Description

Klever-Go KVM: Throttler slot leak in trie account-data sync causes epoch bootstrap / state sync DoS

Summary

The account-data trie syncers leak bounded throttler slots on error paths in syncDataTrie(). Each failed trie sync permanently consumes one slot from the NumGoRoutinesThrottler, and the slot is never returned unless the sync succeeds or the root hash was already present.

I confirmed this on the current default branch develop at commit 9640d63 (observed on May 20, 2026). I also confirmed the bug with a runtime PoC using the real timeout path in trieSyncer.StartSyncing(): two timed-out sync attempts are enough to exhaust a throttler with capacity 2.

This affects the epoch bootstrap path because syncUserAccountsState() and syncKappAccountsState() create bounded throttlers and abort bootstrap immediately if the syncer returns an error. Once enough trie-root sync attempts fail, the syncer cannot make forward progress and bootstrap fails.

Affected Components

data/syncer/userAccountsSyncer.go

data/syncer/kappAccountsSyncer.go

data/trie/sync.go

core/throttler/numGoRoutinesThrottler.go

core/bootstrap/process.go

Affected Version

Verified on:

develop HEAD 9640d63

Please check whether the same code is present in supported 1.7.x releases.

Suggested Severity

High

Vulnerability Details

Root Cause

Both account-data syncers call StartProcessing() before creating / starting the trie syncer, but they only call EndProcessing() on the success path and on the duplicate-root early return.

userAccountsSyncer.syncDataTrie():

func (u *userAccountsSyncer) syncDataTrie(rootHash []byte, ssh data.SyncStatisticsHandler, ctx context.Context) error {
    u.throttler.StartProcessing()

    u.syncerMutex.Lock()
    if _, ok := u.dataTries[string(rootHash)]; ok {
        u.syncerMutex.Unlock()
        u.throttler.EndProcessing()
        return nil...
Show more

Missing slot release paths

After StartProcessing(), the following error paths return without EndProcessing():

trie.NewTrie(...) returns an error

trie.NewTrieSyncer(...) returns an error

trieSyncer.StartSyncing(...) returns an error

Why this matters

NumGoRoutinesThrottler is a strict bounded counter:

  func (ngrt *NumGoRoutinesThrottler) CanProcess() bool {
      valCounter := atomic.LoadInt32(&ngrt.counter)
      return valCounter < ngrt.max
  }

  func (ngrt *NumGoRoutinesThrottler) StartProcessing() {
      atomic.AddInt32(&ngrt.counter, 1)
  }...
Show more

So after enough failures, further roots stop progressing and the sync operation eventually returns time is out.

Bootstrap impact

Epoch bootstrap uses these syncers directly and aborts on any error:

  err = e.syncUserAccountsState(e.epochStartMeta.Header.TrieRoot)
  if err != nil {
      return nil, nil, err
  }

  err = e.syncKappAccountsState(e.epochStartMeta.Header.KAppsTrieRoot)
  if err != nil {
      return nil, nil, err...
Show more

The throttlers for these paths are real bounded throttlers created from numConcurrentTrieSyncers.

Proof of Concept

I verified the bug with the real timeout path, not only with a canceled context.

The PoC below uses:

a real NumGoRoutinesThrottler with capacity 2

a real trieSyncer.StartSyncing()

an empty trie-node cache and a request handler that never supplies nodes

a short sync timeout (1s) so StartSyncing() returns trie.ErrTimeIsOut

After the first failed sync, one slot remains leaked. After the second failed sync, the throttler is exhausted.

PoC test

  package syncer

  import (
        "context"
        "testing"
        "time"

        commonmock "github.com/klever-io/klever-go/common/mock"...
Show more

Command used

  go test ./data/syncer -run 'TestPOC_(User|Kapp)AccountsSyncer_LeaksThrottlerSlotOnTrieTimeout' -count=1

Result

  ok    github.com/klever-io/klever-go/data/syncer      4.005s

This confirms the leak with the real timeout path from trieSyncer.StartSyncing().

Impact

An attacker who can repeatedly cause trie-node sync failures or timeouts during bootstrap can consume the bounded sync throttler until no capacity remains.

Once enough slots are leaked:

additional account-data trie sync attempts stop making progress

the parent loop waits until context timeout

SyncAccounts() fails

epoch bootstrap fails

This is a core node availability issue. It affects fresh/restarting nodes and validators that need to bootstrap or resync state.

This is not a theoretical issue:

StartSyncing() performs network-dependent trie-node retrieval

it already has explicit timeout / failure paths

the leaked throttler slots are confirmed by runtime PoC

Recommended Fix

Release the slot with defer immediately after StartProcessing() and cancel the defer only if ownership is intentionally transferred, which is not the case here.

Example fix pattern:

  func (u *userAccountsSyncer) syncDataTrie(rootHash []byte, ssh data.SyncStatisticsHandler, ctx context.Context) error {
      u.throttler.StartProcessing()
      defer u.throttler.EndProcessing()

      u.syncerMutex.Lock()
      defer u.syncerMutex.Unlock()

      if _, ok := u.dataTries[string(rootHash)]; ok {...
Show more

The same pattern should be applied to:

data/syncer/userAccountsSyncer.go

data/syncer/kappAccountsSyncer.go

References

data/syncer/userAccountsSyncer.go

data/syncer/kappAccountsSyncer.go

data/trie/sync.go

core/throttler/numGoRoutinesThrottler.go

core/bootstrap/process.go

SECURITY.md

Mitigation

Aliases

References