Personal tools

CA:Problematic Practices

From MozillaWiki

Jump to: navigation, search

Contents

Potentially problematic CA practices

This page contains draft comments about various CA practices that have been the subject of discussion in past CA evaluations. In general these practices are not explicitly addressed by the Mozilla CA certificate policy, and we do not necessarily consider them security risks. However we want to highlight them because they've occasioned controversy in the past and have in some cases caused approval of applications to be delayed. Some of these practices may be addressed in future versions of the policy.

Long-lived DV certificates

A domain-validated SSL certificate attests only to ownership and control of a domain name, and the owner of a domain name may have acquired it from others. It is therefore possible for the previous owner of the domain to have a still-valid DV certificate for the domain. If such a valid certificate (and associated private key) were to be used in conjunction with a DNS spoofing attack it would allow a malicious site to masquerade as a legitimate site and bypass the protection afforded by SSL.

Some CAs issue DV SSL certificates that have expiration times several years in the future. This increases the time during which the possibility of such an attack exists.

Important: According to section 6 of the Mozilla CA Certificate Inclusion Policy CAs who issue long-lived SSL certs must "verify that all of the information that is included in SSL certificates remains current and correct at time intervals of thirty-nine months or less"

Wildcard DV SSL certificates

Some CAs issue domain-validated SSL certificates that can function as wildcard certificates, e.g., a certificate for *.example.com where the CA verifies only ownership and control of the example.com domain, and the certificate subscriber can then use the certificate with any site foo.example.com, bar.example.com, etc. This means that a subscriber could establish malicious SSL-protected web site that are deliberately named in imitation of legitimate sites, e.g., paypal.example.com, without knowledge of the CA. Concerns have been expressed that wildcard SSL certificates should not be issued except to subscribers whose actual identity has been validated with organizational validation (OV). (There are no EV wildcard certificates.)

Email Address Prefixes for DV Certs

For domain-validated SSL certificates, many CAs use an email challenge-response mechanism to verify that the SSL certificate subscriber owns/controls the domain to be included in the certificate. Some CAs allow applicants to select an address from a predetermined list to be used for this verification.

Offering too many options for the email address prefix increases the risk of issuing a certificate to a subscriber who does not own/control the domain. Therefore, the list of email address prefixes should be limited.

Mozilla's recommendation is to limit the set of verification addresses to the following. Discussion is underway to make this a mandatory part of Mozilla's requirements.

  • admin@domain
  • administrator@domain
  • webmaster@domain
  • hostmaster@domain
  • postmaster@domain
  • Plus any address listed in the technical or administrative contact field of the domain's WHOIS record, regardless of the addresses' domains.

The list above is case-insensitive. However, when the email verification message is sent, it should be sent to the address with the same capitalization as specified by the certificate subscriber. For example, a certificate subscriber requests that validation be sent to PostMaster@foo.com, and this is allowed because a case-insensitive comparison to the list of acceptable email addresses succeeds. The verification message will be sent to PostMaster@foo.com, with the capitalization that was specified by the certificate subscriber.

Delegation of Domain / Email validation to third parties

Domain and Email validation are core-requirements of the Mozilla CA Policy and should always be incorporated into the issuing CAs procedures whenever possible. Registration Authorities (RA) or other third parties performing such functions must provide attestations about their procedures and/or should be audited together with the issuing CA. The CA must demonstrate clear and efficient controls attesting the performance of its RAs. Delegation of domain/email validation to third parties should generally be avoided.

Issuing end entity certificates directly from roots

Some CAs issue end entity certificates directly from the root (i.e., signed using the root CA private key). This is not as secure as using an offline root and issuing certificates using a subordinate CA.

Allowing external entities to operate subordinate CAs

Some CAs authorize external entities to operate their own CAs as subordinate CAs under the original CA's root. This raises concerns relating to whether or not such external entities are audited in a manner equivalent to the root CA, as well as what legal and technical arrangements constrain the external entities.

Where a root from a CA signs an intermediate certificate used by an external CA to then sign subsidiary intermediate certificates or subscriber certificates, that situation needs to be disclosed. That disclosure should include documentation of what requirements are imposed by the CA owning the root upon the operations of external CAs. Further, the public audit report for the CA owning the root must indicate how and when the operations of the external CAs have been reviewed for compliance with those documented requirements.

