freebsd-skq/lib/libalias/alias_db.c

2951 lines
77 KiB
C
Raw Normal View History

/* -*- mode: c; tab-width: 8; c-basic-indent: 4; -*- */
/*-
* 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.
*/
2001-09-30 21:03:33 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
/*
Alias_db.c encapsulates all data structures used for storing
packet aliasing data. Other parts of the aliasing software
access data through functions provided in this file.
Data storage is based on the notion of a "link", which is
established for ICMP echo/reply packets, UDP datagrams and
TCP stream connections. A link stores the original source
and destination addresses. For UDP and TCP, it also stores
source and destination port numbers, as well as an alias
port number. Links are also used to store information about
fragments.
There is a facility for sweeping through and deleting old
links as new packets are sent through. A simple timeout is
used for ICMP and UDP links. TCP links are left alone unless
there is an incomplete connection, in which case the link
can be deleted after a certain amount of time.
Initial version: August, 1996 (cjm)
Version 1.4: September 16, 1996 (cjm)
Facility for handling incoming links added.
Version 1.6: September 18, 1996 (cjm)
ICMP data handling simplified.
Version 1.7: January 9, 1997 (cjm)
Fragment handling simplified.
Saves pointers for unresolved fragments.
Permits links for unspecified remote ports
or unspecified remote addresses.
Fixed bug which did not properly zero port
table entries after a link was deleted.
Cleaned up some obsolete comments.
Version 1.8: January 14, 1997 (cjm)
Fixed data type error in StartPoint().
(This error did not exist prior to v1.7
and was discovered and fixed by Ari Suutari)
Version 1.9: February 1, 1997
Optionally, connections initiated from packet aliasing host
machine will will not have their port number aliased unless it
conflicts with an aliasing port already being used. (cjm)
All options earlier being #ifdef'ed are now available through
a new interface, SetPacketAliasMode(). This allows run time
control (which is now available in PPP+pktAlias through the
'alias' keyword). (ee)
Added ability to create an alias port without
either destination address or port specified.
port type = ALIAS_PORT_UNKNOWN_DEST_ALL (ee)
Removed K&R style function headers
and general cleanup. (ee)
Added packetAliasMode to replace compiler #defines's (ee)
Allocates sockets for partially specified
ports if ALIAS_USE_SOCKETS defined. (cjm)
Version 2.0: March, 1997
SetAliasAddress() will now clean up alias links
if the aliasing address is changed. (cjm)
PacketAliasPermanentLink() function added to support permanent
links. (J. Fortes suggested the need for this.)
Examples:
(192.168.0.1, port 23) <-> alias port 6002, unknown dest addr/port
(192.168.0.2, port 21) <-> alias port 3604, known dest addr
unknown dest port
These permanent links allow for incoming connections to
machines on the local network. They can be given with a
user-chosen amount of specificity, with increasing specificity
meaning more security. (cjm)
Quite a bit of rework to the basic engine. The portTable[]
array, which kept track of which ports were in use was replaced
by a table/linked list structure. (cjm)
SetExpire() function added. (cjm)
DeleteLink() no longer frees memory association with a pointer
to a fragment (this bug was first recognized by E. Eklund in
v1.9).
Version 2.1: May, 1997 (cjm)
Packet aliasing engine reworked so that it can handle
multiple external addresses rather than just a single
host address.
PacketAliasRedirectPort() and PacketAliasRedirectAddr()
added to the API. The first function is a more generalized
version of PacketAliasPermanentLink(). The second function
implements static network address translation.
Version 3.2: July, 2000 (salander and satoh)
2002-07-01 11:19:40 +00:00
Added FindNewPortGroup to get contiguous range of port values.
Added QueryUdpTcpIn and QueryUdpTcpOut to look for an aliasing
link but not actually add one.
2002-07-01 11:19:40 +00:00
Added FindRtspOut, which is closely derived from FindUdpTcpOut,
except that the alias port (from FindNewPortGroup) is provided
as input.
See HISTORY file for additional revisions.
*/
/* System include files */
1999-09-21 01:26:49 +00:00
#include <errno.h>
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/queue.h>
#include <sys/socket.h>
#include <sys/time.h>
#include <sys/types.h>
/* BSD network include files */
#include <netinet/in_systm.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <arpa/inet.h>
#include "alias.h"
#include "alias_local.h"
/*
Constants (note: constants are also defined
near relevant functions or structs)
*/
/* Sizes of input and output link tables */
#define LINK_TABLE_OUT_SIZE 101
#define LINK_TABLE_IN_SIZE 4001
/* Parameters used for cleanup of expired links */
#define ALIAS_CLEANUP_INTERVAL_SECS 60
#define ALIAS_CLEANUP_MAX_SPOKES 30
/* Timeouts (in seconds) for different link types */
#define ICMP_EXPIRE_TIME 60
#define UDP_EXPIRE_TIME 60
#define PROTO_EXPIRE_TIME 60
#define FRAGMENT_ID_EXPIRE_TIME 10
#define FRAGMENT_PTR_EXPIRE_TIME 30
/* TCP link expire time for different cases */
/* When the link has been used and closed - minimal grace time to
allow ACKs and potential re-connect in FTP (XXX - is this allowed?) */
#ifndef TCP_EXPIRE_DEAD
# define TCP_EXPIRE_DEAD 10
#endif
/* When the link has been used and closed on one side - the other side
is allowed to still send data */
#ifndef TCP_EXPIRE_SINGLEDEAD
# define TCP_EXPIRE_SINGLEDEAD 90
#endif
/* When the link isn't yet up */
#ifndef TCP_EXPIRE_INITIAL
# define TCP_EXPIRE_INITIAL 300
#endif
/* When the link is up */
#ifndef TCP_EXPIRE_CONNECTED
# define TCP_EXPIRE_CONNECTED 86400
#endif
/* Dummy port number codes used for FindLinkIn/Out() and AddLink().
These constants can be anything except zero, which indicates an
unknown port number. */
#define NO_DEST_PORT 1
#define NO_SRC_PORT 1
/* Data Structures
The fundamental data structure used in this program is
"struct alias_link". Whenever a TCP connection is made,
a UDP datagram is sent out, or an ICMP echo request is made,
a link record is made (if it has not already been created).
The link record is identified by the source address/port
and the destination address/port. In the case of an ICMP
echo request, the source port is treated as being equivalent
with the 16-bit ID number of the ICMP packet.
The link record also can store some auxiliary data. For
TCP connections that have had sequence and acknowledgment
modifications, data space is available to track these changes.
A state field is used to keep track in changes to the TCP
connection state. ID numbers of fragments can also be
stored in the auxiliary space. Pointers to unresolved
fragments can also be stored.
The link records support two independent chainings. Lookup
tables for input and out tables hold the initial pointers
the link chains. On input, the lookup table indexes on alias
port and link type. On output, the lookup table indexes on
source address, destination address, source port, destination
port and link type.
*/
struct ack_data_record /* used to save changes to ACK/sequence numbers */
{
u_long ack_old;
u_long ack_new;
int delta;
int active;
};
struct tcp_state /* Information about TCP connection */
{
int in; /* State for outside -> inside */
int out; /* State for inside -> outside */
int index; /* Index to ACK data array */
int ack_modified; /* Indicates whether ACK and sequence numbers */
/* been modified */
};
#define N_LINK_TCP_DATA 3 /* Number of distinct ACK number changes
saved for a modified TCP stream */
struct tcp_dat
{
struct tcp_state state;
struct ack_data_record ack[N_LINK_TCP_DATA];
int fwhole; /* Which firewall record is used for this hole? */
};
struct server /* LSNAT server pool (circular list) */
{
struct in_addr addr;
u_short port;
struct server *next;
};
struct alias_link /* Main data structure */
{
struct in_addr src_addr; /* Address and port information */
struct in_addr dst_addr;
struct in_addr alias_addr;
struct in_addr proxy_addr;
u_short src_port;
u_short dst_port;
u_short alias_port;
u_short proxy_port;
struct server *server;
int link_type; /* Type of link: TCP, UDP, ICMP, proto, frag */
/* values for link_type */
#define LINK_ICMP IPPROTO_ICMP
#define LINK_UDP IPPROTO_UDP
#define LINK_TCP IPPROTO_TCP
#define LINK_FRAGMENT_ID (IPPROTO_MAX + 1)
#define LINK_FRAGMENT_PTR (IPPROTO_MAX + 2)
#define LINK_ADDR (IPPROTO_MAX + 3)
#define LINK_PPTP (IPPROTO_MAX + 4)
int flags; /* indicates special characteristics */
int pflags; /* protocol-specific flags */
/* flag bits */
#define LINK_UNKNOWN_DEST_PORT 0x01
#define LINK_UNKNOWN_DEST_ADDR 0x02
#define LINK_PERMANENT 0x04
#define LINK_PARTIALLY_SPECIFIED 0x03 /* logical-or of first two bits */
#define LINK_UNFIREWALLED 0x08
int timestamp; /* Time link was last accessed */
int expire_time; /* Expire time for link */
int sockfd; /* socket descriptor */
LIST_ENTRY(alias_link) list_out; /* Linked list of pointers for */
LIST_ENTRY(alias_link) list_in; /* input and output lookup tables */
union /* Auxiliary data */
{
char *frag_ptr;
struct in_addr frag_addr;
struct tcp_dat *tcp;
} data;
};
2002-07-01 11:19:40 +00:00
/* Global Variables
The global variables listed here are only accessed from
2002-07-01 11:19:40 +00:00
within alias_db.c and so are prefixed with the static
designation.
