Re: [xml] file path vs URI [patch]

[Petr Sumbera <Petr Sumbera Sun COM>]

While I was looking more into the code I came up with rather simple 
sanity patches which solve both of my issues (described bellow in my 
previous email).

Some more background:

xmlCtxtReadFile() has on its input parameter 'filename' which is 
described as filename or URL. This function calls 
xmlLoadExternalEntity() where on input is expected URL. I have added 
code which handles filenames starting with //.

Practically the same issue is with xmlCtxtReadFile() which calls also 
xmlLoadExternalEntity().

Please see my attached patch and consider changing libxml2 in similar 
way. Any comments welcomed.

Thanks,

Petr

Petr Sumbera wrote:
> Hi guys,
>
> Sorry for maybe not well described or at least strange formulation as 
> I'm not XML expert at all.
>
> In UNIX systems file path starting with "//" is being accepted and it's 
> processed without complaining. For example:
>
> "cat //etc/passwd" will show content of /etc/passwd
>
> But this seems to be against URI (RFC 2396) or at least its 
> implementation in libxml2.
>
> http://www.ietf.org/rfc/rfc2396.txt
>
> libxml2 understands such path/URI(?) as Relative reference where "//" is 
>  understood as network. RFC 2396 says about Relative reference in 
> chapter 1.4 following:
>
> "In contrast, a relative identifier refers to a resource by describing 
> the difference within a hierarchical namespace between the current 
> context and an absolute identifier of the resource."
>
> Does libxml2 really follow up this? There seems to be simple fall back 
> from Absolute reference to Relative (in xmlParseURIReference()). I don't 
> see context anywhere.
>
> Simple example why I'm concerned:
>
> -bash-3.2$ head -n 2 /var/svc/manifest/network/ntp.xml
> <?xml version="1.0"?>
> <!DOCTYPE service_bundle SYSTEM 
> "/usr/share/lib/xml/dtd/service_bundle.dtd.1">
> -bash-3.2$ xmllint --valid --noout //var/svc/manifest/network/ntp.xml
> //var/svc/manifest/network/ntp.xml:2: I/O error : failed to load 
> external entity "//var/usr/share/lib/xml/dtd/service_bundle.dtd.1"
> <!DOCTYPE service_bundle SYSTEM 
> "/usr/share/lib/xml/dtd/service_bundle.dtd.1">
>
>        ^
> //var/svc/manifest/network/ntp.xml:2: warning: failed to load external 
> entity "//var/usr/share/lib/xml/dtd/service_bundle.dtd.1"
> <!DOCTYPE service_bundle SYSTEM 
> "/usr/share/lib/xml/dtd/service_bundle.dtd.1">
>
>        ^
> //var/svc/manifest/network/ntp.xml:15: validity error : Validation 
> failed: no DTD found !
> <service_bundle type='manifest' name='SUNWntpr:xntpd'>
>                                                      ^
> -bash-3.2$
>
> ---
>
> Note that DTD is searched in wrong path 
> "//var/usr/share/lib/xml/dtd/service_bundle.dtd.1" instead of 
> "/usr/share/lib/xml/dtd/service_bundle.dtd.1" (basically var is 
> considered as server)
>
> It may be problem of xmllint of not converting (sanitizing) path into 
> URI. Is it?
>
> And why I'm bothering about all of this? We have some utility which 
> suffers with the same symptoms after updating libxml2 from 2.6.10 to 
> 2.6.23. As it uses such functions like xmlReadFile (where on input is 
> filename) it can be similar to problem described above. The difference 
> is that the problem described above was there even in 2.6.10, while in 
> this version our utility was ok. Any idea here?
>
> Thanks for any comment,
>
> Petr

--- libxml2-2.6.30/parser.c.orig        Tue Aug 14 06:43:11 2007
+++ libxml2-2.6.30/parser.c     Fri Nov  2 04:42:37 2007
@@ -12259,6 +12259,11 @@
     if (options)
        xmlCtxtUseOptions(ctxt, options);
     ctxt->linenumbers = 1;
+
+    /* sanitize filename starting with // so it can be used as URI */
+    if (filename != NULL && strlen(filename) >= 2
+       && filename[0]=='/' && filename[1]=='/')
+       filename++;
     
     inputStream = xmlLoadExternalEntity(filename, NULL, ctxt);
     if (inputStream == NULL) {
@@ -13516,6 +13521,11 @@
 
     xmlCtxtReset(ctxt);
 
+    /* sanitize filename starting with // so it can be used as URI */
+    if (filename != NULL && strlen(filename) >= 2
+        && filename[0]=='/' && filename[1]=='/')
+        filename++;
+
     stream = xmlLoadExternalEntity(filename, NULL, ctxt);
     if (stream == NULL) {
         return (NULL);