]> The Tcpdump Group git mirrors - libpcap/blobdiff - gencode.c
The value pointed to by "gen_pf_ifname()"'s argument isn't modified, so
[libpcap] / gencode.c
index 6eab6ced4f0847b3d9d9e14b816df09ddddf4574..f8765fb5dcb96ccb32582ff6c49010c3b7e81880 100644 (file)
--- a/gencode.c
+++ b/gencode.c
  */
 #ifndef lint
 static const char rcsid[] =
-    "@(#) $Header: /tcpdump/master/libpcap/gencode.c,v 1.113 2000-07-11 00:37:04 assar Exp $ (LBL)";
+    "@(#) $Header: /tcpdump/master/libpcap/gencode.c,v 1.191 2003-05-02 08:37:43 guy Exp $ (LBL)";
 #endif
 
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 
+#ifdef WIN32
+#include <pcap-stdinc.h>
+#else /* WIN32 */
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <sys/time.h>
+#endif /* WIN32 */
+
+/*
+ * XXX - why was this included even on UNIX?
+ */
+#ifdef __MINGW32__
+#include "IP6_misc.h"
+#endif
+
+#ifndef WIN32
+
 #ifdef __NetBSD__
 #include <sys/param.h>
 #endif
 
-struct mbuf;
-struct rtentry;
-#include <net/if.h>
-
 #include <netinet/in.h>
-#include <netinet/if_ether.h>
+
+#endif /* WIN32 */
 
 #include <stdlib.h>
+#include <string.h>
 #include <memory.h>
 #include <setjmp.h>
 #include <stdarg.h>
@@ -50,13 +62,27 @@ struct rtentry;
 #include "pcap-int.h"
 
 #include "ethertype.h"
+#include "nlpid.h"
+#include "llc.h"
 #include "gencode.h"
+#include "atmuni31.h"
+#include "sunatmpos.h"
 #include "ppp.h"
-#include <pcap-namedb.h>
+#include "sll.h"
+#include "arcnet.h"
+#include "pf.h"
 #ifdef INET6
-#include <netdb.h>
-#include <sys/socket.h>
+#ifndef WIN32
+#include <netdb.h>     /* for "struct addrinfo" */
+#endif /* WIN32 */
 #endif /*INET6*/
+#include <pcap-namedb.h>
+
+#define ETHERMTU       1500
+
+#ifndef IPPROTO_SCTP
+#define IPPROTO_SCTP 132
+#endif
 
 #ifdef HAVE_OS_PROTO_H
 #include "os-proto.h"
@@ -68,6 +94,9 @@ struct rtentry;
 static jmp_buf top_ctx;
 static pcap_t *bpf_pcap;
 
+/* Hack for updating VLAN offsets. */
+static u_int   orig_linktype = -1, orig_nl = -1, orig_nl_nosnap = -1;
+
 /* XXX */
 #ifdef PCAP_FDDIPAD
 int    pcap_fddipad = PCAP_FDDIPAD;
@@ -124,18 +153,28 @@ static inline void syntax(void);
 static void backpatch(struct block *, struct block *);
 static void merge(struct block *, struct block *);
 static struct block *gen_cmp(u_int, u_int, bpf_int32);
+static struct block *gen_cmp_gt(u_int, u_int, bpf_int32);
 static struct block *gen_mcmp(u_int, u_int, bpf_int32, bpf_u_int32);
 static struct block *gen_bcmp(u_int, u_int, const u_char *);
+static struct block *gen_ncmp(bpf_u_int32, bpf_u_int32, bpf_u_int32,
+    bpf_u_int32, bpf_u_int32, int);
 static struct block *gen_uncond(int);
 static inline struct block *gen_true(void);
 static inline struct block *gen_false(void);
+static struct block *gen_ether_linktype(int);
 static struct block *gen_linktype(int);
+static struct block *gen_snap(bpf_u_int32, bpf_u_int32, u_int);
+static struct block *gen_llc(int);
 static struct block *gen_hostop(bpf_u_int32, bpf_u_int32, int, int, u_int, u_int);
 #ifdef INET6
 static struct block *gen_hostop6(struct in6_addr *, struct in6_addr *, int, int, u_int, u_int);
 #endif
+static struct block *gen_ahostop(const u_char *, int);
 static struct block *gen_ehostop(const u_char *, int);
 static struct block *gen_fhostop(const u_char *, int);
+static struct block *gen_thostop(const u_char *, int);
+static struct block *gen_wlanhostop(const u_char *, int);
+static struct block *gen_ipfchostop(const u_char *, int);
 static struct block *gen_dnhostop(bpf_u_int32, int, u_int);
 static struct block *gen_host(bpf_u_int32, bpf_u_int32, int, int);
 #ifdef INET6
@@ -156,11 +195,15 @@ struct block *gen_portop6(int, int, int);
 static struct block *gen_port6(int, int, int);
 #endif
 static int lookup_proto(const char *, int);
+static struct block *gen_protochain(int, int, int);
 static struct block *gen_proto(int, int, int);
 static struct slist *xfer_to_x(struct arth *);
 static struct slist *xfer_to_a(struct arth *);
+static struct block *gen_mac_multicast(int);
 static struct block *gen_len(int, int);
 
+static struct block *gen_msg_abbrev(int type);
+
 static void *
 newchunk(n)
        u_int n;
@@ -183,6 +226,8 @@ newchunk(n)
                        bpf_error("out of memory");
                size = CHUNK0SIZE << k;
                cp->m = (void *)malloc(size);
+               if (cp->m == NULL)
+                       bpf_error("out of memory");
                memset((char *)cp->m, 0, size);
                cp->n_left = size;
                if (n > size)
@@ -276,11 +321,13 @@ pcap_compile(pcap_t *p, struct bpf_program *program,
        root = NULL;
        bpf_pcap = p;
        if (setjmp(top_ctx)) {
+               lex_cleanup();
                freechunks();
                return (-1);
        }
 
        netmask = mask;
+
        snaplen = pcap_snapshot(p);
        if (snaplen == 0) {
                snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
@@ -307,6 +354,7 @@ pcap_compile(pcap_t *p, struct bpf_program *program,
        program->bf_insns = icode_to_fcode(root, &len);
        program->bf_len = len;
 
+       lex_cleanup();
        freechunks();
        return (0);
 }
@@ -320,43 +368,29 @@ pcap_compile_nopcap(int snaplen_arg, int linktype_arg,
                    struct bpf_program *program,
             char *buf, int optimize, bpf_u_int32 mask)
 {
-       extern int n_errors;
-       int len;
+       pcap_t *p;
+       int ret;
 
-       n_errors = 0;
-       root = NULL;
-       bpf_pcap = NULL;
-       if (setjmp(top_ctx)) {
-               freechunks();
+       p = pcap_open_dead(linktype_arg, snaplen_arg);
+       if (p == NULL)
                return (-1);
-       }
-
-       netmask = mask;
-
-       /* XXX needed? I don't grok the use of globals here. */
-       snaplen = snaplen_arg;
-
-       lex_init(buf ? buf : "");
-       init_linktype(linktype_arg);
-       (void)pcap_parse();
-
-       if (n_errors)
-               syntax();
-
-       if (root == NULL)
-               root = gen_retblk(snaplen_arg);
+       ret = pcap_compile(p, program, buf, optimize, mask);
+       pcap_close(p);
+       return (ret);
+}
 
-       if (optimize) {
-               bpf_optimize(&root);
-               if (root == NULL ||
-                   (root->s.code == (BPF_RET|BPF_K) && root->s.k == 0))
-                       bpf_error("expression rejects all packets");
+/*
+ * Clean up a "struct bpf_program" by freeing all the memory allocated
+ * in it.
+ */
+void
+pcap_freecode(struct bpf_program *program)
+{
+       program->bf_len = 0;
+       if (program->bf_insns != NULL) {
+               free((char *)program->bf_insns);
+               program->bf_insns = NULL;
        }
-       program->bf_insns = icode_to_fcode(root, &len);
-       program->bf_len = len;
-
-       freechunks();
-       return (0);
 }
 
 /*
@@ -459,6 +493,24 @@ gen_cmp(offset, size, v)
        return b;
 }
 
+static struct block *
+gen_cmp_gt(offset, size, v)
+       u_int offset, size;
+       bpf_int32 v;
+{
+       struct slist *s;
+       struct block *b;
+
+       s = new_stmt(BPF_LD|BPF_ABS|size);
+       s->s.k = offset;
+
+       b = new_block(JMP(BPF_JGT));
+       b->stmts = s;
+       b->s.k = v;
+
+       return b;
+}
+
 static struct block *
 gen_mcmp(offset, size, v, mask)
        u_int offset, size;
@@ -514,13 +566,111 @@ gen_bcmp(offset, size, v)
        return b;
 }
 
+static struct block *
+gen_ncmp(datasize, offset, mask, jtype, jvalue, reverse)
+       bpf_u_int32 datasize, offset, mask, jtype, jvalue;
+       int reverse;
+{
+       struct slist *s;
+       struct block *b;
+       s = new_stmt(BPF_LD|datasize|BPF_ABS);
+       s->s.k = offset;
+       if (mask != 0xffffffff) {
+               s->next = new_stmt(BPF_ALU|BPF_AND|BPF_K);
+               s->next->s.k = mask;
+       }
+       b = new_block(JMP(jtype));
+       b->stmts = s;
+       b->s.k = jvalue;
+       if (reverse && (jtype == BPF_JGT || jtype == BPF_JGE))
+               gen_not(b);
+       return b;
+}
+
 /*
  * Various code constructs need to know the layout of the data link
- * layer.  These variables give the necessary offsets.  off_linktype
- * is set to -1 for no encapsulation, in which case, IP is assumed.
+ * layer.  These variables give the necessary offsets.
+ */
+
+/*
+ * This is the offset of the beginning of the MAC-layer header.
+ * It's usually 0, except for ATM LANE.
+ */
+static u_int off_mac;
+
+/*
+ * "off_linktype" is the offset to information in the link-layer header
+ * giving the packet type.
+ *
+ * For Ethernet, it's the offset of the Ethernet type field.
+ *
+ * For link-layer types that always use 802.2 headers, it's the
+ * offset of the LLC header.
+ *
+ * For PPP, it's the offset of the PPP type field.
+ *
+ * For Cisco HDLC, it's the offset of the CHDLC type field.
+ *
+ * For BSD loopback, it's the offset of the AF_ value.
+ *
+ * For Linux cooked sockets, it's the offset of the type field.
+ *
+ * It's set to -1 for no encapsulation, in which case, IP is assumed.
  */
 static u_int off_linktype;
+
+/*
+ * TRUE if the link layer includes an ATM pseudo-header.
+ */
+static int is_atm = 0;
+
+/*
+ * TRUE if "lane" appeared in the filter; it causes us to generate
+ * code that assumes LANE rather than LLC-encapsulated traffic in SunATM.
+ */
+static int is_lane = 0;
+
+/*
+ * These are offsets for the ATM pseudo-header.
+ */
+static u_int off_vpi;
+static u_int off_vci;
+static u_int off_proto;
+
+/*
+ * This is the offset of the first byte after the ATM pseudo_header,
+ * or -1 if there is no ATM pseudo-header.
+ */
+static u_int off_payload;
+
+/*
+ * These are offsets to the beginning of the network-layer header.
+ *
+ * If the link layer never uses 802.2 LLC:
+ *
+ *     "off_nl" and "off_nl_nosnap" are the same.
+ *
+ * If the link layer always uses 802.2 LLC:
+ *
+ *     "off_nl" is the offset if there's a SNAP header following
+ *     the 802.2 header;
+ *
+ *     "off_nl_nosnap" is the offset if there's no SNAP header.
+ *
+ * If the link layer is Ethernet:
+ *
+ *     "off_nl" is the offset if the packet is an Ethernet II packet
+ *     (we assume no 802.3+802.2+SNAP);
+ *
+ *     "off_nl_nosnap" is the offset if the packet is an 802.3 packet
+ *     with an 802.2 header following it.
+ */
 static u_int off_nl;
+static u_int off_nl_nosnap;
+
 static int linktype;
 
 static void
@@ -529,11 +679,39 @@ init_linktype(type)
 {
        linktype = type;
 
+       /*
+        * Assume it's not raw ATM with a pseudo-header, for now.
+        */
+       off_mac = 0;
+       is_atm = 0;
+       is_lane = 0;
+       off_vpi = -1;
+       off_vci = -1;
+       off_proto = -1;
+       off_payload = -1;
+
+       orig_linktype = -1;
+       orig_nl = -1;
+       orig_nl_nosnap = -1;
+
        switch (type) {
 
+       case DLT_ARCNET:
+               off_linktype = 2;
+               off_nl = 6;             /* XXX in reality, variable! */
+               off_nl_nosnap = 6;      /* no 802.2 LLC */
+               return;
+
+       case DLT_ARCNET_LINUX:
+               off_linktype = 4;
+               off_nl = 8;             /* XXX in reality, variable! */
+               off_nl_nosnap = 8;      /* no 802.2 LLC */
+               return;
+
        case DLT_EN10MB:
                off_linktype = 12;
-               off_nl = 14;
+               off_nl = 14;            /* Ethernet II */
+               off_nl_nosnap = 17;     /* 802.3+802.2 */
                return;
 
        case DLT_SLIP:
@@ -543,6 +721,7 @@ init_linktype(type)
                 */
                off_linktype = -1;
                off_nl = 16;
+               off_nl_nosnap = 16;     /* no 802.2 LLC */
                return;
 
        case DLT_SLIP_BSDOS:
@@ -550,60 +729,232 @@ init_linktype(type)
                off_linktype = -1;
                /* XXX end */
                off_nl = 24;
+               off_nl_nosnap = 24;     /* no 802.2 LLC */
                return;
 
        case DLT_NULL:
+       case DLT_LOOP:
                off_linktype = 0;
                off_nl = 4;
+               off_nl_nosnap = 4;      /* no 802.2 LLC */
+               return;
+
+       case DLT_ENC:
+               off_linktype = 0;
+               off_nl = 12;
+               off_nl_nosnap = 12;     /* no 802.2 LLC */
+               return;
+
+       case DLT_PFLOG:
+               off_linktype = 0;
+               off_nl = 28;
+               off_nl_nosnap = 28;     /* no 802.2 LLC */
                return;
 
        case DLT_PPP:
-       case DLT_CHDLC:
+       case DLT_C_HDLC:                /* BSD/OS Cisco HDLC */
+       case DLT_PPP_SERIAL:            /* NetBSD sync/async serial PPP */
                off_linktype = 2;
                off_nl = 4;
+               off_nl_nosnap = 4;      /* no 802.2 LLC */
+               return;
+
+       case DLT_PPP_ETHER:
+               /*
+                * This does no include the Ethernet header, and
+                * only covers session state.
+                */
+               off_linktype = 6;
+               off_nl = 8;
+               off_nl_nosnap = 8;      /* no 802.2 LLC */
                return;
 
        case DLT_PPP_BSDOS:
                off_linktype = 5;
                off_nl = 24;
+               off_nl_nosnap = 24;     /* no 802.2 LLC */
                return;
 
        case DLT_FDDI:
                /*
                 * FDDI doesn't really have a link-level type field.
-                * We assume that SSAP = SNAP is being used and pick
-                * out the encapsulated Ethernet type.
+                * We set "off_linktype" to the offset of the LLC header.
+                *
+                * To check for Ethernet types, we assume that SSAP = SNAP
+                * is being used and pick out the encapsulated Ethernet type.
+                * XXX - should we generate code to check for SNAP?
                 */
-               off_linktype = 19;
+               off_linktype = 13;
 #ifdef PCAP_FDDIPAD
                off_linktype += pcap_fddipad;
 #endif
-               off_nl = 21;
+               off_nl = 21;            /* FDDI+802.2+SNAP */
+               off_nl_nosnap = 16;     /* FDDI+802.2 */
 #ifdef PCAP_FDDIPAD
                off_nl += pcap_fddipad;
+               off_nl_nosnap += pcap_fddipad;
 #endif
                return;
 
        case DLT_IEEE802:
-               off_linktype = 20;
-               off_nl = 22;
+               /*
+                * Token Ring doesn't really have a link-level type field.
+                * We set "off_linktype" to the offset of the LLC header.
+                *
+                * To check for Ethernet types, we assume that SSAP = SNAP
+                * is being used and pick out the encapsulated Ethernet type.
+                * XXX - should we generate code to check for SNAP?
+                *
+                * XXX - the header is actually variable-length.
+                * Some various Linux patched versions gave 38
+                * as "off_linktype" and 40 as "off_nl"; however,
+                * if a token ring packet has *no* routing
+                * information, i.e. is not source-routed, the correct
+                * values are 20 and 22, as they are in the vanilla code.
+                *
+                * A packet is source-routed iff the uppermost bit
+                * of the first byte of the source address, at an
+                * offset of 8, has the uppermost bit set.  If the
+                * packet is source-routed, the total number of bytes
+                * of routing information is 2 plus bits 0x1F00 of
+                * the 16-bit value at an offset of 14 (shifted right
+                * 8 - figure out which byte that is).
+                */
+               off_linktype = 14;
+               off_nl = 22;            /* Token Ring+802.2+SNAP */
+               off_nl_nosnap = 17;     /* Token Ring+802.2 */
+               return;
+
+       case DLT_IEEE802_11:
+               /*
+                * 802.11 doesn't really have a link-level type field.
+                * We set "off_linktype" to the offset of the LLC header.
+                *
+                * To check for Ethernet types, we assume that SSAP = SNAP
+                * is being used and pick out the encapsulated Ethernet type.
+                * XXX - should we generate code to check for SNAP?
+                *
+                * XXX - the header is actually variable-length.  We
+                * assume a 24-byte link-layer header, as appears in
+                * data frames in networks with no bridges.
+                */
+               off_linktype = 24;
+               off_nl = 32;            /* 802.11+802.2+SNAP */
+               off_nl_nosnap = 27;     /* 802.11+802.2 */
+               return;
+
+       case DLT_PRISM_HEADER:
+               /*
+                * Same as 802.11, but with an additional header before
+                * the 802.11 header, containing a bunch of additional
+                * information including radio-level information.
+                *
+                * The header is 144 bytes long.
+                *
+                * XXX - same variable-length header problem; at least
+                * the Prism header is fixed-length.
+                */
+               off_linktype = 144+24;
+               off_nl = 144+32;        /* Prism+802.11+802.2+SNAP */
+               off_nl_nosnap = 144+27; /* Prism+802.11+802.2 */
+               return;
+
+       case DLT_IEEE802_11_RADIO:
+               /*
+                * Same as 802.11, but with an additional header before
+                * the 802.11 header, containing a bunch of additional
+                * information including radio-level information.
+                *
+                * The header is 64 bytes long.
+                *
+                * XXX - same variable-length header problem, only
+                * more so; this header is also variable-length,
+                * with the length being the 32-bit big-endian
+                * number at an offset of 4 from the beginning
+                * of the radio header.
+                */
+               off_linktype = 64+24;
+               off_nl = 64+32;         /* Radio+802.11+802.2+SNAP */
+               off_nl_nosnap = 64+27;  /* Radio+802.11+802.2 */
                return;
 
        case DLT_ATM_RFC1483:
+       case DLT_ATM_CLIP:      /* Linux ATM defines this */
                /*
                 * assume routed, non-ISO PDUs
                 * (i.e., LLC = 0xAA-AA-03, OUT = 0x00-00-00)
                 */
-               off_linktype = 6;
-               off_nl = 8;
+               off_linktype = 0;
+               off_nl = 8;             /* 802.2+SNAP */
+               off_nl_nosnap = 3;      /* 802.2 */
+               return;
+
+       case DLT_SUNATM:
+               /*
+                * Full Frontal ATM; you get AALn PDUs with an ATM
+                * pseudo-header.
+                */
+               is_atm = 1;
+               off_vpi = SUNATM_VPI_POS;
+               off_vci = SUNATM_VCI_POS;
+               off_proto = PROTO_POS;
+               off_mac = -1;   /* LLC-encapsulated, so no MAC-layer header */  
+               off_payload = SUNATM_PKT_BEGIN_POS;
+               off_linktype = off_payload;
+               off_nl = off_payload+8;         /* 802.2+SNAP */
+               off_nl_nosnap = off_payload+3;  /* 802.2 */
                return;
 
        case DLT_RAW:
                off_linktype = -1;
                off_nl = 0;
+               off_nl_nosnap = 0;      /* no 802.2 LLC */
+               return;
+
+       case DLT_LINUX_SLL:     /* fake header for Linux cooked socket */
+               off_linktype = 14;
+               off_nl = 16;
+               off_nl_nosnap = 16;     /* no 802.2 LLC */
+               return;
+
+       case DLT_LTALK:
+               /*
+                * LocalTalk does have a 1-byte type field in the LLAP header,
+                * but really it just indicates whether there is a "short" or
+                * "long" DDP packet following.
+                */
+               off_linktype = -1;
+               off_nl = 0;
+               off_nl_nosnap = 0;      /* no 802.2 LLC */
+               return;
+
+       case DLT_IP_OVER_FC:
+               /*
+                * RFC 2625 IP-over-Fibre-Channel doesn't really have a
+                * link-level type field.  We set "off_linktype" to the
+                * offset of the LLC header.
+                *
+                * To check for Ethernet types, we assume that SSAP = SNAP
+                * is being used and pick out the encapsulated Ethernet type.
+                * XXX - should we generate code to check for SNAP? RFC
+                * 2625 says SNAP should be used.
+                */
+               off_linktype = 16;
+               off_nl = 24;            /* IPFC+802.2+SNAP */
+               off_nl_nosnap = 19;     /* IPFC+802.2 */
+               return;
+
+       case DLT_FRELAY:
+               /*
+                * XXX - we should set this to handle SNAP-encapsulated
+                * frames (NLPID of 0x80).
+                */
+               off_linktype = -1;
+               off_nl = 0;
+               off_nl_nosnap = 0;      /* no 802.2 LLC */
                return;
        }
-       bpf_error("unknown data link type 0x%x", linktype);
+       bpf_error("unknown data link type %d", linktype);
        /* NOTREACHED */
 }
 
@@ -634,2201 +985,4407 @@ gen_false()
        return gen_uncond(0);
 }
 
+/*
+ * Byte-swap a 32-bit number.
+ * ("htonl()" or "ntohl()" won't work - we want to byte-swap even on
+ * big-endian platforms.)
+ */
+#define        SWAPLONG(y) \
+((((y)&0xff)<<24) | (((y)&0xff00)<<8) | (((y)&0xff0000)>>8) | (((y)>>24)&0xff))
+
 static struct block *
-gen_linktype(proto)
+gen_ether_linktype(proto)
        register int proto;
 {
        struct block *b0, *b1;
 
-       /* If we're not using encapsulation, we're done */
-       if (off_linktype == -1)
-               return gen_true();
+       switch (proto) {
 
-       switch (linktype) {
+       case LLCSAP_ISONS:
+               /*
+                * OSI protocols always use 802.2 encapsulation.
+                * XXX - should we check both the DSAP and the
+                * SSAP, like this, or should we check just the
+                * DSAP?
+                */
+               b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+               gen_not(b0);
+               b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                            ((LLCSAP_ISONS << 8) | LLCSAP_ISONS));
+               gen_and(b0, b1);
+               return b1;
 
-       case DLT_SLIP:
-               return gen_false();
+       case LLCSAP_IP:
+               b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+               gen_not(b0);
+               b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                            ((LLCSAP_IP << 8) | LLCSAP_IP));
+               gen_and(b0, b1);
+               return b1;
 
-       case DLT_PPP:
-               if (proto == ETHERTYPE_IP)
-                       proto = PPP_IP;                 /* XXX was 0x21 */
-#ifdef INET6
-               else if (proto == ETHERTYPE_IPV6)
-                       proto = PPP_IPV6;
-#endif
-               break;
+       case LLCSAP_NETBEUI:
+               /*
+                * NetBEUI always uses 802.2 encapsulation.
+                * XXX - should we check both the DSAP and the
+                * SSAP, like this, or should we check just the
+                * DSAP?
+                */
+               b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+               gen_not(b0);
+               b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                            ((LLCSAP_NETBEUI << 8) | LLCSAP_NETBEUI));
+               gen_and(b0, b1);
+               return b1;
 
-       case DLT_PPP_BSDOS:
-               switch (proto) {
+       case LLCSAP_IPX:
+               /*
+                * Check for;
+                *
+                *      Ethernet_II frames, which are Ethernet
+                *      frames with a frame type of ETHERTYPE_IPX;
+                *
+                *      Ethernet_802.3 frames, which are 802.3
+                *      frames (i.e., the type/length field is
+                *      a length field, <= ETHERMTU, rather than
+                *      a type field) with the first two bytes
+                *      after the Ethernet/802.3 header being
+                *      0xFFFF;
+                *
+                *      Ethernet_802.2 frames, which are 802.3
+                *      frames with an 802.2 LLC header and
+                *      with the IPX LSAP as the DSAP in the LLC
+                *      header;
+                *
+                *      Ethernet_SNAP frames, which are 802.3
+                *      frames with an LLC header and a SNAP
+                *      header and with an OUI of 0x000000
+                *      (encapsulated Ethernet) and a protocol
+                *      ID of ETHERTYPE_IPX in the SNAP header.
+                *
+                * XXX - should we generate the same code both
+                * for tests for LLCSAP_IPX and for ETHERTYPE_IPX?
+                */
 
-               case ETHERTYPE_IP:
-                       b0 = gen_cmp(off_linktype, BPF_H, PPP_IP);
-                       b1 = gen_cmp(off_linktype, BPF_H, PPP_VJC);
-                       gen_or(b0, b1);
-                       b0 = gen_cmp(off_linktype, BPF_H, PPP_VJNC);
-                       gen_or(b1, b0);
-                       return b0;
+               /*
+                * This generates code to check both for the
+                * IPX LSAP (Ethernet_802.2) and for Ethernet_802.3.
+                */
+               b0 = gen_cmp(off_linktype + 2, BPF_B, (bpf_int32)LLCSAP_IPX);
+               b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)0xFFFF);
+               gen_or(b0, b1);
 
-#ifdef INET6
-               case ETHERTYPE_IPV6:
-                       proto = PPP_IPV6;
-                       /* more to go? */
-                       break;
-#endif
+               /*
+                * Now we add code to check for SNAP frames with
+                * ETHERTYPE_IPX, i.e. Ethernet_SNAP.
+                */
+               b0 = gen_snap(0x000000, ETHERTYPE_IPX, 14);
+               gen_or(b0, b1);
 
-               case ETHERTYPE_DN:
-                       proto = PPP_DECNET;
-                       break;
+               /*
+                * Now we generate code to check for 802.3
+                * frames in general.
+                */
+               b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+               gen_not(b0);
 
