Fixes from Phil Wood:

[libpcap] / pcap-linux.c

/*
 *  pcap-linux.c: Packet capture interface to the Linux kernel
 *
 *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
 *                     Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
 *  
 *  License: BSD
 *  
 *  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.
 *  3. The names of the authors may not be used to endorse or promote
 *     products derived from this software without specific prior
 *     written permission.
 *  
 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
 *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
 *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
 */
#ifndef lint
static const char rcsid[] =
    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.72 2001-12-10 05:49:40 guy Exp $ (LBL)";
#endif

/*
 * Known problems with 2.0[.x] kernels:
 *
 *   - The loopback device gives every packet twice; on 2.2[.x] kernels,
 *     if we use PF_PACKET, we can filter out the transmitted version
 *     of the packet by using data in the "sockaddr_ll" returned by
 *     "recvfrom()", but, on 2.0[.x] kernels, we have to use
 *     PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a
 *     "sockaddr_pkt" which doesn't give us enough information to let
 *     us do that.
 *
 *   - We have to set the interface's IFF_PROMISC flag ourselves, if
 *     we're to run in promiscuous mode, which means we have to turn
 *     it off ourselves when we're done; the kernel doesn't keep track
 *     of how many sockets are listening promiscuously, which means
 *     it won't get turned off automatically when no sockets are
 *     listening promiscuously.  We catch "pcap_close()" and, for
 *     interfaces we put into promiscuous mode, take them out of
 *     promiscuous mode - which isn't necessarily the right thing to
 *     do, if another socket also requested promiscuous mode between
 *     the time when we opened the socket and the time when we close
 *     the socket.
 *
 *   - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()"
 *     return the amount of data that you could have read, rather than
 *     the amount that was returned, so we can't just allocate a buffer
 *     whose size is the snapshot length and pass the snapshot length
 *     as the byte count, and also pass MSG_TRUNC, so that the return
 *     value tells us how long the packet was on the wire.
 *
 *     This means that, if we want to get the actual size of the packet,
 *     so we can return it in the "len" field of the packet header,
 *     we have to read the entire packet, not just the part that fits
 *     within the snapshot length, and thus waste CPU time copying data
 *     from the kernel that our caller won't see.
 *
 *     We have to get the actual size, and supply it in "len", because
 *     otherwise, the IP dissector in tcpdump, for example, will complain
 *     about "truncated-ip", as the packet will appear to have been
 *     shorter, on the wire, than the IP header said it should have been.
 */


#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include "pcap-int.h"
#include "sll.h"

#include <errno.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/ioctl.h>
#include <sys/utsname.h>
#include <net/if.h>
#include <netinet/in.h>
#include <linux/if_ether.h>
#include <net/if_arp.h>

/*
 * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
 * sockets rather than SOCK_PACKET sockets.
 *
 * To use them, we include <linux/if_packet.h> rather than
 * <netpacket/packet.h>; we do so because
 *
 *      some Linux distributions (e.g., Slackware 4.0) have 2.2 or
 *      later kernels and libc5, and don't provide a <netpacket/packet.h>
 *      file;
 *
 *      not all versions of glibc2 have a <netpacket/packet.h> file
 *      that defines stuff needed for some of the 2.4-or-later-kernel
 *      features, so if the system has a 2.4 or later kernel, we
 *      still can't use those features.
 *
 * We're already including a number of other <linux/XXX.h> headers, and
 * this code is Linux-specific (no other OS has PF_PACKET sockets as
 * a raw packet capture mechanism), so it's not as if you gain any
 * useful portability by using <netpacket/packet.h>
 *
 * XXX - should we just include <linux/if_packet.h> even if PF_PACKET
 * isn't defined?  It only defines one data structure in 2.0.x, so
 * it shouldn't cause any problems.
 */
#ifdef PF_PACKET  
# include <linux/if_packet.h>

 /*
  * On at least some Linux distributions (for example, Red Hat 5.2),
  * there's no <netpacket/packet.h> file, but PF_PACKET is defined if
  * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define
  * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of
  * the PACKET_xxx stuff.
  *
  * So we check whether PACKET_HOST is defined, and assume that we have
  * PF_PACKET sockets only if it is defined.
  */
# ifdef PACKET_HOST
#  define HAVE_PF_PACKET_SOCKETS
# endif /* PACKET_HOST */
#endif /* PF_PACKET */

#ifdef SO_ATTACH_FILTER
#include <linux/types.h>
#include <linux/filter.h>
#endif

#ifndef __GLIBC__
typedef int             socklen_t;
#endif

#ifndef MSG_TRUNC
/*
 * This is being compiled on a system that lacks MSG_TRUNC; define it
 * with the value it has in the 2.2 and later kernels, so that, on
 * those kernels, when we pass it in the flags argument to "recvfrom()"
 * we're passing the right value and thus get the MSG_TRUNC behavior
 * we want.  (We don't get that behavior on 2.0[.x] kernels, because
 * they didn't support MSG_TRUNC.)
 */
#define MSG_TRUNC       0x20
#endif

#define MAX_LINKHEADER_SIZE     256

/* 
 * When capturing on all interfaces we use this as the buffer size. 
 * Should be bigger then all MTUs that occur in real life.
 * 64kB should be enough for now.
 */
#define BIGGER_THAN_ALL_MTUS    (64*1024)

/*
 * Prototypes for internal functions
 */
static void map_arphrd_to_dlt(pcap_t *, int);
static int live_open_old(pcap_t *, char *, int, int, char *);
static int live_open_new(pcap_t *, char *, int, int, char *);
static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);

/*
 * Wrap some ioctl calls
 */
#ifdef HAVE_PF_PACKET_SOCKETS
static int      iface_get_id(int fd, const char *device, char *ebuf);
#endif
static int      iface_get_mtu(int fd, const char *device, char *ebuf);
static int      iface_get_arptype(int fd, const char *device, char *ebuf);
#ifdef HAVE_PF_PACKET_SOCKETS
static int      iface_bind(int fd, int ifindex, char *ebuf);
#endif
static int      iface_bind_old(int fd, const char *device, char *ebuf);

#ifdef SO_ATTACH_FILTER
static int      fix_program(pcap_t *handle, struct sock_fprog *fcode);
static int      fix_offset(struct bpf_insn *p);
static int      set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
static int      reset_kernel_filter(pcap_t *handle);

static struct sock_filter       total_insn
        = BPF_STMT(BPF_RET | BPF_K, 0);
static struct sock_fprog        total_fcode
        = { 1, &total_insn };
#endif

