Operators and functions to use in Cedar

This topic describes the built-in operators and functions that you can use to build your expressions using the Cedar policy language.

Not all expressions that you can evaluate will necessarily validate when using Cedar’s policy validator. This situation is similar to that of most programming languages. For example, in Java the following code does not type check, even though executing it will never result in an error.

if (false) { return 1 == "hello"; } else { return true; }

A key difference between Java and Cedar is that Java typechecking is mandatory – you cannot run a Java program that does not typecheck – whereas for Cedar policy validation is optional – it is still possible to evaluate policies that do not validate. This allows you to get up and running with Cedar faster, and to write more expressive policies, if need be. Of course, the restrictions imposed by validation come with the benefit that valid policies are sure not to exhibit most kinds of evaluation error. See the policy validation section for more information.

When giving examples below, we will indicate whether the example evaluates properly (and to what), and also whether it validates. All expressions that fail to evaluate will also fail to validate, but not vice versa.

Topics on this page

Overview of operators

The operators use the following syntax structures:

  • Unary operators – A unary operator takes one operand. Place the operand after the operator.

    <operator> operand

// Uses the logical NOT operator and evaluates to the
// inverse of the value of the boolean operand
! a

  • Binary operators – A binary operator takes two operands. Place one operand before the operator and one after. Some binary operators are commutative. See the description of each operator to understand where operand order matters.

    firstOperand <operator> secondOperand

// Evaluates to true if both operands have the same type and value
a == b

// Evaluates to true if the first operand is within the
// hierarchy of the second operand
c in d

Functions use one of two styles of syntax:

  • Constructor-style: Place any operands in parentheses after the function name, separating them with commas.

    function(firstOperand, secondOperand, …)

// creates an IP address
ip("127.0.0.1")

  • Method-style: Append the function name to the end of the target parameter, separating them with a . (period) character. Place any operands in parentheses after the function name, separating them with commas.

    firstOperand.function(secondOperand, thirdOperand, …)

// Evaluates to true if the any of the set member
// elements b, c, or d is an element of set a
a.containsAny([b, c, d])

String operators and functions

Use these operators and functions to compare strings or convert them to other types.

like (string matching with wildcard)

Usage: <string> like <string possibly with wildcards>

Binary operator that evaluates to true if the string in the left operand matches the pattern string in the right operand. The pattern string can include one or more asterisks (*) as wildcard characters that match 0 or more of any character.

To match a literal asterisk character, use the escaped \* sequence in the pattern string.

Consider a query with the following context:

"context": {
    "location": "s3://bucketA/redTeam/some/thing"
}

In that scenario, the following expression returns true.

context.location like "s3:*"         //true

More Examples: 

"eggs" like "ham*"                                             //false
"eggs" like "*ham"                                             //false
"eggs" like "*ham*"                                            //false
"ham and eggs" like "ham*"                                     //true
"ham and eggs" like "*ham"                                     //false
"ham and eggs" like "*ham*"                                    //true
"ham and eggs" like "*h*a*m*"                                  //true
"eggs and ham" like "ham*"                                     //false
"eggs and ham" like "*ham"                                     //true
"eggs, ham, and spinach" like "ham*"                           //false
"eggs, ham, and spinach" like "*ham"                           //false
"eggs, ham, and spinach" like "*ham*"                          //true
"Gotham" like "ham*"                                           //false
"Gotham" like "*ham"                                           //true
"ham" like "ham"                                               //true
"ham" like "ham*"                                              //true
"ham" like "*ham"                                              //true
"ham" like "*h*a*m*"                                           //true
"ham and ham" like "ham*"                                      //true
"ham and ham" like "*ham"                                      //true
"ham" like "*ham and eggs*"                                    //false
"\\afterslash" like "\\*"                                      //true
"string\\with\\backslashes" like "string\\with\\backslashes"   //true
"string\\with\\backslashes" like "string*with*backslashes"     //true
"string*with*stars" like "string\*with\*stars"                 //true

decimal() (parse string and convert to decimal)

Usage: decimal(<string>)

Function that parses the string and tries to convert it to type decimal. If the string doesn’t represent a valid decimal value, it generates an error.

To be interpreted successfully as a decimal value, the string must contain a decimal separator (.) and at least one digit before and at least one digit after the separator. There can be no more than 4 digits after the separator. The value must be within the valid range of the decimal type, from -922337203685477.5808 to 922337203685477.5807.

Cedar can properly evaluate decimal(e) where e is any Cedar expression that evaluates to a valid string. For example, the expression decimal(if true then "1.1" else "2.1") will evaluate to the decimal number 1.1. However, Cedar’s policy validator only permits e to be a string literal that will not result in an error or overflow.

Examples:

In the examples below, suppose context.time is "12.25" while context.date is "12/27/91". Examples labeled error indicate both a validation and evaluation error. Unlabeled examples evaluate and validate correctly.

decimal("1.0")
decimal("-1.0")
decimal("123.456")
decimal("0.1234")
decimal("-0.0123")
decimal("55.1")
decimal("00.000")
decimal(context.time)            //Evaluates //Doesn't validate (parameter not a string literal)
decimal(context.date)            //error - invalid format (not valid as parameter not a string literal)
decimal("1234")                  //error - missing decimal
decimal("1.0.")                  //error - stray period at end
decimal("1.")                    //error - missing fractional part
decimal(".1")                    //error - missing whole number part
decimal("1.a")                   //error - invalid fractional part
decimal("-.")                    //error - invalid format
decimal("1000000000000000.0")    //error - overflow
decimal("922337203685477.5808")  //error - overflow
decimal("0.12345")               //error - too many fractional digits

ip() (parse string and convert to ipaddr)

Usage: ip(<string>)

Function that parses the string and attempts to convert it to type ipaddr. If the string doesn’t represent a valid IP address or range, then the ip() expression generates an error when evaluated.

