Jump to: navigation, search

V4/Automatic Certificate Request Generation/Mapping Rules

Overview

Certificate Mapping Rule objects, as defined in V4/Automatic_Certificate_Request_Generation#Mapping_data_to_fields, need a mechanism for defining how to use the data stored in IPA objects to construct the configuration strings that are passed to the CSR generation helper. This document aims to elaborate on the features required for this mapping, and describe the current implementation as well as possible alternatives.

Requirements

There are two different types of mapping rules used in CSR data formatting:

Data rules
Describe how to format an individual data item
Syntax rules
Describe how to combine data rules in order to create a complete certificate field

Data rules

  • Include data attribute values
  • Include literal values
  • Include values requested from the user via a prompt
  • For certutil: replace commas with a different separator character (; or +) within a DN-type subject alternative name
  • For certutil: shell quoting of parameters to prevent injection attacks
  • Suppress rendering of a rule if necessary data is missing

Syntax rules

  • Distinguish between generated config files and command-line arguments
  • For openssl: produce and reference new config file sections
  • For openssl: place config value in root or extensions section as appropriate
  • Combine several values for same field (e.g. comma-separated list)
  • Suppress rendering of a rule if none of the data rules it contains are going to be rendered

Planned implementation

We will use Jinja2 for templating. Data rules and syntax rules will both be snippets of Jinja2 markup. The formatting process will be as follows:

  1. Syntax rules will be rendered using Jinja2. Data rules (rule text, not rendered) will be passed as the datarules attribute.
  2. Rendered syntax rules will be processed by the Formatter class for the selected CSR generation helper. The formatter combines these partial rules into a full template for the config.
  3. The template will be rendered using Jinja2. Several objects from the IPA server (discussed below) will be available in the context for this rendering.
  4. The final rendered template will be returned to the caller, labeled with its function (e.g. a command line or a config file)

Data available during final render

subject
dict of data from the IPA object corresponding to the subject of this certificate
config
dict of IPA configuration data (result of the config_show API call)

Special features

Some less-than-obvious features will be required to support generating the needed configs. Some of them will require extensions to the Jinja2 language, others just require specific usage of the tools that Jinja2 already provides.

{% section %}{% endsection %}
This tag will be introduced to make the included block into a new section, as for an openssl config file.
Easy escaping
Some tags, like {% section %}, are only meaningful during the second (full-template) render. During the first render they will be implemented to just insert the same tag in the output.
Location specifier
Rules must be able to provide guidance to the formatter about where they belong in the config file (for example, whether a rule is an extension or goes in the base [req] section). This is accomplished by setting a template variable in the syntax rule, for example {% set location = "req" %}

Example

Example data rules:

email={{subject.email}}
O={{config.ipacertificatesubjectbase}}\nCN={{subject.username}}

Example syntax rule:

subjectAltName=@{% section %}{{datarules|join('\n')}}{% endsection %}

Example composed config template:

[ req ]
prompt = no
encrypt_key = no

distinguished_name = {% section %}O={{config.ipacertificatesubjectbase}}
CN={{subject.username}}{% endsection %}

req_extensions = exts

[ exts ]
subjectAltName=@{% section %}email={{subject.email}}{% endsection %}

Current status

The code currently under review implements most of the features described above, with some exceptions:

Macros, not tags

It was convenient to use jinja2 macros to implement some of the #Special features instead of adding new tags which requires extending the parser. So, instead of {% section %}{% endsection %}, one has to use {% call section() %}{% endcall %} to enclose the contents of an openssl config section. Similarly, the features that allow suppressing rules with no data are implemented using jinja2 macros. If the macros turn out to be problematic, we may want to move to using tags after all.

Rule suppression

It is important that the final output does not contain partially-constructed strings, reference empty sections, or provide command-line flags missing their arguments. So, we must be able to prevent rendering an entire rule, when some parts of it can not be rendered due to missing data. That is currently implemented using three macros: syntaxrule, datarule, and datafield.

  • syntaxrule wraps the contents of a syntax rule and renders it only when at least one of the included datarules will be rendered.
  • datarule wraps the contents of a data rule and renders it only when all of the included datafields contain data. It informs the enclosing syntaxrule whether the data rule will be rendered.
  • datafield wraps an individual data item. If the wrapped value is empty, the enclosing datarule will not be rendered.

The syntaxrule and datarule macros are applied automatically by the framework code, so users should only be concerned about datafield. All data items in the data rules must be marked as such using this macro, for example:

email:{{ipa.datafield(subject.mail.0)|quote}}

Alternatives considered

Template languages

Several possible tools were considered and tested for implementing these relationships before settling on jinja2:

  • For prototyping only: built-in python code. This could take advantage of the ability to query the API from within FreeIPA. However, it would be unsafe to allow administrators to add new mappings that run arbitrary code, so this does not satisfy the goal of giving administrators the ability to define their own mappings.
  • Jinja2. Nice because it doesn't reinvent the wheel by defining a new syntax, but would add a dependency to FreeIPA. Would also probably need to use its sandboxing features to prevent becoming a vector for arbitrary code execution.
  • Custom syntax, perhaps similar to NIS format specifiers. Would only need to define syntax for what we need, but a whole language would need to be defined and implemented.

Data interpolation

This blog post (archived here) analyzes some alternative ways of using jinja2 to format rules, besides the current option of substituting data rules into syntax rules.

Rule suppression

In retrospect, the definitions of the syntaxrule, datarule, and datafield macros are difficult to understand and modify, in part because it is difficult to avoid introducing unintended whitespace into the produced output. As a result of this, the code has a couple of minor bugs where sections are produced that should not be, which are difficult to fix because of the brittleness of the macros. We may want to consider an alternative implementation, such as:

  • It may be possible to write a more flexible implementation as a jinja2 tag rather than macros. Although not necessarily easier to understand, a parser extension might be able to handle the relationship between syntax rules, data rules, data fields, and openssl sections better because it has access to the internal AST of the template.
  • Instead of simply rendering things to see if they produce output, we could explicitly tag data rules with the data items they depend on. Then we could automatically insert {% if %} statements into the code to suppress things when those data items are unavailable. Of course, if the tagged data items were incorrect the suppression would not work correctly.
  • Finally, if we were interpolating user data during each render rather than just the final one (this solution for example) we could easily drop any rules that didn't produce output. However, then we would be at much higher risk of template injection attacks.