Next Previous Contents

5. Filters

5.1 Introduction

BIRD contains a simple programming language. (No, it can't yet read mail :-). There are two objects in this language: filters and functions. Filters are interpreted by BIRD core when a route is being passed between protocols and routing tables. The filter language contains control structures such as if's and switches, but it allows no loops. An example of a filter using many features can be found in filter/test.conf.

Filter gets the route, looks at its attributes and modifies some of them if it wishes. At the end, it decides whether to pass the changed route through (using accept) or whether to reject it. A simple filter looks like this:

filter not_too_far
        int var;
        if defined( rip_metric ) then
                var = rip_metric;
        else {
                var = 1;
                rip_metric = 1;
        if rip_metric > 10 then
                reject "RIP metric is too big";
                accept "ok";

As you can see, a filter has a header, a list of local variables, and a body. The header consists of the filter keyword followed by a (unique) name of filter. The list of local variables consists of type name; pairs where each pair declares one local variable. The body consists of { statements }. Each statement is terminated by a ;. You can group several statements to a single compound statement by using braces ({ statements }) which is useful if you want to make a bigger block of code conditional.

BIRD supports functions, so that you don not have to repeat the same blocks of code over and over. Functions can have zero or more parameters and they can have local variables. If the function returns value, then you should always specify its return type. Direct recursion is possible. Function definitions look like this:

function name() -> int
        int local_variable;
        int another_variable = 5;
        return 42;

function with_parameters(int parameter) -> pair
        print parameter;
        return (1, 2);

Like in C programming language, variables are declared inside function body, either at the beginning, or mixed with other statements. Declarations may contain initialization. You can also declare variables in nested blocks, such variables have scope restricted to such block. There is a deprecated syntax to declare variables after the function line, but before the first {. Functions are called like in C: name(); with_parameters(5);. Function may return values using the return [expr] command. Returning a value exits from current function (this is similar to C).

Filters are defined in a way similar to functions except they cannot have explicit parameters and cannot return. They get a route table entry as an implicit parameter, it is also passed automatically to any functions called. The filter must terminate with either accept or reject statement. If there is a runtime error in filter, the route is rejected.

A nice trick to debug filters is to use show route filter name from the command line client. An example session might look like:

pavel@bug:~/bird$ ./birdc -s bird.ctl
BIRD 0.0.0 ready.
bird> show route         dev eth0 [direct1 23:21] (240)    dev tunl1 [direct1 23:21] (240)        dev lo [direct1 23:21] (240)
bird> show route ?
show route [<prefix>] [table <t>] [filter <f>] [all] [primary]...
bird> show route filter { if ~ net then accept; }        dev lo [direct1 23:21] (240)

5.2 Data types

Each variable and each value has certain type. Booleans, integers and enums are incompatible with each other (that is to prevent you from shooting oneself in the foot).


This is a boolean type, it can have only two values, true and false. Boolean is the only type you can use in if statements.


This is a general integer type. It is an unsigned 32bit type; i.e., you can expect it to store values from 0 to 4294967295. Overflows are not checked. You can use 0x1234 syntax to write hexadecimal values.


This is a pair of two short integers. Each component can have values from 0 to 65535. Literals of this type are written as (1234,5678). The same syntax can also be used to construct a pair from two arbitrary integer expressions (for example (1+2,a)).

Operators .asn and .data can be used to extract corresponding components of a pair: (asn, data).


This is a dotted quad of numbers used to represent router IDs (and others). Each component can have a value from 0 to 255. Literals of this type are written like IPv4 addresses.


This is a string of characters. There are no ways to modify strings in filters. You can pass them between functions, assign them to variables of type string, print such variables, use standard string comparison operations (e.g. =, !=, <, >, <=, >=), but you can't concatenate two strings. String literals are written as "This is a string constant". Additionally matching (~, !~) operators could be used to match a string value against a shell pattern (represented also as a string).


