netinet/libalias/libalias.3

.Dd July, 1997
.Dt "libalias" 3
.Os
.Sh NAME
.Nm "libalias"
Packet Aliasing Library.  A collection of
functions for aliasing and de-aliasing
of IP packets, intended for masquerading and
network address translation (NAT).

.Sh SYNOPSIS
.Fd #include <sys/types.h>
.Fd #include <netinet/in.h>
.Fd #include <alias.h>

Function prototypes are given in the main body
of the text.

.Sh CONTENTS
.Bd -literal -offset left
1. Introduction
2. Initialization and Control
    2.1 PacketAliasInit()
    2.2 PacketAliasUninit()
    2.3 PacketAliasSetAddress()
    2.4 PacketAliasSetMode()
    2.5 PacketAliasSetFWBase()
3. Packet Handling
    3.1 PacketAliasIn()
    3.2 PacketAliasOut()
4. Port and Address Redirection
    4.1 PacketAliasRedirectPort()
    4.2 PacketAliasRedirectAddr()
    4.3 PacketAliasRedirectDelete()
    4.4 PacketAliasProxyRule()
5. Fragment Handling
    5.1 PacketAliasSaveFragment()
    5.2 PacketAliasGetFragment()
    5.3 PacketAliasFragmentIn()
6. Miscellaneous Functions
    6.1 PacketAliasSetTarget()
    6.2 PacketAliasCheckNewLink()
    6.3 PacketAliasInternetChecksum()
7. Authors
8. Acknowledgments

Appendix A: Conceptual Background
    A.1 Aliasing Links
    A.2 Static and Dynamic Links
    A.3 Partially Specified Links
    A.4 Dynamic Link Creation
.Ed

.Sh 1. Introduction
This library is a moderately portable
set of functions designed to assist
in the process of IP masquerading and
network address translation.  Outgoing
packets from a local network with
unregistered IP addresses can be aliased
to appear as if they came from an
accessible IP address.  Incoming packets
are then de-aliased so that they are sent
to the correct machine on the local network.

A certain amount of flexibility is built
into the packet aliasing engine.  In
the simplest mode of operation, a
many-to-one address mapping takes place
between local network and the packet
aliasing host.  This is known as IP
masquerading.  In addition, one-to-one
mappings between local and public addresses
can also be implemented, which is known as
static NAT.  In between these extremes,
different groups of private addresses
can be linked to different public addresses,
comprising several distinct many-to-one
mappings.  Also, a given public address
and port can be statically redirected to
a private address/port.

The packet aliasing engine was designed
to operate in user space outside of the
kernel, without any access to private
kernel data structure, but the source code
can also be ported to a kernel environment.

.Sh 2. Initialization and Control
Two specific functions, PacketAliasInit()
and PacketAliasSetAddress(), must always be
called before any packet handling may be
performed.  In addition, the operating mode
of the packet aliasing engine can be customized
by calling PacketAliasSetMode().
.Ss 2.1 PacketAliasInit()

.Ft void
.Fn PacketAliasInit "void"

This function has no argument or return
value and is used to initialize internal
data structures. The following mode bits
are always set after calling
PacketAliasInit().  See section 2.3 for
the meaning of these mode bits.
.Bd -literal -offset indent
    PKT_ALIAS_USE_SAME_PORTS
    PKT_ALIAS_USE_SOCKETS
    PKT_ALIAS_RESET_ON_ADDR_CHANGE

.Ed
This function will always return the packet
aliasing engine to the same initial state.
PacketAliasSetAddress() must be called afterwards,
and any desired changes from the default mode
bits listed above require a call to
PacketAliasSetMode().

It is mandatory that this function be called
at the beginning of a program prior to any
packet handling.
.Ss 2.2 PacketAliasUninit()

.Ft void
.Fn PacketAliasUninit "void"

This function has no argument or return
value and is used to clear any resources
attached to internal data structures.

This functions should be called when a
program stop using the aliasing engine;
it do, among other things, clear out any
firewall holes.  To provide backwards
compatibility and extra security, it is
added to the atexit() chain by
PacketAliasInit().  Calling it multiple
times is harmless.
.Ss 2.3 PacketAliasSetAddress()

.Ft void
.Fn PacketAliasSetAddress "struct in_addr addr"

This function sets the source address to which
outgoing packets from the local area network
are aliased.  All outgoing packets are remapped
to this address unless overridden by a static
address mapping established by
PacketAliasRedirectAddr().

