go-ethereum/eth/protocols/snap/sync.go

// Copyright 2020 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.

package snap

import (
	"bytes"
	"encoding/json"
	"errors"
	"fmt"
	gomath "math"
	"math/big"
	"math/rand"
	"sort"
	"sync"
	"sync/atomic"
	"time"

	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/common/math"
	"github.com/ethereum/go-ethereum/core/rawdb"
	"github.com/ethereum/go-ethereum/core/state"
	"github.com/ethereum/go-ethereum/core/types"
	"github.com/ethereum/go-ethereum/crypto"
	"github.com/ethereum/go-ethereum/ethdb"
	"github.com/ethereum/go-ethereum/event"
	"github.com/ethereum/go-ethereum/log"
	"github.com/ethereum/go-ethereum/p2p/msgrate"
	"github.com/ethereum/go-ethereum/rlp"
	"github.com/ethereum/go-ethereum/trie"
	"github.com/ethereum/go-ethereum/trie/trienode"
	"golang.org/x/crypto/sha3"
)

const (
	// minRequestSize is the minimum number of bytes to request from a remote peer.
	// This number is used as the low cap for account and storage range requests.
	// Bytecode and trienode are limited inherently by item count (1).
	minRequestSize = 64 * 1024

	// maxRequestSize is the maximum number of bytes to request from a remote peer.
	// This number is used as the high cap for account and storage range requests.
	// Bytecode and trienode are limited more explicitly by the caps below.
	maxRequestSize = 512 * 1024

	// maxCodeRequestCount is the maximum number of bytecode blobs to request in a
	// single query. If this number is too low, we're not filling responses fully
	// and waste round trip times. If it's too high, we're capping responses and
	// waste bandwidth.
	//
	// Deployed bytecodes are currently capped at 24KB, so the minimum request
	// size should be maxRequestSize / 24K. Assuming that most contracts do not
	// come close to that, requesting 4x should be a good approximation.
	maxCodeRequestCount = maxRequestSize / (24 * 1024) * 4

	// maxTrieRequestCount is the maximum number of trie node blobs to request in
	// a single query. If this number is too low, we're not filling responses fully
	// and waste round trip times. If it's too high, we're capping responses and
	// waste bandwidth.
	maxTrieRequestCount = maxRequestSize / 512

	// trienodeHealRateMeasurementImpact is the impact a single measurement has on
	// the local node's trienode processing capacity. A value closer to 0 reacts
	// slower to sudden changes, but it is also more stable against temporary hiccups.
	trienodeHealRateMeasurementImpact = 0.005

	// minTrienodeHealThrottle is the minimum divisor for throttling trie node
	// heal requests to avoid overloading the local node and excessively expanding
	// the state trie breadth wise.
	minTrienodeHealThrottle = 1

	// maxTrienodeHealThrottle is the maximum divisor for throttling trie node
	// heal requests to avoid overloading the local node and exessively expanding
	// the state trie bedth wise.
	maxTrienodeHealThrottle = maxTrieRequestCount

	// trienodeHealThrottleIncrease is the multiplier for the throttle when the
	// rate of arriving data is higher than the rate of processing it.
	trienodeHealThrottleIncrease = 1.33

	// trienodeHealThrottleDecrease is the divisor for the throttle when the
	// rate of arriving data is lower than the rate of processing it.
	trienodeHealThrottleDecrease = 1.25
)

var (
	// accountConcurrency is the number of chunks to split the account trie into
	// to allow concurrent retrievals.
	accountConcurrency = 16

	// storageConcurrency is the number of chunks to split the a large contract
	// storage trie into to allow concurrent retrievals.
	storageConcurrency = 16
)

// ErrCancelled is returned from snap syncing if the operation was prematurely
// terminated.
var ErrCancelled = errors.New("sync cancelled")

// accountRequest tracks a pending account range request to ensure responses are
// to actual requests and to validate any security constraints.
//
// Concurrency note: account requests and responses are handled concurrently from
// the main runloop to allow Merkle proof verifications on the peer's thread and
// to drop on invalid response. The request struct must contain all the data to
// construct the response without accessing runloop internals (i.e. task). That
// is only included to allow the runloop to match a response to the task being
// synced without having yet another set of maps.
type accountRequest struct {
	peer string    // Peer to which this request is assigned
	id   uint64    // Request ID of this request
	time time.Time // Timestamp when the request was sent

	deliver chan *accountResponse // Channel to deliver successful response on
	revert  chan *accountRequest  // Channel to deliver request failure on
	cancel  chan struct{}         // Channel to track sync cancellation
	timeout *time.Timer           // Timer to track delivery timeout
	stale   chan struct{}         // Channel to signal the request was dropped

	origin common.Hash // First account requested to allow continuation checks
	limit  common.Hash // Last account requested to allow non-overlapping chunking

	task *accountTask // Task which this request is filling (only access fields through the runloop!!)
}

// accountResponse is an already Merkle-verified remote response to an account
// range request. It contains the subtrie for the requested account range and
// the database that's going to be filled with the internal nodes on commit.
type accountResponse struct {
	task *accountTask // Task which this request is filling

	hashes   []common.Hash         // Account hashes in the returned range
	accounts []*types.StateAccount // Expanded accounts in the returned range

	cont bool // Whether the account range has a continuation
}

// bytecodeRequest tracks a pending bytecode request to ensure responses are to
// actual requests and to validate any security constraints.
//
// Concurrency note: bytecode requests and responses are handled concurrently from
// the main runloop to allow Keccak256 hash verifications on the peer's thread and
// to drop on invalid response. The request struct must contain all the data to
// construct the response without accessing runloop internals (i.e. task). That
// is only included to allow the runloop to match a response to the task being
// synced without having yet another set of maps.
type bytecodeRequest struct {
	peer string    // Peer to which this request is assigned
	id   uint64    // Request ID of this request
	time time.Time // Timestamp when the request was sent

	deliver chan *bytecodeResponse // Channel to deliver successful response on
	revert  chan *bytecodeRequest  // Channel to deliver request failure on
	cancel  chan struct{}          // Channel to track sync cancellation
	timeout *time.Timer            // Timer to track delivery timeout
	stale   chan struct{}          // Channel to signal the request was dropped

	hashes []common.Hash // Bytecode hashes to validate responses
	task   *accountTask  // Task which this request is filling (only access fields through the runloop!!)
}

// bytecodeResponse is an already verified remote response to a bytecode request.
type bytecodeResponse struct {
	task *accountTask // Task which this request is filling

	hashes []common.Hash // Hashes of the bytecode to avoid double hashing
	codes  [][]byte      // Actual bytecodes to store into the database (nil = missing)
}

// storageRequest tracks a pending storage ranges request to ensure responses are
// to actual requests and to validate any security constraints.
//
// Concurrency note: storage requests and responses are handled concurrently from
// the main runloop to allow Merkle proof verifications on the peer's thread and
// to drop on invalid response. The request struct must contain all the data to
// construct the response without accessing runloop internals (i.e. tasks). That
// is only included to allow the runloop to match a response to the task being
// synced without having yet another set of maps.
type storageRequest struct {
	peer string    // Peer to which this request is assigned
	id   uint64    // Request ID of this request
	time time.Time // Timestamp when the request was sent

	deliver chan *storageResponse // Channel to deliver successful response on
	revert  chan *storageRequest  // Channel to deliver request failure on
	cancel  chan struct{}         // Channel to track sync cancellation
	timeout *time.Timer           // Timer to track delivery timeout
	stale   chan struct{}         // Channel to signal the request was dropped

	accounts []common.Hash // Account hashes to validate responses
	roots    []common.Hash // Storage roots to validate responses

	origin common.Hash // First storage slot requested to allow continuation checks
	limit  common.Hash // Last storage slot requested to allow non-overlapping chunking

	mainTask *accountTask // Task which this response belongs to (only access fields through the runloop!!)
	subTask  *storageTask // Task which this response is filling (only access fields through the runloop!!)
}

// storageResponse is an already Merkle-verified remote response to a storage
// range request. It contains the subtries for the requested storage ranges and
// the databases that's going to be filled with the internal nodes on commit.
type storageResponse struct {
	mainTask *accountTask // Task which this response belongs to
	subTask  *storageTask // Task which this response is filling

	accounts []common.Hash // Account hashes requested, may be only partially filled
	roots    []common.Hash // Storage roots requested, may be only partially filled

	hashes [][]common.Hash // Storage slot hashes in the returned range
	slots  [][][]byte      // Storage slot values in the returned range

	cont bool // Whether the last storage range has a continuation
}

// trienodeHealRequest tracks a pending state trie request to ensure responses
// are to actual requests and to validate any security constraints.
//
// Concurrency note: trie node requests and responses are handled concurrently from
// the main runloop to allow Keccak256 hash verifications on the peer's thread and
// to drop on invalid response. The request struct must contain all the data to
// construct the response without accessing runloop internals (i.e. task). That
// is only included to allow the runloop to match a response to the task being
// synced without having yet another set of maps.
type trienodeHealRequest struct {
	peer string    // Peer to which this request is assigned
	id   uint64    // Request ID of this request
	time time.Time // Timestamp when the request was sent

	deliver chan *trienodeHealResponse // Channel to deliver successful response on
	revert  chan *trienodeHealRequest  // Channel to deliver request failure on
	cancel  chan struct{}              // Channel to track sync cancellation
	timeout *time.Timer                // Timer to track delivery timeout
	stale   chan struct{}              // Channel to signal the request was dropped

	paths  []string      // Trie node paths for identifying trie node
	hashes []common.Hash // Trie node hashes to validate responses

	task *healTask // Task which this request is filling (only access fields through the runloop!!)
}

// trienodeHealResponse is an already verified remote response to a trie node request.
type trienodeHealResponse struct {
	task *healTask // Task which this request is filling

	paths  []string      // Paths of the trie nodes
	hashes []common.Hash // Hashes of the trie nodes to avoid double hashing
	nodes  [][]byte      // Actual trie nodes to store into the database (nil = missing)
}

// bytecodeHealRequest tracks a pending bytecode request to ensure responses are to
// actual requests and to validate any security constraints.
//
// Concurrency note: bytecode requests and responses are handled concurrently from
// the main runloop to allow Keccak256 hash verifications on the peer's thread and
// to drop on invalid response. The request struct must contain all the data to
// construct the response without accessing runloop internals (i.e. task). That
// is only included to allow the runloop to match a response to the task being
// synced without having yet another set of maps.
type bytecodeHealRequest struct {
	peer string    // Peer to which this request is assigned
	id   uint64    // Request ID of this request
	time time.Time // Timestamp when the request was sent

	deliver chan *bytecodeHealResponse // Channel to deliver successful response on
	revert  chan *bytecodeHealRequest  // Channel to deliver request failure on
	cancel  chan struct{}              // Channel to track sync cancellation
	timeout *time.Timer                // Timer to track delivery timeout
	stale   chan struct{}              // Channel to signal the request was dropped

	hashes []common.Hash // Bytecode hashes to validate responses
	task   *healTask     // Task which this request is filling (only access fields through the runloop!!)
}

// bytecodeHealResponse is an already verified remote response to a bytecode request.
type bytecodeHealResponse struct {
	task *healTask // Task which this request is filling

	hashes []common.Hash // Hashes of the bytecode to avoid double hashing
	codes  [][]byte      // Actual bytecodes to store into the database (nil = missing)
}

// accountTask represents the sync task for a chunk of the account snapshot.
type accountTask struct {
	// These fields get serialized to leveldb on shutdown
	Next     common.Hash                    // Next account to sync in this interval
	Last     common.Hash                    // Last account to sync in this interval
	SubTasks map[common.Hash][]*storageTask // Storage intervals needing fetching for large contracts

	// These fields are internals used during runtime
	req  *accountRequest  // Pending request to fill this task
	res  *accountResponse // Validate response filling this task
	pend int              // Number of pending subtasks for this round

	needCode  []bool // Flags whether the filling accounts need code retrieval
	needState []bool // Flags whether the filling accounts need storage retrieval
	needHeal  []bool // Flags whether the filling accounts's state was chunked and need healing

	codeTasks  map[common.Hash]struct{}    // Code hashes that need retrieval
	stateTasks map[common.Hash]common.Hash // Account hashes->roots that need full state retrieval

	genBatch ethdb.Batch     // Batch used by the node generator
	genTrie  *trie.StackTrie // Node generator from storage slots

	done bool // Flag whether the task can be removed
}

// storageTask represents the sync task for a chunk of the storage snapshot.
type storageTask struct {
	Next common.Hash // Next account to sync in this interval
	Last common.Hash // Last account to sync in this interval

	// These fields are internals used during runtime
	root common.Hash     // Storage root hash for this instance
	req  *storageRequest // Pending request to fill this task

	genBatch ethdb.Batch     // Batch used by the node generator
	genTrie  *trie.StackTrie // Node generator from storage slots

	done bool // Flag whether the task can be removed
}

// healTask represents the sync task for healing the snap-synced chunk boundaries.
type healTask struct {
	scheduler *trie.Sync // State trie sync scheduler defining the tasks

	trieTasks map[string]common.Hash   // Set of trie node tasks currently queued for retrieval, indexed by node path
	codeTasks map[common.Hash]struct{} // Set of byte code tasks currently queued for retrieval, indexed by code hash
}

// SyncProgress is a database entry to allow suspending and resuming a snapshot state
// sync. Opposed to full and fast sync, there is no way to restart a suspended
// snap sync without prior knowledge of the suspension point.
type SyncProgress struct {
	Tasks []*accountTask // The suspended account tasks (contract tasks within)

	// Status report during syncing phase
	AccountSynced  uint64             // Number of accounts downloaded
	AccountBytes   common.StorageSize // Number of account trie bytes persisted to disk
	BytecodeSynced uint64             // Number of bytecodes downloaded
	BytecodeBytes  common.StorageSize // Number of bytecode bytes downloaded
	StorageSynced  uint64             // Number of storage slots downloaded
	StorageBytes   common.StorageSize // Number of storage trie bytes persisted to disk

	// Status report during healing phase
	TrienodeHealSynced uint64             // Number of state trie nodes downloaded
	TrienodeHealBytes  common.StorageSize // Number of state trie bytes persisted to disk
	BytecodeHealSynced uint64             // Number of bytecodes downloaded
	BytecodeHealBytes  common.StorageSize // Number of bytecodes persisted to disk
}

