trie, cmd/geth: add archive verify command, Walk(), and archiver improvements

Add Trie.Walk() for exhaustive traversal that resolves expired nodes
with hash verification. Add `archive verify` subcommand that walks
the full state (account + storage tries) to validate all archived
data can be correctly resurrected.

Delete both the journal KV entry and file after archiving to force
geth to restart with a bare disk layer, rewinding the chain head to
the persistent disk state and re-executing blocks.

Also adds markSubtreeDirty() to resolveExpiredNodeData() so that all
nodes in a resolved expired subtree are captured in the NodeSet during
commit — preventing them from being lost between diff layers and the
raw DB.

cmd/geth/archivecmd.go

@@ -17,8 +17,10 @@
 package main
 
 import (
+	"encoding/binary"
 	"errors"
 	"fmt"
+	"os"
 	"path/filepath"
 	"slices"
 	"time"
@@ -26,10 +28,14 @@ import (
 	"github.com/ethereum/go-ethereum/cmd/utils"
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
+	"github.com/ethereum/go-ethereum/core/types"
+	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/rlp"
 	"github.com/ethereum/go-ethereum/trie"
 	"github.com/ethereum/go-ethereum/trie/archive"
+	"github.com/ethereum/go-ethereum/triedb/database"
 	"github.com/urfave/cli/v2"
 )
 
@@ -51,14 +57,73 @@ var (
 	}
 
 	// Commands
+	archiveCheckNodeFlag = &cli.StringFlag{
+		Name:  "owner",
+		Usage: "Owner hash (hex) for the trie node to check",
+	}
+	archiveCheckPathFlag = &cli.StringFlag{
+		Name:  "path",
+		Usage: "Path (hex nibbles) of the trie node to check",
+	}
+
 	archiveCommand = &cli.Command{
 		Name:  "archive",
 		Usage: "Archive state trie nodes to reduce database size",
 		Subcommands: []*cli.Command{
 			archiveGenerateCmd,
+			archiveVerifyCmd,
+			archiveDeleteJournalCmd,
+			archiveCheckNodeCmd,
 		},
 	}
 
+	archiveCheckNodeCmd = &cli.Command{
+		Name:   "check-node",
+		Usage:  "Check if a specific trie node exists in the raw DB",
+		Action: archiveCheckNode,
+		Flags: slices.Concat([]cli.Flag{
+			archiveCheckNodeFlag,
+			archiveCheckPathFlag,
+		}, utils.NetworkFlags, utils.DatabaseFlags),
+	}
+
+	archiveDeleteJournalCmd = &cli.Command{
+		Name:   "delete-journal",
+		Usage:  "Delete the pathdb journal to force a clean restart",
+		Action: archiveDeleteJournal,
+		Flags:  slices.Concat(utils.NetworkFlags, utils.DatabaseFlags),
+		Description: `
+Deletes the pathdb journal (TrieJournal key and merkle.journal file) from the
+database. This forces geth to restart with a bare disk layer, discarding any
+in-memory diff layers that may be inconsistent with archived state.
+
+Use this after running 'archive generate' if geth was started in between and
+recreated the journal.
+
+Examples:
+  geth archive delete-journal --datadir /path/to/datadir
+  geth archive delete-journal --hoodi
+`,
+	}
+
+	archiveVerifyCmd = &cli.Command{
+		Name:   "verify",
+		Usage:  "Verify all archived nodes can be correctly resurrected",
+		Action: archiveVerify,
+		Flags:  slices.Concat(utils.NetworkFlags, utils.DatabaseFlags),
+		Description: `
+Walks the entire state trie, resolving every expired node from the archive
+file and verifying that the reconstructed subtree hash matches the original.
+Also walks all storage tries referenced by accounts.
+
+The database is opened read-only. No modifications are made.
+
+Examples:
+  geth archive verify --datadir /path/to/datadir
+  geth archive verify --hoodi
+`,
+	}
+
 	archiveGenerateCmd = &cli.Command{
 		Name:      "generate",
 		Usage:     "Generate archive files from height-3 subtrees",
@@ -77,22 +142,56 @@ that references the archive file offset and size.
 Height is measured from leaves: leaves=0, parents=1, etc. A height-3 node
 has leaves at most 3 levels below it.
 
+The archiver reads trie nodes directly from the persistent database layer,
+bypassing any in-memory diff layers. This ensures consistency between the
+data it reads and the data it modifies.
+
 Examples:
-  # Archive from head state
+  # Archive from the persistent disk state
   geth archive generate --datadir /path/to/datadir
 
   # Dry run to see what would be archived
   geth archive generate --dry-run --datadir /path/to/datadir
 
-  # Archive from a specific state root
-  geth archive generate 0x1234...abcd --datadir /path/to/datadir
-
   # Custom output and compaction interval
   geth archive generate --output /path/to/archive --compaction-interval 500
 `,
 	}
 )
 
+// rawDBNodeReader implements database.NodeReader by reading trie nodes directly
+// from the raw key-value database, bypassing pathdb's in-memory diff layers.
+// This ensures the archiver sees the same trie state it modifies.
+type rawDBNodeReader struct {
+	db ethdb.KeyValueReader
+}
+
+func (r *rawDBNodeReader) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
+	var blob []byte
+	if owner == (common.Hash{}) {
+		blob = rawdb.ReadAccountTrieNode(r.db, path)
+	} else {
+		blob = rawdb.ReadStorageTrieNode(r.db, owner, path)
+	}
+	// Skip hash verification: the raw DB may contain expiredNode markers
+	// (blob[0] == 0x00) which have different hashes than the original nodes.
+	return blob, nil
+}
+
+// rawDBNodeDatabase implements database.NodeDatabase using direct raw DB reads.
+type rawDBNodeDatabase struct {
+	db   ethdb.KeyValueReader
+	root common.Hash
+}
+
+func (d *rawDBNodeDatabase) NodeReader(stateRoot common.Hash) (database.NodeReader, error) {
+	// Only allow reading the persistent disk root state
+	if stateRoot != d.root {
+		return nil, fmt.Errorf("raw DB reader only supports disk root %x, got %x", d.root, stateRoot)
+	}
+	return &rawDBNodeReader{db: d.db}, nil
+}
+
 func archiveGenerate(ctx *cli.Context) error {
 	// 1. Setup node and databases
 	stack, _ := makeConfigNode(ctx)
@@ -109,27 +208,25 @@ func archiveGenerate(ctx *cli.Context) error {
 		return fmt.Errorf("archive generation requires path-based state scheme, got: %s", scheme)
 	}
 
-	triedb := utils.MakeTrieDatabase(ctx, stack, chaindb, false, false, false)
+	// 2. Determine the persistent disk state root.
-	defer triedb.Close()
+	//
-
+	// The archiver reads and writes directly to the raw key-value database,
-	// 2. Determine state root
+	// bypassing pathdb's in-memory diff layers. This avoids the inconsistency
-	var root common.Hash
+	// where diff layers shadow expiredNode markers written to disk.
-	if ctx.NArg() > 0 {
+	//
-		root = common.HexToHash(ctx.Args().First())
+	// The disk root is computed by hashing the account trie root node stored
-		log.Info("Using specified state root", "root", root)
+	// in the raw database. This root corresponds to the last state that was
-	} else {
+	// fully persisted (i.e., PersistentStateID), which matches the canonical
-		headBlock := rawdb.ReadHeadBlock(chaindb)
+	// chain head.
-		if headBlock == nil {
+	rootBlob := rawdb.ReadAccountTrieNode(chaindb, nil)
-			return errors.New("no head block found - specify a state root or sync the chain first")
+	if len(rootBlob) == 0 {
-		}
-		root = headBlock.Root()
-		log.Info("Using head block state", "number", headBlock.NumberU64(), "root", root)
-	}
-
-	// Verify the state exists
-	if !rawdb.HasAccountTrieNode(chaindb, nil) {
 		return errors.New("state trie not found in database")
 	}
+	root := crypto.Keccak256Hash(rootBlob)
+	log.Info("Using persistent disk state root", "root", root)
+
+	// Create a raw DB node reader that bypasses pathdb layers
+	nodeDB := &rawDBNodeDatabase{db: chaindb, root: root}
 
 	// 3. Open archive writer (unless dry-run)
 	var writer *archive.ArchiveWriter
@@ -153,7 +250,7 @@ func archiveGenerate(ctx *cli.Context) error {
 	// 4. Create and run archiver
 	archiver := trie.NewArchiver(
 		chaindb,
-		triedb,
+		nodeDB,
 		writer,
 		ctx.Uint64(archiveCompactionIntervalFlag.Name),
 		dryRun,
@@ -174,6 +271,27 @@ func archiveGenerate(ctx *cli.Context) error {
 		}
 	}
 
