The Gem::Security implements cryptographic signatures for gems. The section below is a step-by-step guide to using signed gems and generating your own.
In order to start signing your gems, you’ll need to build a private key and a self-signed certificate. Here’s how:
# build a private key and certificate for yourself: $ gem cert --build you@example.com
This could take anywhere from a few seconds to a minute or two, depending on the speed of your computer (public key algorithms aren’t exactly the speediest crypto algorithms in the world). When it’s finished, you’ll see the files “gem-private_key.pem” and “gem-public_cert.pem” in the current directory.
First things first: Move both files to ~/.gem if you don’t already have a key and certificate in that directory. Ensure the file permissions make the key unreadable by others (by default the file is saved securely).
Keep your private key hidden; if it’s compromised, someone can sign packages as you (note: PKI has ways of mitigating the risk of stolen keys; more on that later).
In RubyGems 2 and newer there is no extra work to sign a gem. RubyGems will automatically find your key and certificate in your home directory and use them to sign newly packaged gems.
If your certificate is not self-signed (signed by a third party) RubyGems will attempt to load the certificate chain from the trusted certificates. Use gem cert --add signing_cert.pem to add your signers as trusted certificates. See below for further information on certificate chains.
If you build your gem it will automatically be signed. If you peek inside your gem file, you’ll see a couple of new files have been added:
$ tar tf your-gem-1.0.gem metadata.gz metadata.gz.sig # metadata signature data.tar.gz data.tar.gz.sig # data signature checksums.yaml.gz checksums.yaml.gz.sig # checksums signature
If you wish to store your key in a separate secure location you’ll need to set your gems up for signing by hand. To do this, set the signing_key and cert_chain in the gemspec before packaging your gem:
s.signing_key = '/secure/path/to/gem-private_key.pem' s.cert_chain = %w[/secure/path/to/gem-public_cert.pem]
When you package your gem with these options set RubyGems will automatically load your key and certificate from the secure paths.
Now let’s verify the signature. Go ahead and install the gem, but add the following options: -P HighSecurity, like this:
# install the gem with using the security policy "HighSecurity" $ sudo gem install your.gem -P HighSecurity
The -P option sets your security policy – we’ll talk about that in just a minute. Eh, what’s this?
$ gem install -P HighSecurity your-gem-1.0.gem
ERROR: While executing gem ... (Gem::Security::Exception)
root cert /CN=you/DC=example is not trusted
The culprit here is the security policy. RubyGems has several different security policies. Let’s take a short break and go over the security policies. Here’s a list of the available security policies, and a brief description of each one:
NoSecurity - Well, no security at all. Signed packages are treated like unsigned packages.
LowSecurity - Pretty much no security. If a package is signed then RubyGems will make sure the signature matches the signing certificate, and that the signing certificate hasn’t expired, but that’s it. A malicious user could easily circumvent this kind of security.
MediumSecurity - Better than LowSecurity and NoSecurity, but still fallible. Package contents are verified against the signing certificate, and the signing certificate is checked for validity, and checked against the rest of the certificate chain (if you don’t know what a certificate chain is, stay tuned, we’ll get to that). The biggest improvement over LowSecurity is that MediumSecurity won’t install packages that are signed by untrusted sources. Unfortunately, MediumSecurity still isn’t totally secure – a malicious user can still unpack the gem, strip the signatures, and distribute the gem unsigned.
HighSecurity - Here’s the bugger that got us into this mess. The HighSecurity policy is identical to the MediumSecurity policy, except that it does not allow unsigned gems. A malicious user doesn’t have a whole lot of options here; they can’t modify the package contents without invalidating the signature, and they can’t modify or remove signature or the signing certificate chain, or RubyGems will simply refuse to install the package. Oh well, maybe they’ll have better luck causing problems for CPAN users instead :).
The reason RubyGems refused to install your shiny new signed gem was because it was from an untrusted source. Well, your code is infallible (naturally), so you need to add yourself as a trusted source:
# add trusted certificate gem cert --add ~/.gem/gem-public_cert.pem
You’ve now added your public certificate as a trusted source. Now you can install packages signed by your private key without any hassle. Let’s try the install command above again:
# install the gem with using the HighSecurity policy (and this time # without any shenanigans) $ gem install -P HighSecurity your-gem-1.0.gem Successfully installed your-gem-1.0 1 gem installed
This time RubyGems will accept your signed package and begin installing.