// SyncPending is analogous to SyncProgress, but it's used to report on pending
// ephemeral sync progress that doesn't get persisted into the database.
type SyncPending struct {
	TrienodeHeal uint64 // Number of state trie nodes pending
	BytecodeHeal uint64 // Number of bytecodes pending
}

// SyncPeer abstracts out the methods required for a peer to be synced against
// with the goal of allowing the construction of mock peers without the full
// blown networking.
type SyncPeer interface {
	// ID retrieves the peer's unique identifier.
	ID() string

	// RequestAccountRange fetches a batch of accounts rooted in a specific account
	// trie, starting with the origin.
	RequestAccountRange(id uint64, root, origin, limit common.Hash, bytes uint64) error

	// RequestStorageRanges fetches a batch of storage slots belonging to one or
	// more accounts. If slots from only one account is requested, an origin marker
	// may also be used to retrieve from there.
	RequestStorageRanges(id uint64, root common.Hash, accounts []common.Hash, origin, limit []byte, bytes uint64) error

	// RequestByteCodes fetches a batch of bytecodes by hash.
	RequestByteCodes(id uint64, hashes []common.Hash, bytes uint64) error

	// RequestTrieNodes fetches a batch of account or storage trie nodes rooted in
	// a specific state trie.
	RequestTrieNodes(id uint64, root common.Hash, paths []TrieNodePathSet, bytes uint64) error

	// Log retrieves the peer's own contextual logger.
	Log() log.Logger
}

// Syncer is an Ethereum account and storage trie syncer based on snapshots and
// the  snap protocol. It's purpose is to download all the accounts and storage
// slots from remote peers and reassemble chunks of the state trie, on top of
// which a state sync can be run to fix any gaps / overlaps.
//
// Every network request has a variety of failure events:
//   - The peer disconnects after task assignment, failing to send the request
//   - The peer disconnects after sending the request, before delivering on it
//   - The peer remains connected, but does not deliver a response in time
//   - The peer delivers a stale response after a previous timeout
//   - The peer delivers a refusal to serve the requested state
type Syncer struct {
	db     ethdb.KeyValueStore // Database to store the trie nodes into (and dedup)
	scheme string              // Node scheme used in node database

	root    common.Hash    // Current state trie root being synced
	tasks   []*accountTask // Current account task set being synced
	snapped bool           // Flag to signal that snap phase is done
	healer  *healTask      // Current state healing task being executed
	update  chan struct{}  // Notification channel for possible sync progression

	peers    map[string]SyncPeer // Currently active peers to download from
	peerJoin *event.Feed         // Event feed to react to peers joining
	peerDrop *event.Feed         // Event feed to react to peers dropping
	rates    *msgrate.Trackers   // Message throughput rates for peers

	// Request tracking during syncing phase
	statelessPeers map[string]struct{} // Peers that failed to deliver state data
	accountIdlers  map[string]struct{} // Peers that aren't serving account requests
	bytecodeIdlers map[string]struct{} // Peers that aren't serving bytecode requests
	storageIdlers  map[string]struct{} // Peers that aren't serving storage requests

	accountReqs  map[uint64]*accountRequest  // Account requests currently running
	bytecodeReqs map[uint64]*bytecodeRequest // Bytecode requests currently running
	storageReqs  map[uint64]*storageRequest  // Storage requests currently running

	accountSynced  uint64             // Number of accounts downloaded
	accountBytes   common.StorageSize // Number of account trie bytes persisted to disk
	bytecodeSynced uint64             // Number of bytecodes downloaded
	bytecodeBytes  common.StorageSize // Number of bytecode bytes downloaded
	storageSynced  uint64             // Number of storage slots downloaded
	storageBytes   common.StorageSize // Number of storage trie bytes persisted to disk

	extProgress *SyncProgress // progress that can be exposed to external caller.

	// Request tracking during healing phase
	trienodeHealIdlers map[string]struct{} // Peers that aren't serving trie node requests
	bytecodeHealIdlers map[string]struct{} // Peers that aren't serving bytecode requests

	trienodeHealReqs map[uint64]*trienodeHealRequest // Trie node requests currently running
	bytecodeHealReqs map[uint64]*bytecodeHealRequest // Bytecode requests currently running

	trienodeHealRate      float64       // Average heal rate for processing trie node data
	trienodeHealPend      atomic.Uint64 // Number of trie nodes currently pending for processing
	trienodeHealThrottle  float64       // Divisor for throttling the amount of trienode heal data requested
	trienodeHealThrottled time.Time     // Timestamp the last time the throttle was updated

	trienodeHealSynced uint64             // Number of state trie nodes downloaded
	trienodeHealBytes  common.StorageSize // Number of state trie bytes persisted to disk
	trienodeHealDups   uint64             // Number of state trie nodes already processed
	trienodeHealNops   uint64             // Number of state trie nodes not requested
	bytecodeHealSynced uint64             // Number of bytecodes downloaded
	bytecodeHealBytes  common.StorageSize // Number of bytecodes persisted to disk
	bytecodeHealDups   uint64             // Number of bytecodes already processed
	bytecodeHealNops   uint64             // Number of bytecodes not requested

	stateWriter        ethdb.Batch        // Shared batch writer used for persisting raw states
	accountHealed      uint64             // Number of accounts downloaded during the healing stage
	accountHealedBytes common.StorageSize // Number of raw account bytes persisted to disk during the healing stage
	storageHealed      uint64             // Number of storage slots downloaded during the healing stage
	storageHealedBytes common.StorageSize // Number of raw storage bytes persisted to disk during the healing stage

	startTime time.Time // Time instance when snapshot sync started
	logTime   time.Time // Time instance when status was last reported

	pend sync.WaitGroup // Tracks network request goroutines for graceful shutdown
	lock sync.RWMutex   // Protects fields that can change outside of sync (peers, reqs, root)
}

// NewSyncer creates a new snapshot syncer to download the Ethereum state over the
// snap protocol.
func NewSyncer(db ethdb.KeyValueStore, scheme string) *Syncer {
	return &Syncer{
		db:     db,
		scheme: scheme,

		peers:    make(map[string]SyncPeer),
		peerJoin: new(event.Feed),
		peerDrop: new(event.Feed),
		rates:    msgrate.NewTrackers(log.New("proto", "snap")),
		update:   make(chan struct{}, 1),

		accountIdlers:  make(map[string]struct{}),
		storageIdlers:  make(map[string]struct{}),
		bytecodeIdlers: make(map[string]struct{}),

		accountReqs:  make(map[uint64]*accountRequest),
		storageReqs:  make(map[uint64]*storageRequest),
		bytecodeReqs: make(map[uint64]*bytecodeRequest),

		trienodeHealIdlers: make(map[string]struct{}),
		bytecodeHealIdlers: make(map[string]struct{}),

		trienodeHealReqs:     make(map[uint64]*trienodeHealRequest),
		bytecodeHealReqs:     make(map[uint64]*bytecodeHealRequest),
		trienodeHealThrottle: maxTrienodeHealThrottle, // Tune downward instead of insta-filling with junk
		stateWriter:          db.NewBatch(),

		extProgress: new(SyncProgress),
	}
}

// Register injects a new data source into the syncer's peerset.
func (s *Syncer) Register(peer SyncPeer) error {
	// Make sure the peer is not registered yet
	id := peer.ID()

	s.lock.Lock()
	if _, ok := s.peers[id]; ok {
		log.Error("Snap peer already registered", "id", id)

		s.lock.Unlock()
		return errors.New("already registered")
	}
	s.peers[id] = peer
	s.rates.Track(id, msgrate.NewTracker(s.rates.MeanCapacities(), s.rates.MedianRoundTrip()))

	// Mark the peer as idle, even if no sync is running
	s.accountIdlers[id] = struct{}{}
	s.storageIdlers[id] = struct{}{}
	s.bytecodeIdlers[id] = struct{}{}
	s.trienodeHealIdlers[id] = struct{}{}
	s.bytecodeHealIdlers[id] = struct{}{}
	s.lock.Unlock()

	// Notify any active syncs that a new peer can be assigned data
	s.peerJoin.Send(id)
	return nil
}

// Unregister injects a new data source into the syncer's peerset.
func (s *Syncer) Unregister(id string) error {
	// Remove all traces of the peer from the registry
	s.lock.Lock()
	if _, ok := s.peers[id]; !ok {
		log.Error("Snap peer not registered", "id", id)

		s.lock.Unlock()
		return errors.New("not registered")
	}
	delete(s.peers, id)
	s.rates.Untrack(id)

	// Remove status markers, even if no sync is running
	delete(s.statelessPeers, id)

	delete(s.accountIdlers, id)
	delete(s.storageIdlers, id)
	delete(s.bytecodeIdlers, id)
	delete(s.trienodeHealIdlers, id)
	delete(s.bytecodeHealIdlers, id)
	s.lock.Unlock()

	// Notify any active syncs that pending requests need to be reverted
	s.peerDrop.Send(id)
	return nil
}

// Sync starts (or resumes a previous) sync cycle to iterate over a state trie
// with the given root and reconstruct the nodes based on the snapshot leaves.
// Previously downloaded segments will not be redownloaded of fixed, rather any
// errors will be healed after the leaves are fully accumulated.
func (s *Syncer) Sync(root common.Hash, cancel chan struct{}) error {
	// Move the trie root from any previous value, revert stateless markers for
	// any peers and initialize the syncer if it was not yet run
	s.lock.Lock()
	s.root = root
	s.healer = &healTask{
		scheduler: state.NewStateSync(root, s.db, s.onHealState, s.scheme),
		trieTasks: make(map[string]common.Hash),
		codeTasks: make(map[common.Hash]struct{}),
	}
	s.statelessPeers = make(map[string]struct{})
	s.lock.Unlock()

	if s.startTime == (time.Time{}) {
		s.startTime = time.Now()
	}
	// Retrieve the previous sync status from LevelDB and abort if already synced
	s.loadSyncStatus()
	if len(s.tasks) == 0 && s.healer.scheduler.Pending() == 0 {
		log.Debug("Snapshot sync already completed")
		return nil
	}
	defer func() { // Persist any progress, independent of failure
		for _, task := range s.tasks {
			s.forwardAccountTask(task)
		}
		s.cleanAccountTasks()
		s.saveSyncStatus()
	}()

	log.Debug("Starting snapshot sync cycle", "root", root)

	// Flush out the last committed raw states
	defer func() {
		if s.stateWriter.ValueSize() > 0 {
			s.stateWriter.Write()
			s.stateWriter.Reset()
		}
	}()
	defer s.report(true)
	// commit any trie- and bytecode-healing data.
	defer s.commitHealer(true)

	// Whether sync completed or not, disregard any future packets
	defer func() {
		log.Debug("Terminating snapshot sync cycle", "root", root)
		s.lock.Lock()
		s.accountReqs = make(map[uint64]*accountRequest)
		s.storageReqs = make(map[uint64]*storageRequest)
		s.bytecodeReqs = make(map[uint64]*bytecodeRequest)
		s.trienodeHealReqs = make(map[uint64]*trienodeHealRequest)
		s.bytecodeHealReqs = make(map[uint64]*bytecodeHealRequest)
		s.lock.Unlock()
	}()
	// Keep scheduling sync tasks
	peerJoin := make(chan string, 16)
	peerJoinSub := s.peerJoin.Subscribe(peerJoin)
	defer peerJoinSub.Unsubscribe()

	peerDrop := make(chan string, 16)
	peerDropSub := s.peerDrop.Subscribe(peerDrop)
	defer peerDropSub.Unsubscribe()

	// Create a set of unique channels for this sync cycle. We need these to be
	// ephemeral so a data race doesn't accidentally deliver something stale on
	// a persistent channel across syncs (yup, this happened)
	var (
		accountReqFails      = make(chan *accountRequest)
		storageReqFails      = make(chan *storageRequest)
		bytecodeReqFails     = make(chan *bytecodeRequest)
		accountResps         = make(chan *accountResponse)
		storageResps         = make(chan *storageResponse)
		bytecodeResps        = make(chan *bytecodeResponse)
		trienodeHealReqFails = make(chan *trienodeHealRequest)
		bytecodeHealReqFails = make(chan *bytecodeHealRequest)
		trienodeHealResps    = make(chan *trienodeHealResponse)
		bytecodeHealResps    = make(chan *bytecodeHealResponse)
	)
	for {
		// Remove all completed tasks and terminate sync if everything's done
		s.cleanStorageTasks()
		s.cleanAccountTasks()
		if len(s.tasks) == 0 && s.healer.scheduler.Pending() == 0 {
			return nil
		}
		// Assign all the data retrieval tasks to any free peers
		s.assignAccountTasks(accountResps, accountReqFails, cancel)
		s.assignBytecodeTasks(bytecodeResps, bytecodeReqFails, cancel)
		s.assignStorageTasks(storageResps, storageReqFails, cancel)

		if len(s.tasks) == 0 {
			// Sync phase done, run heal phase
			s.assignTrienodeHealTasks(trienodeHealResps, trienodeHealReqFails, cancel)
			s.assignBytecodeHealTasks(bytecodeHealResps, bytecodeHealReqFails, cancel)
		}
		// Update sync progress
		s.lock.Lock()
		s.extProgress = &SyncProgress{
			AccountSynced:      s.accountSynced,
			AccountBytes:       s.accountBytes,
			BytecodeSynced:     s.bytecodeSynced,
			BytecodeBytes:      s.bytecodeBytes,
			StorageSynced:      s.storageSynced,
			StorageBytes:       s.storageBytes,
			TrienodeHealSynced: s.trienodeHealSynced,
			TrienodeHealBytes:  s.trienodeHealBytes,
			BytecodeHealSynced: s.bytecodeHealSynced,
			BytecodeHealBytes:  s.bytecodeHealBytes,
		}
		s.lock.Unlock()
		// Wait for something to happen
		select {
		case <-s.update:
			// Something happened (new peer, delivery, timeout), recheck tasks
		case <-peerJoin:
			// A new peer joined, try to schedule it new tasks
		case id := <-peerDrop:
			s.revertRequests(id)
		case <-cancel:
			return ErrCancelled

		case req := <-accountReqFails:
			s.revertAccountRequest(req)
		case req := <-bytecodeReqFails:
			s.revertBytecodeRequest(req)
		case req := <-storageReqFails:
			s.revertStorageRequest(req)
		case req := <-trienodeHealReqFails:
			s.revertTrienodeHealRequest(req)
		case req := <-bytecodeHealReqFails:
			s.revertBytecodeHealRequest(req)

		case res := <-accountResps:
			s.processAccountResponse(res)
		case res := <-bytecodeResps:
			s.processBytecodeResponse(res)
		case res := <-storageResps:
			s.processStorageResponse(res)
		case res := <-trienodeHealResps:
			s.processTrienodeHealResponse(res)
		case res := <-bytecodeHealResps:
			s.processBytecodeHealResponse(res)
		}
		// Report stats if something meaningful happened
		s.report(false)
	}
}

