Quelques digressions sous GPL

To content | To menu | To search

Friday, November 13 2015

On endpoint security, vendors and root access.

Endpoint security typically comes in two flavors: with or without a local agent. They both do the same thing - reach out to your endpoints and run a bunch of tests - but one will reach out to your systems over SSH, while the other will require a local agent to be deployed on all endpoints. Both approaches often share the same flaw: the servers that operate the security solution have the keys to become root on all your endpoints. These servers become targets of choice: take control of them, and you are root across the infrastructure.

I have evaluated many endpoint security solutions over the past few years, and I too often run into this broken approach to access control. In extreme cases, vendors are even bold enough to sell hosted services that require their customers to grant root accesses to SaaS operating as blackboxes. These vendors are successful, so I imagine they find customers who think sharing root accesses that way is acceptable. I am not one of them.

For some, trust is a commodity that should be outsource-able. They see trust as something you can write into a contract, and not worry about it afterward. To some extend, contracting does help with trust. More often than not, however, trust is earned over time, and contracts only seal the trust relationship both parties have already established.

I trust AWS because they have a proven track record of doing things securely. I did not use to, but time and experience have changed my mind. You, however, young startup that freshly released a fancy new security product I am about to evaluate, I do not yet trust you. You will have to earn that trust over time, and I won't make it easy.

This is where my issue with most endpoint security solutions lies: I do not want to trust random security vendors with root accesses to my servers. Mistakes happen, they will get hacked some day, or leak their password in a git commit or a pastebin, and I do not wish my organization to be a collateral damage of their operational deficiencies.

Endpoint security without blindly trusting the security platform is doable. MIG is designed around the concept of non-trustable infrastructure. This is achieved by requiring all actions sent to MIG agents to be signed using keys that are not stored on the MIG servers, but on the laptops of investigators, the same way SSH keys are managed. If the MIG servers get hacked, some data may leak, but no access will be compromised.

Another aspect that we included in MIG is the notion that endpoint security can be done without arbitrary remote code exception. Most solutions will happily run code that come from the trusted central platform, effectively opening a backdoor into the infrastructure. MIG does not allow this. Agents will only run specific investigative tasks that have been pre-compiled into modules. There is no vector for remote code execution, such that an investigator's key leaking would not allow an attacker to elevate access to being root on endpoints. This approach does limit the capabilities of the platform - we can only investigate what MIG supports - but if remote code execution is really what you need, you probably should be looking into a provisioning tool, or pssh, but not an endpoint security solution.

While I do take MIG as an example, I am not advocating it as a better solution to all things. Rather, I am advocating for proper access controls in endpoint security solutions. Any security product that has the potential to compromise your entire infrastructure if taken over is bad, and should not be trusted. Even if it brings some security benefits. You should not have to compromise on this. Vendors should not ask customers to accept that risk, and just trust them to keep their servers secure. Doing endpoint security the safe way is possible, it's just a matter of engineering it right.

Sunday, October 4 2015

SHA1/SHA256 certificate switching with HAProxy

SHA-1 certificates are on their way out, and you should upgrade to a SHA-256 certificate as soon as possible... unless you have very old clients and must maintain SHA-1 compatibility for a while.

If you are in this situation, you need to either force your clients to upgrade (difficult) or implement some form of certificate selection logic: we call that "cert switching".

The most deterministic selection method is to serve SHA-256 certificates to clients that present a TLS1.2 CLIENT HELLO that explicitly announces their support for SHA256-RSA (0x0401) in the signature_algorithms extension.

signaturealgorithmsextensions.png

Modern web browsers will send this extension. However, I am not aware of any open source load balancer that is currently able to inspect the content of the signature_algorithms extension. It may come in the future, but for now the easiest way to achieve cert switching is to use HAProxy SNI ACLs: if a client presents the SNI extension, direct it to a backend that presents a SHA-256 certificate. If it doesn't present the extension, assume that it's an old client that speaks SSLv3 or some broken version of TLS, and present it a SHA-1 cert.

This can be achieved in HAProxy by chaining frontend and backends:

haproxy-cert-switching.png

global
        ssl-default-bind-ciphers ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-AES256-GCM-SHA384:DHE-RSA-AES128-GCM-SHA256:DHE-DSS-AES128-GCM-SHA256:kEDH+AESGCM:ECDHE-RSA-AES128-SHA256:ECDHE-ECDSA-AES128
-SHA256:ECDHE-RSA-AES128-SHA:ECDHE-ECDSA-AES128-SHA:ECDHE-RSA-AES256-SHA384:ECDHE-ECDSA-AES256-SHA384:ECDHE-RSA-AES256-SHA:ECDHE-ECDSA-AES256-SHA:DHE-RSA-AES128-SHA256:DHE-RSA-AES128-SHA:DHE-DSS-AES128-SHA256:DHE-RSA-AES256-SHA256:DHE-DSS-AES256-SHA:DHE-R
SA-AES256-SHA:!aNULL:!eNULL:!EXPORT:!DES:!RC4:!3DES:!MD5:!PSK

frontend https-in
        bind 0.0.0.0:443
        mode tcp
        tcp-request inspect-delay 5s
        tcp-request content accept if { req_ssl_hello_type 1 }
        use_backend jve_https if { req.ssl_sni -i jve.linuxwall.info }