This is a sequences of arbitrary bytes. There are no ways to modify bytestrings in filters. You can pass them between function, assign them to variables of type bytestring, print such values, compare bytestings (=, !=).

Bytestring literals are written as a sequence of hexadecimal digit pairs, optionally colon-separated. A bytestring specified this way must be either at least 16 bytes (32 digits) long, or prefixed by the hex: prefix: 01:23:45:67:89:ab:cd:ef:01:23:45:67:89:ab:cd:ef, 0123456789abcdef0123456789abcdef, hex:, hex:12:34:56, hex:12345678.

A bytestring can be made from a hex string using from_hex() function. Source strings can use any number of dots, colons, hyphens and spaces as byte separators: from_hex(" 12.34 56:78 ab-cd-ef ").


This type can hold a single IP address. The IPv4 addresses are stored as IPv4-Mapped IPv6 addresses so one data type for both of them is used. Whether the address is IPv4 or not may be checked by .is_v4 which returns a bool. IP addresses are written in the standard notation ( or fec0:3:4::1). You can apply special operator .mask(num) on values of type ip. It masks out all but first num bits from the IP address. So = is true.


This type can hold a network prefix consisting of IP address, prefix length and several other values. This is the key in route tables.

Prefixes may be of several types, which can be determined by the special operator .type. The type may be:

NET_IP4 and NET_IP6 prefixes hold an IP prefix. The literals are written as ipaddress/pxlen. There are two special operators on these: .ip which extracts the IP address from the pair, and .len, which separates prefix length from the pair. So = 16 is true.

NET_IP6_SADR nettype holds both destination and source IPv6 prefix. The literals are written as ipaddress/pxlen from ipaddress/pxlen, where the first part is the destination prefix and the second art is the source prefix. They support the same operators as IP prefixes, but just for the destination part. They also support .src and .dst operators to get respective parts of the address as separate NET_IP6 values.

NET_VPN4 and NET_VPN6 prefixes hold an IP prefix with VPN Route Distinguisher (RFC 4364). They support the same special operators as IP prefixes, and also .rd which extracts the Route Distinguisher. Their literals are written as vpnrd ipprefix

NET_ROA4 and NET_ROA6 prefixes hold an IP prefix range together with an ASN. They support the same special operators as IP prefixes, and also .maxlen which extracts maximal prefix length, and .asn which extracts the ASN.

NET_FLOW4 and NET_FLOW6 hold an IP prefix together with a flowspec rule. Filters currently do not support much flowspec parsing, only .src and .dst operators to get source and destination parts of the flowspec as separate NET_IP4 / NET_IP6 values.

NET_MPLS holds a single MPLS label and its handling is currently not implemented.


This is a route distinguisher according to RFC 4364. There are three kinds of RD's: asn:32bit int, asn4:16bit int and IPv4 address:32bit int


This is a specialized type used to represent BGP extended community values. It is essentially a 64bit value, literals of this type are usually written as (kind, key, value), where kind is a kind of extended community (e.g. rt / ro for a route target / route origin communities), the format and possible values of key and value are usually integers, but it depends on the used kind. Similarly to pairs, ECs can be constructed using expressions for key and value parts, (e.g. (ro, myas, 3*10), where myas is an integer variable).


This is a specialized type used to represent BGP large community values. It is essentially a triplet of 32bit values, where the first value is reserved for the AS number of the issuer, while meaning of remaining parts is defined by the issuer. Literals of this type are written as (123, 456, 789), with any integer values. Similarly to pairs, LCs can be constructed using expressions for its parts, (e.g. (myas, 10+20, 3*10), where myas is an integer variable).

Operators .asn, .data1, and .data2 can be used to extract corresponding components of LCs: (asn, data1, data2).

int|pair|quad|ip|prefix|ec|lc|enum set

Filters recognize four types of sets. Sets are similar to strings: you can pass them around but you can't modify them. Literals of type int set look like [ 1, 2, 5..7 ]. As you can see, both simple values and ranges are permitted in sets.

