go-ethereum/core/state/journal.go

// Copyright 2016 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 state

import (
	"fmt"
	"maps"
	"slices"
	"sort"

	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/crypto"
	"github.com/holiman/uint256"
)

// stateBytesPerSlot is the number of "state-creation bytes" billed for a slot
// transitioning from zero to non-zero within a call frame, and refunded when
// such a slot is cleared back to zero whose tx-original value was also zero.
const stateBytesPerSlot = 64

// stateBytesPerAccount is the per-account overhead billed when a brand-new
// account is created in a call frame.
const stateBytesPerAccount = 120

// frameRange is a half-open interval [start, end) of journal entry indices,
// used to record the slice of entries occupied by a closed child call frame.
type frameRange struct {
	start, end int
}

type revision struct {
	id           int
	journalIndex int
	// closedChildren holds the [start, end) ranges of child call frames that
	// have been closed under this revision via closeSnapshot. Together with
	// journalIndex (this frame's own start) and the current journal length
	// (this frame's tentative end) they describe the slice of entries that
	// belong directly to this frame, with descendant frames' entries excluded.
	//
	// Invariant: ranges are appended in increasing order, are non-overlapping,
	// and lie entirely within [journalIndex, len(entries)).
	closedChildren []frameRange
	// childStateBytes is the sum of state-creation bytes that this frame's
	// successful child frames (and their successful descendants, transitively)
	// produced via closeSnapshot. It is propagated upwards each time a child
	// closes, so that if THIS frame is later reverted, the caller can recover
	// the total amount that was emitted for state changes which the revert is
	// now throwing away.
	childStateBytes int
}

// journalEntry is a modification entry in the state change journal that can be
// reverted on demand.
type journalEntry interface {
	// revert undoes the changes introduced by this journal entry.
	revert(*StateDB)

	// dirtied returns the Ethereum address modified by this journal entry.
	// indicates false if no address was changed.
	dirtied() (common.Address, bool)

	// copy returns a deep-copied journal entry.
	copy() journalEntry
}

// journal contains the list of state modifications applied since the last state
// commit. These are tracked to be able to be reverted in the case of an execution
// exception or request for reversal.
type journal struct {
	entries []journalEntry         // Current changes tracked by the journal
	dirties map[common.Address]int // Dirty accounts and the number of changes

	validRevisions []revision
	nextRevisionId int
}

// newJournal creates a new initialized journal.
func newJournal() *journal {
	return &journal{
		dirties: make(map[common.Address]int),
	}
}

// reset clears the journal, after this operation the journal can be used anew.
// It is semantically similar to calling 'newJournal', but the underlying slices
// can be reused.
func (j *journal) reset() {
	j.entries = j.entries[:0]
	j.validRevisions = j.validRevisions[:0]
	clear(j.dirties)
	j.nextRevisionId = 0
}

// snapshot returns an identifier for the current revision of the state.
func (j *journal) snapshot() int {
	id := j.nextRevisionId
	j.nextRevisionId++
	j.validRevisions = append(j.validRevisions, revision{id: id, journalIndex: j.length()})
	return id
}

// revertToSnapshot reverts all state changes made since the given revision.
//
// It returns the sum of state-creation bytes that successful child frames
// nested within the reverted scope(s) had previously emitted via
// closeSnapshot. The caller can use this figure to undo whatever bookkeeping
// (e.g. gas charging) it did at the time those bytes were reported, since the
// state changes those bytes were paying for are now being thrown away.
func (j *journal) revertToSnapshot(revid int, s *StateDB) int {
	// Find the snapshot in the stack of valid snapshots.
	idx := sort.Search(len(j.validRevisions), func(i int) bool {
		return j.validRevisions[i].id >= revid
	})
	if idx == len(j.validRevisions) || j.validRevisions[idx].id != revid {
		panic(fmt.Errorf("revision id %v cannot be reverted", revid))
	}
	snapshot := j.validRevisions[idx].journalIndex

