xref: /freebsd/sys/netinet/libalias/libalias.3 (revision cc81ddd9db63bfb4f842b2ec152934b46da50a36)

.\"-
.\" Copyright (c) 2001 Charles Mott <cm@linktel.net>
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\"    notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\"    notice, this list of conditions and the following disclaimer in the
.\"    documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
.\" HOWEVER CAUSED AND ON 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 ADVISED OF THE POSSIBILITY OF
.\" SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd October 1, 2006
.Dt LIBALIAS 3
.Os
.Sh NAME
.Nm libalias
.Nd packet aliasing library for masquerading and network address translation
.Sh SYNOPSIS
.In sys/types.h
.In netinet/in.h
.In alias.h
.Pp
Function prototypes are given in the main body of the text.
.Sh DESCRIPTION
The
.Nm
library is a collection of functions for aliasing and de-aliasing of IP
packets, intended for masquerading and network address translation (NAT).
.Sh 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.
.Pp
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.
.Pp
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 INITIALIZATION AND CONTROL
One special function,
.Fn LibAliasInit ,
must always be called before any packet handling may be performed and
the returned instance pointer passed to all the other functions.
Normally, the
.Fn LibAliasSetAddress
function is called afterwards, to set the default aliasing address.
In addition, the operating mode of the packet aliasing engine can be
customized by calling
.Fn LibAliasSetMode .
.Pp
.Ft "struct libalias *"
.Fn LibAliasInit "struct libalias *"
.Bd -ragged -offset indent
This function is used to initialize
internal data structures.
When called the first time, a
.Dv NULL
pointer should be passed as an argument.
The following mode bits are always set after calling
.Fn LibAliasInit .
See the description of
.Fn LibAliasSetMode
below for the meaning of these mode bits.
.Pp
.Bl -item -offset indent -compact
.It
.Dv PKT_ALIAS_SAME_PORTS
.It
.Dv PKT_ALIAS_USE_SOCKETS
.It
.Dv PKT_ALIAS_RESET_ON_ADDR_CHANGE
.El
.Pp
This function will always return the packet aliasing engine to the same
initial state.
The
.Fn LibAliasSetAddress
function is normally called afterwards, and any desired changes from the
default mode bits listed above require a call to
.Fn LibAliasSetMode .
.Pp
It is mandatory that this function be called at the beginning of a program
prior to any packet handling.
.Ed
.Pp
.Ft void
.Fn LibAliasUninit "struct libalias *"
.Bd -ragged -offset indent
This function has no return value and is used to clear any
resources attached to internal data structures.
.Pp
This functions should be called when a program stops using the aliasing
engine; it does, amongst other things, clear out any firewall holes.
To provide backwards compatibility and extra security, it is added to
the
.Xr atexit 3
chain by
.Fn LibAliasInit .
.Ed
.Pp
.Ft void
.Fn LibAliasSetAddress "struct libalias *" "struct in_addr addr"
.Bd -ragged -offset indent
This function sets the source address to which outgoing packets from the
local area network are aliased.
All outgoing packets are re-mapped to this address unless overridden by a
static address mapping established by
.Fn LibAliasRedirectAddr .
If this function is not called, and no static rules match, an outgoing
packet retains its source address.
.Pp
If the
.Dv 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.
This is useful for interfaces such as
.Xr ppp 8 ,
where the IP
address may or may not change on successive dial-up attempts.
.Pp
If the
.Dv 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).
.Pp
It is mandatory that this function be called prior to any packet handling.
.Ed
.Pp
.Ft unsigned int
.Fn LibAliasSetMode "struct libalias *" "unsigned int flags" "unsigned int mask"
.Bd -ragged -offset indent
This function sets or clears mode bits
according to the value of
.Fa flags .
Only bits marked in
.Fa mask
are affected.
The following mode bits are defined in
.In alias.h :
.Bl -tag -width indent
.It Dv PKT_ALIAS_LOG
Enables logging into
.Pa /var/log/alias.log .
Each time an aliasing link is created or deleted, the log file is appended
with the current number of ICMP, TCP and UDP links.
Mainly useful for debugging when the log file is viewed continuously with
.Xr tail 1 .
.It Dv 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
.Fn ( LibAliasIn
returns
.Dv PKT_ALIAS_IGNORED
code)
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 Dv 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 numbers.
This can be done as long as the quintuple (proto, alias addr, alias port,
remote addr, remote port) is unique.