For pair sets, expressions like (123,*) can be used to denote ranges (in that case (123,0)..(123,65535)). You can also use (123,5..100) for range (123,5)..(123,100). You can also use * and a..b expressions in the first part of a pair, note that such expressions are translated to a set of intervals, which may be memory intensive. E.g. (*,4..20) is translated to (0,4..20), (1,4..20), (2,4..20), ... (65535, 4..20).

EC sets use similar expressions like pair sets, e.g. (rt, 123, 10..20) or (ro, 123, *). Expressions requiring the translation (like (rt, *, 3)) are not allowed (as they usually have 4B range for ASNs).

Also LC sets use similar expressions like pair sets. You can use ranges and wildcards, but if one field uses that, more specific (later) fields must be wildcards. E.g., (10, 20..30, *) or (10, 20, 30..40) is valid, while (10, *, 20..30) or (10, 20..30, 40) is not valid.

You can also use expressions for int, pair, EC and LC set values. However, it must be possible to evaluate these expressions before daemon boots. So you can use only constants inside them. E.g.

         define one=1;
         define myas=64500;
         int set odds;
         pair set ps;
         ec set es;

         odds = [ one, 2+1, 6-one, 2*2*2-1, 9, 11 ];
         ps = [ (1,one+one), (3,4)..(4,8), (5,*), (6,3..6), (7..9,*) ];
         es = [ (rt, myas, 3*10), (rt, myas+one, 0..16*16*16-1), (ro, myas+2, *) ];

Sets of prefixes are special: their literals does not allow ranges, but allows prefix patterns that are written as ipaddress/pxlen{low,high}. Prefix ip1/len1 matches prefix pattern ip2/len2{l,h} if the first min(len1, len2) bits of ip1 and ip2 are identical and l <= len1 <= h. A valid prefix pattern has to satisfy low <= high, but pxlen is not constrained by low or high. Obviously, a prefix matches a prefix set literal if it matches any prefix pattern in the prefix set literal.

There are also two shorthands for prefix patterns: address/len+ is a shorthand for address/len{len,maxlen} (where maxlen is 32 for IPv4 and 128 for IPv6), that means network prefix address/len and all its subnets. address/len- is a shorthand for address/len{0,len}, that means network prefix address/len and all its supernets (network prefixes that contain it).

For example, [,,,{16,24} ] matches prefix, all subprefixes of, all superprefixes of and prefixes 4.X.X.X whose prefix length is 16 to 24. [{20,24} ] matches all prefixes (regardless of IP address) whose prefix length is 20 to 24, [ ] matches any prefix that contains IP address ~ [{15,17} ] is true, but ~ [ ] is false.

Cisco-style patterns like ge 16 le 24 can be expressed in BIRD as{16,24}, le 24 as{16,24} and ge 24 as{24,32}.

It is not possible to mix IPv4 and IPv6 prefixes in a prefix set. It is currently possible to mix IPv4 and IPv6 addresses in an ip set, but that behavior may change between versions without any warning; don't do it unless you are more than sure what you are doing. (Really, don't do it.)


Enumeration types are fixed sets of possibilities. You can't define your own variables of such type, but some route attributes are of enumeration type. Enumeration types are incompatible with each other.


BGP path is a list of autonomous system numbers. You can't write literals of this type. There are several special operators on bgppaths:

P.first returns the first ASN (the neighbor ASN) in path P.

P.last returns the last ASN (the source ASN) in path P.

P.last_nonaggregated returns the last ASN in the non-aggregated part of the path P.

Both first and last return zero if there is no appropriate ASN, for example if the path contains an AS set element as the first (or the last) part. If the path ends with an AS set, last_nonaggregated may be used to get last ASN before any AS set.

P.len returns the length of path P.

P.empty makes the path P empty. Can't be used as a value, always modifies the object.