/*
 *  Get a handle for a live capture from the given device. You can 
 *  pass NULL as device to get all packages (without link level 
 *  information of course). If you pass 1 as promisc the interface
 *  will be set to promiscous mode (XXX: I think this usage should 
 *  be deprecated and functions be added to select that later allow
 *  modification of that values -- Torsten).
 *  
 *  See also pcap(3).
 */
pcap_t *
pcap_open_live(char *device, int snaplen, int promisc, int to_ms, char *ebuf)
{
        pcap_t          *handle;
        int             mtu;
        struct utsname  utsname;

        /* Allocate a handle for this session. */

        handle = malloc(sizeof(*handle));
        if (handle == NULL) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
                         pcap_strerror(errno));
                return NULL;
        }

        /* Initialize some components of the pcap structure. */

        memset(handle, 0, sizeof(*handle));
        handle->snapshot        = snaplen;
        handle->md.timeout      = to_ms;

        /*
         * NULL and "any" are special devices which give us the hint to 
         * monitor all devices.
         */
        if (!device || strcmp(device, "any") == 0) {
                device                  = NULL;
                handle->md.device       = strdup("any");
                if (promisc) {
                        promisc = 0;
                        /* Just a warning. */
                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                            "Promiscuous mode not supported on the \"any\" device");
                }
        
        } else
                handle->md.device       = strdup(device);

        if (handle->md.device == NULL) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE, "strdup: %s",
                         pcap_strerror(errno) );
                free(handle);
                return NULL;
        }

        /* 
         * Current Linux kernels use the protocol family PF_PACKET to 
         * allow direct access to all packets on the network while 
         * older kernels had a special socket type SOCK_PACKET to 
         * implement this feature.
         * While this old implementation is kind of obsolete we need
         * to be compatible with older kernels for a while so we are 
         * trying both methods with the newer method preferred.
         */

        if (! (live_open_new(handle, device, promisc, to_ms, ebuf) ||
               live_open_old(handle, device, promisc, to_ms, ebuf)) )
        {
                /* 
                 * Both methods to open the packet socket failed. Tidy
                 * up and report our failure (ebuf is expected to be
                 * set by the functions above). 
                 */

                free(handle->md.device);
                free(handle);
                return NULL;
        }

        /*
         * Compute the buffer size.
         *
         * If we're using SOCK_PACKET, this might be a 2.0[.x] kernel,
         * and might require special handling - check.
         */
        if (handle->md.sock_packet && (uname(&utsname) < 0 ||
            strncmp(utsname.release, "2.0", 3) == 0)) {
                /*
                 * We're using a SOCK_PACKET structure, and either
                 * we couldn't find out what kernel release this is,
                 * or it's a 2.0[.x] kernel.
                 *
                 * In the 2.0[.x] kernel, a "recvfrom()" on
                 * a SOCK_PACKET socket, with MSG_TRUNC set, will
                 * return the number of bytes read, so if we pass
                 * a length based on the snapshot length, it'll
                 * return the number of bytes from the packet
                 * copied to userland, not the actual length
                 * of the packet.
                 *
                 * This means that, for example, the IP dissector
                 * in tcpdump will get handed a packet length less
                 * than the length in the IP header, and will
                 * complain about "truncated-ip".
                 *
                 * So we don't bother trying to copy from the
                 * kernel only the bytes in which we're interested,
                 * but instead copy them all, just as the older
                 * versions of libpcap for Linux did.
                 *
                 * The buffer therefore needs to be big enough to
                 * hold the largest packet we can get from this
                 * device.  Unfortunately, we can't get the MRU
                 * of the network; we can only get the MTU.  The
                 * MTU may be too small, in which case a packet larger
                 * than the buffer size will be truncated *and* we
                 * won't get the actual packet size.
                 *
                 * However, if the snapshot length is larger than
                 * the buffer size based on the MTU, we use the
                 * snapshot length as the buffer size, instead;
                 * this means that with a sufficiently large snapshot
                 * length we won't artificially truncate packets
                 * to the MTU-based size.
                 *
                 * This mess just one of many problems with packet
                 * capture on 2.0[.x] kernels; you really want a
                 * 2.2[.x] or later kernel if you want packet capture
                 * to work well.
                 */
                mtu = iface_get_mtu(handle->fd, device, ebuf);
                if (mtu == -1) {
                        close(handle->fd);
                        free(handle->md.device);
                        free(handle);
                        return NULL;
                }
                handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
                if (handle->bufsize < handle->snapshot)
                        handle->bufsize = handle->snapshot;
        } else {
                /*
                 * This is a 2.2[.x] or later kernel (we know that
                 * either because we're not using a SOCK_PACKET
                 * socket - PF_PACKET is supported only in 2.2
                 * and later kernels - or because we checked the
                 * kernel version).
                 *
                 * We can safely pass "recvfrom()" a byte count
                 * based on the snapshot length.
                 */
                handle->bufsize = handle->snapshot;
        }

        /* Allocate the buffer */

        handle->buffer   = malloc(handle->bufsize + handle->offset);
        if (!handle->buffer) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "malloc: %s", pcap_strerror(errno));
                close(handle->fd);
                free(handle->md.device);
                free(handle);
                return NULL;
        }

        return handle;
}

/*
 *  Read at most max_packets from the capture stream and call the callback
 *  for each of them. Returns the number of packets handled or -1 if an
 *  error occured. 
 */
int
pcap_read(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
{
        /*
         * Currently, on Linux only one packet is delivered per read,
         * so we don't loop.
         */
        return pcap_read_packet(handle, callback, user);
}

/*
 *  Read a packet from the socket calling the handler provided by 
 *  the user. Returns the number of packets received or -1 if an
 *  error occured.
 */
static int
pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
{
        u_char                  *bp;
        int                     offset;
#ifdef HAVE_PF_PACKET_SOCKETS
        struct sockaddr_ll      from;
        struct sll_header       *hdrp;
#else
        struct sockaddr         from;
#endif
        socklen_t               fromlen;
        int                     packet_len, caplen;
        struct pcap_pkthdr      pcap_header;

#ifdef HAVE_PF_PACKET_SOCKETS
        /*
         * If this is a cooked device, leave extra room for a
         * fake packet header.
         */
        if (handle->md.cooked)
                offset = SLL_HDR_LEN;
        else
                offset = 0;
#else
        /*
         * This system doesn't have PF_PACKET sockets, so it doesn't
         * support cooked devices.
         */
        offset = 0;
#endif

