Skip to content

Latest commit

 

History

History
495 lines (433 loc) · 22.7 KB

config-guide.md

File metadata and controls

495 lines (433 loc) · 22.7 KB
Raw

A VPP Configuration Utility

Configuration Guide

vppcfg consumes YAML files of a specific format. Their validity is asserted by two main types of validation:

  1. syntax checks are performed by Yamale and this ensures that all fields in the YAML file are correctly formed, that field-names are correctly spelled, that no extra fields are given, and their values are of the correct type.
  2. semantic validations are performed to ensure that configurations are safely applyable to a running VPP. Note: Some semantic checks are stricter than VPP, because applying them may leave the dataplane in a non-recoverable state.

For the curious, the Yamale syntax validation lives in this schema. If you want to get started quickly and don't mind cargo-culting, take a look at this example.

Basic structure

The YAML configuration file has the following structure, consisting of several maps of a given object type, which specify names of those objects:

loopbacks:
  loop0:
    [ Loopback Configuration ]

bondethernets:
  BondEthernet0:
    [ BondEthernet (bond) Configuration ]

vxlan_tunnels:
  vxlan_tunnel0:
    [ VXLAN (tunnel) Configuration ]

bridgedomains:
  bd1:
    [ BridgeDomain Configuration ]

interfaces:
  GigabitEthernet3/0/0:
    [ Interface Configuration ]
  BondEthernet0:
    [ BondEthernet (interface) Configuration ]
  vxlan_tunnel0:
    [ VXLAN (interface) Configuration ]

Object names are strictly enforced, they must be unique in their scope, and they are case sensitive. For example, any loopback MUST be named loopN, and any bondethernet MUST be named BondEthernetN (note here the camel case). A distinction is made between the object and the resulting interface: A BondEthernet occurs twice in the configuration. The first time, in the bondethernets section, the bond configuration is specified. That bond configuration yields an interface in VPP named BondEthernetN, which occurs in the interfaces section where it can then be manipulated like any other interface (eg. have IP addresses, Linux Control Plane names, sub-interfaces and so on). The same is true for VXLAN tunnels, the only currently supported tunnel type.

Loopbacks

Loopbacks are required to be named loopN where N in [0,4096). The configuration allows the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • lcp: A Linux Control Plane interface pair LIP. If specified, the loopback will be presented in Linux under this name. Its name may be at most 15 characters long, and match the regular expression [a-z]+[a-z0-9-]*.
  • mtu: An integer value between [128,9216], noting the (packet) MTU of the loopback. It will default to 1500 if not specified.
  • addresses: A list of between one and six IPv4 or IPv6 addresses including prefixlen in CIDR format. VPP requires IP addresses to be unique in the entire dataplane, with one notable exception: Multiple IP addresses in the same prefix/len can be added on one and the same interface.
  • mpls: An optional boolean that configures MPLS on the interface or sub-interface. The default value is false, if the field is not specified, which means MPLS will not be enabled. This allows BVIs, represented by Loopbacks, to participate in MPLS .

Although VPP would allow it, vppcfg does not allow for loopbacks to have sub-interfaces.

Examples:

loopbacks:
  loop0:
    description: "loopback with default 1500 byte MTU"
    lcp: lo0
    addresses: [ 10.0.0.1/32, 2001:db8::1/128 ]
  loop1:
    lcp: bvi1
    mtu: 9000
    addresses: [ 10.0.1.1/24, 10.0.1.2/24, 2001:db8:1::1/64 ]
    mpls: true

