Basic Packet-Sniffer Construction
from the Ground Up
Part 1
by
Chad Renfro
raw_sock@hotmail.com
Packet sniffers are applications used by network administrators to monitor and
validate network traffic. Sniffers are programs used to read packets that travel across
the network at various levels of the OSI layer. And like most security tools sniffers too
can be used for both good and destructive purposes. On the light-side of network
administration sniffers help quickly track down problems such as bottlenecks and
misplaced filters. However on the dark-side sniffers can be used to reap tremendous
amounts of havoc by gathering legitimate user names and passwords so that other
machines can be quickly compromised. Hopefully this paper will be used to help
administrators gain control of their networks by being able to analyze network traffic
not only by using preconstructed sniffers but by being able to create their own. This
paper will look at the packet sniffer from the bottem up, looking in depth at the sniffer
core and then gradualy adding functionality to the application. The example included
here will help illustrate some rather cumbersome issues when dealing with network
programing. In no way will this single paper teach a person to write a complete sniffing
application like tcpdump or sniffit. It will however teach some very fundamental issues
that are inherent to all packet sniffers. Like how the packets are accessed on the network
and how to work with the packets at different layers.
The most basic sniffer...
Sniffer #1.
This sniffer will illustrate the use of the SOCK_RAW device and show how to gather
packets from the network and print out some simple header information to std_out.
Although the basic premise is that packet sniffers operate in a promiscuous mode which
listens to all packets weather or not the packet is destined for the machines mac address,
this example will collect packets in a non-promiscuous mode . This will let usconcentrate
on the SOCK_RAW device for the first example. To operate this same code in a
promiscous mode the network card may be put in a promiscous mode manually. To do
this type this in after the log in :
> su -
Password : ********
# ifconfig eth0 promisc
This will now set the network interface eth0 in promiscous mode.
/************************simple_Tcp_sniff.c********************/
1. #include <stdio.h>
2. #include <sys/socket.h>
3. #include <netinet/in.h>
4. #include <arpa/inet.h>
5. #include "headers.h"
6. int main()
7. {
8. int sock, bytes_recieved, fromlen;
9. char buffer[65535];
10. struct sockaddr_in from;
11. struct ip *ip;
12. struct tcp *tcp;
13.
14. sock = socket(AF_INET, SOCK_RAW, IPPROTO_TCP);
15. while(1)
16. {
17. fromlen = sizeof from;
18. bytes_recieved = recvfrom(sock, buffer, sizeof buffer, 0,
(struct sockaddr *)&from, &fromlen);
19. printf("\nBytes received ::: %5d\n",bytes_recieved);
20. printf("Source address ::: %s\n",inet_ntoa(from.sin_addr));
21. ip = (struct ip *)buffer;
22. printf("IP header length ::: %d\n",ip->ip_length);
23. printf("Protocol ::: %d\n",ip->ip_protocol);
24. tcp = (struct tcp *)(buffer + (4*ip->ip_length));
25. printf("Source port ::: %d\n",ntohs(tcp->tcp_source_port);
26. printf("Dest port ::: %d\n",ntohs(tcp->tcp_dest_port));
27. }
28. }
/***********************EOF**********************************/
What this means :
Line 1-4 :
These are the header files required to use some needed c functions we will use later
<stdio.h> = functions like printf and std_out
<sys/socket.h> = this will give access to the SOCK_RAW and the
IPPROTO_TCP defines
<netinet/in.h> = structs like the sockaddr_in
<arpa/inet.h> = lets us use the functions to do network to host byte
order conversions
line 5 :
This is the header file headers.h that is also included with this program to give standard
structures to access the ip and tcp fields. The structures identify each field in the ip and
tcp header for instance :
struct ip {
unsigned int ip_length:4; /* length of ip-header in 32-bit