        # fallback to backward compatible sha1
        default_backend jve_https_sha1

backend jve_https
        mode tcp
        server jve_https 127.0.0.1:1665
frontend jve_https
        bind 127.0.0.1:1665 ssl no-sslv3 no-tlsv10 crt /etc/haproxy/certs/jve_sha256.pem tfo
        mode http
        option forwardfor
        use_backend jve

backend jve_https_sha1
        mode tcp
        server jve_https 127.0.0.1:1667
frontend jve_https_sha1
        bind 127.0.0.1:1667 ssl crt /etc/haproxy/certs/jve_sha1.pem tfo ciphers ECDHE-RSA-AES128-GCM-SHA256:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-AES256-GCM-SHA384:DHE-RSA-AES128-GCM-SHA256:DHE-DSS-AES128-GCM-SHA256:kEDH+AESGCM:ECDHE-RSA-AES128-SHA256:ECDHE-ECDSA-AES128-SHA256:ECDHE-RSA-AES128-SHA:ECDHE-ECDSA-AES128-SHA:ECDHE-RSA-AES256-SHA384:ECDHE-ECDSA-AES256-SHA384:ECDHE-RSA-AES256-SHA:ECDHE-ECDSA-AES256-SHA:DHE-RSA-AES128-SHA256:DHE-RSA-AES128-SHA:DHE-DSS-AES128-SHA256:DHE-RSA-AES256-SHA256:DHE-DSS-AES256-SHA:DHE-RSA-AES256-SHA:ECDHE-RSA-DES-CBC3-SHA:ECDHE-ECDSA-DES-CBC3-SHA:AES128-GCM-SHA256:AES256-GCM-SHA384:AES128-SHA256:AES256-SHA256:AES128-SHA:AES256-SHA:AES:DES-CBC3-SHA:HIGH:!aNULL:!eNULL:!EXPORT:!DES:!RC4:!MD5:!PSK:!aECDH:!EDH-DSS-DES-CBC3-SHA:!EDH-RSA-DES-CBC3-SHA:!KRB5-DES-CBC3-SHA
        mode http
        option forwardfor
        use_backend jve

backend jve
        rspadd Strict-Transport-Security:\ max-age=15768000
        server jve 172.16.0.6:80 maxconn 128

The configuration above receives inbound traffic in the frontend called "https-in". That frontend is in TCP mode and inspects the CLIENT HELLO coming from the client for the value of the SNI extension. If that value exists and matches our target site, it sends the connection to the backend named "jve_https", which redirects to a frontend also named "jve_https" where the SHA256 certificate is configured and served to the client.

If the client fails to present a CLIENT HELLO with SNI, or presents a SNI that doesn't match our target site, it is redirected to the "https_jve_sha1" backend, then to its corresponding frontend where a SHA1 certificate is served. That frontend also supports an older ciphersuite to accommodate older clients.

Both frontends eventually redirect to a single backend named "jve" which sends traffic to the destination web servers.

This is a very simple configuration, and eventually it could be improved using better ACLs (HAproxy regularly adds news ones), but for a basic cert switching configuration, it gets the job done!

Thursday, October 1 2015

Introducing SOPS: a manager of encrypted files for secrets distribution

Automating the distribution of secrets and credentials to components of an infrastructure is a hard problem. We know how to encrypt secrets and share them between humans, but extending that trust to systems is difficult. Particularly when these systems follow devops principles and are created and destroyed without human intervention. The issue boils down to establishing the initial trust of a system that just joined the infrastructure, and providing it access to the secrets it needs to configure itself.

The initial trust

In many infrastructures, even highly dynamic ones, the initial trust is established by a human. An example is seen in Puppet by the way certificates are issued: when a new system attempts to join a Puppetmaster, an administrator must, by default, manually approve the issuance of the certificate the system needs. This is cumbersome, and many puppetmasters are configured to auto-sign new certificates to work around that issue. This is obviously not recommended and far from ideal.

AWS provides a more flexible approach to trusting new systems. It uses a powerful mechanism of roles and identities. In AWS, it is possible to verify that a new system has been granted a specific role at creation, and it is possible to map that role to specific resources. Instead of trusting new systems directly, the administrator trusts the AWS permission model and its automation infrastructure. As long as AWS keys are safe, and the AWS API is secure, we can assume that trust is maintained and systems are who they say they are.

KMS, Trust and secrets distributionKMS_Benefit_Key.png

Using the AWS trust model, we can create fine grained access controls to Amazon's Key Management Service (KMS). KMS is a service that encrypts and decrypts data with AES_GCM, using keys that are never visible to users of the service. Each KMS master key has a set of role-based access controls, and individual roles are permitted to encrypt or decrypt using the master key. KMS helps solve the problem of distributing keys, by shifting it into an access control problem that can be solved using AWS's trust model.

Since KMS's inception a few months ago, a number of projects have popped up to use its capabilities to distribute secrets: credstash and sneaker are such examples. Today I'm introducing sops: a secrets editor that uses KMS and PGP to manage encrypted files.


SOPS: Secrets OPerationS

A few weeks ago, Mozilla's Services Operations team started revisiting the issue of distributing secrets to EC2 instances, with a goal to store these secrets encrypted until the very last moment, when they need to be decrypted on target systems. Not unlike many other organizations that operate sufficiently complex automation, we found this to be a hard problem with a number of prerequisites:

  1. Secrets must be stored in YAML files for easy integration into hiera
  2. Secrets must be stored in GIT, and when a new CloudFormation stack is built, the current HEAD is pinned to the stack. (This allows secrets to be changed in GIT without impacting the current stack that may autoscale).
  3. Encrypt entries separately. Encrypting entire files as blobs makes git conflict resolution almost impossible. Encrypting each entry separately is much easier to manage.
  4. Secrets must always be encrypted on disk (admin laptop, upstream git repo, jenkins and S3) and only be decrypted on the target systems

Daniel Thornton and I brainstormed a number of ideas, and eventually ended-up with a workflow similar to the one described below.

kms_yaml_small.JPG

The idea behind SOPS is to provide a wrapper around a text editor that takes care of the encryption and decryption transparently. When creating a new file, sops generates a data encryption key "Kd" that is itself encrypted with one or more KMS master keys and PGP public keys. Kd is used to encrypt the content of the file with AES256-GCM. In order to decrypt the files, sops must have access to any of the KMS or PGP master keys.