Bridge Domains

BridgeDomains are required to be named bdN where N in [1, 16777216). Note that bridgedomain bd0 is reserved and cannot be used. The configuration allows the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • mtu: An integer value between [128,9216], noting the (packet) MTU of the bridgedomain. It will default to 1500 if not specified. All member interfaces, including the BVI, are required to have the same MTU as their bridge.
  • bvi: An optional bridge virtual interface (sometimes also referred to as an IRB) which refers to an existing loopback interface by name (ie loop0).
  • interfaces: A list of zero or more interfaces or sub-interfaces that are bridge members. If the bridge has a BVI, it MUST NOT appear in this list. Bridges are allowed to exist with no member interfaces.
  • settings: A map of bridge-domain settings to further manipulate its behavior:
    • learn: A boolean that turns learning on/off. Default True.
    • unicast-flood: A boolean that turns unicast flooding on/off. Default True.
    • unknown-unicast-flood: A boolean that turns unknown unicast flooding on/off. Default True.
    • unicast-forward: A boolean that turns unicast forwarding on/off. Default True.
    • arp-termination: A boolean that turns termination and response of ARP Requests on/off. Default False.
    • arp-unicast-forward: A boolean that turns L2 arp-unicast forwarding on/off. Default False.
    • mac-age-minutes: An integer between [0,256) that drives the ARP timeout on the bridge in minutes, where 0 means do not age out, which is the default.

Any member sub-interfaces that are added, will automatically be configured to tag-rewrite the number of tags they have, so a simple dot1q sub-interface will be configured as pop 1, while a QinQ or QinAD sub-interface will be configured as pop 2. Conversely, when interfaces are removed from the bridge, their tag-rewriting will be disabled.

Examples:

bridgedomains:
  bd10:
    mtu: 2000
    bvi: loop1
    interfaces: [ BondEthernet0.500, HundredGigabitEthernet12/0/1, vxlan_tunnel1 ]
  bd11:
    description: "No members, default 1500 byte MTU, with (default) settings"
    settings:
      learn: True
      unicast-flood: True
      unknown-unicast-flood: True
      unicast-forward: True
      arp-termination: False
      arp-unicast-forward: False
      mac-age-minutes: 0

Caveat: The flooding of unknown-unicast can be turned on or off, but flooding to a specific interface (as opposed to all interfaces which is the default), is not supported.

BondEthernets

BondEthernets are required to be named BondEthernetN (note the camelcase) where N in [0,4294967294). The configuration allows the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • interfaces: A list of zero or more interfaces that are bond members. The interfaces must be PHYs, and in their interface configuration, members are allowed only to set the MTU.
  • mode: A mode to run the LAG in. Can be one of 'round-robin', 'active-backup', 'xor', 'broadcast' or 'lacp'. The default is LACP.
  • load-balance: A loadbalancing strategy to use, if the mode is either XOR or LACP. Can be one of 'l2', 'l23', or 'l34'. The default is l34, which hashes on the source and destination IPs and ports.

Note that the configuration object here only specifies the link aggregation and its members. BondEthernets are expected to occur as well in the interfaces section, where their sub-interfaces and IP addresses and so on are specified.

Examples:

bondethernets:
  BondEthernet0:
    description: "Core: LACP to fsw0.lab.ipng.ch"
    interfaces: [ GigabitEthernet1/0/0, GigabitEthernet1/0/1 ]
    mode: lacp
    load-balance: l2
  BondEthernet1:
    description: "Core: RR LAG"
    interfaces: [ GigabitEthernet3/0/0, GigabitEthernet3/0/1 ]
    mode: round-robin

VXLAN Tunnels

VXLAN Tunnels are required to be named vxlan_tunnelN (note the underscore), where N in [0,2G). The configuration allows the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • local: A required IPv4 or IPv6 address for our (source) side of the tunnel.
  • remote: A required IPv4 or IPv6 address for their (destination) side of the tunnel.
  • vni: A Virtual Network Indentifier, a required integer number between [1,16M).

Local and Remote sides of the tunnel MUST have the same address family.

Caveat: VXLAN tunnels are currently only possible as unicast (src/dst), with static source and destination ports (4789), and with a decap-next of L2. Also, VNIs must be globally unique. In a future release of vppcfg, these fields will be configurable, and VNI reuses will be allowed between different dst endpoints.