-               case ETHERTYPE_ATALK:
-                       proto = PPP_APPLE;
-                       break;
+               /*
+                * Now add the check for 802.3 frames before the
+                * check for Ethernet_802.2 and Ethernet_802.3,
+                * as those checks should only be done on 802.3
+                * frames, not on Ethernet frames.
+                */
+               gen_and(b0, b1);
 
-               case ETHERTYPE_NS:
-                       proto = PPP_NS;
-                       break;
-               }
-               break;
+               /*
+                * Now add the check for Ethernet_II frames, and
+                * do that before checking for the other frame
+                * types.
+                */
+               b0 = gen_cmp(off_linktype, BPF_H, (bpf_int32)ETHERTYPE_IPX);
+               gen_or(b0, b1);
+               return b1;
 
-       case DLT_NULL:
-               /* XXX */
-               if (proto == ETHERTYPE_IP)
-                       return (gen_cmp(0, BPF_W, (bpf_int32)htonl(AF_INET)));
-#ifdef INET6
-               else if (proto == ETHERTYPE_IPV6)
-                       return (gen_cmp(0, BPF_W, (bpf_int32)htonl(AF_INET6)));
-#endif
-               else
-                       return gen_false();
-       }
-       return gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
-}
+       case ETHERTYPE_ATALK:
+       case ETHERTYPE_AARP:
+               /*
+                * EtherTalk (AppleTalk protocols on Ethernet link
+                * layer) may use 802.2 encapsulation.
+                */
 
-static struct block *
-gen_hostop(addr, mask, dir, proto, src_off, dst_off)
-       bpf_u_int32 addr;
-       bpf_u_int32 mask;
-       int dir, proto;
-       u_int src_off, dst_off;
-{
-       struct block *b0, *b1;
-       u_int offset;
-
-       switch (dir) {
-
-       case Q_SRC:
-               offset = src_off;
-               break;
-
-       case Q_DST:
-               offset = dst_off;
-               break;
+               /*
+                * Check for 802.2 encapsulation (EtherTalk phase 2?);
+                * we check for an Ethernet type field less than
+                * 1500, which means it's an 802.3 length field.
+                */
+               b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+               gen_not(b0);
 
-       case Q_AND:
-               b0 = gen_hostop(addr, mask, Q_SRC, proto, src_off, dst_off);
-               b1 = gen_hostop(addr, mask, Q_DST, proto, src_off, dst_off);
+               /*
+                * 802.2-encapsulated ETHERTYPE_ATALK packets are
+                * SNAP packets with an organization code of
+                * 0x080007 (Apple, for Appletalk) and a protocol
+                * type of ETHERTYPE_ATALK (Appletalk).
+                *
+                * 802.2-encapsulated ETHERTYPE_AARP packets are
+                * SNAP packets with an organization code of
+                * 0x000000 (encapsulated Ethernet) and a protocol
+                * type of ETHERTYPE_AARP (Appletalk ARP).
+                */
+               if (proto == ETHERTYPE_ATALK)
+                       b1 = gen_snap(0x080007, ETHERTYPE_ATALK, 14);
+               else    /* proto == ETHERTYPE_AARP */
+                       b1 = gen_snap(0x000000, ETHERTYPE_AARP, 14);
                gen_and(b0, b1);
-               return b1;
 
-       case Q_OR:
-       case Q_DEFAULT:
-               b0 = gen_hostop(addr, mask, Q_SRC, proto, src_off, dst_off);
-               b1 = gen_hostop(addr, mask, Q_DST, proto, src_off, dst_off);
+               /*
+                * Check for Ethernet encapsulation (Ethertalk
+                * phase 1?); we just check for the Ethernet
+                * protocol type.
+                */
+               b0 = gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
+
                gen_or(b0, b1);
                return b1;
 
        default:
-               abort();
+               if (proto <= ETHERMTU) {
+                       /*
+                        * This is an LLC SAP value, so the frames
+                        * that match would be 802.2 frames.
+                        * Check that the frame is an 802.2 frame
+                        * (i.e., that the length/type field is
+                        * a length field, <= ETHERMTU) and
+                        * then check the DSAP.
+                        */
+                       b0 = gen_cmp_gt(off_linktype, BPF_H, ETHERMTU);
+                       gen_not(b0);
+                       b1 = gen_cmp(off_linktype + 2, BPF_B, (bpf_int32)proto);
+                       gen_and(b0, b1);
+                       return b1;
+               } else {
+                       /*
+                        * This is an Ethernet type, so compare
+                        * the length/type field with it (if
+                        * the frame is an 802.2 frame, the length
+                        * field will be <= ETHERMTU, and, as
+                        * "proto" is > ETHERMTU, this test
+                        * will fail and the frame won't match,
+                        * which is what we want).
+                        */
+                       return gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
+               }
        }
-       b0 = gen_linktype(proto);
-       b1 = gen_mcmp(offset, BPF_W, (bpf_int32)addr, mask);
-       gen_and(b0, b1);
-       return b1;
 }
 
-#ifdef INET6
 static struct block *
-gen_hostop6(addr, mask, dir, proto, src_off, dst_off)
-       struct in6_addr *addr;
-       struct in6_addr *mask;
-       int dir, proto;
-       u_int src_off, dst_off;
+gen_linktype(proto)
+       register int proto;
 {
-       struct block *b0, *b1;
-       u_int offset;
-       u_int32_t *a, *m;
+       struct block *b0, *b1, *b2;
 
-       switch (dir) {
+       switch (linktype) {
 
-       case Q_SRC:
-               offset = src_off;
+       case DLT_EN10MB:
+               return gen_ether_linktype(proto);
                break;
 
-       case Q_DST:
-               offset = dst_off;
-               break;
+       case DLT_C_HDLC:
+               switch (proto) {
 
-       case Q_AND:
-               b0 = gen_hostop6(addr, mask, Q_SRC, proto, src_off, dst_off);
-               b1 = gen_hostop6(addr, mask, Q_DST, proto, src_off, dst_off);
-               gen_and(b0, b1);
-               return b1;
+               case LLCSAP_ISONS:
+                       proto = (proto << 8 | LLCSAP_ISONS);
+                       /* fall through */
 
-       case Q_OR:
-       case Q_DEFAULT:
-               b0 = gen_hostop6(addr, mask, Q_SRC, proto, src_off, dst_off);
-               b1 = gen_hostop6(addr, mask, Q_DST, proto, src_off, dst_off);
-               gen_or(b0, b1);
-               return b1;
+               default:
+                       return gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
+                       break;
+               }
+               break;
 
-       default:
-               abort();
-       }
-       /* this order is important */
-       a = (u_int32_t *)addr;
-       m = (u_int32_t *)mask;
-       b1 = gen_mcmp(offset + 12, BPF_W, ntohl(a[3]), ntohl(m[3]));
-       b0 = gen_mcmp(offset + 8, BPF_W, ntohl(a[2]), ntohl(m[2]));
-       gen_and(b0, b1);
-       b0 = gen_mcmp(offset + 4, BPF_W, ntohl(a[1]), ntohl(m[1]));
-       gen_and(b0, b1);
-       b0 = gen_mcmp(offset + 0, BPF_W, ntohl(a[0]), ntohl(m[0]));
-       gen_and(b0, b1);
-       b0 = gen_linktype(proto);
-       gen_and(b0, b1);
-       return b1;
-}
-#endif /*INET6*/
+       case DLT_IEEE802_11:
+       case DLT_PRISM_HEADER:
+       case DLT_IEEE802_11_RADIO:
+       case DLT_FDDI:
+       case DLT_IEEE802:
+       case DLT_ATM_RFC1483:
+       case DLT_ATM_CLIP:
+       case DLT_IP_OVER_FC:
+               return gen_llc(proto);
+               break;
 
-static struct block *
-gen_ehostop(eaddr, dir)
-       register const u_char *eaddr;
-       register int dir;
-{
-       register struct block *b0, *b1;
+       case DLT_SUNATM:
+               /*
+                * If "is_lane" is set, check for a LANE-encapsulated
+                * version of this protocol, otherwise check for an
+                * LLC-encapsulated version of this protocol.
+                *
+                * We assume LANE means Ethernet, not Token Ring.
+                */
+               if (is_lane) {
+                       /*
+                        * Check that the packet doesn't begin with an
+                        * LE Control marker.  (We've already generated
+                        * a test for LANE.)
+                        */
+                       b0 = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H, 0xFF00);
+                       gen_not(b0);
 
-       switch (dir) {
-       case Q_SRC:
-               return gen_bcmp(6, 6, eaddr);
+                       /*
+                        * Now generate an Ethernet test.
+                        */
+                       b1 = gen_ether_linktype(proto);
+                       gen_and(b0, b1);
+                       return b1;
+               } else {
+                       /*
+                        * Check for LLC encapsulation and then check the
+                        * protocol.
+                        */
+                       b0 = gen_atmfield_code(A_PROTOTYPE, PT_LLC, BPF_JEQ, 0);
+                       b1 = gen_llc(proto);
+                       gen_and(b0, b1);
+                       return b1;
+               }
 
-       case Q_DST:
-               return gen_bcmp(0, 6, eaddr);
+       case DLT_LINUX_SLL:
+               switch (proto) {
 
-       case Q_AND:
-               b0 = gen_ehostop(eaddr, Q_SRC);
-               b1 = gen_ehostop(eaddr, Q_DST);
-               gen_and(b0, b1);
-               return b1;
+               case LLCSAP_IP:
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_2);
+                       b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                                    ((LLCSAP_IP << 8) | LLCSAP_IP));
+                       gen_and(b0, b1);
+                       return b1;
 
-       case Q_DEFAULT:
-       case Q_OR:
-               b0 = gen_ehostop(eaddr, Q_SRC);
-               b1 = gen_ehostop(eaddr, Q_DST);
-               gen_or(b0, b1);
-               return b1;
-       }
-       abort();
-       /* NOTREACHED */
-}
+               case LLCSAP_ISONS:
+                       /*
+                        * OSI protocols always use 802.2 encapsulation.
+                        * XXX - should we check both the DSAP and the
+                        * LSAP, like this, or should we check just the
+                        * DSAP?
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_2);
+                       b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                                    ((LLCSAP_ISONS << 8) | LLCSAP_ISONS));
+                       gen_and(b0, b1);
+                       return b1;
 
-/*
- * Like gen_ehostop, but for DLT_FDDI
- */
-static struct block *
-gen_fhostop(eaddr, dir)
-       register const u_char *eaddr;
-       register int dir;
-{
-       struct block *b0, *b1;
+               case LLCSAP_NETBEUI:
+                       /*
+                        * NetBEUI always uses 802.2 encapsulation.
+                        * XXX - should we check both the DSAP and the
+                        * LSAP, like this, or should we check just the
+                        * DSAP?
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_2);
+                       b1 = gen_cmp(off_linktype + 2, BPF_H, (bpf_int32)
+                                    ((LLCSAP_NETBEUI << 8) | LLCSAP_NETBEUI));
+                       gen_and(b0, b1);
+                       return b1;
 
-       switch (dir) {
-       case Q_SRC:
-#ifdef PCAP_FDDIPAD
-               return gen_bcmp(6 + 1 + pcap_fddipad, 6, eaddr);
-#else
-               return gen_bcmp(6 + 1, 6, eaddr);
-#endif
+               case LLCSAP_IPX:
+                       /*
+                        *      Ethernet_II frames, which are Ethernet
+                        *      frames with a frame type of ETHERTYPE_IPX;
+                        *
+                        *      Ethernet_802.3 frames, which have a frame
+                        *      type of LINUX_SLL_P_802_3;
+                        *
+                        *      Ethernet_802.2 frames, which are 802.3
+                        *      frames with an 802.2 LLC header (i.e, have
+                        *      a frame type of LINUX_SLL_P_802_2) and
+                        *      with the IPX LSAP as the DSAP in the LLC
+                        *      header;
+                        *
+                        *      Ethernet_SNAP frames, which are 802.3
+                        *      frames with an LLC header and a SNAP
+                        *      header and with an OUI of 0x000000
+                        *      (encapsulated Ethernet) and a protocol
+                        *      ID of ETHERTYPE_IPX in the SNAP header.
+                        *
+                        * First, do the checks on LINUX_SLL_P_802_2
+                        * frames; generate the check for either
+                        * Ethernet_802.2 or Ethernet_SNAP frames, and
+                        * then put a check for LINUX_SLL_P_802_2 frames
+                        * before it.
+                        */
+                       b0 = gen_cmp(off_linktype + 2, BPF_B,
+                           (bpf_int32)LLCSAP_IPX);
+                       b1 = gen_snap(0x000000, ETHERTYPE_IPX,
+                           off_linktype + 2);
+                       gen_or(b0, b1);
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_2);
+                       gen_and(b0, b1);
 
-       case Q_DST:
-#ifdef PCAP_FDDIPAD
-               return gen_bcmp(0 + 1 + pcap_fddipad, 6, eaddr);
-#else
-               return gen_bcmp(0 + 1, 6, eaddr);
-#endif
+                       /*
+                        * Now check for 802.3 frames and OR that with
+                        * the previous test.
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_3);
+                       gen_or(b0, b1);
 
-       case Q_AND:
-               b0 = gen_fhostop(eaddr, Q_SRC);
-               b1 = gen_fhostop(eaddr, Q_DST);
-               gen_and(b0, b1);
-               return b1;
+                       /*
+                        * Now add the check for Ethernet_II frames, and
+                        * do that before checking for the other frame
+                        * types.
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H,
+                           (bpf_int32)ETHERTYPE_IPX);
+                       gen_or(b0, b1);
+                       return b1;
 
-       case Q_DEFAULT:
-       case Q_OR:
-               b0 = gen_fhostop(eaddr, Q_SRC);
-               b1 = gen_fhostop(eaddr, Q_DST);
-               gen_or(b0, b1);
-               return b1;
-       }
-       abort();
-       /* NOTREACHED */
-}
+               case ETHERTYPE_ATALK:
+               case ETHERTYPE_AARP:
+                       /*
+                        * EtherTalk (AppleTalk protocols on Ethernet link
+                        * layer) may use 802.2 encapsulation.
+                        */
 
-/*
- * This is quite tricky because there may be pad bytes in front of the
- * DECNET header, and then there are two possible data packet formats that
- * carry both src and dst addresses, plus 5 packet types in a format that
- * carries only the src node, plus 2 types that use a different format and
- * also carry just the src node.
- *
- * Yuck.
- *
- * Instead of doing those all right, we just look for data packets with
- * 0 or 1 bytes of padding.  If you want to look at other packets, that
- * will require a lot more hacking.
- *
- * To add support for filtering on DECNET "areas" (network numbers)
- * one would want to add a "mask" argument to this routine.  That would
- * make the filter even more inefficient, although one could be clever
- * and not generate masking instructions if the mask is 0xFFFF.
- */
-static struct block *
-gen_dnhostop(addr, dir, base_off)
-       bpf_u_int32 addr;
-       int dir;
-       u_int base_off;
-{
-       struct block *b0, *b1, *b2, *tmp;
-       u_int offset_lh;        /* offset if long header is received */
-       u_int offset_sh;        /* offset if short header is received */
+                       /*
+                        * Check for 802.2 encapsulation (EtherTalk phase 2?);
+                        * we check for the 802.2 protocol type in the
+                        * "Ethernet type" field.
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H, LINUX_SLL_P_802_2);
 
-       switch (dir) {
+                       /*
+                        * 802.2-encapsulated ETHERTYPE_ATALK packets are
+                        * SNAP packets with an organization code of
+                        * 0x080007 (Apple, for Appletalk) and a protocol
+                        * type of ETHERTYPE_ATALK (Appletalk).
+                        *
+                        * 802.2-encapsulated ETHERTYPE_AARP packets are
+                        * SNAP packets with an organization code of
+                        * 0x000000 (encapsulated Ethernet) and a protocol
+                        * type of ETHERTYPE_AARP (Appletalk ARP).
+                        */
+                       if (proto == ETHERTYPE_ATALK)
+                               b1 = gen_snap(0x080007, ETHERTYPE_ATALK,
+                                   off_linktype + 2);
+                       else    /* proto == ETHERTYPE_AARP */
+                               b1 = gen_snap(0x000000, ETHERTYPE_AARP,
+                                   off_linktype + 2);
+                       gen_and(b0, b1);
 
-       case Q_DST:
-               offset_sh = 1;  /* follows flags */
-               offset_lh = 7;  /* flgs,darea,dsubarea,HIORD */
-               break;
+                       /*
+                        * Check for Ethernet encapsulation (Ethertalk
+                        * phase 1?); we just check for the Ethernet
+                        * protocol type.
+                        */
+                       b0 = gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
 
-       case Q_SRC:
-               offset_sh = 3;  /* follows flags, dstnode */
-               offset_lh = 15; /* flgs,darea,dsubarea,did,sarea,ssub,HIORD */
+                       gen_or(b0, b1);
+                       return b1;
+
+               default:
+                       if (proto <= ETHERMTU) {
+                               /*
+                                * This is an LLC SAP value, so the frames
+                                * that match would be 802.2 frames.
+                                * Check for the 802.2 protocol type
+                                * in the "Ethernet type" field, and
+                                * then check the DSAP.
+                                */
+                               b0 = gen_cmp(off_linktype, BPF_H,
+                                   LINUX_SLL_P_802_2);
+                               b1 = gen_cmp(off_linktype + 2, BPF_B,
+                                    (bpf_int32)proto);
+                               gen_and(b0, b1);
+                               return b1;
+                       } else {
+                               /*
+                                * This is an Ethernet type, so compare
+                                * the length/type field with it (if
+                                * the frame is an 802.2 frame, the length
+                                * field will be <= ETHERMTU, and, as
+                                * "proto" is > ETHERMTU, this test
+                                * will fail and the frame won't match,
+                                * which is what we want).
+                                */
+                               return gen_cmp(off_linktype, BPF_H,
+                                   (bpf_int32)proto);
+                       }
+               }
                break;
 
-       case Q_AND:
-               /* Inefficient because we do our Calvinball dance twice */
-               b0 = gen_dnhostop(addr, Q_SRC, base_off);
-               b1 = gen_dnhostop(addr, Q_DST, base_off);
-               gen_and(b0, b1);
-               return b1;
+       case DLT_SLIP:
+       case DLT_SLIP_BSDOS:
+       case DLT_RAW:
+               /*
+                * These types don't provide any type field; packets
+                * are always IP.
+                *
+                * XXX - for IPv4, check for a version number of 4, and,
+                * for IPv6, check for a version number of 6?
+                */
+               switch (proto) {
 
-       case Q_OR:
-       case Q_DEFAULT:
-               /* Inefficient because we do our Calvinball dance twice */
-               b0 = gen_dnhostop(addr, Q_SRC, base_off);
-               b1 = gen_dnhostop(addr, Q_DST, base_off);
-               gen_or(b0, b1);
-               return b1;
+               case ETHERTYPE_IP:
+#ifdef INET6
+               case ETHERTYPE_IPV6:
+#endif
+                       return gen_true();              /* always true */
 
-       default:
-               abort();
-       }
-       b0 = gen_linktype(ETHERTYPE_DN);
-       /* Check for pad = 1, long header case */
-       tmp = gen_mcmp(base_off + 2, BPF_H,
-           (bpf_int32)ntohs(0x0681), (bpf_int32)ntohs(0x07FF));
-       b1 = gen_cmp(base_off + 2 + 1 + offset_lh,
-           BPF_H, (bpf_int32)ntohs(addr));
-       gen_and(tmp, b1);
-       /* Check for pad = 0, long header case */
-       tmp = gen_mcmp(base_off + 2, BPF_B, (bpf_int32)0x06, (bpf_int32)0x7);
-       b2 = gen_cmp(base_off + 2 + offset_lh, BPF_H, (bpf_int32)ntohs(addr));
-       gen_and(tmp, b2);
-       gen_or(b2, b1);
-       /* Check for pad = 1, short header case */
-       tmp = gen_mcmp(base_off + 2, BPF_H,
-           (bpf_int32)ntohs(0x0281), (bpf_int32)ntohs(0x07FF));
-       b2 = gen_cmp(base_off + 2 + 1 + offset_sh,
-           BPF_H, (bpf_int32)ntohs(addr));
-       gen_and(tmp, b2);
-       gen_or(b2, b1);
-       /* Check for pad = 0, short header case */
-       tmp = gen_mcmp(base_off + 2, BPF_B, (bpf_int32)0x02, (bpf_int32)0x7);
-       b2 = gen_cmp(base_off + 2 + offset_sh, BPF_H, (bpf_int32)ntohs(addr));
-       gen_and(tmp, b2);
-       gen_or(b2, b1);
+               default:
+                       return gen_false();             /* always false */
+               }
+               break;
 
-       /* Combine with test for linktype */
-       gen_and(b0, b1);
-       return b1;
-}
+       case DLT_PPP:
+       case DLT_PPP_SERIAL:
+       case DLT_PPP_ETHER:
+               /*
+                * We use Ethernet protocol types inside libpcap;
+                * map them to the corresponding PPP protocol types.
+                */
+               switch (proto) {
 
-static struct block *
-gen_host(addr, mask, proto, dir)
-       bpf_u_int32 addr;
-       bpf_u_int32 mask;
-       int proto;
-       int dir;
-{
-       struct block *b0, *b1;
+               case ETHERTYPE_IP:
+                       proto = PPP_IP;
+                       break;
 
-       switch (proto) {
+#ifdef INET6
+               case ETHERTYPE_IPV6:
+                       proto = PPP_IPV6;
+                       break;
+#endif
 
-       case Q_DEFAULT:
-               b0 = gen_host(addr, mask, Q_IP, dir);
-               if (off_linktype != -1) {
-                   b1 = gen_host(addr, mask, Q_ARP, dir);
-                   gen_or(b0, b1);
-                   b0 = gen_host(addr, mask, Q_RARP, dir);
-                   gen_or(b1, b0);
-               }
-               return b0;
+               case ETHERTYPE_DN:
+                       proto = PPP_DECNET;
+                       break;
 
-       case Q_IP:
-               return gen_hostop(addr, mask, dir, ETHERTYPE_IP,
-                                 off_nl + 12, off_nl + 16);
+               case ETHERTYPE_ATALK:
+                       proto = PPP_APPLE;
+                       break;
 
-       case Q_RARP:
-               return gen_hostop(addr, mask, dir, ETHERTYPE_REVARP,
-                                 off_nl + 14, off_nl + 24);
+               case ETHERTYPE_NS:
+                       proto = PPP_NS;
+                       break;
 
-       case Q_ARP:
-               return gen_hostop(addr, mask, dir, ETHERTYPE_ARP,
-                                 off_nl + 14, off_nl + 24);
+               case LLCSAP_ISONS:
+                       proto = PPP_OSI;
+                       break;
 
-       case Q_TCP:
-               bpf_error("'tcp' modifier applied to host");
+               case LLCSAP_8021D:
+                       /*
+                        * I'm assuming the "Bridging PDU"s that go
+                        * over PPP are Spanning Tree Protocol
+                        * Bridging PDUs.
+                        */
+                       proto = PPP_BRPDU;
+                       break;
 
-       case Q_UDP:
-               bpf_error("'udp' modifier applied to host");
+               case LLCSAP_IPX:
+                       proto = PPP_IPX;
+                       break;
+               }
+               break;
 
-       case Q_ICMP:
-               bpf_error("'icmp' modifier applied to host");
+       case DLT_PPP_BSDOS:
+               /*
+                * We use Ethernet protocol types inside libpcap;
+                * map them to the corresponding PPP protocol types.
+                */
+               switch (proto) {
 
-       case Q_IGMP:
-               bpf_error("'igmp' modifier applied to host");
+               case ETHERTYPE_IP:
+                       b0 = gen_cmp(off_linktype, BPF_H, PPP_IP);
+                       b1 = gen_cmp(off_linktype, BPF_H, PPP_VJC);
+                       gen_or(b0, b1);
+                       b0 = gen_cmp(off_linktype, BPF_H, PPP_VJNC);
+                       gen_or(b1, b0);
+                       return b0;
 
-       case Q_IGRP:
-               bpf_error("'igrp' modifier applied to host");
+#ifdef INET6
+               case ETHERTYPE_IPV6:
+                       proto = PPP_IPV6;
+                       /* more to go? */
+                       break;
+#endif
 
-       case Q_PIM:
-               bpf_error("'pim' modifier applied to host");
+               case ETHERTYPE_DN:
+                       proto = PPP_DECNET;
+                       break;
 
-       case Q_ATALK:
-               bpf_error("ATALK host filtering not implemented");
+               case ETHERTYPE_ATALK:
+                       proto = PPP_APPLE;
+                       break;
 
-       case Q_DECNET:
-               return gen_dnhostop(addr, dir, off_nl);
+               case ETHERTYPE_NS:
+                       proto = PPP_NS;
+                       break;
 
-       case Q_SCA:
-               bpf_error("SCA host filtering not implemented");
+               case LLCSAP_ISONS:
+                       proto = PPP_OSI;
+                       break;
 
-       case Q_LAT:
-               bpf_error("LAT host filtering not implemented");
+               case LLCSAP_8021D:
+                       /*
+                        * I'm assuming the "Bridging PDU"s that go
+                        * over PPP are Spanning Tree Protocol
+                        * Bridging PDUs.
+                        */
+                       proto = PPP_BRPDU;
+                       break;
 
-       case Q_MOPDL:
-               bpf_error("MOPDL host filtering not implemented");
+               case LLCSAP_IPX:
+                       proto = PPP_IPX;
+                       break;
+               }
+               break;
 
-       case Q_MOPRC:
-               bpf_error("MOPRC host filtering not implemented");
+       case DLT_NULL:
+       case DLT_LOOP:
+       case DLT_ENC:
+       case DLT_PFLOG:
+               /*
+                * For DLT_NULL, the link-layer header is a 32-bit
+                * word containing an AF_ value in *host* byte order,
+                * and for DLT_ENC, the link-layer header begins
+                * with a 32-bit work containing an AF_ value in
+                * host byte order.
+                *
+                * In addition, if we're reading a saved capture file,
+                * the host byte order in the capture may not be the
+                * same as the host byte order on this machine.
+                *
+                * For DLT_LOOP, the link-layer header is a 32-bit
+                * word containing an AF_ value in *network* byte order.
+                *
+                * XXX - AF_ values may, unfortunately, be platform-
+                * dependent; for example, FreeBSD's AF_INET6 is 24
+                * whilst NetBSD's and OpenBSD's is 26.
+                *
+                * This means that, when reading a capture file, just
+                * checking for our AF_INET6 value won't work if the
+                * capture file came from another OS.
+                *
+                * XXX - what's the byte order for DLT_PFLOG?
+                */
+               switch (proto) {
+
+               case ETHERTYPE_IP:
+                       proto = AF_INET;
+                       break;
 
 #ifdef INET6
-       case Q_IPV6:
-               bpf_error("'ip6' modifier applied to ip host");
+               case ETHERTYPE_IPV6:
+                       proto = AF_INET6;
+                       break;
+#endif
 