While you’re waiting for RubyGems to work it’s magic, have a look at some of the other security commands by running gem help cert:
Options:
-a, --add CERT Add a trusted certificate.
-l, --list [FILTER] List trusted certificates where the
subject contains FILTER
-r, --remove FILTER Remove trusted certificates where the
subject contains FILTER
-b, --build EMAIL_ADDR Build private key and self-signed
certificate for EMAIL_ADDR
-C, --certificate CERT Signing certificate for --sign
-K, --private-key KEY Key for --sign or --build
-A, --key-algorithm ALGORITHM Select key algorithm for --build from RSA, DSA, or EC. Defaults to RSA.
-s, --sign CERT Signs CERT with the key from -K
and the certificate from -C
-d, --days NUMBER_OF_DAYS Days before the certificate expires
-R, --re-sign Re-signs the certificate from -C with the key from -K
We’ve already covered the --build option, and the --add, --list, and --remove commands seem fairly straightforward; they allow you to add, list, and remove the certificates in your trusted certificate list. But what’s with this --sign option?
To answer that question, let’s take a look at “certificate chains”, a concept I mentioned earlier. There are a couple of problems with self-signed certificates: first of all, self-signed certificates don’t offer a whole lot of security. Sure, the certificate says Yukihiro Matsumoto, but how do I know it was actually generated and signed by matz himself unless he gave me the certificate in person?
The second problem is scalability. Sure, if there are 50 gem authors, then I have 50 trusted certificates, no problem. What if there are 500 gem authors? 1000? Having to constantly add new trusted certificates is a pain, and it actually makes the trust system less secure by encouraging RubyGems users to blindly trust new certificates.
Here’s where certificate chains come in. A certificate chain establishes an arbitrarily long chain of trust between an issuing certificate and a child certificate. So instead of trusting certificates on a per-developer basis, we use the PKI concept of certificate chains to build a logical hierarchy of trust. Here’s a hypothetical example of a trust hierarchy based (roughly) on geography:
--------------------------
| rubygems@rubygems.org |
--------------------------
|
-----------------------------------
| |
---------------------------- -----------------------------
| seattlerb@seattlerb.org | | dcrubyists@richkilmer.com |
---------------------------- -----------------------------
| | | |
--------------- ---------------- ----------- --------------
| drbrain | | zenspider | | pabs@dc | | tomcope@dc |
--------------- ---------------- ----------- --------------
Now, rather than having 4 trusted certificates (one for drbrain, zenspider, pabs@dc, and tomecope@dc), a user could actually get by with one certificate, the “rubygems@rubygems.org” certificate.
Here’s how it works:
I install “rdoc-3.12.gem”, a package signed by “drbrain”. I’ve never heard of “drbrain”, but his certificate has a valid signature from the “seattle.rb@seattlerb.org” certificate, which in turn has a valid signature from the “rubygems@rubygems.org” certificate. Voila! At this point, it’s much more reasonable for me to trust a package signed by “drbrain”, because I can establish a chain to “rubygems@rubygems.org”, which I do trust.
