Are there any known cases of HTTPS websites leaking the private key of their SSL certificate? Is it even technically possible for a bad website admin to misconfigure a site to send the private key as part of the certificate chain?

up vote 37 down vote accepted

Are there any known cases of HTTPS websites leaking the private key of their SSL certificate?

Yes - the Heartbleed bug involved memory leaks out of the HTTP server such that:

We attacked ourselves from outside, without leaving a trace. Without using any privileged information or credentials we were able steal from ourselves the secret keys used for our X.509 certificates, ...

Aside from bugs like that,

Is it even technically possible for a bad website admin to misconfigure a site to send the private key as part of the certificate chain?

Sure. If you specify the wrong file in, for example, SSLCertificateChainFile then Boom! There goes the private key.

As @duskwuff points out in the comments, there are protections against this. Neither Apache nor NGINX will send a key PEM that's included in a certificate file; they will silently strip it out (which, I'm willing to bet, is a protection put in place after some number of events where people did what I suggested might work).

Other misconfigurations, such as incorrect web root combined with loose permissions or excessive web server privileges, will also leak the key, but those misconfigurations are both mundane and extreme (e.g., you really have to be trying to break things that badly).

Doing so is not recommended.

  • 2
    I had completely forgotten about Heartbleed. Wikipedia has a good page about Heartbleed - Root causes, possible lessons, and reactions – safesploit Nov 26 at 22:47
  • If you specify the wrong file in, for example, SSLCertificateChainFile then Boom! There goes the private key. can you provide reference that this is what will actually happen? AFAIK, this shouldn't happen as people commonly put all the three SSLCertificate*File into the same file. – Lie Ryan Nov 26 at 23:07
  • 4
    @LieRyan, how about an easier-to-understand misconfiguration that has the same impact: specifying / as DocumentRoot? – Mark Nov 27 at 4:15
  • I think putting the private key to serve as static file as clearly out of scope for this question. I interpret the question as the more likely scenario of how keys can be leaked from a configuration that looks reasonable. – Lie Ryan Nov 27 at 7:09
  • 9
    "If you specify the wrong file in, for example, SSLCertificateChainFile" -- then the server will simply fail to start up. The private key isn't a certificate file. – duskwuff Nov 27 at 7:12

Yes, via either misconfiguration or as-yet unknown misconfigurations. I won't repeat the content of @gowenfawr 's answer.

It is worth mentioning as an aside that there are a number of possible misconfigurations that do not directly leak the key but could allow an attacker to decipher parts of the communication. Part of the work towards TLS 1.3 aims to mitigate this via removing support for certain ciphers and other potentially insecure configurations (https://blog.cloudflare.com/rfc-8446-aka-tls-1-3/).

Likewise there are other things that can cause loss of confidentiality; such as the SSL v2 vulnerability known as DROWN (https://drownattack.com/) where previously SSL v2 was known to be insecure but many installations left it enabled for compatibility reasons.

While not ideal, this led to the possibility of extracting the session key from a server that used the same keys or certificate but happened to be running SSL v2 via a padding oracle, even exposing the contents of TLS 1.2 services using the same certificate (after observing a number of connections between the victim and the server).

DROWN shows that merely supporting SSLv2 is a threat to modern servers and clients. It allows an attacker to decrypt modern TLS connections between up-to-date clients and servers by sending probes to a server that supports SSLv2 and uses the same private key.

In terms of directly leaking the key, you can also serve up your private key on a misconfigured web server, or via a directory traversal (or similar) issue

  • 3
    Nit: DROWN doesn't extract, or otherwise recover, the server's configured privatekey, only learn a per-session secret (which gives the per-session working keys). – dave_thompson_085 Nov 27 at 7:02
  • Thanks for the feedback @dave_thompson_085 and Maarten-Bodewes. I've edited the answer so hopefully it is not misleading anyone, but it probably applies less to the question now ;) – richard Nov 28 at 22:26
  • @MaartenBodewes+ 1.3 prohibits plain-RSA keyexchange i.e. ClientKX = RSA-encrypt(premaster) and (unless PSK) requires [EC]DHE thus PFS so stealing the server key does as you say still allow impersonation -- unless and until (discovered and) revoked -- but does not allow decrypting prior data. (<=1.2 of course can do PFS, but doesn't require it.) – dave_thompson_085 Nov 29 at 1:09
  • @dave_thompson_085 I know all that. The comment was for a previous answer in which it was unclear what "the key" was. The answer has been cleared up now, so I just removed the comment. – Maarten Bodewes Nov 29 at 2:30

Are there any known cases of HTTPS websites leaking the private key of their SSL certificate?

Strictly speaking, a private key should have permissions dr--------, with root:root. So, only the root user can read the certificate. If this is misconfigured and the web server has access to the private key, then in some circumstances such as the web server is compromised. Then here we could see the private key being 'leaked', unwittingly by the web server. This, of course, applies to any program which has read access to the private key though.

Is it even technically possible for a bad website admin to misconfigure a site to send the private key as part of the certificate chain?

By the configuration, I did with Apache 2, no! So, one of my web server configurations involves:

    SSLCertificateFile /etc/apache2/ssl/safesploit.com.cert.pem
    SSLCertificateKeyFile /etc/apache2/ssl/safesploit.com.key.pem
    SSLCertificateChainFile /etc/apache2/ssl/fullchain.pem

So, while I understand your concern regarding a 'bad admin' placing the private key within the fullchain, it simply is not possible within Apache 2 vanilla, unless Apache was modified to accept this type of configuration.

For context:

  • Public keys -r--r--r-- root root
  • Private keys -r-------- root root
  • 6
    Hang on... if a webserver cannot access the private key, how exactly is it supposed to serve https? – Shadow Nov 27 at 5:03
  • 2
    @Shadow: In theory, it could use an external key storage which only allows making signatures, such as a PKCS#11 module in Apache, or something similar to ssh-agent in OpenSSH's sshd. But I think safesploit's post is talking about the more usual type of privilege separation, in which the 'listener' httpd process can access the private key but the 'worker' processes (which interpret HTTP requests) cannot. – grawity Nov 27 at 7:33
  • 2
    Sorry, but that's a bit too theoretic in my opinion. If the root is required to act as a signing server then please indicate this. Or are you proposing to run the server as root? Anyway, the private key is needed to place the signature within the server message of the handshake - please indicate how this is performed in your scheme. – Maarten Bodewes Nov 27 at 12:45
  • @Shadow - apache generally starts as root (required to be able to bind to port 80 and/or 443), which would also allow it to read the private key file. After this initial startup is complete, it then drops root privileges and/or spawns unprivileged children so that the request-handling processes don't have root powers. – Dave Sherohman Nov 27 at 12:46
  • 1
    Regarding Apache and the private key -r-------- root root How does apache threads, running as non-root, handle private keys. @DaveSherohman is correct, Apache does start with root privileges so it can bind to privileged ports, and then the main httpd process continues to run as the root user, but the child processes run as a less privileged user. How Apache Works. – safesploit Nov 27 at 13:22

I discovered a flaw in Traefik not too long ago where an API endpoint was leaking the private key for their SSL certificate (CVE-2018-15598).

https://twitter.com/edoverflow/status/1031267058615508994

enter image description here

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.