*/
2002-07-01 11:19:40 +00:00
int packetAliasMode; /* Mode flags */
/* - documented in alias.h */
static struct in_addr aliasAddress; /* Address written onto source */
/* field of IP packet. */
static struct in_addr targetAddress; /* IP address incoming packets */
/* are sent to if no aliasing */
/* link already exists */
static struct in_addr nullAddress; /* Used as a dummy parameter for */
/* some function calls */
static LIST_HEAD(, alias_link)
linkTableOut[LINK_TABLE_OUT_SIZE]; /* Lookup table of pointers to */
/* chains of link records. Each */
static LIST_HEAD(, alias_link) /* link record is doubly indexed */
linkTableIn[LINK_TABLE_IN_SIZE]; /* into input and output lookup */
/* tables. */
static int icmpLinkCount; /* Link statistics */
static int udpLinkCount;
static int tcpLinkCount;
static int pptpLinkCount;
static int protoLinkCount;
static int fragmentIdLinkCount;
static int fragmentPtrLinkCount;
static int sockCount;
static int cleanupIndex; /* Index to chain of link table */
/* being inspected for old links */
static int timeStamp; /* System time in seconds for */
/* current packet */
static int lastCleanupTime; /* Last time IncrementalCleanup() */
/* was called */
static int houseKeepingResidual; /* used by HouseKeeping() */
static int deleteAllLinks; /* If equal to zero, DeleteLink() */
/* will not remove permanent links */
static FILE *monitorFile; /* File descriptor for link */
/* statistics monitoring file */
static int newDefaultLink; /* Indicates if a new aliasing */
/* link has been created after a */
/* call to PacketAliasIn/Out(). */
2002-07-01 11:19:40 +00:00
#ifndef NO_FW_PUNCH
static int fireWallFD = -1; /* File descriptor to be able to */
/* control firewall. Opened by */
/* PacketAliasSetMode on first */
/* setting the PKT_ALIAS_PUNCH_FW */
/* flag. */
#endif
/* Internal utility routines (used only in alias_db.c)
Lookup table starting points:
StartPointIn() -- link table initial search point for
incoming packets
StartPointOut() -- link table initial search point for
outgoing packets
2002-07-01 11:19:40 +00:00
Miscellaneous:
SeqDiff() -- difference between two TCP sequences
ShowAliasStats() -- send alias statistics to a monitor file
*/
/* Local prototypes */
static u_int StartPointIn(struct in_addr, u_short, int);
static u_int StartPointOut(struct in_addr, struct in_addr,
u_short, u_short, int);
static int SeqDiff(u_long, u_long);
static void ShowAliasStats(void);
#ifndef NO_FW_PUNCH
/* Firewall control */
static void InitPunchFW(void);
static void UninitPunchFW(void);
static void ClearFWHole(struct alias_link *link);
#endif
/* Log file control */
static void InitPacketAliasLog(void);
static void UninitPacketAliasLog(void);
static u_int
StartPointIn(struct in_addr alias_addr,
u_short alias_port,
int link_type)
{
u_int n;
n = alias_addr.s_addr;
if (link_type != LINK_PPTP)
n += alias_port;
n += link_type;
return(n % LINK_TABLE_IN_SIZE);
}
static u_int
StartPointOut(struct in_addr src_addr, struct in_addr dst_addr,
u_short src_port, u_short dst_port, int link_type)
{
u_int n;
n = src_addr.s_addr;
n += dst_addr.s_addr;
if (link_type != LINK_PPTP) {
2002-07-01 11:19:40 +00:00
n += src_port;
n += dst_port;
}
n += link_type;
return(n % LINK_TABLE_OUT_SIZE);
}
static int
SeqDiff(u_long x, u_long y)
{
/* Return the difference between two TCP sequence numbers */
/*
This function is encapsulated in case there are any unusual
arithmetic conditions that need to be considered.
*/
return (ntohl(y) - ntohl(x));
}
static void
ShowAliasStats(void)
{
/* Used for debugging */
if (monitorFile)
{
fprintf(monitorFile, "icmp=%d, udp=%d, tcp=%d, pptp=%d, proto=%d, frag_id=%d frag_ptr=%d",
icmpLinkCount,
udpLinkCount,
tcpLinkCount,
pptpLinkCount,
protoLinkCount,
fragmentIdLinkCount,
fragmentPtrLinkCount);
fprintf(monitorFile, " / tot=%d (sock=%d)\n",
icmpLinkCount + udpLinkCount
+ tcpLinkCount
+ pptpLinkCount
+ protoLinkCount
+ fragmentIdLinkCount
+ fragmentPtrLinkCount,
sockCount);
fflush(monitorFile);
}
}
/* Internal routines for finding, deleting and adding links
Port Allocation:
GetNewPort() -- find and reserve new alias port number
GetSocket() -- try to allocate a socket for a given port
Link creation and deletion:
CleanupAliasData() - remove all link chains from lookup table
IncrementalCleanup() - look for stale links in a single chain
DeleteLink() - remove link
2002-07-01 11:19:40 +00:00
AddLink() - add link
ReLink() - change link
Link search:
FindLinkOut() - find link for outgoing packets
FindLinkIn() - find link for incoming packets
Port search:
2002-07-01 11:19:40 +00:00
FindNewPortGroup() - find an available group of ports
*/
/* Local prototypes */
static int GetNewPort(struct alias_link *, int);
static u_short GetSocket(u_short, int *, int);
static void CleanupAliasData(void);
static void IncrementalCleanup(void);
static void DeleteLink(struct alias_link *);
static struct alias_link *
AddLink(struct in_addr, struct in_addr, struct in_addr,
u_short, u_short, int, int);
static struct alias_link *
ReLink(struct alias_link *,
struct in_addr, struct in_addr, struct in_addr,
u_short, u_short, int, int);
static struct alias_link *
FindLinkOut(struct in_addr, struct in_addr, u_short, u_short, int, int);
static struct alias_link *
FindLinkIn(struct in_addr, struct in_addr, u_short, u_short, int, int);
#define ALIAS_PORT_BASE 0x08000
#define ALIAS_PORT_MASK 0x07fff
#define ALIAS_PORT_MASK_EVEN 0x07ffe
#define GET_NEW_PORT_MAX_ATTEMPTS 20
#define GET_ALIAS_PORT -1
#define GET_ALIAS_ID GET_ALIAS_PORT
#define FIND_EVEN_ALIAS_BASE 1
/* GetNewPort() allocates port numbers. Note that if a port number
is already in use, that does not mean that it cannot be used by
another link concurrently. This is because GetNewPort() looks for
unused triplets: (dest addr, dest port, alias port). */
static int
GetNewPort(struct alias_link *link, int alias_port_param)
{
int i;
int max_trials;
u_short port_sys;
u_short port_net;
/*
Description of alias_port_param for GetNewPort(). When
this parameter is zero or positive, it precisely specifies
the port number. GetNewPort() will return this number
without check that it is in use.
When this parameter is GET_ALIAS_PORT, it indicates to get a randomly
selected port number.
*/
2002-07-01 11:19:40 +00:00
if (alias_port_param == GET_ALIAS_PORT)
{
/*
* The aliasing port is automatically selected
* by one of two methods below:
*/
max_trials = GET_NEW_PORT_MAX_ATTEMPTS;
if (packetAliasMode & PKT_ALIAS_SAME_PORTS)
{
/*
* When the PKT_ALIAS_SAME_PORTS option is
* chosen, the first try will be the
* actual source port. If this is already
* in use, the remainder of the trials
* will be random.
*/
port_net = link->src_port;
port_sys = ntohs(port_net);
}
else
{
/* First trial and all subsequent are random. */
port_sys = random() & ALIAS_PORT_MASK;
port_sys += ALIAS_PORT_BASE;
port_net = htons(port_sys);
}
}
else if (alias_port_param >= 0 && alias_port_param < 0x10000)
{
link->alias_port = (u_short) alias_port_param;
return(0);
}
else
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/GetNewPort(): ");
fprintf(stderr, "input parameter error\n");
#endif
return(-1);
}
/* Port number search */
for (i=0; i<max_trials; i++)
{
int go_ahead;
struct alias_link *search_result;
search_result = FindLinkIn(link->dst_addr, link->alias_addr,
link->dst_port, port_net,
link->link_type, 0);
if (search_result == NULL)
go_ahead = 1;
else if (!(link->flags & LINK_PARTIALLY_SPECIFIED)
&& (search_result->flags & LINK_PARTIALLY_SPECIFIED))
go_ahead = 1;
else
go_ahead = 0;
if (go_ahead)
{
if ((packetAliasMode & PKT_ALIAS_USE_SOCKETS)
&& (link->flags & LINK_PARTIALLY_SPECIFIED)
2002-07-01 11:19:40 +00:00
&& ((link->link_type == LINK_TCP) ||
(link->link_type == LINK_UDP)))
{
if (GetSocket(port_net, &link->sockfd, link->link_type))
{
link->alias_port = port_net;
return(0);
}
}
else
{
link->alias_port = port_net;
return(0);
}
}
port_sys = random() & ALIAS_PORT_MASK;
port_sys += ALIAS_PORT_BASE;
port_net = htons(port_sys);
}
#ifdef DEBUG
fprintf(stderr, "PacketAlias/GetnewPort(): ");
fprintf(stderr, "could not find free port\n");
#endif
return(-1);
}
2002-07-01 11:19:40 +00:00
static u_short
GetSocket(u_short port_net, int *sockfd, int link_type)
{
int err;
int sock;
struct sockaddr_in sock_addr;
if (link_type == LINK_TCP)
sock = socket(AF_INET, SOCK_STREAM, 0);
else if (link_type == LINK_UDP)
sock = socket(AF_INET, SOCK_DGRAM, 0);
else
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/GetSocket(): ");
fprintf(stderr, "incorrect link type\n");
#endif
return(0);
}
if (sock < 0)
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/GetSocket(): ");
fprintf(stderr, "socket() error %d\n", *sockfd);
#endif
return(0);
}
sock_addr.sin_family = AF_INET;
sock_addr.sin_addr.s_addr = htonl(INADDR_ANY);
sock_addr.sin_port = port_net;
err = bind(sock,
(struct sockaddr *) &sock_addr,
sizeof(sock_addr));
if (err == 0)
{
sockCount++;
*sockfd = sock;
return(1);
}
else
{
close(sock);
return(0);
}
}
2002-07-01 11:19:40 +00:00
/* FindNewPortGroup() returns a base port number for an available
range of contiguous port numbers. Note that if a port number
is already in use, that does not mean that it cannot be used by
another link concurrently. This is because FindNewPortGroup()
looks for unused triplets: (dest addr, dest port, alias port). */
int
FindNewPortGroup(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short src_port,
u_short dst_port,
2002-07-01 11:19:40 +00:00
u_short port_count,
u_char proto,
u_char align)
{
int i, j;
int max_trials;
u_short port_sys;
int link_type;
/*
* Get link_type from protocol
*/
switch (proto)
{
case IPPROTO_UDP:
link_type = LINK_UDP;
break;
case IPPROTO_TCP:
link_type = LINK_TCP;
break;
default:
return (0);
break;
}
/*
* The aliasing port is automatically selected
* by one of two methods below:
*/
max_trials = GET_NEW_PORT_MAX_ATTEMPTS;
if (packetAliasMode & PKT_ALIAS_SAME_PORTS) {
/*
* When the ALIAS_SAME_PORTS option is
* chosen, the first try will be the
* actual source port. If this is already
* in use, the remainder of the trials
* will be random.
