freebsd-dev/share/man/man4/netgraph.4

.\" Copyright (c) 1996-1999 Whistle Communications, Inc.
.\" All rights reserved.
.\" 
.\" Subject to the following obligations and disclaimer of warranty, use and
.\" redistribution of this software, in source or object code forms, with or
.\" without modifications are expressly permitted by Whistle Communications;
.\" provided, however, that:
.\" 1. Any and all reproductions of the source or object code must include the
.\"    copyright notice above and the following disclaimer of warranties; and
.\" 2. No rights are granted, in any manner or form, to use Whistle
.\"    Communications, Inc. trademarks, including the mark "WHISTLE
.\"    COMMUNICATIONS" on advertising, endorsements, or otherwise except as
.\"    such appears in the above copyright notice or in the software.
.\" 
.\" THIS SOFTWARE IS BEING PROVIDED BY WHISTLE COMMUNICATIONS "AS IS", AND
.\" TO THE MAXIMUM EXTENT PERMITTED BY LAW, WHISTLE COMMUNICATIONS MAKES NO
.\" REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED, REGARDING THIS SOFTWARE,
.\" INCLUDING WITHOUT LIMITATION, ANY AND ALL IMPLIED WARRANTIES OF
.\" MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT.
.\" WHISTLE COMMUNICATIONS DOES NOT WARRANT, GUARANTEE, OR MAKE ANY
.\" REPRESENTATIONS REGARDING THE USE OF, OR THE RESULTS OF THE USE OF THIS
.\" SOFTWARE IN TERMS OF ITS CORRECTNESS, ACCURACY, RELIABILITY OR OTHERWISE.
.\" IN NO EVENT SHALL WHISTLE COMMUNICATIONS BE LIABLE FOR ANY DAMAGES
.\" RESULTING FROM OR ARISING OUT OF ANY USE OF THIS SOFTWARE, INCLUDING
.\" WITHOUT LIMITATION, ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
.\" PUNITIVE, OR CONSEQUENTIAL DAMAGES, PROCUREMENT OF SUBSTITUTE GOODS OR
.\" SERVICES, LOSS OF USE, DATA OR PROFITS, HOWEVER CAUSED AND UNDER ANY
.\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
.\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
.\" THIS SOFTWARE, EVEN IF WHISTLE COMMUNICATIONS IS ADVISED OF THE POSSIBILITY
.\" OF SUCH DAMAGE.
.\" 
.\" Authors: Julian Elischer <julian@FreeBSD.org>
.\"          Archie Cobbs <archie@FreeBSD.org>
.\"
.\" $FreeBSD$
.\" $Whistle: netgraph.4,v 1.7 1999/01/28 23:54:52 julian Exp $
.\"
.Dd January 19, 1999
.Dt NETGRAPH 4
.Os FreeBSD
.Sh NAME
.Nm netgraph
.Nd graph based kernel networking subsystem
.Sh DESCRIPTION
The
.Nm 
system provides a uniform and modular system for the implementation
of kernel objects which perform various networking functions. The objects,
known as 
.Em nodes ,
can be arranged into arbitrarily complicated graphs. Nodes have
.Em hooks
which are used to connect two nodes together, forming the edges in the graph.
Nodes communicate along the edges to process data, implement protocols, etc.
.Pp
The aim of
.Nm 
is to supplement rather than replace the existing kernel networking
infrastructure.  It provides:
.Pp
.Bl -bullet -compact -offset 2n
.It
A flexible way of combining protocol and link level drivers
.It
A modular way to implement new protocols
.It
A common framework for kernel entities to inter-communicate
.It
A reasonably fast, kernel-based implementation
.El
.Sh Nodes and Types
The most fundamental concept in
.Nm
is that of a
.Em node .
All nodes implement a number of predefined methods which allow them
to interact with other nodes in a well defined manner.
.Pp
Each node has a
.Em type ,
which is a static property of the node determined at node creation time.
A node's type is described by a unique
.Tn ASCII
type name.
The type implies what the node does and how it may be connected
to other nodes.
.Pp
In object-oriented language, types are classes and nodes are instances
of their respective class. All node types are subclasses of the generic node
type, and hence inherit certain common functionality and capabilities
(e.g., the ability to have an
.Tn ASCII
name).
.Pp
Nodes may be assigned a globally unique
.Tn ASCII
name which can be
used to refer to the node.
The name must not contain the characters
.Dq .\&
or
.Dq \&:
and is limited to
.Dv "NG_NODELEN + 1"
characters (including NUL byte).
.Pp
Each node instance has a unique
.Em ID number
which is expressed as a 32-bit hex value. This value may be used to
refer to a node when there is no
.Tn ASCII
name assigned to it.
.Sh Hooks
Nodes are connected to other nodes by connecting a pair of
.Em hooks ,
one from each node. Data flows bidirectionally between nodes along
connected pairs of hooks.  A node may have as many hooks as it
needs, and may assign whatever meaning it wants to a hook.
.Pp
Hooks have these properties:
.Pp
.Bl -bullet -compact -offset 2n
.It
A hook has an
.Tn ASCII
name which is unique among all hooks
on that node (other hooks on other nodes may have the same name).
The name must not contain a
.Dq .\&
or a
.Dq \&:
and is
limited to
.Dv "NG_HOOKLEN + 1"
characters (including NUL byte).