You must provide a clear description of the subordinate CAs that are operated by external third parties, and an explanation as to how the CP/CPS and audits ensure the third parties are in compliance with Mozilla's CA Certificate Policy requirements as per the Subordinate CA Checklist.

Distributing generated private keys in PKCS#12 files

It is reported that some CAs generate the key pairs for their subscribers, rather than having the subscribers generate their own key pairs, and once generated, those CAs distribute the private key, together with the issued public key certificate and its chain, to the subscriber in a PKCS#12 file. The issues include:

  • The user doesn't know or control who else possesses and can use his private key (decrypt his private messages or forge his signature), and
  • The distribution channels used (e.g. unencrypted email) may not be adequately secured.

CAs must never generate the key pairs for signer or SSL certificates. CAs may only generate the key pairs for SMIME encryption certificates. Distribution or transfer of certificates in PKCS#12 form through unsecure electronic channels is not allowed. If a PKCS#12 file is distributed via a physical data storage device, then

  • The storage must be packaged in a way that the opening of the package causes irrecoverable physical damage. (e.g. a security seal)
  • The PKCS#12 file must have a sufficiently secure password, and the password must not be transferred together with the storage.

Certificates referencing hostnames or private IP addresses

The standard model for SSL on the web assumes that an SSL certificate references a domain name that is resolvable using the public DNS infrastructure (e.g., "www.example.com") or an IP address that is reachable from the public Internet. However it is also possible to include in a certificate a hostname not resolvable through the public DNS (e.g., "home") or a private IP address (e.g., 192.168.1.101); for example, this might be done for a corporate intranet with SSL-enabled servers behind a firewall and employees who don't want to enter fully-qualified domain names.

We consider this a problematic practice for a public CA because a subscriber who obtains a certificate of this type could in theory use it in contexts other than the one for which the certificate was obtained, and in particular could use it to help enable an SSL MITM attack on users in other organizations who are using the same hostname or IP address for their own SSL-enabled servers. (Depending on the hostnames and private IP addresses used, this vulnerability might also affect users of home networks with SSL-enabled home gateway devices.)

Guidance on the Deprecation of Internal Server Names and Reserved IP Addresses

CA/Browser Forum Baseline Requirements for the Issuance and Management of Publicly-Trusted Certificates, BR 9.2.1: “As of the Effective Date of these Requirements, prior to the issuance of a Certificate with a subjectAlternativeName (SAN) extension or Subject Common Name field containing a Reserved IP Address or Internal Server Name, the CA shall notify the Applicant that the use of such Certificates has been deprecated by the CA / Browser Forum and that the practice will be eliminated by October 2016. Also as of the Effective Date, the CA shall not issue a certificate with an Expiry Date later than 1 November 2015 with a SAN or Subject Common Name field containing a Reserved IP Address or Internal Server Name. As from 1 October 2016, CAs shall revoke all unexpired Certificates.”

It is also a problematic practice to issue a certificate with non resolvable DNS or private IP and resolvable DNS adresses together.

It is not standards compliant for printable ASCII representations of IP addresses to be placed in any certificate field that is intended to hold DNS names, including the subject common name and the DNSName field of the Subject Alternative Names extension. There is a place in a certificate specifically intended to be where IP (v4 or v6) addresses may be placed. It is in the Subject Alternative Names extension. The SubjectAltNames extension has places for both additional DNS names and for IP addresses. The place for IP addresses takes them in binary form, not in printable ASCII (e.g. dotted decimal) form. See bug 553754.

Issuing SSL Certificates for Internal Domains

It has come to our attention that some Certification Authorities may have mistakenly issued SSL certificates to non-existent .int domain names. This appears to have happened because the .int domain may have been confused with internal domain names, and not all of the CAs and RAs may be aware that .int is an ICANN approved TLD.

Section 7 of Mozilla’s CA Certificate Policy states that CAs need to take “reasonable measures to verify that the entity submitting the certificate signing request has registered the domain(s) referenced in the certificate.” There are different interpretations as to what this means in regards to internal domain names such as non-valid TLDs, hostnames, and IP addresses. However, there is consensus that there are problems associated with issuing certificates for servers on internal networks under the same CA hierarchy as certificates for servers on public networks. Mozilla is currently discussing whether the CA Certificate Policy should be updated to add more explicit requirements on this practice, or even to disallow it altogether.