If the PKT_ALIAS_RESET_ON_ADDR_CHANGE mode bit
is set (the default mode of operation), then
the internal aliasing link tables will be reset
any time the aliasing address changes, as if
PacketAliasReset() were called.  This is useful
for interfaces such as ppp where the IP
address may or may not change on successive
dial-up attempts.

If the PKT_ALIAS_RESET_ON_ADDR_CHANGE mode bit
is set to zero, this function can also be used to
dynamically change the aliasing address on a
packet to packet basis (it is a low overhead
call).

It is mandatory that this function be called
prior to any packet handling.
.Ss 2.4 PacketAliasSetMode()

.Ft unsigned int
.Fn PacketAliasSetMode "unsigned int mode" "unsigned int mask"

This function sets or clears mode bits
according to the value of
.Em mode .
Only bits marked in
.Em mask
are affected.  The following mode bits are
defined in alias.h:
.Bl -hang -offset left
.It PKT_ALIAS_LOG.
Enables logging /var/log/alias.log.  The log file
shows total numbers of links (icmp, tcp, udp) each
time an aliasing link is created or deleted.  Mainly
useful for debugging when the log file is viewed
continuously with "tail -f".
.It PKT_ALIAS_DENY_INCOMING.
If this mode bit is set, all incoming packets
associated with new TCP connections or new
UDP transactions will be marked for being
ignored (PacketAliasIn() return code
PKT_ALIAS_IGNORED) by the calling program.
Response packets to connections or transactions
initiated from the packet aliasing host or
local network will be unaffected.  This mode
bit is useful for implementing a one-way firewall.
.It PKT_ALIAS_SAME_PORTS.
If this mode bit is set, the packet aliasing
engine will attempt to leave the alias port
numbers unchanged from the actual local port
number.  This can be done as long as the
quintuple (proto, alias addr, alias port,
remote addr, remote port) is unique.  If a
conflict exists, an new aliasing port number is
chosen even if this mode bit is set.
.It PKT_ALIAS_USE_SOCKETS.
This bit should be set when the the packet
aliasing host originates network traffic as
well as forwards it.  When the packet aliasing
host is waiting for a connection from an
unknown host address or unknown port number
(e.g. an FTP data connection), this mode bit
specifies that a socket be allocated as a place
holder to prevent port conflicts.  Once a
connection is established, usually within a
minute or so, the socket is closed.
.It PKT_ALIAS_UNREGISTERED_ONLY.
If this mode bit is set, traffic on the
local network which does not originate from
unregistered address spaces will be ignored.
Standard Class A, B and C unregistered addresses
are:
.Bd -literal -offset indent
    10.0.0.0     ->   10.255.255.255   (Class A subnet)
    172.16.0.0   ->   172.31.255.255   (Class B subnets)
    192.168.0.0  ->   192.168.255.255  (Class C subnets)

.Ed
This option is useful in the case that
packet aliasing host has both registered and
unregistered subnets on different interfaces.
The registered subnet is fully accessible to
the outside world, so traffic from it doesn't
need to be passed through the packet aliasing
engine.
.It PKT_ALIAS_RESET_ON_ADDR_CHANGE.
When this mode bit is set and
PacketAliasSetAddress() is called to change
the aliasing address, the internal link table
of the packet aliasing engine will be cleared.
This operating mode is useful for ppp links
where the interface address can sometimes
change or remain the same between dial-ups.
If this mode bit is not set, it the link table
will never be reset in the event of an
address change.
.It PKT_ALIAS_PUNCH_FW.
This option make libalias `punch holes' in an
ipfw based firewall for FTP/IRC DCC connections.
The holes punched are bound by from/to IP address
and port; it will not be possible to use a hole
for another connection.  A hole is removed when
the connection that use it die.  To cater for
unexpected death of a program using libalias (e.g
kill -9), changing the state of the flag will
clear the entire ipfw range allocated for holes.
This will also happen on the initial call to
PacketAliasSetFWBase().  This call must happen
prior to setting this flag.

.El

.Ss 2.5 PacketAliasSetFWBase()

.Ft void
.Fn PacketAliasSetFWBase "unsigned int base" "unsigned int num"

Set IPFW range allocated for punching firewall holes (with the
PKT_ALIAS_PUNCH_FW flag).  The range will be cleared for all rules on
initialization.

.Sh 3. Packet Handling
The packet handling functions are used to
modify incoming (remote->local) and outgoing
(local->remote) packets.  The calling program
is responsible for receiving and sending
packets via network interfaces.

