Loading...

XML

Word

Printable

Details

Type: Bug
Resolution: Unresolved
Priority: P2: Important
Fix Version/s: None
Affects Version/s: 5.9.4
Component/s: WebKit
Labels:
None

Description

Hi,

DNS prefetching can be disabled (and is disabled by default, actually) when using Qt WebKit [1]. However, when a web page explicitely enables DNS prefetching by using a header (X-DNS-Prefetch-Control: on) or a HTML meta http-equiv tag (<meta http-equiv="x-dns-prefetch-control" content="on">), DNS prefetching is enabled anyway, ignoring the settings in QWebSettings (this behavior is inherited from upstream). The following snippet of code shows what is happening (Source/WebCore/dom/Document.cpp):

  4729	void Document::initDNSPrefetch()
  4730	{
  4731	    Settings* settings = this->settings();
  4732	
  4733	    m_haveExplicitlyDisabledDNSPrefetch = false;
  4734	    m_isDNSPrefetchEnabled = settings && settings->dnsPrefetchingEnabled() && securityOrigin()->protocol() == "http";
  4735	
  4736	    // Inherit DNS prefetch opt-out from parent frame    
  4737	    if (Document* parent = parentDocument()) {
  4738	        if (!parent->isDNSPrefetchEnabled())
  4739	            m_isDNSPrefetchEnabled = false;
  4740	    }
  4741	}
  4742	
  4743	void Document::parseDNSPrefetchControlHeader(const String& dnsPrefetchControl)
  4744	{
  4745	    if (equalIgnoringCase(dnsPrefetchControl, "on") && !m_haveExplicitlyDisabledDNSPrefetch) {
  4746	        m_isDNSPrefetchEnabled = true;
  4747	        return;
  4748	    }
  4749	
  4750	    m_isDNSPrefetchEnabled = false;
  4751	    m_haveExplicitlyDisabledDNSPrefetch = true;
  4752	}

On line 4734, when initializing DNS prefetching, the settings regarding DNS prefetching are taken into account. However, when a website enables DNS prefetching with a header or a HTML meta http-equiv tag, the settings are ignored and DNS prefetching is enabled anyway (line 4746).

To me, that is somewhat unexpected since I instructed QWebSettings to disable DNS prefetching.

Here is a patch that would solve the problem (the patch is written such that the diff with upstream stays small; I can take a different approach if desired):

diff --git a/Source/WebCore/dom/Document.cpp b/Source/WebCore/dom/Document.cpp
index af63b141f..3ff5a2f25 100644
--- a/Source/WebCore/dom/Document.cpp
+++ b/Source/WebCore/dom/Document.cpp
@@ -4743,7 +4743,6 @@ void Document::initDNSPrefetch()
 void Document::parseDNSPrefetchControlHeader(const String& dnsPrefetchControl)
 {
     if (equalIgnoringCase(dnsPrefetchControl, "on") && !m_haveExplicitlyDisabledDNSPrefetch) {
-        m_isDNSPrefetchEnabled = true;
         return;
     }

Would you accept a patch like this? If so, against which git branch should I submit a gerrit review?

I doubt upstream will take such a patch, as the code seems to be written like this intentionally and furthermore, the behavior is actually described in the documentation [2].

Thanks to Jens Mueller for contacting Trojita about this issue [3].

1: https://doc.qt.io/archives/qt-5.5/qwebsettings.html#WebAttribute-enum
2: https://www.chromium.org/developers/design-documents/dns-prefetching#TOC-DNS-Prefetch-Control
3: https://bugs.kde.org/show_bug.cgi?id=390452