If a conflict exists, a new aliasing port number is chosen even if this
mode bit is set.
.It Dv PKT_ALIAS_USE_SOCKETS
This bit should be set when 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 Dv 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
.Pp
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 does not need to be passed through the packet aliasing engine.
.It Dv PKT_ALIAS_RESET_ON_ADDR_CHANGE
When this mode bit is set and
.Fn LibAliasSetAddress
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
.Xr ppp 8
links where the interface address can sometimes change or remain the same
between dial-up attempts.
If this mode bit is not set, the link table will never be reset in the event
of an address change.
.It Dv PKT_ALIAS_PUNCH_FW
This option makes
.Nm
`punch holes' in an
.Xr ipfirewall 4
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 uses it dies.
To cater to unexpected death of a program using
.Nm
(e.g.\& kill -9),
changing the state of the flag will clear the entire firewall range
allocated for holes.
This will also happen on the initial call to
.Fn LibAliasSetFWBase .
This call must happen prior to setting this flag.
.It Dv PKT_ALIAS_REVERSE
This option makes
.Nm
reverse the way it handles incoming and outgoing packets, allowing it
to be fed with data that passes through the internal interface rather
than the external one.
.It Dv PKT_ALIAS_PROXY_ONLY
This option tells
.Nm
to obey transparent proxy rules only.
Normal packet aliasing is not performed.
See
.Fn LibAliasProxyRule
below for details.
.El
.Ed
.Pp
.Ft void
.Fn LibAliasSetFWBase "struct libalias *" "unsigned int base" "unsigned int num"
.Bd -ragged -offset indent
Set firewall range allocated for punching firewall holes (with the
.Dv PKT_ALIAS_PUNCH_FW
flag).
The range will be cleared for all rules on initialization.
.Ed
.Pp
.Ft void
.Fn LibAliasSkinnyPort "struct libalias *" "unsigned int port"
.Bd -ragged -offset indent
Set the TCP port used by the Skinny Station protocol.
Skinny is used by Cisco IP phones to communicate with
Cisco Call Managers to set up voice over IP calls.
If this is not set, Skinny aliasing will not be done.
The typical port used by Skinny is 2000.
.Ed
.Sh PACKET HANDLING
The packet handling functions are used to modify incoming (remote to local)
and outgoing (local to remote) packets.
The calling program is responsible for receiving and sending packets via
network interfaces.
.Pp
Along with
.Fn LibAliasInit
and
.Fn LibAliasSetAddress ,
the two packet handling functions,
.Fn LibAliasIn
and
.Fn LibAliasOut ,
comprise minimal set of functions needed for a basic IP masquerading
implementation.
.Pp
.Ft int
.Fn LibAliasIn "struct libalias *" "char *buffer" "int maxpacketsize"
.Bd -ragged -offset indent
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
.Fa buffer ,
and
.Fa maxpacketsize
indicates the size of the data structure containing the packet and should
be at least as large as the actual packet size.
.Pp
Return codes:
.Bl -tag -width indent
.It Dv PKT_ALIAS_OK
The packet aliasing process was successful.
.It Dv PKT_ALIAS_IGNORED
The packet was ignored and not de-aliased.
This can happen if the protocol is unrecognized, possibly an ICMP message
type is not handled or if incoming packets for new connections are being
ignored (if
.Dv PKT_ALIAS_DENY_INCOMING
mode bit was set by
.Fn LibAliasSetMode ) .
.It Dv 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
.Fn LibAliasSaveFragment
until a header fragment is found.
.It Dv 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
.Fn LibAliasGetFragment
and de-alias them with
.Fn LibAliasFragmentIn .
.It Dv PKT_ALIAS_ERROR
An internal error within the packet aliasing engine occurred.
.El
.Ed
.Pp
.Ft int
.Fn LibAliasOut "struct libalias *" "char *buffer" "int maxpacketsize"
.Bd -ragged -offset indent
An outgoing packet coming from the local network to a remote machine is
aliased by this function.
The IP packet is pointed to by
.Fa buffer ,
and
.Fa 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 has to be modified and can account for changes in packet
length.
Well known examples of such protocols are FTP and IRC DCC.
.Pp
Return codes:
.Bl -tag -width indent
.It Dv PKT_ALIAS_OK
The packet aliasing process was successful.
