1058551 - [e10s] The data: URI kitchen web page does not work properly

Status: VERIFIED FIXED

(Core :: DOM: Navigation, defect)

Description

[Alice0775 White]

Steps To Reproduce:
1. Open http://software.hixie.ch/utilities/cgi/data/data
2. Click [Generate] button

Actual Results:
Page transitioned as follows:

(Your browser should have redirected you automatically. If you see this it probably has a bug.)
Generated data: URI link appears as <a> link

Expected Results:
Page transitioned and data UTI should be loaded as a actual web page

Comment 1

[Alice0775 White]

s/UTI/URI/

Comment 2

[Chris Peterson [:cpeterson]]

Move old M2's low-priority bugs to M6 milestone.

Comment 3

[Blake Kaplan (:mrbkap) (inactive)]

Attachment: patch v1

I had basically finished this patch by the time that I found bug 590682. That being said, I don't think we need to block redirects to data: URIs on such a generic mechanism.

This patch is larger than I'd like because the HTTP redirect stuff basically assumes that we're going to need a channel in the parent as well as one in the child. data: is probably the only protocol that doesn't need this (and redirects to data: are probably quite rare), so I don't think that adding a little more code here is a terrible thing.

Jason, let me know if there's a better reviewer for this patch.

Comment 4

[Jason Duell]

Comment on attachment 8580838 [details] [diff] [review]
patch v1

I'm giving this review to honza.

Comment 5

[Honza Bambas (:mayhemer)]

Comment on attachment 8580838 [details] [diff] [review]
patch v1

Review of attachment 8580838 [details] [diff] [review]:
-----------------------------------------------------------------

LGTM, thanks!

::: netwerk/ipc/PNecko.ipdl
@@ +78,5 @@
>    SpeculativeConnect(URIParams uri);
>    HTMLDNSPrefetch(nsString hostname, uint16_t flags);
>    CancelHTMLDNSPrefetch(nsString hostname, uint16_t flags, nsresult reason);
> +
> +  PDataChannel(uint32_t channelId);

comment what the channelId is

::: netwerk/protocol/data/DataChannelParent.h
@@ +19,5 @@
> +// to data:, we need to implement the HTTP redirection API, which requires a
> +// class that implements nsIParentChannel, which calls NS_LinkRedirectChannels.
> +// In addition, the HTTP code blindly calls NS_NewChannel on the redirected-to
> +// URI, so we're guaranteed to create an nsDataChannel in the parent, as well as
> +// in the child.

You would need some IPC (at least placeholding) for data: anyway - even when we would create the target channel asynchronously on child (and let parent connect).  The thing is that on the parent there are checks made on source+target channels whether to redirect or not.  But these need to run on the content as well.  So you need some mirroring here.

@@ +39,5 @@
> +    NS_IMETHOD OnDataAvailable(nsIRequest *aRequest,
> +                               nsISupports *aContext,
> +                               nsIInputStream *aInputStream,
> +                               uint64_t aOffset,
> +                               uint32_t aCount) MOZ_OVERRIDE MOZ_FINAL;

why not use NS_DECL_NSIREQUESTOBSEVER and STREAMLISTENER?

Comment 6

[Blake Kaplan (:mrbkap) (inactive)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=b130e931a8fa (with Honza's comments addressed)

Comment 7

[Blake Kaplan (:mrbkap) (inactive)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/6fb980e58948

Comment 8

[Phil Ringnalda (:philor)]

https://hg.mozilla.org/mozilla-central/rev/6fb980e58948

Comment 9

[Moin Shaikh]

QA Whiteboard: [bugday-20150624][good first verify][verified in Firefox 39.0 b7. Operating System - Windows 7 (64bit) only]

Status: Resolved. Worked as expected.

Comment 10

[Ashickur Rahman]

Attachment: Selection_003.png

I checked in latest Nightly Build ID 20150701030207, both in e10s and non-e10s . In both case the page is blank for me.

Comment 11

[Martijn Wargers (dead)]

(In reply to Ashickur Rahman from comment #10)
> Created attachment 8628738 [details]
> Selection_003.png
> 
> I checked in latest Nightly Build ID 20150701030207, both in e10s and
> non-e10s . In both case the page is blank for me.

Yes, because there is nothing to show for on the generated page.