp, FTAG);
            err = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
            dsl_pool_config_exit(dp, FTAG);
            if (err != 0) {
                if (hard)
                    continue;
                break;
            }
            if (dsl_dataset_phys(ds)->ds_prev_snap_txg > txg)
                txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
            err = traverse_dataset(ds, txg, flags, func, arg);
            dsl_dataset_rele(ds, FTAG);
            if (err != 0)
                break;
        }
    }
    if (err == ESRCH)
        err = 0;
    return (err);
}