Default of security.certerrors.mitm.auto_enable_enterprise_roots is compromising confidentiality of Windows users in public environments
Categories
(Core :: Security: PSM, defect)
Tracking
()
People
(Reporter: mozsecurity, Unassigned)
Details
(Keywords: sec-other)
User Agent: Mozilla/5.0 (Windows NT 6.3; Win64; x64; rv:72.0) Gecko/20100101 Firefox/72.0
Steps to reproduce:
Imagine you are attending a higher educational institution or university. You are using your BYOD and followed your admin's advice to import the Enterprise Wi-Fi CA certificate in the Windows store.
Your admin had read https://wiki.geant.org/display/H2eduroam/EAP+Server+Certificate+considerations and thus recommended Firefox to safeguard your confidentiality (as he knew that Chrome uses the Windows store and isn't trustworthy either (or maybe he only always used the Enterprise edition)).
Fragments from the eduroam wiki:
"So, as a general recommendation: if you have the required expertise, it is suggested to set up a private CA exclusively for your IdP's eduroam service."
"Your CA may incidentally yield the power on such client devices of your own user base to inspect their web or other traffic (if you actively abuse it and modify your IT infrastructure to enable this). We do not endorse or encourage this in any way."
"We recommend that you inform your users how best to restrict the power of the CA (e.g. with CA installation instructions which point to a dedicate Wi-Fi store [Android 4.3+]; or with the advice not to use browsers which use the built-in CA store of the device [MS Windows])."
Actual results:
Default of security.certerrors.mitm.auto_enable_enterprise_roots causes Windows' CA store (where the private enterprise CA certificate should be stored) to be automatically trusted since Firefox 68 and potentially enables admins (against their will) (or other MiTM) to spy on/manipulate web sessions of Windows users without their notice.
Expected results:
Unless Microsoft provides a separate dedicated Wi-Fi certificate store for Enterprise WLAN certificates you must not automatically trust the Windows certificate store. At least you have to inform users about the trusted certificates and ask them whether they really trust them for browser sessions.
| Reporter | ||
Comment 1•6 years ago
|
||
You might want to vote for a suggestion to create a dedicated Enterprise Wi-Fi certificate store on Windows (Feedback-Hub app). MSRC told me to place the suggestion on the Feedback-Hub (maybe I should have choosen bug instead of suggestion).
But better you protect innocent users from possibly malign certificates yourself.
| Reporter | ||
Comment 2•6 years ago
|
||
Please do not disclose this until another browser vendor had time to fix this issue too (which doesn't mean that you must not change the default setting in the meantime).
Comment 3•6 years ago
|
||
JC, I think you know who makes decisions about this?
Comment 5•6 years ago
|
||
It's sad that this default change makes Firefox traffic potentially less secure in this situation, but that fact that it changes our behavior to be equally at risk as the dominant browser in the marketplace and the OS default browser would seem a big hint that the problem is not the Windows Root Store but this WiFi configuration. Many enterprises require the ability to MITM traffic for monitoring purposes.
| Reporter | ||
Comment 6•6 years ago
|
||
Firefox has an own trust store. This is a potential which should be used.
For enterprises: if they use Firefox ESR on equipment owned by them, it might be preferable. But I don't think it is preferable for BYOD devices.
| Reporter | ||
Comment 7•6 years ago
|
||
The people over at Chromium think this is no bug and no security issue and closed issue 1048353 and made it viewable to public.
For me their claims say that I can't recommend their browsers for Windows (they have no trust store, they use the Windows trust store).
They claim that someone should change a property after importing the CA certificate.
If you can tell me how someone can properly implement EAP-TTLS (you can't use public CA certificates, otherwise your APs are open to everyone ...) or where the magic button to remove the trust for browsers, but not for WPA2 Enterprise (RADIUS) authentication verification I would be glad (MS details in KB814394 - I can't see how to stop the trust for browsers only).
In my opinion it would help if there was an X509v3 extension preventing browsers to trust, but Windows to accept it for WiFi only (RADIUS server verification).
| Reporter | ||
Comment 8•6 years ago
|
||
Regarding comment 5 + 6: imagine there would be an WiFi network at university operated by a governmental phone company. You probably don't want them to be able to MiTM your traffic when you are at home. If you don't have IT expertise you don't know that you would have to remove the CA certificate at home.
I don't think that it is the right way to think "others are compromising privacy, integrity, confidentiality - so there is no reason for us to protect it".
By the way: I don't know what EU GDPR lawyers would say about this.
| Reporter | ||
Comment 9•6 years ago
|
||
Probably this should be seen in a broader scope: the default makes Firefox vulnerable to things like superfish. I think it was a security decision (and not a decision of power) to not trust the OS CA certificate store but to have an own CA certificate store.
Automatically trusting the OS CA certificate store could make sense in the opposite direction: CA certificates marked as fraudulent could be treated as not trustworthy for Firefox too.
Regarding the enterprise MiTM for proxies, firewalls, ... I think the configuration JSON/group policy where enterprises can manually import CA certificates is the correct way (for enterprise owned devices). I think automatically importing the OS CA truststore was only a first attempt when there was no other way for configuring this (except by putting certX.db in special folders). Maybe you can discuss that in the ESR working group, but I think that enterprises don't want to have things like superfish.
