Writing a Basic Packet Capture Engine

Hi :-), this section consists of a discussion on how to write a simple packet capture engine.  The goal is to demonstrate methods of capturing and filtering multiple packets to aid in packet analysis.  All the juicy info on disecting IP packets and forging new ones are reserved for later sections.. Yes I can see your dissapointment, but you must admit that a program that captures a single packet at a time is pretty much useless.

We'll start by looking at:

  • int pcap_loop(pcap_t *p, int cnt, pcap_handler callback, u_char *user)

When pcap_loop(..) is called it will grab cnt packets (it will loop infinitely when cnt is -1) and pass them to the callback function which is of type pcap_handler. and what is pcap handler?? well lets go to the handy dandy header file..

typedef void (*pcap_handler)(u_char *arg, const struct pcap_pkthdr *, const u_char *);

We are interested in arguments 2 and 3, the pcap packet header and a const u_char consisting of the packet. The first argument (arg) is a pointer to data that you passed to pcap_loop initially as the last argument.  This is used to pass data to the packet processing routine without having to resort to globals.

As a primer, lets write a q&d program that will loop and getn packets, then exit.

Download testpcap2.c Here or just cut and paste from below

* file:   testpcap2.c
* date:   2001-Mar-14 12:14:19 AM 
* Description: Q&D proggy to demonstrate the use of pcap_loop

#include <pcap.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netinet/if_ether.h>

/* callback function that is passed to pcap_loop(..) and called each time 
 * a packet is recieved                                                    */
void my_callback(u_char *useless,const struct pcap_pkthdr* pkthdr,const u_char*
    static int count = 1;
    fprintf(stdout,"%d, ",count);
    if(count == 4)
        fprintf(stdout,"Come on baby sayyy you love me!!! ");
    if(count == 7)
        fprintf(stdout,"Tiiimmmeesss!! ");

int main(int argc,char **argv)
    int i;
    char *dev; 
    char errbuf[PCAP_ERRBUF_SIZE];
    pcap_t* descr;
    const u_char *packet;
    struct pcap_pkthdr hdr;     /* pcap.h */
    struct ether_header *eptr;  /* net/ethernet.h */

    if(argc != 2){ fprintf(stdout,"Usage: %s numpackets\n",argv[0]);return 0;}

    dev = pcap_lookupdev(errbuf);
    if(dev == NULL)
    { printf("%s\n",errbuf); exit(1); }
    descr = pcap_open_live(dev,BUFSIZ,0,-1,errbuf);
    if(descr == NULL)
    { printf("pcap_open_live(): %s\n",errbuf); exit(1); }

    /* allright here we call pcap_loop(..) and pass in our callback function */
    /* int pcap_loop(pcap_t *p, int cnt, pcap_handler callback, u_char *user)*/
    /* If you are wondering what the user argument is all about, so am I!!   */

    fprintf(stdout,"\nDone processing packets... wheew!\n");
    return 0;

Allright then, lets give her a whirl!

[root@pepe libpcap]# gcc testpcap2.c -lpcap
[root@pepe libpcap]# ./a.out 7 1, 2, 3, 4, Come on baby sayyy you love me!!! 5, 6, 7, Tiiimmmeesss!!  Done processing packets... wheew!
[root@pepe libpcap]#

So as you can see, my_callback(...) was actually called 7 times before exiting.  If you are testing your program by pinging an external machine the packets come slow enough to see them arrive in real time..  We could certainly put all of our packet analysis code in my_callback(..) and call it is done deal.  But as good little coders we certainly aren't satisfied with such an easy and straightforward solution!  The first problem is that pcap_loop(..) blocks indefinatly if no packet can be read.  While this may be the desired behaviour it would be nice to timeout on the reads.  Remember way back when we talked aboutpcap_open_live(..)?  One of the arguments you can specify is a timeout value in miliseconds.  pcap_loop actually ignores this argument, but pcap_dispatch(..) doesn't! So if we want our main looping mechanism to time-out replace pcap_loop() with pcap_dispatch(). Here is a description of pcap_dispatch(..)  shamelessly stripped from the man page

