Insecure handling of invalid TLS certificates

[krumelmonster]

Search before asking

• I had searched in the issues and found no similar issues.

Operating System

• macOS
• Windows
• Linux

Operating System Version

Windows 10, Arch Linux and others

It happens on the web browser too?

No, it just happens on the Desktop app

Rocket.Chat Desktop App Version

4.4.0, 3.7.5 and others

Rocket.Chat Server Version

any

Describe the bug

This vulnerability has been reported confidentially 2022-05-30 and the responsible disclosure timeline is elapsed.

Summary

1. The Rocket.Chat app allows accepting insecure TLS-certificate for any request made even if a properly secured connection to the server is already established using a valid TLS-Certificate. In contrast, web browsers would completely block requests on an invalid TLS resource from a secure website.
2. The respective dialog is designed in an insecure way that makes it very likely for users to unwillingly accept an invalid/forged certificate, this is in stark contrast to the properly designed TLS error dialogs of web browsers.
   1. The default action is "Yes" to accept the invalid certificate, therefore pressing Enter or the Spacebar accepts the invalid certificate
   2. It is not possible to review the certificate as only the arbitrary name of the certificate is displayed. Working with e.g. internal domain names (the only reason why you may want to use Rocket!Chat with invalid certs in my opinion) would require reviewing the certificates fingerprint; Only reviewing the certificate name cannot validate a certificate as this name can be arbitrarily chosen by the attacker who can publicly learn the name of the valid certificate to be forged.
   3. The attacker-chosen name is not clearly separated from Rocket.Chats dialog Message Do you trust certificate "[...]"?, allowing attackers to change the question by means of social engineering, further tricking the user to choose "Yes". (e.g. Do you trust certificate from "Microsoft" and want to keep your installation of "Windows"?, see screenshot in comment below)
   4. The dialog can be presented to the user while not interacting with Rocket.Chat at all further separating the dialog message from its Meaning.
   5. Closing the app is not possible while presented with the invalid certificate dialog
   6. Presenting a new forged certificate will present the user with a new dialog. In combination with v, this can result in a situation where accepting the forged certificate is the only way for a user to close the app

Screenshots

For comparison: How Browsers handle invalid certificates when opening a site (note that in the given situation, where a secure site is already present, browsers would not even display a dialog and silently block the attack)

Firefox:

Chrome:

Impact

If a user can be tricked intro pressing the Enter key, the Spacebar key or the Yes button, an attacker could gain full control over their account.

How to Reproduce

1. Have the app up and running on a TLS-secured RC instance
2. Spoof the IP-Address or Domain Name of the server (the former can be done on the client, within the clients LAN, on any intermediate network router, within the servers LAN or on the server; the latter can be done anywhere in the DNS chain). One simple way of simulating this attack is to assign the servers public IP to the client or the router of the client LAN, but in a real-world attack, the attack does not require access to either the server or the client.
3. present an arbitrary TLS certificate. Optional: repeatedly rotate this certificate, e.g. repeatedly generating certificates like this openssl req -new -x509 -days 365 -nodes -out yourboss.pem -keyout yourboss.key -subj '/CN=your boss'
4. observe the insecure dialog/many dialogs if certificates are rotated

Describe your Expected behavior

When solving this issue, please consider whether it is neccessary to allow insecure TLS at all, and if so, how much work you want to put into allowing this in a secure way. I see three options:

1. Forbid invalid TLS-Certificates altogether, the dialog could say: "Cannot connect to Server : TLS error: ." -- no options just ok. This is how many applications handle such situations and this change would be a very easy and complete fix of this vulnerability. I believe it should still be possible to set up self-signed certificates and the such by working with the systems certificate store.

2. Redesign the certificate error dialog in more a secure way, minimal requirements for this are:

   • Pressing enter may never lead to accepting the invalid certificate (denying the exception must be the default)
   • It must be made clear to the user that adding an exception without manually comparing certificate fingerprints will expose their account credentials to a potential attacker (e.g. via a confirmation checkbox "I understand the risk"/"I confirm this certificate")
   • The invalid ssl certificate must in all detail be displayed to the user before it can be manually verified and then set to be trusted. The certificate information supplied by the potential attacker must be clearly visually separated from the dialog question and security warning supplied by the Rocket.Chat Electron App.
   • Once the security exception has been denied, it must not be promted for again (not even if the invalid certificate gets replaced by a different invalid certificate) or -- more properly: Show this dialog only when first connecting to a workspace (on startup/initial workspace setup) and fall back to option one if the error appears while a workspace is already connected.

   Implementing solution 2 will certainly be a lot of work and only be effective if all above requirements are met.

3. A compromise: Option one but allow skilled users/administrators to install custom trusted certificate files either via the gui or by placing them in Rocket!Chats config directory.

4. For a very quick mitigation: Forbid invalid TLS-Certificates altogether (who needs this anyway?) but with an opt-in to invalid certificates from the Settings dialog, make it very clear in the dialog that enabling this makes Rocket.Chat insecure

Anything else

RocketChat/Rocket.Chat#27175
1586249

Are you willing to submit a code contribution?