Cedar can properly evaluate ip(e) where e is any Cedar expression that evaluates to a valid string. For example, the expression ip(if true then "1.1.1.1/24" else "2.1.1.1/32") will evaluate to the IP address 1.1.1.1/24. However, Cedar’s policy validator only permits e to be a string literal.

Examples:

In the examples below, suppose context.addr is "12.25.27.15" while context.date is "12/27/91". Examples labeled error indicate both a validation and evaluation error. Unlabeled examples evaluate and validate correctly.

ip("127.0.0.1")
ip("::1")
ip("127.0.0.1/24")
ip("ffee::/64")
ip("ff00::2")
ip("::2")
ip(context.addr)                    //Evaluates //Doesn't validate (parameter not a string literal)
ip(context.time)                    //error - invalid format (not valid as parameter not a string literal)
ip("380.0.0.1")                     //error – invalid IPv4 address
ip("ab.ab.ab.ab")                   //error – invalid IPv4 address
ip("127.0.0.1/8/24")                //error – invalid CIDR notation
ip("fee::/64::1")                   //error – invalid IPv6 address
ip("fzz::1")                        //error – invalid character in address
ip([127,0,0,1])                     //error – invalid operand type
"127.0.0.1".ip()                    //error – invalid call style

The following examples all evaluate correctly, but the last two do not validate as the validator requires == expressions to be applied to expressions of the same type (see the discussion of == below).

ip("127.0.0.1") == ip("127.0.0.1")            //true
ip("192.168.0.1") == ip("8.8.8.8")            //false
ip("192.168.0.1/24") == ip("8.8.8.8/8")       //false
ip("192.168.0.1/24") == ip("192.168.0.8/24")  //false - different host address
ip("127.0.0.1") == ip("::1")                  //false – different IP versions
ip("127.0.0.1") == ip("192.168.0.1/24")       //false - address compared to range
ip("127.0.0.1") == "127.0.0.1"                //false – different types //Doesn't validate
ip("::1") == 1                                //false – different types //Doesn't validate

Comparison operators and functions

Use these operators to compare two values. An expression that uses one of these operators evaluates to a boolean true or false. You can then combine multiple expressions using the logical operators.

== (equality)

Usage: <any type> == <any type>

Binary operator that compares two operands of any type and evaluates to true only if they are exactly the same type and the same value.

While Cedar can evaluate expressions e1 == e2 when e1 and e2 have different types (usually giving the result false), such comparison expressions are not accepted by the policy validator. In particular, policies containing equality expressions e1 == e2 are only validated when

  1. Both e1 and e2 have the same type, or
  2. Both have entity type, though that type need not be the same

Examples:

All of the examples below evaluate successfully, and are labeled with their evaluation result. Those examples that do not validate (the last two) are labeled as such.

1 == 1                          //true
"something" == "something"      //true
"Something" == "something"      //false
[1, -33, 707] == [1, -33]       //false
[1, 2, 40] == [1, 2, 40]        //true
[1, 2, 40] == [1, 40, 2]        //true
[1, -2, 40] == [1, 40]          //false
[1, 1, 1, 2, 40] == [40, 1, 2]  //true
[1, 1, 2, 1, 40, 2, 1, 2, 40, 1] == [1, 40, 1, 2]   //true
true == true                    //true
context.device_properties == {"os": "Windows", "version": 11}
                                //true if context.device_properties represents a Windows 11 computer
User::"alice" == User::"alice"  //true
User::"alice" == User::"bob"    //false -- two different entities of same type
User::"alice" == Admin::"alice" //false -- entities of two different types //Validates
5 == "5"                        //false -- operands have two different types //Doesn't validate
"alice" == User::"alice"        //false -- operands have two different types //Doesn't validate

!= (inequality)

Usage: <any type> != <any type>

Binary operator that compares two operands of any type and evaluates to true if the operands have different values or are of different types. You can use != only in when and unless clauses. As with the == operator, the validator only accepts policies that use != on two expressions of (possibly differing) entity type, or the same non-entity type.

Example: 

forbid (principal, action, resource)
when{
    resource.tag != "public"
};

< (long integer ‘less than’)

Usage: <long> < <long>

Binary operator that compares two long integer operands and evaluates to true if the left operand is numerically less than the right operand. If either operand is not a long then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

3 < 303               //true
principal.age < 22    //true (assuming principal.age is 21)
3 < "3"               //error - operator not allowed on non-long
false < true          //error - operator not allowed on non-long
"" < "zzz"            //error - operator not allowed on non-long
[1, 2] < [47, 0]      //error - operator not allowed on non-long

.lessThan() (decimal ‘less than’)

Usage: <decimal>.lessThan(<decimal>)

Function that compares two decimal operands and evaluates to true if the left operand is numerically less than the right operand. If either operand is not a decimal then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

decimal("1.23").lessThan(decimal("1.24"))     //true
decimal("1.23").lessThan(decimal("1.23"))     //false
decimal("123.45").lessThan(decimal("1.23"))   //false
decimal("-1.23").lessThan(decimal("1.23"))    //true
decimal("-1.23").lessThan(decimal("-1.24"))   //false
decimal("1.1").lessThan(2)                    //error -- not a decimal operand
ip("1.1.2.3").lessThan(decimal("1.2"))        //error -- not a decimal operand