SOPS can be used to encrypt YAML, JSON and TEXT files. In TEXT mode, the content of the file is treated as a blob, the same way PGP would encrypt an entire file. In YAML and JSON modes, however, the content of the file is manipulated as a tree where keys are stored in cleartext, and values are encrypted. hiera-eyaml does something similar, and over the years we learned to appreciate its benefits, namely:

  • diff are meaningful. If a single value of a file is modified, only that value will show up in the diff. The diff is still limited to only showing encrypted data, but that information is already more granular that indicating that an entire file has changed.
  • conflicts are easier to resolve. If multiple users are working on the same encrypted files, as long as they don't modify the same values, changes are easy to merge. This is an improvement over the PGP encryption approach where unsolvable conflicts often happen when multiple users work on the same file.
# edit a file
$ sops example.yaml file written to example.yaml
# take a look at the diff
$ git diff example.yaml diff --git a/example.yaml b/example.yaml index 00fe479..5f40330 100644 --- a/example.yaml +++ b/example.yaml @@ -1,5 +1,5 @@ # The secrets below are unreadable without access to one of the sops master key -myapp1: ENC[AES256_GCM,data:Tr7oo=,iv:1vw=,aad:eo=,tag:ka=]
+myapp1: ENC[AES256_GCM,data:krm,iv:0Y=,aad:KPyE=,tag:oIA==]
app2: db: user: ENC[AES256_GCM,data:YNKE,iv:H4JQ=,aad:jk0=,tag:Neg==]

Below are two examples of SOPS encrypted files. The first one in YAML, the second one in JSON:

YAML

cleartext:

# The secrets below are unreadable without access to one of the sops master key
myapp1: t00m4nys3cr3tzupdated
app2:
    db:
        user: eve
        password: c4r1b0u
    # private key for secret operations in app2
    key: |
        -----BEGIN RSA PRIVATE KEY-----
        MIIBPAIBAAJBAPTMNIyHuZtpLYc7VsHQtwOkWYobkUblmHWRmbXzlAX6K8tMf3Wf
        ImcbNkqAKnELzFAPSBeEMhrBN0PyOC9lYlMCAwEAAQJBALXD4sjuBn1E7Y9aGiMz
        bJEBuZJ4wbhYxomVoQKfaCu+kH80uLFZKoSz85/ySauWE8LgZcMLIBoiXNhDKfQL
        vHECIQD6tCG9NMFWor69kgbX8vK5Y+QL+kRq+9HK6yZ9a+hsLQIhAPn4Ie6HGTjw
        fHSTXWZpGSan7NwTkIu4U5q2SlLjcZh/AiEA78NYRRBwGwAYNUqzutGBqyXKUl4u
        Erb0xAEyVV7e8J0CIQC8VBY8f8yg+Y7Kxbw4zDYGyb3KkXL10YorpeuZR4LuQQIg
        bKGPkMM4w5blyE1tqGN0T7sJwEx+EUOgacRNqM2ljVA=
        -----END RSA PRIVATE KEY-----
number: 1234567890
an_array:
- secretuser1
- secretuser2
- somelongvalueAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
- some other value

encrypted:

# The secrets below are unreadable without access to one of the sops master key
myapp1: ENC[AES256_GCM,data:krwEdH2fxWRexFuvZHS816Wz46Lm,iv:0STqWePc0HOPuDn2EizQdNepx9ksx0guHGeKrshlYSY=,aad:Krl8HyPGQmnIWIZh74Ib+y0OdiVEvRDBv3jTdMGSPyE=,tag:oI2THtQeUX4ZLNnbrdel2A==]
app2:
    db:
        user: ENC[AES256_GCM,data:YNKE,iv:H9CDb4aUHBJeF2MSTKHQuOwlLxQVdx12AhT0+Dob4JQ=,aad:jlF2KvytlQIgyMpOoO/BiQbukiMwrh1j94Oys+YMgk0=,tag:NeDysIHV9CGtMAQq9i4vMg==]
        password: ENC[AES256_GCM,data:p673JCgHYw==,iv:EOOeivCp/Fd80xFdMYX0QeZn6orGTK8CeckmipjKqYY=,aad:UAhi/SHK0aCzptnFkFG4dW8Vv1ASg7TDHD6lui9mmKQ=,tag:QE6uuhRx+cGInwSVdmxXzA==]
    # private key for secret operations in app2
    key: |-
        ENC[AES256_GCM,data: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,iv:Dl61tsemKH0fdmNul/PmEEsRYFAh8GorR8GRupus/EM=,aad:Ft2aSYYukD1x8pMj1WvmodLjJV6waPy5FqdlImWyQKA=,tag:EPg4KpWqni/buCFjFL857A==]
number: ENC[AES256_GCM,data:XMrBalgZ9tvBxQ==,iv:XyEAAaIzVy/2trnJhLrjMInLg8tMI4CAX9+ccnj3T1Y=,aad:JOlAkP159UxDjL1CrumTuQDqgW2+VOIwz7bdfaJIIn4=,tag:WOHOMJS4nhSdj/aQcGbU1A==]
an_array:
- ENC[AES256_GCM,data:td1aAv4s4cOzSo0=,iv:ErVqte7GpQ3JfzVpVRf7pWSQZDHn6W0iAntKWFsMqio=,aad:RiYy8fKX/yVY7KRgXSOIzydT0+TwK7WGzSFSy+1GmVM=,tag:aSGLCmNZsGcBjxEGvNQRwA==]
- ENC[AES256_GCM,data:2K8C418jef8zoAY=,iv:cXE4Hwdl4ZHzAHHyyXqaIMFs0mn65JUehDdaw/aM0WI=,aad:RlAgUZUZ1DvxD9/lZQk9KOHKl4L+fYETaAdpDVekCaA=,tag:CORSBzis6Vy45dEvT/UtMg==]
- ENC[AES256_GCM,data:hbcOBbsaWmlnrpeuwLfh1ttsi8zj/pxMc1LYqhdksT/oQb80g2z0FE4QwUVb7VV+x98LAWHofVyV8Q==,iv:/sXHXde82r2FyG3Z3vC5x8zONB14RwC0GmtkiYEUNLI=,aad:BQb8l5fZzF/aa/EYnrOQvRfGUTq9QmJOAR/zmgOfYDA=,tag:fjNeg3Manjl6B2U2oflRhg==]
- ENC[AES256_GCM,data:LLHkzGobqL53ws6E2zglkA==,iv:g9z3zz4DUzJr4Cim0SVqKF736w2mZoItqbB0TcsGrQU=,aad:Odrvz0loqFdd9wKJz0ULMX/lyEQcX8WaHE59MgeXkcI=,tag:V+rV/AeZ4uEgtwGhlamTag==]
sops:
    kms:
    -   enc: CiC6yCOtzsnFhkfdIslYZ0bAf//gYLYCmIu87B3sy/5yYxKnAQEBAQB4usgjrc7JxYZH3SLJWGdGwH//4GC2ApiLvOwd7Mv+cmMAAAB+MHwGCSqGSIb3DQEHBqBvMG0CAQAwaAYJKoZIhvcNAQcBMB4GCWCGSAFlAwQBLjARBAyGdRODuYMHbA8Ozj8CARCAO7opMolPJUmBXd39Zlp0L2H9fzMKidHm1vvaF6nNFq0ClRY7FlIZmTm4JfnOebPseffiXFn9tG8cq7oi
        enc_ts: 1439568549.245995
        arn: arn:aws:kms:us-east-1:656532927350:key/920aff2e-c5f1-4040-943a-047fa387b27e
    pgp:
    -   fp: 85D77543B3D624B63CEA9E6DBC17301B491B3F21
        enc: |
            -----BEGIN PGP MESSAGE-----
            Version: GnuPG v1