// cleanPath is used to remove the dangling nodes in the stackTrie.
func (s *Syncer) cleanPath(batch ethdb.Batch, owner common.Hash, path []byte) {
	if owner == (common.Hash{}) && rawdb.ExistsAccountTrieNode(s.db, path) {
		rawdb.DeleteAccountTrieNode(batch, path)
		deletionGauge.Inc(1)
	}
	if owner != (common.Hash{}) && rawdb.ExistsStorageTrieNode(s.db, owner, path) {
		rawdb.DeleteStorageTrieNode(batch, owner, path)
		deletionGauge.Inc(1)
	}
	lookupGauge.Inc(1)
}

// loadSyncStatus retrieves a previously aborted sync status from the database,
// or generates a fresh one if none is available.
func (s *Syncer) loadSyncStatus() {
	var progress SyncProgress

	if status := rawdb.ReadSnapshotSyncStatus(s.db); status != nil {
		if err := json.Unmarshal(status, &progress); err != nil {
			log.Error("Failed to decode snap sync status", "err", err)
		} else {
			for _, task := range progress.Tasks {
				log.Debug("Scheduled account sync task", "from", task.Next, "last", task.Last)
			}
			s.tasks = progress.Tasks
			for _, task := range s.tasks {
				task := task // closure for task.genBatch in the stacktrie writer callback

				task.genBatch = ethdb.HookedBatch{
					Batch: s.db.NewBatch(),
					OnPut: func(key []byte, value []byte) {
						s.accountBytes += common.StorageSize(len(key) + len(value))
					},
				}
				options := trie.NewStackTrieOptions()
				options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
					rawdb.WriteTrieNode(task.genBatch, common.Hash{}, path, hash, blob, s.scheme)
				})
				if s.scheme == rawdb.PathScheme {
					// Configure the dangling node cleaner and also filter out boundary nodes
					// only in the context of the path scheme. Deletion is forbidden in the
					// hash scheme, as it can disrupt state completeness.
					options = options.WithCleaner(func(path []byte) {
						s.cleanPath(task.genBatch, common.Hash{}, path)
					})
					// Skip the left boundary if it's not the first range.
					// Skip the right boundary if it's not the last range.
					options = options.WithSkipBoundary(task.Next != (common.Hash{}), task.Last != common.MaxHash, boundaryAccountNodesGauge)
				}
				task.genTrie = trie.NewStackTrie(options)
				for accountHash, subtasks := range task.SubTasks {
					for _, subtask := range subtasks {
						subtask := subtask // closure for subtask.genBatch in the stacktrie writer callback

						subtask.genBatch = ethdb.HookedBatch{
							Batch: s.db.NewBatch(),
							OnPut: func(key []byte, value []byte) {
								s.storageBytes += common.StorageSize(len(key) + len(value))
							},
						}
						owner := accountHash // local assignment for stacktrie writer closure
						options := trie.NewStackTrieOptions()
						options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
							rawdb.WriteTrieNode(subtask.genBatch, owner, path, hash, blob, s.scheme)
						})
						if s.scheme == rawdb.PathScheme {
							// Configure the dangling node cleaner and also filter out boundary nodes
							// only in the context of the path scheme. Deletion is forbidden in the
							// hash scheme, as it can disrupt state completeness.
							options = options.WithCleaner(func(path []byte) {
								s.cleanPath(subtask.genBatch, owner, path)
							})
							// Skip the left boundary if it's not the first range.
							// Skip the right boundary if it's not the last range.
							options = options.WithSkipBoundary(subtask.Next != common.Hash{}, subtask.Last != common.MaxHash, boundaryStorageNodesGauge)
						}
						subtask.genTrie = trie.NewStackTrie(options)
					}
				}
			}
			s.lock.Lock()
			defer s.lock.Unlock()

			s.snapped = len(s.tasks) == 0

			s.accountSynced = progress.AccountSynced
			s.accountBytes = progress.AccountBytes
			s.bytecodeSynced = progress.BytecodeSynced
			s.bytecodeBytes = progress.BytecodeBytes
			s.storageSynced = progress.StorageSynced
			s.storageBytes = progress.StorageBytes

			s.trienodeHealSynced = progress.TrienodeHealSynced
			s.trienodeHealBytes = progress.TrienodeHealBytes
			s.bytecodeHealSynced = progress.BytecodeHealSynced
			s.bytecodeHealBytes = progress.BytecodeHealBytes
			return
		}
	}
	// Either we've failed to decode the previous state, or there was none.
	// Start a fresh sync by chunking up the account range and scheduling
	// them for retrieval.
	s.tasks = nil
	s.accountSynced, s.accountBytes = 0, 0
	s.bytecodeSynced, s.bytecodeBytes = 0, 0
	s.storageSynced, s.storageBytes = 0, 0
	s.trienodeHealSynced, s.trienodeHealBytes = 0, 0
	s.bytecodeHealSynced, s.bytecodeHealBytes = 0, 0

	var next common.Hash
	step := new(big.Int).Sub(
		new(big.Int).Div(
			new(big.Int).Exp(common.Big2, common.Big256, nil),
			big.NewInt(int64(accountConcurrency)),
		), common.Big1,
	)
	for i := 0; i < accountConcurrency; i++ {
		last := common.BigToHash(new(big.Int).Add(next.Big(), step))
		if i == accountConcurrency-1 {
			// Make sure we don't overflow if the step is not a proper divisor
			last = common.MaxHash
		}
		batch := ethdb.HookedBatch{
			Batch: s.db.NewBatch(),
			OnPut: func(key []byte, value []byte) {
				s.accountBytes += common.StorageSize(len(key) + len(value))
			},
		}
		options := trie.NewStackTrieOptions()
		options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
			rawdb.WriteTrieNode(batch, common.Hash{}, path, hash, blob, s.scheme)
		})
		if s.scheme == rawdb.PathScheme {
			// Configure the dangling node cleaner and also filter out boundary nodes
			// only in the context of the path scheme. Deletion is forbidden in the
			// hash scheme, as it can disrupt state completeness.
			options = options.WithCleaner(func(path []byte) {
				s.cleanPath(batch, common.Hash{}, path)
			})
			// Skip the left boundary if it's not the first range.
			// Skip the right boundary if it's not the last range.
			options = options.WithSkipBoundary(next != common.Hash{}, last != common.MaxHash, boundaryAccountNodesGauge)
		}
		s.tasks = append(s.tasks, &accountTask{
			Next:     next,
			Last:     last,
			SubTasks: make(map[common.Hash][]*storageTask),
			genBatch: batch,
			genTrie:  trie.NewStackTrie(options),
		})
		log.Debug("Created account sync task", "from", next, "last", last)
		next = common.BigToHash(new(big.Int).Add(last.Big(), common.Big1))
	}
}

// saveSyncStatus marshals the remaining sync tasks into leveldb.
func (s *Syncer) saveSyncStatus() {
	// Serialize any partial progress to disk before spinning down
	for _, task := range s.tasks {
		if err := task.genBatch.Write(); err != nil {
			log.Error("Failed to persist account slots", "err", err)
		}
		for _, subtasks := range task.SubTasks {
			for _, subtask := range subtasks {
				if err := subtask.genBatch.Write(); err != nil {
					log.Error("Failed to persist storage slots", "err", err)
				}
			}
		}
	}
	// Store the actual progress markers
	progress := &SyncProgress{
		Tasks:              s.tasks,
		AccountSynced:      s.accountSynced,
		AccountBytes:       s.accountBytes,
		BytecodeSynced:     s.bytecodeSynced,
		BytecodeBytes:      s.bytecodeBytes,
		StorageSynced:      s.storageSynced,
		StorageBytes:       s.storageBytes,
		TrienodeHealSynced: s.trienodeHealSynced,
		TrienodeHealBytes:  s.trienodeHealBytes,
		BytecodeHealSynced: s.bytecodeHealSynced,
		BytecodeHealBytes:  s.bytecodeHealBytes,
	}
	status, err := json.Marshal(progress)
	if err != nil {
		panic(err) // This can only fail during implementation
	}
	rawdb.WriteSnapshotSyncStatus(s.db, status)
}

// Progress returns the snap sync status statistics.
func (s *Syncer) Progress() (*SyncProgress, *SyncPending) {
	s.lock.Lock()
	defer s.lock.Unlock()
	pending := new(SyncPending)
	if s.healer != nil {
		pending.TrienodeHeal = uint64(len(s.healer.trieTasks))
		pending.BytecodeHeal = uint64(len(s.healer.codeTasks))
	}
	return s.extProgress, pending
}

// cleanAccountTasks removes account range retrieval tasks that have already been
// completed.
func (s *Syncer) cleanAccountTasks() {
	// If the sync was already done before, don't even bother
	if len(s.tasks) == 0 {
		return
	}
	// Sync wasn't finished previously, check for any task that can be finalized
	for i := 0; i < len(s.tasks); i++ {
		if s.tasks[i].done {
			s.tasks = append(s.tasks[:i], s.tasks[i+1:]...)
			i--
		}
	}
	// If everything was just finalized just, generate the account trie and start heal
	if len(s.tasks) == 0 {
		s.lock.Lock()
		s.snapped = true
		s.lock.Unlock()

		// Push the final sync report
		s.reportSyncProgress(true)
	}
}

// cleanStorageTasks iterates over all the account tasks and storage sub-tasks
// within, cleaning any that have been completed.
func (s *Syncer) cleanStorageTasks() {
	for _, task := range s.tasks {
		for account, subtasks := range task.SubTasks {
			// Remove storage range retrieval tasks that completed
			for j := 0; j < len(subtasks); j++ {
				if subtasks[j].done {
					subtasks = append(subtasks[:j], subtasks[j+1:]...)
					j--
				}
			}
			if len(subtasks) > 0 {
				task.SubTasks[account] = subtasks
				continue
			}
			// If all storage chunks are done, mark the account as done too
			for j, hash := range task.res.hashes {
				if hash == account {
					task.needState[j] = false
				}
			}
			delete(task.SubTasks, account)
			task.pend--

			// If this was the last pending task, forward the account task
			if task.pend == 0 {
				s.forwardAccountTask(task)
			}
		}
	}
}

// assignAccountTasks attempts to match idle peers to pending account range
// retrievals.
func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *accountRequest, cancel chan struct{}) {
	s.lock.Lock()
	defer s.lock.Unlock()

	// Sort the peers by download capacity to use faster ones if many available
	idlers := &capacitySort{
		ids:  make([]string, 0, len(s.accountIdlers)),
		caps: make([]int, 0, len(s.accountIdlers)),
	}
	targetTTL := s.rates.TargetTimeout()
	for id := range s.accountIdlers {
		if _, ok := s.statelessPeers[id]; ok {
			continue
		}
		idlers.ids = append(idlers.ids, id)
		idlers.caps = append(idlers.caps, s.rates.Capacity(id, AccountRangeMsg, targetTTL))
	}
	if len(idlers.ids) == 0 {
		return
	}
	sort.Sort(sort.Reverse(idlers))

	// Iterate over all the tasks and try to find a pending one
	for _, task := range s.tasks {
		// Skip any tasks already filling
		if task.req != nil || task.res != nil {
			continue
		}
		// Task pending retrieval, try to find an idle peer. If no such peer
		// exists, we probably assigned tasks for all (or they are stateless).
		// Abort the entire assignment mechanism.
		if len(idlers.ids) == 0 {
			return
		}
		var (
			idle = idlers.ids[0]
			peer = s.peers[idle]
			cap  = idlers.caps[0]
		)
		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]

		// Matched a pending task to an idle peer, allocate a unique request id
		var reqid uint64
		for {
			reqid = uint64(rand.Int63())
			if reqid == 0 {
				continue
			}
			if _, ok := s.accountReqs[reqid]; ok {
				continue
			}
			break
		}
		// Generate the network query and send it to the peer
		req := &accountRequest{
			peer:    idle,
			id:      reqid,
			time:    time.Now(),
			deliver: success,
			revert:  fail,
			cancel:  cancel,
			stale:   make(chan struct{}),
			origin:  task.Next,
			limit:   task.Last,
			task:    task,
		}
		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
			peer.Log().Debug("Account range request timed out", "reqid", reqid)
			s.rates.Update(idle, AccountRangeMsg, 0, 0)
			s.scheduleRevertAccountRequest(req)
		})
		s.accountReqs[reqid] = req
		delete(s.accountIdlers, idle)

		s.pend.Add(1)
		go func(root common.Hash) {
			defer s.pend.Done()

			// Attempt to send the remote request and revert if it fails
			if cap > maxRequestSize {
				cap = maxRequestSize
			}
			if cap < minRequestSize { // Don't bother with peers below a bare minimum performance
				cap = minRequestSize
			}
			if err := peer.RequestAccountRange(reqid, root, req.origin, req.limit, uint64(cap)); err != nil {
				peer.Log().Debug("Failed to request account range", "err", err)
				s.scheduleRevertAccountRequest(req)
			}
		}(s.root)

		// Inject the request into the task to block further assignments
		task.req = req
	}
}

// assignBytecodeTasks attempts to match idle peers to pending code retrievals.
func (s *Syncer) assignBytecodeTasks(success chan *bytecodeResponse, fail chan *bytecodeRequest, cancel chan struct{}) {
	s.lock.Lock()
	defer s.lock.Unlock()

	// Sort the peers by download capacity to use faster ones if many available
	idlers := &capacitySort{
		ids:  make([]string, 0, len(s.bytecodeIdlers)),
		caps: make([]int, 0, len(s.bytecodeIdlers)),
	}
	targetTTL := s.rates.TargetTimeout()
	for id := range s.bytecodeIdlers {
		if _, ok := s.statelessPeers[id]; ok {
			continue
		}
		idlers.ids = append(idlers.ids, id)
		idlers.caps = append(idlers.caps, s.rates.Capacity(id, ByteCodesMsg, targetTTL))
	}
	if len(idlers.ids) == 0 {
		return
	}
	sort.Sort(sort.Reverse(idlers))