-       case Q_ICMPV6:
-               bpf_error("'icmp6' modifier applied to host");
-#endif /* INET6 */
+               default:
+                       /*
+                        * Not a type on which we support filtering.
+                        * XXX - support those that have AF_ values
+                        * #defined on this platform, at least?
+                        */
+                       return gen_false();
+               }
 
-       case Q_AH:
-               bpf_error("'ah' modifier applied to host");
+               if (linktype == DLT_NULL || linktype == DLT_ENC) {
+                       /*
+                        * The AF_ value is in host byte order, but
+                        * the BPF interpreter will convert it to
+                        * network byte order.
+                        *
+                        * If this is a save file, and it's from a
+                        * machine with the opposite byte order to
+                        * ours, we byte-swap the AF_ value.
+                        *
+                        * Then we run it through "htonl()", and
+                        * generate code to compare against the result.
+                        */
+                       if (bpf_pcap->sf.rfile != NULL &&
+                           bpf_pcap->sf.swapped)
+                               proto = SWAPLONG(proto);
+                       proto = htonl(proto);
+               }
+               return (gen_cmp(0, BPF_W, (bpf_int32)proto));
 
-       case Q_ESP:
-               bpf_error("'esp' modifier applied to host");
+       case DLT_ARCNET:
+       case DLT_ARCNET_LINUX:
+               /*
+                * XXX should we check for first fragment if the protocol
+                * uses PHDS?
+                */
+               switch (proto) {
 
-       default:
-               abort();
-       }
-       /* NOTREACHED */
-}
+               default:
+                       return gen_false();
 
 #ifdef INET6
-static struct block *
-gen_host6(addr, mask, proto, dir)
-       struct in6_addr *addr;
-       struct in6_addr *mask;
-       int proto;
-       int dir;
-{
-       switch (proto) {
-
-       case Q_DEFAULT:
-               return gen_host6(addr, mask, Q_IPV6, dir);
+               case ETHERTYPE_IPV6:
+                       return (gen_cmp(off_linktype, BPF_B,
+                               (bpf_int32)ARCTYPE_INET6));
+#endif /* INET6 */
 
-       case Q_IP:
-               bpf_error("'ip' modifier applied to ip6 host");
+               case ETHERTYPE_IP:
+                       b0 = gen_cmp(off_linktype, BPF_B, 
+                                    (bpf_int32)ARCTYPE_IP);
+                       b1 = gen_cmp(off_linktype, BPF_B,
+                                    (bpf_int32)ARCTYPE_IP_OLD);
+                       gen_or(b0, b1);
+                       return (b1);
 
-       case Q_RARP:
-               bpf_error("'rarp' modifier applied to ip6 host");
+               case ETHERTYPE_ARP:
+                       b0 = gen_cmp(off_linktype, BPF_B,
+                                    (bpf_int32)ARCTYPE_ARP);
+                       b1 = gen_cmp(off_linktype, BPF_B, 
+                                    (bpf_int32)ARCTYPE_ARP_OLD);
+                       gen_or(b0, b1);
+                       return (b1);
 
-       case Q_ARP:
-               bpf_error("'arp' modifier applied to ip6 host");
+               case ETHERTYPE_REVARP:
+                       return (gen_cmp(off_linktype, BPF_B,
+                                       (bpf_int32)ARCTYPE_REVARP));
 
-       case Q_TCP:
-               bpf_error("'tcp' modifier applied to host");
+               case ETHERTYPE_ATALK:
+                       return (gen_cmp(off_linktype, BPF_B,
+                                       (bpf_int32)ARCTYPE_ATALK));
+               }
+               break;
 
-       case Q_UDP:
-               bpf_error("'udp' modifier applied to host");
+       case DLT_LTALK:
+               switch (proto) {
+               case ETHERTYPE_ATALK:
+                       return gen_true();
+               default:
+                       return gen_false();
+               }
+               break;
 
-       case Q_ICMP:
-               bpf_error("'icmp' modifier applied to host");
+       case DLT_FRELAY:
+               /*
+                * XXX - assumes a 2-byte Frame Relay header with
+                * DLCI and flags.  What if the address is longer?
+                */
+               switch (proto) {
 
-       case Q_IGMP:
-               bpf_error("'igmp' modifier applied to host");
+               case ETHERTYPE_IP:
+                       /*
+                        * Check for the special NLPID for IP.
+                        */
+                       return gen_cmp(2, BPF_H, (0x03<<8) | 0xcc);
 
-       case Q_IGRP:
-               bpf_error("'igrp' modifier applied to host");
+#ifdef INET6
+               case ETHERTYPE_IPV6:
+                       /*
+                        * Check for the special NLPID for IPv6.
+                        */
+                       return gen_cmp(2, BPF_H, (0x03<<8) | 0x8e);
+#endif
 
-       case Q_PIM:
-               bpf_error("'pim' modifier applied to host");
+               case LLCSAP_ISONS:
+                       /*
+                        * Check for several OSI protocols.
+                        *
+                        * Frame Relay packets typically have an OSI
+                        * NLPID at the beginning; we check for each
+                        * of them.
+                        *
+                        * What we check for is the NLPID and a frame
+                        * control field of UI, i.e. 0x03 followed
+                        * by the NLPID.
+                        */
+                       b0 = gen_cmp(2, BPF_H, (0x03<<8) | ISO8473_CLNP);
+                       b1 = gen_cmp(2, BPF_H, (0x03<<8) | ISO9542_ESIS);
+                       b2 = gen_cmp(2, BPF_H, (0x03<<8) | ISO10589_ISIS);
+                       gen_or(b1, b2);
+                       gen_or(b0, b2);
+                       return b2;
+
+               default:
+                       return gen_false();
+               }
+               break;
+       }
 
-       case Q_ATALK:
-               bpf_error("ATALK host filtering not implemented");
+       /*
+        * All the types that have no encapsulation should either be
+        * handled as DLT_SLIP, DLT_SLIP_BSDOS, and DLT_RAW are, if
+        * all packets are IP packets, or should be handled in some
+        * special case, if none of them are (if some are and some
+        * aren't, the lack of encapsulation is a problem, as we'd
+        * have to find some other way of determining the packet type).
+        *
+        * Therefore, if "off_linktype" is -1, there's an error.
+        */
+       if (off_linktype == -1)
+               abort();
 
-       case Q_DECNET:
-               bpf_error("'decnet' modifier applied to ip6 host");
+       /*
+        * Any type not handled above should always have an Ethernet
+        * type at an offset of "off_linktype".  (PPP is partially
+        * handled above - the protocol type is mapped from the
+        * Ethernet and LLC types we use internally to the corresponding
+        * PPP type - but the PPP type is always specified by a value
+        * at "off_linktype", so we don't have to do the code generation
+        * above.)
+        */
+       return gen_cmp(off_linktype, BPF_H, (bpf_int32)proto);
+}
 
-       case Q_SCA:
-               bpf_error("SCA host filtering not implemented");
+/*
+ * Check for an LLC SNAP packet with a given organization code and
+ * protocol type; we check the entire contents of the 802.2 LLC and
+ * snap headers, checking for DSAP and SSAP of SNAP and a control
+ * field of 0x03 in the LLC header, and for the specified organization
+ * code and protocol type in the SNAP header.
+ */
+static struct block *
+gen_snap(orgcode, ptype, offset)
+       bpf_u_int32 orgcode;
+       bpf_u_int32 ptype;
+       u_int offset;
+{
+       u_char snapblock[8];
+
+       snapblock[0] = LLCSAP_SNAP;     /* DSAP = SNAP */
+       snapblock[1] = LLCSAP_SNAP;     /* SSAP = SNAP */
+       snapblock[2] = 0x03;            /* control = UI */
+       snapblock[3] = (orgcode >> 16); /* upper 8 bits of organization code */
+       snapblock[4] = (orgcode >> 8);  /* middle 8 bits of organization code */
+       snapblock[5] = (orgcode >> 0);  /* lower 8 bits of organization code */
+       snapblock[6] = (ptype >> 8);    /* upper 8 bits of protocol type */
+       snapblock[7] = (ptype >> 0);    /* lower 8 bits of protocol type */
+       return gen_bcmp(offset, 8, snapblock);
+}
 
-       case Q_LAT:
-               bpf_error("LAT host filtering not implemented");
-
-       case Q_MOPDL:
-               bpf_error("MOPDL host filtering not implemented");
+/*
+ * Check for a given protocol value assuming an 802.2 LLC header.
+ */
+static struct block *
+gen_llc(proto)
+       int proto;
+{
+       /*
+        * XXX - handle token-ring variable-length header.
+        */
+       switch (proto) {
 
-       case Q_MOPRC:
-               bpf_error("MOPRC host filtering not implemented");
+       case LLCSAP_IP:
+               return gen_cmp(off_linktype, BPF_H, (long)
+                            ((LLCSAP_IP << 8) | LLCSAP_IP));
 
-       case Q_IPV6:
-               return gen_hostop6(addr, mask, dir, ETHERTYPE_IPV6,
-                                 off_nl + 8, off_nl + 24);
+       case LLCSAP_ISONS:
+               return gen_cmp(off_linktype, BPF_H, (long)
+                            ((LLCSAP_ISONS << 8) | LLCSAP_ISONS));
 
-       case Q_ICMPV6:
-               bpf_error("'icmp6' modifier applied to host");
+       case LLCSAP_NETBEUI:
+               return gen_cmp(off_linktype, BPF_H, (long)
+                            ((LLCSAP_NETBEUI << 8) | LLCSAP_NETBEUI));
 
-       case Q_AH:
-               bpf_error("'ah' modifier applied to host");
+       case LLCSAP_IPX:
+               /*
+                * XXX - are there ever SNAP frames for IPX on
+                * non-Ethernet 802.x networks?
+                */
+               return gen_cmp(off_linktype, BPF_B, (bpf_int32)LLCSAP_IPX);
 
-       case Q_ESP:
-               bpf_error("'esp' modifier applied to host");
+       case ETHERTYPE_ATALK:
+               /*
+                * 802.2-encapsulated ETHERTYPE_ATALK packets are
+                * SNAP packets with an organization code of
+                * 0x080007 (Apple, for Appletalk) and a protocol
+                * type of ETHERTYPE_ATALK (Appletalk).
+                *
+                * XXX - check for an organization code of
+                * encapsulated Ethernet as well?
+                */
+               return gen_snap(0x080007, ETHERTYPE_ATALK, off_linktype);
 
        default:
-               abort();
+               /*
+                * XXX - we don't have to check for IPX 802.3
+                * here, but should we check for the IPX Ethertype?
+                */
+               if (proto <= ETHERMTU) {
+                       /*
+                        * This is an LLC SAP value, so check
+                        * the DSAP.
+                        */
+                       return gen_cmp(off_linktype, BPF_B, (bpf_int32)proto);
+               } else {
+                       /*
+                        * This is an Ethernet type; we assume that it's
+                        * unlikely that it'll appear in the right place
+                        * at random, and therefore check only the
+                        * location that would hold the Ethernet type
+                        * in a SNAP frame with an organization code of
+                        * 0x000000 (encapsulated Ethernet).
+                        *
+                        * XXX - if we were to check for the SNAP DSAP and
+                        * LSAP, as per XXX, and were also to check for an
+                        * organization code of 0x000000 (encapsulated
+                        * Ethernet), we'd do
+                        *
+                        *      return gen_snap(0x000000, proto,
+                        *          off_linktype);
+                        *
+                        * here; for now, we don't, as per the above.
+                        * I don't know whether it's worth the extra CPU
+                        * time to do the right check or not.
+                        */
+                       return gen_cmp(off_linktype+6, BPF_H, (bpf_int32)proto);
+               }
        }
-       /* NOTREACHED */
 }
-#endif /*INET6*/
 
-#ifndef INET6
 static struct block *
-gen_gateway(eaddr, alist, proto, dir)
-       const u_char *eaddr;
-       bpf_u_int32 **alist;
-       int proto;
-       int dir;
+gen_hostop(addr, mask, dir, proto, src_off, dst_off)
+       bpf_u_int32 addr;
+       bpf_u_int32 mask;
+       int dir, proto;
+       u_int src_off, dst_off;
 {
-       struct block *b0, *b1, *tmp;
+       struct block *b0, *b1;
+       u_int offset;
 
-       if (dir != 0)
-               bpf_error("direction applied to 'gateway'");
+       switch (dir) {
 
-       switch (proto) {
-       case Q_DEFAULT:
-       case Q_IP:
-       case Q_ARP:
-       case Q_RARP:
-               if (linktype == DLT_EN10MB)
-                       b0 = gen_ehostop(eaddr, Q_OR);
-               else if (linktype == DLT_FDDI)
-                       b0 = gen_fhostop(eaddr, Q_OR);
-               else
-                       bpf_error(
-                           "'gateway' supported only on ethernet or FDDI");
+       case Q_SRC:
+               offset = src_off;
+               break;
 
-               b1 = gen_host(**alist++, 0xffffffff, proto, Q_OR);
-               while (*alist) {
-                       tmp = gen_host(**alist++, 0xffffffff, proto, Q_OR);
-                       gen_or(b1, tmp);
-                       b1 = tmp;
-               }
-               gen_not(b1);
+       case Q_DST:
+               offset = dst_off;
+               break;
+
+       case Q_AND:
+               b0 = gen_hostop(addr, mask, Q_SRC, proto, src_off, dst_off);
+               b1 = gen_hostop(addr, mask, Q_DST, proto, src_off, dst_off);
                gen_and(b0, b1);
                return b1;
-       }
-       bpf_error("illegal modifier of 'gateway'");
-       /* NOTREACHED */
-}
-#endif
-
-struct block *
-gen_proto_abbrev(proto)
-       int proto;
-{
-#ifdef INET6
-       struct block *b0;
-#endif
-       struct block *b1;
-
-       switch (proto) {
 
-       case Q_TCP:
-               b1 = gen_proto(IPPROTO_TCP, Q_IP, Q_DEFAULT);
-#ifdef INET6
-               b0 = gen_proto(IPPROTO_TCP, Q_IPV6, Q_DEFAULT);
+       case Q_OR:
+       case Q_DEFAULT:
+               b0 = gen_hostop(addr, mask, Q_SRC, proto, src_off, dst_off);
+               b1 = gen_hostop(addr, mask, Q_DST, proto, src_off, dst_off);
                gen_or(b0, b1);
-#endif
-               break;
+               return b1;
 
-       case Q_UDP:
-               b1 = gen_proto(IPPROTO_UDP, Q_IP, Q_DEFAULT);
-#ifdef INET6
-               b0 = gen_proto(IPPROTO_UDP, Q_IPV6, Q_DEFAULT);
-               gen_or(b0, b1);
-#endif
-               break;
+       default:
+               abort();
+       }
+       b0 = gen_linktype(proto);
+       b1 = gen_mcmp(offset, BPF_W, (bpf_int32)addr, mask);
+       gen_and(b0, b1);
+       return b1;
+}
 
-       case Q_ICMP:
-               b1 = gen_proto(IPPROTO_ICMP, Q_IP, Q_DEFAULT);
-               break;
+#ifdef INET6
+static struct block *
+gen_hostop6(addr, mask, dir, proto, src_off, dst_off)
+       struct in6_addr *addr;
+       struct in6_addr *mask;
+       int dir, proto;
+       u_int src_off, dst_off;
+{
+       struct block *b0, *b1;
+       u_int offset;
+       u_int32_t *a, *m;
 
-#ifndef        IPPROTO_IGMP
-#define        IPPROTO_IGMP    2
-#endif
+       switch (dir) {
 
-       case Q_IGMP:
-               b1 = gen_proto(IPPROTO_IGMP, Q_IP, Q_DEFAULT);
+       case Q_SRC:
+               offset = src_off;
                break;
 
-#ifndef        IPPROTO_IGRP
-#define        IPPROTO_IGRP    9
-#endif
-       case Q_IGRP:
-               b1 = gen_proto(IPPROTO_IGRP, Q_IP, Q_DEFAULT);
+       case Q_DST:
+               offset = dst_off;
                break;
 
-#ifndef IPPROTO_PIM
-#define IPPROTO_PIM    103
-#endif
+       case Q_AND:
+               b0 = gen_hostop6(addr, mask, Q_SRC, proto, src_off, dst_off);
+               b1 = gen_hostop6(addr, mask, Q_DST, proto, src_off, dst_off);
+               gen_and(b0, b1);
+               return b1;
 
-       case Q_PIM:
-               b1 = gen_proto(IPPROTO_PIM, Q_IP, Q_DEFAULT);
-#ifdef INET6
-               b0 = gen_proto(IPPROTO_PIM, Q_IPV6, Q_DEFAULT);
+       case Q_OR:
+       case Q_DEFAULT:
+               b0 = gen_hostop6(addr, mask, Q_SRC, proto, src_off, dst_off);
+               b1 = gen_hostop6(addr, mask, Q_DST, proto, src_off, dst_off);
                gen_or(b0, b1);
-#endif
-               break;
+               return b1;
 
-       case Q_IP:
-               b1 =  gen_linktype(ETHERTYPE_IP);
-               break;
+       default:
+               abort();
+       }
+       /* this order is important */
+       a = (u_int32_t *)addr;
+       m = (u_int32_t *)mask;
+       b1 = gen_mcmp(offset + 12, BPF_W, ntohl(a[3]), ntohl(m[3]));
+       b0 = gen_mcmp(offset + 8, BPF_W, ntohl(a[2]), ntohl(m[2]));
+       gen_and(b0, b1);
+       b0 = gen_mcmp(offset + 4, BPF_W, ntohl(a[1]), ntohl(m[1]));
+       gen_and(b0, b1);
+       b0 = gen_mcmp(offset + 0, BPF_W, ntohl(a[0]), ntohl(m[0]));
+       gen_and(b0, b1);
+       b0 = gen_linktype(proto);
+       gen_and(b0, b1);
+       return b1;
+}
+#endif /*INET6*/
 
-       case Q_ARP:
-               b1 =  gen_linktype(ETHERTYPE_ARP);
-               break;
+static struct block *
+gen_ehostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
+{
+       register struct block *b0, *b1;
 
-       case Q_RARP:
-               b1 =  gen_linktype(ETHERTYPE_REVARP);
-               break;
+       switch (dir) {
+       case Q_SRC:
+               return gen_bcmp(off_mac + 6, 6, eaddr);
 
-       case Q_LINK:
-               bpf_error("link layer applied in wrong context");
+       case Q_DST:
+               return gen_bcmp(off_mac + 0, 6, eaddr);
 
-       case Q_ATALK:
-               b1 =  gen_linktype(ETHERTYPE_ATALK);
-               break;
+       case Q_AND:
+               b0 = gen_ehostop(eaddr, Q_SRC);
+               b1 = gen_ehostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
 
-       case Q_DECNET:
-               b1 =  gen_linktype(ETHERTYPE_DN);
-               break;
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_ehostop(eaddr, Q_SRC);
+               b1 = gen_ehostop(eaddr, Q_DST);
+               gen_or(b0, b1);
+               return b1;
+       }
+       abort();
+       /* NOTREACHED */
+}
 
-       case Q_SCA:
-               b1 =  gen_linktype(ETHERTYPE_SCA);
-               break;
+/*
+ * Like gen_ehostop, but for DLT_FDDI
+ */
+static struct block *
+gen_fhostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
+{
+       struct block *b0, *b1;
 
-       case Q_LAT:
-               b1 =  gen_linktype(ETHERTYPE_LAT);
-               break;
+       switch (dir) {
+       case Q_SRC:
+#ifdef PCAP_FDDIPAD
+               return gen_bcmp(6 + 1 + pcap_fddipad, 6, eaddr);
+#else
+               return gen_bcmp(6 + 1, 6, eaddr);
+#endif
 
-       case Q_MOPDL:
-               b1 =  gen_linktype(ETHERTYPE_MOPDL);
-               break;
-
-       case Q_MOPRC:
-               b1 =  gen_linktype(ETHERTYPE_MOPRC);
-               break;
-
-#ifdef INET6
-       case Q_IPV6:
-               b1 = gen_linktype(ETHERTYPE_IPV6);
-               break;
-
-#ifndef IPPROTO_ICMPV6
-#define IPPROTO_ICMPV6 58
+       case Q_DST:
+#ifdef PCAP_FDDIPAD
+               return gen_bcmp(0 + 1 + pcap_fddipad, 6, eaddr);
+#else
+               return gen_bcmp(0 + 1, 6, eaddr);
 #endif
-       case Q_ICMPV6:
-               b1 = gen_proto(IPPROTO_ICMPV6, Q_IPV6, Q_DEFAULT);
-               break;
-#endif /* INET6 */
 
-#ifndef IPPROTO_AH
-#define IPPROTO_AH     51
-#endif
-       case Q_AH:
-               b1 = gen_proto(IPPROTO_AH, Q_IP, Q_DEFAULT);
-#ifdef INET6
-               b0 = gen_proto(IPPROTO_AH, Q_IPV6, Q_DEFAULT);
-               gen_or(b0, b1);
-#endif
-               break;
+       case Q_AND:
+               b0 = gen_fhostop(eaddr, Q_SRC);
+               b1 = gen_fhostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
 
-#ifndef IPPROTO_ESP
-#define IPPROTO_ESP    50
-#endif
-       case Q_ESP:
-               b1 = gen_proto(IPPROTO_ESP, Q_IP, Q_DEFAULT);
-#ifdef INET6
-               b0 = gen_proto(IPPROTO_ESP, Q_IPV6, Q_DEFAULT);
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_fhostop(eaddr, Q_SRC);
+               b1 = gen_fhostop(eaddr, Q_DST);
                gen_or(b0, b1);
-#endif
-               break;
-
-       default:
-               abort();
+               return b1;
        }
-       return b1;
-}
-
-static struct block *
-gen_ipfrag()
-{
-       struct slist *s;
-       struct block *b;
-
-       /* not ip frag */
-       s = new_stmt(BPF_LD|BPF_H|BPF_ABS);
-       s->s.k = off_nl + 6;
-       b = new_block(JMP(BPF_JSET));
-       b->s.k = 0x1fff;
-       b->stmts = s;
-       gen_not(b);
-
-       return b;
+       abort();
+       /* NOTREACHED */
 }
 
+/*
+ * Like gen_ehostop, but for DLT_IEEE802 (Token Ring)
+ */
 static struct block *
-gen_portatom(off, v)
-       int off;
-       bpf_int32 v;
+gen_thostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
 {
-       struct slist *s;
-       struct block *b;
+       register struct block *b0, *b1;
 
-       s = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
-       s->s.k = off_nl;
+       switch (dir) {
+       case Q_SRC:
+               return gen_bcmp(8, 6, eaddr);
 
-       s->next = new_stmt(BPF_LD|BPF_IND|BPF_H);
-       s->next->s.k = off_nl + off;
+       case Q_DST:
+               return gen_bcmp(2, 6, eaddr);
 
-       b = new_block(JMP(BPF_JEQ));
-       b->stmts = s;
-       b->s.k = v;
+       case Q_AND:
+               b0 = gen_thostop(eaddr, Q_SRC);
+               b1 = gen_thostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
 
-       return b;
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_thostop(eaddr, Q_SRC);
+               b1 = gen_thostop(eaddr, Q_DST);
+               gen_or(b0, b1);
+               return b1;
+       }
+       abort();
+       /* NOTREACHED */
 }
 
-#ifdef INET6
+/*
+ * Like gen_ehostop, but for DLT_IEEE802_11 (802.11 wireless LAN)
+ */
 static struct block *
-gen_portatom6(off, v)
-       int off;
-       bpf_int32 v;
-{
-       return gen_cmp(off_nl + 40 + off, BPF_H, v);
-}
-#endif/*INET6*/
-
-struct block *
-gen_portop(port, proto, dir)
-       int port, proto, dir;
+gen_wlanhostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
 {
-       struct block *b0, *b1, *tmp;
-
-       /* ip proto 'proto' */
-       tmp = gen_cmp(off_nl + 9, BPF_B, (bpf_int32)proto);
-       b0 = gen_ipfrag();
-       gen_and(tmp, b0);
+       register struct block *b0, *b1, *b2;
+       register struct slist *s;
 
        switch (dir) {
        case Q_SRC:
-               b1 = gen_portatom(0, (bpf_int32)port);
-               break;
+               /*
+                * Oh, yuk.
+                *
+                *      For control frames, there is no SA.
+                *
+                *      For management frames, SA is at an
+                *      offset of 10 from the beginning of
+                *      the packet.
+                *
+                *      For data frames, SA is at an offset
+                *      of 10 from the beginning of the packet
+                *      if From DS is clear, at an offset of
+                *      16 from the beginning of the packet
+                *      if From DS is set and To DS is clear,
+                *      and an offset of 24 from the beginning
+                *      of the packet if From DS is set and To DS
+                *      is set.
+                */
 
-       case Q_DST:
-               b1 = gen_portatom(2, (bpf_int32)port);
-               break;
+               /*
+                * Generate the tests to be done for data frames
+                * with From DS set.
+                *
+                * First, check for To DS set, i.e. check "link[1] & 0x01".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x01; /* To DS */
+               b1->stmts = s;
 
-       case Q_OR:
-       case Q_DEFAULT:
-               tmp = gen_portatom(0, (bpf_int32)port);
-               b1 = gen_portatom(2, (bpf_int32)port);
-               gen_or(tmp, b1);
-               break;
+               /*
+                * If To DS is set, the SA is at 24.
+                */
+               b0 = gen_bcmp(24, 6, eaddr);
+               gen_and(b1, b0);
 
-       case Q_AND:
-               tmp = gen_portatom(0, (bpf_int32)port);
-               b1 = gen_portatom(2, (bpf_int32)port);
-               gen_and(tmp, b1);
-               break;
+               /*
+                * Now, check for To DS not set, i.e. check
+                * "!(link[1] & 0x01)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b2 = new_block(JMP(BPF_JSET));
+               b2->s.k = 0x01; /* To DS */
+               b2->stmts = s;
+               gen_not(b2);
 
-       default:
-               abort();
-       }
-       gen_and(b0, b1);
+               /*
+                * If To DS is not set, the SA is at 16.
+                */
+               b1 = gen_bcmp(16, 6, eaddr);
+               gen_and(b2, b1);
 
-       return b1;
-}
+               /*
+                * Now OR together the last two checks.  That gives
+                * the complete set of checks for data frames with
+                * From DS set.
+                */
+               gen_or(b1, b0);
 
-static struct block *
-gen_port(port, ip_proto, dir)
-       int port;
-       int ip_proto;
-       int dir;
-{
-       struct block *b0, *b1, *tmp;
+               /*
+                * Now check for From DS being set, and AND that with
+                * the ORed-together checks.
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x02; /* From DS */
+               b1->stmts = s;
+               gen_and(b1, b0);
 
