]> The Tcpdump Group git mirrors - libpcap/blobdiff - pcap-linux.c
Another change from Debian.
[libpcap] / pcap-linux.c
index e96c0479190d3b2e536f66d4787466c974f9c6be..e14de811ed3e49691a2c49afff1cc8429f9112b0 100644 (file)
@@ -34,7 +34,7 @@
 
 #ifndef lint
 static const char rcsid[] _U_ =
-    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.129.2.23 2008-08-06 07:45:59 guy Exp $ (LBL)";
+    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.129.2.35 2008-12-14 22:00:04 guy Exp $ (LBL)";
 #endif
 
 /*
@@ -93,7 +93,7 @@ static const char rcsid[] _U_ =
 #include <sys/ioctl.h>
 #include <sys/utsname.h>
 #include <sys/mman.h>
-#include <net/if.h>
+#include <linux/if.h>
 #include <netinet/in.h>
 #include <linux/if_ether.h>
 #include <net/if_arp.h>
@@ -177,6 +177,11 @@ static const char rcsid[] _U_ =
   * uses many ring related structs and macros */
 # ifdef TPACKET_HDRLEN
 #  define HAVE_PACKET_RING
+#  ifdef TPACKET2_HDRLEN
+#   define HAVE_TPACKET2
+#  else
+#   define TPACKET_V1  0
+#  endif /* TPACKET2_HDRLEN */
 # endif /* TPACKET_HDRLEN */
 #endif /* PF_PACKET */
 
@@ -240,11 +245,18 @@ static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
 static void pcap_cleanup_linux(pcap_t *);
 
+union thdr {
+       struct tpacket_hdr      *h1;
+       struct tpacket2_hdr     *h2;
+       void                    *raw;
+};
+
 #ifdef HAVE_PACKET_RING
-#define RING_GET_FRAME(h) (((struct tpacket_hdr**)h->buffer)[h->offset])
+#define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
 
 static void destroy_ring(pcap_t *handle);
 static int create_ring(pcap_t *handle);
+static int prepare_tpacket_socket(pcap_t *handle);
 static void pcap_cleanup_linux_mmap(pcap_t *);
 static int pcap_read_linux_mmap(pcap_t *, int, pcap_handler , u_char *);
 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *);
@@ -262,10 +274,12 @@ static int        iface_get_mtu(int fd, const char *device, char *ebuf);
 static int     iface_get_arptype(int fd, const char *device, char *ebuf);
 #ifdef HAVE_PF_PACKET_SOCKETS
 static int     iface_bind(int fd, int ifindex, char *ebuf);
+#ifdef IW_MODE_MONITOR
 static int     has_wext(int sock_fd, const char *device, char *ebuf);
+#endif /* IW_MODE_MONITOR */
 static int     enter_rfmon_mode_wext(pcap_t *handle, int sock_fd,
     const char *device);
-#endif
+#endif /* HAVE_PF_PACKET_SOCKETS */
 static int     iface_bind_old(int fd, const char *device, char *ebuf);
 
 #ifdef SO_ATTACH_FILTER
@@ -285,6 +299,12 @@ pcap_create(const char *device, char *ebuf)
 {
        pcap_t *handle;
 
+       /*
+        * A null device name is equivalent to the "any" device.
+        */
+       if (device == NULL)
+               device = "any";
+
 #ifdef HAVE_DAG_API
        if (strstr(device, "dag")) {
                return dag_create(device, ebuf);
@@ -326,10 +346,9 @@ pcap_can_set_rfmon_linux(pcap_t *p)
        struct iwreq ireq;
 #endif
 
-       if (p->opt.source == NULL) {
+       if (strcmp(p->opt.source, "any") == 0) {
                /*
-                * This is equivalent to the "any" device, and we don't
-                * support monitor mode on it.
+                * Monitor mode makes no sense on the "any" device.
                 */
                return 0;
        }
@@ -491,7 +510,6 @@ pcap_activate_linux(pcap_t *handle)
 {
        const char      *device;
        int             status = 0;
-       int             activate_ok = 0;
 
        device = handle->opt.source;
 
