Re: [xml] C14N issue with digital signature due to pointer comparison

[Aleksey Sanin <aleksey aleksey com>]

The tradeoff here is speed. Again, if you are using standard LibXML2
API then everything will work as expected. I have no idea why do you
need to create new xmlNsPtr objects - this sounds like a waste of
CPU/memory to me but you might have your reasons. But this also breaks
a few assumptions on how LibXML2 represents the DOM tree in memory
and you are on your own in this case.

Aleksey

On 3/20/14, 1:26 AM, Frank Gross wrote:
> Hi,
>
>  Actually, I have an additional layer to emulate DOM compliant API over
> libxml. In that layer, I have a global list of xmlNsPtr for all
> namespaces encountered by the nodes (and referenced in node->ns) of a
> single XML document. And for each namespace definition (xmlns:xxx) set
> on a node, a new xmlNsPtr is created and referenced in node->nsDef.  So
> during canonicalization process, when a node looks for a parent
> namespace definition via xmlSearchNs, it can happen that the xmlNsPtr
> reference from the node is different from the parent node->nsDef , even
> if the href and prefix value are the same. Then of course simple pointer
> comparison doesn't work. So I was asking myself if such situation cannot
> happen to other people, and maybe change the pointer comparison because
> depending on how people build and manipulate the document, the test is
> not always accurate.
>
>
> Frank
>
>
> Le 20/03/2014 03:16, Aleksey Sanin a écrit :
> > How do you manipulate the XML tree? If you are using "official" LibXML2
> > function then I believe this code should work just fine unless there is
> > a bug in the strings dictionary.
> >
> > Aleksey
> >
> > On 3/12/14, 10:28 AM, Frank Gross wrote:
> > > Hi,
> > >
> > >    I'm getting some trouble to verify a XML signature because the
> > > xmlC14NProcessNamespacesAxis() function does a xmlNsPtr pointer
> > > comparison to decide whether a sub node belongs to the same default
> > > namespace as an ancestor. But if the sub node has been manipulated by
> > > program (in my case) to point to another xmlNsPtr with same values, the
> > > canonicalization process breaks. Wouldn't it be better to check the
> > > namespace href values instead as in following patch ?
> > >
> > > diff --git a/lib-xmlsoft-libxml2/src/c14n.c
> > > b/lib-xmlsoft-libxml2/src/c14n.c
> > > index 9c3cad2..f73e709 100644
> > > --- a/lib-xmlsoft-libxml2/src/c14n.c
> > > +++ b/lib-xmlsoft-libxml2/src/c14n.c
> > > @@ -623,7 +623,7 @@ xmlC14NProcessNamespacesAxis(xmlC14NCtxPtr ctx,
> > > xmlNodePtr cur, int visible)
> > >       for(ns = n->nsDef; ns != NULL; ns = ns->next) {
> > >           tmp = xmlSearchNs(cur->doc, cur, ns->prefix);
> > >
> > > -        if((tmp == ns) && !xmlC14NIsXmlNs(ns) && xmlC14NIsVisible(ctx,
> > > ns, cur)) {
> > > +        if((xmlStrEqual(tmp->href,ns->href)) &&
> > > (xmlStrEqual(tmp->prefix,ns->prefix)) && !xmlC14NIsXmlNs(ns) &&
> > > xmlC14NIsVisible(ctx, ns, cur)) {
> > >           already_rendered = xmlC14NVisibleNsStackFind(ctx->ns_rendered,
> > > ns);
> > >           if(visible) {
> > >                   xmlC14NVisibleNsStackAdd(ctx->ns_rendered, ns, cur);
> > >
> > > Regards,
> > > Frank