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: releng/11.0/sys/netinet/libalias/libalias.3 298995 2016-05-03 18:05:43Z pfg $
.Dd December 25, 2013 .Dt LIBALIAS 3 .Os .Sh NAME .Nm libalias .Nd packet aliasing library for masquerading and network address translation .Sh SYNOPSIS n sys/types.h n netinet/in.h n alias.h
p 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.
p 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 the 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. .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 must be 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 .
p .Ft "struct libalias *" .Fn LibAliasInit "struct libalias *" d -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.
p l -item -offset indent -compact t .Dv PKT_ALIAS_SAME_PORTS t .Dv PKT_ALIAS_USE_SOCKETS t .Dv PKT_ALIAS_RESET_ON_ADDR_CHANGE .El
p 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 .
p It is mandatory that this function be called at the beginning of a program prior to any packet handling. .Ed
p .Ft void .Fn LibAliasUninit "struct libalias *" d -ragged -offset indent This function has no return value and is used to clear any resources attached to internal data structures.
p This function should be called when a program stops using the aliasing engine; amongst other things, it clears out any firewall holes. To provide backwards compatibility and extra security, it is added to the .Xr atexit 3 chain by .Fn LibAliasInit . .Ed
p .Ft void .Fn LibAliasSetAddress "struct libalias *" "struct in_addr addr" d -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 has not been called, and no static rules match, an outgoing packet retains its source address.
p 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.
p 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).
p It is mandatory that this function be called prior to any packet handling. .Ed
p .Ft unsigned int .Fn LibAliasSetMode "struct libalias *" "unsigned int flags" "unsigned int mask" d -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 n alias.h : l -tag -width indent t Dv PKT_ALIAS_LOG Enables logging into
a /var/log/alias.log . Each time an aliasing link is created or deleted, the log file is appended to 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 . t 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
o .Fn LibAliasIn returns .Dv PKT_ALIAS_IGNORED code
c 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. t 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. t 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. t 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:
p 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)
p This option is useful in the case that the 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. t 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. t Dv PKT_ALIAS_PUNCH_FW This option makes .Nm .Dq 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 clearing will also happen on the initial call to .Fn LibAliasSetFWBase , which must happen prior to setting this flag. t 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. t 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. t Dv PKT_ALIAS_SKIP_GLOBAL This option is used by
a ipfw_nat only. Specifying it as a flag to .Fn LibAliasSetMode has no effect. See section .Sx NETWORK ADDRESS TRANSLATION in .Xr ipfw 8 for more details. .El .Ed
p .Ft void .Fn LibAliasSetFWBase "struct libalias *" "unsigned int base" "unsigned int num" d -ragged -offset indent Set the firewall range allocated for punching firewall holes (with the .Dv PKT_ALIAS_PUNCH_FW flag). The range is cleared for all rules on initialization. .Ed
p .Ft void .Fn LibAliasSkinnyPort "struct libalias *" "unsigned int port" d -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.
p Along with .Fn LibAliasInit and .Fn LibAliasSetAddress , the two packet handling functions, .Fn LibAliasIn and .Fn LibAliasOut , comprise the minimal set of functions needed for a basic IP masquerading implementation.
p .Ft int .Fn LibAliasIn "struct libalias *" "char *buffer" "int maxpacketsize" d -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.
p Return codes: l -tag -width indent t Dv PKT_ALIAS_OK The packet aliasing process was successful. t Dv PKT_ALIAS_IGNORED The packet was ignored and not de-aliased. This can happen if the protocol is unrecognized, as for an ICMP message type that is not handled, or if incoming packets for new connections are being ignored (if the .Dv PKT_ALIAS_DENY_INCOMING mode bit was set using .Fn LibAliasSetMode ) . t 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. t 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 . t Dv PKT_ALIAS_ERROR An internal error within the packet aliasing engine occurred. .El .Ed
p .Ft int .Fn LibAliasOut "struct libalias *" "char *buffer" "int maxpacketsize" d -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.
p Return codes: l -tag -width indent t Dv PKT_ALIAS_OK The packet aliasing process was successful. t 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. t 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.
p .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 d -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 n netinet/in.h .
p 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.
p If the link is further set up to operate with 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.
p 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. The remote port specification will almost always 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.
p 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.
p 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
p .Ft struct alias_link * .Fo LibAliasRedirectAddr .Fa "struct libalias *" .Fa "struct in_addr local_addr" .Fa "struct in_addr alias_addr" .Fc d -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 .