@@ -506,12 +524,11 @@ pcap_activate_linux(pcap_t *handle)
        handle->stats_op = pcap_stats_linux;
 
        /*
-        * NULL and "any" are special devices which give us the hint to
-        * monitor all devices.
+        * The "any" device is a special device which causes us not
+        * to bind to a particular device and thus to look at all
+        * devices.
         */
-       if (!device || strcmp(device, "any") == 0) {
-               device                  = NULL;
-               handle->md.device       = strdup("any");
+       if (strcmp(device, "any") == 0) {
                if (handle->opt.promisc) {
                        handle->opt.promisc = 0;
                        /* Just a warning. */
@@ -519,10 +536,9 @@ pcap_activate_linux(pcap_t *handle)
                            "Promiscuous mode not supported on the \"any\" device");
                        status = PCAP_WARNING_PROMISC_NOTSUP;
                }
+       }
 
-       } else
-               handle->md.device       = strdup(device);
-
+       handle->md.device       = strdup(device);
        if (handle->md.device == NULL) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
                         pcap_strerror(errno) );
@@ -540,28 +556,57 @@ pcap_activate_linux(pcap_t *handle)
         */
 
        if ((status = activate_new(handle)) == 1) {
-               activate_ok = 1;
                /*
+                * Success.
                 * Try to use memory-mapped access.
                 */
-               if (activate_mmap(handle) == 1)
-                       return 0;       /* we succeeded; nothing more to do */
+               switch (activate_mmap(handle)) {
+
+               case 1:
+                       /* we succeeded; nothing more to do */
+                       return 0;
+
+               case 0:
+                       /*
+                        * Kernel doesn't support it - just continue
+                        * with non-memory-mapped access.
+                        */
+                       break;
+
+               case -1:
+                       /*
+                        * We failed to set up to use it, or kernel
+                        * supports it, but we failed to enable it;
+                        * return an error.  handle->errbuf contains
+                        * an error message.
+                        */
+                       status = PCAP_ERROR;
+                       goto fail;
+               }
        }
        else if (status == 0) {
                /* Non-fatal error; try old way */
-               if ((status = activate_old(handle)) == 1)
-                       activate_ok = 1;
-       }
-       if (!activate_ok) {
+               if ((status = activate_old(handle)) != 1) {
+                       /*
+                        * Both methods to open the packet socket failed.
+                        * Tidy up and report our failure (handle->errbuf
+                        * is expected to be set by the functions above).
+                        */
+                       goto fail;
+               }
+       } else {
                /*
-                * Both methods to open the packet socket failed. Tidy
-                * up and report our failure (ebuf is expected to be
-                * set by the functions above).
+                * Fatal error with the new way; just fail.
+                * status has the error return; if it's PCAP_ERROR,
+                * handle->errbuf has been set appropriately.
                 */
                goto fail;
        }
 
-       if (handle->opt.buffer_size == 0) {
+       /*
+        * We set up the socket, but not with memory-mapped access.
+        */
+       if (handle->opt.buffer_size != 0) {
                /*
                 * Set the socket buffer size to the specified value.
                 */
@@ -629,8 +674,7 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
 #else
        struct sockaddr         from;
 #endif
-       int                     packet_len, caplen;
-       struct pcap_pkthdr      pcap_header;
+#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
        struct iovec            iov;
        struct msghdr           msg;
        struct cmsghdr          *cmsg;
@@ -638,6 +682,12 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                struct cmsghdr  cmsg;
                char            buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))];
        } cmsg_buf;
+#else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
+       socklen_t               fromlen;
+#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
+       int                     packet_len, caplen;
+       struct pcap_pkthdr      pcap_header;
+
 #ifdef HAVE_PF_PACKET_SOCKETS
        /*
         * If this is a cooked device, leave extra room for a
@@ -672,6 +722,7 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
         */
        bp = handle->buffer + handle->offset;
 
+#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
        msg.msg_name            = &from;
        msg.msg_namelen         = sizeof(from);
        msg.msg_iov             = &iov;
@@ -682,6 +733,7 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
 
        iov.iov_len             = handle->bufsize - offset;
        iov.iov_base            = bp + offset;
+#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
 
        do {
                /*
@@ -697,7 +749,15 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                        return -2;
                }
 