*/
port_sys = ntohs(src_port);
} else {
/* First trial and all subsequent are random. */
if (align == FIND_EVEN_ALIAS_BASE)
port_sys = random() & ALIAS_PORT_MASK_EVEN;
else
port_sys = random() & ALIAS_PORT_MASK;
port_sys += ALIAS_PORT_BASE;
}
/* Port number search */
for (i = 0; i < max_trials; i++) {
struct alias_link *search_result;
2002-07-01 11:19:40 +00:00
for (j = 0; j < port_count; j++)
if (0 != (search_result = FindLinkIn(dst_addr, alias_addr,
dst_port, htons(port_sys + j),
link_type, 0)))
break;
/* Found a good range, return base */
if (j == port_count)
return (htons(port_sys));
/* Find a new base to try */
if (align == FIND_EVEN_ALIAS_BASE)
port_sys = random() & ALIAS_PORT_MASK_EVEN;
else
port_sys = random() & ALIAS_PORT_MASK;
port_sys += ALIAS_PORT_BASE;
}
#ifdef DEBUG
fprintf(stderr, "PacketAlias/FindNewPortGroup(): ");
fprintf(stderr, "could not find free port(s)\n");
#endif
return(0);
}
static void
CleanupAliasData(void)
{
struct alias_link *link;
int i, icount;
icount = 0;
for (i=0; i<LINK_TABLE_OUT_SIZE; i++)
{
link = LIST_FIRST(&linkTableOut[i]);
while (link != NULL)
{
struct alias_link *link_next;
link_next = LIST_NEXT(link, list_out);
icount++;
DeleteLink(link);
link = link_next;
}
}
cleanupIndex =0;
}
static void
IncrementalCleanup(void)
{
int icount;
struct alias_link *link;
icount = 0;
link = LIST_FIRST(&linkTableOut[cleanupIndex++]);
while (link != NULL)
{
int idelta;
struct alias_link *link_next;
link_next = LIST_NEXT(link, list_out);
idelta = timeStamp - link->timestamp;
switch (link->link_type)
{
case LINK_TCP:
if (idelta > link->expire_time)
{
struct tcp_dat *tcp_aux;
2002-07-01 11:19:40 +00:00
tcp_aux = link->data.tcp;
if (tcp_aux->state.in != ALIAS_TCP_STATE_CONNECTED
|| tcp_aux->state.out != ALIAS_TCP_STATE_CONNECTED)
{
DeleteLink(link);
icount++;
}
}
break;
default:
if (idelta > link->expire_time)
{
DeleteLink(link);
icount++;
}
break;
}
link = link_next;
}
if (cleanupIndex == LINK_TABLE_OUT_SIZE)
cleanupIndex = 0;
}
static void
DeleteLink(struct alias_link *link)
{
/* Don't do anything if the link is marked permanent */
if (deleteAllLinks == 0 && link->flags & LINK_PERMANENT)
return;
#ifndef NO_FW_PUNCH
/* Delete associated firewall hole, if any */
ClearFWHole(link);
#endif
/* Free memory allocated for LSNAT server pool */
if (link->server != NULL) {
struct server *head, *curr, *next;
head = curr = link->server;
do {
next = curr->next;
free(curr);
} while ((curr = next) != head);
}
/* Adjust output table pointers */
LIST_REMOVE(link, list_out);
/* Adjust input table pointers */
LIST_REMOVE(link, list_in);
/* Close socket, if one has been allocated */
if (link->sockfd != -1)
{
sockCount--;
close(link->sockfd);
}
/* Link-type dependent cleanup */
switch(link->link_type)
{
case LINK_ICMP:
icmpLinkCount--;
break;
case LINK_UDP:
udpLinkCount--;
break;
case LINK_TCP:
tcpLinkCount--;
free(link->data.tcp);
break;
case LINK_PPTP:
pptpLinkCount--;
break;
case LINK_FRAGMENT_ID:
fragmentIdLinkCount--;
break;
case LINK_FRAGMENT_PTR:
fragmentPtrLinkCount--;
if (link->data.frag_ptr != NULL)
free(link->data.frag_ptr);
break;
case LINK_ADDR:
break;
default:
protoLinkCount--;
break;
}
/* Free memory */
free(link);
/* Write statistics, if logging enabled */
if (packetAliasMode & PKT_ALIAS_LOG)
{
ShowAliasStats();
}
}
static struct alias_link *
AddLink(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_port_param, /* if less than zero, alias */
int link_type) /* port will be automatically */
{ /* chosen. If greater than */
u_int start_point; /* zero, equal to alias port */
struct alias_link *link;
link = malloc(sizeof(struct alias_link));
if (link != NULL)
{
/* Basic initialization */
link->src_addr = src_addr;
link->dst_addr = dst_addr;
link->alias_addr = alias_addr;
link->proxy_addr.s_addr = INADDR_ANY;
link->src_port = src_port;
link->dst_port = dst_port;
link->proxy_port = 0;
link->server = NULL;
link->link_type = link_type;
link->sockfd = -1;
link->flags = 0;
link->pflags = 0;
link->timestamp = timeStamp;
/* Expiration time */
switch (link_type)
{
case LINK_ICMP:
link->expire_time = ICMP_EXPIRE_TIME;
break;
case LINK_UDP:
link->expire_time = UDP_EXPIRE_TIME;
break;
case LINK_TCP:
link->expire_time = TCP_EXPIRE_INITIAL;
break;
case LINK_PPTP:
link->flags |= LINK_PERMANENT; /* no timeout. */
break;
case LINK_FRAGMENT_ID:
link->expire_time = FRAGMENT_ID_EXPIRE_TIME;
break;
case LINK_FRAGMENT_PTR:
link->expire_time = FRAGMENT_PTR_EXPIRE_TIME;
break;
case LINK_ADDR:
break;
default:
link->expire_time = PROTO_EXPIRE_TIME;
break;
}
/* Determine alias flags */
if (dst_addr.s_addr == INADDR_ANY)
link->flags |= LINK_UNKNOWN_DEST_ADDR;
if (dst_port == 0)
link->flags |= LINK_UNKNOWN_DEST_PORT;
/* Determine alias port */
if (GetNewPort(link, alias_port_param) != 0)
{
free(link);
return(NULL);
}
/* Link-type dependent initialization */
switch(link_type)
{
struct tcp_dat *aux_tcp;
case LINK_ICMP:
icmpLinkCount++;
break;
case LINK_UDP:
udpLinkCount++;
break;
case LINK_TCP:
aux_tcp = malloc(sizeof(struct tcp_dat));
if (aux_tcp != NULL)
{
int i;
tcpLinkCount++;
aux_tcp->state.in = ALIAS_TCP_STATE_NOT_CONNECTED;
aux_tcp->state.out = ALIAS_TCP_STATE_NOT_CONNECTED;
aux_tcp->state.index = 0;
aux_tcp->state.ack_modified = 0;
for (i=0; i<N_LINK_TCP_DATA; i++)
aux_tcp->ack[i].active = 0;
aux_tcp->fwhole = -1;
link->data.tcp = aux_tcp;
}
else
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/AddLink: ");
fprintf(stderr, " cannot allocate auxiliary TCP data\n");
#endif
free(link);
return (NULL);
}
break;
case LINK_PPTP:
pptpLinkCount++;
break;
case LINK_FRAGMENT_ID:
fragmentIdLinkCount++;
break;
case LINK_FRAGMENT_PTR:
fragmentPtrLinkCount++;
break;
case LINK_ADDR:
break;
default:
protoLinkCount++;
break;
}
/* Set up pointers for output lookup table */
2002-07-01 11:19:40 +00:00
start_point = StartPointOut(src_addr, dst_addr,
src_port, dst_port, link_type);
LIST_INSERT_HEAD(&linkTableOut[start_point], link, list_out);
/* Set up pointers for input lookup table */
2002-07-01 11:19:40 +00:00
start_point = StartPointIn(alias_addr, link->alias_port, link_type);
LIST_INSERT_HEAD(&linkTableIn[start_point], link, list_in);
}
else
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/AddLink(): ");
fprintf(stderr, "malloc() call failed.\n");
#endif
}
if (packetAliasMode & PKT_ALIAS_LOG)
{
ShowAliasStats();
}
return(link);
}
static struct alias_link *
ReLink(struct alias_link *old_link,
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_port_param, /* if less than zero, alias */
int link_type) /* port will be automatically */
{ /* chosen. If greater than */
struct alias_link *new_link; /* zero, equal to alias port */
new_link = AddLink(src_addr, dst_addr, alias_addr,
src_port, dst_port, alias_port_param,
link_type);
#ifndef NO_FW_PUNCH