.It
A hook is always connected to another hook. That is, hooks are
created at the time they are connected, and breaking an edge by
removing either hook destroys both hooks.
.It
A hook can be set into a state where incoming packets are always queued
by the input queuing system, rather than being delivered directly. This
is used when the two joined nodes need to be decoupled, e.g. if they are
running at different processor priority levels. (spl)
.El
.Pp
A node may decide to assign special meaning to some hooks. 
For example, connecting to the hook named
.Dq debug
might trigger
the node to start sending debugging information to that hook.
.Sh Data Flow
Two types of information flow between nodes: data messages and
control messages. Data messages are passed in mbuf chains along the edges
in the graph, one edge at a time. The first mbuf in a chain must have the
.Dv M_PKTHDR
flag set. Each node decides how to handle data coming in on its hooks.
.Pp
Control messages are type-specific C structures sent from one node
directly to some arbitrary other node.  Control messages have a common
header format, followed by type-specific data, and are binary structures
for efficiency.  However, node types also may support conversion of the
type specific data between binary and
.Tn ASCII
for debugging and human interface purposes (see the
.Dv NGM_ASCII2BINARY
and
.Dv NGM_BINARY2ASCII
generic control messages below).  Nodes are not required to support
these conversions.
.Pp
There are three ways to address a control message. If
there is a sequence of edges connecting the two nodes, the message
may be
.Dq source routed
by specifying the corresponding sequence
of 
.Tn ASCII
hook names as the destination address for the message (relative
addressing). If the destination is adjacent to the source, then the source
node may simply specify (as a pointer in the code) the hook across which the
message should be sent. Otherwise, the recipient node global
.Tn ASCII
name
(or equivalent ID based name) is used as the destination address
for the message (absolute addressing).  The two types of
.Tn ASCII
addressing
may be combined, by specifying an absolute start node and a sequence
of hooks. Only the 
.Tn ASCII 
addressing modes are available to control programs outside the kernel,
as use of direct pointers is limited of course to kernel modules.
.Pp
Messages often represent commands that are followed by a reply message
in the reverse direction. To facilitate this, the recipient of a
control message is supplied with a
.Dq return address
that is suitable for addressing a reply. 
.Pp
Each control message contains a 32 bit value called a
.Em typecookie
indicating the type of the message, i.e., how to interpret it.
Typically each type defines a unique typecookie for the messages
that it understands.  However, a node may choose to recognize and
implement more than one type of message.
.Pp
If a message is delivered to an address that implies that it arrived
at that node through a particular hook, (as opposed to having been directly
addressed using its ID or global name), then that hook is identified to the
receiving node. This allows a message to be rerouted or passed on, should
a node decide that this is required, in much the same way that data packets
are passed around between nodes. A set of standard 
messages for flow control and link management purposes are
defined by the base system that are usually
passed around in this manner. Flow control message would usually travel
in the opposite direction to the data to which they pertain.
.Sh Netgraph is (usually) Functional
In order to minimize latency, most
.Nm
operations are functional.
That is, data and control messages are delivered by making function
calls rather than by using queues and mailboxes.  For example, if node
A wishes to send a data mbuf to neighboring node B, it calls the
generic
.Nm
data delivery function. This function in turn locates
node B and calls B's
.Dq receive data
method. There are exceptions to this.
.Pp
Each node has an input queue, and some operations can be considered to
be 'writers' in that they alter the state of the node. Obviously in an SMP
world it would be bad if the state of a node were changed while another
data packet were transiting the node. For this purpose, the input queue
implements a 
.Em reader/writer
semantic so that when there is a writer in the node, all other requests
are queued, and while there are readers, a writer, and any following
packets are queued. In the case where there is no reason to queue the
data, the input method is called directly, as mentionned above.
.Pp
A node may declare that all requests should be considered as writers,
or that requests coming in over a particular hook should be considered to 
be a writer, or even that packets leaving or entering across a particular
hook should always be queued, rather than delivered directly (often useful
for interrupt routines who want to get back to the hardware quickly).
By default, all controll message packets are considered to be writers
unless specifically declared to be a reader in their definition. (see
NGM_READONLY in ng_message.h)
.Pp
While this mode of operation
results in good performance, it has a few implications for node
developers:
.Pp
.Bl -bullet -compact -offset 2n
.It
Whenever a node delivers a data or control message, the node
may need to allow for the possibility of receiving a returning
message before the original delivery function call returns.
.It
Netgraph nodes and support routines generally run at
.Fn splnet .
However, some nodes may want to send data and control messages
from a different priority level. Netgraph supplies a mechanism which
utilizes the NETISR system to move message and data delivery to 
.Fn splnet .
Nodes that run at other priorities (e.g. interfaces) can be directly
linked to other nodes so that the combination runs at the other priority,
however any interaction with nodes running at splnet MUST be achieved via the 
queueing functions, (which use the
.Fn netisr 
feature of the kernel).
Note that messages are always received at
.Fn splnet .
.It
It's possible for an infinite loop to occur if the graph contains cycles.
.El
.Pp
So far, these issues have not proven problematical in practice.
.Sh Interaction With Other Parts of the Kernel
A node may have a hidden interaction with other components of the
kernel outside of the
.Nm
subsystem, such as device hardware,
kernel protocol stacks, etc.  In fact, one of the benefits of
.Nm
is the ability to join disparate kernel networking entities together in a
consistent communication framework.