+#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
                packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC);
+#else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
+               fromlen = sizeof(from);
+               packet_len = recvfrom(
+                       handle->fd, bp + offset,
+                       handle->bufsize - offset, MSG_TRUNC,
+                       (struct sockaddr *) &from, &fromlen);
+#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
        } while (packet_len == -1 && (errno == EINTR || errno == ENETDOWN));
 
        /* Check if an error occured */
@@ -784,7 +844,7 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                hdrp->sll_protocol = from.sll_protocol;
        }
 
-#ifdef HAVE_PACKET_AUXDATA
+#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
        for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
                struct tpacket_auxdata *aux;
                unsigned int len;
@@ -812,8 +872,8 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
 
                packet_len += VLAN_TAG_LEN;
        }
-#endif /* HAVE_PACKET_AUXDATA */
-#endif
+#endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */
+#endif /* HAVE_PF_PACKET_SOCKETS */
 
        /*
         * XXX: According to the kernel source we should get the real
@@ -1630,19 +1690,21 @@ static int
 activate_new(pcap_t *handle)
 {
 #ifdef HAVE_PF_PACKET_SOCKETS
+       const char              *device = handle->opt.source;
+       int                     is_any_device = (strcmp(device, "any") == 0);
        int                     sock_fd = -1, arptype, val;
        int                     err = 0;
        struct packet_mreq      mr;
-       const char* device = handle->opt.source;
 
        /*
-        * Open a socket with protocol family packet. If a device is
-        * given we try to open it in raw mode otherwise we use
-        * the cooked interface.
+        * Open a socket with protocol family packet. If the
+        * "any" device was specified, we open a SOCK_DGRAM
+        * socket for the cooked interface, otherwise we first
+        * try a SOCK_RAW socket for the raw interface.
         */
-       sock_fd = device ?
-               socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))
-             : socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL));
+       sock_fd = is_any_device ?
+               socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
+               socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 
        if (sock_fd == -1) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
@@ -1677,7 +1739,7 @@ activate_new(pcap_t *handle)
         * to cooked mode if we have an unknown interface type
         * or a type we know doesn't work well in raw mode.
         */
-       if (device) {
+       if (!is_any_device) {
                /* Assume for now we don't need cooked mode. */
                handle->md.cooked = 0;
 
@@ -1792,15 +1854,23 @@ activate_new(pcap_t *handle)
                }
        } else {
                /*
-                * This is cooked mode.
+                * The "any" device.
+                */
+               if (handle->opt.rfmon) {
+                       /*
+                        * It doesn't support monitor mode.
+                        */
+                       return PCAP_ERROR_RFMON_NOTSUP;
+               }
+
+               /*
+                * It uses cooked mode.
                 */
                handle->md.cooked = 1;
                handle->linktype = DLT_LINUX_SLL;
 
                /*
                 * We're not bound to a device.
-                * XXX - true?  Or true only if we're using
-                * the "any" device?
                 * For now, we're using this as an indication
                 * that we can't transmit; stop doing that only
                 * if we figure out how to transmit in cooked
@@ -1825,10 +1895,13 @@ activate_new(pcap_t *handle)
 
        /*
         * Hmm, how can we set promiscuous mode on all interfaces?
-        * I am not sure if that is possible at all.
+        * I am not sure if that is possible at all.  For now, we
+        * silently ignore attempts to turn promiscuous mode on
+        * for the "any" device (so you don't have to explicitly
+        * disable it in programs such as tcpdump).
         */
 
-       if (device && handle->opt.promisc) {
+       if (!is_any_device && handle->opt.promisc) {
                memset(&mr, 0, sizeof(mr));
                mr.mr_ifindex = handle->md.ifindex;
                mr.mr_type    = PACKET_MR_PROMISC;
@@ -1897,8 +1970,11 @@ activate_mmap(pcap_t *handle)
                /* by default request 2M for the ring buffer */
                handle->opt.buffer_size = 2*1024*1024;
        }
+       ret = prepare_tpacket_socket(handle);
+       if (ret != 1)
+               return ret;
        ret = create_ring(handle);
-       if (ret == 0)
+       if (ret != 1)
                return ret;
 