        /* Receive a single packet from the kernel */

        bp = handle->buffer + handle->offset;
        do {
                fromlen = sizeof(from);
                packet_len = recvfrom( 
                        handle->fd, bp + offset,
                        handle->bufsize - offset, MSG_TRUNC, 
                        (struct sockaddr *) &from, &fromlen);
        } while (packet_len == -1 && errno == EINTR);

        /* Check if an error occured */

        if (packet_len == -1) {
                if (errno == EAGAIN)
                        return 0;       /* no packet there */
                else {
                        snprintf(handle->errbuf, sizeof(handle->errbuf),
                                 "recvfrom: %s", pcap_strerror(errno));
                        return -1;
                }
        }

#ifdef HAVE_PF_PACKET_SOCKETS
        /*
         * If this is from the loopback device, reject outgoing packets;
         * we'll see the packet as an incoming packet as well, and
         * we don't want to see it twice.
         *
         * We can only do this if we're using PF_PACKET; the address
         * returned for SOCK_PACKET is a "sockaddr_pkt" which lacks
         * the relevant packet type information.
         */
        if (!handle->md.sock_packet &&
            from.sll_ifindex == handle->md.lo_ifindex &&
            from.sll_pkttype == PACKET_OUTGOING)
                return 0;
#endif

#ifdef HAVE_PF_PACKET_SOCKETS
        /*
         * If this is a cooked device, fill in the fake packet header.
         */
        if (handle->md.cooked) {
                /*
                 * Add the length of the fake header to the length
                 * of packet data we read.
                 */
                packet_len += SLL_HDR_LEN;

                hdrp = (struct sll_header *)bp;

                /*
                 * Map the PACKET_ value to a LINUX_SLL_ value; we
                 * want the same numerical value to be used in
                 * the link-layer header even if the numerical values
                 * for the PACKET_ #defines change, so that programs
                 * that look at the packet type field will always be
                 * able to handle DLT_LINUX_SLL captures.
                 */
                switch (from.sll_pkttype) {

                case PACKET_HOST:
                        hdrp->sll_pkttype = htons(LINUX_SLL_HOST);
                        break;

                case PACKET_BROADCAST:
                        hdrp->sll_pkttype = htons(LINUX_SLL_BROADCAST);
                        break;

                case PACKET_MULTICAST:
                        hdrp->sll_pkttype = htons(LINUX_SLL_MULTICAST);
                        break;

                case PACKET_OTHERHOST:
                        hdrp->sll_pkttype = htons(LINUX_SLL_OTHERHOST);
                        break;

                case PACKET_OUTGOING:
                        hdrp->sll_pkttype = htons(LINUX_SLL_OUTGOING);
                        break;

                default:
                        hdrp->sll_pkttype = -1;
                        break;
                }

                hdrp->sll_hatype = htons(from.sll_hatype);
                hdrp->sll_halen = htons(from.sll_halen);
                memcpy(hdrp->sll_addr, from.sll_addr,
                    (from.sll_halen > SLL_ADDRLEN) ?
                      SLL_ADDRLEN :
                      from.sll_halen);
                hdrp->sll_protocol = from.sll_protocol;
        }
#endif

        /*
         * XXX: According to the kernel source we should get the real 
         * packet len if calling recvfrom with MSG_TRUNC set. It does 
         * not seem to work here :(, but it is supported by this code
         * anyway. 
         * To be honest the code RELIES on that feature so this is really
         * broken with 2.2.x kernels.
         * I spend a day to figure out what's going on and I found out
         * that the following is happening: 
         *
         * The packet comes from a random interface and the packet_rcv 
         * hook is called with a clone of the packet. That code inserts
         * the packet into the receive queue of the packet socket.
         * If a filter is attached to that socket that filter is run
         * first - and there lies the problem. The default filter always
         * cuts the packet at the snaplen:
         *
         * # tcpdump -d
         * (000) ret      #68
         *
         * So the packet filter cuts down the packet. The recvfrom call 
         * says "hey, it's only 68 bytes, it fits into the buffer" with
         * the result that we don't get the real packet length. This 
         * is valid at least until kernel 2.2.17pre6. 
         *
         * We currently handle this by making a copy of the filter
         * program, fixing all "ret" instructions with non-zero
         * operands to have an operand of 65535 so that the filter
         * doesn't truncate the packet, and supplying that modified
         * filter to the kernel.
         */

        caplen = packet_len;
        if (caplen > handle->snapshot)
                caplen = handle->snapshot;

        /* Run the packet filter if not using kernel filter */
        if (!handle->md.use_bpf && handle->fcode.bf_insns) {
                if (bpf_filter(handle->fcode.bf_insns, bp,
                                packet_len, caplen) == 0)
                {
                        /* rejected by filter */
                        return 0;
                }
        }

        /* Fill in our own header data */

        if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
                snprintf(handle->errbuf, sizeof(handle->errbuf),
                         "ioctl: %s", pcap_strerror(errno));
                return -1;
        }
        pcap_header.caplen      = caplen;
        pcap_header.len         = packet_len;

        /*
         * Count the packet.
         *
         * Arguably, we should count them before we check the filter,
         * as on many other platforms "ps_recv" counts packets
         * handed to the filter rather than packets that passed
         * the filter, but if filtering is done in the kernel, we
         * can't get a count of packets that passed the filter,
         * and that would mean the meaning of "ps_recv" wouldn't
         * be the same on all Linux systems.
         *
         * XXX - it's not the same on all systems in any case;
         * ideally, we should have a "get the statistics" call
         * that supplies more counts and indicates which of them
         * it supplies, so that we supply a count of packets
         * handed to the filter only on platforms where that
         * information is available.
         *
         * We count them here even if we can get the packet count
         * from the kernel, as we can only determine at run time
         * whether we'll be able to get it from the kernel (if
         * HAVE_TPACKET_STATS isn't defined, we can't get it from
         * the kernel, but if it is defined, the library might
         * have been built with a 2.4 or later kernel, but we
         * might be running on a 2.2[.x] kernel without Alexey
         * Kuznetzov's turbopacket patches, and thus the kernel
         * might not be able to supply those statistics).  We
         * could, I guess, try, when opening the socket, to get
         * the statistics, and if we can not increment the count
         * here, but it's not clear that always incrementing
         * the count is more expensive than always testing a flag
         * in memory.
         */
        handle->md.stat.ps_recv++;