	// Iterate over all the tasks and try to find a pending one
	for _, task := range s.tasks {
		// Skip any tasks not in the bytecode retrieval phase
		if task.res == nil {
			continue
		}
		// Skip tasks that are already retrieving (or done with) all codes
		if len(task.codeTasks) == 0 {
			continue
		}
		// Task pending retrieval, try to find an idle peer. If no such peer
		// exists, we probably assigned tasks for all (or they are stateless).
		// Abort the entire assignment mechanism.
		if len(idlers.ids) == 0 {
			return
		}
		var (
			idle = idlers.ids[0]
			peer = s.peers[idle]
			cap  = idlers.caps[0]
		)
		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]

		// Matched a pending task to an idle peer, allocate a unique request id
		var reqid uint64
		for {
			reqid = uint64(rand.Int63())
			if reqid == 0 {
				continue
			}
			if _, ok := s.bytecodeReqs[reqid]; ok {
				continue
			}
			break
		}
		// Generate the network query and send it to the peer
		if cap > maxCodeRequestCount {
			cap = maxCodeRequestCount
		}
		hashes := make([]common.Hash, 0, cap)
		for hash := range task.codeTasks {
			delete(task.codeTasks, hash)
			hashes = append(hashes, hash)
			if len(hashes) >= cap {
				break
			}
		}
		req := &bytecodeRequest{
			peer:    idle,
			id:      reqid,
			time:    time.Now(),
			deliver: success,
			revert:  fail,
			cancel:  cancel,
			stale:   make(chan struct{}),
			hashes:  hashes,
			task:    task,
		}
		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
			peer.Log().Debug("Bytecode request timed out", "reqid", reqid)
			s.rates.Update(idle, ByteCodesMsg, 0, 0)
			s.scheduleRevertBytecodeRequest(req)
		})
		s.bytecodeReqs[reqid] = req
		delete(s.bytecodeIdlers, idle)

		s.pend.Add(1)
		go func() {
			defer s.pend.Done()

			// Attempt to send the remote request and revert if it fails
			if err := peer.RequestByteCodes(reqid, hashes, maxRequestSize); err != nil {
				log.Debug("Failed to request bytecodes", "err", err)
				s.scheduleRevertBytecodeRequest(req)
			}
		}()
	}
}

// assignStorageTasks attempts to match idle peers to pending storage range
// retrievals.
func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *storageRequest, cancel chan struct{}) {
	s.lock.Lock()
	defer s.lock.Unlock()

	// Sort the peers by download capacity to use faster ones if many available
	idlers := &capacitySort{
		ids:  make([]string, 0, len(s.storageIdlers)),
		caps: make([]int, 0, len(s.storageIdlers)),
	}
	targetTTL := s.rates.TargetTimeout()
	for id := range s.storageIdlers {
		if _, ok := s.statelessPeers[id]; ok {
			continue
		}
		idlers.ids = append(idlers.ids, id)
		idlers.caps = append(idlers.caps, s.rates.Capacity(id, StorageRangesMsg, targetTTL))
	}
	if len(idlers.ids) == 0 {
		return
	}
	sort.Sort(sort.Reverse(idlers))

	// Iterate over all the tasks and try to find a pending one
	for _, task := range s.tasks {
		// Skip any tasks not in the storage retrieval phase
		if task.res == nil {
			continue
		}
		// Skip tasks that are already retrieving (or done with) all small states
		if len(task.SubTasks) == 0 && len(task.stateTasks) == 0 {
			continue
		}
		// Task pending retrieval, try to find an idle peer. If no such peer
		// exists, we probably assigned tasks for all (or they are stateless).
		// Abort the entire assignment mechanism.
		if len(idlers.ids) == 0 {
			return
		}
		var (
			idle = idlers.ids[0]
			peer = s.peers[idle]
			cap  = idlers.caps[0]
		)
		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]

		// Matched a pending task to an idle peer, allocate a unique request id
		var reqid uint64
		for {
			reqid = uint64(rand.Int63())
			if reqid == 0 {
				continue
			}
			if _, ok := s.storageReqs[reqid]; ok {
				continue
			}
			break
		}
		// Generate the network query and send it to the peer. If there are
		// large contract tasks pending, complete those before diving into
		// even more new contracts.
		if cap > maxRequestSize {
			cap = maxRequestSize
		}
		if cap < minRequestSize { // Don't bother with peers below a bare minimum performance
			cap = minRequestSize
		}
		storageSets := cap / 1024

		var (
			accounts = make([]common.Hash, 0, storageSets)
			roots    = make([]common.Hash, 0, storageSets)
			subtask  *storageTask
		)
		for account, subtasks := range task.SubTasks {
			for _, st := range subtasks {
				// Skip any subtasks already filling
				if st.req != nil {
					continue
				}
				// Found an incomplete storage chunk, schedule it
				accounts = append(accounts, account)
				roots = append(roots, st.root)
				subtask = st
				break // Large contract chunks are downloaded individually
			}
			if subtask != nil {
				break // Large contract chunks are downloaded individually
			}
		}
		if subtask == nil {
			// No large contract required retrieval, but small ones available
			for account, root := range task.stateTasks {
				delete(task.stateTasks, account)

				accounts = append(accounts, account)
				roots = append(roots, root)

				if len(accounts) >= storageSets {
					break
				}
			}
		}
		// If nothing was found, it means this task is actually already fully
		// retrieving, but large contracts are hard to detect. Skip to the next.
		if len(accounts) == 0 {
			continue
		}
		req := &storageRequest{
			peer:     idle,
			id:       reqid,
			time:     time.Now(),
			deliver:  success,
			revert:   fail,
			cancel:   cancel,
			stale:    make(chan struct{}),
			accounts: accounts,
			roots:    roots,
			mainTask: task,
			subTask:  subtask,
		}
		if subtask != nil {
			req.origin = subtask.Next
			req.limit = subtask.Last
		}
		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
			peer.Log().Debug("Storage request timed out", "reqid", reqid)
			s.rates.Update(idle, StorageRangesMsg, 0, 0)
			s.scheduleRevertStorageRequest(req)
		})
		s.storageReqs[reqid] = req
		delete(s.storageIdlers, idle)

		s.pend.Add(1)
		go func(root common.Hash) {
			defer s.pend.Done()

			// Attempt to send the remote request and revert if it fails
			var origin, limit []byte
			if subtask != nil {
				origin, limit = req.origin[:], req.limit[:]
			}
			if err := peer.RequestStorageRanges(reqid, root, accounts, origin, limit, uint64(cap)); err != nil {
				log.Debug("Failed to request storage", "err", err)
				s.scheduleRevertStorageRequest(req)
			}
		}(s.root)

		// Inject the request into the subtask to block further assignments
		if subtask != nil {
			subtask.req = req
		}
	}
}

// assignTrienodeHealTasks attempts to match idle peers to trie node requests to
// heal any trie errors caused by the snap sync's chunked retrieval model.
func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fail chan *trienodeHealRequest, cancel chan struct{}) {
	s.lock.Lock()
	defer s.lock.Unlock()

	// Sort the peers by download capacity to use faster ones if many available
	idlers := &capacitySort{
		ids:  make([]string, 0, len(s.trienodeHealIdlers)),
		caps: make([]int, 0, len(s.trienodeHealIdlers)),
	}
	targetTTL := s.rates.TargetTimeout()
	for id := range s.trienodeHealIdlers {
		if _, ok := s.statelessPeers[id]; ok {
			continue
		}
		idlers.ids = append(idlers.ids, id)
		idlers.caps = append(idlers.caps, s.rates.Capacity(id, TrieNodesMsg, targetTTL))
	}
	if len(idlers.ids) == 0 {
		return
	}
	sort.Sort(sort.Reverse(idlers))

	// Iterate over pending tasks and try to find a peer to retrieve with
	for len(s.healer.trieTasks) > 0 || s.healer.scheduler.Pending() > 0 {
		// If there are not enough trie tasks queued to fully assign, fill the
		// queue from the state sync scheduler. The trie synced schedules these
		// together with bytecodes, so we need to queue them combined.
		var (
			have = len(s.healer.trieTasks) + len(s.healer.codeTasks)
			want = maxTrieRequestCount + maxCodeRequestCount
		)
		if have < want {
			paths, hashes, codes := s.healer.scheduler.Missing(want - have)
			for i, path := range paths {
				s.healer.trieTasks[path] = hashes[i]
			}
			for _, hash := range codes {
				s.healer.codeTasks[hash] = struct{}{}
			}
		}
		// If all the heal tasks are bytecodes or already downloading, bail
		if len(s.healer.trieTasks) == 0 {
			return
		}
		// Task pending retrieval, try to find an idle peer. If no such peer
		// exists, we probably assigned tasks for all (or they are stateless).
		// Abort the entire assignment mechanism.
		if len(idlers.ids) == 0 {
			return
		}
		var (
			idle = idlers.ids[0]
			peer = s.peers[idle]
			cap  = idlers.caps[0]
		)
		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]

		// Matched a pending task to an idle peer, allocate a unique request id
		var reqid uint64
		for {
			reqid = uint64(rand.Int63())
			if reqid == 0 {
				continue
			}
			if _, ok := s.trienodeHealReqs[reqid]; ok {
				continue
			}
			break
		}
		// Generate the network query and send it to the peer
		if cap > maxTrieRequestCount {
			cap = maxTrieRequestCount
		}
		cap = int(float64(cap) / s.trienodeHealThrottle)
		if cap <= 0 {
			cap = 1
		}
		var (
			hashes   = make([]common.Hash, 0, cap)
			paths    = make([]string, 0, cap)
			pathsets = make([]TrieNodePathSet, 0, cap)
		)
		for path, hash := range s.healer.trieTasks {
			delete(s.healer.trieTasks, path)

			paths = append(paths, path)
			hashes = append(hashes, hash)
			if len(paths) >= cap {
				break
			}
		}
		// Group requests by account hash
		paths, hashes, _, pathsets = sortByAccountPath(paths, hashes)
		req := &trienodeHealRequest{
			peer:    idle,
			id:      reqid,
			time:    time.Now(),
			deliver: success,
			revert:  fail,
			cancel:  cancel,
			stale:   make(chan struct{}),
			paths:   paths,
			hashes:  hashes,
			task:    s.healer,
		}
		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
			peer.Log().Debug("Trienode heal request timed out", "reqid", reqid)
			s.rates.Update(idle, TrieNodesMsg, 0, 0)
			s.scheduleRevertTrienodeHealRequest(req)
		})
		s.trienodeHealReqs[reqid] = req
		delete(s.trienodeHealIdlers, idle)

		s.pend.Add(1)
		go func(root common.Hash) {
			defer s.pend.Done()

			// Attempt to send the remote request and revert if it fails
			if err := peer.RequestTrieNodes(reqid, root, pathsets, maxRequestSize); err != nil {
				log.Debug("Failed to request trienode healers", "err", err)
				s.scheduleRevertTrienodeHealRequest(req)
			}
		}(s.root)
	}
}

// assignBytecodeHealTasks attempts to match idle peers to bytecode requests to
// heal any trie errors caused by the snap sync's chunked retrieval model.
func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fail chan *bytecodeHealRequest, cancel chan struct{}) {
	s.lock.Lock()
	defer s.lock.Unlock()

	// Sort the peers by download capacity to use faster ones if many available
	idlers := &capacitySort{
		ids:  make([]string, 0, len(s.bytecodeHealIdlers)),
		caps: make([]int, 0, len(s.bytecodeHealIdlers)),
	}
	targetTTL := s.rates.TargetTimeout()
	for id := range s.bytecodeHealIdlers {
		if _, ok := s.statelessPeers[id]; ok {
			continue
		}
		idlers.ids = append(idlers.ids, id)
		idlers.caps = append(idlers.caps, s.rates.Capacity(id, ByteCodesMsg, targetTTL))
	}
	if len(idlers.ids) == 0 {
		return
	}
	sort.Sort(sort.Reverse(idlers))

	// Iterate over pending tasks and try to find a peer to retrieve with
	for len(s.healer.codeTasks) > 0 || s.healer.scheduler.Pending() > 0 {
		// If there are not enough trie tasks queued to fully assign, fill the
		// queue from the state sync scheduler. The trie synced schedules these
		// together with trie nodes, so we need to queue them combined.
		var (
			have = len(s.healer.trieTasks) + len(s.healer.codeTasks)
			want = maxTrieRequestCount + maxCodeRequestCount
		)
		if have < want {
			paths, hashes, codes := s.healer.scheduler.Missing(want - have)
			for i, path := range paths {
				s.healer.trieTasks[path] = hashes[i]
			}
			for _, hash := range codes {
				s.healer.codeTasks[hash] = struct{}{}
			}
		}
		// If all the heal tasks are trienodes or already downloading, bail
		if len(s.healer.codeTasks) == 0 {
			return
		}
		// Task pending retrieval, try to find an idle peer. If no such peer
		// exists, we probably assigned tasks for all (or they are stateless).
		// Abort the entire assignment mechanism.
		if len(idlers.ids) == 0 {
			return
		}
		var (
			idle = idlers.ids[0]
			peer = s.peers[idle]
			cap  = idlers.caps[0]
		)
		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]

		// Matched a pending task to an idle peer, allocate a unique request id
		var reqid uint64
		for {
			reqid = uint64(rand.Int63())
			if reqid == 0 {
				continue
			}
			if _, ok := s.bytecodeHealReqs[reqid]; ok {
				continue
			}
			break
		}
		// Generate the network query and send it to the peer
		if cap > maxCodeRequestCount {
			cap = maxCodeRequestCount
		}
		hashes := make([]common.Hash, 0, cap)
		for hash := range s.healer.codeTasks {
			delete(s.healer.codeTasks, hash)

			hashes = append(hashes, hash)
			if len(hashes) >= cap {
				break
			}
		}
		req := &bytecodeHealRequest{
			peer:    idle,
			id:      reqid,
			time:    time.Now(),
			deliver: success,
			revert:  fail,
			cancel:  cancel,
			stale:   make(chan struct{}),
			hashes:  hashes,
			task:    s.healer,
		}
		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
			peer.Log().Debug("Bytecode heal request timed out", "reqid", reqid)
			s.rates.Update(idle, ByteCodesMsg, 0, 0)
			s.scheduleRevertBytecodeHealRequest(req)
		})
		s.bytecodeHealReqs[reqid] = req
		delete(s.bytecodeHealIdlers, idle)

		s.pend.Add(1)
		go func() {
			defer s.pend.Done()