<= (long integer ‘less than or equal’)

Usage: <long> <= <long>

Binary operator that compares two long integer operands and evaluates to true if the left operand is numerically less than or equal to the right operand. If either operand is not a long then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

3 <= 303               //true
principal.age <= 21    //true (assuming principal.age is 21)
3 <= "3"               //error - operator not allowed on non-long
false <= true          //error - operator not allowed on non-long
"" <= "zzz"            //error - operator not allowed on non-long
[1, 2] <= [47, 0]      //error - operator not allowed on non-long

.lessThanOrEqual() (decimal ‘less than or equal’)

Usage: <decimal>.lessThanOrEqual(<decimal>)

Function that compares two decimal operands and evaluates to true if the left operand is numerically less than or equal to the right operand. If either operand is not a decimal then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

decimal("1.23").lessThanOrEqual(decimal("1.24"))    //true
decimal("1.23").lessThanOrEqual(decimal("1.23"))    //true
decimal("123.45").lessThanOrEqual(decimal("1.23"))  //false
decimal("-1.23").lessThanOrEqual(decimal("1.23"))   //true
decimal("-1.23").lessThanOrEqual(decimal("-1.24"))  //false
decimal("1.1").lessThanOrEqual(2)                   //error -- not a decimal operand
ip("1.1.2.3").lessThanOrEqual(decimal("1.2"))       //error -- not a decimal operand

> (long integer ‘greater than’)

Usage: <long> > <long>

Binary operator that compares two long integer operands and evaluates to true if the left operand is numerically greater than the right operand. If either operand is not a long then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

3 > 303                //false
principal.age > 22     //false (assuming principal.age is 21)
3 > "3"                //error - operand is a non-long
false > true           //error - operands are not long integers
"some" > "thing"       //error - operands are not long integers

.greaterThan() (decimal ‘greater than’)

Usage: <decimal>.greaterThan(<decimal>)

Function that compares two decimal operands and evaluates to true if the left operand is numerically greater than the right operand. If either operand is not a decimal then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

decimal("1.23").greaterThan(decimal("1.24"))    //false
decimal("1.23").greaterThan(decimal("1.23"))    //false
decimal("123.45").greaterThan(decimal("1.23"))  //true
decimal("-1.23").greaterThan(decimal("1.23"))   //false
decimal("-1.23").greaterThan(decimal("-1.24"))  //true
decimal("1.1").greaterThan(2)                   //error -- not a decimal operand
ip("1.1.2.3").greaterThan(decimal("1.2"))       //error -- not a decimal operand

The greaterThan function must take two decimal operands or else it will produce an error when evaluated, per the last two examples. The policy validator also rejects also any expression that attempts to call greaterThan on non-decimal values.

>= (long integer ‘greater than or equal’)

Usage: <long> >= <long>

Binary operator that compares two long integer operands and evaluates to true if the left operand is numerically greater than or equal to the right operand. If either operand is not a long then evaluation (and validation) results in an error.

Examples:

In the following examples, //error indicates both an evaluation and a validation error.

3 >= 303               //false
principal.age >= 21    //true (assuming principal.age is 21)
3 >= "3"               //error - operand is a non-long
false >= true          //error - operands are not long integers
"some" >= "thing"      //error - operands are not long integers

As shown in the examples, evaluating an expression with >= when the operators are not both long numbers results in an error. The policy validator also rejects also any expression that attempts to compare two values with >= that do not have type long.

.greaterThanOrEqual() (decimal ‘greater than or equal’)

Usage: <decimal>.greaterThanOrEqual(<decimal>)

Function that compares two decimal operands and evaluates to true if the left operand is numerically greater than or equal to the right operand. If either operand is not a decimal then evaluation (and validation) results in an error.

Examples: 

decimal("1.23").greaterThanOrEqual(decimal("1.24"))    //false
decimal("1.23").greaterThanOrEqual(decimal("1.23"))    //true
decimal("123.45").greaterThanOrEqual(decimal("1.23"))  //true
decimal("-1.23").greaterThanOrEqual(decimal("1.23"))   //false
decimal("-1.23").greaterThanOrEqual(decimal("-1.24"))  //true

Logical operators

Use these operators on boolean values or expressions.

&& (AND)

Usage: <boolean> && <boolean>

Binary operator that evaluates to false if the first evaluates to false, or if the first evaluates to true and the second evaluates to false. Evaluates to true only if both arguments evaluate to true.

In the following policy, the when condition is true if both principal.numberOfLaptops < 5 and principal.jobLevel > 6 evaluate to true.

permit (principal, action == Action::"remoteAccess", resource)
when {
    principal.numberOfLaptops < 5 &&
    principal.jobLevel > 6
};

The && operator uses short circuit evaluation. If the first argument is false, then the expression immediately evaluates to false and the second argument isn’t evaluated. This approach is useful when the second argument might result in an error if evaluated. You can use the first argument to test that the second argument is a valid expression.

The following policy is satisfied only if the principal has the attribute level and the level > 5.

permit (principal, action == Action:"read", resource)
when {
    principal has level &&
    principal.level > 5
};

The > comparison in this expression can only succeed if the principal entity has a level attribute. If it doesn’t, then principal.level sub-expression would evaluate to an error. The expression that is the first operand of && uses the has operator to ensure that the principal entity does have such an attribute. If that evaluates to false, then the second operand to && isn’t evaluated.