	// Sum the child-state-bytes carried by every revision being unwound. When
	// revertToSnapshot tears down multiple stacked frames at once, each of
	// them may itself have closed children whose bytes were inherited but
	// never bubbled further up; collecting all of them here lets the caller
	// undo the full subtree's emissions in one go.
	var refund int
	for i := idx; i < len(j.validRevisions); i++ {
		refund += j.validRevisions[i].childStateBytes
	}

	// Replay the journal to undo changes and remove invalidated snapshots
	j.revert(s, snapshot)
	j.validRevisions = j.validRevisions[:idx]
	return refund
}

// closeSnapshot marks the end of the call frame identified by revid without
// reverting any state. The frame's entry range [snapshot_index, current_length)
// is recorded on its parent revision so callers can later iterate the parent's
// own entries while skipping over closed children (and, transitively, their
// descendants — descendant ranges are absorbed into the closing child's range
// when the descendant itself was closed earlier under that child).
//
// closeSnapshot must be invoked in LIFO order: revid must identify the topmost
// snapshot. It panics otherwise. The corresponding revision is popped, so a
// subsequent revertToSnapshot on the same id is no longer valid.
//
// It returns the net state-creation bytes attributable to THIS frame's own
// storage changes (descendant frames' contributions are excluded — they were
// already reported when the descendants closed). For each storage slot that
// this frame touched directly:
//   - if the slot is non-zero now and was zero when the frame first touched
//     it, +stateBytesPerSlot is accumulated;
//   - if the slot is zero now, was non-zero when the frame first touched it,
//     and was zero at the start of the transaction, -stateBytesPerSlot is
//     accumulated.
//
// The returned value is also folded into the parent's childStateBytes (along
// with this frame's own childStateBytes) so a future revertToSnapshot on the
// parent can recover the entire subtree's accumulated bytes.
func (j *journal) closeSnapshot(revid int) int {
	if len(j.validRevisions) == 0 {
		panic(fmt.Errorf("revision id %v cannot be closed: no open snapshot", revid))
	}
	top := len(j.validRevisions) - 1
	if j.validRevisions[top].id != revid {
		panic(fmt.Errorf("revision id %v cannot be closed: top is %v",
			revid, j.validRevisions[top].id))
	}
	rev := &j.validRevisions[top]

	// Compute net state-creation bytes for THIS frame's own slot changes,
	// skipping any entries that lie inside a closed child's range.
	thisBytes := j.computeFrameStateBytes(rev)

	// Record this frame's range and propagate accumulated bytes to the
	// parent. The propagated total is "this frame's own bytes" + "this
	// frame's already-accumulated child bytes": from the parent's vantage
	// point the whole subtree is now a single closed child.
	closed := frameRange{
		start: rev.journalIndex,
		end:   len(j.entries),
	}
	if top > 0 {
		parent := &j.validRevisions[top-1]
		if closed.start < closed.end {
			parent.closedChildren = append(parent.closedChildren, closed)
		}
		parent.childStateBytes += thisBytes + rev.childStateBytes
	}
	// Drop this revision's bookkeeping. The slice is reused by the parent so
	// avoid pinning it via the popped tail.
	rev.closedChildren = nil
	rev.childStateBytes = 0
	j.validRevisions = j.validRevisions[:top]
	return thisBytes
}