.Pp
An example is the node type
.Em socket 
which is both a netgraph node and a
.Xr socket 2
BSD socket in the protocol family
.Dv PF_NETGRAPH .
Socket nodes allow user processes to participate in
.Nm .
Other nodes communicate with socket nodes using the usual methods, and the
node hides the fact that it is also passing information to and from a
cooperating user process.
.Pp
Another example is a device driver that presents
a node interface to the hardware.
.Sh Node Methods
Nodes are notified of the following actions via function calls
to the following node methods (all at
.Fn splnet )
and may accept or reject that action (by returning the appropriate
error code):
.Bl -tag -width xxx
.It Creation of a new node
The constructor for the type is called. If creation of a new node is 
allowed, the constructor must call the generic node creation
function (in object-oriented terms, the superclass constructor)
and then allocate any special resources it needs. For nodes that
correspond to hardware, this is typically done during the device
attach routine. Often a global
.Tn ASCII
name corresponding to the
device name is assigned here as well.
.It Creation of a new hook
The hook is created and tentatively
linked to the node, and the node is told about the name that will be 
used to describe this hook. The node sets up any special data structures
it needs, or may reject the connection, based on the name of the hook.
.It Successful connection of two hooks
After both ends have accepted their
hooks, and the links have been made, the nodes get a chance to
find out who their peer is across the link and can then decide to reject
the connection. Tear-down is automatic. This is also the time at which
a node may decide whether to set a particular hook (or its peer) into 
.Em queuing
mode.
.It Destruction of a hook
The node is notified of a broken connection. The node may consider some hooks
to be critical to operation and others to be expendable: the disconnection
of one hook may be an acceptable event while for another it
may effect a total shutdown for the node.
.It Shutdown of a node
This method allows a node to clean up
and to ensure that any actions that need to be performed
at this time are taken. The method must call the generic (i.e., superclass)
node destructor to get rid of the generic components of the node.
Some nodes (usually associated with a piece of hardware) may be
.Em persistent
in that a shutdown breaks all edges and resets the node,
but doesn't remove it, in which case the generic destructor is not called.
.El
.Sh Sending and Receiving Data
Two other methods are also supported by all nodes:
.Bl -tag -width xxx
.It Receive data message
A
.Em Netgraph queueable reqest item
(usually refered to as an 
.Em item
is recieved by the function.
The item contains a pointer to an mbuf and metadata about the packet.
.Pp
The node is notified on which hook the item arrived,
and can use this information in its processing decision.
The receiving node must always 
.Fn NG_FREE_M
the mbuf chain on completion or error, or pass it on to another node
(or kernel module) which will then be responsible for freeing it.
Similarly the 
.Em item
must be freed if it is not to be passed on to another node, by using the
.Fn NG_FREE_ITEM
macro. If the item still holds references to mbufs or metadata at the time of
freeing then they will also be appropriatly freed.
Therefore, if there is any chance that the mbuf or metadata will be
changed or freed separatly from the item, it is very important
that these fields be retrieved using the 
.Fn NGI_GET_M
and
.Fn NGI_GET_META
macros that also remove the reference within the item. (or multiple frees
of the same object will occur).
.Pp
If it is only required to examine the contents of the mbufs or the 
metadata, then it is possible to use the
.Fn NGI_M
and
.Fn NGI_META
macros to both read and rewrite these fields.
.Pp
In addition to the mbuf chain itself there may also be a pointer to a 
structure describing meta-data about the message
(e.g. priority information). This pointer may be
.Dv NULL
if there is no additional information. The format for this information is
described in 
.Pa sys/netgraph/netgraph.h .
The memory for meta-data must allocated via
.Fn malloc
with type
.Dv M_NETGRAPH_META .
As with the data itself, it is the receiver's responsibility to
.Fn free
the meta-data. If the mbuf chain is freed the meta-data must
be freed at the same time. If the meta-data is freed but the
real data on is passed on, then a
.Dv NULL
pointer must be substituted. It is also the duty of the receiver to free
the request item itself, or to use it to pass the message on further.
.Pp
The receiving node may decide to defer the data by queueing it in the
.Nm
NETISR system (see below). It achieves this by setting the 
.Dv HK_QUEUE
flag in the flags word of the hook on which that data will arrive.
The infrastructure will respect that bit and queue the data for delivery at 
a later time, rather than deliver it directly. A node may decide to set
the bit on the
.Em peer
node, so that it's own output packets are queued. This is used
by device drivers running at different processor priorities to transfer
packet delivery to the splnet() level at which the bulk of
.Nm
runs.
.Pp
The structure and use of meta-data is still experimental, but is
presently used in frame-relay to indicate that management packets
should be queued for transmission
at a higher priority than data packets. This is required for
conformance with Frame Relay standards.
.It Receive control message
This method is called when a control message is addressed to the node.
As with the received data, an
.Em item
is reveived, with a pointer to the control message.
The message can be examined using the 
.Fn NGI_MSG
macro, or completely extracted from the item using the
.Fn NGI_GET_MSG
which also removes the reference within the item.
If the Item still holds a reference to the message when it is freed
(using the 
.Fn NG_FREE_ITEM
macro), then the message will also be freed appropriatly. If the 
reference has been removed the node must free the message itself using the
.Fn NG_FREE_MSG
macro.
A return address is always supplied, giving the address of the node
that originated the message so a reply message can be sent anytime later.