The description of && so far has been from the perspective of evaluation. From the perspective of policy validation, the situation is a little different. In general, the validator will reject any expression e1 && e2 that would evaluate to an error due to either e1 or e2 not having boolean type. However, the validator sometimes is able to take short-circuiting into account. We will elaborate when considering the examples below.

More Examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate. Others evaluate correctly, but some may fail to validate, per the label. Discussion of the reasons for non-validation is given below.

3 && false                                  //error -- first operand is not a boolean
false && 3                                  //Evaluates to false (due to short circuiting) //Validates
(3 == 4) && 3                               //Evaluates to false (due to short circuiting) //Doesn't validate
(User::"alice" == Action::"viewPhoto") && 3 //Evaluates to false //Validates
true && 3                                   //error -- second operand is not a boolean
(false && 3) == 3                           //Evaluates to false //Doesn't validate (== applied to different types)

As mentioned above, validation sometimes is able to account for short-circuiting behavior, but not always. In particular, the validator will accept false && 3 and (User::"alice" == Action::"viewPhoto") && 3, but not (3 == 4) && 3. The reason is that it knows false is always, well, false, so it can model short-circuiting.

|| (OR)

Usage: <boolean> || <boolean>

Binary operator that evaluates to true if the first operand evaluates to true, or the first evaluates to false and the second operand evaluates to true. Evaluates to false if both operands evaluate to false.

This operator uses short circuit evaluation. If the first argument is true, then the expression immediately evaluates to true and the second argument isn’t evaluated. This approach is useful when the second argument might result in an error if evaluated. The first argument should be a test that can determine if the second argument is safe to evaluate. For example, consider the following expression. It evaluates to true if the principal is either missing the age attribute or that attribute is at least 21.

!(principal has age) || principal.age < 21

The second comparison in this expression will evaluate to a boolean only if the age attribute for the principal entity is present. If it is missing, then principal.age will evaluate to an error. The first expression uses the has operator, inverted by the ! NOT operator, to flag that the principal entity is missing the age property. If that evaluates to true, there is no test of the second expression.

The following policy allows if either resource.owner == principal or resource.tag == "public" is true.

permit (principal, action == Action:"read", resource)
when {
    resource.owner == principal ||
    resource.tag == "public"
};

The description so far is from the perspective of evaluation. From the perspective of policy validation, the situation is a little different. In general, the validator will reject any expression e1 || e2 that would evaluate to an error due to either e1 or e2 not having boolean type. The validator sometimes is able to take short-circuiting into account, as discussed using the examples below.

More Examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate. Others evaluate correctly, but some may fail to validate, per the label. Discussion of the reasons for non-validation is given below.

3 || true                  //error (first operand not a boolean)
true || 3                  //Evaluates to true (due to short-circuiting) //Validates
false || 3                 //error (second operand not a boolean)
(3 == 3) || 3              //Evaluates to true (due to short-circuiting) //Doesn't validate

As mentioned above, validation sometimes is able to account for short-circuiting behavior, but not always. In particular, the validator will accept true || 3 but not (3 == 3) && 3.

! (NOT)

Usage: ! <boolean>

Unary operator with only one argument. It inverts the value of the boolean operand from true to false, or from false to true. If either operand is not a boolean then evaluation (and validation) results in an error.

Example:

The following policy forbids if the principal does not belong to Group::”family”.

forbid (principal, action, resource)
when {
    !(principal in Group::"family")
};

You can rewrite the above policy using an unless clause as:

forbid (principal, action, resource)
unless {
  principal in Group::"family"
};

More Examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate.

! true                                //false
! false                               //true
! 8                                   //error
if !true then "hello" else "goodbye"  //"goodbye"

if (CONDITIONAL)

Usage: if <boolean> then <T> else <U>

The if operator returns its evaluated second argument if the first argument evaluates to true, else it returns the evaluated third argument.

The if operator requires its first argument to be a boolean, i.e., to evaluate to either true or false. If it does not, the if evaluates to an error. The second and third arguments can have any type; to be compatible with validation, both arguments usually must have the same type, but sometimes the validator is able to take if’s short-circuiting behavior into account; more details below.

In the following policy, the when condition is true if both principal.numberOfLaptops < 5 and principal.jobLevel > 6 are true.

permit (principal, action == Action::"remoteAccess", resource)
when {
    if principal.numberOfLaptops < 5 then
      principal.jobLevel > 6
    else false
};

The if operator uses short circuit evaluation. When the first argument evaluates to true the third argument is never evaluated. When the first argument evaluates to false, the second argument is never evaluated.

The if operator is a strict generalization of the && and || operators. The expression _e1_ || _e2_ is equivalent to the expression if _e1_ then true else (if _e2_ then true else false). The expression _e1_ && _e2_ is equivalent to the expression if _e1_ then (if _e2_ then true else false) else false. Note that _e1_ || _e2_ is not equivalent to if _e1_ then true else _e2_, due to the possibility of type errors. To see why, consider that false || "foo" produces a type error, while if false then true else "foo" evaluates to "foo".

Note that if and when, though similar in normal English, play different roles in Cedar. The keyword when is part of the policy syntax which simply connects the policy scope to the policy’s condition(s). The keyword if is a part of an expression that can be contained in such a condition, and can be evaluated against a relevant authorization request.

More Examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate. Others evaluate correctly, but some may fail to validate, per the label. Discussion of the reasons for non-validation is given below.

if 1 == 1 then "ok" else "wrong"         //Evaluates to "ok" //Validates
if 1 == 2 then User::"foo" else "ok"     //Evaluates to "ok" //Doesn't validate
if 1 then "wrong" else "wrong"           //error
if false then (1 && "hello") else "ok"   //Evaluates to "ok" (due to short circuiting) //Validates
if true then (1 && "hello") else "ok"    //error

The example if 1 == 1 then "ok" else "wrong" validates because the first operand 1 == 1 has boolean type, and both the second and third operands have the same type (String). The example if 1 == 2 then User::"foo" else "ok" doesn’t validate because the second and third operands do not have the same type. The example if 1 then "wrong" else "wrong" doesn’t validate because the first operand 1 does not have type boolean. The example if false then (1 && "hello") else "ok" is accepted (validates) because the validator is able to consider short-circuiting.

Arithmetic operators

Use these operators to perform arithmetic operations on long integer values.

Notes The arithmetic operators support only values of type long. They don’t support values of type Decimal. There is no operator for arithmetic division.

If you exceed the range available for the long data type by using any of the arithmetic operators, it results in an overflow error. In general, a policy that results in an error is ignored, meaning that a permit policy might unexpectedly fail to allow access, or a forbid policy might unexpectedly fail to block access.

+ (numeric addition)

Usage: <long> + <long>

Binary operator that adds the two long integer values and returns a long integer sum; it evaluates (and validates) to an error if given non-long operands. Addition could result in an overflow evaluation error; such errors are not detected by the validator.

Example:

The following policy returns allow if the context budget plus the context downloaded is greater than 100.

permit (principal, action, resource)
when {
    context.budget + context.downloaded > 100
};

Other examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate, except in the case of overflow, in which case evaluation produces an error but validation does not.

11 + 0                              //11
-1 + 1                              //0
9223372036854775807 + 1             //error - overflow //Validates
7 + "3"                             //error - second operand not a long //Doesn't validate
"lamp" + "la"                       //error - operands not `long` //Doesn't validate

- (numeric subtraction or negation)

Usage: <long> - <long> or - <long>

As a binary operator with two operands, it subtracts the second long integer value from the first and returns a long integer difference. It evaluates (and validates) to an error if given non-long operands. Subtraction could result in an overflow evaluation error (underflow, more precisely); such errors are not detected by the validator.

Examples:

In the following examples, those labeled with //error both fail to evaluate and fail to validate, except in the case of overflow, in which case evaluation produces an error but validation does not.

-3                              //-3
44 - 31                         //13
5 - (-3)                        //8
-9223372036854775807 - 2 + 3    //error - overflow //Validates
7 - "3"                         //error - second operand not a `long` //Doesn't validate

Because the - symbol can mean both unary and binary subtraction, the example -9223372036854775807 - 2 + 3 must use parentheses to disambiguate.

* (numeric multiplication)

Usage: <long> * <long>

Binary operator that multiplies two long integer operands and returns a long integer product. It evaluates (and validates) to an error if given non-long operands. Multiplication could result in an overflow evaluation error; such errors are not detected by the validator.

There is no operator for arithmetic division.

Examples:

In these examples, suppose that resource.value is 3 and context.budget is 4. Examples labeled with //error both fail to evaluate and fail to validate, except in the case of overflow, in which case evaluation produces an error but validation does not.

10 * 20                          //200
resource.value * 10              //30
2 * context.budget > 100         //false
context.budget * resource.value  //depends on entity data
9223372036854775807 * 2          //error - overflow //Validates
5 * (-3)                         //-15
5 * 0                            //0
"5" * 0                          //error - both operands must have type `long` //Doesn't validate

Hierarchy and set membership operators and functions

Use these functions to test if entities are members of a hierarchy or a set.

in (hierarchy membership)

Usage: <entity> in <entity>

Binary operator that evaluates to true if the entity in the left operand is a descendant in the hierarchy under the entity in the right operand. Evaluation (and validation) produces an error if the first (lhs) operand of in is not an entity, or the (rhs) is not an entity or a set thereof (the latter usecase is discussed below).

The in operator is transitive. If A is in B, and B is in C, then A is also in C. This approach allows you to model the concept of a multi-tier hierarchy, for example nesting folders in other folders.

The in operator is reflexive. The expression evaluates to true if the right entity is the same as the left entity. In other words, an entity is always in its own hierarchy. A is always in A.

Examples:

In these examples, assume that the principal in a request is User::"bob", and that User::"bob" has Group::"janefriends" as a parent in the hierarchy, which in turn has Group::"all" as a parent. Examples labeled with //error both fail to evaluate and fail to validate.

principal in User::"bob"             //true by reflexivity
principal in Group::"janefriends"    //true
Group::"janefriends" in Group::"all" //true
principal in Group::"all"            //true by transitivity
Group::"all" in User::"bob"          //false -- in is not symmetric
1 in Group::"janefriends"            //error -- LHS not an entity

Usage: <entity> in set(<entity>)

When the right operand is a set of entities, then the expression evaluates to true if the left operand is in any of the entities in the set. If the left operand is not an entity or the right operand is not an entity or set of entities, then evaluation produces an error. Likewise, the validator requires the lhs to be an entity and the rhs to be an entity or set thereof, where in the latter case all the entities must have the same type.

As an example, consider the following expression.

