You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@shindig.apache.org by jss9920 <js...@gmail.com> on 2009/12/03 15:55:31 UTC

Locked Domain

I would like to start rendering my gadgets on unique locked domains but am
getting a bit confused about where to start...

Since the container generates the gadget iframe urls, it would seem that
implementing locked domain would be a container side affair, but I see
configuration options for locked domain support in the shindig container.js
file and I also see associated Java classes that do a lot of locked domain
work in the Shindig codebase (Java Shindig 1.0 release)...

So is there actually work that needs to be done on both the container side
and the Shindig side to enable locked domain gadgets?

Any pointers on how to get started (even high level) would be helpful and
much appreciated.

Thanks!

Re: Locked Domain

Posted by Kevin Brown <et...@google.com>.
On Thu, Dec 3, 2009 at 6:55 AM, jss9920 <js...@gmail.com> wrote:

> I would like to start rendering my gadgets on unique locked domains but am
> getting a bit confused about where to start...
>
> Since the container generates the gadget iframe urls, it would seem that
> implementing locked domain would be a container side affair, but I see
> configuration options for locked domain support in the shindig container.js
> file and I also see associated Java classes that do a lot of locked domain
> work in the Shindig codebase (Java Shindig 1.0 release)...
>
> So is there actually work that needs to be done on both the container side
> and the Shindig side to enable locked domain gadgets?
>

Yes. Shindig can only enforce the locked domain, but the container still
needs to generate it.

The JSON RPC handler for gadget metadata can provide rendering urls, or,
alternatively, the container could directly use UrlGenerator to get the
appropriate gadget rendering url.


>
> Any pointers on how to get started (even high level) would be helpful and
> much appreciated.
>
> Thanks!
>

Re: Locked Domain

Posted by Jesse <js...@gmail.com>.
* bump *

Any feedback at all on this thread would be appreciated...

Thanks!

--Jesse


On Mon, Dec 7, 2009 at 4:48 PM, Jesse <js...@gmail.com> wrote:

> I'm thinking the answer WRT the security token is to use a short lived
> token on the gagdet URL and swap that out for a longer lived token on the
> Shindig side -- but it still seems that there would be at least some small
> window on page load where the short lived token could still be sniffed/used
> by another gadget on the page for evil.
>
> As for the RPC token -- is that used for anything sensitive?  Do I need to
> be concerned about it being sniffed by other gadgets on the page?
>
> And finally -- I'm still trying to understand what I need to do on the
> Shindig side WRT locked domain support.  I see options in the shindig config
> file to enable locked domain, but it seems like most of the work would be on
> the container side in generating the iframe url's on unique domains...
>
> Please see the rest of this thread for more details.
>
> Any feedback at all on this would be appreciated...
>
> Thanks!
>
> --Jesse
>
>
> On Fri, Dec 4, 2009 at 10:46 AM, Jesse <js...@gmail.com> wrote:
>
>> I've experimented a bit more with locked domain by just setting up my
>> container to prepend a counter to the iframe url for each gadget -- so I end
>> up with iframe urls which look something like:
>>
>> 1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
>> 2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
>> 3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
>> ...
>>
>> (I realize that the urls should stay consistent across page renders for
>> the same gadget, but for testing purposes I just wanted to get the gadgets
>> on unique domains)
>>
>> Everything seems to work just fine with this setup, and I do indeed see
>> additional sandboxing happening in the browser.  I tested this by doing some
>> experimentation in FireFox using FireBug -- before implementing the locked
>> domain changes all of these statements in FireBug execute successfully:
>>
>> cd($("remote_iframe_0").contentWindow);   //change FireBug's context to
>> the window of one of the gadgets
>> console.log(document.location);   //make sure the context is now what I
>> think it should be
>> var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
>> reference to the window of another gadget
>> console.log(otherWindow.location);   //log its location to be sure I have
>> a hold of the iframe I think I do
>> console.log(otherWindow.document.getElementById('someElement').textContent);
>> //log out the content of an element in the other gadget
>>
>> and as soon as I implement locked domain as described above, the last
>> statement fails with a security exception (as expected).
>>
>> But I'm still left with a few questions...
>>
>> 1) Even with the gadgets rendering on their own domains, I can still get
>> to the location property of other gadgets on the page, which means I can
>> parse out things like the security tokens and rpc tokens -- doesn’t that
>> mean one gadget can still spoof another?  How do other containers handle
>> this?
>>
>> 2) I'm still trying to understand what all the locked domain code in
>> Shindig is for...  I can see how some of it would be useful in my container
>> (like the logic in HashLockedDomainService that generates the iframe url
>> based on a hash of the gadget url) -- so is that why it’s there?  To be use
>> on the container side?  But then I also see methods in the
>> LockedDomainService interface that definitely seem Shindig specific like
>> isSafeForOpenProxy and gadgetCanRender, so I'm sure there must be more to it
>> that I'm not understanding...  I've spent a bunch of time digging in the
>> Shindig codebase and searching through the mailing list looking for tidbits
>> but haven’t made much progress so far...
>>
>> Any help or pointers would be greatly appreciated!
>> On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:
>>
>>> I would like to start rendering my gadgets on unique locked domains but
>>> am getting a bit confused about where to start...
>>>
>>> Since the container generates the gadget iframe urls, it would seem that
>>> implementing locked domain would be a container side affair, but I see
>>> configuration options for locked domain support in the shindig container.js
>>> file and I also see associated Java classes that do a lot of locked domain
>>> work in the Shindig codebase (Java Shindig 1.0 release)...
>>>
>>> So is there actually work that needs to be done on both the container
>>> side and the Shindig side to enable locked domain gadgets?
>>>
>>> Any pointers on how to get started (even high level) would be helpful and
>>> much appreciated.
>>>
>>> Thanks!
>>>
>>
>>
>