            hQIMA0t4uZHfl9qgAQ//ZvUMJOLUJyzKa/Uigwh1jKVhx3feHUitVjCWBfVTPgj1
            rRbaTcaF/mYi+rLdW+6kmAg1UEPoVgEBEiBvCTcHjyDzw3m0DoQwvK85nqOpEhkx
            rjU1XAnKZ8LNFfIaj8Xo/L6qzE882gwOhfCPU+QmnkWdijs6dQof06DButQDTx5D
            KlFvr9CgSa52/uPazZ41disho9guS06k+KrV/P2F4jrU5aB5mfP7YZY9mkVcm2bv
            9C5O9neNlXcivgWqKQjB5fmv1Z9yUFAUBNg98wjT8o5Hxz6P6hIbV3f+vn/Vu+VZ
            Qo+E7g3/2ItaT89KAIVXgQdHhwJneoDBVpJ4rYz7LLbcvEyAbipKIY4Fl3Cn1ggH
            9odIZWA6FWZxHNhRVonMVHZ8Jei5NkUdpJltjDmPJpl3B+7XiWg4NS8dp860fLeL
            8nrkR0Z4nVK8DNg+7nQiOxHL9wye6ljWl7/xapJ5r+mYA6eLybsSSlxDo9/OmeON
            CYo3jV8HT8amrXYVi4MyZ3LV2TTyGVPObnthYEN2lPSJmms6ei6t/xKaZtAj6779
            EzGbUP9VpTKKf5tqGcy9MeGEk2p5ed5hJGinrrt92cNIebcMBJpkLQAy++V/fKnH
            Meecaoj1NThBnRguNuz73WSy2C5u/g7OoI50HJJCmoVXY+8D64tWmCZc8Ib2fprS
            XgFcoR9u5yPkLZW8xASpRXfKKTbRTTjAXdYEyaYuuOW2nFWo62/d1mZsT7kY21ja
            AhVVoxwsj45FCuk63bDVceAJJm+9xxufMp0gNW1GUk858VLyE8gn+uAB5zBcS5c=
            =BN/t
            -----END PGP MESSAGE-----
        created_at: 1443203323.058362

JSON

cleartext

{
    "address": {
        "city": "New York", 
        "postalCode": "10021-3100", 
        "state": "NY", 
        "streetAddress": "21 2nd Street"
    }, 
    "age": 25, 
    "firstName": "John", 
    "lastName": "Smith", 
    "phoneNumbers": [
        {
            "number": "212 555-1234", 
            "type": "home"
        }, 
        {
            "number": "646 555-4567", 
            "type": "office"
        }
    ]
}

encrypted:

{
    "address": {
        "city": "ENC[AES256_GCM,data:2wNRKB+Sjjw=,iv:rmATLCPii2WMzcT80Wp9gOpYQqzx6juRmCf9ioz2ZLM=,aad:dj0QZW0BvZVjF1Dn25hOJpcwcVB0qYvEIhGWgxq6YzQ=,tag:wOoPYU+8BA9DiNFlsal3Aw==]", 
        "postalCode": "ENC[AES256_GCM,data:xwWZ/np9Gxv3CQ==,iv:OLwOr7iliPyWWBtKfUUH7E1wQlxJLA6aFxIfNAEC/M0=,aad:8mw5NU8MpyBlrh7XaUqa642jeyJWGqKvduaQ5bWJ5pc=,tag:VFmnc4Ay+yKzyHcrKeEzZQ==]", 
        "state": "ENC[AES256_GCM,data:3jY=,iv:Y2bEgkjdn91Pbf5RgJMbyCsyfhV7XWdDhe8wVwTQue0=,aad:DcA5kW1rrET9TxQ4kn9jHSpoMlkcPKs5O5n9wZjZYCQ=,tag:ad1xdNnFwkqx/8EOKVVHIA==]", 
        "streetAddress": "ENC[AES256_GCM,data:payzP57DGPl5S9Z7uQ==,iv:UIz34fk9zH4z6hYfu0duXmAnI8CqnoRhoaIUqg1YoYA=,aad:hll9Baw40lMjwj7HePQ1o1Lsuh1LCwrE6+bkG4025sg=,tag:FDBhYxMmJ1Wj/uxYxdvVZg==]"
    }, 
    "age": "ENC[AES256_GCM,data:4Y4=,iv:hi1iSH19dHSgG/c7yVbNj4yzueHSmmY46yYqeNCoX5M=,aad:nnyubQyaWeLTcz9k9cMHUlgTwVDMyHf32sWCBm7KWAA=,tag:4lcMjstadzI8K40BoDEfDA==]", 
    "firstName": "ENC[AES256_GCM,data:KVe8Dw==,iv:+eg+Rjvaqa2EEp6ufw9c4hwWwObxRLPmxx3fG6rkyps=,aad:3BdHcorHfbvM2Jcs96zX0JY2VQL5dBNgy7zwhqLNqAU=,tag:5OD6MN9SPhBmXuA81hyxhQ==]", 
    "lastName": "ENC[AES256_GCM,data:1+koqsI=,iv:b2kBxSW4yOnLFc8qoeylkMtiO/6qr4cZ5VTntXTyXO8=,aad:W7HXukq3lUUMj9i57UehILG2NAp8XCgJMYbvgflWJIY=,tag:HOrgi1L+IRP+X5JGMnm7Ig==]", 
    "phoneNumbers": [
        {
            "number": "ENC[AES256_GCM,data:Oo0IxdtBrnfE+bTf,iv:tQ1E/JQ4lHZvj1nQnGL2sKE30sCctjiMCiagS2Yzch8=,aad:P+m5gD3pKfNEOy6t61vbKhEpPtMFI2NZjBPrD/m8T9w=,tag:6iRMUVUEx3UZvUTGTjCdwg==]", 
            "type": "ENC[AES256_GCM,data:M3zOKQ==,iv:pD9RO4BPUVu6AWPo2DprRsOqouN+0HJn+RXQAXhfB2s=,aad:KFBBVEEnSjdmah3i2XmPx7wWEiFPrxpnfKYW4BSolhk=,tag:liwNnip/L6SZ9srn0N5G4g==]"
        }, 
        {
            "number": "ENC[AES256_GCM,data:BI2f/qFUea6UHYQ+,iv:jaVLMju6h7s+AlF7CsPbpUFXO2YtYAqYsCIsyHgfrfI=,aad:N+8sVpdTlY5I+DcvnY018Iyh/QesD7bvwfKHRr7q2L0=,tag:hHPPpQKP4cUIXfh9CFe4dA==]", 
            "type": "ENC[AES256_GCM,data:EfKAdEUP,iv:Td+sGaS8XXRqzY98OK08zmdqsO2EqVGK1/yDTursD8U=,aad:h9zi8s+EBsfR3BQG4r+t+uqeChK4Hw6B9nJCrValXnI=,tag:GxSk1LAQIJNGyUy7AvlanQ==]"
        }
    ], 
    "sops": {
        "kms": [
            {
                "arn": "arn:aws:kms:us-east-1:656532927350:key/920aff2e-c5f1-4040-943a-047fa387b27e", 
                "created_at": 1443204393.48012, 
                "enc": "CiC6yCOtzsnFhkfdIslYZ0bAf//gYLYCmIu87B3sy/5yYxKnAQEBAgB4usgjrc7JxYZH3SLJWGdGwH//4GC2ApiLvOwd7Mv+cmMAAAB+MHwGCSqGSIb3DQEHBqBvMG0CAQAwaAYJKoZIhvcNAQcBMB4GCWCGSAFlAwQBLjARBAwBpvXXfdPzEIyEMxICARCAOy57Odt9ngHHyIjVU8wqMA4QszXdBglNkr+duzKQO316CRoV5r7bO8JwFCb7699qreocJd+RhRH5IIE3"
            }, 
            {
                "arn": "arn:aws:kms:ap-southeast-1:656532927350:key/9006a8aa-0fa6-4c14-930e-a2dfb916de1d", 
                "created_at": 1443204394.74377, 
                "enc": "CiBdfsKZbRNf/Li8Tf2SjeSdP76DineB1sbPjV0TV+meTxKnAQEBAgB4XX7CmW0TX/y4vE39ko3knT++g4p3gdbGz41dE1fpnk8AAAB+MHwGCSqGSIb3DQEHBqBvMG0CAQAwaAYJKoZIhvcNAQcBMB4GCWCGSAFlAwQBLjARBAwag3w44N8+0WBVySwCARCAOzpqMpvzIXV416ycCJd7mn9dBvjqzkUDag/zHlKse57uNN7P0S9GeRVJ6TyJsVNM+GlWx8++F9B+RUE3"
            }
        ], 
        "pgp": [
            {
                "created_at": 1443204394.748745, 
                "enc": "-----BEGIN PGP MESSAGE-----\nVersion: GnuPG v1\n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\n=Z9kg\n-----END PGP MESSAGE-----\n", 
                "fp": "85D77543B3D624B63CEA9E6DBC17301B491B3F21"
            }
        ]
    }
}

As you can see on each key/value pair, only the values are encrypted and keys are kept in clear. It can be argued that this approach leaks sensitive information, but it’s a tradeoff we’re willing to accept in exchange for an increased usability.

Simplifying key management

OpenPGP gets a lot of bad press for being an outdated crypto protocol, and while true, what really made look for alternatives is the difficulty to manage and distribute keys to systems. With KMS, we manage permissions to an API, not keys, and that's a lot easier to do.

But PGP is not dead yet, and we still rely on it heavily as a backup solution: all our files are encrypted with KMS and with one PGP public key, with its private key stored securely for emergency decryption in the event that we lose all our KMS master keys.


That said, nothing prevents you from using SOPS the same way you would use an encrypted PGP file: by referencing the pubkeys of each individual who has access to the file. It can easily be done by providing sops with a comma-separated list of public keys when creating a new file:

$ sops --pgp "E60892BB9BD89A69F759A1A0A3D652173B763E8F, 84050F1D61AF7C230A12217687DF65059EF093D3, 85D77543B3D624B63CEA9E6DBC17301B491B3F21" mynewfile.yaml

Updating the master keys

GnuPG can be a little obscure when it comes to managing the keys that have access to a file. While its command line is powerful, it takes a few minutes to find the right commands and figure out how to provide access to a new member of the team.

In SOPS, managing master keys is easy: they are simply stored as entries of the document under sops->{kms,pgp}. By default, that information is hidden during editing, but calling sops with the "-s" flag will display the master keys in the editor. From there, add new keys by creating entry in the document, or remove them by deleting the lines.