if (new_link != NULL &&
old_link->link_type == LINK_TCP &&
old_link->data.tcp->fwhole > 0) {
PunchFWHole(new_link);
}
#endif
DeleteLink(old_link);
return new_link;
}
static struct alias_link *
_FindLinkOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_short src_port,
u_short dst_port,
int link_type,
int replace_partial_links)
{
u_int i;
struct alias_link *link;
i = StartPointOut(src_addr, dst_addr, src_port, dst_port, link_type);
LIST_FOREACH(link, &linkTableOut[i], list_out)
{
if (link->src_addr.s_addr == src_addr.s_addr
&& link->server == NULL
&& link->dst_addr.s_addr == dst_addr.s_addr
&& link->dst_port == dst_port
&& link->src_port == src_port
&& link->link_type == link_type)
{
link->timestamp = timeStamp;
break;
}
}
/* Search for partially specified links. */
if (link == NULL && replace_partial_links)
{
if (dst_port != 0 && dst_addr.s_addr != INADDR_ANY)
{
link = _FindLinkOut(src_addr, dst_addr, src_port, 0,
link_type, 0);
if (link == NULL)
link = _FindLinkOut(src_addr, nullAddress, src_port,
dst_port, link_type, 0);
}
if (link == NULL &&
(dst_port != 0 || dst_addr.s_addr != INADDR_ANY))
{
link = _FindLinkOut(src_addr, nullAddress, src_port, 0,
link_type, 0);
}
if (link != NULL)
{
link = ReLink(link,
src_addr, dst_addr, link->alias_addr,
src_port, dst_port, link->alias_port,
link_type);
}
}
return(link);
}
static struct alias_link *
FindLinkOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_short src_port,
u_short dst_port,
int link_type,
int replace_partial_links)
{
struct alias_link *link;
link = _FindLinkOut(src_addr, dst_addr, src_port, dst_port,
link_type, replace_partial_links);
if (link == NULL)
{
/* The following allows permanent links to be
specified as using the default source address
(i.e. device interface address) without knowing
in advance what that address is. */
if (aliasAddress.s_addr != INADDR_ANY &&
src_addr.s_addr == aliasAddress.s_addr)
{
link = _FindLinkOut(nullAddress, dst_addr, src_port, dst_port,
link_type, replace_partial_links);
}
}
return(link);
}
static struct alias_link *
_FindLinkIn(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short dst_port,
u_short alias_port,
int link_type,
int replace_partial_links)
{
int flags_in;
u_int start_point;
struct alias_link *link;
struct alias_link *link_fully_specified;
struct alias_link *link_unknown_all;
struct alias_link *link_unknown_dst_addr;
struct alias_link *link_unknown_dst_port;
/* Initialize pointers */
link_fully_specified = NULL;
link_unknown_all = NULL;
link_unknown_dst_addr = NULL;
link_unknown_dst_port = NULL;
/* If either the dest addr or port is unknown, the search
loop will have to know about this. */
flags_in = 0;
if (dst_addr.s_addr == INADDR_ANY)
flags_in |= LINK_UNKNOWN_DEST_ADDR;
if (dst_port == 0)
flags_in |= LINK_UNKNOWN_DEST_PORT;
/* Search loop */
start_point = StartPointIn(alias_addr, alias_port, link_type);
LIST_FOREACH(link, &linkTableIn[start_point], list_in)
{
int flags;
flags = flags_in | link->flags;
if (!(flags & LINK_PARTIALLY_SPECIFIED))
{
if (link->alias_addr.s_addr == alias_addr.s_addr
2002-07-01 11:19:40 +00:00
&& link->alias_port == alias_port
&& link->dst_addr.s_addr == dst_addr.s_addr
&& link->dst_port == dst_port
&& link->link_type == link_type)
{
link_fully_specified = link;
break;
}
}
else if ((flags & LINK_UNKNOWN_DEST_ADDR)
&& (flags & LINK_UNKNOWN_DEST_PORT))
{
if (link->alias_addr.s_addr == alias_addr.s_addr
&& link->alias_port == alias_port
&& link->link_type == link_type)
{
if (link_unknown_all == NULL)
link_unknown_all = link;
}
}
else if (flags & LINK_UNKNOWN_DEST_ADDR)
{
if (link->alias_addr.s_addr == alias_addr.s_addr
&& link->alias_port == alias_port
&& link->link_type == link_type
&& link->dst_port == dst_port)
{
if (link_unknown_dst_addr == NULL)
link_unknown_dst_addr = link;
}
}
else if (flags & LINK_UNKNOWN_DEST_PORT)
{
if (link->alias_addr.s_addr == alias_addr.s_addr
&& link->alias_port == alias_port
&& link->link_type == link_type
&& link->dst_addr.s_addr == dst_addr.s_addr)
{
if (link_unknown_dst_port == NULL)
link_unknown_dst_port = link;
}
}
}
if (link_fully_specified != NULL)
{
link_fully_specified->timestamp = timeStamp;
link = link_fully_specified;
}
else if (link_unknown_dst_port != NULL)
link = link_unknown_dst_port;
else if (link_unknown_dst_addr != NULL)
link = link_unknown_dst_addr;
else if (link_unknown_all != NULL)
link = link_unknown_all;
else
return (NULL);
if (replace_partial_links &&
(link->flags & LINK_PARTIALLY_SPECIFIED || link->server != NULL))
{
struct in_addr src_addr;
u_short src_port;
if (link->server != NULL) { /* LSNAT link */
src_addr = link->server->addr;
src_port = link->server->port;
link->server = link->server->next;
} else {
src_addr = link->src_addr;
src_port = link->src_port;
}
link = ReLink(link,
src_addr, dst_addr, alias_addr,
src_port, dst_port, alias_port,
link_type);
}
return (link);
}
static struct alias_link *
FindLinkIn(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short dst_port,
u_short alias_port,
int link_type,
int replace_partial_links)
{
struct alias_link *link;
link = _FindLinkIn(dst_addr, alias_addr, dst_port, alias_port,
link_type, replace_partial_links);
if (link == NULL)
{
/* The following allows permanent links to be
specified as using the default aliasing address
(i.e. device interface address) without knowing
in advance what that address is. */
if (aliasAddress.s_addr != INADDR_ANY &&
alias_addr.s_addr == aliasAddress.s_addr)
{
link = _FindLinkIn(dst_addr, nullAddress, dst_port, alias_port,
link_type, replace_partial_links);
}
}
return(link);
}
/* External routines for finding/adding links
-- "external" means outside alias_db.c, but within alias*.c --
FindIcmpIn(), FindIcmpOut()
FindFragmentIn1(), FindFragmentIn2()
AddFragmentPtrLink(), FindFragmentPtr()
FindProtoIn(), FindProtoOut()
FindUdpTcpIn(), FindUdpTcpOut()
AddPptp(), FindPptpOutByCallId(), FindPptpInByCallId(),
FindPptpOutByPeerCallId(), FindPptpInByPeerCallId()
FindOriginalAddress(), FindAliasAddress()
(prototypes in alias_local.h)
*/
struct alias_link *
FindIcmpIn(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short id_alias,
int create)
{
struct alias_link *link;
link = FindLinkIn(dst_addr, alias_addr,
NO_DEST_PORT, id_alias,
LINK_ICMP, 0);
if (link == NULL && create && !(packetAliasMode & PKT_ALIAS_DENY_INCOMING))
{
struct in_addr target_addr;
target_addr = FindOriginalAddress(alias_addr);
link = AddLink(target_addr, dst_addr, alias_addr,
id_alias, NO_DEST_PORT, id_alias,
LINK_ICMP);
}
return (link);
}
struct alias_link *
FindIcmpOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_short id,
int create)
{
struct alias_link * link;
link = FindLinkOut(src_addr, dst_addr,
id, NO_DEST_PORT,
LINK_ICMP, 0);
if (link == NULL && create)
{
struct in_addr alias_addr;