        /* Call the user supplied callback function */
        callback(userdata, &pcap_header, bp);

        return 1;
}

/*
 *  Get the statistics for the given packet capture handle.
 *  Reports the number of dropped packets iff the kernel supports
 *  the PACKET_STATISTICS "getsockopt()" argument (2.4 and later
 *  kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket
 *  patches); otherwise, that information isn't available, and we lie
 *  and report 0 as the count of dropped packets.
 */
int
pcap_stats(pcap_t *handle, struct pcap_stat *stats)
{
#ifdef HAVE_TPACKET_STATS
        struct tpacket_stats kstats;
        socklen_t len = sizeof (struct tpacket_stats);

        /*
         * Try to get the packet counts from the kernel.
         */
        if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
                        &kstats, &len) > -1) {
                /*
                 * In "linux/net/packet/af_packet.c", at least in the
                 * 2.4.9 kernel, "tp_packets" is incremented for every
                 * packet that passes the packet filter *and* is
                 * successfully queued on the socket; "tp_drops" is
                 * incremented for every packet dropped because there's
                 * not enough free space in the socket buffer.
                 *
                 * When the statistics are returned for a PACKET_STATISTICS
                 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
                 * so that "tp_packets" counts all packets handed to
                 * the PF_PACKET socket, including packets dropped because
                 * there wasn't room on the socket buffer - but not
                 * including packets that didn't pass the filter.
                 *
                 * In the BSD BPF, the count of received packets is
                 * incremented for every packet handed to BPF, regardless
                 * of whether it passed the filter.
                 *
                 * We can't make "pcap_stats()" work the same on both
                 * platforms, but the best approximation is to return
                 * "tp_packets" as the count of packets and "tp_drops"
                 * as the count of drops.
                 *
                 * Note that "tp_packets" may include packets not yet
                 * processed by libpcap; the packets may have been
                 * queued on the socket but not read yet.  (The equivalent
                 * may be true on other platforms.)
                 */
                handle->md.stat.ps_recv = kstats.tp_packets;
                handle->md.stat.ps_drop = kstats.tp_drops;
        }
        else
        {
                /*
                 * If the error was EOPNOTSUPP, fall through, so that
                 * if you build the library on a system with
                 * "struct tpacket_stats" and run it on a system
                 * that doesn't, it works as it does if the library
                 * is built on a system without "struct tpacket_stats".
                 */
                if (errno != EOPNOTSUPP) {
                        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                            "pcap_stats: %s", pcap_strerror(errno));
                        return -1;
                }
        }
#endif
        /*
         * "ps_recv" counts only packets that passed the filter.
         *
         * "ps_drop" is maintained only on systems that support
         * the PACKET_STATISTICS "getsockopt()" argument.
         */
        *stats = handle->md.stat;
        return 0;
}

/*
 *  Attach the given BPF code to the packet capture device. 
 */
int
pcap_setfilter(pcap_t *handle, struct bpf_program *filter)
{
#ifdef SO_ATTACH_FILTER
        struct sock_fprog       fcode;
        int                     can_filter_in_kernel;
#endif

        if (!handle)
                return -1;
        if (!filter) {
                strncpy(handle->errbuf, "setfilter: No filter specified",
                        sizeof(handle->errbuf));
                return -1;
        }

        /* Make our private copy of the filter */

        if (install_bpf_program(handle, filter) < 0) {
                snprintf(handle->errbuf, sizeof(handle->errbuf),
                         "malloc: %s", pcap_strerror(errno));
                return -1;
        }

        /* 
         * Run user level packet filter by default. Will be overriden if 
         * installing a kernel filter succeeds. 
         */
        handle->md.use_bpf = 0;

        /*
         * If we're reading from a savefile, don't try to install
         * a kernel filter.
         */
        if (handle->sf.rfile != NULL)
                return 0;

        /* Install kernel level filter if possible */

#ifdef SO_ATTACH_FILTER
#ifdef USHRT_MAX
        if (handle->fcode.bf_len > USHRT_MAX) {
                /*
                 * fcode.len is an unsigned short for current kernel. 
                 * I have yet to see BPF-Code with that much
                 * instructions but still it is possible. So for the
                 * sake of correctness I added this check.
                 */
                fprintf(stderr, "Warning: Filter too complex for kernel\n");
                fcode.filter = NULL;
                can_filter_in_kernel = 0;
        } else
#endif /* USHRT_MAX */
        {
                /*
                 * Oh joy, the Linux kernel uses struct sock_fprog instead
                 * of struct bpf_program and of course the length field is
                 * of different size. Pointed out by Sebastian
                 *
                 * Oh, and we also need to fix it up so that all "ret"
                 * instructions with non-zero operands have 65535 as the
                 * operand, and so that, if we're in cooked mode, all
                 * memory-reference instructions use special magic offsets
                 * in references to the link-layer header and assume that
                 * the link-layer payload begins at 0; "fix_program()"
                 * will do that.
                 */
                switch (fix_program(handle, &fcode)) {

                case -1:
                default:
                        /*
                         * Fatal error; just quit.
                         * (The "default" case shouldn't happen; we
                         * return -1 for that reason.)
                         */
                        return -1;

                case 0:
                        /*
                         * The program performed checks that we can't make
                         * work in the kernel.
                         */
                        can_filter_in_kernel = 0;
                        break;

                case 1:
                        /*
                         * We have a filter that'll work in the kernel.
                         */
                        can_filter_in_kernel = 1;
                        break;
                }
        }

        if (can_filter_in_kernel) {
                if (set_kernel_filter(handle, &fcode) == 0)
                {
                        /* Installation succeded - using kernel filter. */
                        handle->md.use_bpf = 1;
                }
                else
                {
                        /* 
                         * Print a warning if we weren't able to install
                         * the filter for a reason other than "this kernel
                         * isn't configured to support socket filters.
                         */
                        if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
                                fprintf(stderr,
                                    "Warning: Kernel filter failed: %s\n", 
                                        pcap_strerror(errno));
                        }
                }
        }

        /*
         * If we're not using the kernel filter, get rid of any kernel
         * filter that might've been there before, e.g. because the
         * previous filter could work in the kernel, or because some other
         * code attached a filter to the socket by some means other than
         * calling "pcap_setfilter()".  Otherwise, the kernel filter may
         * filter out packets that would pass the new userland filter.
         */
        if (!handle->md.use_bpf)
                reset_kernel_filter(handle);

