Chromium Code Reviews| Index: net/docs/certificate-transparency.md |
| diff --git a/net/docs/certificate-transparency.md b/net/docs/certificate-transparency.md |
| new file mode 100644 |
| index 0000000000000000000000000000000000000000..5e948d3292625582c106f9f19847e76ab2d9332e |
| --- /dev/null |
| +++ b/net/docs/certificate-transparency.md |
| @@ -0,0 +1,186 @@ |
| +# Certificate Transparency |
| + |
| +## Overview |
| + |
| +[Certificate Transparency](http://www.certificate-transparency.org/) (CT) is a |
| +protocol designed to fix several structural flaws in the SSL/TLS certificate |
| +ecosystem. Described by [RFC 6962](https://tools.ietf.org/html/rfc6962) and |
| +the ongoing work in [RFC 6962-bis](https://datatracker.ietf.org/doc/draft-ietf-trans-rfc6962-bis/), |
| +it provides a means of providing a public, append-only data structure that |
| +can log certificates issued by [certificate authorities](https://en.wikipedia.org/wiki/Certificate_authority) (CAs). |
| +By logging these certificates, it becomes possible for site operators to |
| +detect when a certificate may have been issued for their domain without their |
| +approval, and allows browsers and the wider ecosystem to verify that CAs |
| +are following their expected and disclosed practices. |
| + |
| +## Certificate Transparency Basics |
| + |
| +Broadly speaking, the goal of supporting Certificate Transparency is to ensure |
| +that certificates that an application trusts will be publicly disclosed in a |
|
Eran Messeri
2017/03/16 12:10:28
very small nit: that ... that
Can the 2nd 'that'
Ryan Sleevi
2017/03/16 14:10:35
Yup. Thanks
|
| +way sufficient for site operators and application developers to ensure that |
| +nothing is wrong. |
| + |
| +At its most basic level, it's possible to simply introduce Certificate |
| +Transparency logs as trusted third parties, much like CAs are trusted third |
| +parties. If the logs are operated by CAs, this may not be much of a security |
| +improvement, but if the logs are operated by non-CA entities, this might serve |
| +as a sufficient counter-balance to the risks. |
| + |
| +However, with more work, it's possible to minimize the trust afforded to CT |
| +logs, and to automatically and cryptographically verify they're complying with |
| +their stated policies. This can provide even greater assurance to application |
| +developers, site operators, and their users, that the security expected from |
| +certificates is actually being provided. |
| + |
| +For a more thorough threat analysis, see |
| +https://datatracker.ietf.org/doc/draft-ietf-trans-threat-analysis/ that |
| +discusses the different risks in Certificate Transparency, and how the |
| +protocol addresses them. |
| + |
| +## Certificate Transparency in `//net` |
| + |
| +A goal of `//net` is to try to ensure that code is 'safe by default' when |
| +used. As part of serving that goal, in order to make a TLS or QUIC connection |
| +using code in `//net`, it's necessary for the `//net` embedder to make |
| +a decision about Certificate Transparency, much like it is necessary to |
| +provide a [`CertVerifier`](../cert/cert_verifier.h) that describes how to |
| +verify the server's certificate. |
| + |
| +Because this is necessary to make a TLS or QUIC connection, this requirement |
| +surfaces upwards through each layer in the stack - applying to things like the |
| +[`HttpNetworkSession`](../http/http_network_session.h) and upwards to the |
| +[`URLRequestContext`](../url_request/url_request_context.h). |
| + |
| +This requirement is expressed by requiring two separate, but related, objects |
| +to be supplied: A [`CTVerifier`](../cert/ct_verifier.h) and a |
| +[`CTPolicyEnforcer`](../cert/ct_policy_enforcer.h), which together can be used |
| +to express an application's policies with respect to Certificate Transparency. |
| + |
| +As part of the goal of ensuring 'safe by default', `//net` also has various |
| +policies related to certificates issues by particular certificate authorities |
| +whose past actions have created unnecessary security risk for TLS connections, |
| +and as a consequence, are required to have their certificates disclosed using |
| +Certificate Transparency in order to ensure that the security provided by |
| +certificates from these CAs matches the level of security and assurance that |
| +other CAs provide. These policies are implemented in |
| +[`TransportSecurityState`](../http/transport_security_state.cc), via the |
| +`ShouldRequireCT` method |
| + |
| +### CTVerifier |
| + |
| +The `CTVerifier` is the core class for parsing and validating the structures |
| +defined in RFC6962 (or future versions), and providing basic information about |
| +the [`SignedCertificateTimestamps`](https://tools.ietf.org/html/rfc6962#section-3.2) |
| +present within the connection. |
| + |
| +### CTPolicyEnforcer |
| + |
| +The `CTPolicyEnforcer` is the core class for expressing an application's |
| +policies around how it expects Certificate Transparency to be used by the |
| +certificates it trusts and the certificate authorities that issue these |
| +certificates. |
| + |
| +The `CTPolicyEnforcer` currently expresses two policies: |
| + * How to treat [Extended Validation](https://cabforum.org/extended-validation-2/) |
| + certificates (those for which a [`CertVerifier`](../cert/cert_verifier.h) |
| + returned `CERT_STATUS_IS_EV`) |
| + * How to treat all certificates, regardless of EV status. |
| + |
| +### TransportSecurityState |
| + |
| +The `TransportSecurityState::ShouldRequireCT` method implements the core logic |
| +for determining whether or not a connection attempt should be rejected if it |