User::"bob" in [Group::"janefriends", Group::"joefriends"]

This expression is similar to

User::"bob" in Group::"janefriends" || User::"bob" in Group::"joefriends"

In other words, if User::"bob" is in either Group, then the expression evaluates to true. However, in does not short-circuit evaluation – it will test membership in all elements. This is important for error reporting. In particular, the following expression

User::"bob" in Group::"janefriends" || User::"bob" in 1

would evaluate to true because the second expression given to || is short-circuited, whereas

User::"bob" in [Group::"janefriends", 1]

evaluates to an error (since 1 is not an entity).

The right operand of in can be any expression that evaluates to a set of entity references, not just a set literal. For example, suppose the query context record contains the following:

{
    "groups ": [Group::"jane_family", Group::"jane_friends "]
}

Then the following two expressions in a policy statement are equivalent:

User::"alice" in context.groups
User::"alice" in [Group::"jane_family", Group::"jane_friends"]

Because the in operator is reflexive, A in A returns true even if the entity A does not exist in the entities passed in with the request. The evaluator treats entity references that are not in the hierarchy as a valid entity. For example:

Stranger::"jimmy" in Stranger::"jimmy"        //true by reflexivity.
Stranger::"jimmy" in Group::"jane_friends"    //false - Stranger::"jimmy" does not refer to an existing entity
Stranger::"jimmy" in [
    Group::"jane_family",
    Stranger::"jimmy"
]                                             //true - Stranger::"jimmy" in Stranger::"jimmy" is true

More Examples:

These examples both fail to evaluate and fail to validate because their operands are invalid.

"some" in ["some", "thing"] //error - these are strings, not entities. Use `contains` for set membership.
"os" in {"os":"Windows "}   //error - use `has` operator to check if a key exists

has (presence of attribute test)

Usage: <entity or record> has <attribute>

boolean operator that evaluates to true if the left operand has a value defined for the specified attribute. Evaluation (and validation) produces an error if lhs is not a record or entity type. We discuss evaluation, first, and validation in more depth further down.

Evaluation

Use this operator to check that a value is present before accessing that value. If an expression accesses an attribute that isn’t present, then evaluation produces an error.

The following example expression first tests whether the entity principal has a defined attribute manager. Because the && operator uses shortcut logic, the second expression to && is evaluated and the attribute accessed only if the has check has deemed it is present.

principal has manager && principal.manager == User::"kirk"

In the following example, assume that the request has the following context:

"context":{
    "role": ["admin", "user"],
    "addr": { "street": "main", "city": "DC"}
    "owner info": { "name": "Alice", "age": 18 }
}

The following condition checks if the context has an attribute role. If the attribute exists, then it checks if it is a set containing the string "admin" as an element.

context has role && context.role.contains("admin")       //true

The attribute name role can be written as an identifier (as in the previous example) or as a string literal. The following expression is equivalent to the previous one:

context has "role" && context.role.contains("admin")     //true

You must check for presence of nested, optional attributes one layer at a time. For example, to check for the presence of principal.custom.project where both custom and project are optional, you must first check if principal has a custom attribute and then check that it principal.custom has a project attribute:

principal has custom && principal.custom has project && principal.custom.project == "greenzone"

If the attribute name is not valid as an identifier, then the string literal syntax must be used for has and attribute values must be accessed with the [] operator instead of using dot syntax. For example, to check if context has an attribute called owner info (with an embedded space), then you could use the following syntax.

context has "owner info" && context["owner info"].name == "Alice"    //true

The following expression returns false because context doesn’t have an attribute tag.

context has tag      //false

Evaluating the following expression results in an error because the left-hand side of the has operator must be an entity or a record.

context.role has admin     //type error

The following expression returns false because the addr sub-record does not have an attribute country. The second expression is not evaluated.

context.addr has country && context.addr.country == "US "    //false

However, consider the case where context does not have the addr sub-record at all:

"context": {
    "role": ["admin", "user"]
}

In that case, then the previous expression that checks for context.addr has country raises a missing-attribute error on context.addr before the has operator is even evaluated, and the entire policy is skipped. If the addr sub-record is optional, you can avoid this error by checking whether addr is present before accessing it with the . operator:

context has addr && context.addr has country && context.addr.country == "US"  // false, with no error

Validation

Validating has expressions relies on information specified in the schema about what entity and record types have what attributes, and which attributes might be optional. Suppose we have expression context has role. It can have the following types:

  • Type boolean if the schema says context has a role attribute which is not required, i.e., it’s optional
  • Type True if the schema says context has a role attribute which is required, which means the expression will always evaluate to true
  • Type False if the schema says context does not have a role attribute, which means that the expression will always evaluate to false

Recall that types True and False are used internally by the validator for simulating the evaluator’s short-circuiting behavior for && and ||.

The validator will reject any has expression whose left-hand operand is not an expression whose type is an entity or record.

.hasTag() (presence of tag test)

Usage: <entity>.hasTag(<expr>)

Method that evalutes to true if the entity on the left has a value defined for the tag name specified on the right. Unlike for attributes with has, for tags the tag name may be any (string-typed) expression, and does not have to be a string literal. Evaluation (and validation) produces an error if <entity> is not an entity or if <expr> does not evaluate to a string.

In all other respects, .hasTag() behaves similarly to has except that it operates on tags instead of attributes. (And only entities, not records, can have tags.)

Calling .hasTag() on an entity type without a tags declaration is valid and will return false without a validation error, because entities of that type cannot have tags.

.getTag() (tag access)

Usage: <entity>.getTag(<expr>)

Method that gets the value of a given tag. The tag name (<expr>) may be any (string-typed) expression, and does not have to be a string literal. Evaluation (and validation) produces an error if <entity> is not an entity or if <expr> does not evaluate to a string.