			// Attempt to send the remote request and revert if it fails
			if err := peer.RequestByteCodes(reqid, hashes, maxRequestSize); err != nil {
				log.Debug("Failed to request bytecode healers", "err", err)
				s.scheduleRevertBytecodeHealRequest(req)
			}
		}()
	}
}

// revertRequests locates all the currently pending requests from a particular
// peer and reverts them, rescheduling for others to fulfill.
func (s *Syncer) revertRequests(peer string) {
	// Gather the requests first, revertals need the lock too
	s.lock.Lock()
	var accountReqs []*accountRequest
	for _, req := range s.accountReqs {
		if req.peer == peer {
			accountReqs = append(accountReqs, req)
		}
	}
	var bytecodeReqs []*bytecodeRequest
	for _, req := range s.bytecodeReqs {
		if req.peer == peer {
			bytecodeReqs = append(bytecodeReqs, req)
		}
	}
	var storageReqs []*storageRequest
	for _, req := range s.storageReqs {
		if req.peer == peer {
			storageReqs = append(storageReqs, req)
		}
	}
	var trienodeHealReqs []*trienodeHealRequest
	for _, req := range s.trienodeHealReqs {
		if req.peer == peer {
			trienodeHealReqs = append(trienodeHealReqs, req)
		}
	}
	var bytecodeHealReqs []*bytecodeHealRequest
	for _, req := range s.bytecodeHealReqs {
		if req.peer == peer {
			bytecodeHealReqs = append(bytecodeHealReqs, req)
		}
	}
	s.lock.Unlock()

	// Revert all the requests matching the peer
	for _, req := range accountReqs {
		s.revertAccountRequest(req)
	}
	for _, req := range bytecodeReqs {
		s.revertBytecodeRequest(req)
	}
	for _, req := range storageReqs {
		s.revertStorageRequest(req)
	}
	for _, req := range trienodeHealReqs {
		s.revertTrienodeHealRequest(req)
	}
	for _, req := range bytecodeHealReqs {
		s.revertBytecodeHealRequest(req)
	}
}

// scheduleRevertAccountRequest asks the event loop to clean up an account range
// request and return all failed retrieval tasks to the scheduler for reassignment.
func (s *Syncer) scheduleRevertAccountRequest(req *accountRequest) {
	select {
	case req.revert <- req:
		// Sync event loop notified
	case <-req.cancel:
		// Sync cycle got cancelled
	case <-req.stale:
		// Request already reverted
	}
}

// revertAccountRequest cleans up an account range request and returns all failed
// retrieval tasks to the scheduler for reassignment.
//
// Note, this needs to run on the event runloop thread to reschedule to idle peers.
// On peer threads, use scheduleRevertAccountRequest.
func (s *Syncer) revertAccountRequest(req *accountRequest) {
	log.Debug("Reverting account request", "peer", req.peer, "reqid", req.id)
	select {
	case <-req.stale:
		log.Trace("Account request already reverted", "peer", req.peer, "reqid", req.id)
		return
	default:
	}
	close(req.stale)

	// Remove the request from the tracked set
	s.lock.Lock()
	delete(s.accountReqs, req.id)
	s.lock.Unlock()

	// If there's a timeout timer still running, abort it and mark the account
	// task as not-pending, ready for rescheduling
	req.timeout.Stop()
	if req.task.req == req {
		req.task.req = nil
	}
}

// scheduleRevertBytecodeRequest asks the event loop to clean up a bytecode request
// and return all failed retrieval tasks to the scheduler for reassignment.
func (s *Syncer) scheduleRevertBytecodeRequest(req *bytecodeRequest) {
	select {
	case req.revert <- req:
		// Sync event loop notified
	case <-req.cancel:
		// Sync cycle got cancelled
	case <-req.stale:
		// Request already reverted
	}
}

// revertBytecodeRequest cleans up a bytecode request and returns all failed
// retrieval tasks to the scheduler for reassignment.
//
// Note, this needs to run on the event runloop thread to reschedule to idle peers.
// On peer threads, use scheduleRevertBytecodeRequest.
func (s *Syncer) revertBytecodeRequest(req *bytecodeRequest) {
	log.Debug("Reverting bytecode request", "peer", req.peer)
	select {
	case <-req.stale:
		log.Trace("Bytecode request already reverted", "peer", req.peer, "reqid", req.id)
		return
	default:
	}
	close(req.stale)

	// Remove the request from the tracked set
	s.lock.Lock()
	delete(s.bytecodeReqs, req.id)
	s.lock.Unlock()

	// If there's a timeout timer still running, abort it and mark the code
	// retrievals as not-pending, ready for rescheduling
	req.timeout.Stop()
	for _, hash := range req.hashes {
		req.task.codeTasks[hash] = struct{}{}
	}
}

// scheduleRevertStorageRequest asks the event loop to clean up a storage range
// request and return all failed retrieval tasks to the scheduler for reassignment.
func (s *Syncer) scheduleRevertStorageRequest(req *storageRequest) {
	select {
	case req.revert <- req:
		// Sync event loop notified
	case <-req.cancel:
		// Sync cycle got cancelled
	case <-req.stale:
		// Request already reverted
	}
}

// revertStorageRequest cleans up a storage range request and returns all failed
// retrieval tasks to the scheduler for reassignment.
//
// Note, this needs to run on the event runloop thread to reschedule to idle peers.
// On peer threads, use scheduleRevertStorageRequest.
func (s *Syncer) revertStorageRequest(req *storageRequest) {
	log.Debug("Reverting storage request", "peer", req.peer)
	select {
	case <-req.stale:
		log.Trace("Storage request already reverted", "peer", req.peer, "reqid", req.id)
		return
	default:
	}
	close(req.stale)

	// Remove the request from the tracked set
	s.lock.Lock()
	delete(s.storageReqs, req.id)
	s.lock.Unlock()

	// If there's a timeout timer still running, abort it and mark the storage
	// task as not-pending, ready for rescheduling
	req.timeout.Stop()
	if req.subTask != nil {
		req.subTask.req = nil
	} else {
		for i, account := range req.accounts {
			req.mainTask.stateTasks[account] = req.roots[i]
		}
	}
}

// scheduleRevertTrienodeHealRequest asks the event loop to clean up a trienode heal
// request and return all failed retrieval tasks to the scheduler for reassignment.
func (s *Syncer) scheduleRevertTrienodeHealRequest(req *trienodeHealRequest) {
	select {
	case req.revert <- req:
		// Sync event loop notified
	case <-req.cancel:
		// Sync cycle got cancelled
	case <-req.stale:
		// Request already reverted
	}
}

// revertTrienodeHealRequest cleans up a trienode heal request and returns all
// failed retrieval tasks to the scheduler for reassignment.
//
// Note, this needs to run on the event runloop thread to reschedule to idle peers.
// On peer threads, use scheduleRevertTrienodeHealRequest.
func (s *Syncer) revertTrienodeHealRequest(req *trienodeHealRequest) {
	log.Debug("Reverting trienode heal request", "peer", req.peer)
	select {
	case <-req.stale:
		log.Trace("Trienode heal request already reverted", "peer", req.peer, "reqid", req.id)
		return
	default:
	}
	close(req.stale)

	// Remove the request from the tracked set
	s.lock.Lock()
	delete(s.trienodeHealReqs, req.id)
	s.lock.Unlock()

	// If there's a timeout timer still running, abort it and mark the trie node
	// retrievals as not-pending, ready for rescheduling
	req.timeout.Stop()
	for i, path := range req.paths {
		req.task.trieTasks[path] = req.hashes[i]
	}
}

// scheduleRevertBytecodeHealRequest asks the event loop to clean up a bytecode heal
// request and return all failed retrieval tasks to the scheduler for reassignment.
func (s *Syncer) scheduleRevertBytecodeHealRequest(req *bytecodeHealRequest) {
	select {
	case req.revert <- req:
		// Sync event loop notified
	case <-req.cancel:
		// Sync cycle got cancelled
	case <-req.stale:
		// Request already reverted
	}
}

// revertBytecodeHealRequest cleans up a bytecode heal request and returns all
// failed retrieval tasks to the scheduler for reassignment.
//
// Note, this needs to run on the event runloop thread to reschedule to idle peers.
// On peer threads, use scheduleRevertBytecodeHealRequest.
func (s *Syncer) revertBytecodeHealRequest(req *bytecodeHealRequest) {
	log.Debug("Reverting bytecode heal request", "peer", req.peer)
	select {
	case <-req.stale:
		log.Trace("Bytecode heal request already reverted", "peer", req.peer, "reqid", req.id)
		return
	default:
	}
	close(req.stale)

	// Remove the request from the tracked set
	s.lock.Lock()
	delete(s.bytecodeHealReqs, req.id)
	s.lock.Unlock()

	// If there's a timeout timer still running, abort it and mark the code
	// retrievals as not-pending, ready for rescheduling
	req.timeout.Stop()
	for _, hash := range req.hashes {
		req.task.codeTasks[hash] = struct{}{}
	}
}

// processAccountResponse integrates an already validated account range response
// into the account tasks.
func (s *Syncer) processAccountResponse(res *accountResponse) {
	// Switch the task from pending to filling
	res.task.req = nil
	res.task.res = res

	// Ensure that the response doesn't overflow into the subsequent task
	last := res.task.Last.Big()
	for i, hash := range res.hashes {
		// Mark the range complete if the last is already included.
		// Keep iteration to delete the extra states if exists.
		cmp := hash.Big().Cmp(last)
		if cmp == 0 {
			res.cont = false
			continue
		}
		if cmp > 0 {
			// Chunk overflown, cut off excess
			res.hashes = res.hashes[:i]
			res.accounts = res.accounts[:i]
			res.cont = false // Mark range completed
			break
		}
	}
	// Iterate over all the accounts and assemble which ones need further sub-
	// filling before the entire account range can be persisted.
	res.task.needCode = make([]bool, len(res.accounts))
	res.task.needState = make([]bool, len(res.accounts))
	res.task.needHeal = make([]bool, len(res.accounts))

	res.task.codeTasks = make(map[common.Hash]struct{})
	res.task.stateTasks = make(map[common.Hash]common.Hash)

	resumed := make(map[common.Hash]struct{})

	res.task.pend = 0
	for i, account := range res.accounts {
		// Check if the account is a contract with an unknown code
		if !bytes.Equal(account.CodeHash, types.EmptyCodeHash.Bytes()) {
			if !rawdb.HasCodeWithPrefix(s.db, common.BytesToHash(account.CodeHash)) {
				res.task.codeTasks[common.BytesToHash(account.CodeHash)] = struct{}{}
				res.task.needCode[i] = true
				res.task.pend++
			}
		}
		// Check if the account is a contract with an unknown storage trie
		if account.Root != types.EmptyRootHash {
			if !rawdb.HasTrieNode(s.db, res.hashes[i], nil, account.Root, s.scheme) {
				// If there was a previous large state retrieval in progress,
				// don't restart it from scratch. This happens if a sync cycle
				// is interrupted and resumed later. However, *do* update the
				// previous root hash.
				if subtasks, ok := res.task.SubTasks[res.hashes[i]]; ok {
					log.Debug("Resuming large storage retrieval", "account", res.hashes[i], "root", account.Root)
					for _, subtask := range subtasks {
						subtask.root = account.Root
					}
					res.task.needHeal[i] = true
					resumed[res.hashes[i]] = struct{}{}
				} else {
					res.task.stateTasks[res.hashes[i]] = account.Root
				}
				res.task.needState[i] = true
				res.task.pend++
			}
		}
	}
	// Delete any subtasks that have been aborted but not resumed. This may undo
	// some progress if a new peer gives us less accounts than an old one, but for
	// now we have to live with that.
	for hash := range res.task.SubTasks {
		if _, ok := resumed[hash]; !ok {
			log.Debug("Aborting suspended storage retrieval", "account", hash)
			delete(res.task.SubTasks, hash)
		}
	}
	// If the account range contained no contracts, or all have been fully filled
	// beforehand, short circuit storage filling and forward to the next task
	if res.task.pend == 0 {
		s.forwardAccountTask(res.task)
		return
	}
	// Some accounts are incomplete, leave as is for the storage and contract
	// task assigners to pick up and fill
}

// processBytecodeResponse integrates an already validated bytecode response
// into the account tasks.
func (s *Syncer) processBytecodeResponse(res *bytecodeResponse) {
	batch := s.db.NewBatch()

	var (
		codes uint64
	)
	for i, hash := range res.hashes {
		code := res.codes[i]

		// If the bytecode was not delivered, reschedule it
		if code == nil {
			res.task.codeTasks[hash] = struct{}{}
			continue
		}
		// Code was delivered, mark it not needed any more
		for j, account := range res.task.res.accounts {
			if res.task.needCode[j] && hash == common.BytesToHash(account.CodeHash) {
				res.task.needCode[j] = false
				res.task.pend--
			}
		}
		// Push the bytecode into a database batch
		codes++
		rawdb.WriteCode(batch, hash, code)
	}
	bytes := common.StorageSize(batch.ValueSize())
	if err := batch.Write(); err != nil {
		log.Crit("Failed to persist bytecodes", "err", err)
	}
	s.bytecodeSynced += codes
	s.bytecodeBytes += bytes

	log.Debug("Persisted set of bytecodes", "count", codes, "bytes", bytes)

	// If this delivery completed the last pending task, forward the account task
	// to the next chunk
	if res.task.pend == 0 {
		s.forwardAccountTask(res.task)
		return
	}
	// Some accounts are still incomplete, leave as is for the storage and contract
	// task assigners to pick up and fill.
}

// processStorageResponse integrates an already validated storage response
// into the account tasks.
func (s *Syncer) processStorageResponse(res *storageResponse) {
	// Switch the subtask from pending to idle
	if res.subTask != nil {
		res.subTask.req = nil
	}
	batch := ethdb.HookedBatch{
		Batch: s.db.NewBatch(),
		OnPut: func(key []byte, value []byte) {
			s.storageBytes += common.StorageSize(len(key) + len(value))
		},
	}
	var (
		slots           int
		oldStorageBytes = s.storageBytes
	)
	// Iterate over all the accounts and reconstruct their storage tries from the
	// delivered slots
	for i, account := range res.accounts {
		// If the account was not delivered, reschedule it
		if i >= len(res.hashes) {
			res.mainTask.stateTasks[account] = res.roots[i]
			continue
		}
		// State was delivered, if complete mark as not needed any more, otherwise
		// mark the account as needing healing
		for j, hash := range res.mainTask.res.hashes {
			if account != hash {
				continue
			}
			acc := res.mainTask.res.accounts[j]