.It Dv PKT_ALIAS_IGNORED
The packet was ignored and not aliased.
This can happen if the protocol is unrecognized, or possibly an ICMP message
type is not handled.
.It Dv PKT_ALIAS_ERROR
An internal error within the packet aliasing engine occurred.
.El
.Ed
.Sh 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.
.Pp
.Ft struct alias_link *
.Fo LibAliasRedirectPort
.Fa "struct libalias *"
.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
.Bd -ragged -offset indent
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
.Fa proto
can be either
.Dv IPPROTO_TCP
or
.Dv IPPROTO_UDP ,
as defined in
.In netinet/in.h .
.Pp
If
.Fa local_addr
or
.Fa alias_addr
is zero, this indicates that the packet aliasing address as established
by
.Fn LibAliasSetAddress
is to be used.
Even if
.Fn LibAliasSetAddress
is called to change the address after
.Fn LibAliasRedirectPort
is called, a zero reference will track this change.
.Pp
If the link is further set up to operate for a load sharing, then
.Fa local_addr
and
.Fa local_port
are ignored, and are selected dynamically from the server pool, as described in
.Fn LibAliasAddServer
below.
.Pp
If
.Fa remote_addr
is zero, this indicates to redirect packets from any remote address.
Likewise, if
.Fa 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 sometimes be useful for firewalling.
If two calls to
.Fn LibAliasRedirectPort
overlap in their address/port specifications, then the most recent call
will have precedence.
.Pp
This function returns a pointer which can subsequently be used by
.Fn LibAliasRedirectDelete .
If
.Dv NULL
is returned, then the function call did not complete successfully.
.Pp
All port numbers should be in network address byte order, so it is necessary
to use
.Xr htons 3
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
.Fa struct in_addr
data type.
.Ed
.Pp
.Ft struct alias_link *
.Fo LibAliasRedirectAddr
.Fa "struct libalias *"
.Fa "struct in_addr local_addr"
.Fa "struct in_addr alias_addr"
.Fc
.Bd -ragged -offset indent
This function designates that all incoming traffic to
.Fa alias_addr
be redirected to
.Fa local_addr .
Similarly, all outgoing traffic from
.Fa local_addr
is aliased to
.Fa alias_addr .
.Pp
If
.Fa local_addr
or
.Fa alias_addr
is zero, this indicates that the packet aliasing address as established by
.Fn LibAliasSetAddress
is to be used.
Even if
.Fn LibAliasSetAddress
is called to change the address after
.Fn LibAliasRedirectAddr
is called, a zero reference will track this change.
.Pp
If the link is further set up to operate for a load sharing, then
.Fa local_addr
is ignored, and is selected dynamically from the server pool, as described in
.Fn LibAliasAddServer
below.
.Pp
If subsequent calls to
.Fn LibAliasRedirectAddr
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.
New traffic generated by any of the local machines, designated in the
several function calls, will be aliased to the same address.
Consider the following example:
.Bd -literal -offset indent
LibAliasRedirectAddr(la, inet_aton("192.168.0.2"),
                        inet_aton("141.221.254.101"));
LibAliasRedirectAddr(la, inet_aton("192.168.0.3"),
                        inet_aton("141.221.254.101"));
LibAliasRedirectAddr(la, inet_aton("192.168.0.4"),
                        inet_aton("141.221.254.101"));
.Ed
.Pp
Any outgoing connections such as
.Xr telnet 1
or
.Xr ftp 1
from 192.168.0.2, 192.168.0.3 and 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.
.Pp
Any calls to
.Fn LibAliasRedirectPort
will have precedence over address mappings designated by
.Fn LibAliasRedirectAddr .
.Pp
This function returns a pointer which can subsequently be used by
.Fn LibAliasRedirectDelete .
If
.Dv NULL
is returned, then the function call did not complete successfully.
.Ed
.Pp
.Ft int
.Fo LibAliasAddServer
.Fa "struct libalias *"
.Fa "struct alias_link *link"
.Fa "struct in_addr addr"
.Fa "u_short port"
.Fc
.Bd -ragged -offset indent
This function sets the
.Fa link
up for Load Sharing using IP Network Address Translation (RFC 2391, LSNAT).
LSNAT operates as follows.
A client attempts to access a server by using the server virtual address.
The LSNAT router transparently redirects the request to one of the hosts
in server pool, selected using a real-time load sharing algorithm.