For validation, .getTag() has the same relationship to .hasTag() as . has to has. See the notes on has.

The .getTag() expression has a return type as specified by the tags declaration on the appropriate entity type in the schema. If .getTag() is used on an entity type without a tags declaration, validation produces an error.

is (entity type test)

Usage: <entity> is <entity-type>

boolean operator that evaluates to true if the left operand is an entity that has the specified entity type and evaluates to false if the left operand is an entity that does not have the specified entity type. Evaluating (and validating) an is expression where the type of an expression that is not an entity results in an error. Validation can also fail if the RHS is not a known entity type (given in the schema).

Using is is helpful when knowing the type of an entity ensures that it has particular attributes or entity relationships. For example, suppose that for requests with action Action::"view", the principal always has type User, but the resource could have type Photo or type User, and only entities of type Photo have an owner attribute. Then we might write the following policy.

permit(principal, action == Action::"view", resource)
when {
  resource is Photo && resource.owner == principal
};

Because of short-circuiting of &&, we know that only if the resource is Photo sub-expression succeeds will we access resource.owner, so we can be confident the attribute is present.

Usage: <entity> is <entity-type> in <entity>

Usage: <entity> is <entity-type> in set(<entity>)

The is operator may optionally be combined with an in operation, in which case the expression is equivalent to <entity> is <entity-type> && <entity> in <entity> or <entity> is <entity-type> && <entity> in set(<entity>).

Examples:

Examples labeled with //error both fail to evaluate and fail to validate.

User::"alice" is User                       //true
principal is User                           //true if `principal` has the `User` entity type
principal is User in Group::"friends"       //true if `principal` has the `User` entity type and is in `Group::"friends"`
ExampleCo::User::"alice" is ExampleCo::User //true
Group::"friends" is User                    //false
ExampleCo::User::"alice" is User            //false - `ExampleCo::User` and `User` are different entity types
"alice" is String                           //error - `is` applies only to entity types, not strings

Validation

Validating is expressions relies on information specified in the schema about what the possible entity types are, and what the principal, resource, and context types can be for particular actions. Suppose we have expression principal is Admin. It can have the following types:

  • Type error if Admin is not declared as an entity type in the schema
  • Type True if the schema says principal surely does have type Admin, which means the expression will always evaluate to true
  • Type False if the schema says principal surely does not have a Admin, which means that the expression will always evaluate to false

Recall that types True and False are used internally by the validator for expressions that definitely evaluate to true or false, respectively. They are used for simulating the evaluator’s short-circuiting behavior for && and ||.

It may seem strange that the validator always knows when an is expression will evaluate to true or false. This is because it always knows when an expression has an entity type, and when it does, what type it must have. In particular, when validating a particular policy, the validator makes sure the policy is valid for every possible principal, resource, and action entity type combination defined by the schema. Consider our example policy up above:

permit(principal, action == Action::"view", resource)
when {
  resource is Photo && resource.owner == principal
};

Recall that for Action::"view" we said that principals always have type User, but resources could be either Photo or User entities. Thus, the validator first considers the case where principal has entity type User and resource has type Photo. In this case, the resource is Photo sub-expression has type True, so the validator also considers the resource.owner == principal sub-expression, which is valid since Photo entities have owner attributes of type User. The validator next considers the case where principal has entity type User and resource has type User. In this case, the resource is Photo sub-expression has type False (since a resource with User type is not a Photo), and thanks to short-circuiting the whole expression can be given type False.

.contains() (single element set membership test)

Usage: <set>.contains(<value>)

Function that evaluates to true if the operand is a member of the receiver on the left side of the function. The receiver must be of type Set or evaluation produces an error. To be accepted by the policy validator, contains must be called on a receiver that is a Set of some type T, with an argument that also has type T.

Examples:

Examples labeled with //error both fail to evaluate and fail to validate. Examples that evaluate to a result may fail to validate.

[1,2,3].contains(1)                             //Evaluates to true //Validates
[1,"something",2].contains(1)                   //Evaluates to true //Doesn't validate (heterogeneous set)
[1,"something",2].contains("Something")         //Evaluates to false (string comparison is case-sensitive) //Doesn't validate (heterogeneous set)
["some", "useful", "tags"].contains("useful")   //Evaluates to true //Validates
[].contains(100)                                //Evaluates to false // Doesn't validate (has empty-set literal)
context.role.contains("admin")                  //Evaluates to true (if the `context.role` set contains string "admin") //Validates
[User::"alice"].contains(principal)             //Evaluates to true (if principal == User::"alice") //Validates
"ham and ham".contains("ham")                   //error - 'contains' is not allowed on strings

A heterogeneous set, as shown in several examples, contains more than one type. None of the validates: false examples is a valid set. See valid sets for more info.

.containsAll() (all element set membership test)

Usage: <set>.containsAll(<set>)

Function that evaluates to true if every member of the operand set is a member of the receiver set. Both the receiver and the operand must be of type set or evaluation results in an error. To be accepted by the validator, the receiver and argument to containsAll must be homogeneous sets of the same type.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error. The remaining examples evaluate to a proper result, but some fail to validate, as indicated in the labels.