			// If the packet contains multiple contract storage slots, all
			// but the last are surely complete. The last contract may be
			// chunked, so check it's continuation flag.
			if res.subTask == nil && res.mainTask.needState[j] && (i < len(res.hashes)-1 || !res.cont) {
				res.mainTask.needState[j] = false
				res.mainTask.pend--
				smallStorageGauge.Inc(1)
			}
			// If the last contract was chunked, mark it as needing healing
			// to avoid writing it out to disk prematurely.
			if res.subTask == nil && !res.mainTask.needHeal[j] && i == len(res.hashes)-1 && res.cont {
				res.mainTask.needHeal[j] = true
			}
			// If the last contract was chunked, we need to switch to large
			// contract handling mode
			if res.subTask == nil && i == len(res.hashes)-1 && res.cont {
				// If we haven't yet started a large-contract retrieval, create
				// the subtasks for it within the main account task
				if tasks, ok := res.mainTask.SubTasks[account]; !ok {
					var (
						keys    = res.hashes[i]
						chunks  = uint64(storageConcurrency)
						lastKey common.Hash
					)
					if len(keys) > 0 {
						lastKey = keys[len(keys)-1]
					}
					// If the number of slots remaining is low, decrease the
					// number of chunks. Somewhere on the order of 10-15K slots
					// fit into a packet of 500KB. A key/slot pair is maximum 64
					// bytes, so pessimistically maxRequestSize/64 = 8K.
					//
					// Chunk so that at least 2 packets are needed to fill a task.
					if estimate, err := estimateRemainingSlots(len(keys), lastKey); err == nil {
						if n := estimate / (2 * (maxRequestSize / 64)); n+1 < chunks {
							chunks = n + 1
						}
						log.Debug("Chunked large contract", "initiators", len(keys), "tail", lastKey, "remaining", estimate, "chunks", chunks)
					} else {
						log.Debug("Chunked large contract", "initiators", len(keys), "tail", lastKey, "chunks", chunks)
					}
					r := newHashRange(lastKey, chunks)
					if chunks == 1 {
						smallStorageGauge.Inc(1)
					} else {
						largeStorageGauge.Inc(1)
					}
					// Our first task is the one that was just filled by this response.
					batch := ethdb.HookedBatch{
						Batch: s.db.NewBatch(),
						OnPut: func(key []byte, value []byte) {
							s.storageBytes += common.StorageSize(len(key) + len(value))
						},
					}
					owner := account // local assignment for stacktrie writer closure
					options := trie.NewStackTrieOptions()
					options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
						rawdb.WriteTrieNode(batch, owner, path, hash, blob, s.scheme)
					})
					if s.scheme == rawdb.PathScheme {
						options = options.WithCleaner(func(path []byte) {
							s.cleanPath(batch, owner, path)
						})
						// Keep the left boundary as it's the first range.
						// Skip the right boundary if it's not the last range.
						options = options.WithSkipBoundary(false, r.End() != common.MaxHash, boundaryStorageNodesGauge)
					}
					tasks = append(tasks, &storageTask{
						Next:     common.Hash{},
						Last:     r.End(),
						root:     acc.Root,
						genBatch: batch,
						genTrie:  trie.NewStackTrie(options),
					})
					for r.Next() {
						batch := ethdb.HookedBatch{
							Batch: s.db.NewBatch(),
							OnPut: func(key []byte, value []byte) {
								s.storageBytes += common.StorageSize(len(key) + len(value))
							},
						}
						options := trie.NewStackTrieOptions()
						options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
							rawdb.WriteTrieNode(batch, owner, path, hash, blob, s.scheme)
						})
						if s.scheme == rawdb.PathScheme {
							// Configure the dangling node cleaner and also filter out boundary nodes
							// only in the context of the path scheme. Deletion is forbidden in the
							// hash scheme, as it can disrupt state completeness.
							options = options.WithCleaner(func(path []byte) {
								s.cleanPath(batch, owner, path)
							})
							// Skip the left boundary as it's not the first range
							// Skip the right boundary if it's not the last range.
							options = options.WithSkipBoundary(true, r.End() != common.MaxHash, boundaryStorageNodesGauge)
						}
						tasks = append(tasks, &storageTask{
							Next:     r.Start(),
							Last:     r.End(),
							root:     acc.Root,
							genBatch: batch,
							genTrie:  trie.NewStackTrie(options),
						})
					}
					for _, task := range tasks {
						log.Debug("Created storage sync task", "account", account, "root", acc.Root, "from", task.Next, "last", task.Last)
					}
					res.mainTask.SubTasks[account] = tasks

					// Since we've just created the sub-tasks, this response
					// is surely for the first one (zero origin)
					res.subTask = tasks[0]
				}
			}
			// If we're in large contract delivery mode, forward the subtask
			if res.subTask != nil {
				// Ensure the response doesn't overflow into the subsequent task
				last := res.subTask.Last.Big()
				// Find the first overflowing key. While at it, mark res as complete
				// if we find the range to include or pass the 'last'
				index := sort.Search(len(res.hashes[i]), func(k int) bool {
					cmp := res.hashes[i][k].Big().Cmp(last)
					if cmp >= 0 {
						res.cont = false
					}
					return cmp > 0
				})
				if index >= 0 {
					// cut off excess
					res.hashes[i] = res.hashes[i][:index]
					res.slots[i] = res.slots[i][:index]
				}
				// Forward the relevant storage chunk (even if created just now)
				if res.cont {
					res.subTask.Next = incHash(res.hashes[i][len(res.hashes[i])-1])
				} else {
					res.subTask.done = true
				}
			}
		}
		// Iterate over all the complete contracts, reconstruct the trie nodes and
		// push them to disk. If the contract is chunked, the trie nodes will be
		// reconstructed later.
		slots += len(res.hashes[i])

		if i < len(res.hashes)-1 || res.subTask == nil {
			// no need to make local reassignment of account: this closure does not outlive the loop
			options := trie.NewStackTrieOptions()
			options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
				rawdb.WriteTrieNode(batch, account, path, hash, blob, s.scheme)
			})
			if s.scheme == rawdb.PathScheme {
				// Configure the dangling node cleaner only in the context of the
				// path scheme. Deletion is forbidden in the hash scheme, as it can
				// disrupt state completeness.
				//
				// Notably, boundary nodes can be also kept because the whole storage
				// trie is complete.
				options = options.WithCleaner(func(path []byte) {
					s.cleanPath(batch, account, path)
				})
			}
			tr := trie.NewStackTrie(options)
			for j := 0; j < len(res.hashes[i]); j++ {
				tr.Update(res.hashes[i][j][:], res.slots[i][j])
			}
			tr.Commit()
		}
		// Persist the received storage segments. These flat state maybe
		// outdated during the sync, but it can be fixed later during the
		// snapshot generation.
		for j := 0; j < len(res.hashes[i]); j++ {
			rawdb.WriteStorageSnapshot(batch, account, res.hashes[i][j], res.slots[i][j])

			// If we're storing large contracts, generate the trie nodes
			// on the fly to not trash the gluing points
			if i == len(res.hashes)-1 && res.subTask != nil {
				res.subTask.genTrie.Update(res.hashes[i][j][:], res.slots[i][j])
			}
		}
	}
	// Large contracts could have generated new trie nodes, flush them to disk
	if res.subTask != nil {
		if res.subTask.done {
			root := res.subTask.genTrie.Commit()
			if err := res.subTask.genBatch.Write(); err != nil {
				log.Error("Failed to persist stack slots", "err", err)
			}
			res.subTask.genBatch.Reset()

			// If the chunk's root is an overflown but full delivery,
			// clear the heal request.
			accountHash := res.accounts[len(res.accounts)-1]
			if root == res.subTask.root && rawdb.HasStorageTrieNode(s.db, accountHash, nil, root) {
				for i, account := range res.mainTask.res.hashes {
					if account == accountHash {
						res.mainTask.needHeal[i] = false
						skipStorageHealingGauge.Inc(1)
					}
				}
			}
		}
		if res.subTask.genBatch.ValueSize() > ethdb.IdealBatchSize {
			if err := res.subTask.genBatch.Write(); err != nil {
				log.Error("Failed to persist stack slots", "err", err)
			}
			res.subTask.genBatch.Reset()
		}
	}
	// Flush anything written just now and update the stats
	if err := batch.Write(); err != nil {
		log.Crit("Failed to persist storage slots", "err", err)
	}
	s.storageSynced += uint64(slots)

	log.Debug("Persisted set of storage slots", "accounts", len(res.hashes), "slots", slots, "bytes", s.storageBytes-oldStorageBytes)

	// If this delivery completed the last pending task, forward the account task
	// to the next chunk
	if res.mainTask.pend == 0 {
		s.forwardAccountTask(res.mainTask)
		return
	}
	// Some accounts are still incomplete, leave as is for the storage and contract
	// task assigners to pick up and fill.
}

// processTrienodeHealResponse integrates an already validated trienode response
// into the healer tasks.
func (s *Syncer) processTrienodeHealResponse(res *trienodeHealResponse) {
	var (
		start = time.Now()
		fills int
	)
	for i, hash := range res.hashes {
		node := res.nodes[i]

		// If the trie node was not delivered, reschedule it
		if node == nil {
			res.task.trieTasks[res.paths[i]] = res.hashes[i]
			continue
		}
		fills++

		// Push the trie node into the state syncer
		s.trienodeHealSynced++
		s.trienodeHealBytes += common.StorageSize(len(node))

		err := s.healer.scheduler.ProcessNode(trie.NodeSyncResult{Path: res.paths[i], Data: node})
		switch err {
		case nil:
		case trie.ErrAlreadyProcessed:
			s.trienodeHealDups++
		case trie.ErrNotRequested:
			s.trienodeHealNops++
		default:
			log.Error("Invalid trienode processed", "hash", hash, "err", err)
		}
	}
	s.commitHealer(false)

	// Calculate the processing rate of one filled trie node
	rate := float64(fills) / (float64(time.Since(start)) / float64(time.Second))

	// Update the currently measured trienode queueing and processing throughput.
	//
	// The processing rate needs to be updated uniformly independent if we've
	// processed 1x100 trie nodes or 100x1 to keep the rate consistent even in
	// the face of varying network packets. As such, we cannot just measure the
	// time it took to process N trie nodes and update once, we need one update
	// per trie node.
	//
	// Naively, that would be:
	//
	//   for i:=0; i<fills; i++ {
	//     healRate = (1-measurementImpact)*oldRate + measurementImpact*newRate
	//   }
	//
	// Essentially, a recursive expansion of HR = (1-MI)*HR + MI*NR.
	//
	// We can expand that formula for the Nth item as:
	//   HR(N) = (1-MI)^N*OR + (1-MI)^(N-1)*MI*NR + (1-MI)^(N-2)*MI*NR + ... + (1-MI)^0*MI*NR
	//
	// The above is a geometric sequence that can be summed to:
	//   HR(N) = (1-MI)^N*(OR-NR) + NR
	s.trienodeHealRate = gomath.Pow(1-trienodeHealRateMeasurementImpact, float64(fills))*(s.trienodeHealRate-rate) + rate

	pending := s.trienodeHealPend.Load()
	if time.Since(s.trienodeHealThrottled) > time.Second {
		// Periodically adjust the trie node throttler
		if float64(pending) > 2*s.trienodeHealRate {
			s.trienodeHealThrottle *= trienodeHealThrottleIncrease
		} else {
			s.trienodeHealThrottle /= trienodeHealThrottleDecrease
		}
		if s.trienodeHealThrottle > maxTrienodeHealThrottle {
			s.trienodeHealThrottle = maxTrienodeHealThrottle
		} else if s.trienodeHealThrottle < minTrienodeHealThrottle {
			s.trienodeHealThrottle = minTrienodeHealThrottle
		}
		s.trienodeHealThrottled = time.Now()

		log.Debug("Updated trie node heal throttler", "rate", s.trienodeHealRate, "pending", pending, "throttle", s.trienodeHealThrottle)
	}
}

func (s *Syncer) commitHealer(force bool) {
	if !force && s.healer.scheduler.MemSize() < ethdb.IdealBatchSize {
		return
	}
	batch := s.db.NewBatch()
	if err := s.healer.scheduler.Commit(batch); err != nil {
		log.Error("Failed to commit healing data", "err", err)
	}
	if err := batch.Write(); err != nil {
		log.Crit("Failed to persist healing data", "err", err)
	}
	log.Debug("Persisted set of healing data", "type", "trienodes", "bytes", common.StorageSize(batch.ValueSize()))
}

// processBytecodeHealResponse integrates an already validated bytecode response
// into the healer tasks.
func (s *Syncer) processBytecodeHealResponse(res *bytecodeHealResponse) {
	for i, hash := range res.hashes {
		node := res.codes[i]

		// If the trie node was not delivered, reschedule it
		if node == nil {
			res.task.codeTasks[hash] = struct{}{}
			continue
		}
		// Push the trie node into the state syncer
		s.bytecodeHealSynced++
		s.bytecodeHealBytes += common.StorageSize(len(node))

		err := s.healer.scheduler.ProcessCode(trie.CodeSyncResult{Hash: hash, Data: node})
		switch err {
		case nil:
		case trie.ErrAlreadyProcessed:
			s.bytecodeHealDups++
		case trie.ErrNotRequested:
			s.bytecodeHealNops++
		default:
			log.Error("Invalid bytecode processed", "hash", hash, "err", err)
		}
	}
	s.commitHealer(false)
}

// forwardAccountTask takes a filled account task and persists anything available
// into the database, after which it forwards the next account marker so that the
// task's next chunk may be filled.
func (s *Syncer) forwardAccountTask(task *accountTask) {
	// Remove any pending delivery
	res := task.res
	if res == nil {
		return // nothing to forward
	}
	task.res = nil

	// Persist the received account segments. These flat state maybe
	// outdated during the sync, but it can be fixed later during the
	// snapshot generation.
	oldAccountBytes := s.accountBytes

	batch := ethdb.HookedBatch{
		Batch: s.db.NewBatch(),
		OnPut: func(key []byte, value []byte) {
			s.accountBytes += common.StorageSize(len(key) + len(value))
		},
	}
	for i, hash := range res.hashes {
		if task.needCode[i] || task.needState[i] {
			break
		}
		slim := types.SlimAccountRLP(*res.accounts[i])
		rawdb.WriteAccountSnapshot(batch, hash, slim)