Re: Locked Domain

Posted by Jesse <js...@gmail.com>.
I'm thinking the answer WRT the security token is to use a short lived token
on the gagdet URL and swap that out for a longer lived token on the Shindig
side -- but it still seems that there would be at least some small window on
page load where the short lived token could still be sniffed/used by another
gadget on the page for evil.

As for the RPC token -- is that used for anything sensitive?  Do I need to
be concerned about it being sniffed by other gadgets on the page?

And finally -- I'm still trying to understand what I need to do on the
Shindig side WRT locked domain support.  I see options in the shindig config
file to enable locked domain, but it seems like most of the work would be on
the container side in generating the iframe url's on unique domains...

Please see the rest of this thread for more details.

Any feedback at all on this would be appreciated...

Thanks!

--Jesse

On Fri, Dec 4, 2009 at 10:46 AM, Jesse <js...@gmail.com> wrote:

> I've experimented a bit more with locked domain by just setting up my
> container to prepend a counter to the iframe url for each gadget -- so I end
> up with iframe urls which look something like:
>
> 1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> 2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> 3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> ...
>
> (I realize that the urls should stay consistent across page renders for the
> same gadget, but for testing purposes I just wanted to get the gadgets on
> unique domains)
>
> Everything seems to work just fine with this setup, and I do indeed see
> additional sandboxing happening in the browser.  I tested this by doing some
> experimentation in FireFox using FireBug -- before implementing the locked
> domain changes all of these statements in FireBug execute successfully:
>
> cd($("remote_iframe_0").contentWindow);   //change FireBug's context to the
> window of one of the gadgets
> console.log(document.location);   //make sure the context is now what I
> think it should be
> var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
> reference to the window of another gadget
> console.log(otherWindow.location);   //log its location to be sure I have a
> hold of the iframe I think I do
> console.log(otherWindow.document.getElementById('someElement').textContent);
> //log out the content of an element in the other gadget
>
> and as soon as I implement locked domain as described above, the last
> statement fails with a security exception (as expected).
>
> But I'm still left with a few questions...
>
> 1) Even with the gadgets rendering on their own domains, I can still get to
> the location property of other gadgets on the page, which means I can parse
> out things like the security tokens and rpc tokens -- doesn’t that mean one
> gadget can still spoof another?  How do other containers handle this?
>
> 2) I'm still trying to understand what all the locked domain code in
> Shindig is for...  I can see how some of it would be useful in my container
> (like the logic in HashLockedDomainService that generates the iframe url
> based on a hash of the gadget url) -- so is that why it’s there?  To be use
> on the container side?  But then I also see methods in the
> LockedDomainService interface that definitely seem Shindig specific like
> isSafeForOpenProxy and gadgetCanRender, so I'm sure there must be more to it
> that I'm not understanding...  I've spent a bunch of time digging in the
> Shindig codebase and searching through the mailing list looking for tidbits
> but haven’t made much progress so far...
>
> Any help or pointers would be greatly appreciated!
> On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:
>
>> I would like to start rendering my gadgets on unique locked domains but am
>> getting a bit confused about where to start...
>>
>> Since the container generates the gadget iframe urls, it would seem that
>> implementing locked domain would be a container side affair, but I see
>> configuration options for locked domain support in the shindig container.js
>> file and I also see associated Java classes that do a lot of locked domain
>> work in the Shindig codebase (Java Shindig 1.0 release)...
>>
>> So is there actually work that needs to be done on both the container side
>> and the Shindig side to enable locked domain gadgets?
>>
>> Any pointers on how to get started (even high level) would be helpful and
>> much appreciated.
>>
>> Thanks!
>>
>
>