alias_addr = FindAliasAddress(src_addr);
link = AddLink(src_addr, dst_addr, alias_addr,
id, NO_DEST_PORT, GET_ALIAS_ID,
LINK_ICMP);
}
return(link);
}
struct alias_link *
FindFragmentIn1(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short ip_id)
{
struct alias_link *link;
link = FindLinkIn(dst_addr, alias_addr,
NO_DEST_PORT, ip_id,
LINK_FRAGMENT_ID, 0);
if (link == NULL)
{
link = AddLink(nullAddress, dst_addr, alias_addr,
NO_SRC_PORT, NO_DEST_PORT, ip_id,
LINK_FRAGMENT_ID);
}
return(link);
}
struct alias_link *
FindFragmentIn2(struct in_addr dst_addr, /* Doesn't add a link if one */
struct in_addr alias_addr, /* is not found. */
u_short ip_id)
{
return FindLinkIn(dst_addr, alias_addr,
NO_DEST_PORT, ip_id,
LINK_FRAGMENT_ID, 0);
}
struct alias_link *
AddFragmentPtrLink(struct in_addr dst_addr,
u_short ip_id)
{
return AddLink(nullAddress, dst_addr, nullAddress,
NO_SRC_PORT, NO_DEST_PORT, ip_id,
LINK_FRAGMENT_PTR);
}
struct alias_link *
FindFragmentPtr(struct in_addr dst_addr,
u_short ip_id)
{
return FindLinkIn(dst_addr, nullAddress,
NO_DEST_PORT, ip_id,
LINK_FRAGMENT_PTR, 0);
}
struct alias_link *
FindProtoIn(struct in_addr dst_addr,
struct in_addr alias_addr,
u_char proto)
{
struct alias_link *link;
link = FindLinkIn(dst_addr, alias_addr,
NO_DEST_PORT, 0,
proto, 1);
if (link == NULL && !(packetAliasMode & PKT_ALIAS_DENY_INCOMING))
{
struct in_addr target_addr;
target_addr = FindOriginalAddress(alias_addr);
link = AddLink(target_addr, dst_addr, alias_addr,
NO_SRC_PORT, NO_DEST_PORT, 0,
proto);
}
return (link);
}
struct alias_link *
FindProtoOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_char proto)
{
struct alias_link *link;
link = FindLinkOut(src_addr, dst_addr,
NO_SRC_PORT, NO_DEST_PORT,
proto, 1);
if (link == NULL)
{
struct in_addr alias_addr;
alias_addr = FindAliasAddress(src_addr);
link = AddLink(src_addr, dst_addr, alias_addr,
NO_SRC_PORT, NO_DEST_PORT, 0,
proto);
}
return (link);
}
struct alias_link *
FindUdpTcpIn(struct in_addr dst_addr,
struct in_addr alias_addr,
u_short dst_port,
u_short alias_port,
u_char proto,
int create)
{
int link_type;
struct alias_link *link;
switch (proto)
{
case IPPROTO_UDP:
link_type = LINK_UDP;
break;
case IPPROTO_TCP:
link_type = LINK_TCP;
break;
default:
return NULL;
break;
}
link = FindLinkIn(dst_addr, alias_addr,
dst_port, alias_port,
link_type, create);
if (link == NULL && create && !(packetAliasMode & PKT_ALIAS_DENY_INCOMING))
{
struct in_addr target_addr;
target_addr = FindOriginalAddress(alias_addr);
link = AddLink(target_addr, dst_addr, alias_addr,
alias_port, dst_port, alias_port,
link_type);
}
return(link);
}
2002-07-01 11:19:40 +00:00
struct alias_link *
FindUdpTcpOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_short src_port,
u_short dst_port,
u_char proto,
int create)
{
int link_type;
struct alias_link *link;
switch (proto)
{
case IPPROTO_UDP:
link_type = LINK_UDP;
break;
case IPPROTO_TCP:
link_type = LINK_TCP;
break;
default:
return NULL;
break;
}
link = FindLinkOut(src_addr, dst_addr, src_port, dst_port, link_type, create);
if (link == NULL && create)
{
struct in_addr alias_addr;
alias_addr = FindAliasAddress(src_addr);
link = AddLink(src_addr, dst_addr, alias_addr,
src_port, dst_port, GET_ALIAS_PORT,
link_type);
}
return(link);
}
struct alias_link *
AddPptp(struct in_addr src_addr,
struct in_addr dst_addr,
struct in_addr alias_addr,
u_int16_t src_call_id)
{
struct alias_link *link;
link = AddLink(src_addr, dst_addr, alias_addr,
src_call_id, 0, GET_ALIAS_PORT,
LINK_PPTP);
return (link);
}
struct alias_link *
FindPptpOutByCallId(struct in_addr src_addr,
struct in_addr dst_addr,
u_int16_t src_call_id)
{
u_int i;
struct alias_link *link;
i = StartPointOut(src_addr, dst_addr, 0, 0, LINK_PPTP);
LIST_FOREACH(link, &linkTableOut[i], list_out)
if (link->link_type == LINK_PPTP &&
link->src_addr.s_addr == src_addr.s_addr &&
link->dst_addr.s_addr == dst_addr.s_addr &&
link->src_port == src_call_id)
break;
return (link);
}
struct alias_link *
FindPptpOutByPeerCallId(struct in_addr src_addr,
struct in_addr dst_addr,
u_int16_t dst_call_id)
{
u_int i;
struct alias_link *link;
i = StartPointOut(src_addr, dst_addr, 0, 0, LINK_PPTP);
LIST_FOREACH(link, &linkTableOut[i], list_out)
if (link->link_type == LINK_PPTP &&
link->src_addr.s_addr == src_addr.s_addr &&
link->dst_addr.s_addr == dst_addr.s_addr &&
link->dst_port == dst_call_id)
break;
return (link);
}
struct alias_link *
FindPptpInByCallId(struct in_addr dst_addr,
struct in_addr alias_addr,
u_int16_t dst_call_id)
{
u_int i;
struct alias_link *link;
i = StartPointIn(alias_addr, 0, LINK_PPTP);
LIST_FOREACH(link, &linkTableIn[i], list_in)
if (link->link_type == LINK_PPTP &&
link->dst_addr.s_addr == dst_addr.s_addr &&
link->alias_addr.s_addr == alias_addr.s_addr &&
link->dst_port == dst_call_id)
break;
return (link);
}
struct alias_link *
FindPptpInByPeerCallId(struct in_addr dst_addr,
struct in_addr alias_addr,
u_int16_t alias_call_id)
{
struct alias_link *link;
link = FindLinkIn(dst_addr, alias_addr,
0/* any */, alias_call_id,
LINK_PPTP, 0);
return (link);
}
2002-07-01 11:19:40 +00:00
struct alias_link *
FindRtspOut(struct in_addr src_addr,
struct in_addr dst_addr,
u_short src_port,
u_short alias_port,
u_char proto)
{
int link_type;
struct alias_link *link;
switch (proto)
{
case IPPROTO_UDP:
link_type = LINK_UDP;
break;
case IPPROTO_TCP:
link_type = LINK_TCP;
break;
default:
return NULL;
break;
}
link = FindLinkOut(src_addr, dst_addr, src_port, 0, link_type, 1);
if (link == NULL)
{
struct in_addr alias_addr;
alias_addr = FindAliasAddress(src_addr);
link = AddLink(src_addr, dst_addr, alias_addr,
src_port, 0, alias_port,
link_type);
}
return(link);
}
struct in_addr
FindOriginalAddress(struct in_addr alias_addr)
{
struct alias_link *link;
2002-07-01 11:19:40 +00:00
link = FindLinkIn(nullAddress, alias_addr,
0, 0, LINK_ADDR, 0);
if (link == NULL)
{
newDefaultLink = 1;
if (targetAddress.s_addr == INADDR_ANY)
return alias_addr;
else if (targetAddress.s_addr == INADDR_NONE)
return (aliasAddress.s_addr != INADDR_ANY) ?
aliasAddress : alias_addr;
else
return targetAddress;
}
else
{
if (link->server != NULL) { /* LSNAT link */
struct in_addr src_addr;
src_addr = link->server->addr;
link->server = link->server->next;
return (src_addr);
} else if (link->src_addr.s_addr == INADDR_ANY)
return (aliasAddress.s_addr != INADDR_ANY) ?
aliasAddress : alias_addr;
else
return link->src_addr;
}
}
struct in_addr
FindAliasAddress(struct in_addr original_addr)
{
struct alias_link *link;
2002-07-01 11:19:40 +00:00
link = FindLinkOut(original_addr, nullAddress,
0, 0, LINK_ADDR, 0);
if (link == NULL)
{
return (aliasAddress.s_addr != INADDR_ANY) ?
aliasAddress : original_addr;
}
else
{
if (link->alias_addr.s_addr == INADDR_ANY)
return (aliasAddress.s_addr != INADDR_ANY) ?