		// If the task is complete, drop it into the stack trie to generate
		// account trie nodes for it
		if !task.needHeal[i] {
			full, err := types.FullAccountRLP(slim) // TODO(karalabe): Slim parsing can be omitted
			if err != nil {
				panic(err) // Really shouldn't ever happen
			}
			task.genTrie.Update(hash[:], full)
		}
	}
	// Flush anything written just now and update the stats
	if err := batch.Write(); err != nil {
		log.Crit("Failed to persist accounts", "err", err)
	}
	s.accountSynced += uint64(len(res.accounts))

	// Task filling persisted, push it the chunk marker forward to the first
	// account still missing data.
	for i, hash := range res.hashes {
		if task.needCode[i] || task.needState[i] {
			return
		}
		task.Next = incHash(hash)
	}
	// All accounts marked as complete, track if the entire task is done
	task.done = !res.cont

	// Stack trie could have generated trie nodes, push them to disk (we need to
	// flush after finalizing task.done. It's fine even if we crash and lose this
	// write as it will only cause more data to be downloaded during heal.
	if task.done {
		task.genTrie.Commit()
	}
	if task.genBatch.ValueSize() > ethdb.IdealBatchSize || task.done {
		if err := task.genBatch.Write(); err != nil {
			log.Error("Failed to persist stack account", "err", err)
		}
		task.genBatch.Reset()
	}
	log.Debug("Persisted range of accounts", "accounts", len(res.accounts), "bytes", s.accountBytes-oldAccountBytes)
}

// OnAccounts is a callback method to invoke when a range of accounts are
// received from a remote peer.
func (s *Syncer) OnAccounts(peer SyncPeer, id uint64, hashes []common.Hash, accounts [][]byte, proof [][]byte) error {
	size := common.StorageSize(len(hashes) * common.HashLength)
	for _, account := range accounts {
		size += common.StorageSize(len(account))
	}
	for _, node := range proof {
		size += common.StorageSize(len(node))
	}
	logger := peer.Log().New("reqid", id)
	logger.Trace("Delivering range of accounts", "hashes", len(hashes), "accounts", len(accounts), "proofs", len(proof), "bytes", size)

	// Whether or not the response is valid, we can mark the peer as idle and
	// notify the scheduler to assign a new task. If the response is invalid,
	// we'll drop the peer in a bit.
	defer func() {
		s.lock.Lock()
		defer s.lock.Unlock()
		if _, ok := s.peers[peer.ID()]; ok {
			s.accountIdlers[peer.ID()] = struct{}{}
		}
		select {
		case s.update <- struct{}{}:
		default:
		}
	}()
	s.lock.Lock()
	// Ensure the response is for a valid request
	req, ok := s.accountReqs[id]
	if !ok {
		// Request stale, perhaps the peer timed out but came through in the end
		logger.Warn("Unexpected account range packet")
		s.lock.Unlock()
		return nil
	}
	delete(s.accountReqs, id)
	s.rates.Update(peer.ID(), AccountRangeMsg, time.Since(req.time), int(size))

	// Clean up the request timeout timer, we'll see how to proceed further based
	// on the actual delivered content
	if !req.timeout.Stop() {
		// The timeout is already triggered, and this request will be reverted+rescheduled
		s.lock.Unlock()
		return nil
	}
	// Response is valid, but check if peer is signalling that it does not have
	// the requested data. For account range queries that means the state being
	// retrieved was either already pruned remotely, or the peer is not yet
	// synced to our head.
	if len(hashes) == 0 && len(accounts) == 0 && len(proof) == 0 {
		logger.Debug("Peer rejected account range request", "root", s.root)
		s.statelessPeers[peer.ID()] = struct{}{}
		s.lock.Unlock()

		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertAccountRequest(req)
		return nil
	}
	root := s.root
	s.lock.Unlock()

	// Reconstruct a partial trie from the response and verify it
	keys := make([][]byte, len(hashes))
	for i, key := range hashes {
		keys[i] = common.CopyBytes(key[:])
	}
	nodes := make(trienode.ProofList, len(proof))
	for i, node := range proof {
		nodes[i] = node
	}
	cont, err := trie.VerifyRangeProof(root, req.origin[:], keys, accounts, nodes.Set())
	if err != nil {
		logger.Warn("Account range failed proof", "err", err)
		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertAccountRequest(req)
		return err
	}
	accs := make([]*types.StateAccount, len(accounts))
	for i, account := range accounts {
		acc := new(types.StateAccount)
		if err := rlp.DecodeBytes(account, acc); err != nil {
			panic(err) // We created these blobs, we must be able to decode them
		}
		accs[i] = acc
	}
	response := &accountResponse{
		task:     req.task,
		hashes:   hashes,
		accounts: accs,
		cont:     cont,
	}
	select {
	case req.deliver <- response:
	case <-req.cancel:
	case <-req.stale:
	}
	return nil
}

// OnByteCodes is a callback method to invoke when a batch of contract
// bytes codes are received from a remote peer.
func (s *Syncer) OnByteCodes(peer SyncPeer, id uint64, bytecodes [][]byte) error {
	s.lock.RLock()
	syncing := !s.snapped
	s.lock.RUnlock()

	if syncing {
		return s.onByteCodes(peer, id, bytecodes)
	}
	return s.onHealByteCodes(peer, id, bytecodes)
}

// onByteCodes is a callback method to invoke when a batch of contract
// bytes codes are received from a remote peer in the syncing phase.
func (s *Syncer) onByteCodes(peer SyncPeer, id uint64, bytecodes [][]byte) error {
	var size common.StorageSize
	for _, code := range bytecodes {
		size += common.StorageSize(len(code))
	}
	logger := peer.Log().New("reqid", id)
	logger.Trace("Delivering set of bytecodes", "bytecodes", len(bytecodes), "bytes", size)

	// Whether or not the response is valid, we can mark the peer as idle and
	// notify the scheduler to assign a new task. If the response is invalid,
	// we'll drop the peer in a bit.
	defer func() {
		s.lock.Lock()
		defer s.lock.Unlock()
		if _, ok := s.peers[peer.ID()]; ok {
			s.bytecodeIdlers[peer.ID()] = struct{}{}
		}
		select {
		case s.update <- struct{}{}:
		default:
		}
	}()
	s.lock.Lock()
	// Ensure the response is for a valid request
	req, ok := s.bytecodeReqs[id]
	if !ok {
		// Request stale, perhaps the peer timed out but came through in the end
		logger.Warn("Unexpected bytecode packet")
		s.lock.Unlock()
		return nil
	}
	delete(s.bytecodeReqs, id)
	s.rates.Update(peer.ID(), ByteCodesMsg, time.Since(req.time), len(bytecodes))

	// Clean up the request timeout timer, we'll see how to proceed further based
	// on the actual delivered content
	if !req.timeout.Stop() {
		// The timeout is already triggered, and this request will be reverted+rescheduled
		s.lock.Unlock()
		return nil
	}

	// Response is valid, but check if peer is signalling that it does not have
	// the requested data. For bytecode range queries that means the peer is not
	// yet synced.
	if len(bytecodes) == 0 {
		logger.Debug("Peer rejected bytecode request")
		s.statelessPeers[peer.ID()] = struct{}{}
		s.lock.Unlock()

		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertBytecodeRequest(req)
		return nil
	}
	s.lock.Unlock()

	// Cross reference the requested bytecodes with the response to find gaps
	// that the serving node is missing
	hasher := sha3.NewLegacyKeccak256().(crypto.KeccakState)
	hash := make([]byte, 32)

	codes := make([][]byte, len(req.hashes))
	for i, j := 0, 0; i < len(bytecodes); i++ {
		// Find the next hash that we've been served, leaving misses with nils
		hasher.Reset()
		hasher.Write(bytecodes[i])
		hasher.Read(hash)

		for j < len(req.hashes) && !bytes.Equal(hash, req.hashes[j][:]) {
			j++
		}
		if j < len(req.hashes) {
			codes[j] = bytecodes[i]
			j++
			continue
		}
		// We've either ran out of hashes, or got unrequested data
		logger.Warn("Unexpected bytecodes", "count", len(bytecodes)-i)
		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertBytecodeRequest(req)
		return errors.New("unexpected bytecode")
	}
	// Response validated, send it to the scheduler for filling
	response := &bytecodeResponse{
		task:   req.task,
		hashes: req.hashes,
		codes:  codes,
	}
	select {
	case req.deliver <- response:
	case <-req.cancel:
	case <-req.stale:
	}
	return nil
}

// OnStorage is a callback method to invoke when ranges of storage slots
// are received from a remote peer.
func (s *Syncer) OnStorage(peer SyncPeer, id uint64, hashes [][]common.Hash, slots [][][]byte, proof [][]byte) error {
	// Gather some trace stats to aid in debugging issues
	var (
		hashCount int
		slotCount int
		size      common.StorageSize
	)
	for _, hashset := range hashes {
		size += common.StorageSize(common.HashLength * len(hashset))
		hashCount += len(hashset)
	}
	for _, slotset := range slots {
		for _, slot := range slotset {
			size += common.StorageSize(len(slot))
		}
		slotCount += len(slotset)
	}
	for _, node := range proof {
		size += common.StorageSize(len(node))
	}
	logger := peer.Log().New("reqid", id)
	logger.Trace("Delivering ranges of storage slots", "accounts", len(hashes), "hashes", hashCount, "slots", slotCount, "proofs", len(proof), "size", size)

	// Whether or not the response is valid, we can mark the peer as idle and
	// notify the scheduler to assign a new task. If the response is invalid,
	// we'll drop the peer in a bit.
	defer func() {
		s.lock.Lock()
		defer s.lock.Unlock()
		if _, ok := s.peers[peer.ID()]; ok {
			s.storageIdlers[peer.ID()] = struct{}{}
		}
		select {
		case s.update <- struct{}{}:
		default:
		}
	}()
	s.lock.Lock()
	// Ensure the response is for a valid request
	req, ok := s.storageReqs[id]
	if !ok {
		// Request stale, perhaps the peer timed out but came through in the end
		logger.Warn("Unexpected storage ranges packet")
		s.lock.Unlock()
		return nil
	}
	delete(s.storageReqs, id)
	s.rates.Update(peer.ID(), StorageRangesMsg, time.Since(req.time), int(size))

	// Clean up the request timeout timer, we'll see how to proceed further based
	// on the actual delivered content
	if !req.timeout.Stop() {
		// The timeout is already triggered, and this request will be reverted+rescheduled
		s.lock.Unlock()
		return nil
	}

	// Reject the response if the hash sets and slot sets don't match, or if the
	// peer sent more data than requested.
	if len(hashes) != len(slots) {
		s.lock.Unlock()
		s.scheduleRevertStorageRequest(req) // reschedule request
		logger.Warn("Hash and slot set size mismatch", "hashset", len(hashes), "slotset", len(slots))
		return errors.New("hash and slot set size mismatch")
	}
	if len(hashes) > len(req.accounts) {
		s.lock.Unlock()
		s.scheduleRevertStorageRequest(req) // reschedule request
		logger.Warn("Hash set larger than requested", "hashset", len(hashes), "requested", len(req.accounts))
		return errors.New("hash set larger than requested")
	}
	// Response is valid, but check if peer is signalling that it does not have
	// the requested data. For storage range queries that means the state being
	// retrieved was either already pruned remotely, or the peer is not yet
	// synced to our head.
	if len(hashes) == 0 && len(proof) == 0 {
		logger.Debug("Peer rejected storage request")
		s.statelessPeers[peer.ID()] = struct{}{}
		s.lock.Unlock()
		s.scheduleRevertStorageRequest(req) // reschedule request
		return nil
	}
	s.lock.Unlock()

	// Reconstruct the partial tries from the response and verify them
	var cont bool

	// If a proof was attached while the response is empty, it indicates that the
	// requested range specified with 'origin' is empty. Construct an empty state
	// response locally to finalize the range.
	if len(hashes) == 0 && len(proof) > 0 {
		hashes = append(hashes, []common.Hash{})
		slots = append(slots, [][]byte{})
	}
	for i := 0; i < len(hashes); i++ {
		// Convert the keys and proofs into an internal format
		keys := make([][]byte, len(hashes[i]))
		for j, key := range hashes[i] {
			keys[j] = common.CopyBytes(key[:])
		}
		nodes := make(trienode.ProofList, 0, len(proof))
		if i == len(hashes)-1 {
			for _, node := range proof {
				nodes = append(nodes, node)
			}
		}
		var err error
		if len(nodes) == 0 {
			// No proof has been attached, the response must cover the entire key
			// space and hash to the origin root.
			_, err = trie.VerifyRangeProof(req.roots[i], nil, keys, slots[i], nil)
			if err != nil {
				s.scheduleRevertStorageRequest(req) // reschedule request
				logger.Warn("Storage slots failed proof", "err", err)
				return err
			}
		} else {
			// A proof was attached, the response is only partial, check that the
			// returned data is indeed part of the storage trie
			proofdb := nodes.Set()

			cont, err = trie.VerifyRangeProof(req.roots[i], req.origin[:], keys, slots[i], proofdb)
			if err != nil {
				s.scheduleRevertStorageRequest(req) // reschedule request
				logger.Warn("Storage range failed proof", "err", err)
				return err
			}
		}
	}
	// Partial tries reconstructed, send them to the scheduler for storage filling
	response := &storageResponse{
		mainTask: req.mainTask,
		subTask:  req.subTask,
		accounts: req.accounts,
		roots:    req.roots,
		hashes:   hashes,
		slots:    slots,
		cont:     cont,
	}
	select {
	case req.deliver <- response:
	case <-req.cancel:
	case <-req.stale:
	}
	return nil
}

// OnTrieNodes is a callback method to invoke when a batch of trie nodes
// are received from a remote peer.
func (s *Syncer) OnTrieNodes(peer SyncPeer, id uint64, trienodes [][]byte) error {
	var size common.StorageSize
	for _, node := range trienodes {
		size += common.StorageSize(len(node))
	}
	logger := peer.Log().New("reqid", id)
	logger.Trace("Delivering set of healing trienodes", "trienodes", len(trienodes), "bytes", size)