If you have issued certificates for internal domains within your CA hierarchy, Mozilla requests that you take the following actions:

  1. Perform an internal audit to look for certificates that have been issued within your CA hierarchy which have .int domain names in the Common Name and/or as DNS Names in the subjectAlternativeName. For each of these certificates, check to see if the certificate subscriber owns/controls that domain name, and revoke the certificate if they do not own/control that domain name.
  2. Review your controls/procedures (both internally and your RAs) for correct identification of internal and external domain names and verification that subscribers own/control the domain name to be included in their certificate. Please refer to these documents:

Mozilla also recommends that you

  1. Implement automated checks to signal a red flag for domains such as .int and null characters in the Common Name and subjectAlternativeName of certificates.
  2. Maintain your own list of ICANN approved TLDs that are eligible to be used for domains in certificates issued within your CA hierarchy. If a new TLD is created by IANA, make an explicit decision whether or not to add the new TLD to your list.

OCSP Responses signed by a certificate under a different root

CAs who issue certificates with OCSP URLs in AIA extensions should make sure that the OCSP responses conform to RFC 2560, and work correctly for Mozilla users without requiring the user to find and install the OCSP responder's certificate, that is, the certificate with which the OCSP response signatures are verified.

RFC 2560, sections 2.2, 2.6, 3.2 and 4.2.2.2 define the requirements for the OCSP response signer's certificate and certificate chain. NSS enforces these requirements exactly.

When an OCSP responder URL is included in end-entity certificates, Firefox will by default attempt to check the certificate's status via OCSP. If the OCSP signer certificate is not the certificate of the CA that issued the certificate in question and is not issued by the CA that issued the certificate in question, the OCSP check will fail with an NSS error code for OCSP, such as SEC_ERROR_OCSP_UNAUTHORIZED_REQUEST or SEC_ERROR_OCSP_UNAUTHORIZED_RESPONSE.

For a detailed explanation about why an OCSP responder should not use a self-signed OCSP responder certificate and depend on Trusted Responder Mode within the Firefox browser, see: Details about OCSP Trusted Responder Mode.

Please test your OCSP responder within the Firefox browser by enforcing OCSP as per our CA Recommended Practices for OCSP.

CRL with critical CIDP Extension

Currently Firefox handles "full" CRLs, but not "partitioned" CRLs. Partitioned CRLs are identified by the presence of a CRL Issuing Distribution Point (CIDP) extension flagged as critical. Firefox is not presently able to load CRLs with critical CIDP extensions. When attempting to load a CRL with a critical CIDP extension, Firefox will return the error code ffffe095, which is equivalent to the negative decimal number -8043. According to the NSS Error Codes this error corresponds to SEC_ERROR_CRL_UNKNOWN_CRITICAL_EXTENSION.

The NSS team hopes to eventually implement partitioned CRLs, and when that work is done, Firefox should allow CRLs with critical CIDP extensions. However, even when that is done, older versions of Firefox will still not be able to load CRLs with critical CIDP extensions.

Our recommendation is to not put critical CIDP extensions into full CRLs, and to make full CRLs available for download when practical.

Generic names for CAs

In various contexts Firefox and other Mozilla-based products display to users the names of root CAs, issuing CAs, and intermediate CAs in general. In some cases CA names are very generic, e.g., "Secure Server CA"; this makes it difficult for users to ascertain who operates the CA without undertaking a detailed investigation.

Our recommendation is that all CA names incorporate an organizational name or product brand name sufficiently unique to allow relatively straightforward identification of the CA.

Additionally, the issuer and subject information in the root certificate should provide clear indication about who owns or operates the certificate. Generic issuer and subject information inhibits the users' ability to establish a chain of trust, and to pursue complaints when appropriate. For instance, the following issuer information would not be acceptable in a root certificate to be included in NSS.

  • CN = Root CA
  • OU = Certification Authorities
  • OU = Services
  • O = admin

There is no information in this issuer that can be linked back to any particular CA. There is no distinguishable company name or brand name. All of the information in this issuer is too generic to do a search on and hope to find the CA.