-       /* ether proto ip */
-       b0 =  gen_linktype(ETHERTYPE_IP);
+               /*
+                * Now check for data frames with From DS not set.
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b2 = new_block(JMP(BPF_JSET));
+               b2->s.k = 0x02; /* From DS */
+               b2->stmts = s;
+               gen_not(b2);
 
-       switch (ip_proto) {
-       case IPPROTO_UDP:
-       case IPPROTO_TCP:
-               b1 = gen_portop(port, ip_proto, dir);
-               break;
+               /*
+                * If From DS isn't set, the SA is at 10.
+                */
+               b1 = gen_bcmp(10, 6, eaddr);
+               gen_and(b2, b1);
 
-       case PROTO_UNDEF:
-               tmp = gen_portop(port, IPPROTO_TCP, dir);
-               b1 = gen_portop(port, IPPROTO_UDP, dir);
-               gen_or(tmp, b1);
-               break;
+               /*
+                * Now OR together the checks for data frames with
+                * From DS not set and for data frames with From DS
+                * set; that gives the checks done for data frames.
+                */
+               gen_or(b1, b0);
 
-       default:
-               abort();
-       }
-       gen_and(b0, b1);
-       return b1;
-}
+               /*
+                * Now check for a data frame.
+                * I.e, check "link[0] & 0x08".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x08;
+               b1->stmts = s;
 
-#ifdef INET6
-struct block *
-gen_portop6(port, proto, dir)
-       int port, proto, dir;
-{
-       struct block *b0, *b1, *tmp;
+               /*
+                * AND that with the checks done for data frames.
+                */
+               gen_and(b1, b0);
 
-       /* ip proto 'proto' */
-       b0 = gen_cmp(off_nl + 6, BPF_B, (bpf_int32)proto);
+               /*
+                * If the high-order bit of the type value is 0, this
+                * is a management frame.
+                * I.e, check "!(link[0] & 0x08)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b2 = new_block(JMP(BPF_JSET));
+               b2->s.k = 0x08;
+               b2->stmts = s;
+               gen_not(b2);
 
-       switch (dir) {
-       case Q_SRC:
-               b1 = gen_portatom6(0, (bpf_int32)port);
-               break;
+               /*
+                * For management frames, the SA is at 10.
+                */
+               b1 = gen_bcmp(10, 6, eaddr);
+               gen_and(b2, b1);
+
+               /*
+                * OR that with the checks done for data frames.
+                * That gives the checks done for management and
+                * data frames.
+                */
+               gen_or(b1, b0);
+
+               /*
+                * If the low-order bit of the type value is 1,
+                * this is either a control frame or a frame
+                * with a reserved type, and thus not a
+                * frame with an SA.
+                *
+                * I.e., check "!(link[0] & 0x04)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x04;
+               b1->stmts = s;
+               gen_not(b1);
+
+               /*
+                * AND that with the checks for data and management
+                * frames.
+                */
+               gen_and(b1, b0);
+               return b0;
 
        case Q_DST:
-               b1 = gen_portatom6(2, (bpf_int32)port);
-               break;
+               /*
+                * Oh, yuk.
+                *
+                *      For control frames, there is no DA.
+                *
+                *      For management frames, DA is at an
+                *      offset of 4 from the beginning of
+                *      the packet.
+                *
+                *      For data frames, DA is at an offset
+                *      of 4 from the beginning of the packet
+                *      if To DS is clear and at an offset of
+                *      16 from the beginning of the packet
+                *      if To DS is set.
+                */
 
-       case Q_OR:
-       case Q_DEFAULT:
-               tmp = gen_portatom6(0, (bpf_int32)port);
-               b1 = gen_portatom6(2, (bpf_int32)port);
-               gen_or(tmp, b1);
-               break;
+               /*
+                * Generate the tests to be done for data frames.
+                *
+                * First, check for To DS set, i.e. "link[1] & 0x01".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x01; /* To DS */
+               b1->stmts = s;
 
-       case Q_AND:
-               tmp = gen_portatom6(0, (bpf_int32)port);
-               b1 = gen_portatom6(2, (bpf_int32)port);
-               gen_and(tmp, b1);
-               break;
+               /*
+                * If To DS is set, the DA is at 16.
+                */
+               b0 = gen_bcmp(16, 6, eaddr);
+               gen_and(b1, b0);
 
-       default:
-               abort();
-       }
-       gen_and(b0, b1);
+               /*
+                * Now, check for To DS not set, i.e. check
+                * "!(link[1] & 0x01)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 1;
+               b2 = new_block(JMP(BPF_JSET));
+               b2->s.k = 0x01; /* To DS */
+               b2->stmts = s;
+               gen_not(b2);
 
-       return b1;
-}
+               /*
+                * If To DS is not set, the DA is at 4.
+                */
+               b1 = gen_bcmp(4, 6, eaddr);
+               gen_and(b2, b1);
 
-static struct block *
-gen_port6(port, ip_proto, dir)
-       int port;
-       int ip_proto;
-       int dir;
-{
-       struct block *b0, *b1, *tmp;
+               /*
+                * Now OR together the last two checks.  That gives
+                * the complete set of checks for data frames.
+                */
+               gen_or(b1, b0);
 
-       /* ether proto ip */
-       b0 =  gen_linktype(ETHERTYPE_IPV6);
+               /*
+                * Now check for a data frame.
+                * I.e, check "link[0] & 0x08".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x08;
+               b1->stmts = s;
 
-       switch (ip_proto) {
-       case IPPROTO_UDP:
-       case IPPROTO_TCP:
-               b1 = gen_portop6(port, ip_proto, dir);
-               break;
+               /*
+                * AND that with the checks done for data frames.
+                */
+               gen_and(b1, b0);
 
-       case PROTO_UNDEF:
-               tmp = gen_portop6(port, IPPROTO_TCP, dir);
-               b1 = gen_portop6(port, IPPROTO_UDP, dir);
-               gen_or(tmp, b1);
-               break;
+               /*
+                * If the high-order bit of the type value is 0, this
+                * is a management frame.
+                * I.e, check "!(link[0] & 0x08)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b2 = new_block(JMP(BPF_JSET));
+               b2->s.k = 0x08;
+               b2->stmts = s;
+               gen_not(b2);
 
-       default:
-               abort();
-       }
-       gen_and(b0, b1);
-       return b1;
-}
-#endif /* INET6 */
+               /*
+                * For management frames, the DA is at 4.
+                */
+               b1 = gen_bcmp(4, 6, eaddr);
+               gen_and(b2, b1);
 
-static int
-lookup_proto(name, proto)
-       register const char *name;
-       register int proto;
-{
-       register int v;
+               /*
+                * OR that with the checks done for data frames.
+                * That gives the checks done for management and
+                * data frames.
+                */
+               gen_or(b1, b0);
 
-       switch (proto) {
+               /*
+                * If the low-order bit of the type value is 1,
+                * this is either a control frame or a frame
+                * with a reserved type, and thus not a
+                * frame with an SA.
+                *
+                * I.e., check "!(link[0] & 0x04)".
+                */
+               s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+               s->s.k = 0;
+               b1 = new_block(JMP(BPF_JSET));
+               b1->s.k = 0x04;
+               b1->stmts = s;
+               gen_not(b1);
 
-       case Q_DEFAULT:
-       case Q_IP:
-               v = pcap_nametoproto(name);
-               if (v == PROTO_UNDEF)
-                       bpf_error("unknown ip proto '%s'", name);
-               break;
+               /*
+                * AND that with the checks for data and management
+                * frames.
+                */
+               gen_and(b1, b0);
+               return b0;
 
-       case Q_LINK:
-               /* XXX should look up h/w protocol type based on linktype */
-               v = pcap_nametoeproto(name);
-               if (v == PROTO_UNDEF)
-                       bpf_error("unknown ether proto '%s'", name);
-               break;
+       case Q_AND:
+               b0 = gen_wlanhostop(eaddr, Q_SRC);
+               b1 = gen_wlanhostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
 
-       default:
-               v = PROTO_UNDEF;
-               break;
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_wlanhostop(eaddr, Q_SRC);
+               b1 = gen_wlanhostop(eaddr, Q_DST);
+               gen_or(b0, b1);
+               return b1;
        }
-       return v;
+       abort();
+       /* NOTREACHED */
 }
 
-#if 0
-struct stmt *
-gen_joinsp(s, n)
-       struct stmt **s;
-       int n;
+/*
+ * Like gen_ehostop, but for RFC 2625 IP-over-Fibre-Channel.
+ * (We assume that the addresses are IEEE 48-bit MAC addresses,
+ * as the RFC states.)
+ */
+static struct block *
+gen_ipfchostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
 {
-       return NULL;
+       register struct block *b0, *b1;
+
+       switch (dir) {
+       case Q_SRC:
+               return gen_bcmp(10, 6, eaddr);
+
+       case Q_DST:
+               return gen_bcmp(2, 6, eaddr);
+
+       case Q_AND:
+               b0 = gen_ipfchostop(eaddr, Q_SRC);
+               b1 = gen_ipfchostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
+
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_ipfchostop(eaddr, Q_SRC);
+               b1 = gen_ipfchostop(eaddr, Q_DST);
+               gen_or(b0, b1);
+               return b1;
+       }
+       abort();
+       /* NOTREACHED */
 }
-#endif
 
-struct block *
-gen_protochain(v, proto, dir)
-       int v;
-       int proto;
+/*
+ * This is quite tricky because there may be pad bytes in front of the
+ * DECNET header, and then there are two possible data packet formats that
+ * carry both src and dst addresses, plus 5 packet types in a format that
+ * carries only the src node, plus 2 types that use a different format and
+ * also carry just the src node.
+ *
+ * Yuck.
+ *
+ * Instead of doing those all right, we just look for data packets with
+ * 0 or 1 bytes of padding.  If you want to look at other packets, that
+ * will require a lot more hacking.
+ *
+ * To add support for filtering on DECNET "areas" (network numbers)
+ * one would want to add a "mask" argument to this routine.  That would
+ * make the filter even more inefficient, although one could be clever
+ * and not generate masking instructions if the mask is 0xFFFF.
+ */
+static struct block *
+gen_dnhostop(addr, dir, base_off)
+       bpf_u_int32 addr;
        int dir;
+       u_int base_off;
 {
-#ifdef NO_PROTOCHAIN
-       return gen_proto(v, proto, dir);
-#else
-       struct block *b0, *b;
-       struct slist *s[100];
-       int fix2, fix3, fix4, fix5;
-       int ahcheck, again, end;
-       int i, max;
-       int reg1 = alloc_reg();
-       int reg2 = alloc_reg();
+       struct block *b0, *b1, *b2, *tmp;
+       u_int offset_lh;        /* offset if long header is received */
+       u_int offset_sh;        /* offset if short header is received */
 
-       memset(s, 0, sizeof(s));
-       fix2 = fix3 = fix4 = fix5 = 0;
+       switch (dir) {
 
-       switch (proto) {
-       case Q_IP:
-       case Q_IPV6:
+       case Q_DST:
+               offset_sh = 1;  /* follows flags */
+               offset_lh = 7;  /* flgs,darea,dsubarea,HIORD */
                break;
+
+       case Q_SRC:
+               offset_sh = 3;  /* follows flags, dstnode */
+               offset_lh = 15; /* flgs,darea,dsubarea,did,sarea,ssub,HIORD */
+               break;
+
+       case Q_AND:
+               /* Inefficient because we do our Calvinball dance twice */
+               b0 = gen_dnhostop(addr, Q_SRC, base_off);
+               b1 = gen_dnhostop(addr, Q_DST, base_off);
+               gen_and(b0, b1);
+               return b1;
+
+       case Q_OR:
        case Q_DEFAULT:
-               b0 = gen_protochain(v, Q_IP, dir);
-               b = gen_protochain(v, Q_IPV6, dir);
-               gen_or(b0, b);
-               return b;
+               /* Inefficient because we do our Calvinball dance twice */
+               b0 = gen_dnhostop(addr, Q_SRC, base_off);
+               b1 = gen_dnhostop(addr, Q_DST, base_off);
+               gen_or(b0, b1);
+               return b1;
+
+       case Q_ISO:
+               bpf_error("ISO host filtering not implemented");
+
        default:
-               bpf_error("bad protocol applied for 'protochain'");
-               /*NOTREACHED*/
+               abort();
        }
+       b0 = gen_linktype(ETHERTYPE_DN);
+       /* Check for pad = 1, long header case */
+       tmp = gen_mcmp(base_off + 2, BPF_H,
+           (bpf_int32)ntohs(0x0681), (bpf_int32)ntohs(0x07FF));
+       b1 = gen_cmp(base_off + 2 + 1 + offset_lh,
+           BPF_H, (bpf_int32)ntohs(addr));
+       gen_and(tmp, b1);
+       /* Check for pad = 0, long header case */
+       tmp = gen_mcmp(base_off + 2, BPF_B, (bpf_int32)0x06, (bpf_int32)0x7);
+       b2 = gen_cmp(base_off + 2 + offset_lh, BPF_H, (bpf_int32)ntohs(addr));
+       gen_and(tmp, b2);
+       gen_or(b2, b1);
+       /* Check for pad = 1, short header case */
+       tmp = gen_mcmp(base_off + 2, BPF_H,
+           (bpf_int32)ntohs(0x0281), (bpf_int32)ntohs(0x07FF));
+       b2 = gen_cmp(base_off + 2 + 1 + offset_sh,
+           BPF_H, (bpf_int32)ntohs(addr));
+       gen_and(tmp, b2);
+       gen_or(b2, b1);
+       /* Check for pad = 0, short header case */
+       tmp = gen_mcmp(base_off + 2, BPF_B, (bpf_int32)0x02, (bpf_int32)0x7);
+       b2 = gen_cmp(base_off + 2 + offset_sh, BPF_H, (bpf_int32)ntohs(addr));
+       gen_and(tmp, b2);
+       gen_or(b2, b1);
 
-       no_optimize = 1; /*this code is not compatible with optimzer yet */
+       /* Combine with test for linktype */
+       gen_and(b0, b1);
+       return b1;
+}
 
-       /*
-        * s[0] is a dummy entry to protect other BPF insn from damaged
-        * by s[fix] = foo with uninitialized variable "fix".  It is somewhat
-        * hard to find interdependency made by jump table fixup.
-        */
-       i = 0;
-       s[i] = new_stmt(0);     /*dummy*/
-       i++;
+static struct block *
+gen_host(addr, mask, proto, dir)
+       bpf_u_int32 addr;
+       bpf_u_int32 mask;
+       int proto;
+       int dir;
+{
+       struct block *b0, *b1;
 
        switch (proto) {
+
+       case Q_DEFAULT:
+               b0 = gen_host(addr, mask, Q_IP, dir);
+               if (off_linktype != -1) {
+                   b1 = gen_host(addr, mask, Q_ARP, dir);
+                   gen_or(b0, b1);
+                   b0 = gen_host(addr, mask, Q_RARP, dir);
+                   gen_or(b1, b0);
+               }
+               return b0;
+
        case Q_IP:
-               b0 = gen_linktype(ETHERTYPE_IP);
+               return gen_hostop(addr, mask, dir, ETHERTYPE_IP,
+                                 off_nl + 12, off_nl + 16);
 
-               /* A = ip->ip_p */
-               s[i] = new_stmt(BPF_LD|BPF_ABS|BPF_B);
-               s[i]->s.k = off_nl + 9;
-               i++;
-               /* X = ip->ip_hl << 2 */
-               s[i] = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
-               s[i]->s.k = off_nl;
-               i++;
-               break;
-#ifdef INET6
-       case Q_IPV6:
-               b0 = gen_linktype(ETHERTYPE_IPV6);
+       case Q_RARP:
+               return gen_hostop(addr, mask, dir, ETHERTYPE_REVARP,
+                                 off_nl + 14, off_nl + 24);
 