Re: Locked Domain

Posted by Kevin Brown <et...@google.com>.
On Fri, Dec 11, 2009 at 12:51 PM, Jesse <js...@gmail.com> wrote:

> Thanks Kevin -- your comments got me going in the right direction.  I was
> testing things out using FireBug's console.log function, and apparently
> console.log was letting me get away with things that I'm not able
> to otherwise.  I was able to use console.log to log out the location of
> another cross-domain iframe on the page, but trying to do something else
> with it like send the location to an alert box threw a security exception.
>
> The only other thing I'm still not clear on is what's happening on
> the Shindig side to enforce locked-domain -- I think I understand two out
> of
> the three methods in the Shindig LockedDomainService interface:
>
> gadgetCanRender -- I think this method is here to keep gadgets from doing
> things like dynamically adding a child iframe to themselves to render on
> another gadgets locked domain (since the locked domains are well known
> since
> they are a hash of the gadget URL).
>
> getLockedDomainForGadget -- Self explanatory...
>
> isSafeForOpenProxy -- This one I havent quite figure out yet.  I plan to
> spend some time next week trying to understand what the default
> HashLockedDomainService is doing here, but if someone can say a few quick
> words in the meantime about what this method is for, I'd appreciate it...
>

"open proxy" requests refer to requests that pass through /gadgets/proxy.
These resources do not need to be served on the same domain as the gadget
itself, and in fact it is a big performance win for them to be served off of
some small set of static hosts (a typical pattern might be www[0...n].
gadgethost.com).

IMO most of this should be enforced with filters rather than in the
servlets, but the argument against this was that it's too easy to screw up.