pcap_dispatch() is used to collect  and  process  packets.  cnt  specifies  the maximum  number of packets to process before returning. A cnt of -1 processes all  the  packets received  in  one buffer. A cnt of 0 processes all packets until an error occurs, EOF is reached, or the  read  times out  (when doing live reads and a non-zero read timeout is specified).  callback specifies a routine  to  be  called with  three arguments: a u_char pointer which is passed in from pcap_dispatch(), a pointer to the pcap_pkthdr  struct (which precede the actual network headers and data), and a u_char pointer to the packet data. The number  of  packets read is returned.  Zero is returned when EOF is reached in a ``savefile.'' A return of -1 indicates an error in which case pcap_perror() or pcap_geterr() may be used to display the error text.

In many applications using packet capture, you are not going to be interested in every packet on your network.  Take the following scenario.  Little Johny just bought the coolest new internet game to hit the markets.  Little Johny wants to be the first kid to hack up a bot for the game, but unlike all other little kiddies, Johny is going to write his own packet capture engine instead of using something canned.  Little Johnny uses tcpdump and notices that when the game starts up and he connects to the server.. it is connecting to on port 26112.  What should little Johnny do to only capture packets to or from port 26112?  Enter... pcap_compile(..) and pcap_setfilter(...) !!!

Note that we could read in all packets, sort through them one by one to pick out the subset we are interested in.  However, each callback into user space is (of course) a kernel crossing and could become quite expenise.  Luckily libpcap provides an interface where you can specify exactly which packets you are interested as bpf (berkeley packet filter) programs. In brief, to do this you need to pass a filter program as a string to pcap_compile() and then set it as a filter.... the problem is that the pcap man page doesn't provide any detail of what the filter program should look like (at least mine doesn't).  Is all lost!? No! because we have the handy dandy program tcpdump and its man page. You should have tcpdump already installed on your machine (which tcpdump) but if you don't I highly suggest you put it on.  Tcpdump is pretty much a wrapper of libpcap.  What is useful to us at the moment is that it accepts filter programs from the command line! Aha! a reference. The tcpdump man page explicitly describes the syntax and semantics of the filter language, which is (of course) pretty straight forward.  Here are the pertinent sections from my man pages..

              The program consists of one or more primitives.  Primitives usu
              ally  consist  of  an  id  (name or number) preceded by one or more
              qualifiers.  There are three different kinds of qualifier:

              type   qualifiers say what kind of thing  the  id  name  or  number
                     refers  to.   Possible  types are host, net and port.  E.g.,
                     `host foo', `net 128.3', `port 20'.  If  there  is  no  type
                     qualifier, host is assumed.

              dir    qualifiers specify a particular transfer direction to and/or
                     from id.  Possible directions are src, dst, src or  dst  and
                     src  and dst.  E.g., `src foo', `dst net 128.3', `src or dst
                     port ftp-data'.  If there is no dir qualifier, src or dst is
                     assumed.  For `null' link layers (i.e. point to point proto
                     cols such as slip) the inbound and outbound  qualifiers  can
                     be used to specify a desired direction.

              proto  qualifiers  restrict  the  match  to  a particular protocol.
                     Possible protos are: ether, fddi,  ip,  arp,  rarp,  decnet,
                     lat, sca, moprc, mopdl, tcp and udp.  E.g., `ether src foo',
                     `arp net 128.3', `tcp port 21'.  If there is no proto quali
                     fier,  all  protocols  consistent with the type are assumed.
                     E.g., `src foo' means `(ip or arp or rarp) src foo'  (except
                     the latter is not legal syntax), `net bar' means `(ip or arp
                     or rarp) net bar' and `port 53' means  `(tcp  or  udp)  port

              In  addition  to the above, there are some special `primitive' key
              words that don't follow  the  pattern:  gateway,  broadcast,  less,
              greater  and  arithmetic  expressions.   All of these are described

              More complex filter expressions are built up  by  using  the  words
              and,  or  and  not  to combine primitives.  E.g., `host foo and not
              port ftp and not port ftp-data'.  To save typing, identical  quali
              fier  lists can be omitted.  E.g., `tcp dst port ftp or ftp-data or
              domain' is exactly the same as `tcp dst port ftp or  tcp  dst  port
              ftp-data or tcp dst port domain'.