+	if !dryRun {
+		// Delete the pathdb journal. The archiver modified the raw DB
+		// underneath the diff layers, so the journal's buffered state is
+		// inconsistent. Deleting forces geth to restart with a bare disk
+		// layer and rewind the chain head to the disk state.
+		if err := chaindb.Delete([]byte("TrieJournal")); err != nil {
+			log.Warn("Failed to delete pathdb journal key", "err", err)
+		}
+		log.Info("Deleted pathdb journal to force clean restart")
+
+		// Delete journal file(s) - check both legacy and current locations
+		for _, dir := range []string{"triedb", ""} {
+			for _, name := range []string{"merkle.journal", "verkle.journal"} {
+				journalFile := filepath.Join(stack.ResolvePath(dir), name)
+				if err := os.Remove(journalFile); err == nil {
+					log.Info("Deleted journal file", "path", journalFile)
+				}
+			}
+		}
+	}
+
 	// 6. Print summary
 	var archiveSize uint64
 	if writer != nil {
@@ -194,6 +312,258 @@ func archiveGenerate(ctx *cli.Context) error {
 	return nil
 }
 
+func archiveVerify(ctx *cli.Context) error {
+	stack, _ := makeConfigNode(ctx)
+	defer stack.Close()
+
+	// Open database read-only
+	chaindb := utils.MakeChainDatabase(ctx, stack, true)
+	defer chaindb.Close()
+
+	scheme := cycleCheckScheme(ctx, chaindb)
+	if scheme != rawdb.PathScheme {
+		return fmt.Errorf("archive verify requires path-based state scheme, got: %s", scheme)
+	}
+
+	// Set archive data dir so ArchivedNodeResolver can find the file
+	// ResolvePath("") returns the node's data directory (e.g. .ethereum/hoodi/geth),
+	// but ArchivedNodeResolver expects the instance directory (.ethereum/hoodi)
+	// since it appends "geth/nodearchive" itself.
+	archive.ArchiveDataDir = filepath.Dir(stack.ResolvePath(""))
+
+	// Compute disk root
+	rootBlob := rawdb.ReadAccountTrieNode(chaindb, nil)
+	if len(rootBlob) == 0 {
+		return errors.New("state trie not found in database")
+	}
+	root := crypto.Keccak256Hash(rootBlob)
+	log.Info("Verifying archived nodes", "root", root)
+
+	nodeDB := &rawDBNodeDatabase{db: chaindb, root: root}
+
+	// Open account trie
+	accountTrie, err := trie.New(trie.StateTrieID(root), nodeDB)
+	if err != nil {
+		return fmt.Errorf("failed to open account trie: %w", err)
+	}
+
+	var (
+		totalAccounts     int
+		totalStorageTries int
+		totalLeaves       int
+		totalExpired      int
+		totalErrors       int
+		start             = time.Now()
+		lastLog           = time.Now()
+	)
+
+	// Walk the account trie — this resolves all expired nodes and verifies hashes
+	accountStats, err := accountTrie.Walk(func(path []byte, value []byte) error {
+		totalAccounts++
+		if time.Since(lastLog) > 30*time.Second {
+			log.Info("Verification progress",
+				"accounts", totalAccounts,
+				"storageTries", totalStorageTries,
+				"leaves", totalLeaves,
+				"expired", totalExpired,
+				"errors", totalErrors)
+			lastLog = time.Now()
+		}
+
+		// Decode account to check for storage trie
+		var acc types.StateAccount
+		if err := rlp.DecodeBytes(value, &acc); err != nil {
+			log.Warn("Failed to decode account", "err", err)
+			totalErrors++
+			return nil // continue walking
+		}
+		if acc.Root == types.EmptyRootHash {
+			return nil
+		}
+
+		// Open and walk storage trie.
+		// path is hex-nibble encoded (with a 16 terminator from the trie key),
+		// so convert nibble pairs back to the 32-byte account hash.
+		nibbles := path
+		if len(nibbles) > 0 && nibbles[len(nibbles)-1] == 16 {
+			nibbles = nibbles[:len(nibbles)-1]
+		}
+		keyBytes := make([]byte, len(nibbles)/2)
+		for i := 0; i < len(nibbles); i += 2 {
+			keyBytes[i/2] = nibbles[i]<<4 | nibbles[i+1]
+		}
+		accountHash := common.BytesToHash(keyBytes)
+		storageID := trie.StorageTrieID(root, accountHash, acc.Root)
+		storageTrie, err := trie.New(storageID, nodeDB)
+		if err != nil {
+			log.Warn("Failed to open storage trie", "account", accountHash, "err", err)
+			totalErrors++
+			return nil
+		}
+
+		storageStats, err := storageTrie.Walk(func(spath []byte, svalue []byte) error {
+			return nil
+		})
+		if err != nil {
+			log.Warn("Storage trie walk failed", "account", accountHash, "err", err)
+			totalErrors++
+			return nil
+		}
+		totalStorageTries++
+		totalLeaves += storageStats.Leaves
+		totalExpired += storageStats.ExpiredResolved
+		return nil
+	})
+	if err != nil {
+		return fmt.Errorf("account trie walk failed: %w", err)
+	}
+
+	totalLeaves += accountStats.Leaves
+	totalExpired += accountStats.ExpiredResolved
+
+	log.Info("Archive verification complete",
+		"accounts", totalAccounts,
+		"storageTries", totalStorageTries,
+		"totalLeaves", totalLeaves,
+		"expiredResolved", totalExpired,
+		"errors", totalErrors,
+		"elapsed", common.PrettyDuration(time.Since(start)))
+
+	if totalErrors > 0 {
+		return fmt.Errorf("verification completed with %d errors", totalErrors)
+	}
+	return nil
+}
+
+func archiveDeleteJournal(ctx *cli.Context) error {
+	stack, _ := makeConfigNode(ctx)
+	defer stack.Close()
+
+	chaindb := utils.MakeChainDatabase(ctx, stack, false)
+	defer chaindb.Close()
+
+	// Delete the pathdb journal KV key
+	if err := chaindb.Delete([]byte("TrieJournal")); err != nil {
+		log.Warn("Failed to delete pathdb journal key", "err", err)
+	} else {
+		log.Info("Deleted pathdb journal key (TrieJournal)")
+	}
+
+	// Delete the journal file(s) - check both legacy and current locations
+	for _, dir := range []string{"triedb", ""} {
+		for _, name := range []string{"merkle.journal", "verkle.journal"} {
+			journalFile := filepath.Join(stack.ResolvePath(dir), name)
+			if err := os.Remove(journalFile); err == nil {
+				log.Info("Deleted journal file", "path", journalFile)
+			} else if !os.IsNotExist(err) {
+				log.Warn("Failed to delete journal file", "path", journalFile, "err", err)
+			}
+		}
+	}
+
+	return nil
+}
+
+func archiveCheckNode(ctx *cli.Context) error {
+	stack, _ := makeConfigNode(ctx)
+	defer stack.Close()
+
+	chaindb := utils.MakeChainDatabase(ctx, stack, true)
+	defer chaindb.Close()
+
+	ownerHex := ctx.String(archiveCheckNodeFlag.Name)
+	pathHex := ctx.String(archiveCheckPathFlag.Name)
+
+	if ownerHex == "" {
+		return errors.New("--owner flag is required")
+	}
+
+	owner := common.HexToHash(ownerHex)
+
+	// Parse path: hex nibbles like "08" → []byte{0, 8}
+	var path []byte
+	for _, c := range pathHex {
+		var nibble byte
+		switch {
+		case c >= '0' && c <= '9':
+			nibble = byte(c - '0')
+		case c >= 'a' && c <= 'f':
+			nibble = byte(c-'a') + 10
+		case c >= 'A' && c <= 'F':
+			nibble = byte(c-'A') + 10
+		default:
+			return fmt.Errorf("invalid hex char in path: %c", c)
+		}
+		path = append(path, nibble)
+	}
+
+	log.Info("Checking node in raw DB", "owner", owner, "path", fmt.Sprintf("%x", path))
+
+	// Read the node directly from the raw DB
+	isAccount := owner == (common.Hash{})
+
+	// Check the target path and all prefixes up to root
+	for i := len(path); i >= 0; i-- {
+		subpath := path[:i]
+		var blob []byte
+		if isAccount {
+			blob = rawdb.ReadAccountTrieNode(chaindb, subpath)
+		} else {
+			blob = rawdb.ReadStorageTrieNode(chaindb, owner, subpath)
+		}
+
+		status := "MISSING"
+		details := ""
+		if len(blob) > 0 {
+			if blob[0] == 0x00 {
+				status = "EXPIRED"
+				if len(blob) == 17 {
+					offset := binary.BigEndian.Uint64(blob[1:9])
+					size := binary.BigEndian.Uint64(blob[9:17])
+					details = fmt.Sprintf("offset=%d size=%d", offset, size)
+				}
+			} else {
+				status = fmt.Sprintf("PRESENT (%d bytes, first=0x%02x)", len(blob), blob[0])
+			}
+		}
+		label := "prefix"
+		if i == len(path) {
+			label = "TARGET"
+		}
+		if i == 0 {
+			label = "ROOT"
+		}
+		log.Info("Node check",
+			"label", label,
+			"path", fmt.Sprintf("%x", subpath),
+			"pathLen", i,
+			"status", status,
+			"details", details)
+	}
+
+	// Also check a few child paths to see what's below the target
+	for nibble := byte(0); nibble < 16; nibble++ {
+		childPath := append(append([]byte{}, path...), nibble)
+		var blob []byte
+		if isAccount {
+			blob = rawdb.ReadAccountTrieNode(chaindb, childPath)
+		} else {
+			blob = rawdb.ReadStorageTrieNode(chaindb, owner, childPath)
+		}
+		if len(blob) > 0 {
+			status := fmt.Sprintf("PRESENT (%d bytes, first=0x%02x)", len(blob), blob[0])
+			if blob[0] == 0x00 && len(blob) == 17 {
+				offset := binary.BigEndian.Uint64(blob[1:9])
+				size := binary.BigEndian.Uint64(blob[9:17])
+				status = fmt.Sprintf("EXPIRED offset=%d size=%d", offset, size)
+			}
+			log.Info("Child node", "path", fmt.Sprintf("%x", childPath), "status", status)
+		}
+	}
+
+	return nil
+}
+
 // cycleCheckScheme returns the state scheme for the database.
 // It's a helper to check what scheme is in use.
 func cycleCheckScheme(ctx *cli.Context, db ethdb.Database) string {