        /*
         * Free up the copy of the filter that was made by "fix_program()".
         */
        if (fcode.filter != NULL)
                free(fcode.filter);
#endif /* SO_ATTACH_FILTER */

        return 0;
}

/*
 *  Linux uses the ARP hardware type to identify the type of an 
 *  interface. pcap uses the DLT_xxx constants for this. This 
 *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
 *  constant, as arguments, and sets "handle->linktype" to the
 *  appropriate DLT_XXX constant and sets "handle->offset" to
 *  the appropriate value (to make "handle->offset" plus link-layer
 *  header length be a multiple of 4, so that the link-layer payload
 *  will be aligned on a 4-byte boundary when capturing packets).
 *  (If the offset isn't set here, it'll be 0; add code as appropriate
 *  for cases where it shouldn't be 0.)
 *  
 *  Sets the link type to -1 if unable to map the type.
 */
static void map_arphrd_to_dlt(pcap_t *handle, int arptype)
{
        switch (arptype) {

        case ARPHRD_ETHER:
        case ARPHRD_METRICOM:
        case ARPHRD_LOOPBACK:
                handle->linktype = DLT_EN10MB;
                handle->offset = 2;
                break;

        case ARPHRD_EETHER:
                handle->linktype = DLT_EN3MB;
                break;

        case ARPHRD_AX25:
                handle->linktype = DLT_AX25;
                break;

        case ARPHRD_PRONET:
                handle->linktype = DLT_PRONET;
                break;

        case ARPHRD_CHAOS:
                handle->linktype = DLT_CHAOS;
                break;

#ifndef ARPHRD_IEEE802_TR
#define ARPHRD_IEEE802_TR 800   /* From Linux 2.4 */
#endif
        case ARPHRD_IEEE802_TR:
        case ARPHRD_IEEE802:
                handle->linktype = DLT_IEEE802;
                handle->offset = 2;
                break;

        case ARPHRD_ARCNET:
                handle->linktype = DLT_ARCNET;
                break;

        case ARPHRD_FDDI:
                handle->linktype = DLT_FDDI;
                handle->offset = 3;
                break;

#ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
#define ARPHRD_ATM 19
#endif
        case ARPHRD_ATM:
                /*
                 * The Classical IP implementation in ATM for Linux
                 * supports both what RFC 1483 calls "LLC Encapsulation",
                 * in which each packet has an LLC header, possibly
                 * with a SNAP header as well, prepended to it, and
                 * what RFC 1483 calls "VC Based Multiplexing", in which
                 * different virtual circuits carry different network
                 * layer protocols, and no header is prepended to packets.
                 *
                 * They both have an ARPHRD_ type of ARPHRD_ATM, so
                 * you can't use the ARPHRD_ type to find out whether
                 * captured packets will have an LLC header, and,
                 * while there's a socket ioctl to *set* the encapsulation
                 * type, there's no ioctl to *get* the encapsulation type.
                 *
                 * This means that
                 *
                 *      programs that dissect Linux Classical IP frames
                 *      would have to check for an LLC header and,
                 *      depending on whether they see one or not, dissect
                 *      the frame as LLC-encapsulated or as raw IP (I
                 *      don't know whether there's any traffic other than
                 *      IP that would show up on the socket, or whether
                 *      there's any support for IPv6 in the Linux
                 *      Classical IP code);
                 *
                 *      filter expressions would have to compile into
                 *      code that checks for an LLC header and does
                 *      the right thing.
                 *
                 * Both of those are a nuisance - and, at least on systems
                 * that support PF_PACKET sockets, we don't have to put
                 * up with those nuisances; instead, we can just capture
                 * in cooked mode.  That's what we'll do.
                 */
                handle->linktype = DLT_LINUX_SLL;
                break;

#ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
#define ARPHRD_IEEE80211 801
#endif
        case ARPHRD_IEEE80211:
                handle->linktype = DLT_IEEE802_11;
                break;

        case ARPHRD_PPP:
                /*
                 * Some PPP code in the kernel supplies no link-layer
                 * header whatsoever to PF_PACKET sockets; other PPP
                 * code supplies PPP link-layer headers ("syncppp.c");
                 * some PPP code might supply random link-layer
                 * headers (PPP over ISDN - there's code in Ethereal,
                 * for example, to cope with PPP-over-ISDN captures
                 * with which the Ethereal developers have had to cope,
                 * heuristically trying to determine which of the
                 * oddball link-layer headers particular packets have).
                 *
                 * As such, we just punt, and run all PPP interfaces
                 * in cooked mode.
                 */
                handle->linktype = DLT_LINUX_SLL;
                break;

        case ARPHRD_HDLC:
                handle->linktype = DLT_C_HDLC;
                break;

        /* Not sure if this is correct for all tunnels, but it
         * works for CIPE */
        case ARPHRD_TUNNEL:
#ifndef ARPHRD_SIT
#define ARPHRD_SIT 776  /* From Linux 2.2.14 */
#endif
        case ARPHRD_SIT:
        case ARPHRD_CSLIP:
        case ARPHRD_SLIP6:
        case ARPHRD_CSLIP6:
        case ARPHRD_ADAPT:
        case ARPHRD_SLIP:
                /*
                 * XXX - should some of those be mapped to DLT_LINUX_SLL
                 * instead?  Should we just map all of them to DLT_LINUX_SLL?
                 */
                handle->linktype = DLT_RAW;
                break;

        case ARPHRD_LOCALTLK:
                handle->linktype = DLT_LTALK;
                break;

        default:
                handle->linktype = -1;
                break;
        }
}

/* ===== Functions to interface to the newer kernels ================== */

/*
 *  Try to open a packet socket using the new kernel interface.
 *  Returns 0 on failure.
 *  FIXME: 0 uses to mean success (Sebastian)
 */
static int
live_open_new(pcap_t *handle, char *device, int promisc, 
              int to_ms, char *ebuf)
{
#ifdef HAVE_PF_PACKET_SOCKETS
        int                     sock_fd = -1, device_id, arptype;
        struct packet_mreq      mr;

        /* One shot loop used for error handling - bail out with break */

        do {
                /*
                 * Open a socket with protocol family packet. If a device is
                 * given we try to open it in raw mode otherwise we use 
                 * the cooked interface. 
                 */
                sock_fd = device ? 
                        socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))
                      : socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL));

                if (sock_fd == -1) {
                        snprintf(ebuf, PCAP_ERRBUF_SIZE, "socket: %s",
                                 pcap_strerror(errno) );
                        break;
                }

                /* It seems the kernel supports the new interface. */
                handle->md.sock_packet = 0;