The return address is retrieved from the
.Em item
using the
.Fn NGI_RETADDR
macro and is of type 
.Em ng_ID_t.
All control messages and replies are
allocated with
.Fn malloc
type
.Dv M_NETGRAPH_MSG ,
however it is more usual to use the
.Fn NG_MKMESSAGE
and
.Fn NG_MKRESPONSE
macros to allocate and fill out a message.
Messages must be freed using the 
.Fn NG_FREE_MSG
macro.
.Pp
If the message was delivered via a specific hook, that hook will
also be made known, which allows the use of such things as flow-control
messages, and status change messages, where the node may want to forward
the message out another hook to that on which it arrived.
.El
.Pp
Much use has been made of reference counts, so that nodes being
free'd of all references are automatically freed, and this behaviour
has been tested and debugged to present a consistent and trustworthy
framework for the
.Dq type module
writer to use.
.Sh Addressing
The 
.Nm
framework provides an unambiguous and simple to use method of specifically
addressing any single node in the graph. The naming of a node is 
independent of its type, in that another node, or external component
need not know anything about the node's type in order to address it so as 
to send it a generic message type. Node and hook names should be
chosen so as to make addresses meaningful.
.Pp
Addresses are either absolute or relative. An absolute address begins
with a node name, (or ID), followed by a colon, followed by a sequence of hook
names separated by periods. This addresses the node reached by starting
at the named node and following the specified sequence of hooks.
A relative address includes only the sequence of hook names, implicitly
starting hook traversal at the local node.
.Pp
There are a couple of special possibilities for the node name.
The name
.Dq .\&
(referred to as
.Dq \&.: )
always refers to the local node.
Also, nodes that have no global name may be addressed by their ID numbers,
by enclosing the hex representation of the ID number within square brackets.
Here are some examples of valid netgraph addresses:
.Bd -literal -offset 4n -compact

  .:
  [3f]:
  foo:
  .:hook1
  foo:hook1.hook2
  [d80]:hook1
.Ed
.Pp
Consider the following set of nodes might be created for a site with
a single physical frame relay line having two active logical DLCI channels,
with RFC-1490 frames on DLCI 16 and PPP frames over DLCI 20:
.Pp
.Bd -literal
[type SYNC ]                  [type FRAME]                 [type RFC1490]
[ "Frame1" ](uplink)<-->(data)[<un-named>](dlci16)<-->(mux)[<un-named>  ]
[    A     ]                  [    B     ](dlci20)<---+    [     C      ]
                                                      |
                                                      |      [ type PPP ]
                                                      +>(mux)[<un-named>]
                                                             [    D     ]
.Ed
.Pp
One could always send a control message to node C from anywhere
by using the name
.Em "Frame1:uplink.dlci16" .
In this case, node C would also be notified that the message
reached it via its hook 
.Dq mux .
Similarly, 
.Em "Frame1:uplink.dlci20"
could reliably be used to reach node D, and node A could refer
to node B as
.Em ".:uplink" ,
or simply
.Em "uplink" .
Conversely, B can refer to A as
.Em "data" .
The address
.Em "mux.data"
could be used by both nodes C and D to address a message to node A.
.Pp
Note that this is only for
.Em control messages .
In each of these cases, where a relative addressing mode is
used, the recipient is notified of the hook on which the
message arrived, as well as 
the originating node.
This allows the option of hop-by-hop distibution of messages and
state information.
Data messages are 
.Em only
routed one hop at a time, by specifying the departing
hook, with each node making
the next routing decision. So when B receives a frame on hook
.Dq data
it decodes the frame relay header to determine the DLCI,
and then forwards the unwrapped frame to either C or D.
.Pp
In a similar way, flow control messages may be routed in the reverse
direction to outgoing data. For example a "buffer nearly full" message from 
.Em "Frame1:
would be passed to node 
.Em B
which might decide to send similar messages to both nodes
.Em C
and 
.Em D .
The nodes would use 
.Em "Direct hook pointer"
addressing to route the messages. The message may have travelled from
.Em "Frame1:
to
.Em B
as a synchronous reply, saving time and cycles.
.Pp
A similar graph might be used to represent multi-link PPP running
over an ISDN line:
.Pp
.Bd -literal
[ type BRI ](B1)<--->(link1)[ type MPP  ]
[  "ISDN1" ](B2)<--->(link2)[ (no name) ]
[          ](D) <-+
                  |
 +----------------+
 |
 +->(switch)[ type Q.921 ](term1)<---->(datalink)[ type Q.931 ]
            [ (no name)  ]                       [ (no name)  ]
.Ed
.Sh Netgraph Structures
Structures are defined in 
.Pa sys/netgraph/netgraph.h 
(for kernel sructures only of interest to nodes)
and
.Pa sys/netgraph/ng_message.h 
(for message definitions also of interest to user programs).
.Pp
The two basic object types that are of interest to node authors are 
.Em nodes
and
.Em hooks .
These two objects have the following
properties that are also of interest to the node writers.
.Bl -tag -width xxx
.It struct  ng_node
Node authors should always use the following typedef to declare
their pointers, and should never actually declare the structure.
