Netlink specification support for legacy Generic Netlink families

This document describes the many additional quirks and properties required to describe older Generic Netlink families which form the genetlink-legacy protocol level.

Specification

Globals

Attributes listed directly at the root level of the spec file.

version

Generic Netlink family version, default is 1.

version has historically been used to introduce family changes which may break backwards compatibility. Since compatibility breaking changes are generally not allowed version is very rarely used.

Attribute type nests

New Netlink families should use multi-attr to define arrays. Older families (e.g. genetlink control family) attempted to define array types reusing attribute type to carry information.

For reference the multi-attr array may look like this:

[ARRAY-ATTR]
  [INDEX (optionally)]
  [MEMBER1]
  [MEMBER2]
[SOME-OTHER-ATTR]
[ARRAY-ATTR]
  [INDEX (optionally)]
  [MEMBER1]
  [MEMBER2]

where ARRAY-ATTR is the array entry type.

array-nest

array-nest creates the following structure:

[SOME-OTHER-ATTR]
[ARRAY-ATTR]
  [ENTRY]
    [MEMBER1]
    [MEMBER2]
  [ENTRY]
    [MEMBER1]
    [MEMBER2]

It wraps the entire array in an extra attribute (hence limiting its size to 64kB). The ENTRY nests are special and have the index of the entry as their type instead of normal attribute type.

type-value

type-value is a construct which uses attribute types to carry information about a single object (often used when array is dumped entry-by-entry).

type-value can have multiple levels of nesting, for example genetlink’s policy dumps create the following structures:

[POLICY-IDX]
  [ATTR-IDX]
    [POLICY-INFO-ATTR1]
    [POLICY-INFO-ATTR2]

Where the first level of nest has the policy index as it’s attribute type, it contains a single nest which has the attribute index as its type. Inside the attr-index nest are the policy attributes. Modern Netlink families should have instead defined this as a flat structure, the nesting serves no good purpose here.

Operations

Enum (message ID) model

unified

Modern families use the unified message ID model, which uses a single enumeration for all messages within family. Requests and responses share the same message ID. Notifications have separate IDs from the same space. For example given the following list of operations:

-
  name: a
  value: 1
  do: ...
-
  name: b
  do: ...
-
  name: c
  value: 4
  notify: a
-
  name: d
  do: ...

Requests and responses for operation a will have the ID of 1, the requests and responses of b - 2 (since there is no explicit value it’s previous operation + 1). Notification c will use the ID of 4, operation d 5 etc.

directional

The directional model splits the ID assignment by the direction of the message. Messages from and to the kernel can’t be confused with each other so this conserves the ID space (at the cost of making the programming more cumbersome).

In this case value attribute should be specified in the request reply sections of the operations (if an operation has both do and dump the IDs are shared, value should be set in do). For notifications the value is provided at the op level but it only allocates a reply (i.e. a “from-kernel” ID). Let’s look at an example:

-
  name: a
  do:
    request:
      value: 2
      attributes: ...
    reply:
      value: 1
      attributes: ...
-
  name: b
  notify: a
-
  name: c
  notify: a
  value: 7
-
  name: d
  do: ...

In this case a will use 2 when sending the message to the kernel and expects message with ID 1 in response. Notification b allocates a “from-kernel” ID which is 2. c allocates “from-kernel” ID of 7. If operation d does not set values explicitly in the spec it will be allocated 3 for the request (a is the previous operation with a request section and the value of 2) and 8 for response (c is the previous operation in the “from-kernel” direction).

Other quirks

Structures

Legacy families can define C structures both to be used as the contents of an attribute and as a fixed message header. Structures are defined in definitions and referenced in operations or attributes.

members

• name - The attribute name of the struct member

• type - One of the scalar types u8, u16, u32, u64, s8, s16, s32, s64, string, binary or bitfield32.

• byte-order - big-endian or little-endian

• doc, enum, enum-as-flags, display-hint - Same as for attribute definitions

Note that structures defined in YAML are implicitly packed according to C conventions. For example, the following struct is 4 bytes, not 6 bytes:

struct {
        u8 a;
        u16 b;
        u8 c;
}

Any padding must be explicitly added and C-like languages should infer the need for explicit padding from whether the members are naturally aligned.

Here is the struct definition from above, declared in YAML:

definitions:
  -
    name: message-header
    type: struct
    members:
      -
        name: a
        type: u8
      -
        name: b
        type: u16
      -
        name: c
        type: u8

Fixed Headers

Fixed message headers can be added to operations using fixed-header. The default fixed-header can be set in operations and it can be set or overridden for each operation.

operations:
  fixed-header: message-header
  list:
    -
      name: get
      fixed-header: custom-header
      attribute-set: message-attrs

Attributes

A binary attribute can be interpreted as a C structure using a struct property with the name of the structure definition. The struct property implies sub-type: struct so it is not necessary to specify a sub-type.

attribute-sets:
  -
    name: stats-attrs
    attributes:
      -
        name: stats
        type: binary
        struct: vport-stats

C Arrays

Legacy families also use binary attributes to encapsulate C arrays. The sub-type is used to identify the type of scalar to extract.

attributes:
  -
    name: ports
    type: binary
    sub-type: u32

Multi-message DO

New Netlink families should never respond to a DO operation with multiple replies, with NLM_F_MULTI set. Use a filtered dump instead.

At the spec level we can define a dumps property for the do, perhaps with values of combine and multi-object depending on how the parsing should be implemented (parse into a single reply vs list of objects i.e. pretty much a dump).