property - Properties, a selection mechanism for algorithm implementations
As of OpenSSL 3.0, a new method has been introduced to decide which of multiple
implementations of an algorithm will be used. The method is centered around
the concept of properties. Each implementation defines a number of properties
and when an algorithm is being selected, filters based on these properties can
be used to choose the most appropriate implementation of the algorithm.
Properties are like variables, they are referenced by name and
have a value assigned.
Property names fall into two categories: those reserved by the OpenSSL project
and user defined names. A reserved property name consists of a single
C-style identifier (except for leading underscores not being permitted), which
begins with a letter and can be followed by any number of letters, numbers and
underscores. Property names are case-insensitive, but OpenSSL will only use
lowercase letters.
A user defined property name is similar, but it must
consist of two or more C-style identifiers, separated by periods. The last
identifier in the name can be considered the 'true' property name, which is
prefixed by some sort of 'namespace'. Providers for example could include
their name in the prefix and use property names like
<provider_name>.<property_name>
<provider_name>.<algorithm_name>.<property_name>
A property is a name=value pair. A property definition is a
sequence of comma separated properties. There can be any number of properties
in a definition, however each name must be unique. For example: ""
defines an empty property definition (i.e., no restriction);
"my.foo=bar" defines a property named my.foo which has a
string value bar and "iteration.count=3" defines a property
named iteration.count which has a numeric value of 3. The full
syntax for property definitions appears below.
Each implementation of an algorithm can define any number of properties. For
example, the default provider defines the property provider=default for
all of its algorithms. Likewise, OpenSSL's FIPS provider defines
provider=fips and the legacy provider defines provider=legacy
for all of their algorithms.
A property query clause is a single conditional test. For example,
"fips=yes", "provider!=default" or
"?iteration.count=3". The first two represent mandatory clauses,
such clauses must match for any algorithm to even be under
consideration. The third clause represents an optional clause. Matching such
clauses is not a requirement, but any additional optional match counts in
favor of the algorithm. More details about that in the Lookups section.
A property query is a sequence of comma separated property query
clauses. It is an error if a property name appears in more than one query
clause. The full syntax for property queries appears below, but the available
syntactic features are:
- = is an infix operator providing an equality test.
- != is an infix operator providing an inequality test.
- ? is a prefix operator that means that the following clause is
optional but preferred.
- - is a prefix operator that means any global query clause involving
the following property name should be ignored.
- "..." is a quoted string. The quotes are not included in
the body of the string.
- '...' is a quoted string. The quotes are not included in the body
of the string.
When an algorithm is looked up, a property query is used to determine the best
matching algorithm. All mandatory query clauses must be present and the
implementation that additionally has the largest number of matching optional
query clauses will be used. If there is more than one such optimal candidate,
the result will be chosen from amongst those in an indeterminate way. Ordering
of optional clauses is not significant.
In order to permit a more concise expression of boolean properties, there is one
short cut: a property name alone (e.g. "my.property") is exactly
equivalent to "my.property=yes" in both definitions and queries.
Two levels of property query are supported. A context based property query that
applies to all fetch operations and a local property query. Where both the
context and local queries include a clause with the same name, the local
clause overrides the context clause.
It is possible for a local property query to remove a clause in
the context property query by preceding the property name with a '-'. For
example, a context property query that contains "fips=yes" would
normally result in implementations that have "fips=yes".
However, if the setting of the "fips" property is
irrelevant to the operations being performed, the local property query can
include the clause "-fips". Note that the local property query
could not use "fips=no" because that would disallow any
implementations with "fips=yes" rather than not caring about the
setting.
The lexical syntax in EBNF is given by:
Definition ::= PropertyName ( '=' Value )?
( ',' PropertyName ( '=' Value )? )*
Query ::= PropertyQuery ( ',' PropertyQuery )*
PropertyQuery ::= '-' PropertyName
| '?'? ( PropertyName (( '=' | '!=' ) Value)?)
Value ::= NumberLiteral | StringLiteral
StringLiteral ::= QuotedString | UnquotedString
QuotedString ::= '"' [^"]* '"' | "'" [^']* "'"
UnquotedString ::= [A-Za-z] [^{space},]+
NumberLiteral ::= '0' ( [0-7]* | 'x' [0-9A-Fa-f]+ ) | '-'? [1-9] [0-9]+
PropertyName ::= [A-Za-z] [A-Za-z0-9_]* ( '.' [A-Za-z] [A-Za-z0-9_]* )*
The flavour of EBNF being used is defined by:
<https://www.w3.org/TR/2010/REC-xquery-20101214/#EBNFNotation>.
Properties were added in OpenSSL 3.0
Copyright 2019-2023 The OpenSSL Project Authors. All Rights Reserved.
Licensed under the Apache License 2.0 (the "License").
You may not use this file except in compliance with the License. You can
obtain a copy in the file LICENSE in the source distribution or at
<https://www.openssl.org/source/license.html>.