node/node.go

@@ -38,6 +38,7 @@ import (
 	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/p2p"
 	"github.com/ethereum/go-ethereum/rpc"
+	"github.com/ethereum/go-ethereum/trie/archive"
 	"github.com/gofrs/flock"
 )
 
@@ -85,6 +86,7 @@ func New(conf *Config) (*Node, error) {
 			return nil, err
 		}
 		conf.DataDir = absdatadir
+		archive.ArchiveDataDir = absdatadir
 	}
 	if conf.Logger == nil {
 		conf.Logger = log.New()

trie/archive/archive.go

@@ -22,6 +22,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"path/filepath"
 
 	"github.com/ethereum/go-ethereum/rlp"
 )
@@ -45,13 +46,16 @@ type Record struct {
 	Value []byte
 }
 
+// ArchiveDataDir is the data directory where the archive file is stored.
+var ArchiveDataDir string
+
 // ArchivedNodeResolver takes a buffer containing the archive data
 // held by an expiring node (an offset and a size) and returns a
 // list of records, which is a list of serialized leaf nodes. The
 // caller knows the context (MPT, binary trie) and is responsible
 // for decoding the nodes.
 func ArchivedNodeResolver(offset, size uint64) ([]*Record, error) {
-	file, err := os.Open("nodearchive")
+	file, err := os.Open(filepath.Join(ArchiveDataDir, "geth", "nodearchive"))
 	if err != nil {
 		return nil, fmt.Errorf("error opening archive file: %w", err)
 	}

trie/archiver.go

@@ -19,6 +19,7 @@ package trie
 import (
 	"encoding/binary"
 	"fmt"
+	"time"
 
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core/rawdb"
@@ -39,6 +40,7 @@ type subtreeInfo struct {
 	height    int               // Height of subtree (from leaves)
 	leaves    []*archive.Record // All leaf records (relative path + encoded node)
 	nodePaths [][]byte          // Paths of all nodes to delete
+	rootHash  common.Hash       // Hash of the original subtree root (for verification)
 }
 
 // Archiver handles the archival process of trie nodes.
@@ -79,7 +81,7 @@ func NewArchiver(db ethdb.Database, triedb database.NodeDatabase,
 }
 
 // ProcessState archives subtrees from the given state root.
-// It processes the account trie first, then all storage tries.
+// It processes storage tries first, then the account trie.
 func (a *Archiver) ProcessState(root common.Hash) error {
 	a.stateRoot = root
 
@@ -141,7 +143,12 @@ func (a *Archiver) processTrie(owner common.Hash, t *Trie) error {
 	subtrees := a.findHeight3Subtrees(t.root, nil, owner)
 	log.Info("Found subtrees to archive", "owner", owner, "count", len(subtrees))
 
-	for _, info := range subtrees {
+	lastLog := time.Now()
+	for i, info := range subtrees {
+		if time.Since(lastLog) > 30*time.Second {
+			log.Info("Archiving subtrees", "owner", owner, "progress", fmt.Sprintf("%d/%d", i, len(subtrees)), "archived", a.subtreesArchived)
+			lastLog = time.Now()
+		}
 		if err := a.archiveSubtree(info); err != nil {
 			log.Warn("Failed to archive subtree", "path", common.Bytes2Hex(info.path), "err", err)
 			continue
@@ -159,18 +166,52 @@ func (a *Archiver) processTrie(owner common.Hash, t *Trie) error {
 // findHeight3Subtrees recursively finds all subtrees with height == 3.
 // Height is measured from leaves: leaves=0, their parents=1, etc.
 func (a *Archiver) findHeight3Subtrees(n node, path []byte, owner common.Hash) []*subtreeInfo {
-	info := a.computeSubtreeInfo(n, path, owner)
+	info, err := a.computeSubtreeInfo(n, path, owner)
+	if err != nil {
+		// computeSubtreeInfo failed (e.g. unresolvable hashNode within the
+		// subtree). We cannot archive this node as-is, but deeper children
+		// may still form valid height-3 subtrees. Recurse into them.
+		log.Debug("computeSubtreeInfo failed, trying children", "path", common.Bytes2Hex(path), "err", err)
+		return a.findSubtreesInChildren(n, path, owner)
+	}
 	if info == nil {
 		return nil
 	}
 
 	// If this subtree has height 3, it's a candidate for archival
 	if info.height == 3 {
+		// Capture the original subtree root hash for verification.
+		// The hash is available from the node that was passed in:
+		// - hashNode: the hash IS the node
+		// - fullNode/shortNode: loaded from DB, flags.hash is set
+		switch nn := n.(type) {
+		case hashNode:
+			info.rootHash = common.BytesToHash(nn)
+		case *fullNode:
+			if nn.flags.hash != nil {
+				info.rootHash = common.BytesToHash(nn.flags.hash)
+			}
+		case *shortNode:
+			if nn.flags.hash != nil {
+				info.rootHash = common.BytesToHash(nn.flags.hash)
+			}
+		}
 		return []*subtreeInfo{info}
 	}
 
 	// If height > 3, recurse into children to find height-3 subtrees
 	if info.height > 3 {
+		return a.findSubtreesInChildren(n, path, owner)
+	}
+
+	// Height < 3: no archivable subtrees here
+	return nil
+}
+
+// findSubtreesInChildren recurses into the children of a node to find
+// height-3 subtrees. Used both by the normal height > 3 path and as a
+// fallback when computeSubtreeInfo fails for a node.
+func (a *Archiver) findSubtreesInChildren(n node, path []byte, owner common.Hash) []*subtreeInfo {
 	var results []*subtreeInfo
 	switch n := n.(type) {
 	case *fullNode:
@@ -191,22 +232,20 @@ func (a *Archiver) findHeight3Subtrees(n node, path []byte, owner common.Hash) [
 		}
 	}
 	return results
-	}
-
-	// Height < 3: no archivable subtrees here
-	return nil
 }
 