.Pp
typedef struct ng_node *node_p;
.Pp
The following properties are associated with a node, and can be
accessed in the following manner:
.Bl -bullet -compact -offset 2n
.Pp
.It
Validity
.Pp
A driver or interrupt routine may want to check whether
the node is still valid. It is assumed that the caller holds a reference
on the node so it will not have been freed, however it may have been
disabled or otherwise shut down. Using the
.Fn NG_NODE_IS_VALID "node"
macro will return this state. Eventually it should be almost impossible
for code to run in an invalid node but at this time that work has not been
completed.
.Pp
.It
node ID
.Pp
Of type 
.Em ng_ID_t ,
This property can be retrieved using the macro 
.Fn NG_NODE_ID "node".
.Pp
.It
node name
.Pp
Optional globally unique name, null terminated string. If there
is a value in here, it is the name of the node.
.Pp
if (
.Fn NG_NODE_NAME "node"
[0]) ....
.Pp
if (strncmp(
.Fn NG_NODE_NAME "node"
, "fred", NG_NODELEN)) ...
.Pp
.It
A node dependent opaque cookie
.Pp
You may place anything of type 
.Em pointer
here.
Use the macros
.Fn NG_NODE_SET_PRIVATE "node, value"
and
.Fn NG_NODE_PRIVATE "node"
to set and retrieve this property.
.Pp
.It
number of hooks
.Pp
Use
.Fn NG_NODE_NUMHOOKS "node" 
to retrieve this value.
.Pp
.It
hooks
.Pp
The node may have a number of hooks.
A traversal method is provided to allow all the hooks to be
tested for some condition.
.Fn NG_NODE_FOREACH_HOOK "node, fn, arg, rethook"
where fn is a function that will be called for each hook
with the form
.Fn fn "hook, arg"
and returning 0 to terminate the search. If the search is terminated, then 
.Em rethook
will be set to the hook at which the search was terminated.
.El
.It struct  ng_hook 
Node authors should always use the following typedef to declare
their hook pointers.
.Pp
typedef struct ng_hook *hook_p;
.Pp
The following properties are associated with a hook, and can be
accessed in the following manner:
.Bl -bullet -compact -offset 2n
.Pp
.It
A node dependent opaque cookie.
.Pp
You may place anything of type 
.Em pointer
here.
Use the macros
.Fn NG_HOOK_SET_PRIVATE "hook, value"
and
.Fn NG_HOOK_PRIVATE "hook"
to set and retrieve this property.
.Pp
.It
An associate node.
.Pp
You may use the macro 
.Fn NG_HOOK_NODE "hook"
to find the associated node.
.Pp
.It
A peer hook
.Pp
The other hook in this connected pair. Of type hook_p. You can
use
.Fn NG_HOOK_PEER "hook"
to find the peer.
.Pp
.It
references
.Pp
.Fn NG_HOOK_REF "hook"
and
.Fn NG_HOOK_UNREF "hook"
increment and decrement the hook reference count accordingly.
After decrement you should always sume the hook has been freed.
In fact the macro may set it to NULL.
.El
.Pp
The maintenance of the names, reference counts, and linked list
of hooks for each node is handled automatically by the
.Nm
subsystem.
Typically a node's private info contains a back-pointer to the node or hook
structure, which counts as a new reference that must be included
in the reference count for the node. When the node constructor is called
there is already a reference for this calculated in, so that
when the node is destroyed, it should remember to do a
.Fn NG_NODE_UNREF
on the node.
.Pp
From a hook you can obtain the corresponding node, and from
a node, it is possible to traverse all the active hooks.
.Pp
A current example of how to define a node can always be seen in 
.Em sys/netgraph/ng_sample.c
and should be used as a starting point for new node writers.
.El
.Sh Netgraph Message Structure
Control messages have the following structure:
.Bd -literal
#define NG_CMDSTRLEN    15      /* Max command string (16 with null) */

struct ng_mesg {
  struct ng_msghdr {
    u_char      version;        /* Must equal NG_VERSION */
    u_char      spare;          /* Pad to 2 bytes */
    u_short     arglen;         /* Length of cmd/resp data */
    u_long      flags;          /* Message status flags */
    u_long      token;          /* Reply should have the same token */
    u_long      typecookie;     /* Node type understanding this message */
    u_long      cmd;            /* Command identifier */
    u_char      cmdstr[NG_CMDSTRLEN+1]; /* Cmd string (for debug) */
  } header;
  char  data[0];                /* Start of cmd/resp data */
};

#define NG_ABI_VERSION  5               /* Netgraph kernel ABI version */
#define NG_VERSION      4               /* Netgraph message version */
#define NGF_ORIG        0x0000          /* Command */
#define NGF_RESP        0x0001          /* Response */
.Ed
.Pp
Control messages have the fixed header shown above, followed by a 
variable length data section which depends on the type cookie
and the command. Each field is explained below:
.Bl -tag -width xxx
.It Dv version
Indicates the version of the netgraph message protocol itself. The current version is
.Dv NG_VERSION .
.It Dv arglen
This is the length of any extra arguments, which begin at
.Dv data .
.It Dv flags
Indicates whether this is a command or a response control message.
.It Dv token
The
.Dv token
is a means by which a sender can match a reply message to the
corresponding command message; the reply always has the same token.
.Pp
.It Dv typecookie
The corresponding node type's unique 32-bit value.
If a node doesn't recognize the type cookie it must reject the message
by returning
.Er EINVAL .
.Pp
Each type should have an include file that defines the commands,
argument format, and cookie for its own messages.
The typecookie
insures that the same header file was included by both sender and
receiver; when an incompatible change in the header file is made,
the typecookie
.Em must
be changed.