Multiple sessions may be initiated from the same client, and each session
could be directed to a different host based on load balance across server
pool hosts at the time.
If load share is desired for just a few specific services, the configuration
on LSNAT could be defined to restrict load share for just the services
desired.
.Pp
Currently, only the simplest selection algorithm is implemented, where a
host is selected on a round-robin basis only, without regard to load on
the host.
.Pp
First, the
.Fa link
is created by either
.Fn LibAliasRedirectPort
or
.Fn LibAliasRedirectAddr .
Then,
.Fn LibAliasAddServer
is called multiple times to add entries to the
.Fa link Ns 's
server pool.
.Pp
For links created with
.Fn LibAliasRedirectAddr ,
the
.Fa port
argument is ignored and could have any value, e.g.\& htons(~0).
.Pp
This function returns 0 on success, \-1 otherwise.
.Ed
.Pp
.Ft int
.Fn LibAliasRedirectDynamic "struct libalias *" "struct alias_link *link"
.Bd -ragged -offset indent
This function marks the specified static redirect rule entered by
.Fn LibAliasRedirectPort
as dynamic.
This can be used to e.g.\& dynamically redirect a single TCP connection,
after which the rule is removed.
Only fully specified links can be made dynamic.
(See the
.Sx STATIC AND DYNAMIC LINKS
and
.Sx PARTIALLY SPECIFIED ALIASING LINKS
sections below for a definition of static vs.\& dynamic,
and partially vs.\& fully specified links.)
.Pp
This function returns 0 on success, \-1 otherwise.
.Ed
.Pp
.Ft void
.Fn LibAliasRedirectDelete "struct libalias *" "struct alias_link *link"
.Bd -ragged -offset indent
This function will delete a specific static redirect rule entered by
.Fn LibAliasRedirectPort
or
.Fn LibAliasRedirectAddr .
The parameter
.Fa link
is the pointer returned by either of the redirection functions.
If an invalid pointer is passed to
.Fn LibAliasRedirectDelete ,
then a program crash or unpredictable operation could result, so it is
necessary to be careful using this function.
.Ed
.Pp
.Ft int
.Fn LibAliasProxyRule "struct libalias *" "const char *cmd"
.Bd -ragged -offset indent
The passed
.Fa 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 -width indent
.It Cm 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
.Cm encode_ip_hdr
is specified, the original destination address and port are passed
as an extra IP option.
If
.Cm encode_tcp_stream
is specified, the original destination address and port are passed
as the first piece of data in the TCP stream in the format
.Dq Li DEST Ar IP port .
.It Cm port Ar portnum
Only packets with the destination port
.Ar portnum
are proxied.
.It Cm server Ar host Ns Op : Ns Ar portnum
This specifies the
.Ar host
and
.Ar portnum
that the data is to be redirected to.
.Ar host
must be an IP address rather than a DNS host name.
If
.Ar portnum
is not specified, the destination port number is not changed.
.Pp
The
.Ar server
specification is mandatory unless the
.Cm delete
command is being used.
.It Cm rule Ar index
Normally, each call to
.Fn LibAliasProxyRule
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 LibAliasProxyRule
that do not specify a rule are assigned rule 0.
.It Cm 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 Cm proto tcp | udp
If specified, only packets of the given protocol type are matched.
.It Cm src Ar IP Ns Op / Ns Ar 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 Cm dst Ar IP Ns Op / Ns Ar 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
.Pp
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.
.Ed
.Pp
.Ft struct alias_link *
.Fo LibAliasRedirectProto
.Fa "struct libalias *"
.Fa "struct in_addr local_addr"
.Fa "struct in_addr remote_addr"
.Fa "struct in_addr alias_addr"
.Fa "u_char proto"
.Fc
.Bd -ragged -offset indent
This function specifies that any IP packet with protocol number of
.Fa proto
from a given remote address to an alias address be
redirected to a specified local address.
.Pp
If
.Fa local_addr
or
.Fa alias_addr
is zero, this indicates that the packet aliasing address as established
by
.Fn LibAliasSetAddress
is to be used.
Even if
.Fn LibAliasSetAddress
is called to change the address after
.Fn LibAliasRedirectProto
is called, a zero reference will track this change.
.Pp
If
.Fa remote_addr
is zero, this indicates to redirect packets from any remote address.
Non-zero remote addresses can sometimes be useful for firewalling.
