167475 - [URL] Disable external and returning no data protocol handlers in all cases, excluding <A HREF=>

Status: VERIFIED FIXED

(Core :: DOM: Navigation, defect)

Description

[Manko]

As we can see in bug 163648, external protocols can cause a lot of security
issues. But exploits for this bug are dangerous mainly if external protocol
handler is being requested automatically from HTML code via <IMG
SRC="externalprotocol:URL">, <IFRAME SRC="externalprotocol:URL"> and other
similar cases.

More, with relation to common sense, invoking an external protocol is absurd in
this case, because <ANYTAG SRC="..."> is request to return some data in browser,
not for launch external application.

So, disable external protocols in all cases, excluding <A HREF=>, can solve this
problem.

Marking severity critical according to 163648.

Comment 1

[Manko]

Another way to exploit URL with external protocols might be using Javascript.
Therefore, we must disaple external protocols in JS operators:
window.open()
location.href=
location.replace()
etc.

Comment 2

[Jesse Ruderman]

It's not hard for a malicious site to get a visitor to click a link.  Requiring
a click or an equivalent keyboard action can be useful for limiting how much a
web site can annoy you (pop-up windows, etc.) but I don't think it's useful for
larger security issues.

Comment 3

[Daniel Veditz [:dveditz]]

I agree, WONTFIX. Other bugs are already discussing blocking external protocol
handlers, we don't need to do additional work to base the decision on context.

Comment 4

[Manko]

I can't agree with WONTFIX. Have we at least one reason for enabling external
protocols in <IFRAME SRC> or <IMG SRC>?

Assuming that user has enabled some protocol, for example, irc: to start IRC
chat by pressing a link. He has opened web page with HTML code:
<iframe|img src="irc:some.url.there">. And he has seen IRC window to appear on
screen suddenly (without his deliberate request or confirmation) and empty
iframe (or broken image icon in case <IMG SRC=>) in page. Absurd, isn't it?

More, imagine that you have received HTML email with such code...

Comment 5

[Daniel Veditz [:dveditz]]

re-opening for reconsideration. This doesn't solve the problem of untrusted
protocols, but even for trusted ones it doesn't make much sense in these kinds
of places.

Comment 6

[Manko]

Changing severity from enhancemet to normal - it isn't enhancement anyway.

Comment 7

[Daniel Veditz [:dveditz]]

It is working as-designed, you would like a change == enhancement. enhancement
doesn't mean "less than trivial" despite the position on the severity list, it's
really a different kind of bug report. Enhancements can be quite important.

--> darin who had some ideas how we could do this, although the changes probably
aren't in necko land.

Comment 8

[Matthias Versen [:Matti]]

*** Bug 172498 has been marked as a duplicate of this bug. ***

Comment 9

[Manko]

I would propose my guess-work to solve this problem. Sorry if I don't understand
this issue correctly, this is only my imhoes and afaiks ;)

Before implementing external protocols Mozilla did follow this algorithm:
+-----------------------+    +-------------------------------------+
| User clicks on a link |    |  HTML tag refers to external object |
+-----------------------+    +-------------------------------------+
                   |               |
                   +---------------+
                          |
                          V
            /------------------------------\
           /                                \
  +--yes--< Is this is an internal protocol? >--no---+
  |        \                                /        |
  |         \------------------------------/         |
  |                                                  |
  V                                                  V
+----------------------------+                  +--------+
| Call Necko to request data |                  | Ignore |
+----------------------------+                  +--------+


This algorithm Mozilla does follow now:
+-----------------------+    +-------------------------------------+
| User clicks on a link |    | HTML object refers to external data |
+-----------------------+    +-------------------------------------+
                   |               |
                   +---------------+
                          |
                          V
            /------------------------------\
           /                                \
  +--yes--< Is this is an internal protocol? >--no--+
  |        \                                /       |
  |         \------------------------------/        |
  |                                                 |
  V                                                 V
+----------------------------+      +--------------------------------+
| Call Necko to request data |      | Call external protocol handler |
+----------------------------+      +--------------------------------+


Proposed algorithm:
+-----------------------+        +-------------------------------------+
| User clicks on a link |        | HTML object refers to external data |
+-----------------------+        +-------------------------------------+
             |                                                    |
             V                                                    |
            /---------------------------\                         |
           /                             \                        |
  +--yes--< Is this an internal protocol? >--no--+                |
  |        \                             /       |                |
  |         \---------------------------/        V                |
  |                                       +-------------------+   |
  |                                       | User confirmation |   |
  |                                       | (see bug 167473)  |   |
  |                                       +-------------------+   |
  |                                              |                |
  |                                              V                |
  |                          +--------------------------------+   |
  |                          | Call external protocol handler |   |
  |                          +--------------------------------+   |
  |                                                               |
  |                                  +--------------------------- +
  |                                  |
  |                                  V
  |             /------------------------------\
  |            /                                \
  |   +--yes--< Is this is an internal protocol? >--no--+
  |   |        \                                /       |
  |   |         \------------------------------/        |
  |   |                                                 |
  V   V                                                 V
+----------------------------+                     +--------+
| Call Necko to request data |                     | Ignore |
+----------------------------+                     +--------+

How do you think, is it correct?