| +does not comply with an application's Certificate Transparency policy. |
| + |
| +The implementation in `//net` provides a default implementation that tries to |
| +ensure maximum security, by causing connections that fail to abide by an |
| +application's CT policy that are from CAs known to have security issues in the |
| +past. |
| + |
| +Embedders can customize or override this by providing a |
| +`TransportSecurityState::RequireCTDelegate` implementation, which allows |
|
Eran Messeri
2017/03/16 12:10:27
Do I understand correctly that if said TransportSe
Ryan Sleevi
2017/03/16 14:10:35
Nope, because the CTPolicyEnforcer is still called
|
| +applications to inspect the connection information and determine whether |
| +CT should be required, CT should not be required, or whether the default logic |
| +in `//net` should be used. |
| + |
| +## Certificate Transparency in Chromium |
| + |
| +As part of the open-source implementation of Chrome, the policies related to |
| +how Chromium code treats Certificate Transparency are documented at |
| +https://www.chromium.org/Home/chromium-security/certificate-transparency . This |
| +page includes the policies for how Chromium determines an acceptable set of |
| +CT logs and what CT-related information is expected to accompany certificates, |
| +both for EV and non-EV. |
| + |
| +The implementation of these policies lives within [`//net/cert`](../cert), and |
| +include: |
| + * [`ct_known_logs.h`](../cert/ct_known_logs.h): The set of CT logs known and |
| + qualified according to Chromium's |
| + [Certificate Transparency Log Policy](https://www.chromium.org/Home/chromium-security/certificate-transparency/log-policy) |
| + * ['multi_log_ct_verifier.h`](../cert/multi_log_ct_verifier.h): Capable of |
| + parsing SCTs from a variety of logs and validating their signatures, using |
| + the keys and information provided by `ct_known_logs.h`. |
| + * [`ct_policy_enforcer.h`](../cert/ct_policy_enforcer.h): A base class that |
| + implements the Certificate Transparency in Chrome Policy, for both EV and |
| + non-EV certificates. |
| + |
| +## Certificate Transparency for `//net` Consumers |
| + |
| +Not every TLS connection may need the security assurances that |
| +Certificate Transparency aims to provide. For example, some consumers of |
| +`//net` APIs in Chromium use mutual authentication with self-signed |
| +certificates and which are authenticated out-of-band. For these connections, |
| +Certificate Transparency is not relevant, and it's not necessary to parse |
| +or enforce CT related information. |
| + |
| +For these cases, the approach is: |
| + * [`do_nothing_ct_verifier.h`](../cert/do_nothing_ct_verifier.h): A no-op |
| + CTVerifier that does not parse or verify CT-related information. |
| + * A derived `CTPolicyEnforcer` implementation that indicates all |
| + certificates comply with its policies. |
| + **TODO(rsleevi):** Provide a DoNothingCTPolicyEnforcer |
| + |
| +As documented in these classes, care should be taken before using these, as |
| +they provide much weaker security guarantees. In general, emailing |
| +net-dev@chromium.org or discussing it during a security review is the right |
| +answer, and documenting at the instantiation points why it is safe and |
| +acceptable to use these classes. |
| + |
| +## Certificate Transparency for `//net` Embedders |
| + |
| +For projects and third party products that embed `//net`, the policies |
| +that are included as part of the open-source repository may not be |
| +appropriate. This is because the implementations may rely implicitly |
| +or explicitly on several key guarantees that come from Google-branded |
| +distributions and products, and may not be appropriate for other cases. |
| + |
| +These key expectations are: |
| + * A release cycle aligned with Chrome releases; that is, every six weeks, |
| + and on the same versions as Chrome releases. |
| + * Widespread support for automatic updates. |
| + * That [`base::GetBuildTime()`](../../base/build_time.h) will reflect, to |
| + some degree, when the tree was branched and/or released, and will not |
| + be re-generated on recompilation. That is, this implies is_official_build |
| + for binaries released to end-users, but is not enforced in code so that |
| + developers can accurately test release behavior. |
| + * Support for dynamic [`base::FieldTrial`](../../base/metrics/field_trial.h) |
| + configurations. |
| + |
| +For projects that don't support automatic updates, or which measure 'stable' |
| +on the order of months to years, or which don't have tools suitable to |
| +respond to changes in the Certificate Authority and Certificate Transparency |
| +ecosystem, it may not be appropriate to enable Certificate Transparency |
| +support yet. |
| + |
| +These issues are not unique or particular to Certificate Transparency - in |
| +many ways, they're similar to issues already faced with determining which |
| +certificate authorities are trusted and how to successfully validate a |
| +TLS server's certificate. However, as the Certificate Transparency ecosystem |
| +is still growing, it may be suitable to disable support until some of the |
| +solutions to these challenges stablize. |
| + |
| +To opt-out of enforcing Certificate Transparency, using the `DoNothing` |
| +variants discussed above provides a suitable implementation that will opt to |
| +'fail open' instead. This may provide less security, but provides greater |
| +stability, and minimizes the risk that these `//net` embedding clients |
| +might cause to the CT ecosystem or receive from enabling CT. |