 // computeSubtreeInfo computes height and collects leaves for a subtree.
-// Returns nil if the node is nil or an error occurs during resolution.
+// Returns (nil, nil) if the node is nil, already expired, or has no leaves.
-func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *subtreeInfo {
+// Returns (nil, error) if any constituent node could not be resolved — the
+// caller MUST NOT archive a subtree when an error is returned, as the leaf
+// set would be incomplete.
+func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) (*subtreeInfo, error) {
 	switch n := n.(type) {
 	case nil:
-		return nil
+		return nil, nil
 
 	case valueNode:
 		// Leaf: height 0
-		// Encode the leaf as a shortNode for archive storage
 		return &subtreeInfo{
 			path:   copyBytes(path),
 			owner:  owner,
@@ -216,13 +255,16 @@ func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *s
 				Value: []byte(n),
 			}},
 			nodePaths: [][]byte{copyBytes(path)},
-		}
+		}, nil
 
 	case *shortNode:
 		childPath := append(append([]byte{}, path...), n.Key...)
-		childInfo := a.computeSubtreeInfo(n.Val, childPath, owner)
+		childInfo, err := a.computeSubtreeInfo(n.Val, childPath, owner)
+		if err != nil {
+			return nil, fmt.Errorf("shortNode key=%x: %w", n.Key, err)
+		}
 		if childInfo == nil {
-			return nil
+			return nil, nil
 		}
 
 		// Adjust relative paths in leaves to include this node's key
@@ -236,7 +278,7 @@ func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *s
 			height:    childInfo.height + 1,
 			leaves:    childInfo.leaves,
 			nodePaths: append([][]byte{copyBytes(path)}, childInfo.nodePaths...),
-		}
+		}, nil
 
 	case *fullNode:
 		var (
@@ -247,7 +289,10 @@ func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *s
 		for i, child := range n.Children[:16] {
 			if child != nil {
 				childPath := append(append([]byte{}, path...), byte(i))
-				childInfo := a.computeSubtreeInfo(child, childPath, owner)
+				childInfo, err := a.computeSubtreeInfo(child, childPath, owner)
+				if err != nil {
+					return nil, fmt.Errorf("fullNode child[%x]: %w", i, err)
+				}
 				if childInfo != nil {
 					if childInfo.height+1 > maxHeight {
 						maxHeight = childInfo.height + 1
@@ -263,7 +308,7 @@ func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *s
 		}
 
 		if len(allLeaves) == 0 {
-			return nil
+			return nil, nil
 		}
 
 		return &subtreeInfo{
@@ -272,21 +317,20 @@ func (a *Archiver) computeSubtreeInfo(n node, path []byte, owner common.Hash) *s
 			height:    maxHeight,
 			leaves:    allLeaves,
 			nodePaths: allPaths,
-		}
+		}, nil
 
 	case hashNode:
 		resolved, err := a.resolveNode(n, path, owner)
 		if err != nil {
-			log.Debug("Failed to resolve hashNode", "path", common.Bytes2Hex(path), "err", err)
+			return nil, fmt.Errorf("failed to resolve hashNode at path %s: %w", common.Bytes2Hex(path), err)
-			return nil
 		}
 		return a.computeSubtreeInfo(resolved, path, owner)
 
 	case *expiredNode:
 		// Already archived, skip
-		return nil
+		return nil, nil
 	}
-	return nil
+	return nil, nil
 }
 
 // archiveSubtree writes leaves to archive and replaces subtree with expiredNode.
@@ -312,7 +356,25 @@ func (a *Archiver) archiveSubtree(info *subtreeInfo) error {
 		return fmt.Errorf("failed to sync archive: %w", err)
 	}
 
-	// 3. Batch database operations
+	// 3. Verify archive round-trip: reconstruct trie from records and
+	// check that the hash matches the original subtree root. This
+	// catches any data corruption before we delete the original nodes.
+	if info.rootHash != (common.Hash{}) {
+		reconstructed, err := archiveRecordsToNode(info.leaves)
+		if err != nil {
+			return fmt.Errorf("archive verification failed: cannot reconstruct trie from records: %w", err)
+		}
+		h := newHasher(false)
+		gotHash := common.BytesToHash(h.hash(reconstructed, true))
+		returnHasherToPool(h)
+		if gotHash != info.rootHash {
+			return fmt.Errorf("archive verification failed: hash mismatch at path %s owner %s: got %s want %s (leaves=%d offset=%d size=%d)",
+				common.Bytes2Hex(info.path), info.owner, gotHash, info.rootHash,
+				len(info.leaves), offset, size)
+		}
+	}
+
+	// 4. Batch database operations
 	batch := a.db.NewBatch()
 
 	// Delete all nodes in subtree (except the root which we'll overwrite)

trie/committer.go

@@ -79,6 +79,8 @@ func (c *committer) commit(path []byte, n node, parallel bool) node {
 		return cn
 	case hashNode:
 		return cn
+	case *expiredNode:
+		return cn
 	default:
 		// nil, valuenode shouldn't be committed
 		panic(fmt.Sprintf("%T: invalid node: %v", n, n))

trie/expired_node.go

@@ -17,6 +17,7 @@
 package trie
 
 import (
+	"bytes"
 	"encoding/binary"
 	"fmt"
 
@@ -33,11 +34,12 @@ const expiredNodeMarker = 0x00
 type expiredNode struct {
 	offset          uint64
 	size            uint64
+	cachedHash      hashNode
 	archiveResolver archive.ResolverFn
 }
 
 func (n *expiredNode) cache() (hashNode, bool) {
-	return nil, true
+	return n.cachedHash, n.cachedHash == nil
 }
 
 func (n *expiredNode) encode(w rlp.EncoderBuffer) {
@@ -62,36 +64,101 @@ func (n *expiredNode) SetArchiveResolver(resolver archive.ResolverFn) {
 	n.archiveResolver = resolver
 }
 
+// resolveExpiredNodeData resolves an expired node from the archive, verifies
+// the reconstructed subtree hash, and stamps the cached hash onto the root.
+// Returns an error if the archive data is corrupted (hash mismatch).
+func resolveExpiredNodeData(n *expiredNode) (node, error) {
+	records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+	if err != nil {
+		return nil, fmt.Errorf("failed to resolve expired node: %w", err)
+	}
+	resolved, err := archiveRecordsToNode(records)
+	if err != nil {
+		return nil, fmt.Errorf("failed to rebuild expired node from archive: %w", err)
+	}
+	// Verify hash integrity: if the original hash is known, check that the
+	// reconstructed subtree produces the same hash. A mismatch means the
+	// archive is corrupted (e.g. missing leaves due to unresolvable hashNodes
+	// during archival) and any data from it is unreliable.
+	if n.cachedHash != nil {
+		h := newHasher(false)
+		gotHash := h.hash(resolved, true)
+		returnHasherToPool(h)
+		if !bytes.Equal(gotHash, n.cachedHash) {
+			return nil, fmt.Errorf("expired node hash mismatch at offset=%d size=%d: archive data is corrupted (expected %x got %x, %d records)",
+				n.offset, n.size, []byte(n.cachedHash), gotHash, len(records))
+		}
+		// Stamp the original hash onto the resolved subtree root so the
+		// hasher returns it directly instead of re-computing.
+		switch nn := resolved.(type) {
+		case *fullNode:
+			nn.flags.hash = n.cachedHash
+		case *shortNode:
+			nn.flags.hash = n.cachedHash
+		}
+	}
+	// Mark the entire resolved subtree as dirty. This is critical for
+	// correctness with pathdb's diff layer model: when a trie with expired
+	// nodes is modified and committed, the committer only captures dirty
+	// nodes into the NodeSet (which becomes the diff layer). Without this
+	// marking, resolved-but-unmodified sibling nodes within the subtree
+	// would exist nowhere — not in any diff layer (they're clean) and not
+	// in the raw DB (the archiver deleted them). Subsequent trie accesses
+	// from higher diff layers would fall through to the disk layer, find
+	// nothing, and produce MissingNodeError.
+	//
+	// For read-only tries (only get operations, no commit), this dirty
+	// marking is harmless — the nodes are discarded when the trie is GC'd.
+	markSubtreeDirty(resolved)
+	return resolved, nil
+}
+
+// markSubtreeDirty recursively marks all fullNode and shortNode in the
+// subtree as dirty, preserving any cached hashes. This ensures the
+// committer will capture them in the NodeSet during trie commit.
+func markSubtreeDirty(n node) {
+	switch n := n.(type) {
+	case *fullNode:
+		n.flags.dirty = true
+		for _, child := range n.Children[:16] {
+			if child != nil {
+				markSubtreeDirty(child)
+			}
+		}
+	case *shortNode:
+		n.flags.dirty = true
+		markSubtreeDirty(n.Val)
+	}
+	// valueNode, hashNode, nil: no flags to mark
+}
+
 func archiveRecordsToNode(records []*archive.Record) (node, error) {
 	if len(records) == 0 {
 		return nil, archive.EmptyArchiveRecord
 	}
-	if len(records) == 1 {
-		return buildLeafFromRecord(records[0])
-	}
 
