I-Regexp: An Interoperable Regexp Format

Status of This Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶

This Internet-Draft will expire on 17 May 2022.¶

Copyright Notice

Copyright (c) 2021 IETF Trust and the persons identified as the document authors. All rights reserved.¶

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶

Table of Contents

1. Introduction

Data modeling formats (YANG, CDDL) as well as query languages (jsonpath) often need a regular expression (regexp) sublanguage. There are many dialects of regular expressions in use in platforms, programming languages, and data modeling formats.¶

While regular expressions originally were intended to provide a Boolean matching function, they have turned into parsing functions for many applications, with capture groups, greedy/lazy/possessive variants, etc. Language features such as backreferences allow specifying languages that actually are context-free (Chomsky type 2) instead of the regular languages (Chomsky type 3) that regular expressions are named for.¶

YANG (Section 9.4.5 of [RFC7950]) and CDDL (Section 3.8.3 of [RFC8610]) have adopted the regexp language from W3C Schema [XSD2]. XSD regexp is a pure matching language, i.e., XSD regexps can be used to match a string against them and yield a simple true or false result. XSD regexps are not as widely implemented as programming language regexp dialects such as those of Perl, Python, Ruby, Go [RE2], or JavaScript (ECMAScript) [ECMA-262]. The latter are often in a state of continuous development; in the best case (ECMAScript) there is a complete specification which however is highly complex (Section 21.2 of [ECMA-262] comprises 62 pages) and evolves on a yearly timeline, with significant additions. Regexp dialects such as PCRE [PCRE2] have evolved to cover a common set of functions available in parsing regexp dialects, offered in a widely available library.¶

With continuing accretion of complex features, parsing regexp libraries have become susceptible to bugs and performance degradation, in particular those that can be exploited in DoS attacks. The library RE2 that is compatible with Go language regexps strives to be immune to DoS attacks, making it attractive to applications such as query languages where an attacker could control the input. The problem remains that other bugs in such libraries can lead to exploitable vulnerabilities; at the time of writing, the Common Vulnerabilities and Exposures (CVE) system has 131 entries that mention the word "regex" [REGEX-CVE] (not all, but many of which are such bugs, with 23 matches for arbitrary code execution).¶

Implementations of YANG and CDDL often struggle with providing true XSD regexps; some instead cheat by providing one of the parsing regexp varieties, sometime without even advertising this fact.¶

A matching regexp that does not use the more complex XSD features (Section 3) can usually be converted into a parsing regexp of many dialects by simply surrounding it with anchors of that dialect (e.g., ^ or \A and $ or \z). If the original matching regexps exceed the envelope of compatibility between dialects, this can lead to interoperability problems, or, worse, security vulnerabilities. Also, features of the target dialect such as capture groups may be triggered inadvertently, reducing performance.¶

The present specification defines an interoperable regexp flavor for matching, I-Regexp. This flavor is a subset of XSD regexps. It also comes with defined rules for converting the regexp into common parsing regexp dialects.¶

2. Requirements

I-Regexps should handle the vast majority of practical cases where a matching regexp is needed in a data model specification or a query language expression.¶

A brief survey of published RFCs yielded the regexp patterns in Appendix A (with no attempt at completeness). These should be covered by I-Regexps, both syntactically and with their intended semantics.¶

3. Subsetting XSD Regexps

XSD Regexps are relatively easy to implement or map to widely implemented parsing regexp dialects, with a small number of notable exceptions:¶

• Character class subtraction. This is a very useful feature in many specifications, but it is unfortunately mostly absent from parsing regexp dialects.¶

  • Issue: This absence can often be addressed by translating character class subtraction into positive character classes (possibly requiring significant expansion) and/or inserting negative lookahead assertions (which are not universally supported by regexp libraries, most notably not by RE2 [RE2]). This specification therefore opts for leaving out character class subtraction, but that decision is up for discussion.¶

• Unicode. While there is no doubt that a regexp flavor meant to last needs to be Unicode enabled, there are a number of aspects of this that need discussion. First of all, predefined character classes such as \w may be meant to be ASCII only, or they may encompass all letters and digits defined in Unicode. The latter is usually of interest in query languages, while the former is of interest to a subset of applications in data model specifications. Second, not all regexp implementations that one might want to map I-Regexps to will support accesses to Unicode tables that enable executing on constructs such as \p{IsCoptic}.¶

  • Issue: The ASCII focus can partially be addressed by adding a constraint that the matched text has to be ASCII in the first place. This often is all that is needed where regexps are used to define lexical elements of a computer language. The access to Unicode tables can simply be ruled out. (Note that RFC 6643 contains a lone instance of \p{IsBasicLatin}{0,255}, which is needed to describe a transition from a legacy character set to Unicode. The author believes that this would be a rare application and can be left out. RFC2622 contains [[:digit:]], [[:alpha:]], [[:alnum:]], albeit in a specification for the flex tool; this is intended to be close to \d, \p{L}, \w in an ASCII subset.)¶