P.prepend(A) prepends ASN A to path P and returns the result.

P.delete(A) deletes all instances of ASN A from from path P and returns the result. A may also be an integer set, in that case the operator deletes all ASNs from path P that are also members of set A.

P.filter(A) deletes all ASNs from path P that are not members of integer set A, and returns the result. I.e., filter do the same as delete with inverted set A.

Methods prepend, delete and filter keep the original object intact as long as you use the result in any way. You can also write e.g. P.prepend(A); as a standalone statement. This variant does modify the original object with the result of the operation.


BGP masks are patterns used for BGP path matching (using path ~ [= 2 3 5 * =] syntax). The masks resemble wildcard patterns as used by UNIX shells. Autonomous system numbers match themselves, * matches any (even empty) sequence of arbitrary AS numbers and ? matches one arbitrary AS number. For example, if bgp_path is 4 3 2 1, then: bgp_path ~ [= * 4 3 * =] is true, but bgp_path ~ [= * 4 5 * =] is false. There is also + operator which matches one or multiple instances of previous expression, e.g. [= 1 2+ 3 =] matches both path 1 2 3 and path 1 2 2 2 3, but not 1 3 nor 1 2 4 3. Note that while * and ? are wildcard-style operators, + is regex-style operator.

BGP mask expressions can also contain integer expressions enclosed in parenthesis and integer variables, for example [= * 4 (1+2) a =]. You can also use ranges (e.g. [= * 3..5 2 100..200 * =]) and sets (e.g. [= 1 2 [3, 5, 7] * =]).


Clist is similar to a set, except that unlike other sets, it can be modified. The type is used for community list (a set of pairs) and for cluster list (a set of quads). There exist no literals of this type. There are special operators on clists:

C.len returns the length of clist C.

C.empty makes the list C empty. Can't be used as a value, always modifies the object.

C.add(P) adds pair (or quad) P to clist C and returns the result. If item P is already in clist C, it does nothing. P may also be a clist, in that case all its members are added; i.e., it works as clist union.

C.delete(P) deletes pair (or quad) P from clist C and returns the result. If clist C does not contain item P, it does nothing. P may also be a pair (or quad) set, in that case the operator deletes all items from clist C that are also members of set P. Moreover, P may also be a clist, which works analogously; i.e., it works as clist difference.

C.filter(P) deletes all items from clist C that are not members of pair (or quad) set P, and returns the result. I.e., filter do the same as delete with inverted set P. P may also be a clist, which works analogously; i.e., it works as clist intersection.

Methods add, delete and filter keep the original object intact as long as you use the result in any way. You can also write e.g. P.add(A); as a standalone statement. This variant does modify the original object with the result of the operation.

C.min returns the minimum element of clist C.

C.max returns the maximum element of clist C.

Operators .min, .max can be used together with filter to extract the community from the specific subset of communities (e.g. localpref or prepend) without the need to check every possible value (e.g. filter(bgp_community, [(23456, 1000..1099)]).min).


Eclist is a data type used for BGP extended community lists. Eclists are very similar to clists, but they are sets of ECs instead of pairs. The same operations (like add, delete or ~ and !~ membership operators) can be used to modify or test eclists, with ECs instead of pairs as arguments.


Lclist is a data type used for BGP large community lists. Like eclists, lclists are very similar to clists, but they are sets of LCs instead of pairs. The same operations (like add, delete or ~ and !~ membership operators) can be used to modify or test lclists, with LCs instead of pairs as arguments.