Rotation

Rotating data keys is also trivial: sops provides a rotation flag "-r" that will generate a new data key Kd and re-encrypt all values in the file with it. Coupled with in-place encryption/decryption, it is easy to rotate all the keys on a group of files:

for file in $(find . -type f -name "*.yaml"); do
        sops -d -i $file
        sops -e -i -r $file
done

Something that should be done every few months for good practice ;)

Assuming roles

SOPS has the ability to use KMS in multiple AWS accounts by assuming roles in each account. Being able to assume roles is a nice feature of AWS that allows administrator to establish trust relationships between accounts, typically from the most secure account to the least secure one. In our use-case, we use roles to indicate that a user of the Master AWS account is allowed to make use of KMS master keys in development and staging AWS accounts. Using roles, a single file can be encrypted with KMS keys in multiple accounts, thus increasing reliability and ease of use.

Check it out, and contribute!

SOPS is available on Github at http://github.com/mozilla/sops and on Pypi at https://pypi.python.org/pypi/sops. We are progressively reaching a stable stage, with a goal to support Python 2.6.6 to 3.4.

Screenshot_2015-10-01_23-07-55.png

Sunday, September 13 2015

Investigating SSH authorized keys across the infrastructure using MIG

One of the challenge of operating an organization like Mozilla is dealing with the heterogeneity of the platform. Each group is free to define its own operational practices, as long as they respect strong security rules. We don't centralize a lot, and when we do, we do it in a way to doesn't slow down devops.

The real challenge on the infosec side is being able to investigate infrastructures that are managed in many different ways. We look for anomalies, and one that recently received our focus is finding bad ~/.ssh/authorized_keys files.

Solving that problem involved adding some functionalities to MIG's file investigation module to assert the content of files, as well as writing a little bit of Python. Not only did this method help us find files that needed updating, but it also provided a way to assert the content of authorized_keys files moving forward.

Let's dive in.

LDAP all the things!

We have a really good LDAP database, results of tons of hard work from the folks in Mozilla IT. We use it for a lot of things, from providing a hierarchical view of Mozilla to showing your personal photo in the organization's phonebook. We also use it to store GPG Fingerprints and, what interests us today, SSH Public Keys.

LDAP users are in charge of their keys. They have an admin panel where they can add and remove keys, to facilitate regular rotations. On the infra side, Puppet pulls the public keys and writes them into the users authorized_keys files. As long as LDAP is up to date, and Puppet runs, authorized_keys files contain the proper keys.

But bugs happen, and sometimes, for various reasons, configurations don't get updated when they should be, and files go out of date. This is where we need an external mechanism to find the systems where configurations go stale, and fix them.

Asserting the content of a file

The most common way to verify the integrity of a file is by using a checksum, like a sha256sum. Unfortunately, it is very rare that a given file would always be exactly the same across the infrastructure. That is particularly true in our environment, because we often add a header with a generation date to authorized_keys files.

# HEADER: This file was autogenerated at Mon Jul 27 14:24:07 +0000 2015

That header means the checksum will change on every machine, and we cannot use a checksum approach to assert the content of a file. Instead, we need to use a regular expression.

Content regexes have been present in MIG for a while now, and are probably the most used feature in investigations. But until recently, content regexes were limited to finding things that exist in a file, such as an IOC. The file module would stop inspecting as soon as a regex matches, even skipping the rest of the file, to accelerate investigations.

To assert the content of a file, we need a different approach. The regex needs to verify that every line of a file match our expectations, and if one line does not match, that means the file has bad content.

Introducing Macroal mode

The first part of the equation is making sure that every line in a file match a given regex. In the file module, we introduced a new option called "macroal" that stands for Match All Content Regexes On All Lines. When activated, this mode tells the file module to continue reading the file until the end, and flag the file if all lines have match the content regex.

On the MIG command line, this option can be used in the file module with flag ""-macroal". It's a boolean that is off by default.

$ mig file -t local -path /etc -name "^passwd$" -content "^([A-Za-z0-9]|-|_){1,100}:x:[0-9]{1,5}:[0-9]{1,5}:.+" -macroal

The command above finds /etc/passwd and checks that all the lines in the file match the content regex "^([A-Za-z0-9]|-|_){1,100}:x:[0-9]{1,5}:[0-9]{1,5}:.+". If they do, MIG returns a positive result on the file.

In the JSON of the investigation, macroal is stored in the options of a search:

{
    "searches": {
        "s1": {
            "paths": [
                "/etc"
            ],
            "names": [
                "^passwd$"
            ],
            "contents": [
                "^([A-Za-z0-9]|-|_){1,100}:x:[0-9]{1,5}:[0-9]{1,5}:.+"
            ],
            "options": {
                "macroal": true,
                "matchall": true,
                "maxdepth": 1
            }
        }
    }
}

But finding files that match all the lines is not yet what we want. In fact, we want the exact opposite: finding files that have lines that do not match the content regex.

Introducing Mismatch mode

Another improvement we added to the file module is the mismatch mode. It's a simple feature that inverses the behavior of one or several parameters in a file search.

For example, if we know that all versions of RHEL6.6 have /usr/bin/sudo matching a given sha256, we can use the mismatch option to find instances where sudo does not match the expected checksum.

$ mig file -t "environment->>'ident' LIKE 'Red Hat Enterprise Linux Server release 6.6%'" \
> -path /usr/bin -name "^sudo$" \
> -sha256 28d18c50eb23cfd6ac8d39461d5479e19f6f1a5f6b839d34f2eeaf7ce8a3e054 \
> -mismatch sha256

Mismatch allows us to find anomalies, files that don't match our expectations. By combining Macroal and Mismatch in a file search, we can find files that have unexpected content. But we need one last piece: a content regex that can be used to inspect authorized_keys files.

Building regexes for authorized_keys files

An authorized_keys file should only contain one of three type of line:

  1. a comment line that starts with a pound "#" character
  2. an empty line, or a line full of spaces
  3. a ssh public key

Writing a regex for the first two types is easy. A comment line is "^#.+$" and an empty line is "^(\s+)?$".