.Pp
If two calls to
.Fn LibAliasRedirectProto
overlap in their address specifications, then the most recent call
will have precedence.
.Pp
This function returns a pointer which can subsequently be used by
.Fn LibAliasRedirectDelete .
If
.Dv NULL
is returned, then the function call did not complete successfully.
.Ed
.Sh FRAGMENT HANDLING
The functions in this section are used to deal with incoming fragments.
.Pp
Outgoing fragments are handled within
.Fn LibAliasOut
by changing the address according to any applicable mapping set by
.Fn LibAliasRedirectAddr ,
or the default aliasing address set by
.Fn LibAliasSetAddress .
.Pp
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.
.Pp
.Ft int
.Fn LibAliasSaveFragment "struct libalias *" "char *ptr"
.Bd -ragged -offset indent
When
.Fn LibAliasIn
returns
.Dv PKT_ALIAS_UNRESOLVED_FRAGMENT ,
this function can be used to save the pointer to the unresolved fragment.
.Pp
It is implicitly assumed that
.Fa ptr
points to a block of memory allocated by
.Xr malloc 3 .
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.]
.Pp
This function returns
.Dv PKT_ALIAS_OK
if it was successful and
.Dv PKT_ALIAS_ERROR
if there was an error.
.Ed
.Pp
.Ft char *
.Fn LibAliasGetFragment "struct libalias *" "char *buffer"
.Bd -ragged -offset indent
This function can be used to retrieve fragment pointers saved by
.Fn LibAliasSaveFragment .
The IP header fragment pointed to by
.Fa buffer
is the header fragment indicated when
.Fn LibAliasIn
returns
.Dv PKT_ALIAS_FOUND_HEADER_FRAGMENT .
Once a fragment pointer is retrieved, it becomes the calling program's
responsibility to free the dynamically allocated memory for the fragment.
.Pp
The
.Fn LibAliasGetFragment
function can be called sequentially until there are no more fragments
available, at which time it returns
.Dv NULL .
.Ed
.Pp
.Ft void
.Fn LibAliasFragmentIn "struct libalias *" "char *header" "char *fragment"
.Bd -ragged -offset indent
When a fragment is retrieved with
.Fn LibAliasGetFragment ,
it can then be de-aliased with a call to
.Fn LibAliasFragmentIn .
The
.Fa header
argument is the pointer to a header fragment used as a template, and
.Fa fragment
is the pointer to the packet to be de-aliased.
.Ed
.Sh MISCELLANEOUS FUNCTIONS
.Ft void
.Fn LibAliasSetTarget "struct libalias *" "struct in_addr addr"
.Bd -ragged -offset indent
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
.Fn LibAliasSetTarget .
.Pp
If this function is called with an
.Dv INADDR_NONE
address argument, then all new incoming packets go to the address set by
.Fn LibAliasSetAddress .
.Pp
If this function is not called, or is called with an
.Dv INADDR_ANY
address argument, then all new incoming packets go to the address specified
in the packet.
This allows external machines to talk directly to internal machines if they
can route packets to the machine in question.
.Ed
.Pp
.Ft int
.Fn LibAliasCheckNewLink "struct libalias *"
.Bd -ragged -offset indent
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
.Fn LibAliasSetTarget
is called to change the default target address.
.Ed
.Pp
.Ft u_short
.Fn LibAliasInternetChecksum "struct libalias *" "u_short *buffer" "int nbytes"
.Bd -ragged -offset indent
This is a utility function that does not seem to be available elsewhere and
is included as a convenience.
It computes the internet checksum, which is used in both IP and
protocol-specific headers (TCP, UDP, ICMP).
.Pp
The
.Fa buffer
argument points to the data block to be checksummed, and
.Fa nbytes
is the number of bytes.
The 16-bit checksum field should be zeroed before computing the checksum.
.Pp
Checksums can also be verified by operating on a block of data including
its checksum.
If the checksum is valid,
.Fn LibAliasInternetChecksum
will return zero.
.Ed
.Pp
.Ft int
.Fn LibAliasUnaliasOut "struct libalias *" "char *buffer" "int maxpacketsize"
.Bd -ragged -offset indent
An outgoing packet, which has already been aliased,
has its private address/port information restored by this function.
The IP packet is pointed to by
.Fa buffer ,
and
.Fa maxpacketsize
is provided for error checking purposes.
This function can be used if an already-aliased packet needs to have its
original IP header restored for further processing (e.g.\& logging).