4. Formal definition of I-Regexp

The syntax of I-Regexp is defined by the ABNF specification in Figure 1, with the following additional restriction:¶

• \w or \S MUST NOT occur in negative charClassExpr, i.e., in charClassExpr that include the optional "^" at the start.¶

  • Rationale: an exact implementation of XSD \w or \S in a negative charClassExpr essentially requires character class subtraction, which is not supported in I-Regexp (Section 3).¶
  • Issue: the ABNF grammar could express this restriction (by splitting charClassExpr, CCE1, charClassEsc, and MultiCharExp into positive and negative branches each), but would be harder to read.¶

This syntax is a subset of that of [XSD2]; the semantics of all the constructs allowed by this ABNF grammar are the same as those in [XSD2].¶

i-regexp = branch *( "|" branch )
branch = *piece
piece = atom [ quantifier ]
quantifier = ( %x2A-2B ; '*'-'+'
 / "?" ) / ( "{" quantity "}" )
quantity = QuantExact [ "," [ QuantExact ] ]
QuantExact = 1*%x30-39 ; '0'-'9'

atom = NormalChar / charClass / ( "(" i-regexp ")" )
NormalChar = ( %x00-27 / %x2C-2D ; ','-'-'
 / %x2F-3E ; '/'-'>'
 / %x40-5A ; '@'-'Z'
 / %x5E-7A ; '^'-'z'
 / %x7E-10FFFF )
charClass = "." / SingleCharEsc / charClassEsc / charClassExpr
SingleCharEsc = "\" ( %x28-2B ; '('-'+'
 / %x2D-2E ; '-'-'.'
 / "?" / %x5B-5E ; '['-'^'
 / %s"n" / %s"r" / %s"t" / %x7B-7D ; '{'-'}'
 )
charClassEsc = MultiCharEsc / catEsc / complEsc
MultiCharEsc = "\" ( %s"D" / %s"S" / %s"W" / %s"d" / %s"s" / %s"w" )
charClassExpr = "[" [ "^" ] ( "-" / CCE1 ) *CCE1 [ "-" ] "]"
CCE1 = ( CCchar [ "-" CCchar ] ) / charClassEsc
CCchar = ( %x00-2C / %x2E-5A ; '.'-'Z'
 / %x5E-10FFFF ) / SingleCharEsc
catEsc = %s"\p{" charProp "}"
complEsc = %s"\P{" charProp "}"
charProp = IsCategory / IsBlock
IsCategory = Letters / Marks / Numbers / Punctuation / Separators /
    Symbols / Others
Letters = %s"L" [ ( %x6C-6D ; 'l'-'m'
 / %s"o" / %x74-75 ; 't'-'u'
 ) ]
Marks = %s"M" [ ( %s"c" / %s"e" / %s"n" ) ]
Numbers = %s"N" [ ( %s"d" / %s"l" / %s"o" ) ]
Punctuation = %s"P" [ ( %x63-66 ; 'c'-'f'
 / %s"i" / %s"o" / %s"s" ) ]
Separators = %s"Z" [ ( %s"l" / %s"p" / %s"s" ) ]
Symbols = %s"S" [ ( %s"c" / %s"k" / %s"m" / %s"o" ) ]
Others = %s"C" [ ( %s"c" / %s"f" / %x6E-6F ; 'n'-'o'
 ) ]
IsBlock = %s"Is" 1*( "-" / %x30-39 ; '0'-'9'
 / %x41-5A ; 'A'-'Z'
 / %x61-7A ; 'a'-'z'
 )

About a third of the complexity of this ABNF grammar comes from going into details on the Unicode IsCategory classes. Additional complexity stems from the way hyphens can be used inside character classes to denote ranges; the grammar deliberately excludes questionable usage such as /[a-z-A-Z]/.¶

• Issue: This is essentially XSD regexp without character class subtraction. There is probably potential for simplification in IsBlock (leave out) and possibly in the rather large part for IsCategory as well. The ABNF has been automatically generated and maybe could use some polishing. The ABNF has been verified against Appendix A, but a wider corpus of regular expressions should be examined.¶

5. Mapping I-Regexp to Regexp Dialects

(TBD; these mappings need to be thoroughly verified.)¶

6. IANA Considerations

This document makes no requests of IANA.¶

7. Security considerations

TBD¶

(Discuss security issues of regexp implementations, both DoS and RCE; this is covered in part in Section 1.)¶