// computeFrameStateBytes walks the entries that belong directly to rev (skipping
// any closed-child ranges) and sums the per-step state-creation contribution of
// each individual SSTORE.
//
// State-creation accounting is the per-step sum of three independent
// contributions, each applied locally to its own journal entry:
//
//  1. storageChange (slot SSTORE):
//     - origin != 0                  → 0  (rearranging pre-existing storage)
//     - prev == 0 && new != 0        → +stateBytesPerSlot  (new slot created)
//     - prev != 0 && new == 0        → -stateBytesPerSlot  (in-tx creation undone)
//
//  2. codeChange (SetCode on an account): a brand-new contract publishes its
//     bytecode for the first time. Origin code is implicitly empty in this
//     accounting — we treat the prev-empty/new-non-empty transition as the
//     creation event and bill its byte size, with the inverse transition
//     refunding it for symmetry.
//     - len(prev) == 0 && len(new) > 0 → +len(new)  (code committed)
//     - len(prev) > 0  && len(new) == 0 → -len(prev) (in-tx code committed then cleared)
//
//  3. createObjectChange (account materialised in state): each event adds
//     +stateBytesPerAccount of per-account overhead.
//
// The per-step formulation composes naturally: a frame's bytes is the sum of
// deltas of its own entries, and the sum of every frame's bytes across the
// subtree equals the sum of deltas across all entries — i.e. the same number
// you would get from a single whole-frame walk. Slots/code/accounts whose
// intermediate values bounce across frame boundaries reconcile automatically
// without any need to dedup by "first touch".
func (j *journal) computeFrameStateBytes(rev *revision) int {
	var total int
	zero := common.Hash{}
	visit := func(e journalEntry) {
		switch ch := e.(type) {
		case storageChange:
			switch {
			case ch.origvalue != zero:
				// Slot was already populated at tx-start; any in-tx
				// transition is rearranging existing storage.
			case ch.prevvalue == zero && ch.newvalue != zero:
				total += stateBytesPerSlot
			case ch.prevvalue != zero && ch.newvalue == zero:
				total -= stateBytesPerSlot
			}
		case codeChange:
			switch {
			case len(ch.prevCode) == 0 && len(ch.newCode) > 0:
				total += len(ch.newCode)
			case len(ch.prevCode) > 0 && len(ch.newCode) == 0:
				total -= len(ch.prevCode)
			}
		case createObjectChange:
			total += stateBytesPerAccount
		}
	}
	idx := rev.journalIndex
	for _, child := range rev.closedChildren {
		for ; idx < child.start; idx++ {
			visit(j.entries[idx])
		}
		idx = child.end
	}
	for ; idx < len(j.entries); idx++ {
		visit(j.entries[idx])
	}
	return total
}

// frameEntries invokes visit for each entry that belongs directly to the
// current (topmost) call frame, skipping entries that lie within any closed
// child frame's range. Entries are visited in append order. If no frame is
// open, frameEntries is a no-op.
func (j *journal) frameEntries(visit func(entry journalEntry)) {
	if len(j.validRevisions) == 0 {
		return
	}
	rev := j.validRevisions[len(j.validRevisions)-1]
	idx := rev.journalIndex
	for _, child := range rev.closedChildren {
		for ; idx < child.start; idx++ {
			visit(j.entries[idx])
		}
		idx = child.end
	}
	for ; idx < len(j.entries); idx++ {
		visit(j.entries[idx])
	}
}

// append inserts a new modification entry to the end of the change journal.
func (j *journal) append(entry journalEntry) {
	j.entries = append(j.entries, entry)
	if addr, dirty := entry.dirtied(); dirty {
		j.dirties[addr]++
	}
}

// revert undoes a batch of journalled modifications along with any reverted
// dirty handling too.
func (j *journal) revert(statedb *StateDB, snapshot int) {
	for i := len(j.entries) - 1; i >= snapshot; i-- {
		// Undo the changes made by the operation
		j.entries[i].revert(statedb)

		// Drop any dirty tracking induced by the change
		if addr, dirty := j.entries[i].dirtied(); dirty {
			if j.dirties[addr]--; j.dirties[addr] == 0 {
				delete(j.dirties, addr)
			}
		}
	}
	j.entries = j.entries[:snapshot]
}

// dirty explicitly sets an address to dirty, even if the change entries would
// otherwise suggest it as clean. This method is an ugly hack to handle the RIPEMD
// precompile consensus exception.
func (j *journal) dirty(addr common.Address) {
	j.dirties[addr]++
}

// length returns the current number of entries in the journal.
func (j *journal) length() int {
	return len(j.entries)
}