.Ed
.Sh AUTHORS
.An Charles Mott Aq cm@linktel.net ,
versions 1.0 - 1.8, 2.0 - 2.4.
.An Eivind Eklund Aq 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.
.An Erik Salander Aq erik@whistle.com
added support for PPTP and RTSP.
.An Junichi Satoh Aq junichi@junichi.org
added support for RTSP/PNA.
.An Ruslan Ermilov Aq ru@FreeBSD.org
added support for PPTP and LSNAT as well as general hacking.
.An Paolo Pisati Aq piso@FreeBSD.org
made the library modular, moving support for all
protocols (except for IP, TCP and UDP) to external modules.
.Sh ACKNOWLEDGMENTS
Listed below, in approximate chronological order, are individuals who
have provided valuable comments and/or debugging assistance.
.Pp
.Bd -ragged -offset indent
.An -split
.An Gary Roberts
.An Tom Torrance
.An Reto Burkhalter
.An Martin Renters
.An Brian Somers
.An Paul Traina
.An Ari Suutari
.An Dave Remien
.An J. Fortes
.An Andrzej Bialecki
.An Gordon Burditt
.Ed
.Sh CONCEPTUAL BACKGROUND
This section is intended for those who are planning to modify the source
code or want to create somewhat esoteric applications using the packet
aliasing functions.
.Pp
The conceptual framework under which the packet aliasing engine operates
is described here.
Central to the discussion is the idea of an
.Em 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 ALIASING LINKS
There is a notion of an
.Em aliasing link ,
which is a 7-tuple describing a specific translation:
.Bd -literal -offset indent
(local addr, local port, alias addr, alias port,
 remote addr, remote port, protocol)
.Ed
.Pp
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.
.Pp
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.)
.Pp
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 aliasing IP address.
In cases where conflicts might arise, the aliasing port is chosen so that
uniqueness is maintained.
.Ss 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 PARTIALLY SPECIFIED ALIASING LINKS
Aliasing links can be partially specified, meaning that the remote address
and/or remote port 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.
.Pp
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
.Pp
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 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
.Pp
Address mappings are searched when creating new dynamic links.
.Pp
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 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.
.Pp
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.
.Sh MODULAR ARCHITECTURE (AND Xr ipfw 4 Sh SUPPORT)
One of the latest improvements to
.Nm
was to make its support
for new protocols independent from the rest of the library, giving it
the ability to load/unload support for new protocols at run-time.
To achieve this feature, all the code for protocol handling was moved
to a series of modules outside of the main library.
These modules are compiled from the same sources but work in
different ways, depending on whether they are compiled to work inside a kernel
or as part of the userland library.
.Ss LIBALIAS MODULES IN KERNEL LAND
When compiled for the kernel,
.Nm
modules are plain KLDs recognizable with the
.Pa alias_
prefix.
.Pp
To add support for a new protocol, load the corresponding module.
For example:
.Pp
.Dl "kldload alias_ftp"
.Pp
When support for a protocol is no longer needed, its module can be unloaded:
.Pp
.Dl "kldunload alias_ftp"
.Ss LIBALIAS MODULES IN USERLAND
Due to the differences between kernel and userland (no KLD mechanism,
many different address spaces, etc.), we had to change a bit how to
handle module loading/tracking/unloading in userland.
.Pp
While compiled for a userland
.Nm ,
all the modules are plain libraries, residing in
.Pa /usr/lib ,
and recognizable with the
.Pa libalias_
prefix.
.Pp
There is a configuration file,
.Pa /etc/libalias.conf ,
with the following contents (by default):
.Bd -literal -offset indent
/usr/lib/libalias_cuseeme.so
/usr/lib/libalias_ftp.so
/usr/lib/libalias_irc.so
/usr/lib/libalias_nbt.so
/usr/lib/libalias_pptp.so
/usr/lib/libalias_skinny.so
/usr/lib/libalias_smedia.so
.Ed
.Pp
This file contains the paths to the modules that
.Nm
will load.
To load/unload a new module, just add its path to
.Pa libalias.conf
and call
.Fn LibAliasRefreshModules
from the program.
In case the application provides a
.Dv SIGHUP
signal handler, add a call to
.Fn LibAliasRefreshModules
inside the handler, and everytime you want to refresh the loaded modules,
send it the
.Dv SIGHUP
signal:
.Pp
.Dl "kill -HUP <process_pid>"
.Ss MODULAR ARCHITECURE: HOW IT WORKS
The modular architecture of
.Nm
works similar whether it is running inside the
kernel or in userland.