-	var newnode fullNode
+	// Build the trie incrementally from nil to produce the canonical
+	// MPT structure. Starting with a fullNode would be wrong when the
+	// original subtree root was a shortNode (shared prefix).
+	var root node
 	for i, record := range records {
 		if err := validateRecordPath(record.Path); err != nil {
 			return nil, err
 		}
 
-		// we are not in the case of a single leaf node, so each
-		// path should be at least 2 nibbles (terminator included)
-		if len(record.Path) < 2 || !hasTerm(record.Path) {
-			return nil, fmt.Errorf("invalid record path for non-leaf node #%d: %v", i, record.Path)
-		}
 		key, err := normalizeRecordKey(record.Path)
 		if err != nil {
 			return nil, err
 		}
-		child, err := insertTrieNode(newnode.Children[key[0]], key[1:], valueNode(record.Value))
+		if len(key) < 1 {
+			return nil, fmt.Errorf("empty key in record #%d", i)
+		}
+		root, err = insertTrieNode(root, key, valueNode(record.Value))
 		if err != nil {
 			return nil, err
 		}
-		newnode.Children[key[0]] = child
 	}
-	return &newnode, nil
+	return root, nil
 }
 
 func validateRecordPath(path []byte) error {
@@ -106,14 +173,6 @@ func validateRecordPath(path []byte) error {
 	return nil
 }
 
-func buildLeafFromRecord(record *archive.Record) (node, error) {
-	key, err := normalizeRecordKey(record.Path)
-	if err != nil {
-		return nil, err
-	}
-	return &shortNode{Key: key, Val: valueNode(record.Value)}, nil
-}
-
 // normalizeRecordKey ensures the record path is a hex-nibble key suitable for
 // leaf insertion by guaranteeing a single terminator nibble and preserving any
 // already-terminated path. Empty paths are normalized to a sole terminator.

trie/expired_node_test.go

@@ -19,12 +19,45 @@ package trie
 import (
 	"bytes"
 	"errors"
+	"os"
+	"path/filepath"
 	"testing"
 
 	"github.com/ethereum/go-ethereum/rlp"
 	"github.com/ethereum/go-ethereum/trie/archive"
 )
 
+// setupTestArchive creates a temporary archive directory with an archive file
+// containing the given records, and configures archive.ArchiveDataDir to point
+// to it. It returns the offset and size of the written data, and a cleanup function.
+func setupTestArchive(t *testing.T, records []*archive.Record) (offset, size uint64, cleanup func()) {
+	t.Helper()
+	tmpDir := t.TempDir()
+	gethDir := filepath.Join(tmpDir, "geth")
+	if err := os.MkdirAll(gethDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+
+	writer, err := archive.NewArchiveWriter(filepath.Join(gethDir, "nodearchive"))
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	offset, size, err = writer.WriteSubtree(records)
+	if err != nil {
+		writer.Close()
+		t.Fatal(err)
+	}
+	writer.Close()
+
+	oldDir := archive.ArchiveDataDir
+	archive.ArchiveDataDir = tmpDir
+
+	return offset, size, func() {
+		archive.ArchiveDataDir = oldDir
+	}
+}
+
 func TestExpiredNodeEncodeDecode(t *testing.T) {
 	testCases := []struct {
 		offset uint64
@@ -120,34 +153,36 @@ func TestExpiredNodeInvalidLength(t *testing.T) {
 	}
 }
 
-func TestExpiredNodeNoResolver(t *testing.T) {
+func TestExpiredNodeNoArchiveFile(t *testing.T) {
+	// When no archive file exists, Get should return an error
+	tmpDir := t.TempDir()
+	gethDir := filepath.Join(tmpDir, "geth")
+	if err := os.MkdirAll(gethDir, 0755); err != nil {
+		t.Fatal(err)
+	}
+
+	oldDir := archive.ArchiveDataDir
+	archive.ArchiveDataDir = tmpDir
+	defer func() { archive.ArchiveDataDir = oldDir }()
+
 	tr := NewEmpty(nil)
-	tr.root = &expiredNode{offset: 100}
+	tr.root = &expiredNode{offset: 100, size: 50}
 
 	_, err := tr.Get([]byte("key"))
-	if !errors.Is(err, archive.ErrNoResolver) {
+	if err == nil {
-		t.Errorf("expected archive.ErrNoResolver, got %v", err)
+		t.Error("expected error when archive file doesn't exist")
 	}
 }
 
 func TestExpiredNodeWithResolver(t *testing.T) {
+	records := []*archive.Record{
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("testvalue")},
+	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
+
 	tr := NewEmpty(nil)
-
+	tr.root = &expiredNode{offset: offset, size: size}
-	leafNode := &shortNode{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
-		Val: valueNode([]byte("testvalue")),
-	}
-	encodedLeaf := nodeToBytes(leafNode)
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		if offset == 100 {
-			return []*archive.Record{{Value: encodedLeaf}}, nil
-		}
-		return nil, errors.New("unknown offset")
-	}
-
-	tr.SetArchiveResolver(resolver)
-	tr.root = &expiredNode{offset: 100, size: uint64(len(encodedLeaf)), archiveResolver: resolver}
 
 	val, err := tr.Get([]byte{0x12})
 	if err != nil {
@@ -159,14 +194,10 @@ func TestExpiredNodeWithResolver(t *testing.T) {
 }
 
 func TestExpiredNodeCopy(t *testing.T) {
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return nil, nil
-	}
-
 	original := &expiredNode{
 		offset:          12345,
 		size:            6789,
-		archiveResolver: resolver,
+		archiveResolver: archive.ArchivedNodeResolver,
 	}
 
 	copied := copyNode(original)
@@ -201,18 +232,9 @@ func TestArchiveRecordsToNodeEmpty(t *testing.T) {
 }
 
 func TestArchiveRecordsToNodeMultiple(t *testing.T) {
-	leaf1 := &shortNode{
-		Key: hexToCompact(keybytesToHex([]byte{0x10})),
-		Val: valueNode([]byte("value1")),
-	}
-	leaf2 := &shortNode{
-		Key: hexToCompact(keybytesToHex([]byte{0x20})),
-		Val: valueNode([]byte("value2")),
-	}
-
 	records := []*archive.Record{
-		{Path: []byte{0x01}, Value: nodeToBytes(leaf1)},
+		{Path: []byte{0x01, 16}, Value: []byte("value1")},
-		{Path: []byte{0x02}, Value: nodeToBytes(leaf2)},
+		{Path: []byte{0x02, 16}, Value: []byte("value2")},
 	}
 
 	node, err := archiveRecordsToNode(records)
@@ -234,27 +256,17 @@ func TestArchiveRecordsToNodeMultiple(t *testing.T) {
 }
 
 func TestExpiredNodeGetMultipleRecords(t *testing.T) {
-	leaf1 := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact([]byte{0x02, 0x03, 0x04, 16}),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("value1")},
-		Val: valueNode([]byte("value1")),
+		{Path: []byte{0x04, 0x05, 16}, Value: []byte("value2")},
-	}
-	leaf2 := &shortNode{
-		Key: hexToCompact([]byte{0x05, 0x06, 0x07, 16}),
-		Val: valueNode([]byte("value2")),
-	}
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{0x01}, Value: nodeToBytes(leaf1)},
-			{Path: []byte{0x04}, Value: nodeToBytes(leaf2)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
-	val, err := tr.Get([]byte{0x12, 0x34})
+	val, err := tr.Get([]byte{0x12})
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
@@ -263,10 +275,9 @@ func TestExpiredNodeGetMultipleRecords(t *testing.T) {
 	}
 
 	tr2 := NewEmpty(nil)
-	tr2.SetArchiveResolver(resolver)
+	tr2.root = &expiredNode{offset: offset, size: size}
-	tr2.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
-	val2, err := tr2.Get([]byte{0x45, 0x67})
+	val2, err := tr2.Get([]byte{0x45})
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
@@ -276,20 +287,14 @@ func TestExpiredNodeGetMultipleRecords(t *testing.T) {
 }
 
 func TestExpiredNodeGetKeyNotFound(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("value1")},
-		Val: valueNode([]byte("value1")),
-	}
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{0x01}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	val, err := tr.Get([]byte{0xff, 0xff})
 	if err != nil {
@@ -301,20 +306,14 @@ func TestExpiredNodeGetKeyNotFound(t *testing.T) {
 }
 
 func TestExpiredNodeGetPathMismatch(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("testvalue")},
-		Val: valueNode([]byte("testvalue")),
-	}
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{0x01}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	val, err := tr.Get([]byte{0x19})
 	if err != nil {
@@ -326,20 +325,14 @@ func TestExpiredNodeGetPathMismatch(t *testing.T) {
 }
 