p 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.
p If the link is further set up to operate with load sharing, then the .Fa local_addr argument is ignored, and is selected dynamically from the server pool, as described in .Fn LibAliasAddServer below.
p 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:
p 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"));
p 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.
p Any calls to .Fn LibAliasRedirectPort will have precedence over address mappings designated by .Fn LibAliasRedirectAddr .
p 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
p .Ft int .Fo LibAliasAddServer .Fa "struct libalias *" .Fa "struct alias_link *link" .Fa "struct in_addr addr" .Fa "u_short port" .Fc d -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 the server pool, 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 the load balance across server pool hosts when the sessions are initiated. If load sharing is desired for just a few specific services, the configuration on LSNAT could be defined to restrict load sharing to just the services desired.
p 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.
p 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.
p For links created with .Fn LibAliasRedirectAddr , the .Fa port argument is ignored and could have any value, e.g. htons(~0).
p This function returns 0 on success, -1 otherwise. .Ed
p .Ft int .Fn LibAliasRedirectDynamic "struct libalias *" "struct alias_link *link" d -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.)
p This function returns 0 on success, -1 otherwise. .Ed
p .Ft void .Fn LibAliasRedirectDelete "struct libalias *" "struct alias_link *link" d -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 care is needed when using this function. .Ed
p .Ft int .Fn LibAliasProxyRule "struct libalias *" "const char *cmd" d -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: l -tag -width indent t 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 . t Cm port Ar portnum Only packets with the destination port .Ar portnum are proxied. t 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.
p The .Ar server specification is mandatory unless the .Cm delete command is being used. t 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. t 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. t Cm proto tcp | udp If specified, only packets of the given protocol type are matched. t 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. t 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
p 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
p .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 d -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 will be redirected to a specified local address.
p 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.
p 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.
p If two calls to .Fn LibAliasRedirectProto overlap in their address specifications, then the most recent call will have precedence.
p 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.
p 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 .
p 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.
p .Ft int .Fn LibAliasSaveFragment "struct libalias *" "char *ptr" d -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.
p 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.]
p This function returns .Dv PKT_ALIAS_OK if it was successful and .Dv PKT_ALIAS_ERROR if there was an error. .Ed
p .Ft char * .Fn LibAliasGetFragment "struct libalias *" "char *buffer" d -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.
p The .Fn LibAliasGetFragment function can be called sequentially until there are no more fragments available, at which time it returns .Dv NULL . .Ed
p .Ft void .Fn LibAliasFragmentIn "struct libalias *" "char *header" "char *fragment" d -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 struct alias_link * .Fn AddLink "struct libalias *" "struct in_addr src_addr" "struct in_addr dst_addr" \ "struct in_addr alias_addr" "u_short src_port" "u_short dst_port" \ "int alias_param" "int link_type" d -ragged -offset indent This function adds new state to the instance hash table. The dst_address and/or dst_port may be given as zero, which introduces some dynamic character into the link, since LibAliasSetAddress can change the address that is used. However, in the current implementation, such links can only be used for inbound (ext -> int) traffic. .Ed
p .Ft void .Fn LibAliasSetTarget "struct libalias *" "struct in_addr addr" d -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 .
p 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 .
p 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
p .Ft int .Fn LibAliasCheckNewLink "struct libalias *" d -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
p .Ft u_short .Fn LibAliasInternetChecksum "struct libalias *" "u_short *buffer" "int nbytes" d -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).
p 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.
p 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
p .Ft int .Fn LibAliasUnaliasOut "struct libalias *" "char *buffer" "int maxpacketsize" d -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 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.
p 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: d -literal -offset indent (local addr, local port, alias addr, alias port, remote addr, remote port, protocol) .Ed
p 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.
p 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.)
p 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.
p For instance, a partially specified link might be d -literal -offset indent (192.168.0.4, 23, 204.228.203.215, 8066, 0, 0, tcp) .Ed
p 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. d -literal -offset indent (local addr, alias addr) .Ed
p Address mappings are searched when creating new dynamic links.
p 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.
p 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 Po AND Xr ipfw 4 SUPPORT Pc 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
a alias_ prefix.
p To add support for a new protocol, load the corresponding module. For example:
p .Dl "kldload alias_ftp"
p When support for a protocol is no longer needed, its module can be unloaded:
p .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.
