"Prefetching is not allowed"

[Kaomie]

Hiya,

Sometimes I get this blank page with only the following message:

Prefetching is not allowed due to the various privacy issues that arise.

This usually happens when googling for an addon and clicking on the link that goes directly to an addon page. I am suspecting this is due to some combination of FireFox+Google+Squid, but I have not confirmed exactly what mechanism triggers it.

If you tell me what you do not want us to do, I can probably tweak my FF or Squid configuration to adapt accordingly.

Thanks,

[Shirik]

I've heard of this before, and it's Squid being a bad man.

Prefetching is a pretty simple concept: Let's say you google for something. All those links pop up. While your browser (or in this case, your proxy) is sitting there doing nothing after that page has loaded waiting for you to click a link, it wants to speed things up. So what it does is actually silently follows each of those links to start downloading it ahead of time, in preparation for you to click them. This way, if you do actually click the link, your load time will (appear to) be shorter.

We don't allow this due to various issues.

The problem: When it's blocked, we DO use an HTTP error code. Firefox does prefetching as well (I think IE does also) and detects this error code and knows not to cache that information. Squid, on the other hand, ignores the error code and caches the error page. Thus, if you click the link after it's been prefetched, Squid goes ahead and delivers the (incorrect) error page instead.

I don't actually know the resolution to this unfortunately, I just know that's where the problem lies because it's come up before. Hopefully it's enough information to get you started, though.

Good luck,
-- Shirik

[Dolby]

The problem was the end users proxy.

[Kaomie]

Squid does not do (by default) any prefetching on its own. FireFox does, especially on a Google search result page which always adds rel="prefetch" on the first link returned (which IS really annoying for privacy, your browser will always visit the first page without you ever clicking on any link).

Anyway that does narrow a bit where the issue might be and I will have a closer look at the requests. You probably trigger on the X-moz: prefetch HTTP header, and the error returned is either filtered or misunderstood by Squid (must be HTTP 1.1 or something similar).

Keep you posted.

EDIT: you only return a HTTP/1.0 403 Forbidden (text/html) so this is the normal behavior, Squid will only cache the error page.

Quickest fix is to disable prefetching in FF altoghether, which is probably better anyway:
http://developer.mozilla.org/en/Link...prefetching.3f

And I will not explain the long fix because this is obviously not what we want