// copy returns a deep-copied journal.
func (j *journal) copy() *journal {
	entries := make([]journalEntry, 0, j.length())
	for i := 0; i < j.length(); i++ {
		entries = append(entries, j.entries[i].copy())
	}
	revisions := make([]revision, len(j.validRevisions))
	for i, r := range j.validRevisions {
		revisions[i] = revision{
			id:              r.id,
			journalIndex:    r.journalIndex,
			closedChildren:  slices.Clone(r.closedChildren),
			childStateBytes: r.childStateBytes,
		}
	}
	return &journal{
		entries:        entries,
		dirties:        maps.Clone(j.dirties),
		validRevisions: revisions,
		nextRevisionId: j.nextRevisionId,
	}
}

func (j *journal) logChange(txHash common.Hash) {
	j.append(addLogChange{txhash: txHash})
}

func (j *journal) createObject(addr common.Address) {
	j.append(createObjectChange{account: addr})
}

func (j *journal) createContract(addr common.Address) {
	j.append(createContractChange{account: addr})
}

func (j *journal) destruct(addr common.Address) {
	j.append(selfDestructChange{account: addr})
}

func (j *journal) storageChange(addr common.Address, key, prev, newval, origin common.Hash) {
	j.append(storageChange{
		account:   addr,
		key:       key,
		prevvalue: prev,
		newvalue:  newval,
		origvalue: origin,
	})
}

func (j *journal) transientStateChange(addr common.Address, key, prev common.Hash) {
	j.append(transientStorageChange{
		account:  addr,
		key:      key,
		prevalue: prev,
	})
}

func (j *journal) refundChange(previous uint64) {
	j.append(refundChange{prev: previous})
}

func (j *journal) balanceChange(addr common.Address, previous *uint256.Int) {
	j.append(balanceChange{
		account: addr,
		prev:    previous.Clone(),
	})
}

func (j *journal) setCode(address common.Address, prevCode, newCode []byte) {
	j.append(codeChange{
		account:  address,
		prevCode: prevCode,
		newCode:  newCode,
	})
}

func (j *journal) nonceChange(address common.Address, prev uint64) {
	j.append(nonceChange{
		account: address,
		prev:    prev,
	})
}

func (j *journal) touchChange(address common.Address) {
	j.append(touchChange{
		account: address,
	})
	if address == ripemd {
		// Explicitly put it in the dirty-cache, which is otherwise generated from
		// flattened journals.
		j.dirty(address)
	}
}

func (j *journal) accessListAddAccount(addr common.Address) {
	j.append(accessListAddAccountChange{addr})
}

func (j *journal) accessListAddSlot(addr common.Address, slot common.Hash) {
	j.append(accessListAddSlotChange{
		address: addr,
		slot:    slot,
	})
}

type (
	// Changes to the account trie.
	createObjectChange struct {
		account common.Address
	}
	// createContractChange represents an account becoming a contract-account.
	// This event happens prior to executing initcode. The journal-event simply
	// manages the created-flag, in order to allow same-tx destruction.
	createContractChange struct {
		account common.Address
	}
	selfDestructChange struct {
		account common.Address
	}

	// Changes to individual accounts.
	balanceChange struct {
		account common.Address
		prev    *uint256.Int
	}
	nonceChange struct {
		account common.Address
		prev    uint64
	}
	storageChange struct {
		account   common.Address
		key       common.Hash
		prevvalue common.Hash
		newvalue  common.Hash
		origvalue common.Hash
	}
	codeChange struct {
		account  common.Address
		prevCode []byte
		newCode  []byte
	}

	// Changes to other state values.
	refundChange struct {
		prev uint64
	}
	addLogChange struct {
		txhash common.Hash
	}
	touchChange struct {
		account common.Address
	}

	// Changes to the access list
	accessListAddAccountChange struct {
		address common.Address
	}
	accessListAddSlotChange struct {
		address common.Address
		slot    common.Hash
	}