[1, -22, 34].containsAll([-22, 1])                                //Evaluates to true //Validates
[1, -22, 34].containsAll([-22])                                   //Evaluates to true //Validates
[43, 34].containsAll([34, 43])                                    //Evaluates to true //Validates
[1, -2, 34].containsAll([1, -22])                                 //Evaluates to false //Validates
[1, 34].containsAll([1, 101, 34])                                 //Evaluates to false //Validates
[false, 3, [47, 0], "some"].containsAll([3, "some"])              //Evaluates to true //Doesn't validate (heterogeneous set)
[false, 3, [47, 0], {"2": "ham"}].containsAll([3, {"2": "ham"}])  //Evaluates to true //Doesn't validate (heterogeneous set)
[2, 43].containsAll([])                                           //Evaluates to true //Doesn't validate (emptyset literal)
[].containsAll([2, 43])                                           //Evaluates to false //Doesn't validate (emptyset literal)
[false, 3, [47, 0], "thing"].containsAll("thing")                 //error - operand a string
"ham and eggs".containsAll("ham")                                 //error - prefix and operand are strings
{"2": "ham", "3": "eggs "}.containsAll({"2": "ham"})              //error - prefix and operand are records

Some examples evaluate to a result but fail to validate for one or more of the following reasons:

  • They operate on heterogeneous sets: values of multiple types
  • They reference the empty-set literal []
  • They don’t operate on sets at all. See valid sets for more info.

.containsAny() (any element set membership test)

Usage: <set>.containsAny(<set>)

Function that evaluates to true if any one or more members of the operand set is a member of the receiver set. Both the receiver and the operand must be of type set or evaluation produces an error. To be accepted by the policy validator, calls to containsAny must be on homogeneous sets of the same type.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error. The remaining examples evaluate to a proper result, but some fail to validate, as indicated in the labels.

[1, -22, 34].containsAny([1, -22])                             //Evaluates to true //Validates
[1, -22].containsAny([1, -22, 34])                             //Evaluates to true //Validates
[-22].containsAny([1, -22, 34])                                //Evaluates to true //Validates
[1, 101].containsAny([1, -22, 34])                             //Evaluates to true //Validates
[1, 101].containsAny([-22, 34])                                //Evaluates to false //Validates
["alice","bob","charlie"].containsAny(["david","bob","juan"])  //Evaluates to true //Validates
[].containsAny(["bob"])                                        //Evaluates to false //Doesn't validate (emptyset literal)
["bob"].containsAny([])                                        //Evaluates to false //Doesn't validate (emptyset literal)
"ham".containsAny("ham and eggs")                              //error - operand is a string
{"2": "ham"}.containsAny({"2": "ham", "3": "eggs "})           //error - prefix and operands are records

The examples that evaluate to a result but fail to validate reference the empty-set literal []. See valid sets for more info.

IP address functions

Use these functions to test characteristics of IP addresses and ranges.

.isIpv4() (IPv4 address valid test)

Usage: <ipaddr>.isIpv4()

Evaluates to true if the receiver is an IPv4 address; evaluates (and validates) to an error if receiver does not have ipaddr type. This function takes no operand.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error.

ip("127.0.0.1").isIpv4()     //true
ip("::1").isIpv4()           //false
ip("127.0.0.1/24").isIpv4()  //true
context.foo.isIpv4()         //error if `context.foo` is not an `ipaddr`

.isIpv6() (IPv6 address valid test)

Usage: <ipaddr>.isIpv6()

Function that evaluates to true if the receiver is an IPv6 address; evaluates (and validates) to an error if received does not have ipaddr type. This function takes no operand.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error.

ip("127.0.0.1/24").isIpv6()  //false
ip("ffee::/64").isIpv6()     //true
ip("::1").isIpv6()           //true
context.foo.isIpv6()         //error if `context.foo` is not an `ipaddr`

.isLoopback() (test for IP loopback address)

Usage: <ipaddr>.isLoopback()

Function that evaluates to true if the receiver is a valid loopback address for its IP version type; evaluates (and validates) to an error if receiver does not have ipaddr type. This function takes no operand.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error.

ip("127.0.0.2").isLoopback()  //true
ip("::1").isLoopback()        //true
ip("::2").isLoopback()        //false
context.foo.isLoopback()      //error if `context.foo` is not an `ipaddr`

.isMulticast() (test for multicast address)

Usage: <ipaddr>.isMulticast()

Function that evaluates to true if the receiver is a multicast address for its IP version type; evaluates (and validates) to an error if receiver does not have ipaddr type. This function takes no operand.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error.

ip("127.0.0.1").isMulticast()  //false
ip("ff00::2").isMulticast()    //true
context.foo.isMulticast()      //error if `context.foo` is not an `ipaddr`

.isInRange() (test for inclusion in IP address range)

Usage: <ipaddr>.isInRange(<ipaddr>)

Function that evaluates to true if the receiver is an IP address or a range of addresses that fall completely within the range specified by the operand. This function evaluates (and validates) to an error if either operand does not have ipaddr type.

Examples:

In the examples that follow, those labeled //error both evaluate and validate to an error.

ip("192.168.0.1").isInRange(ip("192.168.0.1/24"))   //true
ip("192.168.0.1").isInRange(ip("192.168.0.1/28"))   //true
ip("192.168.0.75").isInRange(ip("192.168.0.1/24"))  //true
ip("192.168.0.75").isInRange(ip("192.168.0.1/28"))  //false
ip("1:2:3:4::").isInRange(ip("1:2:3:4::/48"))       //true
ip("192.168.0.1").isInRange(ip("1:2:3:4::"))        //false
ip("192.168.0.1").isInRange(1)                      //error - operand is not an ipaddr
context.foo.isInRange(ip("192.168.0.1/24"))         //error if `context.foo` is not an `ipaddr`