words*/
unsigned int ip_version:4; /* set to "4", for Ipv4 */
unsigned char ip_tos; /* type of service*/
unsigned short ip_total_length; /* Total length of ip datagram in
bytes */
unsigned short ip_id; /*identification field*/
unsigned short ip_flags;
unsigned char ip_ttl; /*time-to-live, sets upper limit
for max number of routers to
go through before the packet is
discarded*/
unsigned char ip_protocol; /*identifies the correct transport
protocol */
unsigned short ip_cksum; /*calculated for the ip header ONLY*/
unsigned int ip_source; /*source ip */
unsigned int ip_dest; /*dest ip*/
};
struct tcp {
unsigned short tcp_source_port; /*tcp source port*/
unsigned short tcp_dest_port; /*tcp dest port*/
unsigned int tcp_seqno; /*tcp sequence number,
identifies the byte in the
stream of data*/
unsigned int tcp_ackno; /*contains the next seq num that
the sender expects to recieve*/
unsigned int tcp_res1:4, /*little-endian*/
tcp_hlen:4, /*length of tcp header in 32-bit
words*/
tcp_fin:1, /*Finish flag "fin"*/
tcp_syn:1, /*Synchronize sequence
numbers to start a connection
tcp_rst:1, /*Reset flag */
tcp_psh:1, /*Push, sends data to the
application*/
tcp_ack:1, /*acknowledge*/
tcp_urg:1, /*urgent pointer*/
tcp_res2:2;
unsigned short tcp_winsize; /*maxinum number of bytes able
to recieve*/
unsigned short tcp_cksum; /*checksum to cover the tcp
header and data portion of the
packet*/
unsigned short tcp_urgent; /*vaild only if the urgent flag is
set, used to transmit
emergency data */
};
line 8-13 :
This is the variable declaration section
integers :
sock = socket file descriptor
bytes_recieved = bytes read from the open socket "sock"
fromlen = the size of the from structure char :
buffer = where the ip packet that is read off the
wire will be held buffer will hold a datagram
of 65535 bytes which is the maximum length
of an ip datagram.
Struct sockaddr_in :
struct sockaddr_in {
short int sin_family; /* Address family */
unsigned short int sin_port; /* Port number */
struct in_addr sin_addr; /* Internet address */
unsigned char sin_zero[8]; /* Same size as struct sockaddr */
};
Before we go any further two topics should be covered,byte-ordering and sockaddr
structures. Byte-ordering,is the way that the operating system stores bytes in memory.
There are two ways that this is done first with the low-order byte at the starting address
this is known as "little-endian" or host-byte order. Next bytes can be stored with the
high order byte at the starting address, this is called "big-endian" or network byte order.
The Internet protocol uses >>>>>> network byte order.
This is important because if you are working on an intel based linux box you will be
programming on a little-endian machine and to send data via ip you must convert the
bytes to network-byte order. For examle lets say we are going to store a 2-byte number
in memory say the value is (in hex) 0x0203
First this is how the value is stored on a big-endian machine:
___________
| 02 | 03 |
|_____|_____|
address: 0 1
And here is the same value on a little-endian machine:
___________
|03 | 02 |
|_____|_____|
address: 1 0
The same value is being represented in both examples it is just how we order the bytes
that changes.
The next topic that you must understand is the sockaddr vs. the sockaddr_in structures.
The struct sockaddr is used to hold information about the socket such as the family type
and other address information it looks like :
struct sockaddr {
unsigned short sa_family; /*address family*/
char sa_data[14]; /*address data*/
};
The first element in the structure "sa_family" will be used to reference what the family
type is for the socket, in our sniffer it will be AF_INET. Next the "sa_data" element
holds the destination port and address for the socket. To make it easier to deal with the
sockaddr struct the use of the sockaddr_in structure is commonly used. Sockaddr_in
makes it easier to reference all of the elements that are contained by sockaddr.