5.3 Operators

The filter language supports common integer operators (+,-,*,/), parentheses (a*(b+c)), comparison (a=b, a!=b, a<b, a>=b). Logical operations include unary not (!), and (&&), and or (||). Special operators include (~, !~) for "is (not) element of a set" operation - it can be used on element and set of elements of the same type (returning true if element is contained in the given set), or on two strings (returning true if first string matches a shell-like pattern stored in second string) or on IP and prefix (returning true if IP is within the range defined by that prefix), or on prefix and prefix (returning true if first prefix is more specific than second one) or on bgppath and bgpmask (returning true if the path matches the mask) or on number and bgppath (returning true if the number is in the path) or on bgppath and int (number) set (returning true if any ASN from the path is in the set) or on pair/quad and clist (returning true if the pair/quad is element of the clist) or on clist and pair/quad set (returning true if there is an element of the clist that is also a member of the pair/quad set).

There is one operator related to ROA infrastructure - roa_check(). It examines a ROA table and does RFC 6483 route origin validation for a given network prefix. The basic usage is roa_check(table), which checks the current route (which should be from BGP to have AS_PATH argument) in the specified ROA table and returns ROA_UNKNOWN if there is no relevant ROA, ROA_VALID if there is a matching ROA, or ROA_INVALID if there are some relevant ROAs but none of them match. There is also an extended variant roa_check(table, prefix, asn), which allows to specify a prefix and an ASN as arguments.

5.4 Control structures

Filters support several control structures: conditions, for loops and case switches.

Syntax of a condition is: if boolean expression then commandT; else commandF; and you can use { command1; command2; ... } instead of either command. The else clause may be omitted. If the boolean expression is true, commandT is executed, otherwise commandF is executed.

For loops allow to iterate over elements in compound data like BGP paths or community lists. The syntax is: for [ type ] variable in expr do command; and you can also use compound command like in conditions. The expression is evaluated to a compound data, then for each element from such data the command is executed with the item assigned to the variable. A variable may be an existing one (when just name is used) or a locally defined (when type and name is used). In both cases, it must have the same type as elements.

The case is similar to case from Pascal. Syntax is case expr { else: | num_or_prefix [ .. num_or_prefix]: statement ; [ ... ] }. The expression after case can be of any type which can be on the left side of the ~ operator and anything that could be a member of a set is allowed before :. Multiple commands are allowed without {} grouping. If expr matches one of the : clauses, statements between it and next : statement are executed. If expr matches neither of the : clauses, the statements after else: are executed.

Here is example that uses if and case structures:

if 1234 = i then printn "."; else {
        print "not 1234";
        print "You need {} around multiple commands";

for int asn in bgp_path do {
        printn "ASN: ", asn;
        if asn < 65536 then print " (2B)"; else print " (4B)";

case arg1 {
        2: print "two"; print "I can do more commands without {}";
        3 .. 5: print "three to five";
        else: print "something else";

5.5 Route attributes

A filter is implicitly passed a route, and it can access its attributes just like it accesses variables. There are common route attributes, protocol-specific route attributes and custom route attributes. Most common attributes are mandatory (always defined), while remaining are optional. Attempts to access undefined attribute result in a runtime error; you can check if an attribute is defined by using the defined( attribute ) operator. One notable exception to this rule are attributes of bgppath and *clist types, where undefined value is regarded as empty bgppath/*clist for most purposes.

Attributes can be defined by just setting them in filters. Custom attributes have to be first declared by attribute global option. You can also undefine optional attribute back to non-existence by using the unset( attribute ) operator.

Common route attributes are:

prefix net

The network prefix or anything else the route is talking about. The primary key of the routing table. Read-only. (See the chapter about routes.)

enum scope

The scope of the route. Possible values: SCOPE_HOST for routes local to this host, SCOPE_LINK for those specific for a physical link, SCOPE_SITE and SCOPE_ORGANIZATION for private routes and SCOPE_UNIVERSE for globally visible routes. This attribute is not interpreted by BIRD and can be used to mark routes in filters. The default value for new routes is SCOPE_UNIVERSE.

int preference

Preference of the route. Valid values are 0-65535. (See the chapter about routing tables.)

ip from

The router which the route has originated from.

ip gw

Next hop packets routed using this route should be forwarded to.