Comment 10

[Darin Fisher]

this sounds like 1.2 fodder, no?

Comment 11

[Matthew Middleton (:zzxc)]

This is very important to be fixed ASAP with the recent windows xp flaw that
allows an hcp protocol request to delete any file on your hard disk, wildcards
allowed.  See bug 172498 for an explanation.  Note: that bug was marked as a
dupe of this broader one.  An exploit which will delete everything in
"c:\delthis" with an image tag is included on that bug.

Comment 12

[Darin Fisher]

now that we have a suitable blacklist in place (bug 163648 has been fixed), this
bug is much less urgent.  pushing out to moz 1.3.

Comment 13

[Darin Fisher]

cc'ing bzbarsky since he's done some recent work w/ the unknown protocol handler.

Comment 14

[Boris Zbarsky [:bzbarsky]]

Hmm.... Link clicks call InternalLoad directly.  So we could throttle this in
nsDocShell::LoadURI, perhaps.  Of course if the user types such a URI in the URL
bar we had better load it...

Comment 15

[Darin Fisher]

-> future (if someone wants to own this, please let me know!)

Comment 16

[benc]

-> docshell per bz's comment.

Comment 17

[Boris Zbarsky [:bzbarsky]]

Docshell can't throttle <img> loads.  We need a more general solution.

Comment 18

[Adam Hauner]

*** Bug 243503 has been marked as a duplicate of this bug. ***

Comment 19

[Boris Zbarsky [:bzbarsky]]

*** Bug 244205 has been marked as a duplicate of this bug. ***

Comment 20

[Boris Zbarsky [:bzbarsky]]

This wouldn't be too hard to do with the flag proposed in bug 229168 -- we'd
just not load things with that flag set in places where content is expected
(most places).

Comment 21

[Tom Graham]

Requesting review for 1.0

Comment 22

[Manko]

Changing priority according bug 250180: we'll have more security issues until
tyhis will be fixed.

Changing dependencies: bug 229168 is depending on this, not vice versa.

Comment 23

[Manko]

Oh, sorry, logically i'm right, but bug 229168 have a patch to do this. Begging
your pardons.

Comment 24

[penguinista]

Why not add a whitelist for external handlers, similar to the cookie 
whitelist? The prompting could be similar too... For example: 
 
The site www.blah.com has requested to open external protocol shell. 
[ ] Use my choice for all external protocol requests from this site. 
 
[Show Details] [Allow] [Allow for Session] [Deny] 
 
 
Clicking "Show Details" might provide the argument information and anything 
else that would be relevant. 
 
Thoughts? This would seem to solve all of the external protocol handler issues 
in one fell swoop, IMO. :) 
 

Comment 25

[Andrew Hagen]

Whitelist is bug 173010.

Comment 26

[Daniel Veditz [:dveditz]]

Adding whiteboard comment to hopefully redirect folks to the real whitelist bug.

assigning to me since I don't think Adam's going to be working on this.

Comment 27

[Jesse Ruderman]

See also bug 334426 (especially comments 2 and 3) and bug 266325.

Comment 28

[Daniel Veditz [:dveditz]]

(In reply to comment #27)
> See also bug 334426 (especially comments 2 and 3) and bug 266325.

Thanks Jesse, this is the one I was thinking of.

How is bug 266325 relevant? executables-in-frames generally use http which wouldn't be covered here, plus we *want* it to work since some legitimate sites assume that behavior for their normal download schemes. (The real problem with 266325/284282 is that we preserve the .exe extension on the tmp file. We should give it a bogus extension to prevent accidents in case someone stumbles on the file before we can clean it up, and then rename when the user approves the download.)

Comment 29

[Manko]

Of course, bug 266325 is irrelevant, but similar. I have opened bug 334644 to resolve the problem described in 266325.