                /*
                 * Get the interface index of the loopback device.
                 * If the attempt fails, don't fail, just set the
                 * "md.lo_ifindex" to -1.
                 *
                 * XXX - can there be more than one device that loops
                 * packets back, i.e. devices other than "lo"?  If so,
                 * we'd need to find them all, and have an array of
                 * indices for them, and check all of them in
                 * "pcap_read_packet()".
                 */
                handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", ebuf);

                /*
                 * Default value for offset to align link-layer payload
                 * on a 4-byte boundary.
                 */
                handle->offset   = 0;

                /*
                 * What kind of frames do we have to deal with? Fall back 
                 * to cooked mode if we have an unknown interface type. 
                 */

                if (device) {
                        /* Assume for now we don't need cooked mode. */
                        handle->md.cooked = 0;

                        arptype = iface_get_arptype(sock_fd, device, ebuf);
                        if (arptype == -1) 
                                break;
                        map_arphrd_to_dlt(handle, arptype);
                        if (handle->linktype == -1 ||
                            handle->linktype == DLT_LINUX_SLL ||
                            (handle->linktype == DLT_EN10MB &&
                             (strncmp("isdn", device, 4) == 0 ||
                              strncmp("isdY", device, 4) == 0))) {
                                /*
                                 * Unknown interface type (-1), or a
                                 * device we explicitly chose to run
                                 * in cooked mode (e.g., PPP devices),
                                 * or an ISDN device (whose link-layer
                                 * type we can only determine by using
                                 * APIs that may be different on different
                                 * kernels) - reopen in cooked mode.
                                 */
                                if (close(sock_fd) == -1) {
                                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                                 "close: %s", pcap_strerror(errno));
                                        break;
                                }
                                sock_fd = socket(PF_PACKET, SOCK_DGRAM, 
                                                 htons(ETH_P_ALL));
                                if (sock_fd == -1) {
                                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                                 "socket: %s", pcap_strerror(errno));
                                        break;
                                }
                                handle->md.cooked = 1;

                                if (handle->linktype == -1) {
                                        /*
                                         * Warn that we're falling back on
                                         * cooked mode; we may want to
                                         * update "map_arphrd_to_dlt()"
                                         * to handle the new type.
                                         */
                                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                                "arptype %d not "
                                                "supported by libpcap - "
                                                "falling back to cooked "
                                                "socket",
                                                arptype);
                                }
                                handle->linktype = DLT_LINUX_SLL;
                        }

                        device_id = iface_get_id(sock_fd, device, ebuf);
                        if (device_id == -1)
                                break;

                        if (iface_bind(sock_fd, device_id, ebuf) == -1)
                                break;
                } else {
                        /*
                         * This is cooked mode.
                         */
                        handle->md.cooked = 1;
                        handle->linktype = DLT_LINUX_SLL;

                        /*
                         * XXX - squelch GCC complaints about
                         * uninitialized variables; if we can't
                         * select promiscuous mode on all interfaces,
                         * we should move the code below into the
                         * "if (device)" branch of the "if" and
                         * get rid of the next statement.
                         */
                        device_id = -1;
                }

                /* Select promiscuous mode on/off */

#ifdef SOL_PACKET
                /* 
                 * Hmm, how can we set promiscuous mode on all interfaces?
                 * I am not sure if that is possible at all.
                 */

                if (device) {
                        memset(&mr, 0, sizeof(mr));
                        mr.mr_ifindex = device_id;
                        mr.mr_type    = promisc ? 
                                PACKET_MR_PROMISC : PACKET_MR_ALLMULTI;
                        if (setsockopt(sock_fd, SOL_PACKET, 
                                PACKET_ADD_MEMBERSHIP, &mr, sizeof(mr)) == -1)
                        {
                                snprintf(ebuf, PCAP_ERRBUF_SIZE, 
                                        "setsockopt: %s", pcap_strerror(errno));
                                break;
                        }
                }
#endif

                /* Save the socket FD in the pcap structure */

                handle->fd       = sock_fd;

                return 1;

        } while(0);

        if (sock_fd != -1)
                close(sock_fd);
        return 0;
#else
        strncpy(ebuf, 
                "New packet capturing interface not supported by build " 
                "environment", PCAP_ERRBUF_SIZE);
        return 0;
#endif
}

#ifdef HAVE_PF_PACKET_SOCKETS
/*
 *  Return the index of the given device name. Fill ebuf and return 
 *  -1 on failure.
 */
static int
iface_get_id(int fd, const char *device, char *ebuf)
{
        struct ifreq    ifr;

        memset(&ifr, 0, sizeof(ifr));
        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));

        if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "ioctl: %s", pcap_strerror(errno));
                return -1;
        }

        return ifr.ifr_ifindex;
}

/*
 *  Bind the socket associated with FD to the given device. 
 */
static int
iface_bind(int fd, int ifindex, char *ebuf)
{
        struct sockaddr_ll      sll;

        memset(&sll, 0, sizeof(sll));
        sll.sll_family          = AF_PACKET;
        sll.sll_ifindex         = ifindex;
        sll.sll_protocol        = htons(ETH_P_ALL);

        if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "bind: %s", pcap_strerror(errno));
                return -1;
        }

        return 0;
}

#endif


/* ===== Functions to interface to the older kernels ================== */

/*
 * With older kernels promiscuous mode is kind of interesting because we
 * have to reset the interface before exiting. The problem can't really
 * be solved without some daemon taking care of managing usage counts. 
 * If we put the interface into promiscuous mode, we set a flag indicating
 * that we must take it out of that mode when the interface is closed,
 * and, when closing the interface, if that flag is set we take it out
 * of promiscuous mode.
 */

/*
 * List of pcaps for which we turned promiscuous mode on by hand.
 * If there are any such pcaps, we arrange to call "pcap_close_all()"
 * when we exit, and have it close all of them to turn promiscuous mode
 * off.
 */
static struct pcap *pcaps_to_close;