Write a regex for SSH public keys isn't too complicated, but we need to take a few precautions. A pubkey entry has three section separated by a white space, and we only care about the first two section. The third one, the comments, can be discarded entirely with ".+".

Next, a few things need to be escaped in the public key, as pubkey are base64 encoded and thus include the slash "/" and plus "+" character that have special meaning in regexes.

Awk and Sed can do this very easily:

$ awk '{print $1,$2}' ~/.ssh/authorized_keys | grep -v "^#" | sed "s;\/;\\\/;g" | sed "s;\+;\\\+;g"

The result can be placed into a content regex and given to MIG.

$ mig file -path /home/jvehent/.ssh -name "^authorized_keys$" \
> -content "^((#.+)|(\\s+)|(ssh-rsa\\sAAAAB3NzaC1yc2EAA[...]yFDMZLFlVmQ==\\s.+))$" \
> -macroal -mismatch content

Or in JSON form:

{
    "searches": {
        "jvehent@mozilla.com_ssh_pubkeys": {
            "contents": [
                "^((#.+)|(\\s+)|(ssh-rsa\\sAAAAB3NzaC1yc2EAA[...]yFDMZLFlVmQ==\\s.+))$"
            ],
            "names": [
                "^authorized_keys$"
            ],
            "options": {
                "macroal": true,
                "matchall": true,
                "maxdepth": 1,
                "mismatch": [
                    "content"
                ]
            },
            "paths": [
                "/home/jvehent/.ssh"
            ]
        }
    }
}

Automating the investigation

With a several hundreds pubkeys in LDAP, it is more than necessary to automate the generation of the investigation file. We can do so with Python and a small LDAP helper library called mozlibldap.

The algorithm is very simple: iterate over active LDAP users and retrieve their public keys. Then it find their home directory from LDAP and creates a MIG file search that asserts the content of their authorized_keys file.

The investigation JSON file gets bigs very quickly (2.4MB, and ~40,000 lines), but still runs decently fast on target systems. A single system runs the whole thing in approximately 15 seconds, and since MIG is completely parallelized, running it across the infrastructure takes less than a minute.

Below is the Python script that generate the investigation in MIG's action v2 format.

#!/usr/bin/env python
# This Source Code Form is subject to the terms of the Mozilla Public
# License, v. 2.0. If a copy of the MPL was not distributed with this
# file, You can obtain one at http://mozilla.org/MPL/2.0/.
# Copyright (c) 2015 Mozilla Corporation
# Author: jvehent@mozilla.com

# Requires:
# mozlibldap

from __future__ import print_function
import mozlibldap
import string
import json
import sys

LDAP_URL = 'ldap://someplace.at.mozilla'
LDAP_BIND_DN = 'mail=ldapreadonlyuser,o=com,dc=mozilla'
LDAP_BIND_PASSWD = "readonlyuserpassphrase"


def main():
    lcli = mozlibldap.MozLDAP(LDAP_URL, LDAP_BIND_DN, LDAP_BIND_PASSWD)
    searches = {}

    # get a list of users that have a pubkey in ldap
    users = lcli.get_all_enabled_users_attr('sshPublicKey')
    for user_attr in users:
        search = {}
        user = user_attr[0].split(',', 1)[0].split('=', 1)[1]
        print("current user: "+user, file=sys.stderr)
        keys = user_attr[1]
        if len(keys) == 0:
            continue
        contentre = '^((#.+)|(\s+)'
        for pubkey in keys['sshPublicKey']:
            if len(pubkey) < 5 or not (pubkey.startswith("ssh")):
                continue
            pubkey = string.join(pubkey.split(' ', 2)[:2], '\s')
            pubkey = pubkey.replace('/', '\/')
            pubkey = pubkey.replace('+', '\+')
            contentre += '|({pubkey}\s.+)'.format(pubkey=pubkey)
        contentre += ')$'
        search["names"] = []
        search["names"].append("^authorized_keys$")
        search["contents"] = []
        search["contents"].append(contentre)
        paths = []
        try:
            paths = get_search_paths(lcli, user)
        except:
            continue
        if not paths or len(paths) < 1:
            continue
        search["paths"] = paths
        search["options"] = {}
        search["options"]["matchall"] = True
        search["options"]["macroal"] = True
        search["options"]["maxdepth"] = 1
        search["options"]["mismatch"] = []
        search["options"]["mismatch"].append("content")
        print(json.dumps(search), file=sys.stderr)
        searches[user+"_ssh_pubkeys"] = search
    action = {}
    action["name"] = "Investigate the content of authorized_keys for LDAP users"
    action["target"] = "status='online' AND mode='daemon'"
    action["version"] = 2
    action["operations"] = []
    operation = {}
    operation["module"] = "file"
    operation["parameters"] = {}
    operation["parameters"]["searches"] = searches
    action["operations"].append(operation)
    print(json.dumps(action, indent=4, sort_keys=True))


def get_search_paths(lcli, user):
    paths = []
    res = lcli.query("mail="+user, ['homeDirectory', 'hgHome',
                                    'stageHome', 'svnHome'])
    for attr in res[0][1]:
        try:
            paths.append(res[0][1][attr][0]+"/.ssh")
        except:
            continue
    return paths


if __name__ == "__main__":
    main()

The script write the investigation JSON to stdout and needs to be redirected to a file. We can then use the MIG command line to run the investigation file.

$ ./make-pubkeys-investigation.py > /tmp/investigate_pubkeys.json
$ mig -i /tmp/investigate_pubkeys.json
[info] launching action from file, all flags are ignored
3124 agents will be targeted. ctrl+c to cancel. launching in 5 4 3 2 1 GO
Following action ID 4898767262251.status=inflight......status=completed
- 100.0% done in 34.848325918s
3124 sent, 3124 done, 3124 succeeded
server.example.net /home/bob/.ssh/authorized_keys [lastmodified:2014-05-30 04:04:45 +0000 UTC, mode:-rw-------, size:968] in search 'bob_ssh_pubkeys'
[...]
17 agent have found results

In conclusion

When maintaining the security of a large infrastructure, it is critical to separate the components that perform the configuration from the components that verify the configuration.