From
.Pa alias_mod.c :
.Bd -literal
/* Protocol and userland module handlers chains. */
LIST_HEAD(handler_chain, proto_handler) handler_chain ...
\&...
SLIST_HEAD(dll_chain, dll) dll_chain ...
.Ed
.Pp
.Va handler_chain
keep tracks of all the protocol handlers loaded, while
.Va ddl_chain
takes care of userland modules loaded.
.Pp
.Va handler_chain
is composed of
.Vt "struct proto_handler"
entries:
.Bd -literal
struct proto_handler {
	u_int pri;
	int16_t dir;
	uint8_t proto;
	int (*fingerprint)(struct libalias *la,
		 struct ip *pip, struct alias_data *ah);
	int (*protohandler)(struct libalias *la,
		 struct ip *pip, struct alias_data *ah);
	LIST_ENTRY(proto_handler) entries;
};
.Ed
.Pp
where:
.Bl -inset
.It Va pri
is the priority assigned to a protocol handler, lower
is better.
.It Va dir
is the direction of packets: ingoing or outgoing.
.It Va proto
says at which protocol this packet belongs: IP, TCP or UDP.
.It Va fingerprint
points to the fingerprint function while protohandler points
to the protocol handler function.
.El
.Pp
The
.Va fingerprint
function has the double of scope of checking if the
incoming packet is found and if it belongs to any categories that this
module can handle.
.Pp
The
.Va protohandler
function actually manipulates
the packet to make
.Nm
correctly NAT it.
.Pp
When a packet enters
.Nm ,
if it meets a module hook,
.Va handler_chain
is searched to see if there is an handler that matches
this type of a packet (it checks protocol and direction of packet), then if
more than one handler is found, it starts with the module with
the lowest priority number: it calls the
.Va fingerprint
function and interprets the result.
.Pp
If the result value is equal to 0 then it calls the protocol handler
of this handler and returns.
Otherwise, it proceeds to the next eligible module until the
.Va handler_chain
is exhausted.
.Pp
Inside
.Nm ,
the module hook looks like this:
.Bd -literal -offset indent
struct alias_data ad = {
	lnk,
	&original_address,
	&alias_address,
	&alias_port,
	&ud->uh_sport,          /* original source port */
	&ud->uh_dport,		/* original dest port */
	256                     /* maxpacketsize */
};

\&...

/* walk out chain */
err = find_handler(IN, UDP, la, pip, &ad);
.Ed
.Pp
All data useful to a module are gathered together in an
.Vt alias_data
structure, then
.Fn find_handler
is called.
The
.Fn find_handler
function is responsible for walking out the handler
chain, it receives as input parameters:
.Bl -tag -width indent
.It Fa IN
direction
.It Fa UDP
working protocol
.It Fa la
pointer to this instance of libalias
.It Fa pip
pointer to a
.Vt "struct ip"
.It Fa ad
pointer to
.Vt "struct alias_data"
(see above)
.El
.Pp
In this case,
.Fn find_handler
will search only for modules registered for
supporting INcoming UDP packets.
.Pp
As was mentioned earlier,
.Nm
in userland is a bit different, cause
care has to be taken of module handling too (avoiding duplicate load of
module, avoiding module with same name, etc.) so
.Va dll_chain
was introduced.
.Pp
.Va dll_chain
contains a list of all userland
.Nm
modules loaded.
.Pp
When an application calls
.Fn LibAliasRefreshModules ,
.Nm
first unloads all the loaded modules, then reloads all the modules listed in
.Pa /etc/libalias.conf :
for every module loaded, a new entry to
.Va dll_chain
is added.
.Pp
.Va dll_chain
is composed of
.Vt "struct dll"
entries:
.Bd -literal
struct dll {
	/* name of module */
	char            name[DLL_LEN];
	/*
	 * ptr to shared obj obtained through
	 * dlopen() - use this ptr to get access
	 * to any symbols from a loaded module
	 * via dlsym()
	 */
	void            *handle;
	struct dll      *next;
};
.Ed
.Bl -inset
.It Va name
is the name of the module
.It Va handle
is a pointer to the module obtained through
.Xr dlopen 3
.El
Whenever a module is loaded in userland, an entry is added to
.Va dll_chain ,
then every protocol handler present in that module
is resolved and registered in
.Va handler_chain .