Appendix A. Regexps and Similar Constructs in Published RFCs

This appendix contains a number of regular expressions that have been extracted from published RFCs based on some ad-hoc matching. Multi-line constructions were not included. All regular expressions validate against the ABNF in Figure 1.¶

rfc6021.txt  459 (([0-1](\.[1-3]?[0-9]))|(2\.(0|([1-9]\d*))))
rfc6021.txt  513 \d*(\.\d*){1,127}
rfc6021.txt  529 \d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(\.\d+)?
rfc6021.txt  631 ([0-9a-fA-F]{2}(:[0-9a-fA-F]{2})*)?
rfc6021.txt  647 [0-9a-fA-F]{2}(:[0-9a-fA-F]{2}){5}
rfc6021.txt  933 ((:|[0-9a-fA-F]{0,4}):)([0-9a-fA-F]{0,4}:){0,5}
rfc6021.txt  938 (([^:]+:){6}(([^:]+:[^:]+)|(.*\..*)))|
rfc6021.txt 1026 ((:|[0-9a-fA-F]{0,4}):)([0-9a-fA-F]{0,4}:){0,5}
rfc6021.txt 1031 (([^:]+:){6}(([^:]+:[^:]+)|(.*\..*)))|
rfc6020.txt 6647 [0-9a-fA-F]*
rfc6095.txt 2544 \S(.*\S)?
rfc6110.txt 1583 [aeiouy]*
rfc6110.txt 3222 [A-Z][a-z]*
rfc6536.txt 1583 \*
rfc6536.txt 1632 [^\*].*
rfc6643.txt  524 \p{IsBasicLatin}{0,255}
rfc6728.txt 3480 \S+
rfc6728.txt 3500 \S(.*\S)?
rfc6991.txt  477 (([0-1](\.[1-3]?[0-9]))|(2\.(0|([1-9]\d*))))
rfc6991.txt  525 \d*(\.\d*){1,127}
rfc6991.txt  541 [a-zA-Z_][a-zA-Z0-9\-_.]*
rfc6991.txt  542 .|..|[^xX].*|.[^mM].*|..[^lL].*
rfc6991.txt  571 \d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(\.\d+)?
rfc6991.txt  665 ([0-9a-fA-F]{2}(:[0-9a-fA-F]{2})*)?
rfc6991.txt  693 [0-9a-fA-F]{2}(:[0-9a-fA-F]{2}){5}
rfc6991.txt  725 ([0-9a-fA-F]{2}(:[0-9a-fA-F]{2})*)?
rfc6991.txt  743 [0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-
rfc6991.txt 1041 ((:|[0-9a-fA-F]{0,4}):)([0-9a-fA-F]{0,4}:){0,5}
rfc6991.txt 1046 (([^:]+:){6}(([^:]+:[^:]+)|(.*\..*)))|
rfc6991.txt 1099 [0-9\.]*
rfc6991.txt 1109 [0-9a-fA-F:\.]*
rfc6991.txt 1164 ((:|[0-9a-fA-F]{0,4}):)([0-9a-fA-F]{0,4}:){0,5}
rfc6991.txt 1169 (([^:]+:){6}(([^:]+:[^:]+)|(.*\..*)))|
rfc7407.txt  933 ([0-9a-fA-F]){2}(:([0-9a-fA-F]){2}){0,254}
rfc7407.txt 1494 ([0-9a-fA-F]){2}(:([0-9a-fA-F]){2}){4,31}
rfc7758.txt  703 \d{2}:\d{2}:\d{2}(\.\d+)?
rfc7758.txt 1358 \d{2}:\d{2}:\d{2}(\.\d+)?
rfc7895.txt  349 \d{4}-\d{2}-\d{2}
rfc7950.txt 8323 [0-9a-fA-F]*
rfc7950.txt 8355 [a-zA-Z_][a-zA-Z0-9\-_.]*
rfc7950.txt 8356 [xX][mM][lL].*
rfc8040.txt 4713 \d{4}-\d{2}-\d{2}
rfc8049.txt 6704 [A-Z]{2}
rfc8194.txt  629 \*
rfc8194.txt  637 [0-9]{8}\.[0-9]{6}
rfc8194.txt  905 Z|[\+\-]\d{2}:\d{2}
rfc8194.txt  963 (2((2[4-9])|(3[0-9]))\.).*
rfc8194.txt  974 (([fF]{2}[0-9a-fA-F]{2}):).*
rfc8299.txt 7986 [A-Z]{2}
rfc8341.txt 1878 \*
rfc8341.txt 1927 [^\*].*
rfc8407.txt 1723 [0-9\.]*
rfc8407.txt 1749 [a-zA-Z_][a-zA-Z0-9\-_.]*
rfc8407.txt 1750 .|..|[^xX].*|.[^mM].*|..[^lL].*
rfc8525.txt  550 \d{4}-\d{2}-\d{2}
rfc8776.txt  838 /?([a-zA-Z0-9\-_.]+)(/[a-zA-Z0-9\-_.]+)*
rfc8776.txt  874 ([a-zA-Z0-9\-_.]+:)*
rfc8819.txt  311 [\S ]+
rfc8944.txt  596 [0-9a-fA-F]{2}(:[0-9a-fA-F]{2}){7}

Acknowledgements

This draft has been motivated by the discussion in the IETF JSONPATH WG about whether to include a regexp mechanism into the JSONPath query expression specification, as well as by previous discussions about the YANG pattern and CDDL .regexp features.¶

The basic approach for this draft was inspired by The I-JSON Message Format [RFC7493].¶

Author's Address