On Mon, Jul 9, 2012 at 6:42 PM, Richard L. Barnes <[email protected]> wrote: > On Jul 9, 2012, at 7:24 PM, Adam Barth wrote: >> On Mon, Jul 9, 2012 at 4:19 PM, Richard L. Barnes <[email protected]> wrote: >>> I haven't thought much about this, but a couple of thoughts: >>> >>> The binary prologue means that the document is not valid HTML, so in >>> principle, it shouldn't be accepted as HTML. It makes you wonder what >>> other stuff you could put in there that the browser would stuff into the >>> DOM without it being obvious on the wire, say, to a proxy. I'm imagining >>> things like encrypted / compressed Javascript code that could be unpacked >>> by the more obviously HTML part of the page. >> >> You don't have to imagine. It's specified in HTML5. > > Could you clarify? What is "it"? Reference would be helpful.
You mentioned that you were wondering what "other stuff" you could put there that the browser would stuff into the DOM. The HTML specification [1] defines precisely what DOM you'll get for every possible input, so you don't need to wonder. > Is there really a use case for inserting into the DOM arbitrary octets that > are not syntactically part of the HTML page? This topic has been discussed at length in the HTML working group. It's probably not worth re-hashing it on this list. The short answer is that it's what web sites expect browsers to do. Adam [1] http://whatwg.org/specs/web-apps/current-work/ (or http://www.w3.org/TR/html5/ if you want to see the more official but less up-to-day version). >>> In a related vein, the "Text or Binary" section of >>> draft-ietf-websec-mime-sniff says that nothing scriptable must come out of >>> sniffing a binary blob. Yet in this case, it produced "text/html", which >>> is obviously scriptable. >> >> The browser isn't sniffing HTML in this case. The server sent a >> Content-Type header with text/html. >> >> Adam >> >> >>> On Jul 9, 2012, at 5:05 PM, Adam Barth wrote: >>> >>>> Why is this sniffing gone awry? Nothing bad seems to have happened in >>>> this example. >>>> >>>> Adam >>>> >>>> >>>> On Mon, Jul 9, 2012 at 1:55 PM, Richard L. Barnes <[email protected]> wrote: >>>>> Related to draft-ietf-websec-mime-sniff, an example of sniffing gone awry: >>>>> <http://lcamtuf.coredump.cx/squirrel/> >>>>> >>>>> It's a valid JPEG image that contains and HTML snippet in a comment >>>>> segment. As a result, when a browser loads the URL expecting an image, >>>>> it renders the image content, and when it expects HTML, it skips the >>>>> binary junk at the top and renders the HTML [*]. (In both cases, the >>>>> server reports Content-Type text/html.) What's even more startling is >>>>> that Chrome helpfully adds the binary junk at the top as the first child >>>>> of the <body> element in the parsed DOM! >>>>> >>>>> --Richard >>>>> >>>>> >>>>> [*] At least in Chrome 20.0.1132.47 >>>>> _______________________________________________ >>>>> websec mailing list >>>>> [email protected] >>>>> https://www.ietf.org/mailman/listinfo/websec >>> > _______________________________________________ websec mailing list [email protected] https://www.ietf.org/mailman/listinfo/websec