 func TestExpiredNodeInsert(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("existing")},
-		Val: valueNode([]byte("existing")),
-	}
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	err := tr.Update([]byte{0x45}, []byte("newvalue"))
 	if err != nil {
@@ -356,20 +349,14 @@ func TestExpiredNodeInsert(t *testing.T) {
 }
 
 func TestExpiredNodeUpdate(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("oldvalue")},
-		Val: valueNode([]byte("oldvalue")),
-	}
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	err := tr.Update([]byte{0x12}, []byte("newvalue"))
 	if err != nil {
@@ -386,28 +373,15 @@ func TestExpiredNodeUpdate(t *testing.T) {
 }
 
 func TestExpiredNodeDelete(t *testing.T) {
-	leaf1 := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact([]byte{0x02, 16}),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("value1")},
-		Val: valueNode([]byte("value1")),
+		{Path: []byte{0x04, 0x05, 16}, Value: []byte("value2")},
-	}
-	leaf2 := &shortNode{
-		Key: hexToCompact([]byte{0x05, 16}),
-		Val: valueNode([]byte("value2")),
-	}
-
-	branch := &fullNode{}
-	branch.Children[0x01] = leaf1
-	branch.Children[0x04] = leaf2
-
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		return []*archive.Record{
-			{Path: []byte{}, Value: nodeToBytes(branch)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	err := tr.Delete([]byte{0x12})
 	if err != nil {
@@ -432,22 +406,14 @@ func TestExpiredNodeDelete(t *testing.T) {
 }
 
 func TestTrieCopyPreservesArchiveResolver(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("testvalue")},
-		Val: valueNode([]byte("testvalue")),
-	}
-
-	resolverCalled := false
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		resolverCalled = true
-		return []*archive.Record{
-			{Path: []byte{}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
 
 	trCopy := tr.Copy()
 
@@ -455,36 +421,180 @@ func TestTrieCopyPreservesArchiveResolver(t *testing.T) {
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
-	if !resolverCalled {
-		t.Error("resolver was not called on copied trie")
-	}
 	if string(val) != "testvalue" {
 		t.Errorf("value mismatch: got %q, want %q", val, "testvalue")
 	}
 }
 
-func TestExpiredNodeGetNode(t *testing.T) {
+func TestWalkWithExpiredNodes(t *testing.T) {
-	leaf := &shortNode{
+	records := []*archive.Record{
-		Key: hexToCompact(keybytesToHex([]byte{0x12})),
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("value1")},
-		Val: valueNode([]byte("testvalue")),
+		{Path: []byte{0x04, 0x05, 16}, Value: []byte("value2")},
-	}
+		{Path: []byte{0x07, 0x08, 16}, Value: []byte("value3")},
-
-	resolverCalled := false
-	resolver := func(offset, size uint64) ([]*archive.Record, error) {
-		resolverCalled = true
-		return []*archive.Record{
-			{Path: []byte{}, Value: nodeToBytes(leaf)},
-		}, nil
 	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
 
 	tr := NewEmpty(nil)
-	tr.SetArchiveResolver(resolver)
+	tr.root = &expiredNode{offset: offset, size: size}
-	tr.root = &expiredNode{offset: 100, size: 200, archiveResolver: resolver}
+
+	var leaves []string
+	stats, err := tr.Walk(func(path []byte, value []byte) error {
+		leaves = append(leaves, string(value))
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("Walk failed: %v", err)
+	}
+	if stats.Leaves != 3 {
+		t.Errorf("expected 3 leaves, got %d", stats.Leaves)
+	}
+	if stats.ExpiredResolved != 1 {
+		t.Errorf("expected 1 expired resolved, got %d", stats.ExpiredResolved)
+	}
+	// Verify all values were visited
+	expected := map[string]bool{"value1": true, "value2": true, "value3": true}
+	for _, leaf := range leaves {
+		if !expected[leaf] {
+			t.Errorf("unexpected leaf value: %q", leaf)
+		}
+		delete(expected, leaf)
+	}
+	if len(expected) > 0 {
+		t.Errorf("missing leaves: %v", expected)
+	}
+}
+
+func TestWalkEmptyTrie(t *testing.T) {
+	tr := NewEmpty(nil)
+	stats, err := tr.Walk(func(path []byte, value []byte) error {
+		t.Error("callback should not be called for empty trie")
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("Walk failed: %v", err)
+	}
+	if stats.Leaves != 0 || stats.ExpiredResolved != 0 {
+		t.Errorf("expected zero stats for empty trie, got leaves=%d expired=%d", stats.Leaves, stats.ExpiredResolved)
+	}
+}
+
+func TestWalkCallbackError(t *testing.T) {
+	records := []*archive.Record{
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("value1")},
+	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
+
+	tr := NewEmpty(nil)
+	tr.root = &expiredNode{offset: offset, size: size}
+
+	testErr := errors.New("test error")
+	_, err := tr.Walk(func(path []byte, value []byte) error {
+		return testErr
+	})
+	if !errors.Is(err, testErr) {
+		t.Fatalf("expected test error, got %v", err)
+	}
+}
+
+// TestExpiredNodeResolvedSubtreeDirty verifies that when an expired node is
+// resolved and a sibling leaf is modified, the commit captures ALL resolved
+// nodes (not just the modified path). Without this fix, resolved-but-unmodified
+// nodes would be lost: not in the diff layer (clean) and not in the raw DB
+// (deleted by archiver).
+func TestExpiredNodeResolvedSubtreeDirty(t *testing.T) {
+	// Use large values (>32 bytes) so leaf nodes are NOT embedded in
+	// their parent. This matches production storage tries where
+	// intermediate nodes are large enough to be stored independently.
+	bigVal1 := bytes.Repeat([]byte("A"), 40)
+	bigVal2 := bytes.Repeat([]byte("B"), 40)
+
+	// Create an archive with records under different branches.
+	records := []*archive.Record{
+		{Path: []byte{0x01, 0x02, 16}, Value: bigVal1},
+		{Path: []byte{0x04, 0x05, 16}, Value: bigVal2},
+	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
+
+	tr := NewEmpty(nil)
+	tr.root = &expiredNode{offset: offset, size: size}
+
+	// Insert a value that goes through one branch of the resolved subtree.
+	// This modifies path [1, ...] but leaves path [4, ...] unmodified.
+	if err := tr.Update([]byte{0x12}, bytes.Repeat([]byte("C"), 40)); err != nil {
+		t.Fatalf("Update failed: %v", err)
+	}
+
+	// Commit the trie. The NodeSet should be non-nil because we modified data.
+	_, nodes := tr.Commit(false)
+	if nodes == nil {
+		t.Fatal("expected non-nil NodeSet after modifying expired subtree")
+	}
+
+	// The resolved-but-unmodified sibling (path [4, 5]) should also be
+	// captured in the NodeSet, because markSubtreeDirty ensures all resolved
+	// nodes are dirty. Count the nodes to verify.
+	nodeCount := len(nodes.Nodes)
+	// We expect at least 3 nodes: the root, the modified branch, and the
+	// sibling branch. The exact count depends on trie structure.
+	if nodeCount < 3 {
+		t.Errorf("expected at least 3 nodes in NodeSet (root + modified + sibling), got %d", nodeCount)
+	}
+}
+
+// TestMarkSubtreeDirty verifies that markSubtreeDirty correctly sets the dirty
+// flag on all nodes in a subtree while preserving cached hashes.
+func TestMarkSubtreeDirty(t *testing.T) {
+	// Build a small trie structure
+	leaf1 := &shortNode{Key: []byte{1, 16}, Val: valueNode("v1")}
+	leaf2 := &shortNode{Key: []byte{2, 16}, Val: valueNode("v2")}
+	branch := &fullNode{}
+	branch.Children[1] = leaf1
+	branch.Children[2] = leaf2
+
+	// Set hash but not dirty (as if loaded from DB)
+	branch.flags = nodeFlag{hash: hashNode("testhash"), dirty: false}
+	leaf1.flags = nodeFlag{hash: hashNode("hash1"), dirty: false}
+	leaf2.flags = nodeFlag{hash: hashNode("hash2"), dirty: false}
+
+	markSubtreeDirty(branch)
+
+	// All nodes should be dirty
+	if !branch.flags.dirty {
+		t.Error("branch should be dirty")
+	}
+	if !leaf1.flags.dirty {
+		t.Error("leaf1 should be dirty")
+	}
+	if !leaf2.flags.dirty {
+		t.Error("leaf2 should be dirty")
+	}
+
+	// Hashes should be preserved
+	if !bytes.Equal(branch.flags.hash, hashNode("testhash")) {
+		t.Error("branch hash should be preserved")
+	}
+	if !bytes.Equal(leaf1.flags.hash, hashNode("hash1")) {
+		t.Error("leaf1 hash should be preserved")
+	}
+	if !bytes.Equal(leaf2.flags.hash, hashNode("hash2")) {
+		t.Error("leaf2 hash should be preserved")
+	}
+}
+
+func TestExpiredNodeGetNode(t *testing.T) {
+	records := []*archive.Record{
+		{Path: []byte{0x01, 0x02, 16}, Value: []byte("testvalue")},
+	}
+	offset, size, cleanup := setupTestArchive(t, records)
+	defer cleanup()
+
+	tr := NewEmpty(nil)
+	tr.root = &expiredNode{offset: offset, size: size}
 