Examples:

vxlan_tunnels:
  vxlan_tunnel0:
    description: "Some IPv6 VXLAN tunnel"
    local: 2001:db8::1
    remote: 2001:db8::2
    vni: 100
  vxlan_tunnel1:
    local: 192.0.2.1
    remote: 192.0.2.2
    vni: 101

TAPs

TAPs are virtual L2 (and sometimes L3) devices in the kernel, that are backed by a userspace program. VPP can create a TAP and expose them in a network namespace, and optionally add them to a (Linux) bridge.

TAPs are required to be named tapN where N in [0,1024], but be aware that Linux CP will use TAPs with an instance id that equals their hardware interface id. It is safer to create TAPs from the top of the namespace, for example tap100, see the caveat below on why. The configuration then allows for the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • host: Configuration of the Linux side of the TAP:
    • name: A (mandatory) Linux interface name, at most 15 characters long, matching the regular expression [a-z]+[a-z0-9-]*.
    • mac: The MAC address for the Linux interface, if empty it will be randomly assigned.
    • mtu: The MTU of the Linux interface, if empty it will be set to 1500.
    • bridge: An optional Linux bridge to add the Linux interface into. Note: VPP will expect this bridge to exist, otherwise the addition will silently fail after creating the TAP.
    • namespace: An optional Linux network namespace in which to add the Linux interface, which can be empty (the default) in which case the Linux interface is created in the default namespace.
    • bridge-create: A boolean that determines if vppcfg will create the bridge in the namespace if it does not yet exist, and will set its MTU to the host.mtu value if it does exist. Defaults to False, and can only be True if bridge is given.
    • namespace-create: A boolean that determines if vppcfg will create the network namespace if it does not yet exist. Defaults to False, and can only be True if namespace is given.
  • rx-ring-size: An optional RX ringbuffer size, a value from 8 to 32K, must be a power of two. If it is not specified, it will default to 256.
  • tx-ring-size: An optional TX ringbuffer size, a value from 8 to 32K, must be a power of two. If it is not specified, it will default to 256.

NOTE: The Linux Controlplane (LCP) plugin in VPP also uses TAPs to expose the dataplane (sub-) interfaces in Linux, but for that functionality, refer to the lcp fields in interfaces and loopbacks.

Caveat: syncing changed attributes (with the exception of the bridge name) after the TAP was created is not supported. This is because there are no API setters in VPP. Changing attributes is possible, but operators should expect that the TAP interface gets pruned and recreated.

Caveat: vppcfg will try to ensure a TAP is not created with the same instance ID as a hardware interface, but it can not make strict guarantees, because there exists no API to look the hardware interface id's up. As a rule of thumb, start TAPs at twice the total count of hardware interfaces (PHYs, BondEthernets, VXLAN Tunnels and other TAPs) in the config.

Examples:

taps:
  tap100:
    description: "TAP with MAC, MTU and Bridge"
    host:
      name: vpp-tap100
      mac: f6:18:fe:e7:d2:3a
      mtu: 9000
      namespace: test
      namespace-create: True
      bridge: vpp-br0
      bridge-create: True
    rx-ring-size: 1024
    tx-ring-size: 512