.Ss HOW TO WRITE A MODULE FOR LIBALIAS
There is a module (called
.Pa alias_dummy.[ch] )
in
.Nm
that can be used as a skeleton for future work, here we analyse some parts of that
module.
From
.Pa alias_dummy.c :
.Bd -literal
struct proto_handler handlers [] = {{666, IN|OUT, UDP|TCP,
				    &fingerprint, &protohandler}};
.Ed
.Pp
The variable
.Va handlers
is the
.Dq "most important thing"
in a module
cause it describes the handlers present and lets the outside world use
it in an opaque way.
.Pp
It must ALWAYS be present in every module, and it MUST retain
the name
.Va handlers ,
otherwise attempting to load a module in userland will fail and
complain about missing symbols: for more information about module
load/unload, please refer to
.Fn LibAliasRefreshModules ,
.Fn LibAliasLoadModule
and
.Fn LibAliasUnloadModule
in
.Pa alias.c .
.Pp
.Va handlers
contains all the
.Vt proto_handler
structures present in a module.
.Bd -literal
static int
mod_handler(module_t mod, int type, void *data)
{
	int error;

	switch (type) {
	case MOD_LOAD:
		error = 0;
		attach_handlers(handlers);
		break;
	case MOD_UNLOAD:
		error = 0;
		detach_handlers(handlers;
		break;
	default:
		error = EINVAL;
	}
	return (error);
}
.Ed
When running as KLD,
.Fn mod_handler
register/deregister the module using
.Fn attach_handlers
and
.Fn detach_handlers ,
respectively.
.Pp
Every module must contain at least 2 functions: one fingerprint
function and a protocol handler function.
.Bd -literal
#ifdef _KERNEL
static
#endif
int
fingerprint(struct libalias *la, struct ip *pip, struct alias_data *ah)
{

\&...
}

#ifdef _KERNEL
static
#endif
int
protohandler(struct libalias *la, struct ip *pip,
             struct alias_data *ah)
{

\&...
}
.Ed
and they must accept exactly these input parameters.
.Ss PATCHING AN APPLICATION FOR USERLAND LIBALIAS MODULES
To add module support into an application that uses
.Nm ,
the following simple steps can be followed.
.Bl -enum
.It
Find the main file of an application
(let us call it
.Pa main.c ) .
.It
Add this to the header section of
.Pa main.c ,
if not already present:
.Pp
.Dl "#include <signal.h>"
.Pp
and this just after the header section:
.Pp
.Dl "static void signal_handler(int);"
.It
Add the following line to the init function of an application or,
if it does not have any init function, put it in
.Fn main :
.Pp
.Dl "signal(SIGHUP, signal_handler);"
.Pp
and place the
.Fn signal_handler
function somewhere in
.Pa main.c :
.Bd -literal -offset indent
static void
signal_handler(int sig)
{

	LibAliasRefreshModules();
}
.Ed
.Pp
Otherwise, if an application already traps the
.Dv SIGHUP
signal, just add a call to
.Fn LibAliasRefreshModules
in the signal handler function.
.El
For example, to patch
.Xr natd 8
to use
.Nm
modules, just add the following line to
.Fn RefreshAddr "int sig __unused" :
.Pp
.Dl "LibAliasRefreshModules()"
.Pp
recompile and you are done.
.Ss LOGGING SUPPORT IN KERNEL LAND
When working as KLD,
.Nm
now has log support that
happens on a buffer allocated inside
.Vt "struct libalias"
(from
.Pa alias_local.h ) :
.Bd -literal
struct libalias {
       ...

	/* log descriptor        */
#ifdef	KERNEL_LOG
	char           *logDesc;        /*
					 * ptr to an auto-malloced
					 * memory buffer when libalias
					 * works as kld
					 */
#else
	FILE           *logDesc;	/*
					 * ptr to /var/log/alias.log
					 * when libalias runs as a
					 * userland lib
					 */
#endif

	...
}
.Ed
so all applications using
.Nm
will be able to handle their
own logs, if they want, accessing
.Va logDesc .
Moreover, every change to a log buffer is automatically added to
.Xr syslog 3
with the
.Dv LOG_SECURITY
facility and the
.Dv LOG_INFO
level.