Comment 30

[Manko]

Please add P1 because this has a lot of critical dependencies

Comment 31

[Jesse Ruderman]

I'm changing the severity to "critical" because this bug can be -- and is -- used to perform DoS-like attacks against web users.  (See bug 334426 for a common shock page that abuses this bug.)  I'm also clearing the priority and target milestone fields, which were set before dveditz took the bug and before Last Measure came into existence.  I'm not setting those fields to new values because I'm not the owner.

Is it possible that fixing this is a simple matter of adding a CheckForAbusePoint call, like bug 355482 was? The popup blocking machinery seems to be appropriate here as a way of limiting what user actions can trigger external protocol handlers or previously-unused-protocol dialogs.  I'm not sure that machinery provides a way to limit the number of possibly abusive actions *per click*, but if it doesn't then it should.

Comment 32

[Manko]

Jesse, I guess that we should separate a codeflow between "click a link" event and <ANYTAG SRC=> reference. It still seems to me that comment #9 should be a proper solution.

Comment 33

[Manko]

Changing summary. An internal-handled protocols returning no data (e.g. mailto: or irc: in Seamonkey) should be also denied in <ANYTAG SRC=>.

Comment 34

[Dan Mosedale (:dmosedale, :dmose)]

The addition of web-based protocol handlers changes the landscape here a bit: <iframe src="mailto:foo"> could actually be meaningful/useful in some cases.

Comment 35

[Manko]

Dan, what cases do you mean?

Comment 36

[Dan Mosedale (:dmosedale, :dmose)]

Well, instead of writing your own feedback form, as many sites do today, one could imagine using <iframe src="mailto:[e-mail]>, since this would open up the compose page in your webmail provider in that iframe.  This case might be particularly useful to ISPs who could guarantee that all their users have access to webmail support this.

Comment 37

[Jesse Ruderman]

I don't think we should go out of our way to support <iframe src="mailto:"> for webmail.  It would encourage phishing, and turning off third-party cookies would break it.

If an ISP knows what webmail you use they can just embed it with an http URL.

Comment 38

[Manko]

Agreed with Jessy.

If "mailto:" URLs would be handled by webmail in my system, I would prefer click to email link and see a new window/tab rather than compose a message in the iframe, even at trusted site. But at untrusted one?..

Comment 39

[Dan Mosedale (:dmosedale, :dmose)]

Folks interested in this bug may also be interested in attending the security/design review of the web-based protocol handler: <http://groups.google.com/group/mozilla.dev.planning/browse_thread/thread/d7e55195eafea8cc/693cd7bc73167aea#693cd7bc73167aea>.

Comment 40

[Jeremy Nickurak]

Ran across another "Last Measure" browser exploit.

Careful opening it. It redirects from google (somehow):

http://www.google.com/url?sa=t&source=web&ct=res&cd=1&url=http%3A%2F%2Fmembers.on.nimp.org%2F%3Fu%3Dtimecop&ei=UcVHSpWdItCvtwfnjsCMCg&rct=j&q=nimp+timecop&usg=AFQjCNHg4SY1IP4BptziBA5eGd-gxcxlLg

Comment 42

[Jez]

GNAA 'shock site' pages have a habit of seriously abusing this.  I think this
should be a priority, maybe for Firefox 4.  I had 100+ thunderbird e-mail
composition windows opening up.

Comment 43

[timeless]

Attachment: draft

this seems to make bug 553609 with noscript blocking Java (and flash) tolerable.

note that there are typically many problems w/ such sites:
1. use of various protocol handlers (this bug)
2. trying to download random file types (some other bug)
3. spawning java/flash applets (other bugs)
4. randomly moving windows around (other bugs, already controllable by a preference)

Comment 45

[O. Atsushi (Torisugari)]

This bug is waiting for something? The patch does not seem checked in.

Comment 46

[timeless]

Torisugari, samuel: assuming it still applies (someone should talk to the try-server), it just needs someone to land it.

When this patch received review, I was working for a vendor that used a competing layout engine, so I was generally avoiding bugzilla and certainly not touching Gecko. That is no longer the case.

Comment 47

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Given that the review is 5 years old at this point, it'd probably be proper to get a new review though. (Not from me, I don't know this code well enough, nor am I a peer)

Comment 50

[Tom Ritter [:tjr]]

This is not actually fingerprinting. Fingerprinting external protocols is Bug 680300 - this is about the DOS capability of auto-triggering them to be launched.

Comment 51

[Andrea Marchesini [:baku]]

Attachment: patch updated

This patch blocks the opening of external-protocol URLs if the loading has been flagged as LOAD_DOCUMENT_URI: documents cannot be handled by external resources.

Comment 52

[Andrea Marchesini [:baku]]

Comment on attachment 9026700 [details] [diff] [review]
patch updated

I have a better approach than this.