	// Changes to transient storage
	transientStorageChange struct {
		account       common.Address
		key, prevalue common.Hash
	}
)

func (ch createObjectChange) revert(s *StateDB) {
	delete(s.stateObjects, ch.account)
}

func (ch createObjectChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch createObjectChange) copy() journalEntry {
	return createObjectChange{
		account: ch.account,
	}
}

func (ch createContractChange) revert(s *StateDB) {
	s.getStateObject(ch.account).newContract = false
}

func (ch createContractChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch createContractChange) copy() journalEntry {
	return createContractChange{
		account: ch.account,
	}
}

func (ch selfDestructChange) revert(s *StateDB) {
	obj := s.getStateObject(ch.account)
	if obj != nil {
		obj.selfDestructed = false
	}
}

func (ch selfDestructChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch selfDestructChange) copy() journalEntry {
	return selfDestructChange{
		account: ch.account,
	}
}

var ripemd = common.HexToAddress("0000000000000000000000000000000000000003")

func (ch touchChange) revert(s *StateDB) {
}

func (ch touchChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch touchChange) copy() journalEntry {
	return touchChange{
		account: ch.account,
	}
}

func (ch balanceChange) revert(s *StateDB) {
	s.getStateObject(ch.account).setBalance(ch.prev)
}

func (ch balanceChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch balanceChange) copy() journalEntry {
	return balanceChange{
		account: ch.account,
		prev:    new(uint256.Int).Set(ch.prev),
	}
}

func (ch nonceChange) revert(s *StateDB) {
	s.getStateObject(ch.account).setNonce(ch.prev)
}

func (ch nonceChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch nonceChange) copy() journalEntry {
	return nonceChange{
		account: ch.account,
		prev:    ch.prev,
	}
}

func (ch codeChange) revert(s *StateDB) {
	s.getStateObject(ch.account).setCode(crypto.Keccak256Hash(ch.prevCode), ch.prevCode)
}

func (ch codeChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch codeChange) copy() journalEntry {
	return codeChange{
		account:  ch.account,
		prevCode: ch.prevCode,
		newCode:  ch.newCode,
	}
}

func (ch storageChange) revert(s *StateDB) {
	s.getStateObject(ch.account).setState(ch.key, ch.prevvalue, ch.origvalue)
}

func (ch storageChange) dirtied() (common.Address, bool) {
	return ch.account, true
}

func (ch storageChange) copy() journalEntry {
	return storageChange{
		account:   ch.account,
		key:       ch.key,
		prevvalue: ch.prevvalue,
		newvalue:  ch.newvalue,
		origvalue: ch.origvalue,
	}
}

func (ch transientStorageChange) revert(s *StateDB) {
	s.setTransientState(ch.account, ch.key, ch.prevalue)
}

func (ch transientStorageChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch transientStorageChange) copy() journalEntry {
	return transientStorageChange{
		account:  ch.account,
		key:      ch.key,
		prevalue: ch.prevalue,
	}
}

func (ch refundChange) revert(s *StateDB) {
	s.refund = ch.prev
}

func (ch refundChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch refundChange) copy() journalEntry {
	return refundChange{
		prev: ch.prev,
	}
}

func (ch addLogChange) revert(s *StateDB) {
	logs := s.logs[ch.txhash]
	if len(logs) == 1 {
		delete(s.logs, ch.txhash)
	} else {
		s.logs[ch.txhash] = logs[:len(logs)-1]
	}
	s.logSize--
}

func (ch addLogChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch addLogChange) copy() journalEntry {
	return addLogChange{
		txhash: ch.txhash,
	}
}

func (ch accessListAddAccountChange) revert(s *StateDB) {
	/*
		One important invariant here, is that whenever a (addr, slot) is added, if the
		addr is not already present, the add causes two journal entries:
		- one for the address,
		- one for the (address,slot)
		Therefore, when unrolling the change, we can always blindly delete the
		(addr) at this point, since no storage adds can remain when come upon
		a single (addr) change.
	*/
	s.accessList.DeleteAddress(ch.address)
}

func (ch accessListAddAccountChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch accessListAddAccountChange) copy() journalEntry {
	return accessListAddAccountChange{
		address: ch.address,
	}
}

func (ch accessListAddSlotChange) revert(s *StateDB) {
	s.accessList.DeleteSlot(ch.address, ch.slot)
}

func (ch accessListAddSlotChange) dirtied() (common.Address, bool) {
	return common.Address{}, false
}

func (ch accessListAddSlotChange) copy() journalEntry {
	return accessListAddSlotChange{
		address: ch.address,
		slot:    ch.slot,
	}
}