The de facto method for generating unique type cookies is to take the
seconds from the epoch at the time the header file is written
(i.e., the output of
.Dv "date -u +'%s'" ) .
.Pp
There is a predefined typecookie
.Dv NGM_GENERIC_COOKIE
for the
.Dq generic
node type, and
a corresponding set of generic messages which all nodes understand.
The handling of these messages is automatic.
.It Dv command
The identifier for the message command. This is type specific,
and is defined in the same header file as the typecookie.
.It Dv cmdstr
Room for a short human readable version of
.Dq command
(for debugging purposes only).
.El
.Pp
Some modules may choose to implement messages from more than one 
of the header files and thus recognize more than one type cookie. 
.Sh Control Message ASCII Form
Control messages are in binary format for efficiency.  However, for
debugging and human interface purposes, and if the node type supports
it, control messages may be converted to and from an equivalent
.Tn ASCII
form.  The
.Tn ASCII
form is similar to the binary form, with two exceptions:
.Pp
.Bl -tag -compact -width xxx
.It o
The
.Dv cmdstr
header field must contain the
.Tn ASCII
name of the command, corresponding to the
.Dv cmd
header field.
.It o
The
.Dv args
field contains a NUL-terminated
.Tn ASCII
string version of the message arguments.
.El
.Pp
In general, the arguments field of a control messgage can be any
arbitrary C data type.  Netgraph includes parsing routines to support
some pre-defined datatypes in
.Tn ASCII
with this simple syntax:
.Pp
.Bl -tag -compact -width xxx
.It o
Integer types are represented by base 8, 10, or 16 numbers.
.It o
Strings are enclosed in double quotes and respect the normal
C language backslash escapes.
.It o
IP addresses have the obvious form.
.It o
Arrays are enclosed in square brackets, with the elements listed
consecutively starting at index zero.  An element may have an optional
index and equals sign preceeding it.  Whenever an element
does not have an explicit index, the index is implicitly the previous
element's index plus one.
.It o
Structures are enclosed in curly braces, and each field is specified
in the form 
.Dq fieldname=value .
.It o
Any array element or structure field whose value is equal to its
.Dq default value
may be omitted. For integer types, the default value
is usually zero; for string types, the empty string.
.It o
Array elements and structure fields may be specified in any order.
.El
.Pp
Each node type may define its own arbitrary types by providing
the necessary routines to parse and unparse.
.Tn ASCII
forms defined
for a specific node type are documented in the documentation for
that node type.
.Sh Generic Control Messages
There are a number of standard predefined messages that will work
for any node, as they are supported directly by the framework itself.
These are defined in
.Pa ng_message.h
along with the basic layout of messages and other similar information.
.Bl -tag -width xxx
.It Dv NGM_CONNECT
Connect to another node, using the supplied hook names on either end.
.It Dv NGM_MKPEER
Construct a node of the given type and then connect to it using the
supplied hook names.
.It Dv NGM_SHUTDOWN
The target node should disconnect from all its neighbours and shut down.
Persistent nodes such as those representing physical hardware
might not disappear from the node namespace, but only reset themselves.
The node must disconnect all of its hooks.
This may result in neighbors shutting themselves down, and possibly a
cascading shutdown of the entire connected graph.
.It Dv NGM_NAME
Assign a name to a node. Nodes can exist without having a name, and this
is the default for nodes created using the
.Dv NGM_MKPEER
method. Such nodes can only be addressed relatively or by their ID number.
.It Dv NGM_RMHOOK
Ask the node to break a hook connection to one of its neighbours.
Both nodes will have their
.Dq disconnect
method invoked.
Either node may elect to totally shut down as a result.
.It Dv NGM_NODEINFO
Asks the target node to describe itself. The four returned fields
are the node name (if named), the node type, the node ID and the
number of hooks attached. The ID is an internal number unique to that node.
.It Dv NGM_LISTHOOKS
This returns the information given by
.Dv NGM_NODEINFO ,
but in addition 
includes an array of fields describing each link, and the description for
the node at the far end of that link.
.It Dv NGM_LISTNAMES
This returns an array of node descriptions (as for
.Dv NGM_NODEINFO ")"
where each entry of the array describes a named node.
All named nodes will be described.
.It Dv NGM_LISTNODES
This is the same as
.Dv NGM_LISTNAMES
except that all nodes are listed regardless of whether they have a name or not.
.It Dv NGM_LISTTYPES
This returns a list of all currently installed netgraph types.
.It Dv NGM_TEXT_STATUS
The node may return a text formatted status message.
The status information is determined entirely by the node type.
It is the only "generic" message
that requires any support within the node itself and as such the node may
elect to not support this message. The text response must be less than
.Dv NG_TEXTRESPONSE
bytes in length (presently 1024). This can be used to return general
status information in human readable form.
.It Dv NGM_BINARY2ASCII
This message converts a binary control message to its
.Tn ASCII
form.
The entire control message to be converted is contained within the
arguments field of the
.Dv Dv NGM_BINARY2ASCII
message itself.  If successful, the reply will contain the same control
message in
.Tn ASCII
form.
A node will typically only know how to translate messages that it
itself understands, so the target node of the
.Dv NGM_BINARY2ASCII
is often the same node that would actually receive that message.
.It Dv NGM_ASCII2BINARY
The opposite of
.Dv NGM_BINARY2ASCII .