aliasAddress : original_addr;
else
return link->alias_addr;
}
}
/* External routines for getting or changing link data
(external to alias_db.c, but internal to alias*.c)
SetFragmentData(), GetFragmentData()
SetFragmentPtr(), GetFragmentPtr()
SetStateIn(), SetStateOut(), GetStateIn(), GetStateOut()
GetOriginalAddress(), GetDestAddress(), GetAliasAddress()
GetOriginalPort(), GetAliasPort()
SetAckModified(), GetAckModified()
GetDeltaAckIn(), GetDeltaSeqOut(), AddSeq()
SetProtocolFlags(), GetProtocolFlags()
SetDestCallId()
*/
void
SetFragmentAddr(struct alias_link *link, struct in_addr src_addr)
{
link->data.frag_addr = src_addr;
}
void
GetFragmentAddr(struct alias_link *link, struct in_addr *src_addr)
{
*src_addr = link->data.frag_addr;
}
void
SetFragmentPtr(struct alias_link *link, char *fptr)
{
link->data.frag_ptr = fptr;
}
void
GetFragmentPtr(struct alias_link *link, char **fptr)
{
*fptr = link->data.frag_ptr;
}
void
SetStateIn(struct alias_link *link, int state)
{
/* TCP input state */
switch (state) {
case ALIAS_TCP_STATE_DISCONNECTED:
if (link->data.tcp->state.out != ALIAS_TCP_STATE_CONNECTED)
link->expire_time = TCP_EXPIRE_DEAD;
else
link->expire_time = TCP_EXPIRE_SINGLEDEAD;
break;
case ALIAS_TCP_STATE_CONNECTED:
if (link->data.tcp->state.out == ALIAS_TCP_STATE_CONNECTED)
link->expire_time = TCP_EXPIRE_CONNECTED;
break;
default:
abort();
}
link->data.tcp->state.in = state;
}
void
SetStateOut(struct alias_link *link, int state)
{
/* TCP output state */
switch (state) {
case ALIAS_TCP_STATE_DISCONNECTED:
if (link->data.tcp->state.in != ALIAS_TCP_STATE_CONNECTED)
link->expire_time = TCP_EXPIRE_DEAD;
else
link->expire_time = TCP_EXPIRE_SINGLEDEAD;
break;
case ALIAS_TCP_STATE_CONNECTED:
if (link->data.tcp->state.in == ALIAS_TCP_STATE_CONNECTED)
link->expire_time = TCP_EXPIRE_CONNECTED;
break;
default:
abort();
}
link->data.tcp->state.out = state;
}
int
GetStateIn(struct alias_link *link)
{
/* TCP input state */
return link->data.tcp->state.in;
}
int
GetStateOut(struct alias_link *link)
{
/* TCP output state */
return link->data.tcp->state.out;
}
struct in_addr
GetOriginalAddress(struct alias_link *link)
{
if (link->src_addr.s_addr == INADDR_ANY)
return aliasAddress;
else
return(link->src_addr);
}
struct in_addr
GetDestAddress(struct alias_link *link)
{
return(link->dst_addr);
}
struct in_addr
GetAliasAddress(struct alias_link *link)
{
if (link->alias_addr.s_addr == INADDR_ANY)
return aliasAddress;
else
return link->alias_addr;
}
struct in_addr
GetDefaultAliasAddress()
{
return aliasAddress;
}
void
SetDefaultAliasAddress(struct in_addr alias_addr)
{
aliasAddress = alias_addr;
}
u_short
GetOriginalPort(struct alias_link *link)
{
return(link->src_port);
}
u_short
GetAliasPort(struct alias_link *link)
{
return(link->alias_port);
}
#ifndef NO_FW_PUNCH
static u_short
GetDestPort(struct alias_link *link)
{
return(link->dst_port);
}
#endif
void
SetAckModified(struct alias_link *link)
{
/* Indicate that ACK numbers have been modified in a TCP connection */
link->data.tcp->state.ack_modified = 1;
}
struct in_addr
GetProxyAddress(struct alias_link *link)
{
return link->proxy_addr;
}
void
SetProxyAddress(struct alias_link *link, struct in_addr addr)
{
link->proxy_addr = addr;
}
u_short
GetProxyPort(struct alias_link *link)
{
return link->proxy_port;
}
void
SetProxyPort(struct alias_link *link, u_short port)
{
link->proxy_port = port;
}
int
GetAckModified(struct alias_link *link)
{
/* See if ACK numbers have been modified */
return link->data.tcp->state.ack_modified;
}
int
GetDeltaAckIn(struct ip *pip, struct alias_link *link)
{
/*
Find out how much the ACK number has been altered for an incoming
TCP packet. To do this, a circular list of ACK numbers where the TCP
2002-07-01 11:19:40 +00:00
packet size was altered is searched.
*/
int i;
struct tcphdr *tc;
int delta, ack_diff_min;
u_long ack;
tc = (struct tcphdr *) ((char *) pip + (pip->ip_hl << 2));
ack = tc->th_ack;
delta = 0;
ack_diff_min = -1;
for (i=0; i<N_LINK_TCP_DATA; i++)
{
struct ack_data_record x;
x = link->data.tcp->ack[i];
if (x.active == 1)
{
int ack_diff;
ack_diff = SeqDiff(x.ack_new, ack);
if (ack_diff >= 0)
{
if (ack_diff_min >= 0)
{
if (ack_diff < ack_diff_min)
{
delta = x.delta;
ack_diff_min = ack_diff;
}
}
else
{
delta = x.delta;
ack_diff_min = ack_diff;
}
}
}
}
return (delta);
}
int
GetDeltaSeqOut(struct ip *pip, struct alias_link *link)
{
/*
Find out how much the sequence number has been altered for an outgoing
TCP packet. To do this, a circular list of ACK numbers where the TCP
2002-07-01 11:19:40 +00:00
packet size was altered is searched.
*/
int i;
struct tcphdr *tc;
int delta, seq_diff_min;
u_long seq;
tc = (struct tcphdr *) ((char *) pip + (pip->ip_hl << 2));
seq = tc->th_seq;
delta = 0;
seq_diff_min = -1;
for (i=0; i<N_LINK_TCP_DATA; i++)
{
struct ack_data_record x;
x = link->data.tcp->ack[i];
if (x.active == 1)
{
int seq_diff;
seq_diff = SeqDiff(x.ack_old, seq);
if (seq_diff >= 0)
{
if (seq_diff_min >= 0)
{
if (seq_diff < seq_diff_min)
{
delta = x.delta;
seq_diff_min = seq_diff;
}
}
else
{
delta = x.delta;
seq_diff_min = seq_diff;
}
}
}
}
return (delta);
}
void
AddSeq(struct ip *pip, struct alias_link *link, int delta)
{
/*
When a TCP packet has been altered in length, save this
information in a circular list. If enough packets have
been altered, then this list will begin to overwrite itself.
*/
struct tcphdr *tc;
struct ack_data_record x;
int hlen, tlen, dlen;
int i;
tc = (struct tcphdr *) ((char *) pip + (pip->ip_hl << 2));
hlen = (pip->ip_hl + tc->th_off) << 2;
tlen = ntohs(pip->ip_len);
dlen = tlen - hlen;
x.ack_old = htonl(ntohl(tc->th_seq) + dlen);
x.ack_new = htonl(ntohl(tc->th_seq) + dlen + delta);
x.delta = delta;
x.active = 1;
i = link->data.tcp->state.index;
link->data.tcp->ack[i] = x;
i++;
if (i == N_LINK_TCP_DATA)
link->data.tcp->state.index = 0;
else
link->data.tcp->state.index = i;
}
void
SetExpire(struct alias_link *link, int expire)
{
if (expire == 0)
{
link->flags &= ~LINK_PERMANENT;
DeleteLink(link);
}
else if (expire == -1)
{
link->flags |= LINK_PERMANENT;
}
else if (expire > 0)
{
link->expire_time = expire;
}
else
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/SetExpire(): ");
fprintf(stderr, "error in expire parameter\n");
#endif
}
}
void
ClearCheckNewLink(void)
{
newDefaultLink = 0;
}
void
SetProtocolFlags(struct alias_link *link, int pflags)
{
link->pflags = pflags;;
}
int
GetProtocolFlags(struct alias_link *link)
{
return (link->pflags);
}
void
SetDestCallId(struct alias_link *link, u_int16_t cid)
{
deleteAllLinks = 1;
link = ReLink(link, link->src_addr, link->dst_addr, link->alias_addr,
link->src_port, cid, link->alias_port, link->link_type);
deleteAllLinks = 0;
}
/* Miscellaneous Functions
HouseKeeping()
InitPacketAliasLog()
UninitPacketAliasLog()
*/
/*
Whenever an outgoing or incoming packet is handled, HouseKeeping()
is called to find and remove timed-out aliasing links. Logic exists
to sweep through the entire table and linked list structure
every 60 seconds.
(prototype in alias_local.h)
*/
void
HouseKeeping(void)
{
int i, n, n100;
struct timeval tv;
struct timezone tz;
/*
* Save system time (seconds) in global variable timeStamp for
* use by other functions. This is done so as not to unnecessarily
* waste timeline by making system calls.