Regarding antivirus programs using CA certificates for inspecting https traffic: yes it is convenient for home users, but this could be handled by showing users an option dialog on the first occurence of an unknown CA certificate, which is found to be trusted in the OS certificate store, with some information that she/he should accept it if it is the first run of Firefox after the installation of such a product. Maybe there would be an better option - i.e. an API which the antivirus vendor could call during setup/first start of Firefox to have Firefox open an import dialog for this certificate.
Comment 10•6 years ago
|
||
This seems more like an issue with Windows. As you said, Android has a separate wifi root store that addresses this problem.
We might look at our messaging and UI in terms of communicating the situation to users, but in general this is out of scope in terms of attacks we can mitigate, given the constraints of the operating system.
| Reporter | ||
Comment 11•6 years ago
|
||
I agree that it would be good if there was a separate wifi root store on Windows.
The only way for Firefox to distinguish the certificate would be if someone would introduce a special Certificate Usage type (i.e. Wireless-EAP) in The X.509v3 certificate extensions which admins can set and Firefox must distrust CA certificates with this extension (or maybe such an extension already exists, which doesn't apply to Windows certificate verification for wireless usage?)
As there exists a Netscape Certificate Type - 2.16.840.1.113730.1.1 (who owns that today?) and in some (old) documentation you can read
bit-4 Reserved - this bit is reserved for future use
Maybe this could be used for such CA types?
Otherwise some other X.509v3 extension might be appropriate.
But I think this issue is part of a bigger problem (unless I don't understand it correctly):
Isn't Mozillaʼs CA Certificate Program pretty useless for Windows users if you automatically trust any certificates which somehow came into the OS trust store (maybe including some leftover test certificates from some hardware vendor, ...) - making the responsible selection of trustworthy CA certificates useless?
Comment 12•6 years ago
|
||
Much of the benefit of Mozilla's root CA program is in CA transparency and ensuring adherence to the required standards.
Underlying issues with the security of the operating system are largely out of scope for our threat model. If the OS can be compromised (e.g. by malware or by someone adding a new trust anchor), there's not much that Firefox can meaningfully do to protect the user.
| Reporter | ||
Comment 13•6 years ago
|
||
Not trusting the OS trust anchor by default (as this was the case before 68)? It is Firefox blindly putting trust into the OS trust anchors (which might not adhere to the required standards) without any verification of the found non-built-in certificates.
And I wrote about the request to make it possible for admins to explicitly mark their WiFi CA certificates as non-trustable for browsing (should I move this to an own feature request). You might have to coordinate with other browser vendors to make this possible industry-wide, but I think adding a simple verification of an extension would not be so difficult to implement.
The OS isn't necessarily compromised through the trust anchor (unless some update mechanisms trust this trust anchor - I could imagine MS to have its own trust anchor hard coded into the update components).
It is clear that you can't do much about active attacks coming from inside the OS you are running on, but i.e. leftover certificates in the OS trust store from a not completely cleaned up malware infection (without the remainder of the malware being active) or untrustworthy development certificates would not necessarily harm the confidentiality of the transmitted user data when Firefox has been installed from a trustworthy source and does not trust certificates.
| Reporter | ||
Comment 14•6 years ago
|
||
Firefox does know (or has known) about X509v3 extensions for restricting the allowed DNS names in signed certificates (verified that years ago). Unfortunately I fear this is not applicable to certificates created for WiFi or VPN RADIUS client certificates (i.e. usernames) and I remember having read somewhere that the same CA has to be used for the server and the client certificates (otherwise the server would trust all certificates signed from this CA and everybody could access it) thus making it impossible to use another CA for the server certificate the clients have to trust.
If I understood something incorrectly and those restrictions on the CA do not exist, there would be no problem by having admins use those certificate naming restrictions in their selfsigned CAs (and the users verifying this one time).
| Reporter | ||
Comment 15•6 years ago
|
||
Created bug 1613977 because manual explicit distrust not working.
| Reporter | ||
Comment 16•6 years ago
|
||
As stated in comment 7 Chromium developers considered this to be no security concern and made the details public. As MSRC had told me to make a suggestion in Feedback-Hub and this seems to be public too (https://aka.ms/AA718id), there is no to keep the security-view restrictions in place (you might want to remove my previous comment of another bug id before removing view restrictions).
Currently the only solution seems to tell admins to tell Windows users that if they want to make sure that their browser traffic is protected they have to follow the suggestions on the Mozilla KB and to verify that no untrusted certificate is left in the Firefox trust store:
https://support.mozilla.org/en-US/kb/how-disable-enterprise-roots-preference
| Reporter | ||
Comment 17•6 years ago
|
||
Additional note: After reading Thayler's blog about the change of this setting I remembered that on my system a antivirus intercepting SSL traffic was installed while testing. So it is currently not clear to me if that caused this configuration setting to be enabled (by changing the suggested prefs.js). I had disabled the intercepting of the SSL traffic during the test so I saw my CA certificate, but it might have left settings enabled, causing side-effects which would not be seen on other systems.
I was testing with Firefox and not Firefox ESR (last has other default settings).
Comment 18•6 years ago
|
||
The priority flag is not set for this bug.
:keeler, could you have a look please?
For more information, please visit auto_nag documentation.
Comment 19•6 years ago
|
||
See also the discussion in bug 1614155. There's nothing we as a browser can do here.
| Reporter | ||
Comment 20•6 years ago
|
||
Sorry, but this is not true. At least you could provide a working way for manual distrust for specific certificates (in the case of the wireless/VPN certificate). See comment 15.
Description
•