Interfaces

Interfaces and their sub-interfaces are configured very similarly. Interface names MUST either exist as a PHY in VPP (ie. HundredGigabitEthernet12/0/0) or as a specified BondEthernetN or vxlan_tunnel0 device. The configuration allows the following fields:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • lcp: A Linux Control Plane interface pair LIP. If specified, the interface will be presented in Linux under this name. Its name may be at most 15 characters long, and match the regular expression [a-z]+[a-z0-9-]*. In sub-interfaces, a LIP may only be specified if its direct parent has an LIP as well. In the case of a QinQ or QinAD sub-interface, there must exist an intermediary interface with the correct encapsulation, and it too must have a LIP.
  • mtu: An integer value between [128,9216], noting the MTU of the interface. The MTU for a PHY will be set as its Max Frame Size in addition to its packet MTU. Parents must always have a larger MTU than any of their children (this is done to satisfy Linux Control Plane).
  • addresses: A list of between one and six IPv4 or IPv6 addresses including prefixlen in CIDR format. VPP requires IP addresses to be unique in the entire dataplane, with one notable exception: Multiple IP addresses in the same prefix/len can be added on one and the same interface.
  • l2xc: A Layer2 Cross Connect interface name. An l2xc will be configured, after which this interface cannot have any L3 configuration (IP addresses or LCP), and neither can the target interface.
  • state: An optional string that configures the link admin state, either up or down. If it is not specified, the link is considered admin 'up'.
  • device-type: An optional interface type in VPP. Currently the only supported vlaue is dpdk, and it is used to generate correct mock interfaces if the --novpp flag is used.
  • mpls: An optional boolean that configures MPLS on the interface or sub-interface. The default value is false, if the field is not specified, which means MPLS will not be enabled.

Further, top-level interfaces, that is to say those that do not have an encapsulation, are permitted to have any number of sub-interfaces specified by subid, an integer between [0,2G), which further allow the following field:

  • encapsulation: An encapsulation for the sub-interface:
    • dot1q: An outer Dot1Q tag, an integer between [1,4096).
    • dot1ad: An outer Dot1AD tag, an integer between [1,4096).
    • inner-dot1q: An inner Dot1Q tag, an integer between [1,4096).
    • exact-match: A boolean, signalling the sub-interface should match on the exact number of tags specified. This is required for any L3 interface (carrying an IP address or LCP), but allowed to be False for L2 interfaces (ie. bridge-domain members or L2XC targets).

It's permitted to omit the encapsulation fields, in which case an exact-matching Dot1Q encapsulation with tag value equal to the subid will be configured. Obviously, it is forbidden to specify both dot1q and dot1ad fields at the same time.

Examples:

interfaces:
  HundredGigabitEthernet12/0/0:
    device-type: dpdk
    lcp: "ice0"
    mtu: 9000
    addresses: [ 192.0.2.1/30, 2001:db8:1::1/64 ]
    mpls: true
    sub-interfaces:
      1234:
        mtu: 9000
        lcp: "ice0.dot1q"
        addresses: [ 192.0.2.5/30, 2001:db8:2::1/64 ]
      1235:
        mtu: 1500
        lcp: "ice0.qinq"
        mpls: true
        addresses: [ 192.0.2.9/30, 2001:db8:3::1/64 ]
        encapsulation:
          dot1q: 1234
          inner-dot1q: 1000
          exact-match: True

  BondEthernet0:
    mtu: 9000
    lcp: "bond0"
    sub-interfaces:
      100:
        mtu: 2500
        l2xc: BondEthernet0.200
        encapsulation:
           dot1q: 100
           exact-match: False
      200:
        mtu: 2500
        l2xc: BondEthernet0.100
        encapsulation:
           dot1q: 200
           exact-match: False

Prefix Lists

This construct allows to enumerate a list of IPv4 or IPv6 host addresses and/or networks. Each prefixlist has a name which consists of anywhere between 1 and 56 characters, and it must start with a letter. The prefixlist name any is reserved. The syntax is straight forward:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • members: A list of zero or more entries which can take the form:
    • IPv4 Host: an IPv4 address, eg. 192.0.2.1
    • IPv4 Prefix: an IPv6 prefix, eg. 192.0.2.0/24
    • IPv6 Host: an IPv4 address, eg. 2001:db8::1
    • IPv6 Prefix: an IPv6 prefix, eg. 2001:db8::0/64

NOTE: It is valid to have host addresses with prefixlen, for example 192.168.1.1/24 in other words, the prefix can be either a network or a host.