Comment 53

[Andrea Marchesini [:baku]]

Attachment: iframes should not load non-data URLs

Comment 54

[Andrea Marchesini [:baku]]

Note that we already don't allow loading external protocol URLs in <img> and <script> elements.

Comment 55

[Olli Pettay [:smaug][[e-mail]]]

Comment on attachment 9026737 [details] [diff] [review]
iframes should not load non-data URLs

Why we think this is web compatible?
data:text/html,<iframe src="mailto:[e-mail]"> in some other browsers (Chrome at least) is loaded.
I could easily see some web sites relying on the existing behavior.

Can't r+ without some more data here.

Comment 56

[Andrea Marchesini [:baku]]

> Why we think this is web compatible?

I don't have an answer here. What I can do, is to add telemetry and count how often iframes are loaded with URI_DOES_NOT_RETURN_DATA urls. Is this OK for you?

Comment 57

[Olli Pettay [:smaug][[e-mail]]]

That gives at least some data, so yes please :)
But would be good to understand better what other browsers do too.

Comment 58

[Andrea Marchesini [:baku]]

Attachment: telemetry.patch

Comment 59

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/8810195a2a42
Telemetry to count how often iframes load no-data URLs, r=smaug

Comment 60

[Andrea Marchesini [:baku]]

I just landed a telemetry patch. Let's wait 1 or 2 cycles before the next steps.
I see 2 possible approaches:

1. if the number of iframes with no-data URLs is almost 0, we can land my previous patch. See comment 53 and 55.

2. if we need to support iframes with no-data URLs, we can use the popup blocking algorithm, allowing just the first iframe to load and blocking the next ones.

Comment 61

[Stefan Hindli [:stefan_hindli]]

https://hg.mozilla.org/mozilla-central/rev/8810195a2a42

Comment 62

[Andrea Marchesini [:baku]]

I think I can steal this bug.

Comment 63

[Manko]

What about no-data <IMG SRC="..." and <SCRIPT SRC="..."?

Comment 64

[Andrea Marchesini [:baku]]

HTMLImageElement and HTMLScriptElement are already ignoring external protocols and no-data URLs.

Comment 65

[Manko]

<iframe src="mailto:[e-mail]"> generates popup mail message window with To, CC, Subject, Body etc fields as described in RFC 6068 indeliberately, which possibly may be easily scammed. I strongly doubt that this opportunity is used by respectable sites.

So, collecting statistics we should take into account the degree of site respectability.

Comment 66

[Andrea Marchesini [:baku]]

Comment on attachment 9026737 [details] [diff] [review]
iframes should not load non-data URLs

Review of attachment 9026737 [details] [diff] [review]:
-----------------------------------------------------------------

https://telemetry.mozilla.org/new-pipeline/dist.html#!cumulative=0&end_date=2018-11-29&include_spill=0&keys=__none__!__none__!__none__&max_channel_version=aurora%252F64&measure=USE_COUNTER2_NO_DATA_URL_DOCUMENT&min_channel_version=nightly%252F62&processType=*&product=Firefox&sanitize=1&sort_keys=submissions&start_date=2018-10-15&table=0&trim=1&use_submission_date=0

We are having the first telemetry results about the number of iframes loading no-data-URLs.

Comment 67

[Andrea Marchesini [:baku]]

Attachment: iframes should not load non-data URLs

Comment 68

[Olli Pettay [:smaug][[e-mail]]]

Comment on attachment 9029792 [details] [diff] [review]
iframes should not load non-data URLs

But please don't land before the next merge.

Comment 69

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/a5ad58077bd8
iframes should load just URLs able to return data, r=smaug

Comment 70

[Cristina Coroiu [:ccoroiu]]

https://hg.mozilla.org/mozilla-central/rev/a5ad58077bd8

Comment 73

[Manko]

17 years... Great thanks for all!

Comment 74

[Boris Zbarsky [:bzbarsky]]

Does this fix bug 670328 too?

Comment 75

[Andrea Marchesini [:baku]]

No, it covers just the blocking of no_data_URLs for iframes. You can still do: window.location = mailto:foo.

Comment 78

[Virtual_ManPL [:Virtual] 🇵🇱 - (please needinfo? me - so I will see your comment/reply/question/etc.)]

I'm confirming that bug is fixed, starting in Mozilla Firefox Nightly 66.0a1 (2018-12-12), so I'm marking this bug as VERIFIED. Thank you very much! \o/

Comment 90

[Kohei Yoshino]

Posted site compatibility note: https://www.fxsitecompat.com/en-CA/docs/2019/loading-of-external-protocol-url-in-iframe-is-now-blocked/