Sockaddr_in looks like:
struct sockaddr_in {
short int sin_family; /* Address family */
unsigned short int sin_port; /* Port number */
struct in_addr sin_addr; /* Internet address */
unsigned char sin_zero[8]; /* Same size as struct sockaddr */
};
We will use this struct and declare a variable "from" which will give us the information
on the packet that we will collect from the raw socket. For instance the var
"from.sin_addr" will give access to the packets source address (in
network byte order). The thing to mention here is that all items in the sockaddr_in
structure must be in network-byte order. When we receive the data in the sockaddr_in
struct we must then convert it back to Host-byte order. To do this we can use some
predefined functions to convert back and forth between host and network byteorder.
Here are the functions we will use:
ntohs : this function converts network byte order to host byte order
for a 16-bit short
ntohl : same as above but for a 32-bit long
inet_ntoa : this function converts a 32-bit network binary value to a
dotted decimal ip address
inet_aton : converts a character string address to the 32-bit network
binary value
inet_addr : takes a char string dotted decimal addr and returns a 32-bit
network binary value
To further illustrate ,say I want to know the port number that this packet originated from:
int packet_port; packet_port =ntohs(from.sin_port);
^^^^^
If I want the source IP address of the packet we will use a special function to get it to the
123.123.123.123 format:
char *ip_addr; ip_addr =inet_ntoa(from.sin_addr)
^^^^^^^^^
line 11-12:
struct ip *ip :
struct tcp *tcp :
This is a structure that we defined in our header file "headers.h". This structure is
declared so that we can access individual fields of the ip/tcp header. The structure is like
a transparent slide with predefined fields drawn on it. When a packet is taken off
the wire it is a stream of bits, to make sense of it the "transparency" (or cast) is laid on
top of or over the bits so the individual fields can be referenced.
Line 14 :
sock = socket(AF_INET, SOCK_RAW, IPPROTO_TCP);
This is the most important line in the entire program. Socket() takes three arguments in
this form:
sockfd = socket(int family, int type, int protocol);
The first argument is the family. This could be either AF_UNIX which is used so a process
can communicate with another process on the same host or AF_INET which is used for
internet communication between remote hosts. In this case it will be AF_INET . Next
is the type, the type is usually between 1 of 4 choices (there are others that we will not
discuss here) the main four are :
1. SOCK_DRAM : used for udp datagrams
2. SOCK_STREAM : used for tcp packets
3. SOCK_RAW : used to bypass the transport layer
and directly access the IP layer
4. SOCK_PACKET : this is linux specific, it is similuar to
SOCK_RAW except it accesses the DATA LINK Layer
For our needs we will use the SOCK_RAW type. You must have root acces to open a
raw socket. The last parameter is the protocol,the protocol value specifies what type of
traffic the socket should receive , for normal sockets this value is usally set to "0"
because the socket can figure out if for instance the "type" of SOCK_DGRAM is
specified then the protocol should be UDP.In our case we just want to look at tcp
traffic so we will specify IPPROTO_TCP.
line 15 :
while (1)
The while (1) puts the program into an infinite loop this is necessary so that after the
first packet is processed we will loop around and grab the next.
Line 18:
bytes_recieved = recvfrom(sock, buffer, sizeof buffer, 0, (struct sockaddr *)&from, &fromlen);
Now here is where we are actually reading data from the open socket "sock".The from
struct is also filled in but notice that we are casting "from" from a "sockaddr_in" struct
to a "sockaddr" struct. We do this because the recvfrom() requires a sockaddr type but
to access the separate fields we will continue to use the sockaddr_in structure. The
length of the "from" struct must also be present and passed by address. The recvfrom()
call will return the number of bytes on success and a -1 on error and fill the global var
errno.
This is what we call "blocking-I/O" the recvfrom() will wait here forever until a
datagram on the open socket is ready to be processed. This is opposed to
Non-blocking I/O which is like running a process in the background and move on to
other tasks.
Line 20:
printf("Source address ::: %s\n",inet_ntoa(from.sin_addr));
This printf uses the special function inet_ntoa() to take the value of "from.sin_addr"
which is stored in Network-byte order and outputs a value in a readable ip form such
as 192.168.1.XXX.