              Allowable primitives are:

              dst host host
                     True  if  the  IP  destination  field of the packet is host,
                     which may be either an address or a name.

              src host host
                     True if the IP source field of the packet is host.

              host host
                     True if either the IP source or destination of the packet is
                     host.   Any  of  the above host expressions can be prepended
                     with the keywords, ip, arp, or rarp as in:
                          ip host host
                     which is equivalent to:
                          ether proto \ip and host host
                     If host is a name with multiple IP addresses,  each  address
                     will be checked for a match.

              ether dst ehost
                     True  if  the  ethernet destination address is ehost.  Ehost
                     may be either a name  from  /etc/ethers  or  a  number  (see
                     ethers(3N) for numeric format).

              ether src ehost
                     True if the ethernet source address is ehost.

              ether host ehost
                     True if either the ethernet source or destination address is

              gateway host
                     True if the packet used host as a gateway.  I.e., the ether
                     net  source  or destination address was host but neither the
                     IP source nor the IP destination was host.  Host must  be  a
                     name  and  must be found in both /etc/hosts and /etc/ethers.
                     (An equivalent expression is
                          ether host ehost and not host host
                     which can be used with either names or numbers  for  host  /

              dst net net
                     True  if the IP destination address of the packet has a net
                     work number of net. Net may be either a name from  /etc/net
                     works or a network number (see networks(4) for details).

              src net net
                     True  if  the  IP source address of the packet has a network
                     number of net.

              net net
                     True if either the IP source or destination address  of  the
                     packet has a network number of net.

              net net mask mask
                     True  if  the  IP address matches net with the specific net
                     mask.  May be qualified with src or dst.

              net net/len
                     True if the IP address matches net a netmask len bits  wide.
                     May be qualified with src or dst.

              dst port port
                     True if the packet is ip/tcp or ip/udp and has a destination
                     port value of port.  The port can be a number or a name used
                     in  /etc/services  (see  tcp(4P) and udp(4P)).  If a name is
                     used, both the port number and protocol are checked.   If  a
                     number  or  ambiguous  name is used, only the port number is
                     checked (e.g., dst port 513 will print both tcp/login  traf
                     fic  and  udp/who  traffic,  and port domain will print both
                     tcp/domain and udp/domain traffic).

              src port port
                     True if the packet has a source port value of port.

              port port
                     True if either the source or destination port of the  packet
                     is port.  Any of the above port expressions can be prepended
                     with the keywords, tcp or udp, as in:
                          tcp src port port
                     which matches only tcp packets whose source port is port.

              less length
                     True if the packet has  a  length  less  than  or  equal  to
                     length.  This is equivalent to:
                          len <= length.

              greater length
                     True  if  the  packet  has a length greater than or equal to
                     length.  This is equivalent to:
                          len >= length.

              ip proto protocol
                     True if the packet is an ip packet (see ip(4P)) of  protocol
                     type protocol.  Protocol can be a number or one of the names
                     icmp, igrp, udp, nd, or tcp.  Note that the identifiers tcp,
                     udp,  and  icmp  are  also  keywords and must be escaped via
                     backslash (\), which is \\ in the C-shell.

              ether broadcast
                     True if the packet is an  ethernet  broadcast  packet.   The
                     ether keyword is optional.

              ip broadcast
                     True if the packet is an IP broadcast packet.  It checks for
                     both the all-zeroes and all-ones broadcast conventions,  and
                     looks up the local subnet mask.