Important: Both the O and the CN must be meaningful, and not generic terms such as "admin" or "root". It is not acceptable to have the O be a generic term such as "Admin" because it could mislead users that rely on the issuer details, such as when you hover your mouse over the domain or organization section in the address bar.

Lack of Communication With End Users

CAs should be contactable by, and accept and act upon complaints made by, those relying on their assertions of identity. For CAs included in Mozilla, this will include being responsive to members of the general public, including people who have not purchased products from that CA.

Backdating the notBefore date

Certificates do not contain an issue timestamp, so it is not possible to be certain when they were issued. The notBefore date is the start of the certificate's validity range, and is set by the CA. It should be a reasonable reflection of the date on which the certificate was issued. Minor tweaking for technical compatibility reasons is accepted, but backdating certificates in order to avoid some deadline or code-enforced restriction is not.

Other considerations when updating the CA Certificate Policy

Many of the descriptions of the practices above will provide food for thought when and if we are making further updates to the CA Certificate Policy. Other issues which might be considered at that time include:

Root Count Restrictions

It has been suggested that, when the CA cert policy is revised, we restrict the number of roots any one CA may have to e.g. 3. This is because more roots increases the download size of the product.

Restrict government roots to their TLDs

A suggestion for a future revision of the policy is: we should restrict government run/sponsored roots to only issuing certificates for the corresponding TLD.

There are, of course, questions such as:

  • What defines a government root
  • What if they have all the necessary audits anyway

and so on. These would need to be discussed.

Minimum Key Sizes

One suggestion for a future revision of the CA Cert Policy is that we should specify minimum key sizes, either just for roots or for roots, intermediates and end entity certificates.

The exact restrictions would need to be discussed, but doubtless we would take into account the views of our crypto team and advice from places like NIST.

Dates for Phasing out MD5-based signatures and 1024-bit moduli

Max Time Between Audits

It has been suggested that, when the CA cert policy is revised, we specify the maximum period allowed between audits. WebTrust currently specifies 12 months, and the same is (I understand) recommended for ETSI audits.

Actual Paperwork

It has been suggested that CAs should submit some paperwork by postal mail as well as electronically. A formal inclusion request and general details from the CA in question might help Mozilla in the case of legal problems in the future.

Apparently Apple and Microsoft do require physical paperwork.

Improve definition of "independent"; add idea of "trustworthy"

Currently, the guidelines talk about an auditor having to be both "independent" and "competent". It has been suggested that the definition of independent should be changed to be more like that the inverse of the MPL's definition of You:

"For legal entities, "You" includes any entity which controls, is controlled by, or is under common control with You. For purposes of this definition, "control" means (a) the power, direct or indirect, to cause the direction or management of such entity, whether by contract or otherwise, or (b) ownership of more than fifty percent (50%) of the outstanding shares or beneficial ownership of such entity."

Additionally, a new "trustworthiness" requirement would be added, which would address some of the issues currently listed under "independent", such as being bound to render a true judgement. This is because one could imagine an auditor who was (under the above definition) independent and also competent, but may nevertheless always provide "the right result" on payment of a fee.

Validate all Data included in Certificates

Only data that has been verified to be correct should be included in a certificate. All information that is supplied by the requester must be verified to be correct before it may be included in the certificate. For example, for SSL certificates, alternative names need to be validated just as well as the subject. And for email certificates, if only the email address of the certificate subscriber is verified, then the CN should not include any unverified element that the subscriber supplied.

DNS names in SANs

It would be appropriate for Mozilla CA policy to mandate that CAs put all DNS names for a cert into SANs. It would not be necessary to go beyond that and disallow CAs to ALSO ADDITIONALLY put one DNS name in the subject common name for the benefit of VERY old (more than 12 year old) browsers that don't recognize SANs. It is only necessary to make it clear that ALL the DNS names (not all but one) must go into the SAN.

Some CAs mistakenly believe that one primary DNS name should go into the Subject Common Name and all the others into the SAN. That's wrong. ALL should go into the SAN.

Then, modern browsers should stop paying attention to Subject common names (bug 552346). Doesn't matter what CAs put there as long as browsers don't look there.