We’re excited to introduce CRSLang, a new YAML-based rule language that will replace Seclang in the next major release of OWASP CRS. This represents a significant evolution in how we write, maintain, and deploy WAF rules.
Why CRSLang?
For nearly two decades, the OWASP CRS has relied on ModSecurity’s Seclang syntax. While Seclang has served us well, it comes with significant limitations that have become increasingly apparent as the project has grown:
The Problems with Seclang
- Technology Lock-in: Rules are tightly coupled to ModSecurity syntax, making it difficult to support other WAF engines
- Portability Issues: Direct deployment to alternative WAF platforms requires significant translation efforts
- Complexity: Rules are hard to read, write, and maintain, with all components mixed in long strings
- Learning Curve: The steep barrier to entry discourages new contributors
- Limited Expressiveness: Complex logical conditions (especially OR operations) require workarounds, and there’s no support for template functions
Consider this typical Seclang rule:
# Validate request line against the format specified in the HTTP RFC
SecRule REQUEST_LINE "!@rx (?i)^(?:get /[^#\?]..." \
"id:920100,\
phase:1,\
block,\
t:none,\
msg:'Method is not allowed by policy',\
logdata:'Matched Data: %{MATCHED_VAR} found within %{MATCHED_VAR_NAME}',\
tag:'application-multi',\
tag:'attack-protocol',\
ver:'OWASP_CRS/4.0.0',\
severity:'WARNING'"
Can you quickly understand what this rule does? The metadata, transformations, operators, and actions are all jumbled together in a single string, making it difficult to parse both for humans and machines.
Enter CRSLang
CRSLang is a YAML-based, technology-agnostic rule language designed to address these limitations while maintaining full backward compatibility with existing Seclang rules. It’s important to understand that CRSLang is a language for rule writers, not for WAF engines to parse directly. The accompanying tooling handles the translation to engine-specific formats, taking into account each engine’s quirks and capabilities.
Here’s the same rule in CRSLang format:
rule:
metadata:
comment: |
"Validate request line against the format specified in the HTTP RFC"
phase: 1
id: 920100
message: Invalid HTTP Request Line
severity: WARNING
tags:
- application-multi
- language-multi
version: OWASP_CRS/4.0.0
conditions:
- variables:
- REQUEST_LINE
operator:
negate: true
rx: (?i)^(?:get /[^#\?]...
transformations:
- none
actions:
disruptive: block
non-disruptive:
- logdata: "%{request_line}"
- setvar:
collection: TX
operation: =+
assignments:
- inbound_anomaly_score_pl1: "%{tx.warning_anomaly_score}"
The difference is immediately apparent. The rule structure is clear, with distinct sections for metadata, conditions, and actions.
Key Features
1. Clear Structure and Separation of Concerns
CRSLang separates metadata, conditions, and actions into distinct, easy-to-understand sections. This makes rules easier to read, validate, and maintain.
2. Improved Logical Expressions
Seclang’s support for complex logical conditions is limited and often requires workarounds. CRSLang provides a clean syntax for logical conjunctions (logical “and”), and we are working on extending its logical capabilities. In the following example, the rule has three logical conjunctions, which would be expressed as chained rules in Seclang:
REQUEST_HEADERS:Acceptmust be emptyREQUEST_METHODmust not beOPTIONSREQUEST_HEADERS:User-Agentmust not include any of the strings “AppleWebKit”, “Android”, “Business”, “Enterprise”, or “Entreprise”
rule:
metadata:
phase: 1
id: 920310
message: Request Has an Empty Accept Header
conditions:
- collections:
- name: REQUEST_HEADERS
arguments:
- Accept
operator:
rx: ^$
transformations:
- none
- variables:
- REQUEST_METHOD
operator:
negate: true
rx: ^OPTIONS$
- collections:
- name: REQUEST_HEADERS
arguments:
- User-Agent
operator:
negate: true
pm: AppleWebKit Android Business Enterprise Entreprise
transformations:
- none
actions:
disruptive: block
3. Common Definitions
CRSLang supports reusable rule components through common definitions, making it easier to maintain consistent patterns across your rule set. You can define patterns at a global or scoped level and reference them from multiple rules, ensuring consistency without duplication.
4. Bidirectional Translation
One of CRSLang’s most powerful features is its bidirectional translation capability. We’ve built a robust parser that can:
- Convert existing Seclang rules to CRSLang format
- Generate valid Seclang from CRSLang rules
- Preserve all semantic information during translation
- Maintain comment and documentation context
This means you can gradually migrate to CRSLang while maintaining compatibility with existing tools and deployments. The translation engine ensures that no information is lost in the conversion process, and round-trip testing validates correctness.
The Tooling Vision
CRSLang is designed as the single source of truth for rule writers. The real power lies in the tooling that surrounds it:
- Engine-specific transformation: Rules are automatically transformed into formats for different WAF engines, with transformers that account for each engine’s specific quirks and limitations
- Filtered output: Engine-specific output can be filtered by tag, platform, or other criteria to produce minimal, application-specific rule sets
- Augmented output: Engine-specific output can be enriched with additional information such as tags, comments, and engine-specific metadata
This approach means rule writers only need to learn one language, while the tooling handles the complexity of supporting multiple WAF platforms.
Building on Previous Work
CRSLang builds upon the foundation laid by earlier parser projects. If you’re familiar with our msc_pyparser tool, you’ll recognize the concept of converting ModSecurity rules into structured formats. CRSLang takes this concept further by providing a complete, production-ready language specification with full tooling support.
While msc_pyparser focused on rule manipulation through Python, CRSLang provides a comprehensive language specification with:
- ANTLR-based parsing with support for multiple target languages (Go, Python, Java), based on seclang_parser
- Full support for Seclang v2 and v3
- Validation and testing frameworks
- IDE integration capabilities (syntax highlighting, validation, debugging)
The project is fully open source and available at github.com/coreruleset/crslang.
Looking Forward
CRSLang represents more than just a syntax change — it’s a foundation for the future of OWASP CRS. With this new language, we’re opening doors to:
- Enhanced portability: Support for multiple WAF technologies beyond ModSecurity
- Alternative logical engines: Integration with systems like Google CEL or Wirefilter
- Better tooling: IDE support, linting, automated testing, and debugging capabilities
- Reduced learning curve: Making it easier for new contributors to join the project
- Improved maintainability: Clearer rule structure means fewer bugs and easier updates
Timeline and Migration
CRSLang will replace Seclang as the primary rule language in the next major release of OWASP CRS. We’re committed to making this transition as smooth as possible:
- Current state: CRSLang parser and tooling are available now
- Migration period: Full backward compatibility maintained through translation engine
- Next major release: CRSLang becomes the default format
- Long-term support: Seclang translation capabilities maintained for existing deployments
You can start exploring CRSLang today by:
- Visiting the CRSLang GitHub repository
- Converting your existing rules to see the difference
- Providing feedback to help shape the final specification
- Contributing to the parser and tooling development
Get Involved
We’re excited about this evolution and would love your feedback. Whether you’re a long-time CRS contributor or new to the project, your input is valuable as we shape the future of WAF rule development.
Join the conversation:
- GitHub: github.com/coreruleset/crslang
- OWASP Slack: owasp.org/slack/ (#coreruleset channel)
The future of OWASP CRS is clearer, more maintainable, and more accessible. We can’t wait to see what the community builds with CRSLang.