A few examples:

prefixlists:
  example:
    description: "An example prefixlist with hosts and prefixes"
    members:
      - 192.0.2.1
      - 192.0.2.0/24
      - 2001:db8::1
      - 2001:db8::/64
  empty:
    description: "An empty prefixlist"
    members: []

Access Control Lists

In VPP, a common firewall function is provided by the acl-plugin. The anatomy of this plugin is as follows. First, an ACL consists of one or more Access Control Elements or ACEs. These can match on IPv4 or IPv6 source/destination, an IP protocol, and then for TCP/UDP a range of source- and destination ports, and for ICMP a range of ICMP type and codes. Any matching packets then either perform an action of permit or deny (for stateless) or permit+reflect (stateful). The full syntax is as follows:

  • description: A string, no longer than 64 characters, and excluding the single quote ' and double quote ". This string is currently not used anywhere, and serves for enduser documentation purposes.
  • terms: A list of Access Control Elements:
    • action: What to do upon match, can be either permit, deny or permit+reflect. This is the only required field.
    • family: Which IP address family to match, can be either ipv4, or ipv6 or any, which is the default. If any is used, this term will also operate on any source and destination addresses, and it will emit two ACEs, one for each address family.
    • source: Either an IPv4 or IPv6 host (without prefixlen, eg. 192.0.2.1 or 2001:db8::1), an IPv4 or IPv6 prefix (with prefixlen, eg. 192.0.2.0/24 or 2001:db8::/64), or a reference to the name of an existing prefixlist (eg. trusted). If left empty, this means all IPv4 and IPv6 (ie. [ 0.0.0.0/0, ::/0 ]).
    • destination: Similar to source, but for the destination field of the packets.
    • protocol: The L4 protocol, can be either a numeric value (eg. 6), or a symbolic string value from /etc/protocols (eg. tcp). If omitted, only L3 matches are performed.
    • source-port: When TCP or UDP are specified, this field specified which source port(s) are matched. It can be either a numeric value (eg. 80), a symbolic string value from /etc/services (eg. www), a numeric range with start and/or end ranges (eg. -1024 for all ports from 0-1024 inclusive; or 1024- for all ports from 1024-65535 inclusive, or an actual range 49152-65535). The default keyword any is also permitted, which results in range 0-65535, and is the default if the field is not specified.
    • destination-port: Similar to source-port but for the destination port field in the TCP or UDP header.
    • icmp-type: It can be either a numeric value (eg. 3), a numeric range with start and/or end ranges (eg. -10 for all types from 0-10 inclusive; or 10- for all types from 10-255 inclusive, or an actual range 10-15). The default keyword any is also permitted, which results in range 0-255, and is the default if the field is not specified. This field can only be specified if the protocol field is icmp (1) or ipv6-icmp (58).
    • icmp-code: Similar to icmp-type but for the ICMP code field. This field can only be specified if the protocol field is icmp (1) or ipv6-icmp (58).

An example ACL with four ACE terms:

prefixlists:
  example:
    description: "An example prefixlist with hosts and prefixes"
    members:
      - 192.0.2.1
      - 192.0.2.0/24
      - 2001:db8::1
      - 2001:db8::/64

acls:
  acl01:
     description: "Test ACL"
     terms:
       - description: "Allow a prefixlist, but only for IPv6"
         family: ipv6
         action: permit
         source: example
       - description: "Allow a specific IPv6 TCP flow"
         action: permit
         source: 2001:db8::/64
         destination: 2001:db8:1::/64
         protocol: tcp
         destination-port: www
         source-port: "1024-65535"
       - description: "Allow IPv4 ICMP Destination Unreachable, any code"
         family: ipv4
         action: permit
         protocol: icmp
         icmp-type: 3
         icmp-code: any
       - description: "Deny any IPv4 or IPv6"
         action: deny

One or more of these ACLs are then applied to an interface in either the input or the output direction:

interfaces:
  GigabitEthernet3/0/0:
    acls:
      input: acl01
      output: [ acl02, acl03 ]

The configuration here is tolerant of either a singleton (a literal string referring to the one ACL that must be applied), or a list of strings to more than one ACL, in which case they will be tested in order (with a first-match return value).