While MIG was written primarily as a security investigation platform, its low-level file investigation capabilities can be used to assert the integrity of configurations organization-wide.

This post shows how checks that verify the integrity of SSH Authorized Keys files can be executed using MIG. The checks are designed to consume negligible amounts of resources, and as such should be automated to run every few days in an approach that should be reused for a large amount of sensitive configuration files.

Test your infra, the same way you would test your applications!

Wednesday, August 26 2015

Hosting Go code on Github with a custom import path

We host MIG at https://github.com/mozilla/mig, but while I have tons of respect for the folks at Github, I can't guarantee that we won't use another hosting provider in the future. Telling people to import MIG packages using something of the form "github.com/mozilla/mig/<package>" bothers me, and I've been looking for a better solution for a while.

I bought the domain mig.ninja with the intention to use that as the base import path. I initially tried to use HAProxy to proxy github.com, and somewhat succeeded, but it involved a whole bunch of rewrites that were frankly ugly.

Thankfully, Russ Cox got an even better solution merged into Go 1.4 and I ended up implementing it. Here's how.


Understanding go get

When asked to fetch a package, go get does a number of checks. If the target is on a known hosting site, it fetches the data using a method that is hardcoded (git for github.com, hg for bitbucket, ...). But when using your own domain, go get has no way to know how to fetch the data. To work around that, go lets you specify the vcs method in the import path: import mig.ninja/mig.git The .git indicates to go get that git should be used to retrieve the package. It also doesn't interfere with the code: in the code that import the package, .git is ignored and the package content is accessed using mig.Something.

But that's ugly. No one wants to suffix .git to their import path. There is another, cleaner, solution that use a HTML file to tell go get where the package is located, and which protocol should be used to retrieve it. The file is served from the location of the import path. For an example, let's curl https://mig.ninja/mig:

$ curl https://mig.ninja/mig
<!DOCTYPE html>
<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8"/> <meta name="go-import" content="mig.ninja/mig git https://github.com/mozilla/mig"> <meta http-equiv="refresh" content="0; url=http://mig.mozilla.org"> </head> <body> Nothing to see here; <a href="http://mig.mozilla.org">move along</a>. </body> </html>

The key here is in the <meta> tag named "go-import". When go get requests https://mig.ninja/mig, it hits that HTML file and knows that "mig.ninja/mig" must be retrieved using git from https://github.com/mozilla/mig.

One great aspect of this method, aside from removing the need for .git in the import path, is that "mig.ninja/mig" can now be hosted anywhere as long as the meta tag continues to indicate the authoritative location (in this case: github). It also works nicely with packages under the base repository, such that go get mig.ninja/mig/modules/file works as expected as long as the file is served from that location as well. Note that go get will retrieve the entire repository, not just the target package.


Serving the meta tag from HAProxy

Creating a whole web server for the sole purpose of serving an 11 lines of HTML isn't very appealing. So I reused an existing server that already hosts various things, including this blog, and is powered by HAProxy.

HAProxy can't serve files, but here's the trick, it can serve a custom response at a monitoring uri. I created a new HTTPS backend for mig.ninja that monitors /mig and serves a custom HTTP 200 response.

frontend https-in
        bind 62.210.76.92:443
        mode tcp
        tcp-request inspect-delay 5s
        tcp-request content accept if { req_ssl_hello_type 1 }
        use_backend jve_https if { req_ssl_sni -i jve.linuxwall.info }
        use_backend mig_https if { req_ssl_sni -i mig.ninja }

backend mig_https
        mode tcp
        server mig_https 127.0.0.1:1666

frontend mig_https
        bind 127.0.0.1:1666 ssl no-sslv3 no-tlsv10 crt /etc/certs/mig.ninja.bundle
        mode http
        monitor-uri /mig
        errorfile 200 /etc/haproxy/mig.ninja.200.http
        acl mig_pkg url /mig
        redirect location https://mig.ninja/mig if !mig_pkg

The configuration above uses SNI to serve multiple HTTPS domains from the same IP. When a new connection enters the https_in frontend, it inspects the server_name TLS extension and decides which backend should handle the request. If the request is for mig.ninja, it sends it to the mig_https backend, which forward it to the mig_https frontend. There, the request URI is inspect. If it matches /mig, the file mig.ninja.200.http is returned. Otherwise, a HTTP redirect is returned to send the caller back to https://mig.ninja/mig (in case a longer path, such as mig.ninja/mig/module, was requested).

mig.ninja.200.http is a complete HTTP response, with HTTP headers, body and proper carriage return. HAProxy doesn't process the file at all, it just globs it and sends it back to the client. It looks like this:

HTTP/1.1 200 OK
Cache-Control: no-cache
Connection: close
Content-Type: text/html

<!DOCTYPE html>
<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8"/>
<meta name="go-import" content="mig.ninja/mig git https://github.com/mozilla/mig">
<meta http-equiv="refresh" content="0; url=http://mig.mozilla.org">
</head>
<body>
Nothing to see here; <a href="http://mig.mozilla.org">move along</a>.
</body>
</html>


Building MIG

With all that in place, retrieving mig is now as easy as go get mig.ninja/mig. Icing on the cake, the clients can now be retrieved with go get as well:

$ go get -u mig.ninja/mig/client/mig
$ mig help
$ go get -u mig.ninja/mig/client/mig-console
$ mig-console

Telling everyone to use the right path

We just this in place, you can't guarantee that users of your packages won't directly reference the github import path.

Except that you can! As outlined's in rsc's document, and in go help importpath, it is possible to specify the authoritative location of a package directly in the package itself. This is done by adding a comment with the package location right next to the package name:
package mig /* import "mig.ninja/mig" */

Using the import comment, Go will enforce retrieval from the authoritative source. Other tools will use it as well, for example try accessing MIG's doc from Godoc using the github URL, and you will notice the redirection: https://godoc.org/github.com/mozilla/mig


Trust Github, it's a great service, but controlling your import path is the way to Go ;)

- page 2 of 34 -