eth/fetcher: fix data race in updateCustody

eth/fetcher/blob_fetcher.go

@@ -114,6 +114,7 @@ type BlobFetcher struct {
 	notify    chan *blobTxAnnounce
 	cleanup   chan *payloadDelivery
 	drop      chan *txDrop
+	custodyCh chan types.CustodyBitmap
 	quit      chan struct{}
 	custody   *types.CustodyBitmap
 
@@ -160,6 +161,7 @@ func NewBlobFetcher(fn BlobFetcherFunctions, custody *types.CustodyBitmap, rand
 		notify:     make(chan *blobTxAnnounce),
 		cleanup:    make(chan *payloadDelivery),
 		drop:       make(chan *txDrop),
+		custodyCh:  make(chan types.CustodyBitmap),
 		quit:       make(chan struct{}),
 		full:       make(map[common.Hash]struct{}),
 		partial:    make(map[common.Hash]struct{}),
@@ -221,9 +223,14 @@ func (f *BlobFetcher) Drop(peer string) error {
 	}
 }
 
+// UpdateCustody hands a new custody bitmap to the fetcher loop. The actual
+// swap happens inside the loop so f.custody is never read and written
+// concurrently.
 func (f *BlobFetcher) UpdateCustody(cells types.CustodyBitmap) {
-	// todo use lock or process inside of loop
+	select {
-	f.custody = &cells
+	case f.custodyCh <- cells:
+	case <-f.quit:
+	}
 }
 
 func (f *BlobFetcher) Start() {
@@ -632,6 +639,9 @@ func (f *BlobFetcher) loop() {
 				f.rescheduleTimeout(timeoutTimer, timeoutTrigger)
 			}
 
+		case cells := <-f.custodyCh:
+			f.custody = &cells
+
 		case <-f.quit:
 			return
 		}