*/
gettimeofday(&tv, &tz);
timeStamp = tv.tv_sec;
/* Compute number of spokes (output table link chains) to cover */
n100 = LINK_TABLE_OUT_SIZE * 100 + houseKeepingResidual;
n100 *= timeStamp - lastCleanupTime;
n100 /= ALIAS_CLEANUP_INTERVAL_SECS;
n = n100/100;
/* Handle different cases */
if (n > ALIAS_CLEANUP_MAX_SPOKES)
{
n = ALIAS_CLEANUP_MAX_SPOKES;
lastCleanupTime = timeStamp;
houseKeepingResidual = 0;
for (i=0; i<n; i++)
IncrementalCleanup();
}
else if (n > 0)
{
lastCleanupTime = timeStamp;
houseKeepingResidual = n100 - 100*n;
for (i=0; i<n; i++)
IncrementalCleanup();
}
else if (n < 0)
{
#ifdef DEBUG
fprintf(stderr, "PacketAlias/HouseKeeping(): ");
fprintf(stderr, "something unexpected in time values\n");
#endif
lastCleanupTime = timeStamp;
houseKeepingResidual = 0;
}
}
/* Init the log file and enable logging */
static void
InitPacketAliasLog(void)
{
if ((~packetAliasMode & PKT_ALIAS_LOG)
&& (monitorFile = fopen("/var/log/alias.log", "w")))
{
packetAliasMode |= PKT_ALIAS_LOG;
fprintf(monitorFile,
"PacketAlias/InitPacketAliasLog: Packet alias logging enabled.\n");
}
}
/* Close the log-file and disable logging. */
static void
UninitPacketAliasLog(void)
{
if (monitorFile) {
fclose(monitorFile);
monitorFile = NULL;
}
packetAliasMode &= ~PKT_ALIAS_LOG;
}
/* Outside world interfaces
-- "outside world" means other than alias*.c routines --
PacketAliasRedirectPort()
PacketAliasAddServer()
PacketAliasRedirectProto()
PacketAliasRedirectAddr()
PacketAliasRedirectDynamic()
PacketAliasRedirectDelete()
PacketAliasSetAddress()
PacketAliasInit()
PacketAliasUninit()
PacketAliasSetMode()
(prototypes in alias.h)
*/
/* Redirection from a specific public addr:port to a
private addr:port */
struct alias_link *
PacketAliasRedirectPort(struct in_addr src_addr, u_short src_port,
struct in_addr dst_addr, u_short dst_port,
struct in_addr alias_addr, u_short alias_port,
u_char proto)
{
int link_type;
struct alias_link *link;
switch(proto)
{
case IPPROTO_UDP:
link_type = LINK_UDP;
break;
case IPPROTO_TCP:
link_type = LINK_TCP;
break;
default:
#ifdef DEBUG
fprintf(stderr, "PacketAliasRedirectPort(): ");
fprintf(stderr, "only TCP and UDP protocols allowed\n");
#endif
return NULL;
}
link = AddLink(src_addr, dst_addr, alias_addr,
src_port, dst_port, alias_port,
link_type);
if (link != NULL)
{
link->flags |= LINK_PERMANENT;
}
#ifdef DEBUG
else
{
2002-07-01 11:19:40 +00:00
fprintf(stderr, "PacketAliasRedirectPort(): "
"call to AddLink() failed\n");
}
#endif
return link;
}
/* Add server to the pool of servers */
int
PacketAliasAddServer(struct alias_link *link, struct in_addr addr, u_short port)
{
struct server *server;
server = malloc(sizeof(struct server));
if (server != NULL) {
struct server *head;
server->addr = addr;
server->port = port;
head = link->server;
if (head == NULL)
server->next = server;
else {
struct server *s;
for (s = head; s->next != head; s = s->next);
s->next = server;
server->next = head;
}
link->server = server;
return (0);
} else
return (-1);
}
/* Redirect packets of a given IP protocol from a specific
public address to a private address */
struct alias_link *
PacketAliasRedirectProto(struct in_addr src_addr,
struct in_addr dst_addr,
struct in_addr alias_addr,
u_char proto)
{
struct alias_link *link;
link = AddLink(src_addr, dst_addr, alias_addr,
NO_SRC_PORT, NO_DEST_PORT, 0,
proto);
if (link != NULL)
{
link->flags |= LINK_PERMANENT;
}
#ifdef DEBUG
else
{
2002-07-01 11:19:40 +00:00
fprintf(stderr, "PacketAliasRedirectProto(): "
"call to AddLink() failed\n");
}
#endif
return link;
}
/* Static address translation */
struct alias_link *
PacketAliasRedirectAddr(struct in_addr src_addr,
struct in_addr alias_addr)
{
struct alias_link *link;
link = AddLink(src_addr, nullAddress, alias_addr,
0, 0, 0,
LINK_ADDR);
if (link != NULL)
{
link->flags |= LINK_PERMANENT;
}
#ifdef DEBUG
else
{
2002-07-01 11:19:40 +00:00
fprintf(stderr, "PacketAliasRedirectAddr(): "
"call to AddLink() failed\n");
}
#endif
return link;
}
/* Mark the aliasing link dynamic */
int
PacketAliasRedirectDynamic(struct alias_link *link)
{
if (link->flags & LINK_PARTIALLY_SPECIFIED)
return (-1);
else {
link->flags &= ~LINK_PERMANENT;
return (0);
}
}
void
PacketAliasRedirectDelete(struct alias_link *link)
{
/* This is a dangerous function to put in the API,
because an invalid pointer can crash the program. */
deleteAllLinks = 1;
DeleteLink(link);
deleteAllLinks = 0;
}
void
PacketAliasSetAddress(struct in_addr addr)
{
if (packetAliasMode & PKT_ALIAS_RESET_ON_ADDR_CHANGE
&& aliasAddress.s_addr != addr.s_addr)
CleanupAliasData();
aliasAddress = addr;
}
void
PacketAliasSetTarget(struct in_addr target_addr)
{
targetAddress = target_addr;
}
void
PacketAliasInit(void)
{
int i;
struct timeval tv;
struct timezone tz;
static int firstCall = 1;
if (firstCall == 1)
{
gettimeofday(&tv, &tz);
timeStamp = tv.tv_sec;
lastCleanupTime = tv.tv_sec;
houseKeepingResidual = 0;
for (i=0; i<LINK_TABLE_OUT_SIZE; i++)
LIST_INIT(&linkTableOut[i]);
for (i=0; i<LINK_TABLE_IN_SIZE; i++)
LIST_INIT(&linkTableIn[i]);
atexit(PacketAliasUninit);
firstCall = 0;
}
else
{
deleteAllLinks = 1;
CleanupAliasData();
deleteAllLinks = 0;
}
aliasAddress.s_addr = INADDR_ANY;
targetAddress.s_addr = INADDR_ANY;
icmpLinkCount = 0;
udpLinkCount = 0;
tcpLinkCount = 0;
pptpLinkCount = 0;
protoLinkCount = 0;
fragmentIdLinkCount = 0;
fragmentPtrLinkCount = 0;
sockCount = 0;
cleanupIndex =0;
packetAliasMode = PKT_ALIAS_SAME_PORTS
| PKT_ALIAS_USE_SOCKETS
| PKT_ALIAS_RESET_ON_ADDR_CHANGE;
}
void
PacketAliasUninit(void) {
deleteAllLinks = 1;
CleanupAliasData();
deleteAllLinks = 0;
UninitPacketAliasLog();
#ifndef NO_FW_PUNCH
UninitPunchFW();
#endif
}
/* Change mode for some operations */
unsigned int
PacketAliasSetMode(
unsigned int flags, /* Which state to bring flags to */
unsigned int mask /* Mask of which flags to affect (use 0 to do a
probe for flag values) */
)
{
/* Enable logging? */
if (flags & mask & PKT_ALIAS_LOG)
{
InitPacketAliasLog(); /* Do the enable */
} else
/* _Disable_ logging? */
if (~flags & mask & PKT_ALIAS_LOG) {
UninitPacketAliasLog();
}
#ifndef NO_FW_PUNCH
/* Start punching holes in the firewall? */
if (flags & mask & PKT_ALIAS_PUNCH_FW) {
InitPunchFW();
} else
/* Stop punching holes in the firewall? */
if (~flags & mask & PKT_ALIAS_PUNCH_FW) {
UninitPunchFW();
}
#endif
/* Other flags can be set/cleared without special action */
packetAliasMode = (flags & mask) | (packetAliasMode & ~mask);
return packetAliasMode;
}
int
PacketAliasCheckNewLink(void)
{
return newDefaultLink;
}
#ifndef NO_FW_PUNCH
/*****************
Code to support firewall punching. This shouldn't really be in this
file, but making variables global is evil too.
****************/
#ifndef IPFW2
#define IPFW2 1 /* use new ipfw code */
#endif
/* Firewall include files */
#include <net/if.h>
#include <netinet/ip_fw.h>
#include <string.h>
#include <err.h>
#if IPFW2 /* support for new firewall code */
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
/*
* helper function, updates the pointer to cmd with the length
* of the current command, and also cleans up the first word of
* the new command in case it has been clobbered before.
*/
static ipfw_insn *
next_cmd(ipfw_insn *cmd)
{
cmd += F_LEN(cmd);
bzero(cmd, sizeof(*cmd));
return cmd;
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
}
/*
* A function to fill simple commands of size 1.
* Existing flags are preserved.
*/
static ipfw_insn *
fill_cmd(ipfw_insn *cmd, enum ipfw_opcodes opcode, int size,
int flags, u_int16_t arg)
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
{
cmd->opcode = opcode;
cmd->len = ((cmd->len | flags) & (F_NOT | F_OR)) | (size & F_LEN_MASK);
cmd->arg1 = arg;
return next_cmd(cmd);
}
static ipfw_insn *
fill_ip(ipfw_insn *cmd1, enum ipfw_opcodes opcode, u_int32_t addr)
{
ipfw_insn_ip *cmd = (ipfw_insn_ip *)cmd1;
cmd->addr.s_addr = addr;
return fill_cmd(cmd1, opcode, F_INSN_SIZE(ipfw_insn_u32), 0, 0);
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
}
static ipfw_insn *
fill_one_port(ipfw_insn *cmd1, enum ipfw_opcodes opcode, u_int16_t port)
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
{
ipfw_insn_u16 *cmd = (ipfw_insn_u16 *)cmd1;
cmd->ports[0] = cmd->ports[1] = port;
return fill_cmd(cmd1, opcode, F_INSN_SIZE(ipfw_insn_u16), 0, 0);
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
}
static int
fill_rule(void *buf, int bufsize, int rulenum,
enum ipfw_opcodes action, int proto,
struct in_addr sa, u_int16_t sp, struct in_addr da, u_int16_t dp)
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
{
struct ip_fw *rule = (struct ip_fw *)buf;
ipfw_insn *cmd = (ipfw_insn *)rule->cmd;
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
bzero(buf, bufsize);
rule->rulenum = rulenum;
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
cmd = fill_cmd(cmd, O_PROTO, F_INSN_SIZE(ipfw_insn), 0, proto);
cmd = fill_ip(cmd, O_IP_SRC, sa.s_addr);
cmd = fill_one_port(cmd, O_IP_SRCPORT, sp);
cmd = fill_ip(cmd, O_IP_DST, da.s_addr);
cmd = fill_one_port(cmd, O_IP_DSTPORT, dp);
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
rule->act_ofs = (u_int32_t *)cmd - (u_int32_t *)rule->cmd;