-               /* A = ip6->ip_nxt */
-               s[i] = new_stmt(BPF_LD|BPF_ABS|BPF_B);
-               s[i]->s.k = off_nl + 6;
-               i++;
-               /* X = sizeof(struct ip6_hdr) */
-               s[i] = new_stmt(BPF_LDX|BPF_IMM);
-               s[i]->s.k = 40;
-               i++;
+       case Q_ARP:
+               return gen_hostop(addr, mask, dir, ETHERTYPE_ARP,
+                                 off_nl + 14, off_nl + 24);
+
+       case Q_TCP:
+               bpf_error("'tcp' modifier applied to host");
+
+       case Q_SCTP:
+               bpf_error("'sctp' modifier applied to host");
+
+       case Q_UDP:
+               bpf_error("'udp' modifier applied to host");
+
+       case Q_ICMP:
+               bpf_error("'icmp' modifier applied to host");
+
+       case Q_IGMP:
+               bpf_error("'igmp' modifier applied to host");
+
+       case Q_IGRP:
+               bpf_error("'igrp' modifier applied to host");
+
+       case Q_PIM:
+               bpf_error("'pim' modifier applied to host");
+
+       case Q_VRRP:
+               bpf_error("'vrrp' modifier applied to host");
+
+       case Q_ATALK:
+               bpf_error("ATALK host filtering not implemented");
+
+       case Q_AARP:
+               bpf_error("AARP host filtering not implemented");
+
+       case Q_DECNET:
+               return gen_dnhostop(addr, dir, off_nl);
+
+       case Q_SCA:
+               bpf_error("SCA host filtering not implemented");
+
+       case Q_LAT:
+               bpf_error("LAT host filtering not implemented");
+
+       case Q_MOPDL:
+               bpf_error("MOPDL host filtering not implemented");
+
+       case Q_MOPRC:
+               bpf_error("MOPRC host filtering not implemented");
+
+#ifdef INET6
+       case Q_IPV6:
+               bpf_error("'ip6' modifier applied to ip host");
+
+       case Q_ICMPV6:
+               bpf_error("'icmp6' modifier applied to host");
+#endif /* INET6 */
+
+       case Q_AH:
+               bpf_error("'ah' modifier applied to host");
+
+       case Q_ESP:
+               bpf_error("'esp' modifier applied to host");
+
+       case Q_ISO:
+               bpf_error("ISO host filtering not implemented");
+
+       case Q_ESIS:
+               bpf_error("'esis' modifier applied to host");
+
+       case Q_ISIS:
+               bpf_error("'isis' modifier applied to host");
+
+       case Q_CLNP:
+               bpf_error("'clnp' modifier applied to host");
+
+       case Q_STP:
+               bpf_error("'stp' modifier applied to host");
+
+       case Q_IPX:
+               bpf_error("IPX host filtering not implemented");
+
+       case Q_NETBEUI:
+               bpf_error("'netbeui' modifier applied to host");
+
+       default:
+               abort();
+       }
+       /* NOTREACHED */
+}
+
+#ifdef INET6
+static struct block *
+gen_host6(addr, mask, proto, dir)
+       struct in6_addr *addr;
+       struct in6_addr *mask;
+       int proto;
+       int dir;
+{
+       switch (proto) {
+
+       case Q_DEFAULT:
+               return gen_host6(addr, mask, Q_IPV6, dir);
+
+       case Q_IP:
+               bpf_error("'ip' modifier applied to ip6 host");
+
+       case Q_RARP:
+               bpf_error("'rarp' modifier applied to ip6 host");
+
+       case Q_ARP:
+               bpf_error("'arp' modifier applied to ip6 host");
+
+       case Q_SCTP:
+               bpf_error("'sctp' modifier applied to host");
+
+       case Q_TCP:
+               bpf_error("'tcp' modifier applied to host");
+
+       case Q_UDP:
+               bpf_error("'udp' modifier applied to host");
+
+       case Q_ICMP:
+               bpf_error("'icmp' modifier applied to host");
+
+       case Q_IGMP:
+               bpf_error("'igmp' modifier applied to host");
+
+       case Q_IGRP:
+               bpf_error("'igrp' modifier applied to host");
+
+       case Q_PIM:
+               bpf_error("'pim' modifier applied to host");
+
+       case Q_VRRP:
+               bpf_error("'vrrp' modifier applied to host");
+
+       case Q_ATALK:
+               bpf_error("ATALK host filtering not implemented");
+
+       case Q_AARP:
+               bpf_error("AARP host filtering not implemented");
+
+       case Q_DECNET:
+               bpf_error("'decnet' modifier applied to ip6 host");
+
+       case Q_SCA:
+               bpf_error("SCA host filtering not implemented");
+
+       case Q_LAT:
+               bpf_error("LAT host filtering not implemented");
+
+       case Q_MOPDL:
+               bpf_error("MOPDL host filtering not implemented");
+
+       case Q_MOPRC:
+               bpf_error("MOPRC host filtering not implemented");
+
+       case Q_IPV6:
+               return gen_hostop6(addr, mask, dir, ETHERTYPE_IPV6,
+                                 off_nl + 8, off_nl + 24);
+
+       case Q_ICMPV6:
+               bpf_error("'icmp6' modifier applied to host");
+
+       case Q_AH:
+               bpf_error("'ah' modifier applied to host");
+
+       case Q_ESP:
+               bpf_error("'esp' modifier applied to host");
+
+       case Q_ISO:
+               bpf_error("ISO host filtering not implemented");
+
+       case Q_ESIS:
+               bpf_error("'esis' modifier applied to host");
+
+       case Q_ISIS:
+               bpf_error("'isis' modifier applied to host");
+
+       case Q_CLNP:
+               bpf_error("'clnp' modifier applied to host");
+
+       case Q_STP:
+               bpf_error("'stp' modifier applied to host");
+
+       case Q_IPX:
+               bpf_error("IPX host filtering not implemented");
+
+       case Q_NETBEUI:
+               bpf_error("'netbeui' modifier applied to host");
+
+       default:
+               abort();
+       }
+       /* NOTREACHED */
+}
+#endif /*INET6*/
+
+#ifndef INET6
+static struct block *
+gen_gateway(eaddr, alist, proto, dir)
+       const u_char *eaddr;
+       bpf_u_int32 **alist;
+       int proto;
+       int dir;
+{
+       struct block *b0, *b1, *tmp;
+
+       if (dir != 0)
+               bpf_error("direction applied to 'gateway'");
+
+       switch (proto) {
+       case Q_DEFAULT:
+       case Q_IP:
+       case Q_ARP:
+       case Q_RARP:
+               if (linktype == DLT_EN10MB)
+                       b0 = gen_ehostop(eaddr, Q_OR);
+               else if (linktype == DLT_FDDI)
+                       b0 = gen_fhostop(eaddr, Q_OR);
+               else if (linktype == DLT_IEEE802)
+                       b0 = gen_thostop(eaddr, Q_OR);
+               else if (linktype == DLT_IEEE802_11)
+                       b0 = gen_wlanhostop(eaddr, Q_OR);
+               else if (linktype == DLT_SUNATM && is_lane) {
+                       /*
+                        * Check that the packet doesn't begin with an
+                        * LE Control marker.  (We've already generated
+                        * a test for LANE.)
+                        */
+                       b1 = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H, 0xFF00);
+                       gen_not(b1);
+
+                       /*
+                        * Now check the MAC address.
+                        */
+                       b0 = gen_ehostop(eaddr, Q_OR);
+                       gen_and(b1, b0);
+               } else if (linktype == DLT_IP_OVER_FC)
+                       b0 = gen_ipfchostop(eaddr, Q_OR);
+               else
+                       bpf_error(
+                           "'gateway' supported only on ethernet/FDDI/token ring/802.11/Fibre Channel");
+
+               b1 = gen_host(**alist++, 0xffffffff, proto, Q_OR);
+               while (*alist) {
+                       tmp = gen_host(**alist++, 0xffffffff, proto, Q_OR);
+                       gen_or(b1, tmp);
+                       b1 = tmp;
+               }
+               gen_not(b1);
+               gen_and(b0, b1);
+               return b1;
+       }
+       bpf_error("illegal modifier of 'gateway'");
+       /* NOTREACHED */
+}
+#endif
+
+struct block *
+gen_proto_abbrev(proto)
+       int proto;
+{
+       struct block *b0;
+       struct block *b1;
+
+       switch (proto) {
+
+       case Q_SCTP:
+               b1 = gen_proto(IPPROTO_SCTP, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_SCTP, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+       case Q_TCP:
+               b1 = gen_proto(IPPROTO_TCP, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_TCP, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+       case Q_UDP:
+               b1 = gen_proto(IPPROTO_UDP, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_UDP, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+       case Q_ICMP:
+               b1 = gen_proto(IPPROTO_ICMP, Q_IP, Q_DEFAULT);
+               break;
+
+#ifndef        IPPROTO_IGMP
+#define        IPPROTO_IGMP    2
+#endif
+
+       case Q_IGMP:
+               b1 = gen_proto(IPPROTO_IGMP, Q_IP, Q_DEFAULT);
                break;
+
+#ifndef        IPPROTO_IGRP
+#define        IPPROTO_IGRP    9
 #endif
+       case Q_IGRP:
+               b1 = gen_proto(IPPROTO_IGRP, Q_IP, Q_DEFAULT);
+               break;
+
+#ifndef IPPROTO_PIM
+#define IPPROTO_PIM    103
+#endif
+
+       case Q_PIM:
+               b1 = gen_proto(IPPROTO_PIM, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_PIM, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+#ifndef IPPROTO_VRRP
+#define IPPROTO_VRRP   112
+#endif
+
+       case Q_VRRP:
+               b1 = gen_proto(IPPROTO_VRRP, Q_IP, Q_DEFAULT);
+               break;
+
+       case Q_IP:
+               b1 =  gen_linktype(ETHERTYPE_IP);
+               break;
+
+       case Q_ARP:
+               b1 =  gen_linktype(ETHERTYPE_ARP);
+               break;
+
+       case Q_RARP:
+               b1 =  gen_linktype(ETHERTYPE_REVARP);
+               break;
+
+       case Q_LINK:
+               bpf_error("link layer applied in wrong context");
+
+       case Q_ATALK:
+               b1 =  gen_linktype(ETHERTYPE_ATALK);
+               break;
+
+       case Q_AARP:
+               b1 =  gen_linktype(ETHERTYPE_AARP);
+               break;
+
+       case Q_DECNET:
+               b1 =  gen_linktype(ETHERTYPE_DN);
+               break;
+
+       case Q_SCA:
+               b1 =  gen_linktype(ETHERTYPE_SCA);
+               break;
+
+       case Q_LAT:
+               b1 =  gen_linktype(ETHERTYPE_LAT);
+               break;
+
+       case Q_MOPDL:
+               b1 =  gen_linktype(ETHERTYPE_MOPDL);
+               break;
+
+       case Q_MOPRC:
+               b1 =  gen_linktype(ETHERTYPE_MOPRC);
+               break;
+
+#ifdef INET6
+       case Q_IPV6:
+               b1 = gen_linktype(ETHERTYPE_IPV6);
+               break;
+
+#ifndef IPPROTO_ICMPV6
+#define IPPROTO_ICMPV6 58
+#endif
+       case Q_ICMPV6:
+               b1 = gen_proto(IPPROTO_ICMPV6, Q_IPV6, Q_DEFAULT);
+               break;
+#endif /* INET6 */
+
+#ifndef IPPROTO_AH
+#define IPPROTO_AH     51
+#endif
+       case Q_AH:
+               b1 = gen_proto(IPPROTO_AH, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_AH, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+#ifndef IPPROTO_ESP
+#define IPPROTO_ESP    50
+#endif
+       case Q_ESP:
+               b1 = gen_proto(IPPROTO_ESP, Q_IP, Q_DEFAULT);
+#ifdef INET6
+               b0 = gen_proto(IPPROTO_ESP, Q_IPV6, Q_DEFAULT);
+               gen_or(b0, b1);
+#endif
+               break;
+
+       case Q_ISO:
+               b1 = gen_linktype(LLCSAP_ISONS);
+               break;
+
+       case Q_ESIS:
+               b1 = gen_proto(ISO9542_ESIS, Q_ISO, Q_DEFAULT);
+               break;
+
+       case Q_ISIS:
+               b1 = gen_proto(ISO10589_ISIS, Q_ISO, Q_DEFAULT);
+               break;
+
+       case Q_ISIS_L1: /* all IS-IS Level1 PDU-Types */
+               b0 = gen_proto(ISIS_L1_LAN_IIH, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_PTP_IIH, Q_ISIS, Q_DEFAULT); /* FIXME extract the circuit-type bits */
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L1_LSP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L1_CSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L1_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_L2: /* all IS-IS Level2 PDU-Types */
+               b0 = gen_proto(ISIS_L2_LAN_IIH, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_PTP_IIH, Q_ISIS, Q_DEFAULT); /* FIXME extract the circuit-type bits */
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L2_LSP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L2_CSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L2_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_IIH: /* all IS-IS Hello PDU-Types */
+               b0 = gen_proto(ISIS_L1_LAN_IIH, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_L2_LAN_IIH, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_PTP_IIH, Q_ISIS, Q_DEFAULT);                
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_LSP: 
+               b0 = gen_proto(ISIS_L1_LSP, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_L2_LSP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_SNP:
+               b0 = gen_proto(ISIS_L1_CSNP, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_L2_CSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L1_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               b0 = gen_proto(ISIS_L2_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_CSNP:
+               b0 = gen_proto(ISIS_L1_PSNP, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_L2_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_ISIS_PSNP:
+               b0 = gen_proto(ISIS_L1_PSNP, Q_ISIS, Q_DEFAULT);
+               b1 = gen_proto(ISIS_L2_PSNP, Q_ISIS, Q_DEFAULT);
+               gen_or(b0, b1);
+               break;
+
+       case Q_CLNP:
+               b1 = gen_proto(ISO8473_CLNP, Q_ISO, Q_DEFAULT);
+               break;
+
+       case Q_STP:
+               b1 = gen_linktype(LLCSAP_8021D);
+               break;
+
+       case Q_IPX:
+               b1 = gen_linktype(LLCSAP_IPX);
+               break;
+
+       case Q_NETBEUI:
+               b1 = gen_linktype(LLCSAP_NETBEUI);
+               break;
+
+       default:
+               abort();
+       }
+       return b1;
+}
+
+static struct block *
+gen_ipfrag()
+{
+       struct slist *s;
+       struct block *b;
+
+       /* not ip frag */
+       s = new_stmt(BPF_LD|BPF_H|BPF_ABS);
+       s->s.k = off_nl + 6;
+       b = new_block(JMP(BPF_JSET));
+       b->s.k = 0x1fff;
+       b->stmts = s;
+       gen_not(b);
+
+       return b;
+}
+
+static struct block *
+gen_portatom(off, v)
+       int off;
+       bpf_int32 v;
+{
+       struct slist *s;
+       struct block *b;
+
+       s = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
+       s->s.k = off_nl;
+
+       s->next = new_stmt(BPF_LD|BPF_IND|BPF_H);
+       s->next->s.k = off_nl + off;
+
+       b = new_block(JMP(BPF_JEQ));
+       b->stmts = s;
+       b->s.k = v;
+
+       return b;
+}
+
+#ifdef INET6
+static struct block *
+gen_portatom6(off, v)
+       int off;
+       bpf_int32 v;
+{
+       return gen_cmp(off_nl + 40 + off, BPF_H, v);
+}
+#endif/*INET6*/
+
+struct block *
+gen_portop(port, proto, dir)
+       int port, proto, dir;
+{
+       struct block *b0, *b1, *tmp;
+
+       /* ip proto 'proto' */
+       tmp = gen_cmp(off_nl + 9, BPF_B, (bpf_int32)proto);
+       b0 = gen_ipfrag();
+       gen_and(tmp, b0);
+
+       switch (dir) {
+       case Q_SRC:
+               b1 = gen_portatom(0, (bpf_int32)port);
+               break;
+
+       case Q_DST:
+               b1 = gen_portatom(2, (bpf_int32)port);
+               break;
+
+       case Q_OR:
+       case Q_DEFAULT:
+               tmp = gen_portatom(0, (bpf_int32)port);
+               b1 = gen_portatom(2, (bpf_int32)port);
+               gen_or(tmp, b1);
+               break;
+
+       case Q_AND:
+               tmp = gen_portatom(0, (bpf_int32)port);
+               b1 = gen_portatom(2, (bpf_int32)port);
+               gen_and(tmp, b1);
+               break;
+
+       default:
+               abort();
+       }
+       gen_and(b0, b1);
+
+       return b1;
+}
+
+static struct block *
+gen_port(port, ip_proto, dir)
+       int port;
+       int ip_proto;
+       int dir;
+{
+       struct block *b0, *b1, *tmp;
+
+       /*
+        * ether proto ip
+        *
+        * For FDDI, RFC 1188 says that SNAP encapsulation is used,
+        * not LLC encapsulation with LLCSAP_IP.
+        *
+        * For IEEE 802 networks - which includes 802.5 token ring
+        * (which is what DLT_IEEE802 means) and 802.11 - RFC 1042
+        * says that SNAP encapsulation is used, not LLC encapsulation
+        * with LLCSAP_IP.
+        *
+        * For LLC-encapsulated ATM/"Classical IP", RFC 1483 and
+        * RFC 2225 say that SNAP encapsulation is used, not LLC
+        * encapsulation with LLCSAP_IP.
+        *
+        * So we always check for ETHERTYPE_IP.
+        */
+       b0 =  gen_linktype(ETHERTYPE_IP);
+
+       switch (ip_proto) {
+       case IPPROTO_UDP:
+       case IPPROTO_TCP:
+       case IPPROTO_SCTP:
+               b1 = gen_portop(port, ip_proto, dir);
+               break;
+
+       case PROTO_UNDEF:
+               tmp = gen_portop(port, IPPROTO_TCP, dir);
+               b1 = gen_portop(port, IPPROTO_UDP, dir);
+               gen_or(tmp, b1);
+               tmp = gen_portop(port, IPPROTO_SCTP, dir);
+               gen_or(tmp, b1);
+               break;
+
+       default:
+               abort();
+       }
+       gen_and(b0, b1);
+       return b1;
+}
+
+#ifdef INET6
+struct block *
+gen_portop6(port, proto, dir)
+       int port, proto, dir;
+{
+       struct block *b0, *b1, *tmp;
+
+       /* ip proto 'proto' */
+       b0 = gen_cmp(off_nl + 6, BPF_B, (bpf_int32)proto);
+
+       switch (dir) {
+       case Q_SRC:
+               b1 = gen_portatom6(0, (bpf_int32)port);
+               break;
+
+       case Q_DST:
+               b1 = gen_portatom6(2, (bpf_int32)port);
+               break;
+
+       case Q_OR:
+       case Q_DEFAULT:
+               tmp = gen_portatom6(0, (bpf_int32)port);
+               b1 = gen_portatom6(2, (bpf_int32)port);
+               gen_or(tmp, b1);
+               break;
+
+       case Q_AND:
+               tmp = gen_portatom6(0, (bpf_int32)port);
+               b1 = gen_portatom6(2, (bpf_int32)port);
+               gen_and(tmp, b1);
+               break;
+
+       default:
+               abort();
+       }
+       gen_and(b0, b1);
+
+       return b1;
+}
+
+static struct block *
+gen_port6(port, ip_proto, dir)
+       int port;
+       int ip_proto;
+       int dir;
+{
+       struct block *b0, *b1, *tmp;
+
+       /* ether proto ip */
+       b0 =  gen_linktype(ETHERTYPE_IPV6);
+
+       switch (ip_proto) {
+       case IPPROTO_UDP:
+       case IPPROTO_TCP:
+       case IPPROTO_SCTP:
+               b1 = gen_portop6(port, ip_proto, dir);
+               break;
+
+       case PROTO_UNDEF:
+               tmp = gen_portop6(port, IPPROTO_TCP, dir);
+               b1 = gen_portop6(port, IPPROTO_UDP, dir);
+               gen_or(tmp, b1);
+               tmp = gen_portop6(port, IPPROTO_SCTP, dir);
+               gen_or(tmp, b1);
+               break;
+
+       default:
+               abort();
+       }
+       gen_and(b0, b1);
+       return b1;
+}
+#endif /* INET6 */
+
+static int
+lookup_proto(name, proto)
+       register const char *name;
+       register int proto;
+{
+       register int v;
+
+       switch (proto) {
+
+       case Q_DEFAULT:
+       case Q_IP:
+       case Q_IPV6:
+               v = pcap_nametoproto(name);
+               if (v == PROTO_UNDEF)
+                       bpf_error("unknown ip proto '%s'", name);
+               break;
+
+       case Q_LINK:
+               /* XXX should look up h/w protocol type based on linktype */
+               v = pcap_nametoeproto(name);
+               if (v == PROTO_UNDEF)
+                       bpf_error("unknown ether proto '%s'", name);
+               break;
+
+       case Q_ISO:
+               if (strcmp(name, "esis") == 0)
+                       v = ISO9542_ESIS;
+               else if (strcmp(name, "isis") == 0)
+                       v = ISO10589_ISIS;
+               else if (strcmp(name, "clnp") == 0)
+                       v = ISO8473_CLNP;
+               else
+                       bpf_error("unknown osi proto '%s'", name);
+               break;
+
+       default:
+               v = PROTO_UNDEF;
+               break;
+       }
+       return v;
+}
+
+#if 0
+struct stmt *
+gen_joinsp(s, n)
+       struct stmt **s;
+       int n;
+{
+       return NULL;
+}
+#endif
+
+static struct block *
+gen_protochain(v, proto, dir)
+       int v;
+       int proto;
+       int dir;
+{
+#ifdef NO_PROTOCHAIN
+       return gen_proto(v, proto, dir);
+#else
+       struct block *b0, *b;
+       struct slist *s[100];
+       int fix2, fix3, fix4, fix5;
+       int ahcheck, again, end;
+       int i, max;
+       int reg2 = alloc_reg();
+
+       memset(s, 0, sizeof(s));
+       fix2 = fix3 = fix4 = fix5 = 0;
+
+       switch (proto) {
+       case Q_IP:
+       case Q_IPV6:
+               break;
+       case Q_DEFAULT:
+               b0 = gen_protochain(v, Q_IP, dir);
+               b = gen_protochain(v, Q_IPV6, dir);
+               gen_or(b0, b);
+               return b;
+       default:
+               bpf_error("bad protocol applied for 'protochain'");
+               /*NOTREACHED*/
+       }
+
+       no_optimize = 1; /*this code is not compatible with optimzer yet */
+
+       /*
+        * s[0] is a dummy entry to protect other BPF insn from damaged
+        * by s[fix] = foo with uninitialized variable "fix".  It is somewhat
+        * hard to find interdependency made by jump table fixup.
+        */
+       i = 0;
+       s[i] = new_stmt(0);     /*dummy*/
+       i++;
+
+       switch (proto) {
+       case Q_IP:
+               b0 = gen_linktype(ETHERTYPE_IP);
+
+               /* A = ip->ip_p */
+               s[i] = new_stmt(BPF_LD|BPF_ABS|BPF_B);
+               s[i]->s.k = off_nl + 9;
+               i++;
+               /* X = ip->ip_hl << 2 */
+               s[i] = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
+               s[i]->s.k = off_nl;
+               i++;
+               break;
+#ifdef INET6
+       case Q_IPV6:
+               b0 = gen_linktype(ETHERTYPE_IPV6);
+
+               /* A = ip6->ip_nxt */
+               s[i] = new_stmt(BPF_LD|BPF_ABS|BPF_B);
+               s[i]->s.k = off_nl + 6;
+               i++;
+               /* X = sizeof(struct ip6_hdr) */
+               s[i] = new_stmt(BPF_LDX|BPF_IMM);
+               s[i]->s.k = 40;
+               i++;
+               break;
+#endif
+       default:
+               bpf_error("unsupported proto to gen_protochain");
+               /*NOTREACHED*/
+       }
+
+       /* again: if (A == v) goto end; else fall through; */
+       again = i;
+       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+       s[i]->s.k = v;
+       s[i]->s.jt = NULL;              /*later*/
+       s[i]->s.jf = NULL;              /*update in next stmt*/
+       fix5 = i;
+       i++;
+
+#ifndef IPPROTO_NONE
+#define IPPROTO_NONE   59
+#endif
+       /* if (A == IPPROTO_NONE) goto end */
+       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+       s[i]->s.jt = NULL;      /*later*/
+       s[i]->s.jf = NULL;      /*update in next stmt*/
+       s[i]->s.k = IPPROTO_NONE;
+       s[fix5]->s.jf = s[i];
+       fix2 = i;
+       i++;
+
+#ifdef INET6
+       if (proto == Q_IPV6) {
+               int v6start, v6end, v6advance, j;
+
+               v6start = i;
+               /* if (A == IPPROTO_HOPOPTS) goto v6advance */
+               s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+               s[i]->s.jt = NULL;      /*later*/
+               s[i]->s.jf = NULL;      /*update in next stmt*/
+               s[i]->s.k = IPPROTO_HOPOPTS;
+               s[fix2]->s.jf = s[i];
+               i++;
+               /* if (A == IPPROTO_DSTOPTS) goto v6advance */
+               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+               s[i]->s.jt = NULL;      /*later*/
+               s[i]->s.jf = NULL;      /*update in next stmt*/
+               s[i]->s.k = IPPROTO_DSTOPTS;
+               i++;
+               /* if (A == IPPROTO_ROUTING) goto v6advance */
+               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+               s[i]->s.jt = NULL;      /*later*/
+               s[i]->s.jf = NULL;      /*update in next stmt*/
+               s[i]->s.k = IPPROTO_ROUTING;
+               i++;
+               /* if (A == IPPROTO_FRAGMENT) goto v6advance; else goto ahcheck; */
+               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+               s[i]->s.jt = NULL;      /*later*/
+               s[i]->s.jf = NULL;      /*later*/
+               s[i]->s.k = IPPROTO_FRAGMENT;
+               fix3 = i;
+               v6end = i;
+               i++;
+
+               /* v6advance: */
+               v6advance = i;
+
+               /*
+                * in short,
+                * A = P[X];
+                * X = X + (P[X + 1] + 1) * 8;
+                */
+               /* A = X */
+               s[i] = new_stmt(BPF_MISC|BPF_TXA);
+               i++;
+               /* A = P[X + packet head] */
+               s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
+               s[i]->s.k = off_nl;
+               i++;
+               /* MEM[reg2] = A */
+               s[i] = new_stmt(BPF_ST);
+               s[i]->s.k = reg2;
+               i++;
+               /* A = X */
+               s[i] = new_stmt(BPF_MISC|BPF_TXA);
+               i++;
+               /* A += 1 */
+               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+               s[i]->s.k = 1;
+               i++;
+               /* X = A */
+               s[i] = new_stmt(BPF_MISC|BPF_TAX);
+               i++;
+               /* A = P[X + packet head]; */
+               s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
+               s[i]->s.k = off_nl;
+               i++;
+               /* A += 1 */
+               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+               s[i]->s.k = 1;
+               i++;
+               /* A *= 8 */
+               s[i] = new_stmt(BPF_ALU|BPF_MUL|BPF_K);
+               s[i]->s.k = 8;
+               i++;
+               /* X = A; */
+               s[i] = new_stmt(BPF_MISC|BPF_TAX);
+               i++;
+               /* A = MEM[reg2] */
+               s[i] = new_stmt(BPF_LD|BPF_MEM);
+               s[i]->s.k = reg2;
+               i++;
+
+               /* goto again; (must use BPF_JA for backward jump) */
+               s[i] = new_stmt(BPF_JMP|BPF_JA);
+               s[i]->s.k = again - i - 1;
+               s[i - 1]->s.jf = s[i];
+               i++;
+
+               /* fixup */
+               for (j = v6start; j <= v6end; j++)
+                       s[j]->s.jt = s[v6advance];
+       } else
+#endif
+       {
+               /* nop */
+               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+               s[i]->s.k = 0;
+               s[fix2]->s.jf = s[i];
+               i++;
+       }
+
+       /* ahcheck: */
+       ahcheck = i;
+       /* if (A == IPPROTO_AH) then fall through; else goto end; */
+       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
+       s[i]->s.jt = NULL;      /*later*/
+       s[i]->s.jf = NULL;      /*later*/
+       s[i]->s.k = IPPROTO_AH;
+       if (fix3)
+               s[fix3]->s.jf = s[ahcheck];
+       fix4 = i;
+       i++;
+
+       /*
+        * in short,
+        * A = P[X];
+        * X = X + (P[X + 1] + 2) * 4;
+        */
+       /* A = X */
+       s[i - 1]->s.jt = s[i] = new_stmt(BPF_MISC|BPF_TXA);
+       i++;
+       /* A = P[X + packet head]; */
+       s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
+       s[i]->s.k = off_nl;
+       i++;
+       /* MEM[reg2] = A */
+       s[i] = new_stmt(BPF_ST);
+       s[i]->s.k = reg2;
+       i++;
+       /* A = X */
+       s[i - 1]->s.jt = s[i] = new_stmt(BPF_MISC|BPF_TXA);
+       i++;
+       /* A += 1 */
+       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+       s[i]->s.k = 1;
+       i++;
+       /* X = A */
+       s[i] = new_stmt(BPF_MISC|BPF_TAX);
+       i++;
+       /* A = P[X + packet head] */
+       s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
+       s[i]->s.k = off_nl;
+       i++;
+       /* A += 2 */
+       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+       s[i]->s.k = 2;
+       i++;
+       /* A *= 4 */
+       s[i] = new_stmt(BPF_ALU|BPF_MUL|BPF_K);
+       s[i]->s.k = 4;
+       i++;
+       /* X = A; */
+       s[i] = new_stmt(BPF_MISC|BPF_TAX);
+       i++;
+       /* A = MEM[reg2] */
+       s[i] = new_stmt(BPF_LD|BPF_MEM);
+       s[i]->s.k = reg2;
+       i++;
+
+       /* goto again; (must use BPF_JA for backward jump) */
+       s[i] = new_stmt(BPF_JMP|BPF_JA);
+       s[i]->s.k = again - i - 1;
+       i++;
+
+       /* end: nop */
+       end = i;
+       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
+       s[i]->s.k = 0;
+       s[fix2]->s.jt = s[end];
+       s[fix4]->s.jf = s[end];
+       s[fix5]->s.jt = s[end];
+       i++;
+
+       /*
+        * make slist chain
+        */
+       max = i;
+       for (i = 0; i < max - 1; i++)
+               s[i]->next = s[i + 1];
+       s[max - 1]->next = NULL;
+
+       /*
+        * emit final check
+        */
+       b = new_block(JMP(BPF_JEQ));
+       b->stmts = s[1];        /*remember, s[0] is dummy*/
+       b->s.k = v;
+
+       free_reg(reg2);
+
+       gen_and(b0, b);
+       return b;
+#endif
+}
+
+static struct block *
+gen_proto(v, proto, dir)
+       int v;
+       int proto;
+       int dir;
+{
+       struct block *b0, *b1;
+
+       if (dir != Q_DEFAULT)
+               bpf_error("direction applied to 'proto'");
+
+       switch (proto) {
+       case Q_DEFAULT:
+#ifdef INET6
+               b0 = gen_proto(v, Q_IP, dir);
+               b1 = gen_proto(v, Q_IPV6, dir);
+               gen_or(b0, b1);
+               return b1;
+#else
+               /*FALLTHROUGH*/
+#endif
+       case Q_IP:
+               /*
+                * For FDDI, RFC 1188 says that SNAP encapsulation is used,
+                * not LLC encapsulation with LLCSAP_IP.
+                *
+                * For IEEE 802 networks - which includes 802.5 token ring
+                * (which is what DLT_IEEE802 means) and 802.11 - RFC 1042
+                * says that SNAP encapsulation is used, not LLC encapsulation
+                * with LLCSAP_IP.
+                *
+                * For LLC-encapsulated ATM/"Classical IP", RFC 1483 and
+                * RFC 2225 say that SNAP encapsulation is used, not LLC
+                * encapsulation with LLCSAP_IP.
+                *
+                * So we always check for ETHERTYPE_IP.
+                */
+               b0 = gen_linktype(ETHERTYPE_IP);
+#ifndef CHASE_CHAIN
+               b1 = gen_cmp(off_nl + 9, BPF_B, (bpf_int32)v);
+#else
+               b1 = gen_protochain(v, Q_IP);
+#endif
+               gen_and(b0, b1);
+               return b1;
+
+       case Q_ISO:
+               switch (linktype) {
+
+               case DLT_FRELAY:
+                       /*
+                        * Frame Relay packets typically have an OSI
+                        * NLPID at the beginning; "gen_linktype(LLCSAP_ISONS)"
+                        * generates code to check for all the OSI
+                        * NLPIDs, so calling it and then adding a check
+                        * for the particular NLPID for which we're
+                        * looking is bogus, as we can just check for
+                        * the NLPID.
+                        *
+                        * What we check for is the NLPID and a frame
+                        * control field value of UI, i.e. 0x03 followed
+                        * by the NLPID.
+                        *
+                        * XXX - assumes a 2-byte Frame Relay header with
+                        * DLCI and flags.  What if the address is longer?
+                        *
+                        * XXX - what about SNAP-encapsulated frames?
+                        */
+                       return gen_cmp(2, BPF_H, (0x03<<8) | v);
+                       break;
+
+               case DLT_C_HDLC:
+                       /*
+                        * Cisco uses an Ethertype lookalike - for OSI,
+                        * it's 0xfefe.
+                        */
+                       b0 = gen_linktype(LLCSAP_ISONS<<8 | LLCSAP_ISONS);
+                       /* OSI in C-HDLC is stuffed with a fudge byte */
+                       b1 = gen_cmp(off_nl_nosnap+1, BPF_B, (long)v);
+                       gen_and(b0, b1);
+                       return b1;
+
+               default:
+                       b0 = gen_linktype(LLCSAP_ISONS);
+                       b1 = gen_cmp(off_nl_nosnap, BPF_B, (long)v);
+                       gen_and(b0, b1);
+                       return b1;
+               }
+
+       case Q_ISIS:
+               b0 = gen_proto(ISO10589_ISIS, Q_ISO, Q_DEFAULT);
+               /*
+                * 4 is the offset of the PDU type relative to the IS-IS
+                * header.
+                */
+               b1 = gen_cmp(off_nl_nosnap+4, BPF_B, (long)v);
+               gen_and(b0, b1);
+               return b1;
+
+       case Q_ARP:
+               bpf_error("arp does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_RARP:
+               bpf_error("rarp does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_ATALK:
+               bpf_error("atalk encapsulation is not specifiable");
+               /* NOTREACHED */
+
+       case Q_DECNET:
+               bpf_error("decnet encapsulation is not specifiable");
+               /* NOTREACHED */
+
+       case Q_SCA:
+               bpf_error("sca does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_LAT:
+               bpf_error("lat does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_MOPRC:
+               bpf_error("moprc does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_MOPDL:
+               bpf_error("mopdl does not encapsulate another protocol");
+               /* NOTREACHED */
+
+       case Q_LINK:
+               return gen_linktype(v);
+
+       case Q_UDP:
+               bpf_error("'udp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_TCP:
+               bpf_error("'tcp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_SCTP:
+               bpf_error("'sctp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_ICMP:
+               bpf_error("'icmp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_IGMP:
+               bpf_error("'igmp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_IGRP:
+               bpf_error("'igrp proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_PIM:
+               bpf_error("'pim proto' is bogus");
+               /* NOTREACHED */
+
+       case Q_VRRP:
+               bpf_error("'vrrp proto' is bogus");
+               /* NOTREACHED */
+
+#ifdef INET6
+       case Q_IPV6:
+               b0 = gen_linktype(ETHERTYPE_IPV6);
+#ifndef CHASE_CHAIN
+               b1 = gen_cmp(off_nl + 6, BPF_B, (bpf_int32)v);
+#else
+               b1 = gen_protochain(v, Q_IPV6);
+#endif
+               gen_and(b0, b1);
+               return b1;
+
+       case Q_ICMPV6:
+               bpf_error("'icmp6 proto' is bogus");
+#endif /* INET6 */
+
+       case Q_AH:
+               bpf_error("'ah proto' is bogus");
+
+       case Q_ESP:
+               bpf_error("'ah proto' is bogus");
+
+       case Q_STP:
+               bpf_error("'stp proto' is bogus");
+
+       case Q_IPX:
+               bpf_error("'ipx proto' is bogus");
+
+       case Q_NETBEUI:
+               bpf_error("'netbeui proto' is bogus");
+
+       default:
+               abort();
+               /* NOTREACHED */
+       }
+       /* NOTREACHED */
+}
+
+struct block *
+gen_scode(name, q)
+       register const char *name;
+       struct qual q;
+{
+       int proto = q.proto;
+       int dir = q.dir;
+       int tproto;
+       u_char *eaddr;
+       bpf_u_int32 mask, addr;
+#ifndef INET6
+       bpf_u_int32 **alist;
+#else
+       int tproto6;
+       struct sockaddr_in *sin;
+       struct sockaddr_in6 *sin6;
+       struct addrinfo *res, *res0;
+       struct in6_addr mask128;
+#endif /*INET6*/
+       struct block *b, *tmp;
+       int port, real_proto;
+
+       switch (q.addr) {
+
+       case Q_NET:
+               addr = pcap_nametonetaddr(name);
+               if (addr == 0)
+                       bpf_error("unknown network '%s'", name);
+               /* Left justify network addr and calculate its network mask */
+               mask = 0xffffffff;
+               while (addr && (addr & 0xff000000) == 0) {
+                       addr <<= 8;
+                       mask <<= 8;
+               }
+               return gen_host(addr, mask, proto, dir);
+
+       case Q_DEFAULT:
+       case Q_HOST:
+               if (proto == Q_LINK) {
+                       switch (linktype) {
+
+                       case DLT_EN10MB:
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown ether host '%s'", name);
+                               b = gen_ehostop(eaddr, dir);
+                               free(eaddr);
+                               return b;
+
+                       case DLT_FDDI:
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown FDDI host '%s'", name);
+                               b = gen_fhostop(eaddr, dir);
+                               free(eaddr);
+                               return b;
+
+                       case DLT_IEEE802:
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown token ring host '%s'", name);
+                               b = gen_thostop(eaddr, dir);
+                               free(eaddr);
+                               return b;
+
+                       case DLT_IEEE802_11:
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown 802.11 host '%s'", name);
+                               b = gen_wlanhostop(eaddr, dir);
+                               free(eaddr);
+                               return b;
+
+                       case DLT_IP_OVER_FC:
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown Fibre Channel host '%s'", name);
+                               b = gen_ipfchostop(eaddr, dir);
+                               free(eaddr);
+                               return b;
+
+                       case DLT_SUNATM:
+                               if (!is_lane)
+                                       break;
+
+                               /*
+                                * Check that the packet doesn't begin
+                                * with an LE Control marker.  (We've
+                                * already generated a test for LANE.)
+                                */
+                               tmp = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H,
+                                   0xFF00);
+                               gen_not(tmp);
+
+                               eaddr = pcap_ether_hostton(name);
+                               if (eaddr == NULL)
+                                       bpf_error(
+                                           "unknown ether host '%s'", name);
+                               b = gen_ehostop(eaddr, dir);
+                               gen_and(tmp, b);
+                               free(eaddr);
+                               return b;
+                       }
+
+                       bpf_error("only ethernet/FDDI/token ring/802.11/ATM LANE/Fibre Channel supports link-level host name");
+               } else if (proto == Q_DECNET) {
+                       unsigned short dn_addr = __pcap_nametodnaddr(name);
+                       /*
+                        * I don't think DECNET hosts can be multihomed, so
+                        * there is no need to build up a list of addresses
+                        */
+                       return (gen_host(dn_addr, 0, proto, dir));
+               } else {
+#ifndef INET6
+                       alist = pcap_nametoaddr(name);
+                       if (alist == NULL || *alist == NULL)
+                               bpf_error("unknown host '%s'", name);
+                       tproto = proto;
+                       if (off_linktype == -1 && tproto == Q_DEFAULT)
+                               tproto = Q_IP;
+                       b = gen_host(**alist++, 0xffffffff, tproto, dir);
+                       while (*alist) {
+                               tmp = gen_host(**alist++, 0xffffffff,
+                                              tproto, dir);
+                               gen_or(b, tmp);
+                               b = tmp;
+                       }
+                       return b;
+#else
+                       memset(&mask128, 0xff, sizeof(mask128));
+                       res0 = res = pcap_nametoaddrinfo(name);
+                       if (res == NULL)
+                               bpf_error("unknown host '%s'", name);
+                       b = tmp = NULL;
+                       tproto = tproto6 = proto;
+                       if (off_linktype == -1 && tproto == Q_DEFAULT) {
+                               tproto = Q_IP;
+                               tproto6 = Q_IPV6;
+                       }
+                       for (res = res0; res; res = res->ai_next) {
+                               switch (res->ai_family) {
+                               case AF_INET:
+                                       if (tproto == Q_IPV6)
+                                               continue;
+
+                                       sin = (struct sockaddr_in *)
+                                               res->ai_addr;
+                                       tmp = gen_host(ntohl(sin->sin_addr.s_addr),
+                                               0xffffffff, tproto, dir);
+                                       break;
+                               case AF_INET6:
+                                       if (tproto6 == Q_IP)
+                                               continue;
+
+                                       sin6 = (struct sockaddr_in6 *)
+                                               res->ai_addr;
+                                       tmp = gen_host6(&sin6->sin6_addr,
+                                               &mask128, tproto6, dir);
+                                       break;
+                               default:
+                                       continue;
+                               }
+                               if (b)
+                                       gen_or(b, tmp);
+                               b = tmp;
+                       }
+                       freeaddrinfo(res0);
+                       if (b == NULL) {
+                               bpf_error("unknown host '%s'%s", name,
+                                   (proto == Q_DEFAULT)
+                                       ? ""
+                                       : " for specified address family");
+                       }
+                       return b;
+#endif /*INET6*/
+               }
+
+       case Q_PORT:
+               if (proto != Q_DEFAULT &&
+                   proto != Q_UDP && proto != Q_TCP && proto != Q_SCTP)
+                       bpf_error("illegal qualifier of 'port'");
+               if (pcap_nametoport(name, &port, &real_proto) == 0)
+                       bpf_error("unknown port '%s'", name);
+               if (proto == Q_UDP) {
+                       if (real_proto == IPPROTO_TCP)
+                               bpf_error("port '%s' is tcp", name);
+                       else if (real_proto == IPPROTO_SCTP)
+                               bpf_error("port '%s' is sctp", name);
+                       else
+                               /* override PROTO_UNDEF */
+                               real_proto = IPPROTO_UDP;
+               }
+               if (proto == Q_TCP) {
+                       if (real_proto == IPPROTO_UDP)
+                               bpf_error("port '%s' is udp", name);
+
+                       else if (real_proto == IPPROTO_SCTP)
+                               bpf_error("port '%s' is sctp", name);
+                       else
+                               /* override PROTO_UNDEF */
+                               real_proto = IPPROTO_TCP;
+               }
+               if (proto == Q_SCTP) {
+                       if (real_proto == IPPROTO_UDP)
+                               bpf_error("port '%s' is udp", name);
+
+                       else if (real_proto == IPPROTO_TCP)
+                               bpf_error("port '%s' is tcp", name);
+                       else
+                               /* override PROTO_UNDEF */
+                               real_proto = IPPROTO_SCTP;
+               }
+#ifndef INET6
+               return gen_port(port, real_proto, dir);
+#else
+           {
+               struct block *b;
+               b = gen_port(port, real_proto, dir);
+               gen_or(gen_port6(port, real_proto, dir), b);
+               return b;
+           }
+#endif /* INET6 */
+
+       case Q_GATEWAY:
+#ifndef INET6
+               eaddr = pcap_ether_hostton(name);
+               if (eaddr == NULL)
+                       bpf_error("unknown ether host: %s", name);
+
+               alist = pcap_nametoaddr(name);
+               if (alist == NULL || *alist == NULL)
+                       bpf_error("unknown host '%s'", name);
+               b = gen_gateway(eaddr, alist, proto, dir);
+               free(eaddr);
+               return b;
+#else
+               bpf_error("'gateway' not supported in this configuration");
+#endif /*INET6*/
+
+       case Q_PROTO:
+               real_proto = lookup_proto(name, proto);
+               if (real_proto >= 0)
+                       return gen_proto(real_proto, proto, dir);
+               else
+                       bpf_error("unknown protocol: %s", name);
+
+       case Q_PROTOCHAIN:
+               real_proto = lookup_proto(name, proto);
+               if (real_proto >= 0)
+                       return gen_protochain(real_proto, proto, dir);
+               else
+                       bpf_error("unknown protocol: %s", name);
+
+
+       case Q_UNDEF:
+               syntax();
+               /* NOTREACHED */
+       }
+       abort();
+       /* NOTREACHED */
+}
+
+struct block *
+gen_mcode(s1, s2, masklen, q)
+       register const char *s1, *s2;
+       register int masklen;
+       struct qual q;
+{
+       register int nlen, mlen;
+       bpf_u_int32 n, m;
+
+       nlen = __pcap_atoin(s1, &n);
+       /* Promote short ipaddr */
+       n <<= 32 - nlen;
+
+       if (s2 != NULL) {
+               mlen = __pcap_atoin(s2, &m);
+               /* Promote short ipaddr */
+               m <<= 32 - mlen;
+               if ((n & ~m) != 0)
+                       bpf_error("non-network bits set in \"%s mask %s\"",
+                           s1, s2);
+       } else {
+               /* Convert mask len to mask */
+               if (masklen > 32)
+                       bpf_error("mask length must be <= 32");
+               m = 0xffffffff << (32 - masklen);
+               if ((n & ~m) != 0)
+                       bpf_error("non-network bits set in \"%s/%d\"",
+                           s1, masklen);
+       }
+
+       switch (q.addr) {
+
+       case Q_NET:
+               return gen_host(n, m, q.proto, q.dir);
+
        default:
-               bpf_error("unsupported proto to gen_protochain");
-               /*NOTREACHED*/
+               bpf_error("Mask syntax for networks only");
+               /* NOTREACHED */
        }
+}
 