        /* override some defaults and inherit the other fields from
@@ -1918,6 +1994,56 @@ activate_mmap(pcap_t *handle)
 }
 
 #ifdef HAVE_PACKET_RING
+static int
+prepare_tpacket_socket(pcap_t *handle)
+{
+#ifdef HAVE_TPACKET2
+       socklen_t len;
+       int val;
+#endif
+
+       handle->md.tp_version = TPACKET_V1;
+       handle->md.tp_hdrlen = sizeof(struct tpacket_hdr);
+
+#ifdef HAVE_TPACKET2
+       /* Probe whether kernel supports TPACKET_V2 */
+       val = TPACKET_V2;
+       len = sizeof(val);
+       if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
+               if (errno == ENOPROTOOPT)
+                       return 1;       /* no - just drive on */
+
+               /* Yes - treat as a failure. */
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't get TPACKET_V2 header len on packet socket: %s",
+                   pcap_strerror(errno));
+               return -1;
+       }
+       handle->md.tp_hdrlen = val;
+
+       val = TPACKET_V2;
+       if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
+                      sizeof(val)) < 0) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't activate TPACKET_V2 on packet socket: %s",
+                   pcap_strerror(errno));
+               return -1;
+       }
+       handle->md.tp_version = TPACKET_V2;
+
+       /* Reserve space for VLAN tag reconstruction */
+       val = VLAN_TAG_LEN;
+       if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val,
+                      sizeof(val)) < 0) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't set up reserve on packet socket: %s",
+                   pcap_strerror(errno));
+               return -1;
+       }
+
+#endif /* HAVE_TPACKET2 */
+       return 1;
+}
 
 static void
 compute_ring_block(int frame_size, unsigned *block_size, unsigned *frames_per_block)
@@ -1944,7 +2070,9 @@ create_ring(pcap_t *handle)
         * (and a lot of memory will be unused). 
         * The snap len should be carefully chosen to achive best
         * performance */
-       req.tp_frame_size = TPACKET_ALIGN(handle->snapshot+TPACKET_HDRLEN);
+       req.tp_frame_size = TPACKET_ALIGN(handle->snapshot +
+                                         TPACKET_ALIGN(handle->md.tp_hdrlen) +
+                                         sizeof(struct sockaddr_ll));
        req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
        compute_ring_block(req.tp_frame_size, &req.tp_block_size, &frames_per_block);
        req.tp_block_nr = req.tp_frame_nr / frames_per_block;
@@ -1962,10 +2090,16 @@ retry:
                        req.tp_block_nr = req.tp_frame_nr/frames_per_block;
                        goto retry;
                }
-               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "can't create rx ring on "
-                               "packet socket %d: %d-%s", handle->fd, errno, 
-                               pcap_strerror(errno));
-               return 0;
+               if (errno == ENOPROTOOPT) {
+                       /*
+                        * We don't have ring buffer support in this kernel.
+                        */
+                       return 0;
+               }
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't create rx ring on packet socket: %s",
+                   pcap_strerror(errno));
+               return -1;
        }
 
        /* memory map the rx ring */
@@ -1973,28 +2107,32 @@ retry:
        handle->bp = mmap(0, ringsize, PROT_READ| PROT_WRITE, MAP_SHARED, 
                                        handle->fd, 0);
        if (handle->bp == MAP_FAILED) {
-               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "can't mmap rx ring: %d-%s",
-                       errno, pcap_strerror(errno));
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't mmap rx ring: %s", pcap_strerror(errno));
 
                /* clear the allocated ring on error*/
                destroy_ring(handle);
-               return 0;
+               return -1;
        }
 
        /* allocate a ring for each frame header pointer*/
        handle->cc = req.tp_frame_nr;
-       handle->buffer = malloc(handle->cc * sizeof(struct tpacket_hdr*));
+       handle->buffer = malloc(handle->cc * sizeof(union thdr *));
        if (!handle->buffer) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't allocate ring of frame headers: %s",
+                   pcap_strerror(errno));
+
                destroy_ring(handle);
-               return 0;
+               return -1;
        }
 