p While compiled for a userland .Nm , all the modules are plain libraries, residing in
a /usr/lib , and recognizable with the
a libalias_ prefix.
p There is a configuration file,
a /etc/libalias.conf , with the following contents (by default): d -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
p This file contains the paths to the modules that .Nm will load. To load/unload a new module, just add its path to
a 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 every time you want to refresh the loaded modules, send it the .Dv SIGHUP signal:
p .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
a alias_mod.c : d -literal /* Protocol and userland module handlers chains. */ LIST_HEAD(handler_chain, proto_handler) handler_chain ... ... SLIST_HEAD(dll_chain, dll) dll_chain ... .Ed
p .Va handler_chain keeps track of all the protocol handlers loaded, while .Va ddl_chain tracks which userland modules are loaded.
p .Va handler_chain is composed of .Vt "struct proto_handler" entries: d -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); TAILQ_ENTRY(proto_handler) link; }; .Ed
p where: l -inset t Va pri is the priority assigned to a protocol handler; lower priority is better. t Va dir is the direction of packets: ingoing or outgoing. t Va proto indicates to which protocol this packet belongs: IP, TCP or UDP. t Va fingerprint points to the fingerprint function while protohandler points to the protocol handler function. .El
p The .Va fingerprint function has the dual role of checking if the incoming packet is found, and if it belongs to any categories that this module can handle.
p The .Va protohandler function actually manipulates the packet to make .Nm correctly NAT it.
p 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.
p 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.
p Inside .Nm , the module hook looks like this: d -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
p 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 the handler chain; it receives as input parameters: l -tag -width indent t Fa IN direction t Fa UDP working protocol t Fa la pointer to this instance of libalias t Fa pip pointer to a .Vt "struct ip" t Fa ad pointer to .Vt "struct alias_data" (see above) .El
p In this case, .Fn find_handler will search only for modules registered for supporting INcoming UDP packets.
p As was mentioned earlier, .Nm in userland is a bit different, as care must be taken in module handling as well (avoiding duplicate load of modules, avoiding modules with same name, etc.) so .Va dll_chain was introduced.
p .Va dll_chain contains a list of all userland .Nm modules loaded.
p When an application calls .Fn LibAliasRefreshModules , .Nm first unloads all the loaded modules, then reloads all the modules listed in
a /etc/libalias.conf : for every module loaded, a new entry is added to .Va dll_chain .
p .Va dll_chain is composed of .Vt "struct dll" entries: d -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 l -inset t Va name is the name of the module. t 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
a alias_dummy.[ch] ) in .Nm that can be used as a skeleton for future work. Here we analyse some parts of that module. From
a alias_dummy.c : d -literal struct proto_handler handlers[] = { { .pri = 666, .dir = IN|OUT, .proto = UDP|TCP, .fingerprint = fingerprint, .protohandler= protohandler, }, { EOH } }; .Ed
p The variable .Va handlers is the .Dq "most important thing" in a module since it describes the handlers present and lets the outside world use it in an opaque way.
p 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
a alias.c .
p .Va handlers contains all the .Vt proto_handler structures present in a module. d -literal static int mod_handler(module_t mod, int type, void *data) { int error; switch (type) { case MOD_LOAD: error = LibAliasAttachHandlers(handlers); break; case MOD_UNLOAD: error = LibAliasDetachHandlers(handlers); break; default: error = EINVAL; } return (error); } .Ed When running as KLD, .Fn mod_handler registers/deregisters the module using .Fn LibAliasAttachHandlers and .Fn LibAliasDetachHandlers , respectively.
p Every module must contain at least 2 functions: one fingerprint function and a protocol handler function. d -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. l -enum t Find the main file of an application (let us call it
a main.c ) . t Add this to the header section of
a main.c , if not already present:
p .Dl "#include <signal.h>"
p and this just after the header section:
p .Dl "static void signal_handler(int);" t 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 :
p .Dl "signal(SIGHUP, signal_handler);"
p and place the .Fn signal_handler function somewhere in
a main.c : d -literal -offset indent static void signal_handler(int sig) { LibAliasRefreshModules(); } .Ed
p 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" :
p .Dl "LibAliasRefreshModules()"
p 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
a alias_local.h ) : d -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. .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 Gleb Smirnoff Aq glebius@FreeBSD.org ported the library to kernel space. .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 ACKNOWLEDGEMENTS Listed below, in approximate chronological order, are individuals who have provided valuable comments and/or debugging assistance. d -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