fix: correct cache-control in car responses

Context: ipfs/specs#295


This commit was moved from ipfs/kubo@289e465

gateway/core/corehttp/gateway_handler_car.go

@@ -43,8 +43,14 @@ func (i *gatewayHandler) serveCAR(ctx context.Context, w http.ResponseWriter, r
 	}
 	setContentDispositionHeader(w, name, "attachment")
 
-	// Weak Etag W/ because we can't guarantee byte-for-byte identical  responses
-	// (CAR is streamed, and in theory, blocks may arrive from datastore in non-deterministic order)
+	// Set Cache-Control (same logic as for a regular files)
+	addCacheControlHeaders(w, r, contentPath, rootCid)
+
+	// Weak Etag W/ because we can't guarantee byte-for-byte identical
+	// responses, but still want to benefit from HTTP Caching. Two CAR
+	// responses for the same CID and selector will be logically equivalent,
+	// but when CAR is streamed, then in theory, blocks may arrive from
+	// datastore in non-deterministic order.
 	etag := `W/` + getEtag(r, rootCid)
 	w.Header().Set("Etag", etag)
 
@@ -55,14 +61,10 @@ func (i *gatewayHandler) serveCAR(ctx context.Context, w http.ResponseWriter, r
 	}
 
 	// Make it clear we don't support range-requests over a car stream
-	// Partial downloads and resumes should be handled using
-	// IPLD selectors: https://github.com/ipfs/go-ipfs/issues/8769
+	// Partial downloads and resumes should be handled using requests for
+	// sub-DAGs and IPLD selectors: https://github.com/ipfs/go-ipfs/issues/8769
 	w.Header().Set("Accept-Ranges", "none")
 
-	// Explicit Cache-Control to ensure fresh stream on retry.
-	// CAR stream could be interrupted, and client should be able to resume and get full response, not the truncated one
-	w.Header().Set("Cache-Control", "no-cache, no-transform")
-
 	w.Header().Set("Content-Type", "application/vnd.ipld.car; version=1")
 	w.Header().Set("X-Content-Type-Options", "nosniff") // no funny business in the browsers :^)