Along with PacketAliasInit() and PacketAliasSetAddress(),
the two packet handling functions, PacketAliasIn()
and PacketAliasOut(), comprise minimal set of functions
needed for a basic IP masquerading implementation.
.Ss 3.1 PacketAliasIn()

.Ft int
.Fn PacketAliasIn "char *buffer" "int maxpacketsize"

An incoming packet coming from a remote machine to
the local network is de-aliased by this function.
The IP packet is pointed to by
.Em buffer ,
and
.Em maxpacketsize
indicates the size of the data structure containing
the packet and should be at least as large as the
actual packet size.

Return codes:
.Bl -hang -offset left
.It PKT_ALIAS_ERROR.
An internal error within the packet aliasing
engine occurred.
.It PKT_ALIAS_OK.
The packet aliasing process was successful.
.It PKT_ALIAS_IGNORED.
The packet was ignored and not de-aliased.
This can happen if the protocal is unrecognized,
possibly an ICMP message type is not handled or
if incoming packets for new connections are being
ignored (see PKT_ALIAS_DENY_INCOMING in section
2.2).
.It PKT_ALIAS_UNRESOLVED_FRAGMENT.
This is returned when a fragment cannot be
resolved because the header fragment has not
been sent yet.  In this situation, fragments
must be saved with PacketAliasSaveFragment()
until a header fragment is found.
.It PKT_ALIAS_FOUND_HEADER_FRAGMENT.
The packet aliasing process was successful,
and a header fragment was found.  This is a
signal to retrieve any unresolved fragments
with PacketAliasGetFragment() and de-alias
them with PacketAliasFragmentIn().
.El
.Ss 3.2 PacketAliasOut()

.Ft int
.Fn PacketAliasOut "char *buffer" "int maxpacketsize"

An outgoing packet coming from the local network
to a remote machine is aliased by this function.
The IP packet is pointed to by
.Em buffer r,
and
.Em maxpacketsize
indicates the maximum packet size permissible
should the packet length be changed.  IP encoding
protocols place address and port information in
the encapsulated data stream which have to be
modified and can account for changes in packet
length.  Well known examples of such protocols
are FTP and IRC DCC.

Return codes:
.Bl -hang -offset left
.It PKT_ALIAS_ERROR.
An internal error within the packet aliasing
engine occurred.
.It PKT_ALIAS_OK.
The packet aliasing process was successful.
.It PKT_ALIAS_IGNORED.
The packet was ignored and not de-aliased.
This can happen if the protocal is unrecognized,
or possibly an ICMP message type is not handled.
.El

.Sh 4. Port and Address Redirection
The functions described in this section allow machines
on the local network to be accessible in some degree
to new incoming connections from the external network.
Individual ports can be re-mapped or static network
address translations can be designated.
.Ss 4.1 PacketAliasRedirectPort()

.Ft struct alias_link *
.Fo PacketAliasRedirectPort
.Fa "struct in_addr local_addr"
.Fa "u_short local_port"
.Fa "struct in_addr remote_addr"
.Fa "u_short remote_port"
.Fa "struct in_addr alias_addr"
.Fa "u_short alias_port"
.Fa "u_char proto"
.Fc

This function specifies that traffic from a
given remote address/port to an alias address/port
be redirected to a specified local address/port.
The parameter
.Em proto
can be either IPPROTO_TCP or IPPROTO_UDP, as
defined in <netinet/in.h>.

If
.Em local_addr
or
.Em alias_addr
is zero, this indicates that the packet aliasing
address as established by PacketAliasSetAddress()
is to be used.  Even if PacketAliasAddress() is
called to change the address after PacketAliasRedirectPort()
is called, a zero reference will track this change.

If
.Em remote_addr
is zero, this indicates to redirect packets from
any remote address.  Likewise, if
.Em remote_port
is zero, this indicates to redirect packets originating
from any remote port number.  Almost always, the remote
port specification will be zero, but non-zero remote
addresses can be sometimes be useful for firewalling.
If two calls to PacketAliasRedirectPort() overlap in
their address/port specifications, then the most recent
call will have precedence.

This function returns a pointer which can subsequently
be used by PacketAliasRedirectDelete().  If NULL is
returned, then the function call did not complete
successfully.