 	_, _, err := tr.GetNode(hexToCompact([]byte{0x01, 0x02}))
-	if !resolverCalled {
-		t.Error("resolver was not called during GetNode")
-	}
 	if err != nil && err.Error() != "non-consensus node" {
 		t.Fatalf("unexpected error: %v", err)
 	}

trie/hasher.go

@@ -18,6 +18,7 @@ package trie
 
 import (
 	"bytes"
+	"encoding/binary"
 	"fmt"
 	"sync"
 
@@ -97,6 +98,22 @@ func (h *hasher) hash(n node, force bool) []byte {
 		// hash nodes don't have children, so they're left as were
 		return n
 
+	case *expiredNode:
+		// Return the original subtree hash that was cached when the
+		// expired node was decoded. The parent node references this
+		// hash, so we must return the same value to keep the Merkle
+		// root consistent.
+		if n.cachedHash != nil {
+			return n.cachedHash
+		}
+		// Fallback: hash the marker blob (should not happen in practice
+		// because decodeNodeUnsafe always provides the hash).
+		var buf [1 + 2*8]byte // 17 bytes
+		buf[0] = expiredNodeMarker
+		binary.BigEndian.PutUint64(buf[1:], n.offset)
+		binary.BigEndian.PutUint64(buf[9:], n.size)
+		return h.hashData(buf[:])
+
 	default:
 		panic(fmt.Errorf("unexpected node type, %T", n))
 	}
@@ -214,6 +231,12 @@ func (h *hasher) proofHash(original node) []byte {
 		return bytes.Clone(h.encodeShortNode(n))
 	case *fullNode:
 		return bytes.Clone(h.encodeFullNode(n))
+	case *expiredNode:
+		var buf [1 + 2*8]byte
+		buf[0] = expiredNodeMarker
+		binary.BigEndian.PutUint64(buf[1:], n.offset)
+		binary.BigEndian.PutUint64(buf[9:], n.size)
+		return buf[:]
 	default:
 		panic(fmt.Errorf("unexpected node type, %T", original))
 	}

trie/node.go

@@ -25,6 +25,7 @@ import (
 
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/crypto"
+	"github.com/ethereum/go-ethereum/log"
 	"github.com/ethereum/go-ethereum/rlp"
 	"github.com/ethereum/go-ethereum/trie/archive"
 )
@@ -166,7 +167,8 @@ func decodeNodeUnsafe(hash, buf []byte) (node, error) {
 		}
 		offset := binary.BigEndian.Uint64(buf[1:])
 		size := binary.BigEndian.Uint64(buf[1+archive.OffsetSize:])
-		return &expiredNode{offset: offset, size: size, archiveResolver: archive.ArchivedNodeResolver}, nil
+		log.Debug("Decoded expired node", "offset", offset, "size", size, "hash", common.BytesToHash(hash))
+		return &expiredNode{offset: offset, size: size, cachedHash: hashNode(hash), archiveResolver: archive.ArchivedNodeResolver}, nil
 	}
 	elems, _, err := rlp.SplitList(buf)
 	if err != nil {

trie/proof.go

@@ -25,6 +25,7 @@ import (
 	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/trie/archive"
 )
 
 // Prove constructs a merkle proof for key. The result contains all encoded nodes
@@ -78,6 +79,16 @@ func (t *Trie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
 			// clean cache or the database, they are all in their own
 			// copy and safe to use unsafe decoder.
 			tn = mustDecodeNodeUnsafe(n, blob)
+		case *expiredNode:
+			records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+			if err != nil {
+				return fmt.Errorf("failed to resolve expired node in proof: %w", err)
+			}
+			resolved, err := archiveRecordsToNode(records)
+			if err != nil {
+				return fmt.Errorf("failed to rebuild expired node in proof: %w", err)
+			}
+			tn = resolved
 		default:
 			panic(fmt.Sprintf("%T: invalid node: %v", tn, tn))
 		}
@@ -617,6 +628,8 @@ func get(tn node, key []byte, skipResolved bool) ([]byte, node) {
 			}
 		case hashNode:
 			return key, n
+		case *expiredNode:
+			return key, n
 		case nil:
 			return key, nil
 		case valueNode:

trie/trie.go

@@ -230,32 +230,12 @@ func (t *Trie) get(origNode node, key []byte, pos int) (value []byte, newnode no
 		value, newnode, _, err := t.get(child, key, pos)
 		return value, newnode, true, err
 	case *expiredNode:
-		records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+		log.Debug("Resolving expired node in get()", "owner", t.owner, "offset", n.offset, "size", n.size, "pos", pos)
-		if err != nil {
+		newnode, err := resolveExpiredNodeData(n)
-			return nil, n, false, fmt.Errorf("failed to resolve expired node: %w", err)
-		}
-		newnode, err := archiveRecordsToNode(records)
-		// alternative: don't rebuild, just find the value
-		// for _, record := range records {
-		// 	// make sure that the path up to the node matches
-		// 	if bytes.HasPrefix(key[pos:], record.Path) {
-		// 		resolved, err := decodeNodeUnsafe(nil, record.Value)
-		// 		if err != nil {
-		// 			fmt.Printf("%v %x\n", record.Path, record.Value)
-		// 			return nil, n, false, fmt.Errorf("failed to deserialize RLP node: %w", err)
-		// 		}
-		// 		if leaf, ok := resolved.(*shortNode); ok {
-		// 			// make sure that the key to the leaf also matches
-		// 			if bytes.Equal(key[pos+len(record.Path):], leaf.Key) {
-		// 				return leaf.Val.(valueNode), newnode, true, nil
-		// 			}
-		// 		}
-		// 	}
-		// }
 		if err != nil {
 			return nil, n, false, err
 		}
-		value, _, _, err = t.get(newnode, key, pos+1)
+		value, _, _, err = t.get(newnode, key, pos)
 		return value, newnode, true, err
 	default:
 		panic(fmt.Sprintf("%T: invalid node: %v", origNode, origNode))
@@ -392,12 +372,11 @@ func (t *Trie) getNode(origNode node, path []byte, pos int) (item []byte, newnod
 		return item, newnode, resolved + 1, err
 
 	case *expiredNode:
-		records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+		rn, err := resolveExpiredNodeData(n)
 		if err != nil {
-			return nil, n, 0, fmt.Errorf("failed to resolve expired node: %w", err)
+			return nil, n, 0, err
 		}
-		newnode, err := archiveRecordsToNode(records)
+		item, newnode, resolvedCount, err := t.getNode(rn, path, pos)
-		item, newnode, resolvedCount, err := t.getNode(newnode, path, pos)
 		return item, newnode, resolvedCount + 1, err
 
 	default:
@@ -524,16 +503,16 @@ func (t *Trie) insert(n node, prefix, key []byte, value node) (bool, node, error
 		return true, nn, nil
 
 	case *expiredNode:
-		records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+		log.Info("Resolving expired node in insert()", "owner", t.owner, "offset", n.offset, "size", n.size)
+		rn, err := resolveExpiredNodeData(n)
 		if err != nil {
-			return false, nil, fmt.Errorf("failed to resolve expired node: %w", err)
+			return false, nil, err
 		}
-		nn, err := archiveRecordsToNode(records)
+		dirty, nn, err := t.insert(rn, prefix, key, value)
-		if err != nil {
+		if !dirty || err != nil {
-			return false, nil, fmt.Errorf("failed to rebuild expired node from archive: %w", err)
+			return false, rn, err
 		}
-		dirty, nn, err := t.insert(nn, prefix, key, value)
+		return true, nn, nil
-		return dirty && err == nil, nn, err
 
 	default:
 		panic(fmt.Sprintf("%T: invalid node: %v", n, n))
@@ -697,16 +676,16 @@ func (t *Trie) delete(n node, prefix, key []byte) (bool, node, error) {
 		return true, nn, nil
 