• Yes, I am willing to submit a Pull Request!

[foss-]

How can users track the progress of this issue? Why not keep this issue here open until a fix is released in Rocket.Chat stable version?

[krumelmonster]

As mentioned in the hackerone ticket, I will not further discuss this issue privately. After two years of explaining all the ins and outs of this issue and answering all your questions, including providing suggestions on how to address it, I believe I have given sufficient information. Only yesterday did you move the severity from High to Low, despite Rocket.Chat's stance on this issue in 2022 being:

As a public disclaimer we have detected a process flaw where we were not aware of some researchers questions in our hackerone program, we detected it and changed the process so we can avoid this error again. Regarding the vulnerability, our developers are aware and they're working to deliver the patch we will release new information regarding this vulnerability as soon as the vulnerability has been fixed.

RocketChat/Rocket.Chat#27175

Given this history, it is disheartening that an issue compromising TLS integrity is deemed low-severity. I believe that a vulnerability so likely to render TLS ineffective is inherently low-complexity yet high-risk.

Since you have chosen not to engage in a public dialogue, I will conclude my involvement and retract the offer of working to submit a pull request. It is my opinion that a product marketed with "complete privacy and security" should not leave an Improper TLS Certificate Validation issue unaddressed for over two years. This undermines the product’s security claims and its suitability for use.

[krumelmonster]

I explicitly stated in the hackerone ticket that

There are various ways of relaying the connections to the middleperson but most importantly would this be done on a DNS or an IP level.

This means for example: ISPs, anyone conntected to the same network, anyone aware of your wifi credentials, hosting providers, VPN-providers, CDNs, anyone in control of malware connected to network involved in the communication between client and server and many more can present the user with an unlimited number of dialogs such as this, completely compromising the users Rocket!Chat account including all data and privilege of that account if the user is tricked into clicking the "Yes"-Button, Spacebar or Enter:

Changing the DNS is just one of many ways to exploit this vulnerability. It constitutes an Improper TLS Certificate Validation issue, where invalid certificates are presented to the user as acceptable. Consequently, trust between the client and server cannot be assured, and all user sessions must be considered compromised.

I have made very detailed suggestions on how to address the issue in the hackerone ticket.

[krumelmonster]

What you describe is not the Complexity, it is the Attack Vector and the Attack Vector of this vulnerability is not Local Network or Adjacent Network but rather Network. These are the CVSS definitions:

Network (AV:N)

A vulnerability exploitable with Network access means the vulnerable component is bound to the network stack and the attacker's path is through OSI layer 3 (the network layer). Such a vulnerability is often termed 'remotely exploitable' and can be thought of as an attack being exploitable one or more network hops away (e.g. across layer 3 boundaries from routers).

Adjacent Network (AV:A)

A vulnerability exploitable with Adjacent Network access means the vulnerable component is bound to the network stack, however the attack is limited to the same shared physical (e.g. Bluetooth, IEEE 802.11), or logical (e.g. local IP subnet) network, and cannot be performed across an OSI layer 3 boundary (e.g. a router).

Local Network (AV:L)

A vulnerability exploitable with Local access means that the vulnerable component is not bound to the network stack, and the attacker's path is via read/write/execute capabilities. In some cases, the attacker may be logged in locally in order to exploit the vulnerability, or may rely on User Interaction to execute a malicious file.

Such MITM vulnerabilities are generally regarded AV:N (as illustrated in the cvss examples section) as they neither require network access to be local (as would be with AV:A) nor do they require shell access or similar to either trusted party (as would be with AV:L). I am referring to the examples and definitions provided in the CVSS documentation as that is the scoring system hackerone uses https://www.first.org/cvss/

High Complexity (AC:H) is also not given as AC:H is defined as (emphasis theirs)

The successful attack depends on the evasion or circumvention of security-enhancing techniques in place that would otherwise hinder the attack. These include: Evasion of exploit mitigation techniques. The attacker must have additional methods available to bypass security measures in place. For example, circumvention of address space randomization (ASLR) or data execution prevention (DEP) must be performed for the attack to be successful. Obtaining target-specific secrets. The attacker must gather some target-specific secret before the attack can be successful. A secret is any piece of information that cannot be obtained through any amount of reconnaissance. To obtain the secret the attacker must perform additional attacks or break otherwise secure measures (e.g. knowledge of a secret key may be needed to break a crypto channel). This operation must be performed for each attacked target.

Further applying the definitions of CVSS including the Requirement of user interaction (UI:R) yields a score of 8.8 when assuming Low Complexity (AC:L) and a score of 7.5 when (in my opinion wrongfully) assuming High Complexity (AC:H).

You have however manually changed the severity from 8.3 (High) to Low (below 3.9).

[krumelmonster]

Thank you for working on this! On my test system, "Yes" is still the default selected option. Only difference is I'm seeing is the new warning message which is an improvement.

I don't believe this fully resolves the issue but as the hackerone is still open, I understand this patch is just the first step.

[krumelmonster]

Possibly related electron/electron#18258 "dialog.showMessageBox doesn't respect defaultId option on Windows" and electron/electron#22600 . It does default to "No" now on my linux.