All port numbers are in network address byte order,
so it is necessary to use htons() to convert these
parameters from internally readable numbers to
network byte order.  Addresses are also in network
byte order, which is implicit in the use of the
.Em struct in_addr
data type.
.Ss 4.2 PacketAliasRedirectAddr()

.Ft struct alias_link *
.Fo PacketAliasRedirectAddr
.Fa "struct in_addr local_addr"
.Fa "struct in_addr alias_addr"
.Fc

This function desgnates that all incoming
traffic to
.Em alias_addr
be redirected to
.Em local_addr.
Similarly, all outgoing traffic from
.Em local_addr
is aliased to
.Em alias_addr .

If
.Em local_addr
or
.Em alias_addr
is zero, this indicates that the packet aliasing
address as established by PacketAliasSetAddress()
is to be used.  Even if PacketAliasAddress() is
called to change the address after PacketAliasRedirectAddr()
is called, a zero reference will track this change.

If subsequent calls to PacketAliasRedirectAddr()
use the same aliasing address, all new incoming
traffic to this aliasing address will be redirected
to the local address made in the last function call,
but new traffic all of the local machines designated
in the several function calls will be aliased to
the same address.  Consider the following example:
.Bd -literal -offset left
    PacketAliasRedirectAddr(inet_aton("192.168.0.2"),
                            inet_aton("141.221.254.101"));
    PacketAliasRedirectAddr(inet_aton("192.168.0.3"),
                            inet_aton("141.221.254.101"));
    PacketAliasRedirectAddr(inet_aton("192.168.0.4"),
                            inet_aton("141.221.254.101"));
.Ed

Any outgoing connections such as telnet or ftp
from 192.168.0.2, 102.168.0.3, 192.168.0.4 will
appear to come from 141.221.254.101.  Any incoming
connections to 141.221.254.101 will be directed
to 192.168.0.4.

Any calls to PacketAliasRedirectPort() will
have precedence over address mappings designated
by PacketAliasRedirectAddr().

This function returns a pointer which can subsequently
be used by PacketAliasRedirectDelete().  If NULL is
returned, then the function call did not complete
successfully.
.Ss 4.3 PacketAliasRedirectDelete()

.Ft void
.Fn PacketAliasRedirectDelete "struct alias_link *ptr"

This function will delete a specific static redirect
rule entered by PacketAliasRedirectPort() or
PacketAliasRedirectAddr().  The parameter
.Em ptr
is the pointer returned by either of the redirection
functions.  If an invalid pointer is passed to
PacketAliasRedirectDelete(), then a program crash
or unpredictable operation could result, so it is
necessary to be careful using this function.

.Sh 5. Fragment Handling
The functions in this section are used to deal with
incoming fragments.

Outgoing fragments are handled within PacketAliasOut()
by changing the address according to any
applicable mapping set by PacketAliasRedirectAddress(),
or the default aliasing address set by
PacketAliasSetAddress().

Incoming fragments are handled in one of two ways.
If the header of a fragmented IP packet has already
been seen, then all subsequent fragments will be
re-mapped in the same manner the header fragment
was.  Fragments which arrive before the header
are saved and then retrieved once the header fragment
has been resolved.
.Ss 4.4 PacketAliasProxyRule()

.Ft int
.Fn PacketAliasProxyRule "char *cmd"

The passed
.Ar cmd
string consists of one or more pairs of words.  The first word in each
pair is a token and the second is the value that should be applied for
that token.  Tokens and their argument types are as follows:

.Bl -tag -offset XXX -width XXX
.It type encode_ip_hdr|encode_tcp_stream|no_encode
In order to support transparent proxying, it is necessary to somehow
pass the original address and port information into the new destination
server.  If
.Dq encode_ip_hdr
is specified, the original address and port is passed as an extra IP
option.  If
.Dq encode_tcp_stream
is specified, the original address and port is passed as the first
piece of data in the tcp stream in the format
.Dq DEST Ar IP port .
.It port Ar portnum
Only packets with the destination port
.Ar portnum
are proxied.
.It server Ar host[:portnum]
This specifies the
.Ar host
and
.Ar portnum
that the data is to be redirected to.  If
.Ar portnum
is not specified, the destination port number is not changed.
.Pp
The
.Ar server
specification is mandatory unless the
.Dq delete
command is being used.
.It rule Ar index
Normally, each call to
.Fn PacketAliasProxyRule
inserts the next rule at the start of a linear list of rules.  If an
.Ar index
is specified, the new rule will be checked after all rules with lower
indices.  Calls to
.Fn PacketAliasProxyRule
that do not specify a rule are assigned rule 0.
.It delete Ar index
This token and its argument must not be used with any other tokens.  When
used, all existing rules with the given
.Ar index
are deleted.
.It proto tcp|udp
If specified, only packets of the given protocol type are matched.
.It src Ar IP[/bits]
If specified, only packets with a source address matching the given
.Ar IP
are matched.  If
.Ar bits
is also specified, then the first
.Ar bits
bits of
.Ar IP
are taken as a network specification, and all IP addresses from that
network will be matched.
.It dest Ar IP[/bits]
If specified, only packets with a destination address matching the given
.Ar IP
are matched.  If
.Ar bits
is also specified, then the first
.Ar bits
bits of
.Ar IP
are taken as a network specification, and all IP addresses from that
network will be matched.
.El