Line 21:
ip = (struct ip *)buffer;
This is where we will overlay a predefined structure that will help us to individually
identify the fields in the packet that we pick up from the open socket.
Line 22:
printf("IP header length ::: %d\n",ip->ip_length);
The thing to notice on this line is the "ip->ip_length" this will access a pointer in
memory to the ip header length the important thing to remember is that the length
will be represented in 4-byte words this will be more important later when trying to
access items past the ip header such as the tcp header or the data portion of the packet.
Line 23:
printf("Protocol ::: %d\n",ip->ip_protocol);
This gives access to the type of protocol such as 6 for tcp or 17 for udp.
Line 24:
tcp = (struct tcp *)(buffer + (4*ip->ip_length));
Remember earlier it was mentioned that the ip header length is stored in 4 byte words,
this is where that bit of information becomes important. Here we are trying to get access
to the tcp header fields, to do this we must overlay a structure that has the fields
predefined just as we did with ip. There is one key difference here the ip header fields
were easy to access due to the fact that the beginning of the buffer was also the beginning
of the ip header as so :
|----------------- buffer ----------------|
_________________________________________
| ip header | |
|____________________|____________________|
^
*ip
^
*buffer
So to get access to the ip header we just set a pointer casted as an ip structure to the
beginning of the buffer like "ip = (struct ip *)buffer;". To get access to the tcp header
is a little more difficult due to the fact that we must set a pointer and cast it as a tcp
structure at the beginning of the tcp header which follows the ip header in the buffer
as so :
|----------------- buffer ---------------|
________________________________________
| ip header | tcp header | |
|___________|____________|_______________|
^
*tcp
This is why we use 4*ip->ip_length to find the start of the tcp header.
Line 25-26:
printf("Source port ::: %d\n",ntohs(tcp->tcp_source_port);
printf("Dest port ::: %d\n",ntohs(tcp->tcp_dest_port));
We can now access the source and dest ports which are located in the tcp header via
the structure as defined above.
This will conclude our first very simple tcp sniffer. This was a very basic application
that should help define how to access packets passing on the network and how to use
sockets to access the packets. Hopefully this will be the first of many papers to come,
which each proceeding paper we will add a new or more complex feature to the sniffer. I
should also mention that there a number of great resources on the net that should aid you
in further research in this area :
1. Beej's Guide to Network Programming
This is an awesome paper that really helps
clear up any misconceptions about network programming.
[http://www.ecst.csuchico.edu/~beej/guide/net]
2. TCP/IP Illustrated Vol 1,2,3
W.Richard Stevens
To use the above program, cut out the above code and strip off all
of the line numbers. Save the edited file as sniff.c. Next cut
out the header file headers.h (below) and save it to a file headers.h
in the same directory. Now just compile: gcc -o sniff sniff.c
You should now have the executable "sniff", to run it type
#./sniff
/*************************headers.h**************************/
/*structure of an ip header */
struct ip {
unsigned int ip_length:4; /*little-endian*/
unsigned int ip_version:4;
unsigned char ip_tos;
unsigned short ip_total_length;
unsigned short ip_id;
unsigned short ip_flags;
unsigned char ip_ttl;
unsigned char ip_protocol;
unsigned short ip_cksum;
unsigned int ip_source;
unsigned int ip_dest;
};
/* Structure of a TCP header */
struct tcp {
unsigned short tcp_source_port;
unsigned short tcp_dest_port;
unsigned int tcp_seqno;
unsigned int tcp_ackno;
unsigned int tcp_res1:4, /*little-endian*/
tcp_hlen:4,
tcp_fin:1,
tcp_syn:1,
tcp_rst:1,
tcp_psh:1,
tcp_ack:1,
tcp_urg:1,
tcp_res2:2;
unsigned short tcp_winsize;
unsigned short tcp_cksum;
unsigned short tcp_urgent;
};
/*********************EOF***********************************/