	// Whether or not the response is valid, we can mark the peer as idle and
	// notify the scheduler to assign a new task. If the response is invalid,
	// we'll drop the peer in a bit.
	defer func() {
		s.lock.Lock()
		defer s.lock.Unlock()
		if _, ok := s.peers[peer.ID()]; ok {
			s.trienodeHealIdlers[peer.ID()] = struct{}{}
		}
		select {
		case s.update <- struct{}{}:
		default:
		}
	}()
	s.lock.Lock()
	// Ensure the response is for a valid request
	req, ok := s.trienodeHealReqs[id]
	if !ok {
		// Request stale, perhaps the peer timed out but came through in the end
		logger.Warn("Unexpected trienode heal packet")
		s.lock.Unlock()
		return nil
	}
	delete(s.trienodeHealReqs, id)
	s.rates.Update(peer.ID(), TrieNodesMsg, time.Since(req.time), len(trienodes))

	// Clean up the request timeout timer, we'll see how to proceed further based
	// on the actual delivered content
	if !req.timeout.Stop() {
		// The timeout is already triggered, and this request will be reverted+rescheduled
		s.lock.Unlock()
		return nil
	}

	// Response is valid, but check if peer is signalling that it does not have
	// the requested data. For bytecode range queries that means the peer is not
	// yet synced.
	if len(trienodes) == 0 {
		logger.Debug("Peer rejected trienode heal request")
		s.statelessPeers[peer.ID()] = struct{}{}
		s.lock.Unlock()

		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertTrienodeHealRequest(req)
		return nil
	}
	s.lock.Unlock()

	// Cross reference the requested trienodes with the response to find gaps
	// that the serving node is missing
	var (
		hasher = sha3.NewLegacyKeccak256().(crypto.KeccakState)
		hash   = make([]byte, 32)
		nodes  = make([][]byte, len(req.hashes))
		fills  uint64
	)
	for i, j := 0, 0; i < len(trienodes); i++ {
		// Find the next hash that we've been served, leaving misses with nils
		hasher.Reset()
		hasher.Write(trienodes[i])
		hasher.Read(hash)

		for j < len(req.hashes) && !bytes.Equal(hash, req.hashes[j][:]) {
			j++
		}
		if j < len(req.hashes) {
			nodes[j] = trienodes[i]
			fills++
			j++
			continue
		}
		// We've either ran out of hashes, or got unrequested data
		logger.Warn("Unexpected healing trienodes", "count", len(trienodes)-i)

		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertTrienodeHealRequest(req)
		return errors.New("unexpected healing trienode")
	}
	// Response validated, send it to the scheduler for filling
	s.trienodeHealPend.Add(fills)
	defer func() {
		s.trienodeHealPend.Add(^(fills - 1))
	}()
	response := &trienodeHealResponse{
		paths:  req.paths,
		task:   req.task,
		hashes: req.hashes,
		nodes:  nodes,
	}
	select {
	case req.deliver <- response:
	case <-req.cancel:
	case <-req.stale:
	}
	return nil
}

// onHealByteCodes is a callback method to invoke when a batch of contract
// bytes codes are received from a remote peer in the healing phase.
func (s *Syncer) onHealByteCodes(peer SyncPeer, id uint64, bytecodes [][]byte) error {
	var size common.StorageSize
	for _, code := range bytecodes {
		size += common.StorageSize(len(code))
	}
	logger := peer.Log().New("reqid", id)
	logger.Trace("Delivering set of healing bytecodes", "bytecodes", len(bytecodes), "bytes", size)

	// Whether or not the response is valid, we can mark the peer as idle and
	// notify the scheduler to assign a new task. If the response is invalid,
	// we'll drop the peer in a bit.
	defer func() {
		s.lock.Lock()
		defer s.lock.Unlock()
		if _, ok := s.peers[peer.ID()]; ok {
			s.bytecodeHealIdlers[peer.ID()] = struct{}{}
		}
		select {
		case s.update <- struct{}{}:
		default:
		}
	}()
	s.lock.Lock()
	// Ensure the response is for a valid request
	req, ok := s.bytecodeHealReqs[id]
	if !ok {
		// Request stale, perhaps the peer timed out but came through in the end
		logger.Warn("Unexpected bytecode heal packet")
		s.lock.Unlock()
		return nil
	}
	delete(s.bytecodeHealReqs, id)
	s.rates.Update(peer.ID(), ByteCodesMsg, time.Since(req.time), len(bytecodes))

	// Clean up the request timeout timer, we'll see how to proceed further based
	// on the actual delivered content
	if !req.timeout.Stop() {
		// The timeout is already triggered, and this request will be reverted+rescheduled
		s.lock.Unlock()
		return nil
	}

	// Response is valid, but check if peer is signalling that it does not have
	// the requested data. For bytecode range queries that means the peer is not
	// yet synced.
	if len(bytecodes) == 0 {
		logger.Debug("Peer rejected bytecode heal request")
		s.statelessPeers[peer.ID()] = struct{}{}
		s.lock.Unlock()

		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertBytecodeHealRequest(req)
		return nil
	}
	s.lock.Unlock()

	// Cross reference the requested bytecodes with the response to find gaps
	// that the serving node is missing
	hasher := sha3.NewLegacyKeccak256().(crypto.KeccakState)
	hash := make([]byte, 32)

	codes := make([][]byte, len(req.hashes))
	for i, j := 0, 0; i < len(bytecodes); i++ {
		// Find the next hash that we've been served, leaving misses with nils
		hasher.Reset()
		hasher.Write(bytecodes[i])
		hasher.Read(hash)

		for j < len(req.hashes) && !bytes.Equal(hash, req.hashes[j][:]) {
			j++
		}
		if j < len(req.hashes) {
			codes[j] = bytecodes[i]
			j++
			continue
		}
		// We've either ran out of hashes, or got unrequested data
		logger.Warn("Unexpected healing bytecodes", "count", len(bytecodes)-i)
		// Signal this request as failed, and ready for rescheduling
		s.scheduleRevertBytecodeHealRequest(req)
		return errors.New("unexpected healing bytecode")
	}
	// Response validated, send it to the scheduler for filling
	response := &bytecodeHealResponse{
		task:   req.task,
		hashes: req.hashes,
		codes:  codes,
	}
	select {
	case req.deliver <- response:
	case <-req.cancel:
	case <-req.stale:
	}
	return nil
}

// onHealState is a callback method to invoke when a flat state(account
// or storage slot) is downloaded during the healing stage. The flat states
// can be persisted blindly and can be fixed later in the generation stage.
// Note it's not concurrent safe, please handle the concurrent issue outside.
func (s *Syncer) onHealState(paths [][]byte, value []byte) error {
	if len(paths) == 1 {
		var account types.StateAccount
		if err := rlp.DecodeBytes(value, &account); err != nil {
			return nil // Returning the error here would drop the remote peer
		}
		blob := types.SlimAccountRLP(account)
		rawdb.WriteAccountSnapshot(s.stateWriter, common.BytesToHash(paths[0]), blob)
		s.accountHealed += 1
		s.accountHealedBytes += common.StorageSize(1 + common.HashLength + len(blob))
	}
	if len(paths) == 2 {
		rawdb.WriteStorageSnapshot(s.stateWriter, common.BytesToHash(paths[0]), common.BytesToHash(paths[1]), value)
		s.storageHealed += 1
		s.storageHealedBytes += common.StorageSize(1 + 2*common.HashLength + len(value))
	}
	if s.stateWriter.ValueSize() > ethdb.IdealBatchSize {
		s.stateWriter.Write() // It's fine to ignore the error here
		s.stateWriter.Reset()
	}
	return nil
}

// hashSpace is the total size of the 256 bit hash space for accounts.
var hashSpace = new(big.Int).Exp(common.Big2, common.Big256, nil)

// report calculates various status reports and provides it to the user.
func (s *Syncer) report(force bool) {
	if len(s.tasks) > 0 {
		s.reportSyncProgress(force)
		return
	}
	s.reportHealProgress(force)
}

// reportSyncProgress calculates various status reports and provides it to the user.
func (s *Syncer) reportSyncProgress(force bool) {
	// Don't report all the events, just occasionally
	if !force && time.Since(s.logTime) < 8*time.Second {
		return
	}
	// Don't report anything until we have a meaningful progress
	synced := s.accountBytes + s.bytecodeBytes + s.storageBytes
	if synced == 0 {
		return
	}
	accountGaps := new(big.Int)
	for _, task := range s.tasks {
		accountGaps.Add(accountGaps, new(big.Int).Sub(task.Last.Big(), task.Next.Big()))
	}
	accountFills := new(big.Int).Sub(hashSpace, accountGaps)
	if accountFills.BitLen() == 0 {
		return
	}
	s.logTime = time.Now()
	estBytes := float64(new(big.Int).Div(
		new(big.Int).Mul(new(big.Int).SetUint64(uint64(synced)), hashSpace),
		accountFills,
	).Uint64())
	// Don't report anything until we have a meaningful progress
	if estBytes < 1.0 {
		return
	}
	elapsed := time.Since(s.startTime)
	estTime := elapsed / time.Duration(synced) * time.Duration(estBytes)

	// Create a mega progress report
	var (
		progress = fmt.Sprintf("%.2f%%", float64(synced)*100/estBytes)
		accounts = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.accountSynced), s.accountBytes.TerminalString())
		storage  = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.storageSynced), s.storageBytes.TerminalString())
		bytecode = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.bytecodeSynced), s.bytecodeBytes.TerminalString())
	)
	log.Info("Syncing: state download in progress", "synced", progress, "state", synced,
		"accounts", accounts, "slots", storage, "codes", bytecode, "eta", common.PrettyDuration(estTime-elapsed))
}

// reportHealProgress calculates various status reports and provides it to the user.
func (s *Syncer) reportHealProgress(force bool) {
	// Don't report all the events, just occasionally
	if !force && time.Since(s.logTime) < 8*time.Second {
		return
	}
	s.logTime = time.Now()

	// Create a mega progress report
	var (
		trienode = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.trienodeHealSynced), s.trienodeHealBytes.TerminalString())
		bytecode = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.bytecodeHealSynced), s.bytecodeHealBytes.TerminalString())
		accounts = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.accountHealed), s.accountHealedBytes.TerminalString())
		storage  = fmt.Sprintf("%v@%v", log.FormatLogfmtUint64(s.storageHealed), s.storageHealedBytes.TerminalString())
	)
	log.Info("Syncing: state healing in progress", "accounts", accounts, "slots", storage,
		"codes", bytecode, "nodes", trienode, "pending", s.healer.scheduler.Pending())
}

// estimateRemainingSlots tries to determine roughly how many slots are left in
// a contract storage, based on the number of keys and the last hash. This method
// assumes that the hashes are lexicographically ordered and evenly distributed.
func estimateRemainingSlots(hashes int, last common.Hash) (uint64, error) {
	if last == (common.Hash{}) {
		return 0, errors.New("last hash empty")
	}
	space := new(big.Int).Mul(math.MaxBig256, big.NewInt(int64(hashes)))
	space.Div(space, last.Big())
	if !space.IsUint64() {
		// Gigantic address space probably due to too few or malicious slots
		return 0, errors.New("too few slots for estimation")
	}
	return space.Uint64() - uint64(hashes), nil
}

// capacitySort implements the Sort interface, allowing sorting by peer message
// throughput. Note, callers should use sort.Reverse to get the desired effect
// of highest capacity being at the front.
type capacitySort struct {
	ids  []string
	caps []int
}

func (s *capacitySort) Len() int {
	return len(s.ids)
}

func (s *capacitySort) Less(i, j int) bool {
	return s.caps[i] < s.caps[j]
}

func (s *capacitySort) Swap(i, j int) {
	s.ids[i], s.ids[j] = s.ids[j], s.ids[i]
	s.caps[i], s.caps[j] = s.caps[j], s.caps[i]
}

// healRequestSort implements the Sort interface, allowing sorting trienode
// heal requests, which is a prerequisite for merging storage-requests.
type healRequestSort struct {
	paths     []string
	hashes    []common.Hash
	syncPaths []trie.SyncPath
}

func (t *healRequestSort) Len() int {
	return len(t.hashes)
}

func (t *healRequestSort) Less(i, j int) bool {
	a := t.syncPaths[i]
	b := t.syncPaths[j]
	switch bytes.Compare(a[0], b[0]) {
	case -1:
		return true
	case 1:
		return false
	}
	// identical first part
	if len(a) < len(b) {
		return true
	}
	if len(b) < len(a) {
		return false
	}
	if len(a) == 2 {
		return bytes.Compare(a[1], b[1]) < 0
	}
	return false
}

func (t *healRequestSort) Swap(i, j int) {
	t.paths[i], t.paths[j] = t.paths[j], t.paths[i]
	t.hashes[i], t.hashes[j] = t.hashes[j], t.hashes[i]
	t.syncPaths[i], t.syncPaths[j] = t.syncPaths[j], t.syncPaths[i]
}

// Merge merges the pathsets, so that several storage requests concerning the
// same account are merged into one, to reduce bandwidth.
// OBS: This operation is moot if t has not first been sorted.
func (t *healRequestSort) Merge() []TrieNodePathSet {
	var result []TrieNodePathSet
	for _, path := range t.syncPaths {
		pathset := TrieNodePathSet(path)
		if len(path) == 1 {
			// It's an account reference.
			result = append(result, pathset)
		} else {
			// It's a storage reference.
			end := len(result) - 1
			if len(result) == 0 || !bytes.Equal(pathset[0], result[end][0]) {
				// The account doesn't match last, create a new entry.
				result = append(result, pathset)
			} else {
				// It's the same account as the previous one, add to the storage
				// paths of that request.
				result[end] = append(result[end], pathset[1])
			}
		}
	}
	return result
}

// sortByAccountPath takes hashes and paths, and sorts them. After that, it generates
// the TrieNodePaths and merges paths which belongs to the same account path.
func sortByAccountPath(paths []string, hashes []common.Hash) ([]string, []common.Hash, []trie.SyncPath, []TrieNodePathSet) {
	var syncPaths []trie.SyncPath
	for _, path := range paths {
		syncPaths = append(syncPaths, trie.NewSyncPath([]byte(path)))
	}
	n := &healRequestSort{paths, hashes, syncPaths}
	sort.Sort(n)
	pathsets := n.Merge()
	return n.paths, n.hashes, n.syncPaths, pathsets
}