The --sign option allows all this to happen. A developer creates their build certificate with the --build option, then has their certificate signed by taking it with them to their next regional Ruby meetup (in our hypothetical example), and it’s signed there by the person holding the regional RubyGems signing certificate, which is signed at the next RubyConf by the holder of the top-level RubyGems certificate. At each point the issuer runs the same command:
# sign a certificate with the specified key and certificate # (note that this modifies client_cert.pem!) $ gem cert -K /mnt/floppy/issuer-priv_key.pem -C issuer-pub_cert.pem --sign client_cert.pem
Then the holder of issued certificate (in this case, your buddy “drbrain”), can start using this signed certificate to sign RubyGems. By the way, in order to let everyone else know about his new fancy signed certificate, “drbrain” would save his newly signed certificate as ~/.gem/gem-public_cert.pem
Obviously this RubyGems trust infrastructure doesn’t exist yet. Also, in the “real world”, issuers actually generate the child certificate from a certificate request, rather than sign an existing certificate. And our hypothetical infrastructure is missing a certificate revocation system. These are that can be fixed in the future…
At this point you should know how to do all of these new and interesting things:
build a gem signing key and certificate
adjust your security policy
modify your trusted certificate list
sign a certificate
In case you don’t trust RubyGems you can verify gem signatures manually:
Fetch and unpack the gem
gem fetch some_signed_gem tar -xf some_signed_gem-1.0.gem
Grab the public key from the gemspec
gem spec some_signed_gem-1.0.gem cert_chain | \ ruby -ryaml -e 'puts YAML.load($stdin)' > public_key.crt
Generate a SHA1 hash of the data.tar.gz
openssl dgst -sha1 < data.tar.gz > my.hash
Verify the signature
openssl rsautl -verify -inkey public_key.crt -certin \ -in data.tar.gz.sig > verified.hash
Compare your hash to the verified hash
diff -s verified.hash my.hash
Repeat 5 and 6 with metadata.gz
OpenSSL Reference The .pem files generated by –build and –sign are PEM files. Here’s a couple of useful OpenSSL commands for manipulating them:
# convert a PEM format X509 certificate into DER format: # (note: Windows .cer files are X509 certificates in DER format) $ openssl x509 -in input.pem -outform der -out output.der # print out the certificate in a human-readable format: $ openssl x509 -in input.pem -noout -text
And you can do the same thing with the private key file as well:
# convert a PEM format RSA key into DER format: $ openssl rsa -in input_key.pem -outform der -out output_key.der # print out the key in a human readable format: $ openssl rsa -in input_key.pem -noout -text
There’s no way to define a system-wide trust list.
custom security policies (from a YAML file, etc)
Simple method to generate a signed certificate request
Support for OCSP, SCVP, CRLs, or some other form of cert status check (list is in order of preference)
Support for encrypted private keys
Some sort of semi-formal trust hierarchy (see long-winded explanation above)
Path discovery (for gem certificate chains that don’t have a self-signed root) – by the way, since we don’t have this, THE ROOT OF THE CERTIFICATE CHAIN MUST BE SELF SIGNED if Policy#verify_root is true (and it is for the MediumSecurity and HighSecurity policies)
Better explanation of X509 naming (ie, we don’t have to use email addresses)
Honor AIA field (see note about OCSP above)
Honor extension restrictions
Might be better to store the certificate chain as a PKCS#7 or PKCS#12 file, instead of an array embedded in the metadata.
Paul Duncan <pabs@pablotron.org> pablotron.org/
This module contains various utility methods as module methods.
Format raw random number as Random does
Implementation of an X.509 certificate as specified in RFC 5280. Provides access to a certificate’s attributes and allows certificates to be read from a string, but also supports the creation of new certificates from scratch.
Certificate is capable of handling DER-encoded certificates and certificates encoded in OpenSSL’s PEM format.
raw = File.read "cert.cer" # DER- or PEM-encoded certificate = OpenSSL::X509::Certificate.new raw
A certificate may be encoded in DER format
cert = ... File.open("cert.cer", "wb") { |f| f.print cert.to_der }
or in PEM format
cert = ... File.open("cert.pem", "wb") { |f| f.print cert.to_pem }
X.509 certificates are associated with a private/public key pair, typically a RSA, DSA or ECC key (see also OpenSSL::PKey::RSA, OpenSSL::PKey::DSA and OpenSSL::PKey::EC), the public key itself is stored within the certificate and can be accessed in form of an OpenSSL::PKey. Certificates are typically used to be able to associate some form of identity with a key pair, for example web servers serving pages over HTTPs use certificates to authenticate themselves to the user.
The public key infrastructure (PKI) model relies on trusted certificate authorities (“root CAs”) that issue these certificates, so that end users need to base their trust just on a selected few authorities that themselves again vouch for subordinate CAs issuing their certificates to end users.
The OpenSSL::X509 module provides the tools to set up an independent PKI, similar to scenarios where the ‘openssl’ command line tool is used for issuing certificates in a private PKI.