        /* fill the header ring with proper frame ptr*/
        handle->offset = 0;
        for (i=0; i<req.tp_block_nr; ++i) {
-               u_char *base = &handle->bp[i*req.tp_block_size];
+               void *base = &handle->bp[i*req.tp_block_size];
                for (j=0; j<frames_per_block; ++j, ++handle->offset) {
-                       RING_GET_FRAME(handle) = (struct tpacket_hdr*) base;
+                       RING_GET_FRAME(handle) = base;
                        base += req.tp_frame_size;
                }
        }
@@ -2055,6 +2193,29 @@ pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
        return 0;
 }
 
+static inline union thdr *
+pcap_get_ring_frame(pcap_t *handle, int status)
+{
+       union thdr h;
+
+       h.raw = RING_GET_FRAME(handle);
+       switch (handle->md.tp_version) {
+       case TPACKET_V1:
+               if (status != (h.h1->tp_status ? TP_STATUS_USER :
+                                               TP_STATUS_KERNEL))
+                       return NULL;
+               break;
+#ifdef HAVE_TPACKET2
+       case TPACKET_V2:
+               if (status != (h.h2->tp_status ? TP_STATUS_USER :
+                                               TP_STATUS_KERNEL))
+                       return NULL;
+               break;
+#endif
+       }
+       return h.raw;
+}
+
 static int
 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback, 
                u_char *user)
@@ -2062,7 +2223,8 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
        int pkts = 0;
 
        /* wait for frames availability.*/
-       if ((handle->md.timeout >= 0) && !(RING_GET_FRAME(handle)->tp_status)) {
+       if ((handle->md.timeout >= 0) &&
+           !pcap_get_ring_frame(handle, TP_STATUS_USER)) {
                struct pollfd pollinfo;
                int ret;
 
@@ -2094,16 +2256,46 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                struct sockaddr_ll *sll;
                struct pcap_pkthdr pcaphdr;
                unsigned char *bp;
-               struct tpacket_hdr* thdr = RING_GET_FRAME(handle);
-               if (thdr->tp_status == TP_STATUS_KERNEL)
+               union thdr h;
+               unsigned int tp_len;
+               unsigned int tp_mac;
+               unsigned int tp_snaplen;
+               unsigned int tp_sec;
+               unsigned int tp_usec;
+
+               h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
+               if (!h.raw)
                        break;
 
+               switch (handle->md.tp_version) {
+               case TPACKET_V1:
+                       tp_len     = h.h1->tp_len;
+                       tp_mac     = h.h1->tp_mac;
+                       tp_snaplen = h.h1->tp_snaplen;
+                       tp_sec     = h.h1->tp_sec;
+                       tp_usec    = h.h1->tp_usec;
+                       break;
+#ifdef HAVE_TPACKET2
+               case TPACKET_V2:
+                       tp_len     = h.h2->tp_len;
+                       tp_mac     = h.h2->tp_mac;
+                       tp_snaplen = h.h2->tp_snaplen;
+                       tp_sec     = h.h2->tp_sec;
+                       tp_usec    = h.h2->tp_nsec / 1000;
+                       break;
+#endif
+               default:
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
+                               "unsupported tpacket version %d",
+                               handle->md.tp_version);
+                       return -1;
+               }
                /* perform sanity check on internal offset. */
-               if (thdr->tp_mac+thdr->tp_snaplen > handle->bufsize) {
+               if (tp_mac + tp_snaplen > handle->bufsize) {
                        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                "corrupted frame on kernel ring mac "
                                "offset %d + caplen %d > frame len %d", 
-                               thdr->tp_mac, thdr->tp_snaplen, handle->bufsize);
+                               tp_mac, tp_snaplen, handle->bufsize);
                        return -1;
                }
 