>
> Thanks!
>
> --Jesse
>
> On Wed, Dec 9, 2009 at 2:53 PM, Kevin Brown <et...@google.com> wrote:
>
> > On Fri, Dec 4, 2009 at 7:46 AM, Jesse <js...@gmail.com> wrote:
> >
> > > I've experimented a bit more with locked domain by just setting up my
> > > container to prepend a counter to the iframe url for each gadget -- so
> I
> > > end
> > > up with iframe urls which look something like:
> > >
> > > 1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > > 2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > > 3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > > ...
> > >
> > > (I realize that the urls should stay consistent across page renders for
> > the
> > > same gadget, but for testing purposes I just wanted to get the gadgets
> on
> > > unique domains)
> > >
> > > Everything seems to work just fine with this setup, and I do indeed see
> > > additional sandboxing happening in the browser.  I tested this by doing
> > > some
> > > experimentation in FireFox using FireBug -- before implementing the
> > locked
> > > domain changes all of these statements in FireBug execute successfully:
> > >
> > > cd($("remote_iframe_0").contentWindow);   //change FireBug's context to
> > the
> > > window of one of the gadgets
> > > console.log(document.location);   //make sure the context is now what I
> > > think it should be
> > > var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
> > > reference to the window of another gadget
> > > console.log(otherWindow.location);   //log its location to be sure I
> have
> > a
> > > hold of the iframe I think I do
> > >
> > >
> >
> console.log(otherWindow.document.getElementById('someElement').textContent);
> > > //log out the content of an element in the other gadget
> > >
> > > and as soon as I implement locked domain as described above, the last
> > > statement fails with a security exception (as expected).
> > >
> > > But I'm still left with a few questions...
> > >
> > > 1) Even with the gadgets rendering on their own domains, I can still
> get
> > to
> > > the location property of other gadgets on the page, which means I can
> > parse
> > > out things like the security tokens and rpc tokens -- doesn’t that mean
> > one
> > > gadget can still spoof another?  How do other containers handle this?
> > >
> >
> > If you have locked domain implemented correctly, you can't read the
> > location
> > property. It's write-only on all major browsers that implement the same
> > origin policy (that is to say, all browsers that anyone uses). If you've
> > found an instance where you can read the property, it's either a browser
> > bug, or you aren't actually rendering each gadget on a unique host.
> >
> > Make sure that each domain is on a unique host name, and that the gadgets
> > aren't modifying the value of document.domain.
> >
> >
> > >
> > > 2) I'm still trying to understand what all the locked domain code in
> > > Shindig
> > > is for...  I can see how some of it would be useful in my container
> (like
> > > the logic in HashLockedDomainService that generates the iframe url
> based
> > on
> > > a hash of the gadget url) -- so is that why it’s there?  To be use on
> the
> > > container side?  But then I also see methods in the LockedDomainService
> > > interface that definitely seem Shindig specific like isSafeForOpenProxy
> > and
> > > gadgetCanRender, so I'm sure there must be more to it that I'm not
> > > understanding...  I've spent a bunch of time digging in the Shindig
> > > codebase
> > > and searching through the mailing list looking for tidbits but haven’t
> > made
> > > much progress so far...
> > >
> > > Any help or pointers would be greatly appreciated!
> > > On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:
> > >
> > > > I would like to start rendering my gadgets on unique locked domains
> but
> > > am
> > > > getting a bit confused about where to start...
> > > >
> > > > Since the container generates the gadget iframe urls, it would seem
> > that
> > > > implementing locked domain would be a container side affair, but I
> see
> > > > configuration options for locked domain support in the shindig
> > > container.js
> > > > file and I also see associated Java classes that do a lot of locked
> > > domain
> > > > work in the Shindig codebase (Java Shindig 1.0 release)...
> > > >
> > > > So is there actually work that needs to be done on both the container
> > > side
> > > > and the Shindig side to enable locked domain gadgets?
> > > >
> > > > Any pointers on how to get started (even high level) would be helpful
> > and
> > > > much appreciated.
> > > >
> > > > Thanks!
> > > >
> > >
> >
>

Re: Locked Domain

Posted by Jesse <js...@gmail.com>.
Thanks Kevin -- your comments got me going in the right direction.  I was
testing things out using FireBug's console.log function, and apparently
console.log was letting me get away with things that I'm not able
to otherwise.  I was able to use console.log to log out the location of
another cross-domain iframe on the page, but trying to do something else
with it like send the location to an alert box threw a security exception.

The only other thing I'm still not clear on is what's happening on
the Shindig side to enforce locked-domain -- I think I understand two out of
the three methods in the Shindig LockedDomainService interface:

gadgetCanRender -- I think this method is here to keep gadgets from doing
things like dynamically adding a child iframe to themselves to render on
another gadgets locked domain (since the locked domains are well known since
they are a hash of the gadget URL).

getLockedDomainForGadget -- Self explanatory...

isSafeForOpenProxy -- This one I havent quite figure out yet.  I plan to
spend some time next week trying to understand what the default
HashLockedDomainService is doing here, but if someone can say a few quick
words in the meantime about what this method is for, I'd appreciate it...

Thanks!

--Jesse

On Wed, Dec 9, 2009 at 2:53 PM, Kevin Brown <et...@google.com> wrote:

> On Fri, Dec 4, 2009 at 7:46 AM, Jesse <js...@gmail.com> wrote:
>
> > I've experimented a bit more with locked domain by just setting up my
> > container to prepend a counter to the iframe url for each gadget -- so I
> > end
> > up with iframe urls which look something like:
> >
> > 1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > 2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > 3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> > ...
> >
> > (I realize that the urls should stay consistent across page renders for
> the
> > same gadget, but for testing purposes I just wanted to get the gadgets on
> > unique domains)
> >
> > Everything seems to work just fine with this setup, and I do indeed see
> > additional sandboxing happening in the browser.  I tested this by doing
> > some
> > experimentation in FireFox using FireBug -- before implementing the
> locked
> > domain changes all of these statements in FireBug execute successfully:
> >
> > cd($("remote_iframe_0").contentWindow);   //change FireBug's context to
> the
> > window of one of the gadgets
> > console.log(document.location);   //make sure the context is now what I
> > think it should be
> > var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
> > reference to the window of another gadget
> > console.log(otherWindow.location);   //log its location to be sure I have
> a
> > hold of the iframe I think I do
> >
> >
> console.log(otherWindow.document.getElementById('someElement').textContent);
> > //log out the content of an element in the other gadget
> >
> > and as soon as I implement locked domain as described above, the last
> > statement fails with a security exception (as expected).
> >
> > But I'm still left with a few questions...
> >
> > 1) Even with the gadgets rendering on their own domains, I can still get
> to
> > the location property of other gadgets on the page, which means I can
> parse
> > out things like the security tokens and rpc tokens -- doesn’t that mean
> one
> > gadget can still spoof another?  How do other containers handle this?
> >
>
> If you have locked domain implemented correctly, you can't read the
> location
> property. It's write-only on all major browsers that implement the same
> origin policy (that is to say, all browsers that anyone uses). If you've
> found an instance where you can read the property, it's either a browser
> bug, or you aren't actually rendering each gadget on a unique host.
>
> Make sure that each domain is on a unique host name, and that the gadgets
> aren't modifying the value of document.domain.
>
>
> >
> > 2) I'm still trying to understand what all the locked domain code in
> > Shindig
> > is for...  I can see how some of it would be useful in my container (like
> > the logic in HashLockedDomainService that generates the iframe url based
> on
> > a hash of the gadget url) -- so is that why it’s there?  To be use on the
> > container side?  But then I also see methods in the LockedDomainService
> > interface that definitely seem Shindig specific like isSafeForOpenProxy
> and
> > gadgetCanRender, so I'm sure there must be more to it that I'm not
> > understanding...  I've spent a bunch of time digging in the Shindig
> > codebase
> > and searching through the mailing list looking for tidbits but haven’t
> made
> > much progress so far...
> >
> > Any help or pointers would be greatly appreciated!
> > On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:
> >
> > > I would like to start rendering my gadgets on unique locked domains but
> > am
> > > getting a bit confused about where to start...
> > >
> > > Since the container generates the gadget iframe urls, it would seem
> that
> > > implementing locked domain would be a container side affair, but I see
> > > configuration options for locked domain support in the shindig
> > container.js
> > > file and I also see associated Java classes that do a lot of locked
> > domain
> > > work in the Shindig codebase (Java Shindig 1.0 release)...
> > >
> > > So is there actually work that needs to be done on both the container
> > side
> > > and the Shindig side to enable locked domain gadgets?
> > >
> > > Any pointers on how to get started (even high level) would be helpful
> and
> > > much appreciated.
> > >
> > > Thanks!
> > >
> >
>

Re: Locked Domain

Posted by Kevin Brown <et...@google.com>.
On Fri, Dec 4, 2009 at 7:46 AM, Jesse <js...@gmail.com> wrote:

> I've experimented a bit more with locked domain by just setting up my
> container to prepend a counter to the iframe url for each gadget -- so I
> end
> up with iframe urls which look something like:
>
> 1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> 2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> 3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
> ...
>
> (I realize that the urls should stay consistent across page renders for the
> same gadget, but for testing purposes I just wanted to get the gadgets on
> unique domains)
>
> Everything seems to work just fine with this setup, and I do indeed see
> additional sandboxing happening in the browser.  I tested this by doing
> some
> experimentation in FireFox using FireBug -- before implementing the locked
> domain changes all of these statements in FireBug execute successfully:
>
> cd($("remote_iframe_0").contentWindow);   //change FireBug's context to the
> window of one of the gadgets
> console.log(document.location);   //make sure the context is now what I
> think it should be
> var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
> reference to the window of another gadget
> console.log(otherWindow.location);   //log its location to be sure I have a
> hold of the iframe I think I do
>
> console.log(otherWindow.document.getElementById('someElement').textContent);
> //log out the content of an element in the other gadget
>
> and as soon as I implement locked domain as described above, the last
> statement fails with a security exception (as expected).
>
> But I'm still left with a few questions...
>
> 1) Even with the gadgets rendering on their own domains, I can still get to
> the location property of other gadgets on the page, which means I can parse
> out things like the security tokens and rpc tokens -- doesn’t that mean one
> gadget can still spoof another?  How do other containers handle this?
>

If you have locked domain implemented correctly, you can't read the location
property. It's write-only on all major browsers that implement the same
origin policy (that is to say, all browsers that anyone uses). If you've
found an instance where you can read the property, it's either a browser
bug, or you aren't actually rendering each gadget on a unique host.