-       /* again: if (A == v) goto end; else fall through; */
-       again = i;
-       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-       s[i]->s.k = v;
-       s[i]->s.jt = NULL;              /*later*/
-       s[i]->s.jf = NULL;              /*update in next stmt*/
-       fix5 = i;
-       i++;
+struct block *
+gen_ncode(s, v, q)
+       register const char *s;
+       bpf_u_int32 v;
+       struct qual q;
+{
+       bpf_u_int32 mask;
+       int proto = q.proto;
+       int dir = q.dir;
+       register int vlen;
 
-#ifndef IPPROTO_NONE
-#define IPPROTO_NONE   59
-#endif
-       /* if (A == IPPROTO_NONE) goto end */
-       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-       s[i]->s.jt = NULL;      /*later*/
-       s[i]->s.jf = NULL;      /*update in next stmt*/
-       s[i]->s.k = IPPROTO_NONE;
-       s[fix5]->s.jf = s[i];
-       fix2 = i;
-       i++;
+       if (s == NULL)
+               vlen = 32;
+       else if (q.proto == Q_DECNET)
+               vlen = __pcap_atodn(s, &v);
+       else
+               vlen = __pcap_atoin(s, &v);
 
-#ifdef INET6
-       if (proto == Q_IPV6) {
-               int v6start, v6end, v6advance, j;
+       switch (q.addr) {
 
-               v6start = i;
-               /* if (A == IPPROTO_HOPOPTS) goto v6advance */
-               s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-               s[i]->s.jt = NULL;      /*later*/
-               s[i]->s.jf = NULL;      /*update in next stmt*/
-               s[i]->s.k = IPPROTO_HOPOPTS;
-               s[fix2]->s.jf = s[i];
-               i++;
-               /* if (A == IPPROTO_DSTOPTS) goto v6advance */
-               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-               s[i]->s.jt = NULL;      /*later*/
-               s[i]->s.jf = NULL;      /*update in next stmt*/
-               s[i]->s.k = IPPROTO_DSTOPTS;
-               i++;
-               /* if (A == IPPROTO_ROUTING) goto v6advance */
-               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-               s[i]->s.jt = NULL;      /*later*/
-               s[i]->s.jf = NULL;      /*update in next stmt*/
-               s[i]->s.k = IPPROTO_ROUTING;
-               i++;
-               /* if (A == IPPROTO_FRAGMENT) goto v6advance; else goto ahcheck; */
-               s[i - 1]->s.jf = s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-               s[i]->s.jt = NULL;      /*later*/
-               s[i]->s.jf = NULL;      /*later*/
-               s[i]->s.k = IPPROTO_FRAGMENT;
-               fix3 = i;
-               v6end = i;
-               i++;
+       case Q_DEFAULT:
+       case Q_HOST:
+       case Q_NET:
+               if (proto == Q_DECNET)
+                       return gen_host(v, 0, proto, dir);
+               else if (proto == Q_LINK) {
+                       bpf_error("illegal link layer address");
+               } else {
+                       mask = 0xffffffff;
+                       if (s == NULL && q.addr == Q_NET) {
+                               /* Promote short net number */
+                               while (v && (v & 0xff000000) == 0) {
+                                       v <<= 8;
+                                       mask <<= 8;
+                               }
+                       } else {
+                               /* Promote short ipaddr */
+                               v <<= 32 - vlen;
+                               mask <<= 32 - vlen;
+                       }
+                       return gen_host(v, mask, proto, dir);
+               }
 
-               /* v6advance: */
-               v6advance = i;
+       case Q_PORT:
+               if (proto == Q_UDP)
+                       proto = IPPROTO_UDP;
+               else if (proto == Q_TCP)
+                       proto = IPPROTO_TCP;
+               else if (proto == Q_SCTP)
+                       proto = IPPROTO_SCTP;
+               else if (proto == Q_DEFAULT)
+                       proto = PROTO_UNDEF;
+               else
+                       bpf_error("illegal qualifier of 'port'");
 
-               /*
-                * in short,
-                * A = P[X + 1];
-                * X = X + (P[X] + 1) * 8;
-                */
-               /* A = X */
-               s[i] = new_stmt(BPF_MISC|BPF_TXA);
-               i++;
-               /* MEM[reg1] = A */
-               s[i] = new_stmt(BPF_ST);
-               s[i]->s.k = reg1;
-               i++;
-               /* A += 1 */
-               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-               s[i]->s.k = 1;
-               i++;
-               /* X = A */
-               s[i] = new_stmt(BPF_MISC|BPF_TAX);
-               i++;
-               /* A = P[X + packet head]; */
-               s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
-               s[i]->s.k = off_nl;
-               i++;
-               /* MEM[reg2] = A */
-               s[i] = new_stmt(BPF_ST);
-               s[i]->s.k = reg2;
-               i++;
-               /* X = MEM[reg1] */
-               s[i] = new_stmt(BPF_LDX|BPF_MEM);
-               s[i]->s.k = reg1;
-               i++;
-               /* A = P[X + packet head] */
-               s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
-               s[i]->s.k = off_nl;
-               i++;
-               /* A += 1 */
-               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-               s[i]->s.k = 1;
-               i++;
-               /* A *= 8 */
-               s[i] = new_stmt(BPF_ALU|BPF_MUL|BPF_K);
-               s[i]->s.k = 8;
-               i++;
-               /* X = A; */
-               s[i] = new_stmt(BPF_MISC|BPF_TAX);
-               i++;
-               /* A = MEM[reg2] */
-               s[i] = new_stmt(BPF_LD|BPF_MEM);
-               s[i]->s.k = reg2;
-               i++;
+#ifndef INET6
+               return gen_port((int)v, proto, dir);
+#else
+           {
+               struct block *b;
+               b = gen_port((int)v, proto, dir);
+               gen_or(gen_port6((int)v, proto, dir), b);
+               return b;
+           }
+#endif /* INET6 */
+
+       case Q_GATEWAY:
+               bpf_error("'gateway' requires a name");
+               /* NOTREACHED */
+
+       case Q_PROTO:
+               return gen_proto((int)v, proto, dir);
+
+       case Q_PROTOCHAIN:
+               return gen_protochain((int)v, proto, dir);
+
+       case Q_UNDEF:
+               syntax();
+               /* NOTREACHED */
+
+       default:
+               abort();
+               /* NOTREACHED */
+       }
+       /* NOTREACHED */
+}
+
+#ifdef INET6
+struct block *
+gen_mcode6(s1, s2, masklen, q)
+       register const char *s1, *s2;
+       register int masklen;
+       struct qual q;
+{
+       struct addrinfo *res;
+       struct in6_addr *addr;
+       struct in6_addr mask;
+       struct block *b;
+       u_int32_t *a, *m;
 
-               /* goto again; (must use BPF_JA for backward jump) */
-               s[i] = new_stmt(BPF_JMP|BPF_JA);
-               s[i]->s.k = again - i - 1;
-               s[i - 1]->s.jf = s[i];
-               i++;
+       if (s2)
+               bpf_error("no mask %s supported", s2);
 
-               /* fixup */
-               for (j = v6start; j <= v6end; j++)
-                       s[j]->s.jt = s[v6advance];
-       } else
-#endif
-       {
-               /* nop */
-               s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-               s[i]->s.k = 0;
-               s[fix2]->s.jf = s[i];
-               i++;
+       res = pcap_nametoaddrinfo(s1);
+       if (!res)
+               bpf_error("invalid ip6 address %s", s1);
+       if (res->ai_next)
+               bpf_error("%s resolved to multiple address", s1);
+       addr = &((struct sockaddr_in6 *)res->ai_addr)->sin6_addr;
+
+       if (sizeof(mask) * 8 < masklen)
+               bpf_error("mask length must be <= %u", (unsigned int)(sizeof(mask) * 8));
+       memset(&mask, 0, sizeof(mask));
+       memset(&mask, 0xff, masklen / 8);
+       if (masklen % 8) {
+               mask.s6_addr[masklen / 8] =
+                       (0xff << (8 - masklen % 8)) & 0xff;
        }
 
-       /* ahcheck: */
-       ahcheck = i;
-       /* if (A == IPPROTO_AH) then fall through; else goto end; */
-       s[i] = new_stmt(BPF_JMP|BPF_JEQ|BPF_K);
-       s[i]->s.jt = NULL;      /*later*/
-       s[i]->s.jf = NULL;      /*later*/
-       s[i]->s.k = IPPROTO_AH;
-       if (fix3)
-               s[fix3]->s.jf = s[ahcheck];
-       fix4 = i;
-       i++;
+       a = (u_int32_t *)addr;
+       m = (u_int32_t *)&mask;
+       if ((a[0] & ~m[0]) || (a[1] & ~m[1])
+        || (a[2] & ~m[2]) || (a[3] & ~m[3])) {
+               bpf_error("non-network bits set in \"%s/%d\"", s1, masklen);
+       }
 
-       /*
-        * in short,
-        * A = P[X + 1];
-        * X = X + (P[X] + 2) * 4;
-        */
-       /* A = X */
-       s[i - 1]->s.jt = s[i] = new_stmt(BPF_MISC|BPF_TXA);
-       i++;
-       /* MEM[reg1] = A */
-       s[i] = new_stmt(BPF_ST);
-       s[i]->s.k = reg1;
-       i++;
-       /* A += 1 */
-       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-       s[i]->s.k = 1;
-       i++;
-       /* X = A */
-       s[i] = new_stmt(BPF_MISC|BPF_TAX);
-       i++;
-       /* A = P[X + packet head]; */
-       s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
-       s[i]->s.k = off_nl;
-       i++;
-       /* MEM[reg2] = A */
-       s[i] = new_stmt(BPF_ST);
-       s[i]->s.k = reg2;
-       i++;
-       /* X = MEM[reg1] */
-       s[i] = new_stmt(BPF_LDX|BPF_MEM);
-       s[i]->s.k = reg1;
-       i++;
-       /* A = P[X + packet head] */
-       s[i] = new_stmt(BPF_LD|BPF_IND|BPF_B);
-       s[i]->s.k = off_nl;
-       i++;
-       /* A += 2 */
-       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-       s[i]->s.k = 2;
-       i++;
-       /* A *= 4 */
-       s[i] = new_stmt(BPF_ALU|BPF_MUL|BPF_K);
-       s[i]->s.k = 4;
-       i++;
-       /* X = A; */
-       s[i] = new_stmt(BPF_MISC|BPF_TAX);
-       i++;
-       /* A = MEM[reg2] */
-       s[i] = new_stmt(BPF_LD|BPF_MEM);
-       s[i]->s.k = reg2;
-       i++;
+       switch (q.addr) {
 
-       /* goto again; (must use BPF_JA for backward jump) */
-       s[i] = new_stmt(BPF_JMP|BPF_JA);
-       s[i]->s.k = again - i - 1;
-       i++;
+       case Q_DEFAULT:
+       case Q_HOST:
+               if (masklen != 128)
+                       bpf_error("Mask syntax for networks only");
+               /* FALLTHROUGH */
 
-       /* end: nop */
-       end = i;
-       s[i] = new_stmt(BPF_ALU|BPF_ADD|BPF_K);
-       s[i]->s.k = 0;
-       s[fix2]->s.jt = s[end];
-       s[fix4]->s.jf = s[end];
-       s[fix5]->s.jt = s[end];
-       i++;
+       case Q_NET:
+               b = gen_host6(addr, &mask, q.proto, q.dir);
+               freeaddrinfo(res);
+               return b;
 
-       /*
-        * make slist chain
-        */
-       max = i;
-       for (i = 0; i < max - 1; i++)
-               s[i]->next = s[i + 1];
-       s[max - 1]->next = NULL;
+       default:
+               bpf_error("invalid qualifier against IPv6 address");
+               /* NOTREACHED */
+       }
+}
+#endif /*INET6*/
+
+struct block *
+gen_ecode(eaddr, q)
+       register const u_char *eaddr;
+       struct qual q;
+{
+       struct block *b, *tmp;
+
+       if ((q.addr == Q_HOST || q.addr == Q_DEFAULT) && q.proto == Q_LINK) {
+               if (linktype == DLT_EN10MB)
+                       return gen_ehostop(eaddr, (int)q.dir);
+               if (linktype == DLT_FDDI)
+                       return gen_fhostop(eaddr, (int)q.dir);
+               if (linktype == DLT_IEEE802)
+                       return gen_thostop(eaddr, (int)q.dir);
+               if (linktype == DLT_IEEE802_11)
+                       return gen_wlanhostop(eaddr, (int)q.dir);
+               if (linktype == DLT_SUNATM && is_lane) {
+                       /*
+                        * Check that the packet doesn't begin with an
+                        * LE Control marker.  (We've already generated
+                        * a test for LANE.)
+                        */
+                       tmp = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H, 0xFF00);
+                       gen_not(tmp);
+
+                       /*
+                        * Now check the MAC address.
+                        */
+                       b = gen_ehostop(eaddr, (int)q.dir);
+                       gen_and(tmp, b);
+                       return b;
+               }
+               if (linktype == DLT_IP_OVER_FC)
+                       return gen_ipfchostop(eaddr, (int)q.dir);
+               bpf_error("ethernet addresses supported only on ethernet/FDDI/token ring/802.11/ATM LANE/Fibre Channel");
+       }
+       bpf_error("ethernet address used in non-ether expression");
+       /* NOTREACHED */
+}
 
+void
+sappend(s0, s1)
+       struct slist *s0, *s1;
+{
        /*
-        * emit final check
+        * This is definitely not the best way to do this, but the
+        * lists will rarely get long.
         */
-       b = new_block(JMP(BPF_JEQ));
-       b->stmts = s[1];        /*remember, s[0] is dummy*/
-       b->s.k = v;
+       while (s0->next)
+               s0 = s0->next;
+       s0->next = s1;
+}
 
-       free_reg(reg1);
-       free_reg(reg2);
+static struct slist *
+xfer_to_x(a)
+       struct arth *a;
+{
+       struct slist *s;
 
-       gen_and(b0, b);
-       return b;
-#endif
+       s = new_stmt(BPF_LDX|BPF_MEM);
+       s->s.k = a->regno;
+       return s;
 }
 
-static struct block *
-gen_proto(v, proto, dir)
-       int v;
+static struct slist *
+xfer_to_a(a)
+       struct arth *a;
+{
+       struct slist *s;
+
+       s = new_stmt(BPF_LD|BPF_MEM);
+       s->s.k = a->regno;
+       return s;
+}
+
+struct arth *
+gen_load(proto, index, size)
        int proto;
-       int dir;
+       struct arth *index;
+       int size;
 {
-       struct block *b0, *b1;
+       struct slist *s, *tmp;
+       struct block *b;
+       int regno = alloc_reg();
 
-       if (dir != Q_DEFAULT)
-               bpf_error("direction applied to 'proto'");
+       free_reg(index->regno);
+       switch (size) {
+
+       default:
+               bpf_error("data size must be 1, 2, or 4");
+
+       case 1:
+               size = BPF_B;
+               break;
+
+       case 2:
+               size = BPF_H;
+               break;
 
+       case 4:
+               size = BPF_W;
+               break;
+       }
        switch (proto) {
-       case Q_DEFAULT:
-#ifdef INET6
-               b0 = gen_proto(v, Q_IP, dir);
-               b1 = gen_proto(v, Q_IPV6, dir);
-               gen_or(b0, b1);
-               return b1;
-#else
-               /*FALLTHROUGH*/
-#endif
-       case Q_IP:
-               b0 = gen_linktype(ETHERTYPE_IP);
-#ifndef CHASE_CHAIN
-               b1 = gen_cmp(off_nl + 9, BPF_B, (bpf_int32)v);
-#else
-               b1 = gen_protochain(v, Q_IP);
-#endif
-               gen_and(b0, b1);
-               return b1;
+       default:
+               bpf_error("unsupported index operation");
 
-       case Q_ARP:
-               bpf_error("arp does not encapsulate another protocol");
-               /* NOTREACHED */
+       case Q_LINK:
+               /*
+                * XXX - what about ATM LANE?  Should the index be
+                * relative to the beginning of the AAL5 frame, so
+                * that 0 refers to the beginning of the LE Control
+                * field, or relative to the beginning of the LAN
+                * frame, so that 0 refers, for Ethernet LANE, to
+                * the beginning of the destination address?
+                */
+               s = xfer_to_x(index);
+               tmp = new_stmt(BPF_LD|BPF_IND|size);
+               sappend(s, tmp);
+               sappend(index->s, s);
+               break;
 
+       case Q_IP:
+       case Q_ARP:
        case Q_RARP:
-               bpf_error("rarp does not encapsulate another protocol");
-               /* NOTREACHED */
-
        case Q_ATALK:
-               bpf_error("atalk encapsulation is not specifiable");
-               /* NOTREACHED */
-
        case Q_DECNET:
-               bpf_error("decnet encapsulation is not specifiable");
-               /* NOTREACHED */
-
        case Q_SCA:
-               bpf_error("sca does not encapsulate another protocol");
-               /* NOTREACHED */
-
        case Q_LAT:
-               bpf_error("lat does not encapsulate another protocol");
-               /* NOTREACHED */
-
        case Q_MOPRC:
-               bpf_error("moprc does not encapsulate another protocol");
-               /* NOTREACHED */
-
        case Q_MOPDL:
-               bpf_error("mopdl does not encapsulate another protocol");
-               /* NOTREACHED */
-
-       case Q_LINK:
-               return gen_linktype(v);
+#ifdef INET6
+       case Q_IPV6:
+#endif
+               /* XXX Note that we assume a fixed link header here. */
+               s = xfer_to_x(index);
+               tmp = new_stmt(BPF_LD|BPF_IND|size);
+               tmp->s.k = off_nl;
+               sappend(s, tmp);
+               sappend(index->s, s);
 
-       case Q_UDP:
-               bpf_error("'udp proto' is bogus");
-               /* NOTREACHED */
+               b = gen_proto_abbrev(proto);
+               if (index->b)
+                       gen_and(index->b, b);
+               index->b = b;
+               break;
 
+       case Q_SCTP:
        case Q_TCP:
-               bpf_error("'tcp proto' is bogus");
-               /* NOTREACHED */
-
+       case Q_UDP:
        case Q_ICMP:
-               bpf_error("'icmp proto' is bogus");
-               /* NOTREACHED */
-
        case Q_IGMP:
-               bpf_error("'igmp proto' is bogus");
-               /* NOTREACHED */
-
        case Q_IGRP:
-               bpf_error("'igrp proto' is bogus");
-               /* NOTREACHED */
-
        case Q_PIM:
-               bpf_error("'pim proto' is bogus");
-               /* NOTREACHED */
+       case Q_VRRP:
+               s = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
+               s->s.k = off_nl;
+               sappend(s, xfer_to_a(index));
+               sappend(s, new_stmt(BPF_ALU|BPF_ADD|BPF_X));
+               sappend(s, new_stmt(BPF_MISC|BPF_TAX));
+               sappend(s, tmp = new_stmt(BPF_LD|BPF_IND|size));
+               tmp->s.k = off_nl;
+               sappend(index->s, s);
 