@@ -2116,25 +2308,25 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                 * Note: alternatively it could be possible to stop applying 
                 * the filter when the ring became empty, but it can possibly
                 * happen a lot later... */
-               bp = (unsigned char*)thdr + thdr->tp_mac;
+               bp = (unsigned char*)h.raw + tp_mac;
                run_bpf = (!handle->md.use_bpf) || 
                        ((handle->md.use_bpf>1) && handle->md.use_bpf--);
                if (run_bpf && handle->fcode.bf_insns && 
                                (bpf_filter(handle->fcode.bf_insns, bp,
-                                       thdr->tp_len, thdr->tp_snaplen) == 0))
+                                       tp_len, tp_snaplen) == 0))
                        goto skip;
 
                /* check direction and interface index */
-               sll = (void*)thdr + TPACKET_ALIGN(sizeof(*thdr));
+               sll = (void *)h.raw + TPACKET_ALIGN(handle->md.tp_hdrlen);
                if ((sll->sll_ifindex == handle->md.lo_ifindex) &&
                                        (sll->sll_pkttype == PACKET_OUTGOING))
                        goto skip;
 
                /* get required packet info from ring header */
-               pcaphdr.ts.tv_sec = thdr->tp_sec;
-               pcaphdr.ts.tv_usec = thdr->tp_usec;
-               pcaphdr.caplen = thdr->tp_snaplen;
-               pcaphdr.len = thdr->tp_len;
+               pcaphdr.ts.tv_sec = tp_sec;
+               pcaphdr.ts.tv_usec = tp_usec;
+               pcaphdr.caplen = tp_snaplen;
+               pcaphdr.len = tp_len;
 
                /* if required build in place the sll header*/
                if (handle->md.cooked) {
@@ -2156,7 +2348,9 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                         * don't step on the header when we construct
                         * the sll header.
                         */
-                       if (bp < (u_char *)thdr + TPACKET_HDRLEN) {
+                       if (bp < (u_char *)h.raw +
+                                          TPACKET_ALIGN(handle->md.tp_hdrlen) +
+                                          sizeof(struct sockaddr_ll)) {
                                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                        "cooked-mode frame doesn't have room for sll header");
                                return -1;
@@ -2178,6 +2372,23 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                        pcaphdr.len += SLL_HDR_LEN;
                }
 
+#ifdef HAVE_TPACKET2
+               if (handle->md.tp_version == TPACKET_V2 && h.h2->tp_vlan_tci &&
+                   tp_snaplen >= 2 * ETH_ALEN) {
+                       struct vlan_tag *tag;
+
+                       bp -= VLAN_TAG_LEN;
+                       memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
+
+                       tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
+                       tag->vlan_tpid = htons(ETH_P_8021Q);
+                       tag->vlan_tci = htons(h.h2->tp_vlan_tci);
+
+                       pcaphdr.caplen += VLAN_TAG_LEN;
+                       pcaphdr.len += VLAN_TAG_LEN;
+               }
+#endif
+
                /* pass the packet to the user */
                pkts++;
                callback(user, &pcaphdr, bp);
@@ -2185,7 +2396,16 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
 
 skip:
                /* next packet */
-               thdr->tp_status = TP_STATUS_KERNEL;
+               switch (handle->md.tp_version) {
+               case TPACKET_V1:
+                       h.h1->tp_status = TP_STATUS_KERNEL;
+                       break;
+#ifdef HAVE_TPACKET2
+               case TPACKET_V2:
+                       h.h2->tp_status = TP_STATUS_KERNEL;
+                       break;
+#endif
+               }
                if (++handle->offset >= handle->cc)
                        handle->offset = 0;
 
@@ -2219,7 +2439,7 @@ pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
        for (n=0; n < handle->cc; ++n) {
                if (--handle->offset < 0)
                        handle->offset = handle->cc - 1;
-               if (RING_GET_FRAME(handle)->tp_status != TP_STATUS_KERNEL)
+               if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
                        break;
        }
 
@@ -2316,6 +2536,7 @@ iface_bind(int fd, int ifindex, char *ebuf)
        return 1;
 }
 