              ether multicast
                     True  if  the  packet  is an ethernet multicast packet.  The
                     ether keyword is optional.  This is shorthand for  `ether[0]
                     & 1 != 0'.

              ip multicast
                     True if the packet is an IP multicast packet.

              ether proto protocol
                     True  if the packet is of ether type protocol.  Protocol can
                     be a number or a name like ip, arp,  or  rarp.   Note  these
                     identifiers  are also keywords and must be escaped via back
                     slash (\).  [In the  case  of  FDDI  (e.g.,  `fddi  protocol
                     arp'), the protocol identification comes from the 802.2 Log
                     ical Link Control (LLC) header, which is usually layered  on
                     top  of the FDDI header.  Tcpdump assumes, when filtering on
                     the protocol identifier, that all FDDI  packets  include  an
                     LLC  header,  and  that  the LLC header is in so-called SNAP

              ip, arp, rarp, decnet
                     Abbreviations for:
                          ether proto p where p is one of the above protocols.

              tcp, udp, icmp
                     Abbreviations for:
                          ip proto p
                     where p is one of the above protocols.

              expr relop expr
                     True  if the relation holds, where relop is one of >,
                     <, >=, <=, =, !=, and expr is an arithmetic
                     expression composed  of integer constants (expressed in
                     standard C syntax), the nor mal binary operators [+, -,
                     *, /, &, |], a length  operator, and  special  packet
                     data accessors.  To access data inside the packet, use the
                     following syntax: proto [ expr : size ] Proto is one of
                     ether, fddi, ip, arp,  rarp,  tcp,  udp,  or icmp,  and
                     indicates the protocol layer for the index opera tion.
                     The byte offset, relative to the  indicated  protocol
                     layer, is given by expr.  Size is optional and indicates
                     the number of bytes in the field of interest; it can  be
                     either one,  two,  or four, and defaults to one.  The
                     length opera tor, indicated by the keyword len, gives the
                     length  of  the packet.

                     For example, `ether[0] & 1 != 0' catches all multicast traf
                     fic.  The expression `ip[0] & 0xf != 5' catches all IP pack
                     ets  with  options.  The  expression  `ip[6:2] & 0x1fff = 0'
                     catches only unfragmented datagrams and frag zero  of  frag
                     mented  datagrams.   This check is implicitly applied to the
                     tcp and udp index operations.  For instance,  tcp[0]  always
                     means  the first byte of the TCP header, and never means the
                     first byte of an intervening fragment.

   Primitives may be combined using:

                     A parenthesized group of primitives and operators (parenthe
                     ses are special to the Shell and must be escaped).

                     Negation (`!' or `not').

                     Concatenation (`&&' or `and').

                     Alternation (`||' or `or').

              Negation  has  highest  precedence.   Alternation and concatenation
              have equal precedence and  associate  left  to  right.   Note  that
              explicit  and  tokens, not juxtaposition, are now required for con

              If an identifier is given without a keyword, the most  recent  key
              word is assumed.  For example,
                   not host vs and ace
              is short for
                   not host vs and host ace
              which should not be confused with
                   not ( host vs or ace )

              Expression  arguments  can  be passed to tcpdump as either a single
              argument or as multiple arguments, whichever  is  more  convenient.
              Generally,  if  the expression contains Shell metacharacters, it is
              easier to pass it as a single, quoted argument.  Multiple arguments
              are concatenated with spaces before being parsed.