The entire control message to be converted, in
.Tn ASCII
form, is contained
in the arguments section of the
.Dv NGM_ASCII2BINARY
and need only have the
.Dv flags ,
.Dv cmdstr ,
and
.Dv arglen
header fields filled in, plus the NUL-terminated string version of
the arguments in the arguments field.  If successful, the reply
contains the binary version of the control message.
.El
.Sh Flow Control Messages
In addition to the control messages that affect nodes with respect to the
graph, there are also a number of 
.Em Flow-control
messages defined. At present these are
.Em NOT
handled automatically by the system, so
nodes need to handle them if they are going to be used in a graph utilising
flow control, and will be in the likely path of these messages. The
default action of a node that doesn't understand these messages should
be to pass them onto the next node. Hopefully some helper functions 
will assist in this eventually. These messages are also defined in
.Pa sys/netgraph/ng_message.h
and have a separate cookie
.Em NG_FLOW_COOKIE
to help identify them. They will not be covered in depth here.
.Sh Metadata
Data moving through the
.Nm
system can be accompanied by meta-data that describes some
aspect of that data. The form of the meta-data is a fixed header,
which contains enough information for most uses, and can optionally 
be supplemented by trailing
.Em option
structures, which contain a 
.Em cookie
(see the section on control messages), an identifier, a length and optional
data. If a node does not recognize the cookie associated with an option,
it should ignore that option.
.Pp
Meta data might include such things as priority, discard eligibility,
or special processing requirements. It might also mark a packet for
debug status, etc. The use of meta-data is still experimental.
.Sh INITIALIZATION
The base
.Nm
code may either be statically compiled
into the kernel or else loaded dynamically as a KLD via
.Xr kldload 8 .
In the former case, include
.Pp
.Dl options NETGRAPH
.Pp
in your kernel configuration file. You may also include selected
node types in the kernel compilation, for example:
.Bd -literal -offset indent
options NETGRAPH
options NETGRAPH_SOCKET
options NETGRAPH_ECHO
.Ed
.Pp
Once the
.Nm
subsystem is loaded, individual node types may be loaded at any time
as KLD modules via
.Xr kldload 8 .
Moreover,
.Nm
knows how to automatically do this; when a request to create a new
node of unknown type
.Em type
is made,
.Nm
will attempt to load the KLD module
.Pa ng_type.ko .
.Pp
Types can also be installed at boot time, as certain device drivers
may want to export each instance of the device as a netgraph node.
.Pp
In general, new types can be installed at any time from within the
kernel by calling
.Fn ng_newtype ,
supplying a pointer to the type's
.Dv struct ng_type
structure.
.Pp
The
.Fn NETGRAPH_INIT
macro automates this process by using a linker set.
.Sh EXISTING NODE TYPES
Several node types currently exist. Each is fully documented
in its own man page:
.Bl -tag -width xxx
.It SOCKET
The socket type implements two new sockets in the new protocol domain
.Dv PF_NETGRAPH .
The new sockets protocols are
.Dv NG_DATA
and
.Dv NG_CONTROL ,
both of type
.Dv SOCK_DGRAM .
Typically one of each is associated with a socket node.
When both sockets have closed, the node will shut down. The
.Dv NG_DATA
socket is used for sending and receiving data, while the
.Dv NG_CONTROL
socket is used for sending and receiving control messages.
Data and control messages are passed using the
.Xr sendto 2
and
.Xr recvfrom 2
calls, using a
.Dv struct sockaddr_ng
socket address.
.Pp
.It HOLE
Responds only to generic messages and is a
.Dq black hole
for data, Useful for testing. Always accepts new hooks.
.Pp
.It ECHO
Responds only to generic messages and always echoes data back through the
hook from which it arrived. Returns any non generic messages as their
own response. Useful for testing.  Always accepts new hooks.
.Pp
.It TEE
This node is useful for
.Dq snooping .
It has 4 hooks:
.Dv left ,
.Dv right ,
.Dv left2right ,
and
.Dv right2left .
Data entering from the right is passed to the left and duplicated on
.Dv right2left,
and data entering from the left is passed to the right and
duplicated on
.Dv left2right .
Data entering from
.Dv left2right
is sent to the right and data from
.Dv right2left
to left. 
.Pp
.It RFC1490 MUX
Encapsulates/de-encapsulates frames encoded according to RFC 1490.
Has a hook for the encapsulated packets
.Pq Dq downstream
and one hook
for each protocol (i.e., IP, PPP, etc.).
.Pp
.It FRAME RELAY MUX
Encapsulates/de-encapsulates Frame Relay frames.
Has a hook for the encapsulated packets
.Pq Dq downstream
and one hook
for each DLCI.
.Pp
.It FRAME RELAY LMI
Automatically handles frame relay
.Dq LMI
(link management interface) operations and packets.
Automatically probes and detects which of several LMI standards
is in use at the exchange.
.Pp
.It TTY
This node is also a line discipline. It simply converts between mbuf
frames and sequential serial data, allowing a tty to appear as a netgraph
node. It has a programmable
.Dq hotkey
character.
.Pp
.It ASYNC
This node encapsulates and de-encapsulates asynchronous frames
according to RFC 1662. This is used in conjunction with the TTY node
type for supporting PPP links over asynchronous serial lines.