string proto

The name of the protocol which the route has been imported from. Read-only.

enum source

what protocol has told me about this route. Possible values: RTS_STATIC, RTS_INHERIT, RTS_DEVICE, RTS_RIP, RTS_OSPF, RTS_OSPF_IA, RTS_OSPF_EXT1, RTS_OSPF_EXT2, RTS_BGP, RTS_PIPE, RTS_BABEL.

enum dest

Type of destination the packets should be sent to (RTD_ROUTER for forwarding to a neighboring router, RTD_DEVICE for routing to a directly-connected network, RTD_MULTIPATH for multipath destinations, RTD_BLACKHOLE for packets to be silently discarded, RTD_UNREACHABLE, RTD_PROHIBIT for packets that should be returned with ICMP host unreachable / ICMP administratively prohibited messages). Can be changed, but only to RTD_BLACKHOLE, RTD_UNREACHABLE or RTD_PROHIBIT.

string ifname

Name of the outgoing interface. Sink routes (like blackhole, unreachable or prohibit) and multipath routes have no interface associated with them, so ifname returns an empty string for such routes. Setting it would also change route to a direct one (remove gateway).

int ifindex

Index of the outgoing interface. System wide index of the interface. May be used for interface matching, however indexes might change on interface creation/removal. Zero is returned for routes with undefined outgoing interfaces. Read-only.

bool onlink

Onlink flag means that the specified nexthop is accessible on the interface regardless of IP prefixes configured on the interface. The attribute can be used to configure such next hops by first setting onlink = true and ifname, and then setting gw. Possible use case for setting this flag is to automatically build overlay IP-IP networks on linux.

int weight

Multipath weight of route next hops. Valid values are 1-256. Reading returns the weight of the first next hop, setting it sets weights of all next hops to the specified value. Therefore, this attribute is not much useful for manipulating individual next hops of an ECMP route, but can be used in BGP multipath setup to set weights of individual routes that are merged to one ECMP route during export to the Kernel protocol (with active marge paths option).

int gw_mpls

Outgoing MPLS label attached to route (i.e., incoming MPLS label on the next hop router for this label-switched path). Reading returns the label value and setting it sets it to the start of the label stack. Setting implicit-NULL label (3) disables the MPLS label stack. Only the first next hop and only one label in the label stack supported right now. This is experimental option, will be likely changed in the future to handle full MPLS label stack.

int igp_metric

The optional attribute that can be used to specify a distance to the network for routes that do not have a native protocol metric attribute (like ospf_metric1 for OSPF routes). It is used mainly by BGP to compare internal distances to boundary routers (see below).

int mpls_label

Local MPLS label attached to the route. This attribute is produced by MPLS-aware protocols for labeled routes. It can also be set in import filters to assign static labels, but that also requires static MPLS label policy.

enum mpls_policy

For MPLS-aware protocols, this attribute defines which MPLS label policy will be used for the route. It can be set in import filters to change it on per-route basis. Valid values are MPLS_POLICY_NONE (no label), MPLS_POLICY_STATIC (static label), MPLS_POLICY_PREFIX (per-prefix label), MPLS_POLICY_AGGREGATE (aggregated label), and MPLS_POLICY_VRF (per-VRF label). See MPLS label policy for details.

int mpls_class

When MPLS label policy is set to aggregate, it may be useful to apply more fine-grained aggregation than just one based on next hops. When routes have different value of this attribute, they will not be aggregated under one local label even if they have the same next hops.

Protocol-specific route attributes are described in the corresponding protocol sections.

5.6 Other statements

The following statements are available:

variable = expr

Set variable (or route attribute) to a given value.

accept|reject [ expr ]

Accept or reject the route, possibly printing expr.

return expr

Return expr from the current function, the function ends at this point.

print|printn expr [, expr...]

Prints given expressions; useful mainly while debugging filters. The printn variant does not terminate the line.

Next Previous Contents