+               gen_and(gen_proto_abbrev(proto), b = gen_ipfrag());
+               if (index->b)
+                       gen_and(index->b, b);
 #ifdef INET6
-       case Q_IPV6:
-               b0 = gen_linktype(ETHERTYPE_IPV6);
-#ifndef CHASE_CHAIN
-               b1 = gen_cmp(off_nl + 6, BPF_B, (bpf_int32)v);
-#else
-               b1 = gen_protochain(v, Q_IPV6);
+               gen_and(gen_proto_abbrev(Q_IP), b);
 #endif
-               gen_and(b0, b1);
-               return b1;
-
+               index->b = b;
+               break;
+#ifdef INET6
        case Q_ICMPV6:
-               bpf_error("'icmp6 proto' is bogus");
-#endif /* INET6 */
-
-       case Q_AH:
-               bpf_error("'ah proto' is bogus");
-
-       case Q_ESP:
-               bpf_error("'ah proto' is bogus");
-
-       default:
-               abort();
-               /* NOTREACHED */
+               bpf_error("IPv6 upper-layer protocol is not supported by proto[x]");
+               /*NOTREACHED*/
+#endif
        }
-       /* NOTREACHED */
-}
-
-struct block *
-gen_scode(name, q)
-       register const char *name;
-       struct qual q;
-{
-       int proto = q.proto;
-       int dir = q.dir;
-       int tproto;
-       u_char *eaddr;
-       bpf_u_int32 mask, addr;
-#ifndef INET6
-       bpf_u_int32 **alist;
-#else
-       int tproto6;
-       struct sockaddr_in *sin;
-       struct sockaddr_in6 *sin6;
-       struct addrinfo *res, *res0;
-       struct in6_addr mask128;
-#endif /*INET6*/
-       struct block *b, *tmp;
-       int port, real_proto;
-
-       switch (q.addr) {
-
-       case Q_NET:
-               addr = pcap_nametonetaddr(name);
-               if (addr == 0)
-                       bpf_error("unknown network '%s'", name);
-               /* Left justify network addr and calculate its network mask */
-               mask = 0xffffffff;
-               while (addr && (addr & 0xff000000) == 0) {
-                       addr <<= 8;
-                       mask <<= 8;
-               }
-               return gen_host(addr, mask, proto, dir);
-
-       case Q_DEFAULT:
-       case Q_HOST:
-               if (proto == Q_LINK) {
-                       switch (linktype) {
-
-                       case DLT_EN10MB:
-                               eaddr = pcap_ether_hostton(name);
-                               if (eaddr == NULL)
-                                       bpf_error(
-                                           "unknown ether host '%s'", name);
-                               return gen_ehostop(eaddr, dir);
-
-                       case DLT_FDDI:
-                               eaddr = pcap_ether_hostton(name);
-                               if (eaddr == NULL)
-                                       bpf_error(
-                                           "unknown FDDI host '%s'", name);
-                               return gen_fhostop(eaddr, dir);
-
-                       default:
-                               bpf_error(
-                       "only ethernet/FDDI supports link-level host name");
-                               break;
-                       }
-               } else if (proto == Q_DECNET) {
-                       unsigned short dn_addr = __pcap_nametodnaddr(name);
-                       /*
-                        * I don't think DECNET hosts can be multihomed, so
-                        * there is no need to build up a list of addresses
-                        */
-                       return (gen_host(dn_addr, 0, proto, dir));
-               } else {
-#ifndef INET6
-                       alist = pcap_nametoaddr(name);
-                       if (alist == NULL || *alist == NULL)
-                               bpf_error("unknown host '%s'", name);
-                       tproto = proto;
-                       if (off_linktype == -1 && tproto == Q_DEFAULT)
-                               tproto = Q_IP;
-                       b = gen_host(**alist++, 0xffffffff, tproto, dir);
-                       while (*alist) {
-                               tmp = gen_host(**alist++, 0xffffffff,
-                                              tproto, dir);
-                               gen_or(b, tmp);
-                               b = tmp;
-                       }
-                       return b;
-#else
-                       memset(&mask128, 0xff, sizeof(mask128));
-                       res0 = res = pcap_nametoaddrinfo(name);
-                       if (res == NULL)
-                               bpf_error("unknown host '%s'", name);
-                       b = tmp = NULL;
-                       tproto = tproto6 = proto;
-                       if (off_linktype == -1 && tproto == Q_DEFAULT) {
-                               tproto = Q_IP;
-                               tproto6 = Q_IPV6;
-                       }
-                       for (res = res0; res; res = res->ai_next) {
-                               switch (res->ai_family) {
-                               case AF_INET:
-                                       if (tproto == Q_IPV6)
-                                               continue;
-
-                                       sin = (struct sockaddr_in *)
-                                               res->ai_addr;
-                                       tmp = gen_host(ntohl(sin->sin_addr.s_addr),
-                                               0xffffffff, tproto, dir);
-                                       break;
-                               case AF_INET6:
-                                       if (tproto6 == Q_IP)
-                                               continue;
-
-                                       sin6 = (struct sockaddr_in6 *)
-                                               res->ai_addr;
-                                       tmp = gen_host6(&sin6->sin6_addr,
-                                               &mask128, tproto6, dir);
-                                       break;
-                               }
-                               if (b)
-                                       gen_or(b, tmp);
-                               b = tmp;
-                       }
-                       freeaddrinfo(res0);
-                       if (b == NULL) {
-                               bpf_error("unknown host '%s'%s", name,
-                                   (proto == Q_DEFAULT)
-                                       ? ""
-                                       : " for specified address family");
-                       }
-                       return b;
-#endif /*INET6*/
-               }
-
-       case Q_PORT:
-               if (proto != Q_DEFAULT && proto != Q_UDP && proto != Q_TCP)
-                       bpf_error("illegal qualifier of 'port'");
-               if (pcap_nametoport(name, &port, &real_proto) == 0)
-                       bpf_error("unknown port '%s'", name);
-               if (proto == Q_UDP) {
-                       if (real_proto == IPPROTO_TCP)
-                               bpf_error("port '%s' is tcp", name);
-                       else
-                               /* override PROTO_UNDEF */
-                               real_proto = IPPROTO_UDP;
-               }
-               if (proto == Q_TCP) {
-                       if (real_proto == IPPROTO_UDP)
-                               bpf_error("port '%s' is udp", name);
-                       else
-                               /* override PROTO_UNDEF */
-                               real_proto = IPPROTO_TCP;
-               }
-#ifndef INET6
-               return gen_port(port, real_proto, dir);
-#else
-           {
-               struct block *b;
-               b = gen_port(port, real_proto, dir);
-               gen_or(gen_port6(port, real_proto, dir), b);
-               return b;
-           }
-#endif /* INET6 */
+       index->regno = regno;
+       s = new_stmt(BPF_ST);
+       s->s.k = regno;
+       sappend(index->s, s);
 
-       case Q_GATEWAY:
-#ifndef INET6
-               eaddr = pcap_ether_hostton(name);
-               if (eaddr == NULL)
-                       bpf_error("unknown ether host: %s", name);
+       return index;
+}
 
-               alist = pcap_nametoaddr(name);
-               if (alist == NULL || *alist == NULL)
-                       bpf_error("unknown host '%s'", name);
-               return gen_gateway(eaddr, alist, proto, dir);
-#else
-               bpf_error("'gateway' not supported in this configuration");
-#endif /*INET6*/
+struct block *
+gen_relation(code, a0, a1, reversed)
+       int code;
+       struct arth *a0, *a1;
+       int reversed;
+{
+       struct slist *s0, *s1, *s2;
+       struct block *b, *tmp;
 
-       case Q_PROTO:
-               real_proto = lookup_proto(name, proto);
-               if (real_proto >= 0)
-                       return gen_proto(real_proto, proto, dir);
-               else
-                       bpf_error("unknown protocol: %s", name);
+       s0 = xfer_to_x(a1);
+       s1 = xfer_to_a(a0);
+       if (code == BPF_JEQ) {
+               s2 = new_stmt(BPF_ALU|BPF_SUB|BPF_X);
+               b = new_block(JMP(code));
+               sappend(s1, s2);
+       }
+       else
+               b = new_block(BPF_JMP|code|BPF_X);
+       if (reversed)
+               gen_not(b);
 
-       case Q_PROTOCHAIN:
-               real_proto = lookup_proto(name, proto);
-               if (real_proto >= 0)
-                       return gen_protochain(real_proto, proto, dir);
+       sappend(s0, s1);
+       sappend(a1->s, s0);
+       sappend(a0->s, a1->s);
+
+       b->stmts = a0->s;
+
+       free_reg(a0->regno);
+       free_reg(a1->regno);
+
+       /* 'and' together protocol checks */
+       if (a0->b) {
+               if (a1->b) {
+                       gen_and(a0->b, tmp = a1->b);
+               }
                else
-                       bpf_error("unknown protocol: %s", name);
+                       tmp = a0->b;
+       } else
+               tmp = a1->b;
 
+       if (tmp)
+               gen_and(tmp, b);
 
-       case Q_UNDEF:
-               syntax();
-               /* NOTREACHED */
-       }
-       abort();
-       /* NOTREACHED */
+       return b;
 }
 
-struct block *
-gen_mcode(s1, s2, masklen, q)
-       register const char *s1, *s2;
-       register int masklen;
-       struct qual q;
+struct arth *
+gen_loadlen()
 {
-       register int nlen, mlen;
-       bpf_u_int32 n, m;
+       int regno = alloc_reg();
+       struct arth *a = (struct arth *)newchunk(sizeof(*a));
+       struct slist *s;
 
-       nlen = __pcap_atoin(s1, &n);
-       /* Promote short ipaddr */
-       n <<= 32 - nlen;
+       s = new_stmt(BPF_LD|BPF_LEN);
+       s->next = new_stmt(BPF_ST);
+       s->next->s.k = regno;
+       a->s = s;
+       a->regno = regno;
 
-       if (s2 != NULL) {
-               mlen = __pcap_atoin(s2, &m);
-               /* Promote short ipaddr */
-               m <<= 32 - mlen;
-               if ((n & ~m) != 0)
-                       bpf_error("non-network bits set in \"%s mask %s\"",
-                           s1, s2);
-       } else {
-               /* Convert mask len to mask */
-               if (masklen > 32)
-                       bpf_error("mask length must be <= 32");
-               m = 0xffffffff << (32 - masklen);
-               if ((n & ~m) != 0)
-                       bpf_error("non-network bits set in \"%s/%d\"",
-                           s1, masklen);
-       }
+       return a;
+}
 
-       switch (q.addr) {
+struct arth *
+gen_loadi(val)
+       int val;
+{
+       struct arth *a;
+       struct slist *s;
+       int reg;
 
-       case Q_NET:
-               return gen_host(n, m, q.proto, q.dir);
+       a = (struct arth *)newchunk(sizeof(*a));
 
-       default:
-               bpf_error("Mask syntax for networks only");
-               /* NOTREACHED */
-       }
+       reg = alloc_reg();
+
+       s = new_stmt(BPF_LD|BPF_IMM);
+       s->s.k = val;
+       s->next = new_stmt(BPF_ST);
+       s->next->s.k = reg;
+       a->s = s;
+       a->regno = reg;
+
+       return a;
 }
 
-struct block *
-gen_ncode(s, v, q)
-       register const char *s;
-       bpf_u_int32 v;
-       struct qual q;
+struct arth *
+gen_neg(a)
+       struct arth *a;
 {
-       bpf_u_int32 mask;
-       int proto = q.proto;
-       int dir = q.dir;
-       register int vlen;
+       struct slist *s;
 
-       if (s == NULL)
-               vlen = 32;
-       else if (q.proto == Q_DECNET)
-               vlen = __pcap_atodn(s, &v);
-       else
-               vlen = __pcap_atoin(s, &v);
+       s = xfer_to_a(a);
+       sappend(a->s, s);
+       s = new_stmt(BPF_ALU|BPF_NEG);
+       s->s.k = 0;
+       sappend(a->s, s);
+       s = new_stmt(BPF_ST);
+       s->s.k = a->regno;
+       sappend(a->s, s);
 
-       switch (q.addr) {
+       return a;
+}
 
-       case Q_DEFAULT:
-       case Q_HOST:
-       case Q_NET:
-               if (proto == Q_DECNET)
-                       return gen_host(v, 0, proto, dir);
-               else if (proto == Q_LINK) {
-                       bpf_error("illegal link layer address");
-               } else {
-                       mask = 0xffffffff;
-                       if (s == NULL && q.addr == Q_NET) {
-                               /* Promote short net number */
-                               while (v && (v & 0xff000000) == 0) {
-                                       v <<= 8;
-                                       mask <<= 8;
-                               }
-                       } else {
-                               /* Promote short ipaddr */
-                               v <<= 32 - vlen;
-                               mask <<= 32 - vlen;
-                       }
-                       return gen_host(v, mask, proto, dir);
-               }
+struct arth *
+gen_arth(code, a0, a1)
+       int code;
+       struct arth *a0, *a1;
+{
+       struct slist *s0, *s1, *s2;
 
-       case Q_PORT:
-               if (proto == Q_UDP)
-                       proto = IPPROTO_UDP;
-               else if (proto == Q_TCP)
-                       proto = IPPROTO_TCP;
-               else if (proto == Q_DEFAULT)
-                       proto = PROTO_UNDEF;
-               else
-                       bpf_error("illegal qualifier of 'port'");
+       s0 = xfer_to_x(a1);
+       s1 = xfer_to_a(a0);
+       s2 = new_stmt(BPF_ALU|BPF_X|code);
 
-#ifndef INET6
-               return gen_port((int)v, proto, dir);
-#else
-           {
-               struct block *b;
-               b = gen_port((int)v, proto, dir);
-               gen_or(gen_port6((int)v, proto, dir), b);
-               return b;
-           }
-#endif /* INET6 */
+       sappend(s1, s2);
+       sappend(s0, s1);
+       sappend(a1->s, s0);
+       sappend(a0->s, a1->s);
 
-       case Q_GATEWAY:
-               bpf_error("'gateway' requires a name");
-               /* NOTREACHED */
+       free_reg(a0->regno);
+       free_reg(a1->regno);
+
+       s0 = new_stmt(BPF_ST);
+       a0->regno = s0->s.k = alloc_reg();
+       sappend(a0->s, s0);
+
+       return a0;
+}
+
+/*
+ * Here we handle simple allocation of the scratch registers.
+ * If too many registers are alloc'd, the allocator punts.
+ */
+static int regused[BPF_MEMWORDS];
+static int curreg;
+
+/*
+ * Return the next free register.
+ */
+static int
+alloc_reg()
+{
+       int n = BPF_MEMWORDS;
+
+       while (--n >= 0) {
+               if (regused[curreg])
+                       curreg = (curreg + 1) % BPF_MEMWORDS;
+               else {
+                       regused[curreg] = 1;
+                       return curreg;
+               }
+       }
+       bpf_error("too many registers needed to evaluate expression");
+       /* NOTREACHED */
+}
+
+/*
+ * Return a register to the table so it can
+ * be used later.
+ */
+static void
+free_reg(n)
+       int n;
+{
+       regused[n] = 0;
+}
 
-       case Q_PROTO:
-               return gen_proto((int)v, proto, dir);
+static struct block *
+gen_len(jmp, n)
+       int jmp, n;
+{
+       struct slist *s;
+       struct block *b;
 
-       case Q_PROTOCHAIN:
-               return gen_protochain((int)v, proto, dir);
+       s = new_stmt(BPF_LD|BPF_LEN);
+       b = new_block(JMP(jmp));
+       b->stmts = s;
+       b->s.k = n;
 
-       case Q_UNDEF:
-               syntax();
-               /* NOTREACHED */
+       return b;
+}
 
-       default:
-               abort();
-               /* NOTREACHED */
-       }
-       /* NOTREACHED */
+struct block *
+gen_greater(n)
+       int n;
+{
+       return gen_len(BPF_JGE, n);
 }
 
-#ifdef INET6
+/*
+ * Actually, this is less than or equal.
+ */
 struct block *
-gen_mcode6(s1, s2, masklen, q)
-       register const char *s1, *s2;
-       register int masklen;
-       struct qual q;
+gen_less(n)
+       int n;
 {
-       struct addrinfo *res;
-       struct in6_addr *addr;
-       struct in6_addr mask;
        struct block *b;
-       u_int32_t *a, *m;
 
-       if (s2)
-               bpf_error("no mask %s supported", s2);
+       b = gen_len(BPF_JGT, n);
+       gen_not(b);
 
-       res = pcap_nametoaddrinfo(s1);
-       if (!res)
-               bpf_error("invalid ip6 address %s", s1);
-       if (res->ai_next)
-               bpf_error("%s resolved to multiple address", s1);
-       addr = &((struct sockaddr_in6 *)res->ai_addr)->sin6_addr;
+       return b;
+}
 
-       if (sizeof(mask) * 8 < masklen)
-               bpf_error("mask length must be <= %u", (unsigned int)(sizeof(mask) * 8));
-       memset(&mask, 0xff, masklen / 8);
-       if (masklen % 8) {
-               mask.s6_addr[masklen / 8] =
-                       (0xff << (8 - masklen % 8)) & 0xff;
-       }
+struct block *
+gen_byteop(op, idx, val)
+       int op, idx, val;
+{
+       struct block *b;
+       struct slist *s;
 
-       a = (u_int32_t *)addr;
-       m = (u_int32_t *)&mask;
-       if ((a[0] & ~m[0]) || (a[1] & ~m[1])
-        || (a[2] & ~m[2]) || (a[3] & ~m[3])) {
-               bpf_error("non-network bits set in \"%s/%d\"", s1, masklen);
-       }
+       switch (op) {
+       default:
+               abort();
 
-       switch (q.addr) {
+       case '=':
+               return gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
 
-       case Q_DEFAULT:
-       case Q_HOST:
-               if (masklen != 128)
-                       bpf_error("Mask syntax for networks only");
-               /* FALLTHROUGH */
+       case '<':
+               b = gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
+               b->s.code = JMP(BPF_JGE);
+               gen_not(b);
+               return b;
 
-       case Q_NET:
-               b = gen_host6(addr, &mask, q.proto, q.dir);
-               freeaddrinfo(res);
+       case '>':
+               b = gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
+               b->s.code = JMP(BPF_JGT);
                return b;
 
-       default:
-               bpf_error("invalid qualifier against IPv6 address");
-               /* NOTREACHED */
+       case '|':
+               s = new_stmt(BPF_ALU|BPF_OR|BPF_K);
+               break;
+
+       case '&':
+               s = new_stmt(BPF_ALU|BPF_AND|BPF_K);
+               break;
        }
+       s->s.k = val;
+       b = new_block(JMP(BPF_JEQ));
+       b->stmts = s;
+       gen_not(b);
+
+       return b;
 }
-#endif /*INET6*/
+
+static u_char abroadcast[] = { 0x0 };
 
 struct block *
-gen_ecode(eaddr, q)
-       register const u_char *eaddr;
-       struct qual q;
+gen_broadcast(proto)
+       int proto;
 {
-       if ((q.addr == Q_HOST || q.addr == Q_DEFAULT) && q.proto == Q_LINK) {
+       bpf_u_int32 hostmask;
+       struct block *b0, *b1, *b2;
+       static u_char ebroadcast[] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+
+       switch (proto) {
+
+       case Q_DEFAULT:
+       case Q_LINK:
+               if (linktype == DLT_ARCNET || linktype == DLT_ARCNET_LINUX)
+                       return gen_ahostop(abroadcast, Q_DST);
                if (linktype == DLT_EN10MB)
-                       return gen_ehostop(eaddr, (int)q.dir);
+                       return gen_ehostop(ebroadcast, Q_DST);
                if (linktype == DLT_FDDI)
-                       return gen_fhostop(eaddr, (int)q.dir);
+                       return gen_fhostop(ebroadcast, Q_DST);
+               if (linktype == DLT_IEEE802)
+                       return gen_thostop(ebroadcast, Q_DST);
+               if (linktype == DLT_IEEE802_11)
+                       return gen_wlanhostop(ebroadcast, Q_DST);
+               if (linktype == DLT_IP_OVER_FC)
+                       return gen_ipfchostop(ebroadcast, Q_DST);
+               if (linktype == DLT_SUNATM && is_lane) {
+                       /*
+                        * Check that the packet doesn't begin with an
+                        * LE Control marker.  (We've already generated
+                        * a test for LANE.)
+                        */
+                       b1 = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H, 0xFF00);
+                       gen_not(b1);
+
+                       /*
+                        * Now check the MAC address.
+                        */
+                       b0 = gen_ehostop(ebroadcast, Q_DST);
+                       gen_and(b1, b0);
+                       return b0;
+               }
+               bpf_error("not a broadcast link");
+               break;
+
+       case Q_IP:
+               b0 = gen_linktype(ETHERTYPE_IP);
+               hostmask = ~netmask;
+               b1 = gen_mcmp(off_nl + 16, BPF_W, (bpf_int32)0, hostmask);
+               b2 = gen_mcmp(off_nl + 16, BPF_W,
+                             (bpf_int32)(~0 & hostmask), hostmask);
+               gen_or(b1, b2);
+               gen_and(b0, b2);
+               return b2;
        }
-       bpf_error("ethernet address used in non-ether expression");
-       /* NOTREACHED */
+       bpf_error("only link-layer/IP broadcast filters supported");
 }
 
-void
-sappend(s0, s1)
-       struct slist *s0, *s1;
+/*
+ * Generate code to test the low-order bit of a MAC address (that's
+ * the bottom bit of the *first* byte).
+ */
+static struct block *
+gen_mac_multicast(offset)
+       int offset;
 {
-       /*
-        * This is definitely not the best way to do this, but the
-        * lists will rarely get long.
-        */
-       while (s0->next)
-               s0 = s0->next;
-       s0->next = s1;
+       register struct block *b0;
+       register struct slist *s;
+
+       /* link[offset] & 1 != 0 */
+       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+       s->s.k = offset;
+       b0 = new_block(JMP(BPF_JSET));
+       b0->s.k = 1;
+       b0->stmts = s;
+       return b0;
 }
 
-static struct slist *
-xfer_to_x(a)
-       struct arth *a;
+struct block *
+gen_multicast(proto)
+       int proto;
 {
-       struct slist *s;
+       register struct block *b0, *b1, *b2;
+       register struct slist *s;
 
-       s = new_stmt(BPF_LDX|BPF_MEM);
-       s->s.k = a->regno;
-       return s;
-}
+       switch (proto) {
 
-static struct slist *
-xfer_to_a(a)
-       struct arth *a;
-{
-       struct slist *s;
+       case Q_DEFAULT:
+       case Q_LINK:
+               if (linktype == DLT_ARCNET || linktype == DLT_ARCNET_LINUX)
+                       /* all ARCnet multicasts use the same address */
+                       return gen_ahostop(abroadcast, Q_DST);
 
-       s = new_stmt(BPF_LD|BPF_MEM);
-       s->s.k = a->regno;
-       return s;
-}
+               if (linktype == DLT_EN10MB) {
+                       /* ether[0] & 1 != 0 */
+                       return gen_mac_multicast(0);
+               }
 
-struct arth *
-gen_load(proto, index, size)
-       int proto;
-       struct arth *index;
-       int size;
-{
-       struct slist *s, *tmp;
-       struct block *b;
-       int regno = alloc_reg();
+               if (linktype == DLT_FDDI) {
+                       /*
+                        * XXX TEST THIS: MIGHT NOT PORT PROPERLY XXX
+                        *
+                        * XXX - was that referring to bit-order issues?
+                        */
+                       /* fddi[1] & 1 != 0 */
+                       return gen_mac_multicast(1);
+               }
+
+               if (linktype == DLT_IEEE802) {
+                       /* tr[2] & 1 != 0 */
+                       return gen_mac_multicast(2);
+               }
+
+               if (linktype == DLT_IEEE802_11) {
+                       /*
+                        * Oh, yuk.
+                        *
+                        *      For control frames, there is no DA.
+                        *
+                        *      For management frames, DA is at an
+                        *      offset of 4 from the beginning of
+                        *      the packet.
+                        *
+                        *      For data frames, DA is at an offset
+                        *      of 4 from the beginning of the packet
+                        *      if To DS is clear and at an offset of
+                        *      16 from the beginning of the packet
+                        *      if To DS is set.
+                        */
+
+                       /*
+                        * Generate the tests to be done for data frames.
+                        *
+                        * First, check for To DS set, i.e. "link[1] & 0x01".
+                        */
+                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+                       s->s.k = 1;
+                       b1 = new_block(JMP(BPF_JSET));
+                       b1->s.k = 0x01; /* To DS */
+                       b1->stmts = s;
+
+                       /*
+                        * If To DS is set, the DA is at 16.
+                        */
+                       b0 = gen_mac_multicast(16);
+                       gen_and(b1, b0);
+
+                       /*
+                        * Now, check for To DS not set, i.e. check
+                        * "!(link[1] & 0x01)".
+                        */
+                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+                       s->s.k = 1;
+                       b2 = new_block(JMP(BPF_JSET));
+                       b2->s.k = 0x01; /* To DS */
+                       b2->stmts = s;
+                       gen_not(b2);
+
+                       /*
+                        * If To DS is not set, the DA is at 4.
+                        */
+                       b1 = gen_mac_multicast(4);
+                       gen_and(b2, b1);
+
+                       /*
+                        * Now OR together the last two checks.  That gives
+                        * the complete set of checks for data frames.
+                        */
+                       gen_or(b1, b0);
+
+                       /*
+                        * Now check for a data frame.
+                        * I.e, check "link[0] & 0x08".
+                        */
+                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+                       s->s.k = 0;
+                       b1 = new_block(JMP(BPF_JSET));
+                       b1->s.k = 0x08;
+                       b1->stmts = s;
+
+                       /*
+                        * AND that with the checks done for data frames.
+                        */
+                       gen_and(b1, b0);
+
+                       /*
+                        * If the high-order bit of the type value is 0, this
+                        * is a management frame.
+                        * I.e, check "!(link[0] & 0x08)".
+                        */
+                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+                       s->s.k = 0;
+                       b2 = new_block(JMP(BPF_JSET));
+                       b2->s.k = 0x08;
+                       b2->stmts = s;
+                       gen_not(b2);
+
+                       /*
+                        * For management frames, the DA is at 4.
+                        */
+                       b1 = gen_mac_multicast(4);
+                       gen_and(b2, b1);
 