       To print all packets arriving at or departing from sundown:
              tcpdump host sundown

       To print traffic between helios and either hot or ace:
              tcpdump host helios and \( hot or ace \)

       To print all IP packets between ace and any host except helios:
              tcpdump ip host ace and not helios

       To print all traffic between local hosts and hosts at Berkeley:
              tcpdump net ucb-ether

       To  print  all  ftp  traffic through internet gateway snup: (note that the
       expression is quoted to prevent  the  shell  from  (mis-)interpreting  the
              tcpdump 'gateway snup and (port ftp or ftp-data)'

       To print traffic neither sourced from nor destined for local hosts (if you
       gateway to one other net, this stuff should never make it onto your  local
              tcpdump ip and not net localnet

       To  print  the start and end packets (the SYN and FIN packets) of each TCP
       conversation that involves a non-local host.
              tcpdump 'tcp[13] & 3 != 0 and not src and dst net localnet'

       To print IP packets longer than 576 bytes sent through gateway snup:
              tcpdump 'gateway snup and ip[2:2] > 576'

       To print IP broadcast or multicast packets that were not sent via ethernet
       broadcast or multicast:
              tcpdump 'ether[0] & 1 = 0 and ip[16] >= 224'

       To  print  all  ICMP packets that are not echo requests/replies (i.e., not
       ping packets):
              tcpdump 'icmp[0] != 8 and icmp[0] != 0"

Ok, so that is a lot of info (and probably more than we need) but it gives us a starting point.  Lets give it a shot... on my network I have a linux box and a windoze machine connected to a non switched hub.  Therefore, if I place my ethernet card in promiscuous mode on my linux machine I should be able to see all traffic going to (and coming from) my windows machine.  Lets see if the examples from the man page will work if directly fed to pcap_compile..

Consider the following program... (download here)

* file:   testpcap3.c
* date:   Sat Apr 07 23:23:02 PDT 2001  
* Author: Martin Casado
* Last Modified:2001-Apr-07 11:23:05 PM
* Investigate using filter programs with pcap_compile() and
* pcap_setfilter()

#include <pcap.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <arpa/inet.h>
#include <netinet/if_ether.h> 

void my_callback(u_char *useless,const struct pcap_pkthdr* pkthdr,const u_char*
    static int count = 1;
    fprintf(stdout,"%d, ",count);

int main(int argc,char **argv)
    int i;
    char *dev; 
    char errbuf[PCAP_ERRBUF_SIZE];
    pcap_t* descr;
    const u_char *packet;
    struct pcap_pkthdr hdr;     /* pcap.h                    */
    struct ether_header *eptr;  /* net/ethernet.h            */
    struct bpf_program fp;      /* hold compiled program     */
    bpf_u_int32 maskp;          /* subnet mask               */
    bpf_u_int32 netp;           /* ip                        */

    if(argc != 2){ fprintf(stdout,"Usage: %s \"filter program\"\n"
            ,argv[0]);return 0;}

    dev = pcap_lookupdev(errbuf);
    if(dev == NULL)
    { fprintf(stderr,"%s\n",errbuf); exit(1); }


    descr = pcap_open_live(dev,BUFSIZ,1,-1,errbuf);
    if(descr == NULL)
    { printf("pcap_open_live(): %s\n",errbuf); exit(1); }

    if(pcap_compile(descr,&fp,argv[1],0,netp) == -1)
    { fprintf(stderr,"Error calling pcap_compile\n"); exit(1); }

    if(pcap_setfilter(descr,&fp) == -1)
    { fprintf(stderr,"Error setting filter\n"); exit(1); }


    return 0;

This program accepts a string from the user, (similar to tcpdump) compiles it and sets it as a filter.  Lets go ahead and try it with an example similar to the one in the tcpdump examples..

[root@localhost libpcap]# gcc testpcap3.c -lpcap
[root@localhost libpcap]# ./a.out "host www.google.com"
(** try and ping www.slashdot.org ... nothing **)
(** try and ping www.google.com **)
1, 2, 3, 4, 5, 6, 
(** hurray! **)

It looks like our filter program worked. Lets try to see if we can capture packets from a different machine on the same network.... how about my windows machine when it connects to battle.net..

[root@localhost libpcap]# ./a.out "src"
1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40,

Yes yes!!! we are getting very close to actually having some power, but first we need to find out how to look inside the packets and pull out information. At last! the next section will delve into disecting packets so we can really analyze what is going on in our networks!!