This function is usually used to redirect outgoing connections for
internal machines that are not permitted certain types of internet
access, or to restrict access to certain external machines.

.Ss 5.1 PacketAliasSaveFragment()

.Ft int
.Fn PacketAliasSaveFragment "char *ptr"

When PacketAliasIn() returns
PKT_ALIAS_UNRESOLVED_FRAGMENT, this
function can be used to save the pointer to
the unresolved fragment.

It is implicitly assumed that
.Em ptr
points to a block of memory allocated by
malloc().  If the fragment is never
resolved, the packet aliasing engine will
automatically free the memory after a
timeout period. [Eventually this function
should be modified so that a callback
function for freeing memory is passed as
an argument.]

This function returns PKT_ALIAS_OK if it
was successful and PKT_ALIAS_ERROR if there
was an error.

.Ss 5.2 PacketAliasGetFragment()

.Ft char *
.Fn PacketAliasGetFragment "char *buffer"

This function can be used to retrieve fragment
pointers saved by PacketAliasSaveFragment().
The IP header fragment pointed to by
Em buffer
is the header fragment indicated when
PacketAliasIn() returns PKT_ALIAS_FOUND_HEADER_FRAGMENT.
Once a a fragment pointer is retrieved, it
becomes the calling program's responsibility
to free the dynamically allocated memory for
the fragment.

PacketAliasGetFragment() can be called
sequentially until there are no more fragments
available, at which time it returns NULL.
.Ss 5.3 PacketAliasFragmentIn()

.Ft void
.Fn PacketAliasFragmentIn "char *header" "char *fragment"

When a fragment is retrieved with
PacketAliasGetFragment(), it can then be
de-aliased with a call to PacketAliasFragmentIn().
.Em header
is the pointer to a header fragment used as a
template, and
.Em fragment
is the pointer to the packet to be de-aliased.

.Sh 6. Miscellaneous Functions

.Ss 6.1 PacketAliasSetTarget()

.Ft void
.Fn PacketAliasSetTarget "struct in_addr addr"

When an incoming packet not associated with
any pre-existing aliasing link arrives at the
host machine, it will be sent to the address
indicated by a call to PacketAliasSetTarget().

If this function is not called, or is called
with a zero address argument, then all new
incoming packets go to the address set by
PacketAliasSetAddress.
.Ss 6.2 PacketAliasCheckNewLink()

.Ft int
.Fn PacketAliasCheckNewLink "void"

This function returns a non-zero value when
a new aliasing link is created.  In circumstances
where incoming traffic is being sequentially
sent to different local servers, this function
can be used to trigger when PacketAliasSetTarget()
is called to change the default target address.
.Ss 6.3 PacketAliasInternetChecksum()

.Ft u_short
.Fn PacketAliasInternetChecksum "u_short *buffer" "int nbytes"

This is a utility function that does not seem
to be available elswhere and is included as a
convenience.  It computes the internet checksum,
which is used in both IP and protocol-specific
headers (TCP, UDP, ICMP).

.Em buffer
points to the data block to be checksummed, and
.Em nbytes
is the number of bytes.  The 16-bit checksum
field should be zeroed before computing the checksum.

Checksums can also be verified by operating on a block
of data including its checksum.  If the checksum is
valid, PacketAliasInternetChecksum() will return zero.

.Sh 7. Authors
Charles Mott (cmott@srv.net), versions 1.0 - 1.8, 2.0 - 2.4.

Eivind Eklund (eivind@freebsd.org), versions 1.8b, 1.9 and
2.5.  Added IRC DCC support as well as contributing a number of
architectural improvements; added the firewall bypass
for FTP/IRC DCC.