-       free_reg(index->regno);
-       switch (size) {
+                       /*
+                        * OR that with the checks done for data frames.
+                        * That gives the checks done for management and
+                        * data frames.
+                        */
+                       gen_or(b1, b0);
 
-       default:
-               bpf_error("data size must be 1, 2, or 4");
+                       /*
+                        * If the low-order bit of the type value is 1,
+                        * this is either a control frame or a frame
+                        * with a reserved type, and thus not a
+                        * frame with an SA.
+                        *
+                        * I.e., check "!(link[0] & 0x04)".
+                        */
+                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
+                       s->s.k = 0;
+                       b1 = new_block(JMP(BPF_JSET));
+                       b1->s.k = 0x04;
+                       b1->stmts = s;
+                       gen_not(b1);
 
-       case 1:
-               size = BPF_B;
-               break;
+                       /*
+                        * AND that with the checks for data and management
+                        * frames.
+                        */
+                       gen_and(b1, b0);
+                       return b0;
+               }
 
-       case 2:
-               size = BPF_H;
-               break;
+               if (linktype == DLT_IP_OVER_FC) {
+                       b0 = gen_mac_multicast(2);
+                       return b0;
+               }
 
-       case 4:
-               size = BPF_W;
-               break;
-       }
-       switch (proto) {
-       default:
-               bpf_error("unsupported index operation");
+               if (linktype == DLT_SUNATM && is_lane) {
+                       /*
+                        * Check that the packet doesn't begin with an
+                        * LE Control marker.  (We've already generated
+                        * a test for LANE.)
+                        */
+                       b1 = gen_cmp(SUNATM_PKT_BEGIN_POS, BPF_H, 0xFF00);
+                       gen_not(b1);
 
-       case Q_LINK:
-               s = xfer_to_x(index);
-               tmp = new_stmt(BPF_LD|BPF_IND|size);
-               sappend(s, tmp);
-               sappend(index->s, s);
+                       /* ether[off_mac] & 1 != 0 */
+                       b0 = gen_mac_multicast(off_mac);
+                       gen_and(b1, b0);
+                       return b0;
+               }
+
+               /* Link not known to support multicasts */
                break;
 
        case Q_IP:
-       case Q_ARP:
-       case Q_RARP:
-       case Q_ATALK:
-       case Q_DECNET:
-       case Q_SCA:
-       case Q_LAT:
-       case Q_MOPRC:
-       case Q_MOPDL:
+               b0 = gen_linktype(ETHERTYPE_IP);
+               b1 = gen_cmp(off_nl + 16, BPF_B, (bpf_int32)224);
+               b1->s.code = JMP(BPF_JGE);
+               gen_and(b0, b1);
+               return b1;
+
 #ifdef INET6
        case Q_IPV6:
-#endif
-               /* XXX Note that we assume a fixed link header here. */
-               s = xfer_to_x(index);
-               tmp = new_stmt(BPF_LD|BPF_IND|size);
-               tmp->s.k = off_nl;
-               sappend(s, tmp);
-               sappend(index->s, s);
+               b0 = gen_linktype(ETHERTYPE_IPV6);
+               b1 = gen_cmp(off_nl + 24, BPF_B, (bpf_int32)255);
+               gen_and(b0, b1);
+               return b1;
+#endif /* INET6 */
+       }
+       bpf_error("link-layer multicast filters supported only on ethernet/FDDI/token ring/ARCNET/802.11/ATM LANE/Fibre Channel");
+}
 
-               b = gen_proto_abbrev(proto);
-               if (index->b)
-                       gen_and(index->b, b);
-               index->b = b;
+/*
+ * generate command for inbound/outbound.  It's here so we can
+ * make it link-type specific.  'dir' = 0 implies "inbound",
+ * = 1 implies "outbound".
+ */
+struct block *
+gen_inbound(dir)
+       int dir;
+{
+       register struct block *b0;
+
+       /*
+        * Only some data link types support inbound/outbound qualifiers.
+        */
+       switch (linktype) {
+       case DLT_SLIP:
+               b0 = gen_relation(BPF_JEQ,
+                         gen_load(Q_LINK, gen_loadi(0), 1),
+                         gen_loadi(0),
+                         dir);
                break;
 
-       case Q_TCP:
-       case Q_UDP:
-       case Q_ICMP:
-       case Q_IGMP:
-       case Q_IGRP:
-       case Q_PIM:
-               s = new_stmt(BPF_LDX|BPF_MSH|BPF_B);
-               s->s.k = off_nl;
-               sappend(s, xfer_to_a(index));
-               sappend(s, new_stmt(BPF_ALU|BPF_ADD|BPF_X));
-               sappend(s, new_stmt(BPF_MISC|BPF_TAX));
-               sappend(s, tmp = new_stmt(BPF_LD|BPF_IND|size));
-               tmp->s.k = off_nl;
-               sappend(index->s, s);
+       case DLT_LINUX_SLL:
+               if (dir) {
+                       /*
+                        * Match packets sent by this machine.
+                        */
+                       b0 = gen_cmp(0, BPF_H, LINUX_SLL_OUTGOING);
+               } else {
+                       /*
+                        * Match packets sent to this machine.
+                        * (No broadcast or multicast packets, or
+                        * packets sent to some other machine and
+                        * received promiscuously.)
+                        *
+                        * XXX - packets sent to other machines probably
+                        * shouldn't be matched, but what about broadcast
+                        * or multicast packets we received?
+                        */
+                       b0 = gen_cmp(0, BPF_H, LINUX_SLL_HOST);
+               }
+               break;
 
-               gen_and(gen_proto_abbrev(proto), b = gen_ipfrag());
-               if (index->b)
-                       gen_and(index->b, b);
-#ifdef INET6
-               gen_and(gen_proto_abbrev(Q_IP), b);
-#endif
-               index->b = b;
+       case DLT_PFLOG:
+               b0 = gen_cmp(26, BPF_H,
+                   (bpf_int32)((dir == 0) ? PF_IN : PF_OUT));
                break;
-#ifdef INET6
-       case Q_ICMPV6:
-               bpf_error("IPv6 upper-layer protocol is not supported by proto[x]");
-               /*NOTREACHED*/
-#endif
-       }
-       index->regno = regno;
-       s = new_stmt(BPF_ST);
-       s->s.k = regno;
-       sappend(index->s, s);
 
-       return index;
+       default:
+               bpf_error("inbound/outbound not supported on linktype %d",
+                   linktype);
+               b0 = NULL;
+               /* NOTREACHED */
+       }
+       return (b0);
 }
 
+/* PF firewall log matched interface */
 struct block *
-gen_relation(code, a0, a1, reversed)
-       int code;
-       struct arth *a0, *a1;
-       int reversed;
+gen_pf_ifname(const char *ifname)
 {
-       struct slist *s0, *s1, *s2;
-       struct block *b, *tmp;
-
-       s0 = xfer_to_x(a1);
-       s1 = xfer_to_a(a0);
-       s2 = new_stmt(BPF_ALU|BPF_SUB|BPF_X);
-       b = new_block(JMP(code));
-       if (code == BPF_JGT || code == BPF_JGE) {
-               reversed = !reversed;
-               b->s.k = 0x80000000;
+       if (linktype != DLT_PFLOG) {
+               bpf_error("ifname supported only for DLT_PFLOG");
+               /* NOTREACHED */
        }
-       if (reversed)
-               gen_not(b);
-
-       sappend(s1, s2);
-       sappend(s0, s1);
-       sappend(a1->s, s0);
-       sappend(a0->s, a1->s);
-
-       b->stmts = a0->s;
-
-       free_reg(a0->regno);
-       free_reg(a1->regno);
+       if (strlen(ifname) >= 16) {
+               bpf_error("ifname interface names can't be larger than 16 characters");
+               /* NOTREACHED */
+       }
+       return (gen_bcmp(4, strlen(ifname), (const u_char *)ifname));
+}
 
-       /* 'and' together protocol checks */
-       if (a0->b) {
-               if (a1->b) {
-                       gen_and(a0->b, tmp = a1->b);
-               }
-               else
-                       tmp = a0->b;
-       } else
-               tmp = a1->b;
 
-       if (tmp)
-               gen_and(tmp, b);
+/* PF firewall log rule number */
+struct block *
+gen_pf_rnr(int rnr)
+{
+       if (linktype != DLT_PFLOG) {
+               bpf_error("rnr supported only for DLT_PFLOG");
+               /* NOTREACHED */
+       }
 
-       return b;
+       return (gen_cmp(20, BPF_H, (bpf_int32)rnr));
 }
 
-struct arth *
-gen_loadlen()
+/* PF firewall log reason code */
+struct block *
+gen_pf_reason(int reason)
 {
-       int regno = alloc_reg();
-       struct arth *a = (struct arth *)newchunk(sizeof(*a));
-       struct slist *s;
-
-       s = new_stmt(BPF_LD|BPF_LEN);
-       s->next = new_stmt(BPF_ST);
-       s->next->s.k = regno;
-       a->s = s;
-       a->regno = regno;
+       if (linktype != DLT_PFLOG) {
+               bpf_error("reason supported only for DLT_PFLOG");
+               /* NOTREACHED */
+       }
 
-       return a;
+       return (gen_cmp(22, BPF_H, (bpf_int32)reason));
 }
 
-struct arth *
-gen_loadi(val)
-       int val;
+/* PF firewall log action */
+struct block *
+gen_pf_action(int action)
 {
-       struct arth *a;
-       struct slist *s;
-       int reg;
-
-       a = (struct arth *)newchunk(sizeof(*a));
-
-       reg = alloc_reg();
-
-       s = new_stmt(BPF_LD|BPF_IMM);
-       s->s.k = val;
-       s->next = new_stmt(BPF_ST);
-       s->next->s.k = reg;
-       a->s = s;
-       a->regno = reg;
+       if (linktype != DLT_PFLOG) {
+               bpf_error("action supported only for DLT_PFLOG");
+               /* NOTREACHED */
+       }
 
-       return a;
+       return (gen_cmp(24, BPF_H, (bpf_int32)action));
 }
 
-struct arth *
-gen_neg(a)
-       struct arth *a;
+struct block *
+gen_acode(eaddr, q)
+       register const u_char *eaddr;
+       struct qual q;
 {
-       struct slist *s;
-
-       s = xfer_to_a(a);
-       sappend(a->s, s);
-       s = new_stmt(BPF_ALU|BPF_NEG);
-       s->s.k = 0;
-       sappend(a->s, s);
-       s = new_stmt(BPF_ST);
-       s->s.k = a->regno;
-       sappend(a->s, s);
-
-       return a;
+       if ((q.addr == Q_HOST || q.addr == Q_DEFAULT) && q.proto == Q_LINK) {
+               if (linktype == DLT_ARCNET || linktype == DLT_ARCNET_LINUX)
+                       return gen_ahostop(eaddr, (int)q.dir);
+       }
+       bpf_error("ARCnet address used in non-arc expression");
+       /* NOTREACHED */
 }
 
-struct arth *
-gen_arth(code, a0, a1)
-       int code;
-       struct arth *a0, *a1;
+static struct block *
+gen_ahostop(eaddr, dir)
+       register const u_char *eaddr;
+       register int dir;
 {
-       struct slist *s0, *s1, *s2;
-
-       s0 = xfer_to_x(a1);
-       s1 = xfer_to_a(a0);
-       s2 = new_stmt(BPF_ALU|BPF_X|code);
-
-       sappend(s1, s2);
-       sappend(s0, s1);
-       sappend(a1->s, s0);
-       sappend(a0->s, a1->s);
-
-       free_reg(a1->regno);
-
-       s0 = new_stmt(BPF_ST);
-       a0->regno = s0->s.k = alloc_reg();
-       sappend(a0->s, s0);
+       register struct block *b0, *b1;
 
-       return a0;
-}
+       switch (dir) {
+       /* src comes first, different from Ethernet */
+       case Q_SRC:
+               return gen_bcmp(0, 1, eaddr);
 
-/*
- * Here we handle simple allocation of the scratch registers.
- * If too many registers are alloc'd, the allocator punts.
- */
-static int regused[BPF_MEMWORDS];
-static int curreg;
+       case Q_DST:
+               return gen_bcmp(1, 1, eaddr);
 
-/*
- * Return the next free register.
- */
-static int
-alloc_reg()
-{
-       int n = BPF_MEMWORDS;
+       case Q_AND:
+               b0 = gen_ahostop(eaddr, Q_SRC);
+               b1 = gen_ahostop(eaddr, Q_DST);
+               gen_and(b0, b1);
+               return b1;
 
-       while (--n >= 0) {
-               if (regused[curreg])
-                       curreg = (curreg + 1) % BPF_MEMWORDS;
-               else {
-                       regused[curreg] = 1;
-                       return curreg;
-               }
+       case Q_DEFAULT:
+       case Q_OR:
+               b0 = gen_ahostop(eaddr, Q_SRC);
+               b1 = gen_ahostop(eaddr, Q_DST);
+               gen_or(b0, b1);
+               return b1;
        }
-       bpf_error("too many registers needed to evaluate expression");
+       abort();
        /* NOTREACHED */
 }
 
 /*
- * Return a register to the table so it can
- * be used later.
+ * support IEEE 802.1Q VLAN trunk over ethernet
  */
-static void
-free_reg(n)
-       int n;
+struct block *
+gen_vlan(vlan_num)
+       int vlan_num;
 {
-       regused[n] = 0;
-}
+       struct  block   *b0;
 
-static struct block *
-gen_len(jmp, n)
-       int jmp, n;
-{
-       struct slist *s;
-       struct block *b;
+       /*
+        * Change the offsets to point to the type and data fields within
+        * the VLAN packet.  This is somewhat of a kludge.
+        */
+       if (orig_nl == (u_int)-1) {
+               orig_linktype = off_linktype;   /* save original values */
+               orig_nl = off_nl;
+               orig_nl_nosnap = off_nl_nosnap;
 
-       s = new_stmt(BPF_LD|BPF_LEN);
-       b = new_block(JMP(jmp));
-       b->stmts = s;
-       b->s.k = n;
+               switch (linktype) {
 
-       return b;
-}
+               case DLT_EN10MB:
+                       off_linktype = 16;
+                       off_nl_nosnap = 18;
+                       off_nl = 18;
+                       break;
 
-struct block *
-gen_greater(n)
-       int n;
-{
-       return gen_len(BPF_JGE, n);
-}
+               default:
+                       bpf_error("no VLAN support for data link type %d",
+                                 linktype);
+                       /*NOTREACHED*/
+               }
+       }
 
-struct block *
-gen_less(n)
-       int n;
-{
-       struct block *b;
+       /* check for VLAN */
+       b0 = gen_cmp(orig_linktype, BPF_H, (bpf_int32)ETHERTYPE_8021Q);
 
-       b = gen_len(BPF_JGT, n);
-       gen_not(b);
+       /* If a specific VLAN is requested, check VLAN id */
+       if (vlan_num >= 0) {
+               struct block *b1;
 
-       return b;
+               b1 = gen_cmp(orig_nl, BPF_H, (bpf_int32)vlan_num);
+               gen_and(b0, b1);
+               b0 = b1;
+       }
+
+       return (b0);
 }
 
 struct block *
-gen_byteop(op, idx, val)
-       int op, idx, val;
+gen_atmfield_code(atmfield, jvalue, jtype, reverse)
+       int atmfield;
+       bpf_u_int32 jvalue;
+       bpf_u_int32 jtype;
+       int reverse;
 {
-       struct block *b;
-       struct slist *s;
+       struct block *b0;
 
-       switch (op) {
-       default:
-               abort();
+       switch (atmfield) {
 
-       case '=':
-               return gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
+       case A_VPI:
+               if (!is_atm)
+                       bpf_error("'vpi' supported only on raw ATM");
+               if (off_vpi == -1)
+                       abort();
+               b0 = gen_ncmp(BPF_B, off_vpi, 0xffffffff, (u_int)jtype,
+                   (u_int)jvalue, reverse);
+               break;
 
-       case '<':
-               b = gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
-               b->s.code = JMP(BPF_JGE);
-               gen_not(b);
-               return b;
+       case A_VCI:
+               if (!is_atm)
+                       bpf_error("'vci' supported only on raw ATM");
+               if (off_vci == -1)
+                       abort();
+               b0 = gen_ncmp(BPF_H, off_vci, 0xffffffff, (u_int)jtype,
+                   (u_int)jvalue, reverse);
+               break;
 
-       case '>':
-               b = gen_cmp((u_int)idx, BPF_B, (bpf_int32)val);
-               b->s.code = JMP(BPF_JGT);
-               return b;
+       case A_PROTOTYPE:
+               if (off_proto == -1)
+                       abort();        /* XXX - this isn't on FreeBSD */
+               b0 = gen_ncmp(BPF_B, off_proto, 0x0f, (u_int)jtype,
+                   (u_int)jvalue, reverse);
+               break;
 
-       case '|':
-               s = new_stmt(BPF_ALU|BPF_OR|BPF_K);
+       case A_MSGTYPE:
+               if (off_payload == -1)
+                       abort();
+               b0 = gen_ncmp(BPF_B, off_payload + MSG_TYPE_POS, 0xffffffff,
+                   (u_int)jtype, (u_int)jvalue, reverse);
                break;
 
-       case '&':
-               s = new_stmt(BPF_ALU|BPF_AND|BPF_K);
+       case A_CALLREFTYPE:
+               if (!is_atm)
+                       bpf_error("'callref' supported only on raw ATM");
+               if (off_proto == -1)
+                       abort();
+               b0 = gen_ncmp(BPF_B, off_proto, 0xffffffff, (u_int)jtype,
+                   (u_int)jvalue, reverse);
                break;
-       }
-       s->s.k = val;
-       b = new_block(JMP(BPF_JEQ));
-       b->stmts = s;
-       gen_not(b);
 
-       return b;
+       default:
+               abort();
+       }
+       return b0;
 }
 
 struct block *
-gen_broadcast(proto)
-       int proto;
+gen_atmtype_abbrev(type)
+       int type;
 {
-       bpf_u_int32 hostmask;
-       struct block *b0, *b1, *b2;
-       static u_char ebroadcast[] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
+       struct block *b0, *b1;
 
-       switch (proto) {
+       switch (type) {
 
-       case Q_DEFAULT:
-       case Q_LINK:
-               if (linktype == DLT_EN10MB)
-                       return gen_ehostop(ebroadcast, Q_DST);
-               if (linktype == DLT_FDDI)
-                       return gen_fhostop(ebroadcast, Q_DST);
-               bpf_error("not a broadcast link");
+       case A_METAC:
+               /* Get all packets in Meta signalling Circuit */
+               if (!is_atm)
+                       bpf_error("'metac' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 1, BPF_JEQ, 0);
+               gen_and(b0, b1);
                break;
 
-       case Q_IP:
-               b0 = gen_linktype(ETHERTYPE_IP);
-               hostmask = ~netmask;
-               b1 = gen_mcmp(off_nl + 16, BPF_W, (bpf_int32)0, hostmask);
-               b2 = gen_mcmp(off_nl + 16, BPF_W,
-                             (bpf_int32)(~0 & hostmask), hostmask);
-               gen_or(b1, b2);
-               gen_and(b0, b2);
-               return b2;
+       case A_BCC:
+               /* Get all packets in Broadcast Circuit*/
+               if (!is_atm)
+                       bpf_error("'bcc' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 2, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_OAMF4SC:
+               /* Get all cells in Segment OAM F4 circuit*/
+               if (!is_atm)
+                       bpf_error("'oam4sc' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 3, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_OAMF4EC:
+               /* Get all cells in End-to-End OAM F4 Circuit*/
+               if (!is_atm)
+                       bpf_error("'oam4ec' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 4, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_SC:
+               /*  Get all packets in connection Signalling Circuit */
+               if (!is_atm)
+                       bpf_error("'sc' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 5, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_ILMIC:
+               /* Get all packets in ILMI Circuit */
+               if (!is_atm)
+                       bpf_error("'ilmic' supported only on raw ATM");
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               b1 = gen_atmfield_code(A_VCI, 16, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_LANE:
+               /* Get all LANE packets */
+               if (!is_atm)
+                       bpf_error("'lane' supported only on raw ATM");
+               b1 = gen_atmfield_code(A_PROTOTYPE, PT_LANE, BPF_JEQ, 0);
+
+               /*
+                * Arrange that all subsequent tests assume LANE
+                * rather than LLC-encapsulated packets, and set
+                * the offsets appropriately for LANE-encapsulated
+                * Ethernet.
+                *
+                * "off_mac" is the offset of the Ethernet header,
+                * which is 2 bytes past the ATM pseudo-header
+                * (skipping the pseudo-header and 2-byte LE Client
+                * field).  The other offsets are Ethernet offsets
+                * relative to "off_mac".
+                */
+               is_lane = 1;
+               off_mac = off_payload + 2;      /* MAC header */
+               off_linktype = off_mac + 12;
+               off_nl = off_mac + 14;          /* Ethernet II */
+               off_nl_nosnap = off_mac + 17;   /* 802.3+802.2 */
+               break;
+
+       case A_LLC:
+               /* Get all LLC-encapsulated packets */
+               if (!is_atm)
+                       bpf_error("'llc' supported only on raw ATM");
+               b1 = gen_atmfield_code(A_PROTOTYPE, PT_LLC, BPF_JEQ, 0);
+               is_lane = 0;
+               break;
+
+       default:
+               abort();
        }
-       bpf_error("only ether/ip broadcast filters supported");
+       return b1;
 }
 
-struct block *
-gen_multicast(proto)
-       int proto;
+
+static struct block *
+gen_msg_abbrev(type)
+       int type;
 {
-       register struct block *b0, *b1;
-       register struct slist *s;
+       struct block *b1;
 
-       switch (proto) {
+       /*
+        * Q.2931 signalling protocol messages for handling virtual circuits
+        * establishment and teardown
+        */
+       switch (type) {
 
-       case Q_DEFAULT:
-       case Q_LINK:
-               if (linktype == DLT_EN10MB) {
-                       /* ether[0] & 1 != 0 */
-                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
-                       s->s.k = 0;
-                       b0 = new_block(JMP(BPF_JSET));
-                       b0->s.k = 1;
-                       b0->stmts = s;
-                       return b0;
-               }
+       case A_SETUP:
+               b1 = gen_atmfield_code(A_MSGTYPE, SETUP, BPF_JEQ, 0);
+               break;
 
-               if (linktype == DLT_FDDI) {
-                       /* XXX TEST THIS: MIGHT NOT PORT PROPERLY XXX */
-                       /* fddi[1] & 1 != 0 */
-                       s = new_stmt(BPF_LD|BPF_B|BPF_ABS);
-                       s->s.k = 1;
-                       b0 = new_block(JMP(BPF_JSET));
-                       b0->s.k = 1;
-                       b0->stmts = s;
-                       return b0;
-               }
-               /* Link not known to support multicasts */
+       case A_CALLPROCEED:
+               b1 = gen_atmfield_code(A_MSGTYPE, CALL_PROCEED, BPF_JEQ, 0); 
                break;
 
-       case Q_IP:
-               b0 = gen_linktype(ETHERTYPE_IP);
-               b1 = gen_cmp(off_nl + 16, BPF_B, (bpf_int32)224);
-               b1->s.code = JMP(BPF_JGE);
-               gen_and(b0, b1);
-               return b1;
+       case A_CONNECT:
+               b1 = gen_atmfield_code(A_MSGTYPE, CONNECT, BPF_JEQ, 0);
+               break;                 
 
-#ifdef INET6
-       case Q_IPV6:
-               b0 = gen_linktype(ETHERTYPE_IPV6);
-               b1 = gen_cmp(off_nl + 24, BPF_B, (bpf_int32)255);
-               gen_and(b0, b1);
-               return b1;
-#endif /* INET6 */
+       case A_CONNECTACK:
+               b1 = gen_atmfield_code(A_MSGTYPE, CONNECT_ACK, BPF_JEQ, 0);  
+               break;
+
+       case A_RELEASE:
+               b1 = gen_atmfield_code(A_MSGTYPE, RELEASE, BPF_JEQ, 0);
+               break;
+
+       case A_RELEASE_DONE:
+               b1 = gen_atmfield_code(A_MSGTYPE, RELEASE_DONE, BPF_JEQ, 0);  
+               break;
+
+       default:
+               abort();
        }
-       bpf_error("only IP multicast filters supported on ethernet/FDDI");
+       return b1;
 }
 
-/*
- * generate command for inbound/outbound.  It's here so we can
- * make it link-type specific.  'dir' = 0 implies "inbound",
- * = 1 implies "outbound".
- */
 struct block *
-gen_inbound(dir)
-       int dir;
+gen_atmmulti_abbrev(type)
+       int type;
 {
-       register struct block *b0;
+       struct block *b0, *b1;
 
-       b0 = gen_relation(BPF_JEQ,
-                         gen_load(Q_LINK, gen_loadi(0), 1),
-                         gen_loadi(0),
-                         dir);
-       return (b0);
+       switch (type) {
+
+       case A_OAM:
+               if (!is_atm)
+                       bpf_error("'oam' supported only on raw ATM");
+               b1 = gen_atmmulti_abbrev(A_OAMF4);
+               break;
+
+       case A_OAMF4:
+               if (!is_atm)
+                       bpf_error("'oamf4' supported only on raw ATM");
+               /* OAM F4 type */
+               b0 = gen_atmfield_code(A_VCI, 3, BPF_JEQ, 0); 
+               b1 = gen_atmfield_code(A_VCI, 4, BPF_JEQ, 0);
+               gen_or(b0, b1); 
+               b0 = gen_atmfield_code(A_VPI, 0, BPF_JEQ, 0);
+               gen_and(b0, b1);
+               break;
+
+       case A_CONNECTMSG:
+               /*
+                * Get Q.2931 signalling messages for switched
+                * virtual connection
+                */
+               if (!is_atm)
+                       bpf_error("'connectmsg' supported only on raw ATM");
+               b0 = gen_msg_abbrev(A_SETUP);
+               b1 = gen_msg_abbrev(A_CALLPROCEED);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_CONNECT);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_CONNECTACK);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_RELEASE);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_RELEASE_DONE);
+               gen_or(b0, b1);
+               b0 = gen_atmtype_abbrev(A_SC);
+               gen_and(b0, b1);
+               break;
+
+       case A_METACONNECT:
+               if (!is_atm)
+                       bpf_error("'metaconnect' supported only on raw ATM");
+               b0 = gen_msg_abbrev(A_SETUP);
+               b1 = gen_msg_abbrev(A_CALLPROCEED);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_CONNECT);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_RELEASE);
+               gen_or(b0, b1);
+               b0 = gen_msg_abbrev(A_RELEASE_DONE);
+               gen_or(b0, b1);
+               b0 = gen_atmtype_abbrev(A_METAC);
+               gen_and(b0, b1);
+               break;
+
+       default:
+               abort();
+       }
+       return b1;
 }