.Pp
.It INTERFACE
This node is also a system networking interface. It has hooks representing
each protocol family (IP, AppleTalk, IPX, etc.) and appears in the output of
.Xr ifconfig 8 .
The interfaces are named
.Em ng0 ,
.Em ng1 ,
etc.
.It ONE2MANY
This node implements a simple round-robin multiplexer. It can be used
for example to make several LAN ports act together to get a higher speed
link between two machines.
.It Various PPP related nodes.
There is a full multilink PPP implementation that runs in Netgraph.
The 
.Em Mpd
port can use these modules to make a very low latency high 
capacity ppp system. It also supports 
.Em PPTP
vpns using the
.Em PPTP
node.
.It PPPOE
A server and client side implememtation of PPPoE. Used in conjunction with
either 
.Xr ppp 8
or the 
.Em mpd port.
.It BRIDGE
This node, togther with the ethernet nodes allows a very flexible
bridging system to be implemented.
.It KSOCKET
This intriguing node looks like a socket to the system but diverts
all data to and from the netgraph system for further processing. This allows
such things as UDP tunnels to be almost trivially implemented from the 
command line.
.El
.Pp
Refer to the section at the end of this man page for more nodes types.
.Sh NOTES
Whether a named node exists can be checked by trying to send a control message
to it (e.g.,
.Dv NGM_NODEINFO
).
If it does not exist,
.Er ENOENT
will be returned.
.Pp
All data messages are mbuf chains with the M_PKTHDR flag set.
.Pp
Nodes are responsible for freeing what they allocate.
There are three exceptions:
.Bl -tag -width xxxx
.It 1
Mbufs sent across a data link are never to be freed by the sender. In the
case of error, they should be considered freed.
.It 2
Any meta-data information traveling with the data has the same restriction.
It might be freed by any node the data passes through, and a
.Dv NULL
passed onwards, but the caller will never free it.
Two macros
.Fn NG_FREE_META "meta"
and
.Fn NG_FREE_M "m"
should be used if possible to free data and meta data (see
.Pa netgraph.h ) .
.It 3
Messages sent using
.Fn ng_send_message
are freed by the recipient. As in the case above, the addresses
associated with the message are freed by whatever allocated them so the 
recipient should copy them if it wants to keep that information.
.It 4
Both control mesages and data are delivered and queued with
a netgraph 
.Em item .
The item must be freed using 
.Fn NG_FREE_ITEM "item"
or passed on to another node.
.El
.Sh FILES
.Bl -tag -width xxxxx -compact
.It Pa /sys/netgraph/netgraph.h
Definitions for use solely within the kernel by
.Nm
nodes.
.It Pa /sys/netgraph/ng_message.h
Definitions needed by any file that needs to deal with 
.Nm 
messages.
.It Pa /sys/netgraph/ng_socket.h
Definitions needed to use 
.Nm
socket type nodes.
.It Pa /sys/netgraph/ng_{type}.h
Definitions needed to use 
.Nm
{type}
nodes, including the type cookie definition.
.It Pa /modules/netgraph.ko
Netgraph subsystem loadable KLD module.
.It Pa /modules/ng_{type}.ko
Loadable KLD module for node type {type}.
.It Pa /sys/netgraph/ng_sample.c
 Skeleton netgraph node. Use this as a starting point for new node types.
.El
.Sh USER MODE SUPPORT
There is a library for supporting user-mode programs that wish
to interact with the netgraph system. See
.Xr netgraph 3
for details.
.Pp
Two user-mode support programs,
.Xr ngctl 8
and
.Xr nghook 8 ,
are available to assist manual configuration and debugging.
.Pp
There are a few useful techniques for debugging new node types.
First, implementing new node types in user-mode first
makes debugging easier.
The
.Em tee
node type is also useful for debugging, especially in conjunction with
.Xr ngctl 8
and
.Xr nghook 8 .
.Pp
Also look in /usr/share/examples/netgraph for solutions to several
common networking problems, solved using 
.Nm .
.Sh SEE ALSO
.Xr socket 2 ,
.Xr netgraph 3 ,
.Xr ng_async 4 ,
.Xr ng_bridge 4 ,
.Xr ng_bpf 4 ,
.Xr ng_cisco 4 ,
.Xr ng_ether 4 ,
.Xr ng_echo 4 ,
.Xr ng_ether 4 ,
.Xr ng_frame_relay 4 ,
.Xr ng_hole 4 ,
.Xr ng_iface 4 ,
.Xr ng_ksocket 4 ,
.Xr ng_lmi 4 ,
.Xr ng_mppc 4 ,
.Xr ng_ppp 4 ,
.Xr ng_pppoe 4 ,
.Xr ng_pptpgre 4 ,
.Xr ng_rfc1490 4 ,
.Xr ng_socket 4 ,
.Xr ng_tee 4 ,
.Xr ng_tty 4 ,
.Xr ng_UI 4 ,
.Xr ng_vjc 4 ,
.Xr ng_{type} 4 ,
.Xr ngctl 8 ,
.Xr nghook 8
.Sh HISTORY
The
.Nm
system was designed and first implemented at Whistle Communications, Inc.
in a version of
.Fx 2.2
customized for the Whistle InterJet.
It first made its debut in the main tree in
.Fx 3.4 .
.Sh AUTHORS
.An -nosplit
.An Julian Elischer Aq julian@FreeBSD.org ,
with contributions by
.An Archie Cobbs Aq archie@FreeBSD.org .