 	case *expiredNode:
-		records, err := archive.ArchivedNodeResolver(n.offset, n.size)
+		log.Info("Resolving expired node in delete()", "owner", t.owner, "offset", n.offset, "size", n.size)
+		rn, err := resolveExpiredNodeData(n)
 		if err != nil {
-			return false, nil, fmt.Errorf("failed to resolve expired node: %w", err)
+			return false, nil, err
 		}
-		nn, err := archiveRecordsToNode(records)
+		dirty, nn, err := t.delete(rn, prefix, key)
-		if err != nil {
+		if !dirty || err != nil {
-			return false, nil, fmt.Errorf("failed to rebuild expired node from archive: %w", err)
+			return false, rn, err
 		}
-		dirty, _, err := t.delete(nn, prefix, key)
+		return true, nn, nil
-		return dirty && err == nil, nn, err
 
 	default:
 		panic(fmt.Sprintf("%T: invalid node: %v (%v)", n, n, key))
@@ -742,7 +721,8 @@ func copyNode(n node) node {
 		return &expiredNode{
 			offset:          n.offset,
 			size:            n.size,
-			archiveResolver: archive.ArchivedNodeResolver,
+			cachedHash:      common.CopyBytes(n.cachedHash),
+			archiveResolver: n.archiveResolver,
 		}
 	default:
 		panic(fmt.Sprintf("%T: unknown node type", n))
@@ -750,8 +730,11 @@ func copyNode(n node) node {
 }
 
 func (t *Trie) resolve(n node, prefix []byte) (node, error) {
-	if n, ok := n.(hashNode); ok {
+	switch n := n.(type) {
+	case hashNode:
 		return t.resolveAndTrack(n, prefix)
+	case *expiredNode:
+		return resolveExpiredNodeData(n)
 	}
 	return n, nil
 }
@@ -862,6 +845,58 @@ func (t *Trie) Witness() map[string][]byte {
 	return t.prevalueTracer.Values()
 }
 
+// WalkStats holds statistics from a Walk traversal.
+type WalkStats struct {
+	Leaves          int // Number of leaf nodes visited
+	ExpiredResolved int // Number of expired nodes resolved from archive
+}
+
+// Walk recursively traverses the trie, resolving all nodes including
+// hashNodes and expiredNodes. It calls fn for each leaf found.
+// This triggers hash verification for expired nodes via cachedHash.
+func (t *Trie) Walk(fn func(path []byte, value []byte) error) (WalkStats, error) {
+	return t.walk(t.root, nil, fn)
+}
+
+func (t *Trie) walk(n node, path []byte, fn func([]byte, []byte) error) (WalkStats, error) {
+	switch n := n.(type) {
+	case *shortNode:
+		return t.walk(n.Val, append(append([]byte{}, path...), n.Key...), fn)
+	case *fullNode:
+		var stats WalkStats
+		for i, child := range n.Children[:16] {
+			if child != nil {
+				childStats, err := t.walk(child, append(append([]byte{}, path...), byte(i)), fn)
+				if err != nil {
+					return stats, err
+				}
+				stats.Leaves += childStats.Leaves
+				stats.ExpiredResolved += childStats.ExpiredResolved
+			}
+		}
+		return stats, nil
+	case hashNode:
+		resolved, err := t.resolveAndTrack(n, path)
+		if err != nil {
+			return WalkStats{}, err
+		}
+		return t.walk(resolved, path, fn)
+	case *expiredNode:
+		resolved, err := resolveExpiredNodeData(n)
+		if err != nil {
+			return WalkStats{}, err
+		}
+		childStats, err := t.walk(resolved, path, fn)
+		childStats.ExpiredResolved++
+		return childStats, err
+	case valueNode:
+		return WalkStats{Leaves: 1}, fn(path, []byte(n))
+	case nil:
+		return WalkStats{}, nil
+	}
+	return WalkStats{}, nil
+}
+
 // reset drops the referenced root node and cleans all internal state.
 func (t *Trie) reset() {
 	t.root = nil

triedb/database.go

@@ -399,6 +399,28 @@ func (db *Database) Disk() ethdb.Database {
 	return db.disk
 }
 
+// DiffHead returns the root hash of the topmost diff layer in pathdb.
+// If there are no diff layers or the backend is not pathdb, it returns
+// the zero hash and false.
+func (db *Database) DiffHead() (common.Hash, bool) {
+	pdb, ok := db.backend.(*pathdb.Database)
+	if !ok {
+		return common.Hash{}, false
+	}
+	return pdb.DiffHead()
+}
+
+// DisableStateHistory closes and disables the state history freezer.
+// This is used by the archiver to bypass state history writes during
+// diff layer flushing when state history may have gaps.
+func (db *Database) DisableStateHistory() {
+	pdb, ok := db.backend.(*pathdb.Database)
+	if !ok {
+		return
+	}
+	pdb.DisableStateHistory()
+}
+
 // SnapshotCompleted returns the indicator if the snapshot is completed.
 func (db *Database) SnapshotCompleted() bool {
 	pdb, ok := db.backend.(*pathdb.Database)

triedb/pathdb/database.go

@@ -318,6 +318,30 @@ func (db *Database) Update(root common.Hash, parentRoot common.Hash, block uint6
 	return db.tree.cap(root, maxDiffLayers)
 }
 
+// DiffHead returns the root hash of the topmost diff layer. If there are no
+// diff layers (only the disk layer), it returns the disk layer root and false.
+func (db *Database) DiffHead() (common.Hash, bool) {
+	db.lock.RLock()
+	defer db.lock.RUnlock()
+
+	return db.tree.diffHead()
+}
+
+// DisableStateHistory closes and disables the state history freezer. This is
+// used by the archiver to bypass state history writes during diff layer flushing,
+// since the archiver only needs trie nodes committed to disk and state history
+// may have gaps from unclean shutdowns that prevent sequential appends.
+func (db *Database) DisableStateHistory() {
+	db.lock.Lock()
+	defer db.lock.Unlock()
+
+	if db.stateFreezer != nil {
+		db.stateFreezer.Close()
+		db.stateFreezer = nil
+		log.Info("Disabled state history freezer")
+	}
+}
+
 // Commit traverses downwards the layer tree from a specified layer with the
 // provided state root and all the layers below are flattened downwards. It
 // can be used alone and mostly for test purposes.

triedb/pathdb/history.go

@@ -278,9 +278,17 @@ func truncateFromHead(store ethdb.AncientStore, typ historyType, nhead uint64) (
 		return 0, err
 	}
 	// Ensure that the truncation target falls within the valid range.
-	if ohead < nhead || nhead < otail {
+	if nhead < otail {
 		return 0, fmt.Errorf("%w, %s, tail: %d, head: %d, target: %d", errHeadTruncationOutOfRange, typ, otail, ohead, nhead)
 	}
+	// If the target is ahead of the current head, there's nothing to truncate.
+	// This can happen after unclean shutdowns where the state history was not
+	// fully written.
+	if ohead < nhead {
+		log.Warn("State history shorter than target, nothing to truncate",
+			"type", typ.String(), "head", ohead, "target", nhead)
+		return 0, nil
+	}
 	// Short circuit if nothing to truncate.
 	if ohead == nhead {
 		return 0, nil

triedb/pathdb/layertree.go

@@ -31,6 +31,7 @@ import (
 // of the referenced layer by themselves.
 type layerTree struct {
 	base   *diskLayer
+	head   common.Hash // Root hash of the topmost layer (diff or disk)
 	layers map[common.Hash]layer
 
 	// descendants is a two-dimensional map where the keys represent
@@ -59,6 +60,7 @@ func (tree *layerTree) init(head layer) {
 	defer tree.lock.Unlock()
 
 	current := head
+	tree.head = head.rootHash()
 	tree.layers = make(map[common.Hash]layer)
 	tree.descendants = make(map[common.Hash]map[common.Hash]struct{})
 
@@ -76,6 +78,18 @@ func (tree *layerTree) init(head layer) {
 	tree.lookup = newLookup(head, tree.isDescendant)
 }
 
+// diffHead returns the root hash of the topmost diff layer. If there are no
+// diff layers, returns the disk layer root and false.
+func (tree *layerTree) diffHead() (common.Hash, bool) {
+	tree.lock.RLock()
+	defer tree.lock.RUnlock()
+
+	if _, ok := tree.layers[tree.head].(*diffLayer); ok {
+		return tree.head, true
+	}
+	return tree.base.rootHash(), false
+}
+
 // get retrieves a layer belonging to the given state root.
 func (tree *layerTree) get(root common.Hash) layer {
 	tree.lock.RLock()