cmd = fill_cmd(cmd, action, F_INSN_SIZE(ipfw_insn), 0, 0);
2002-07-01 11:19:40 +00:00
rule->cmd_len = (u_int32_t *)cmd - (u_int32_t *)rule->cmd;
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
return ((char *)cmd - (char *)buf);
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
}
#endif /* IPFW2 */
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
static void ClearAllFWHoles(void);
static int fireWallBaseNum; /* The first firewall entry free for our use */
static int fireWallNumNums; /* How many entries can we use? */
static int fireWallActiveNum; /* Which entry did we last use? */
static char *fireWallField; /* bool array for entries */
#define fw_setfield(field, num) \
do { \
(field)[(num) - fireWallBaseNum] = 1; \
} /*lint -save -e717 */ while(0) /*lint -restore */
#define fw_clrfield(field, num) \
do { \
(field)[(num) - fireWallBaseNum] = 0; \
} /*lint -save -e717 */ while(0) /*lint -restore */
#define fw_tstfield(field, num) ((field)[(num) - fireWallBaseNum])
static void
InitPunchFW(void) {
fireWallField = malloc(fireWallNumNums);
if (fireWallField) {
memset(fireWallField, 0, fireWallNumNums);
if (fireWallFD < 0) {
fireWallFD = socket(AF_INET, SOCK_RAW, IPPROTO_RAW);
}
ClearAllFWHoles();
fireWallActiveNum = fireWallBaseNum;
}
}
static void
UninitPunchFW(void) {
ClearAllFWHoles();
if (fireWallFD >= 0)
close(fireWallFD);
fireWallFD = -1;
if (fireWallField)
free(fireWallField);
fireWallField = NULL;
packetAliasMode &= ~PKT_ALIAS_PUNCH_FW;
}
/* Make a certain link go through the firewall */
void
PunchFWHole(struct alias_link *link) {
int r; /* Result code */
struct ip_fw rule; /* On-the-fly built rule */
int fwhole; /* Where to punch hole */
/* Don't do anything unless we are asked to */
if ( !(packetAliasMode & PKT_ALIAS_PUNCH_FW) ||
fireWallFD < 0 ||
link->link_type != LINK_TCP)
return;
memset(&rule, 0, sizeof rule);
/** Build rule **/
/* Find empty slot */
for (fwhole = fireWallActiveNum;
fwhole < fireWallBaseNum + fireWallNumNums &&
fw_tstfield(fireWallField, fwhole);
fwhole++)
;
if (fwhole == fireWallBaseNum + fireWallNumNums) {
for (fwhole = fireWallBaseNum;
fwhole < fireWallActiveNum &&
fw_tstfield(fireWallField, fwhole);
fwhole++)
;
if (fwhole == fireWallActiveNum) {
/* No rule point empty - we can't punch more holes. */
fireWallActiveNum = fireWallBaseNum;
#ifdef DEBUG
fprintf(stderr, "libalias: Unable to create firewall hole!\n");
#endif
return;
}
}
/* Start next search at next position */
fireWallActiveNum = fwhole+1;
/*
* generate two rules of the form
*
* add fwhole accept tcp from OAddr OPort to DAddr DPort
* add fwhole accept tcp from DAddr DPort to OAddr OPort
*/
#if IPFW2
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
if (GetOriginalPort(link) != 0 && GetDestPort(link) != 0) {
u_int32_t rulebuf[255];
int i;
i = fill_rule(rulebuf, sizeof(rulebuf), fwhole,
O_ACCEPT, IPPROTO_TCP,
GetOriginalAddress(link), ntohs(GetOriginalPort(link)),
GetDestAddress(link), ntohs(GetDestPort(link)) );
r = setsockopt(fireWallFD, IPPROTO_IP, IP_FW_ADD, rulebuf, i);
if (r)
err(1, "alias punch inbound(1) setsockopt(IP_FW_ADD)");
i = fill_rule(rulebuf, sizeof(rulebuf), fwhole,
O_ACCEPT, IPPROTO_TCP,
GetDestAddress(link), ntohs(GetDestPort(link)),
GetOriginalAddress(link), ntohs(GetOriginalPort(link)) );
r = setsockopt(fireWallFD, IPPROTO_IP, IP_FW_ADD, rulebuf, i);
if (r)
err(1, "alias punch inbound(2) setsockopt(IP_FW_ADD)");
}
#else /* !IPFW2, old code to generate ipfw rule */
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
/* Build generic part of the two rules */
rule.fw_number = fwhole;
IP_FW_SETNSRCP(&rule, 1); /* Number of source ports. */
IP_FW_SETNDSTP(&rule, 1); /* Number of destination ports. */
rule.fw_flg = IP_FW_F_ACCEPT | IP_FW_F_IN | IP_FW_F_OUT;
rule.fw_prot = IPPROTO_TCP;
rule.fw_smsk.s_addr = INADDR_BROADCAST;
rule.fw_dmsk.s_addr = INADDR_BROADCAST;
/* Build and apply specific part of the rules */
rule.fw_src = GetOriginalAddress(link);
rule.fw_dst = GetDestAddress(link);
rule.fw_uar.fw_pts[0] = ntohs(GetOriginalPort(link));
rule.fw_uar.fw_pts[1] = ntohs(GetDestPort(link));
/* Skip non-bound links - XXX should not be strictly necessary,
but seems to leave hole if not done. Leak of non-bound links?
(Code should be left even if the problem is fixed - it is a
clear optimization) */
if (rule.fw_uar.fw_pts[0] != 0 && rule.fw_uar.fw_pts[1] != 0) {
r = setsockopt(fireWallFD, IPPROTO_IP, IP_FW_ADD, &rule, sizeof rule);
#ifdef DEBUG
if (r)
err(1, "alias punch inbound(1) setsockopt(IP_FW_ADD)");
#endif
rule.fw_src = GetDestAddress(link);
rule.fw_dst = GetOriginalAddress(link);
rule.fw_uar.fw_pts[0] = ntohs(GetDestPort(link));
rule.fw_uar.fw_pts[1] = ntohs(GetOriginalPort(link));
r = setsockopt(fireWallFD, IPPROTO_IP, IP_FW_ADD, &rule, sizeof rule);
#ifdef DEBUG
if (r)
err(1, "alias punch inbound(2) setsockopt(IP_FW_ADD)");
#endif
}
#endif /* !IPFW2 */
/* Indicate hole applied */
link->data.tcp->fwhole = fwhole;
fw_setfield(fireWallField, fwhole);
}
/* Remove a hole in a firewall associated with a particular alias
link. Calling this too often is harmless. */
static void
ClearFWHole(struct alias_link *link) {
if (link->link_type == LINK_TCP) {
int fwhole = link->data.tcp->fwhole; /* Where is the firewall hole? */
struct ip_fw rule;
if (fwhole < 0)
return;
memset(&rule, 0, sizeof rule); /* useless for ipfw2 */
#if IPFW2
while (!setsockopt(fireWallFD, IPPROTO_IP, IP_FW_DEL,
&fwhole, sizeof fwhole))
;
#else /* !IPFW2 */
rule.fw_number = fwhole;
while (!setsockopt(fireWallFD, IPPROTO_IP, IP_FW_DEL,
&rule, sizeof rule))
;
#endif /* !IPFW2 */
fw_clrfield(fireWallField, fwhole);
link->data.tcp->fwhole = -1;
}
}
/* Clear out the entire range dedicated to firewall holes. */
static void
ClearAllFWHoles(void) {
struct ip_fw rule; /* On-the-fly built rule */
int i;
2002-07-01 11:19:40 +00:00
if (fireWallFD < 0)
return;
memset(&rule, 0, sizeof rule);
for (i = fireWallBaseNum; i < fireWallBaseNum + fireWallNumNums; i++) {
#if IPFW2
The new ipfw code. This code makes use of variable-size kernel representation of rules (exactly the same concept of BPF instructions, as used in the BSDI's firewall), which makes firewall operation a lot faster, and the code more readable and easier to extend and debug. The interface with the rest of the system is unchanged, as witnessed by this commit. The only extra kernel files that I am touching are if_fw.h and ip_dummynet.c, which is quite tied to ipfw. In userland I only had to touch those programs which manipulate the internal representation of firewall rules). The code is almost entirely new (and I believe I have written the vast majority of those sections which were taken from the former ip_fw.c), so rather than modifying the old ip_fw.c I decided to create a new file, sys/netinet/ip_fw2.c . Same for the user interface, which is in sbin/ipfw/ipfw2.c (it still compiles to /sbin/ipfw). The old files are still there, and will be removed in due time. I have not renamed the header file because it would have required touching a one-line change to a number of kernel files. In terms of user interface, the new "ipfw" is supposed to accepts the old syntax for ipfw rules (and produce the same output with "ipfw show". Only a couple of the old options (out of some 30 of them) has not been implemented, but they will be soon. On the other hand, the new code has some very powerful extensions. First, you can put "or" connectives between match fields (and soon also between options), and write things like ipfw add allow ip from { 1.2.3.4/27 or 5.6.7.8/30 } 10-23,25,1024-3000 to any This should make rulesets slightly more compact (and lines longer!), by condensing 2 or more of the old rules into single ones. Also, as an example of how easy the rules can be extended, I have implemented an 'address set' match pattern, where you can specify an IP address in a format like this: 10.20.30.0/26{18,44,33,22,9} which will match the set of hosts listed in braces belonging to the subnet 10.20.30.0/26 . The match is done using a bitmap, so it is essentially a constant time operation requiring a handful of CPU instructions (and a very small amount of memmory -- for a full /24 subnet, the instruction only consumes 40 bytes). Again, in this commit I have focused on functionality and tried to minimize changes to the other parts of the system. Some performance improvement can be achieved with minor changes to the interface of ip_fw_chk_t. This will be done later when this code is settled. The code is meant to compile unmodified on RELENG_4 (once the PACKET_TAG_* changes have been merged), for this reason you will see #ifdef __FreeBSD_version in a couple of places. This should minimize errors when (hopefully soon) it will be time to do the MFC.
2002-06-27 23:02:18 +00:00
int r = i;
while (!setsockopt(fireWallFD, IPPROTO_IP, IP_FW_DEL, &r, sizeof r))
;
#else /* !IPFW2 */
rule.fw_number = i;
while (!setsockopt(fireWallFD, IPPROTO_IP, IP_FW_DEL, &rule, sizeof rule))
;
#endif /* !IPFW2 */
}
memset(fireWallField, 0, fireWallNumNums);
}
#endif
void
PacketAliasSetFWBase(unsigned int base, unsigned int num) {
#ifndef NO_FW_PUNCH
fireWallBaseNum = base;
fireWallNumNums = num;
#endif
}