/*
 * TRUE if we've already called "atexit()" to cause "pcap_close_all()" to
 * be called on exit.
 */
static int did_atexit;

static void     pcap_close_all(void)
{
        struct pcap *handle;

        while ((handle = pcaps_to_close) != NULL)
                pcap_close(handle);
}

void    pcap_close_linux( pcap_t *handle )
{
        struct pcap     *p, *prevp;
        struct ifreq    ifr;

        if (handle->md.clear_promisc) {
                /*
                 * We put the interface into promiscuous mode; take
                 * it out of promiscuous mode.
                 *
                 * XXX - if somebody else wants it in promiscuous mode,
                 * this code cannot know that, so it'll take it out
                 * of promiscuous mode.  That's not fixable in 2.0[.x]
                 * kernels.
                 */
                memset(&ifr, 0, sizeof(ifr));
                strncpy(ifr.ifr_name, handle->md.device, sizeof(ifr.ifr_name));
                if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
                        fprintf(stderr, 
                            "Can't restore interface flags (SIOCGIFFLAGS failed: %s).\n"
                            "Please adjust manually.\n"
                            "Hint: This can't happen with Linux >= 2.2.0.\n",
                            strerror(errno));
                } else {
                        if (ifr.ifr_flags & IFF_PROMISC) {
                                /*
                                 * Promiscuous mode is currently on; turn it
                                 * off.
                                 */
                                ifr.ifr_flags &= ~IFF_PROMISC;
                                if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
                                        fprintf(stderr, 
                                            "Can't restore interface flags (SIOCSIFFLAGS failed: %s).\n"
                                            "Please adjust manually.\n"
                                            "Hint: This can't happen with Linux >= 2.2.0.\n",
                                            strerror(errno));
                                }
                        }
                }

                /*
                 * Take this pcap out of the list of pcaps for which we
                 * have to take the interface out of promiscuous mode.
                 */
                for (p = pcaps_to_close, prevp = NULL; p != NULL;
                    prevp = p, p = p->md.next) {
                        if (p == handle) {
                                /*
                                 * Found it.  Remove it from the list.
                                 */
                                if (prevp == NULL) {
                                        /*
                                         * It was at the head of the list.
                                         */
                                        pcaps_to_close = p->md.next;
                                } else {
                                        /*
                                         * It was in the middle of the list.
                                         */
                                        prevp->md.next = p->md.next;
                                }
                                break;
                        }
                }
        }
        if (handle->md.device != NULL)
                free(handle->md.device);
}

/*
 *  Try to open a packet socket using the old kernel interface.
 *  Returns 0 on failure.
 *  FIXME: 0 uses to mean success (Sebastian)
 */
static int
live_open_old(pcap_t *handle, char *device, int promisc, 
              int to_ms, char *ebuf)
{
        int             sock_fd = -1, arptype;
        struct ifreq    ifr;

        do {
                /* Open the socket */

                sock_fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
                if (sock_fd == -1) {
                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                 "socket: %s", pcap_strerror(errno));
                        break;
                }

                /* It worked - we are using the old interface */
                handle->md.sock_packet = 1;

                /* ...which means we get the link-layer header. */
                handle->md.cooked = 0;

                /* Bind to the given device */

                if (!device) {
                        strncpy(ebuf, "pcap_open_live: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
                                PCAP_ERRBUF_SIZE);
                        break;
                }
                if (iface_bind_old(sock_fd, device, ebuf) == -1)
                        break;

                /* Go to promisc mode */
                if (promisc) {
                        memset(&ifr, 0, sizeof(ifr));
                        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
                        if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
                                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                         "ioctl: %s", pcap_strerror(errno));
                                break;
                        }
                        if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
                                /*
                                 * Promiscuous mode isn't currently on,
                                 * so turn it on, and remember that
                                 * we should turn it off when the
                                 * pcap_t is closed.
                                 */

                                /*
                                 * If we haven't already done so, arrange
                                 * to have "pcap_close_all()" called when
                                 * we exit.
                                 */
                                if (!did_atexit) {
                                        if (atexit(pcap_close_all) == -1) {
                                                /*
                                                 * "atexit()" failed; don't
                                                 * put the interface in
                                                 * promiscuous mode, just
                                                 * give up.
                                                 */
                                                strncpy(ebuf, "atexit failed",
                                                        PCAP_ERRBUF_SIZE);
                                                break;
                                        }
                                }

                                ifr.ifr_flags |= IFF_PROMISC;
                                if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
                                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                                 "ioctl: %s",
                                                 pcap_strerror(errno));
                                        break;
                                }
                                handle->md.clear_promisc = 1;

                                /*
                                 * Add this to the list of pcaps
                                 * to close when we exit.
                                 */
                                handle->md.next = pcaps_to_close;
                                pcaps_to_close = handle;
                        }
                }

                /* All done - fill in the pcap handle */

                arptype = iface_get_arptype(sock_fd, device, ebuf);
                if (arptype == -1)
                        break;

                /* Save the socket FD in the pcap structure */

                handle->fd       = sock_fd;

                /*
                 * Default value for offset to align link-layer payload
                 * on a 4-byte boundary.
                 */
                handle->offset   = 0;

                /*
                 * XXX - handle ISDN types here?  We can't fall back on
                 * cooked sockets, so we'd have to figure out from the
                 * device name what type of link-layer encapsulation
                 * it's using, and map that to an appropriate DLT_
                 * value, meaning we'd map "isdnN" devices to DLT_RAW
                 * (they supply raw IP packets with no link-layer
                 * header) and "isdY" devices to a new DLT_I4L_IP
                 * type that has only an Ethernet packet type as
                 * a link-layer header.
                 */
                map_arphrd_to_dlt(handle, arptype);
                if (handle->linktype == -1 ||
                    handle->linktype == DLT_LINUX_SLL) {
                        snprintf(ebuf, PCAP_ERRBUF_SIZE,
                                 "interface type of %s not supported", device);
                        break;
                }

                return 1;

        } while (0);

        if (sock_fd != -1)
                close(sock_fd);
        return 0;
}

/*
 *  Bind the socket associated with FD to the given device using the 
 *  interface of the old kernels.
 */
static int
iface_bind_old(int fd, const char *device, char *ebuf)
{
        struct sockaddr saddr;

        memset(&saddr, 0, sizeof(saddr));
        strncpy(saddr.sa_data, device, sizeof(saddr.sa_data));
        if (bind(fd, &saddr, sizeof(saddr)) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "bind: %s", pcap_strerror(errno));
                return -1;
        }

        return 0;
}


/* ===== System calls available on all supported kernels ============== */

/*
 *  Query the kernel for the MTU of the given interface. 
 */
static int
iface_get_mtu(int fd, const char *device, char *ebuf)
{
        struct ifreq    ifr;

        if (!device)
                return BIGGER_THAN_ALL_MTUS;

        memset(&ifr, 0, sizeof(ifr));
        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));

        if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "ioctl: %s", pcap_strerror(errno));
                return -1;
        }

        return ifr.ifr_mtu;
}

/*
 *  Get the hardware type of the given interface as ARPHRD_xxx constant.
 */
static int
iface_get_arptype(int fd, const char *device, char *ebuf)
{
        struct ifreq    ifr;

        memset(&ifr, 0, sizeof(ifr));
        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));

        if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                         "ioctl: %s", pcap_strerror(errno));
                return -1;
        }

        return ifr.ifr_hwaddr.sa_family;
}

#ifdef SO_ATTACH_FILTER
static int
fix_program(pcap_t *handle, struct sock_fprog *fcode)
{
        size_t prog_size;
        register int i;
        register struct bpf_insn *p;
        struct bpf_insn *f;
        int len;