+#ifdef IW_MODE_MONITOR
 /*
  * Check whether the device supports the Wireless Extensions.
  * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE
@@ -2324,7 +2545,6 @@ iface_bind(int fd, int ifindex, char *ebuf)
 static int
 has_wext(int sock_fd, const char *device, char *ebuf)
 {
-#ifdef IW_MODE_MONITOR
        struct iwreq ireq;
 
        strncpy(ireq.ifr_ifrn.ifrn_name, device,
@@ -2336,15 +2556,18 @@ has_wext(int sock_fd, const char *device, char *ebuf)
            "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
        if (errno == ENODEV)
                return PCAP_ERROR_NO_SUCH_DEVICE;
-#endif
        return 0;
 }
+#endif
 
 /*
  * Per me si va ne la citta dolente,
  * Per me si va ne l'etterno dolore,
  *     ...
  * Lasciate ogne speranza, voi ch'intrate.
+ *
+ * XXX - airmon-ng does special stuff with the Orinoco driver and the
+ * wlan-ng driver.
  */
 typedef enum {
        MONITOR_WEXT,
@@ -2376,14 +2599,57 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
         *
         * Atheros cards might require that a separate "monitor virtual access
         * point" be created, with later versions of the madwifi driver.
+        * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode
+        * monitor -bssid", which apparently spits out a line "athN"
+        * where "athN" is the monitor mode device.  To leave monitor
+        * mode, it destroys the monitor mode device.
         *
         * Some Intel Centrino adapters might require private ioctls to get
         * radio headers; the ipw2200 and ipw3945 drivers allow you to
         * configure a separate "rtapN" interface to capture in monitor
         * mode without preventing the adapter from operating normally.
+        * (airmon-ng doesn't appear to use that, though.)
         *
         * It would be Truly Wonderful if mac80211 and nl80211 cleaned this
         * up, and if all drivers were converted to mac80211 drivers.
+        *
+        * If interface {if} is a mac80211 driver, the file
+        * /sys/class/net/{if}/phy80211 is a symlink to
+        * /sys/class/ieee80211/{phydev}, for some {phydev}.
+        *
+        * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
+        * least, has a "wmaster0" device and a "wlan0" device; the
+        * latter is the one with the IP address.  Both show up in
+        * "tcpdump -D" output.  Capturing on the wmaster0 device
+        * captures with 802.11 headers.
+        *
+        * airmon-ng searches through /sys/class/net for devices named
+        * monN, starting with mon0; as soon as one *doesn't* exist,
+        * it chooses that as the monitor device name.  If the "iw"
+        * command exists, it does "iw dev {if} interface add {monif}
+        * type monitor", where {monif} is the monitor device.  It
+        * then (sigh) sleeps .1 second, and then configures the
+        * device up.  Otherwise, if /sys/class/ieee80211/{phydev}/add_iface
+        * is a file, it writes {mondev}, without a newline, to that file,
+        * and again (sigh) sleeps .1 second, and then iwconfig's that
+        * device into monitor mode and configures it up.  Otherwise,
+        * you can't do monitor mode.
+        *
+        * All these devices are "glued" together by having the
+        * /sys/class/net/{device}/phy80211 links pointing to the same
+        * place, so, given a wmaster, wlan, or mon device, you can
+        * find the other devices by looking for devices with
+        * the same phy80211 link.
+        *
+        * To turn monitor mode off, delete the monitor interface,
+        * either with "iw dev {monif} interface del" or by sending
+        * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface
+        *
+        * Note: if you try to create a monitor device named "monN", and
+        * there's already a "monN" device, it fails, as least with
+        * the netlink interface (which is what iw uses), with a return
+        * value of -ENFILE.  (Return values are negative errnos.)  We
+        * could probably use that to find an unused device.
         */
        int err;
        struct iwreq ireq;
@@ -2709,6 +2975,12 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                return PCAP_ERROR_RFMON_NOTSUP;
        }
 
+       /*
+        * XXX - airmon-ng does "iwconfig {if} key off" after setting
+        * monitor mode and setting the channel, and then does
+        * "iwconfig up".
+        */
+
        /*
         * Now select the appropriate radio header.
         */
@@ -2904,7 +3176,7 @@ activate_old(pcap_t *handle)
 
        /* Bind to the given device */
 
-       if (!device) {
+       if (strcmp(device, "any") == 0) {
                strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
                        PCAP_ERRBUF_SIZE);
                return PCAP_ERROR;