Make sure that each domain is on a unique host name, and that the gadgets
aren't modifying the value of document.domain.


>
> 2) I'm still trying to understand what all the locked domain code in
> Shindig
> is for...  I can see how some of it would be useful in my container (like
> the logic in HashLockedDomainService that generates the iframe url based on
> a hash of the gadget url) -- so is that why it’s there?  To be use on the
> container side?  But then I also see methods in the LockedDomainService
> interface that definitely seem Shindig specific like isSafeForOpenProxy and
> gadgetCanRender, so I'm sure there must be more to it that I'm not
> understanding...  I've spent a bunch of time digging in the Shindig
> codebase
> and searching through the mailing list looking for tidbits but haven’t made
> much progress so far...
>
> Any help or pointers would be greatly appreciated!
> On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:
>
> > I would like to start rendering my gadgets on unique locked domains but
> am
> > getting a bit confused about where to start...
> >
> > Since the container generates the gadget iframe urls, it would seem that
> > implementing locked domain would be a container side affair, but I see
> > configuration options for locked domain support in the shindig
> container.js
> > file and I also see associated Java classes that do a lot of locked
> domain
> > work in the Shindig codebase (Java Shindig 1.0 release)...
> >
> > So is there actually work that needs to be done on both the container
> side
> > and the Shindig side to enable locked domain gadgets?
> >
> > Any pointers on how to get started (even high level) would be helpful and
> > much appreciated.
> >
> > Thanks!
> >
>

Re: Locked Domain

Posted by Jesse <js...@gmail.com>.
I've experimented a bit more with locked domain by just setting up my
container to prepend a counter to the iframe url for each gadget -- so I end
up with iframe urls which look something like:

1.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
2.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
3.gadgets.mydomain.com/gadgets/ifr?url=...&st=...&rpctoken=...&...
...

(I realize that the urls should stay consistent across page renders for the
same gadget, but for testing purposes I just wanted to get the gadgets on
unique domains)

Everything seems to work just fine with this setup, and I do indeed see
additional sandboxing happening in the browser.  I tested this by doing some
experimentation in FireFox using FireBug -- before implementing the locked
domain changes all of these statements in FireBug execute successfully:

cd($("remote_iframe_0").contentWindow);   //change FireBug's context to the
window of one of the gadgets
console.log(document.location);   //make sure the context is now what I
think it should be
var otherWindow = window.parent.frames["remote_iframe_1"];   //grab a
reference to the window of another gadget
console.log(otherWindow.location);   //log its location to be sure I have a
hold of the iframe I think I do
console.log(otherWindow.document.getElementById('someElement').textContent);
//log out the content of an element in the other gadget

and as soon as I implement locked domain as described above, the last
statement fails with a security exception (as expected).

But I'm still left with a few questions...

1) Even with the gadgets rendering on their own domains, I can still get to
the location property of other gadgets on the page, which means I can parse
out things like the security tokens and rpc tokens -- doesn’t that mean one
gadget can still spoof another?  How do other containers handle this?

2) I'm still trying to understand what all the locked domain code in Shindig
is for...  I can see how some of it would be useful in my container (like
the logic in HashLockedDomainService that generates the iframe url based on
a hash of the gadget url) -- so is that why it’s there?  To be use on the
container side?  But then I also see methods in the LockedDomainService
interface that definitely seem Shindig specific like isSafeForOpenProxy and
gadgetCanRender, so I'm sure there must be more to it that I'm not
understanding...  I've spent a bunch of time digging in the Shindig codebase
and searching through the mailing list looking for tidbits but haven’t made
much progress so far...

Any help or pointers would be greatly appreciated!
On Thu, Dec 3, 2009 at 9:55 AM, jss9920 <js...@gmail.com> wrote:

> I would like to start rendering my gadgets on unique locked domains but am
> getting a bit confused about where to start...
>
> Since the container generates the gadget iframe urls, it would seem that
> implementing locked domain would be a container side affair, but I see
> configuration options for locked domain support in the shindig container.js
> file and I also see associated Java classes that do a lot of locked domain
> work in the Shindig codebase (Java Shindig 1.0 release)...
>
> So is there actually work that needs to be done on both the container side
> and the Shindig side to enable locked domain gadgets?
>
> Any pointers on how to get started (even high level) would be helpful and
> much appreciated.
>
> Thanks!
>