First, we need to create a “self-signed” root certificate. To do so, we need to generate a key first. Please note that the choice of “1” as a serial number is considered a security flaw for real certificates. Secure choices are integers in the two-digit byte range and ideally not sequential but secure random numbers, steps omitted here to keep the example concise.
root_key = OpenSSL::PKey::RSA.new 2048 # the CA's public/private key root_ca = OpenSSL::X509::Certificate.new root_ca.version = 2 # cf. RFC 5280 - to make it a "v3" certificate root_ca.serial = 1 root_ca.subject = OpenSSL::X509::Name.parse "/DC=org/DC=ruby-lang/CN=Ruby CA" root_ca.issuer = root_ca.subject # root CA's are "self-signed" root_ca.public_key = root_key.public_key root_ca.not_before = Time.now root_ca.not_after = root_ca.not_before + 2 * 365 * 24 * 60 * 60 # 2 years validity ef = OpenSSL::X509::ExtensionFactory.new ef.subject_certificate = root_ca ef.issuer_certificate = root_ca root_ca.add_extension(ef.create_extension("basicConstraints","CA:TRUE",true)) root_ca.add_extension(ef.create_extension("keyUsage","keyCertSign, cRLSign", true)) root_ca.add_extension(ef.create_extension("subjectKeyIdentifier","hash",false)) root_ca.add_extension(ef.create_extension("authorityKeyIdentifier","keyid:always",false)) root_ca.sign(root_key, OpenSSL::Digest.new('SHA256'))
The next step is to create the end-entity certificate using the root CA certificate.
key = OpenSSL::PKey::RSA.new 2048 cert = OpenSSL::X509::Certificate.new cert.version = 2 cert.serial = 2 cert.subject = OpenSSL::X509::Name.parse "/DC=org/DC=ruby-lang/CN=Ruby certificate" cert.issuer = root_ca.subject # root CA is the issuer cert.public_key = key.public_key cert.not_before = Time.now cert.not_after = cert.not_before + 1 * 365 * 24 * 60 * 60 # 1 years validity ef = OpenSSL::X509::ExtensionFactory.new ef.subject_certificate = cert ef.issuer_certificate = root_ca cert.add_extension(ef.create_extension("keyUsage","digitalSignature", true)) cert.add_extension(ef.create_extension("subjectKeyIdentifier","hash",false)) cert.sign(root_key, OpenSSL::Digest.new('SHA256'))
The parent class for all constructed encodings. The value attribute of a Constructive is always an Array. Attributes are the same as for ASN1Data, with the addition of tagging.
Most constructed encodings come in the form of a SET or a SEQUENCE. These encodings are represented by one of the two sub-classes of Constructive:
OpenSSL::ASN1::Sequence
Please note that tagged sequences and sets are still parsed as instances of ASN1Data. Find further details on tagged values there.
int = OpenSSL::ASN1::Integer.new(1) str = OpenSSL::ASN1::PrintableString.new('abc') sequence = OpenSSL::ASN1::Sequence.new( [ int, str ] )
int = OpenSSL::ASN1::Integer.new(1) str = OpenSSL::ASN1::PrintableString.new('abc') set = OpenSSL::ASN1::Set.new( [ int, str ] )
An OpenSSL::OCSP::Response contains the status of a certificate check which is created from an OpenSSL::OCSP::Request.
An OpenSSL::OCSP::BasicResponse contains the status of a certificate check which is created from an OpenSSL::OCSP::Request. A BasicResponse is more detailed than a Response.
An OpenSSL::OCSP::CertificateId identifies a certificate to the CA so that a status check can be performed.
Generic exception that is raised if an operation on an RSA PKey fails unexpectedly or in case an instantiation of an instance of RSA fails due to non-conformant input data.
Generic exception class of the Timestamp module.
Immutable and read-only representation of a timestamp response returned from a timestamp server after receiving an associated Request. Allows access to specific information about the response but also allows to verify the Response.
Creates binary representations of a SecBufferDesc structure, including the SecBuffer contained inside.
CSV::InputsScanner receives IO inputs, encoding and the chunk_size. It also controls the life cycle of the object with its methods keep_start, keep_end, keep_back, keep_drop.
CSV::InputsScanner.scan() tries to match with pattern at the current position. If there’s a match, the scanner advances the “scan pointer” and returns the matched string. Otherwise, the scanner returns nil.
CSV::InputsScanner.rest() returns the “rest” of the string (i.e. everything after the scan pointer). If there is no more data (eos? = true), it returns “”.
See Net::HTTPGenericRequest for attributes and methods.
This class represents a response received by the SMTP server. Instances of this class are created by the SMTP class; they should not be directly created by the user. For more information on SMTP responses, view Section 4.2 of RFC 5321