        /*
         * Make a copy of the filter, and modify that copy if
         * necessary.
         */
        prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
        len = handle->fcode.bf_len;
        f = (struct bpf_insn *)malloc(prog_size);
        if (f == NULL) {
                snprintf(handle->errbuf, sizeof(handle->errbuf),
                         "malloc: %s", pcap_strerror(errno));
                return -1;
        }
        memcpy(f, handle->fcode.bf_insns, prog_size);
        fcode->len = len;
        fcode->filter = (struct sock_filter *) f;

        for (i = 0; i < len; ++i) {
                p = &f[i];
                /*
                 * What type of instruction is this?
                 */
                switch (BPF_CLASS(p->code)) {

                case BPF_RET:
                        /*
                         * It's a return instruction; is the snapshot
                         * length a constant, rather than the contents
                         * of the accumulator?
                         */
                        if (BPF_MODE(p->code) == BPF_K) {
                                /*
                                 * Yes - if the value to be returned,
                                 * i.e. the snapshot length, is anything
                                 * other than 0, make it 65535, so that
                                 * the packet is truncated by "recvfrom()",
                                 * not by the filter.
                                 *
                                 * XXX - there's nothing we can easily do
                                 * if it's getting the value from the
                                 * accumulator; we'd have to insert
                                 * code to force non-zero values to be
                                 * 65535.
                                 */
                                if (p->k != 0)
                                        p->k = 65535;
                        }
                        break;

                case BPF_LD:
                case BPF_LDX:
                        /*
                         * It's a load instruction; is it loading
                         * from the packet?
                         */
                        switch (BPF_MODE(p->code)) {

                        case BPF_ABS:
                        case BPF_IND:
                        case BPF_MSH:
                                /*
                                 * Yes; are we in cooked mode?
                                 */
                                if (handle->md.cooked) {
                                        /*
                                         * Yes, so we need to fix this
                                         * instruction.
                                         */
                                        if (fix_offset(p) < 0) {
                                                /*
                                                 * We failed to do so.
                                                 * Return 0, so our caller
                                                 * knows to punt to userland.
                                                 */
                                                return 0;
                                        }
                                }
                                break;
                        }
                        break;
                }
        }
        return 1;       /* we succeeded */
}

static int
fix_offset(struct bpf_insn *p)
{
        /*
         * What's the offset?
         */
        if (p->k >= SLL_HDR_LEN) {
                /*
                 * It's within the link-layer payload; that starts at an
                 * offset of 0, as far as the kernel packet filter is
                 * concerned, so subtract the length of the link-layer
                 * header.
                 */
                p->k -= SLL_HDR_LEN;
        } else if (p->k == 14) {
                /*
                 * It's the protocol field; map it to the special magic
                 * kernel offset for that field.
                 */
                p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
        } else {
                /*
                 * It's within the header, but it's not one of those
                 * fields; we can't do that in the kernel, so punt
                 * to userland.
                 */
                return -1;
        }
        return 0;
}

static int
set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
{
        int total_filter_on = 0;
        int save_mode;
        int ret;
        int save_errno;

        /*
         * The socket filter code doesn't discard all packets queued
         * up on the socket when the filter is changed; this means
         * that packets that don't match the new filter may show up
         * after the new filter is put onto the socket, if those
         * packets haven't yet been read.
         *
         * This means, for example, that if you do a tcpdump capture
         * with a filter, the first few packets in the capture might
         * be packets that wouldn't have passed the filter.
         *
         * We therefore discard all packets queued up on the socket
         * when setting a kernel filter.  (This isn't an issue for
         * userland filters, as the userland filtering is done after
         * packets are queued up.)
         *
         * To flush those packets, we put the socket in read-only mode,
         * and read packets from the socket until there are no more to
         * read.
         *
         * In order to keep that from being an infinite loop - i.e.,
         * to keep more packets from arriving while we're draining
         * the queue - we put the "total filter", which is a filter
         * that rejects all packets, onto the socket before draining
         * the queue.
         *
         * This code deliberately ignores any errors, so that you may
         * get bogus packets if an error occurs, rather than having
         * the filtering done in userland even if it could have been
         * done in the kernel.
         */
        if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER, 
                       &total_fcode, sizeof(total_fcode)) == 0) {
                char drain[1];

                /*
                 * Note that we've put the total filter onto the socket.
                 */
                total_filter_on = 1;

                /*
                 * Save the socket's current mode, and put it in
                 * non-blocking mode; we drain it by reading packets
                 * until we get an error (which we assume is a
                 * "nothing more to be read" error).
                 */
                save_mode = fcntl(handle->fd, F_GETFL, 0);
                if (save_mode != -1 &&
                    fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) >= 0) {
                        while (recv(handle->fd, &drain, sizeof drain,
                               MSG_TRUNC) >= 0)
                                ;
                        fcntl(handle->fd, F_SETFL, save_mode);
                }
        }

        /*
         * Now attach the new filter.
         */
        ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER, 
                         fcode, sizeof(*fcode));
        if (ret == -1 && total_filter_on) {
                /*
                 * Well, we couldn't set that filter on the socket,
                 * but we could set the total filter on the socket.
                 *
                 * This could, for example, mean that the filter was
                 * too big to put into the kernel, so we'll have to
                 * filter in userland; in any case, we'll be doing
                 * filtering in userland, so we need to remove the
                 * total filter so we see packets.
                 */
                save_errno = errno;

                /*
                 * XXX - if this fails, we're really screwed;
                 * we have the total filter on the socket,
                 * and it won't come off.  What do we do then?
                 */
                reset_kernel_filter(handle);

                errno = save_errno;
        }
        return ret;      
}

static int
reset_kernel_filter(pcap_t *handle)
{
        /* setsockopt() barfs unless it get a dummy parameter */
        int dummy;

        return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
                                   &dummy, sizeof(dummy));
}
#endif