.Sh 8. Acknowledgments

Listed below, in approximate chronological
order, are individuals who have provided
valuable comments and/or debugging assistance.

.Bl -inset -compact -offset left
.It Gary Roberts
.It Tom Torrance
.It Reto Burkhalter
.It Martin Renters
.It Brian Somers
.It Paul Traina
.It Ari Suutari
.It Dave Remien
.It J. Fortes
.It Andrzej Bialeki
.It Gordon Burditt
.El

.Sh Appendix: Conceptual Background
This appendix is intended for those who
are planning to modify the source code or want
to create somewhat esoteric applications using
the packet aliasing functions.

The conceptual framework under which the
packet aliasing engine operates is described here.
Central to the discussion is the idea of an
"aliasing link" which  describes the relationship
for a given packet transaction between the local
machine, aliased identity and remote machine.  It
is discussed how such links come into existence
and are destroyed.
.Ss A.1 Aliasing Links
There is a notion of an "aliasing link",
which is 7-tuple describing a specific
translation:
.Bd -literal -offset indent
(local addr, local port, alias addr, alias port,
 remote addr, remote port, protocol)
.Ed

Outgoing packets have the local address and
port number replaced with the alias address
and port number.  Incoming packets undergo the
reverse process.  The packet aliasing engine
attempts to match packets against an internal
table of aliasing links to determine how to
modify a given IP packet.  Both the IP
header and protocol dependent headers are
modified as necessary.  Aliasing links are
created and deleted as necessary according
to network traffic.

Protocols can be TCP, UDP or even ICMP in
certain circumstances.  (Some types of ICMP
packets can be aliased according to sequence
or id number which acts as an equivalent port
number for identifying how individual packets
should be handled.)

Each aliasing link must have a unique
combination of the following five quantities:
alias address/port, remote address/port
and protocol.  This ensures that several
machines on a local network can share the
same aliased IP address.  In cases where
conflicts might arise, the aliasing port
is chosen so that uniqueness is maintained.
.Ss A.2 Static and Dynamic Links
Aliasing links can either be static or dynamic.
Static links persist indefinitely and represent
fixed rules for translating IP packets.  Dynamic
links come into existence for a specific TCP
connection or UDP transaction or ICMP echo
sequence.  For the case of TCP, the connection
can be monitored to see when the associated
aliasing link should be deleted.  Aliasing links
for UDP transactions (and ICMP echo and timestamp
requests) work on a simple timeout rule.  When
no activity is observed on a dynamic link for
a certain amount of time it is automatically
deleted.  Timeout rules also apply to TCP
connections which do not open or close
properly.
.Ss A.3 Partially Specified Aliasing Links
Aliasing links can be partially specified,
meaning that the remote address and/or remote
ports are unknown.  In this case, when a packet
matching the incomplete specification is found,
a fully specified dynamic link is created.  If
the original partially specified link is dynamic,
it will be deleted after the fully specified link
is created, otherwise it will persist.

For instance, a partially specified link might
be
.Bd -literal -offset indent
(192.168.0.4, 23, 204.228.203.215, 8066, 0, 0, tcp)
.Ed

The zeros denote unspecified components for
the remote address and port.  If this link were
static it would have the effect of redirecting
all incoming traffic from port 8066 of
204.228.203.215 to port 23 (telnet) of machine
192.168.0.4 on the local network.  Each
individual telnet connection would initiate
the creation of a distinct dynamic link.
.Ss A.4 Dynamic Link Creation
In addition to aliasing links, there are
also address mappings that can be stored
within the internal data table of the packet
aliasing mechanism.
.Bd -literal -offset indent
(local addr, alias addr)
.Ed

Address mappings are searched when creating
new dynamic links.

All outgoing packets from the local network
automatically create a dynamic link if
they do not match an already existing fully
specified link.  If an address mapping exists
for the the outgoing packet, this determines
the alias address to be used.  If no mapping
exists, then a default address, usually the
address of the packet aliasing host, is used.
If necessary, this default address can be
changed as often as each individual packet
arrives.

The aliasing port number is determined
such that the new dynamic link does not
conflict with any existing links.  In the
default operating mode, the packet aliasing
engine attempts to set the aliasing port
equal to the local port number.  If this
results in a conflict, then port numbers
are randomly chosen until a unique aliasing
link can be established.  In an alternate
operating mode, the first choice of an
aliasing port is also random and unrelated
to the local port number.