]> The Tcpdump Group git mirrors - libpcap/blobdiff - pcap-linux.c
Fix typo.
[libpcap] / pcap-linux.c
index 50883b66738e068345cd4b26d8aea063a173077f..de14a7ecd8c3cc41dfb30d1874e37c6626c4799e 100644 (file)
  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
  *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
  *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ *  Modifications:     Added PACKET_MMAP support
+ *                     Paolo Abeni <[email protected]
+ *                     
+ *                     based on previous works of:
+ *                     Simon Patarin <[email protected]>
+ *                     Phil Wood <[email protected]>
+ *
+ * Monitor-mode support for mac80211 includes code taken from the iw
+ * command; the copyright notice for that code is
+ *
+ * Copyright (c) 2007, 2008    Johannes Berg
+ * Copyright (c) 2007          Andy Lutomirski
+ * Copyright (c) 2007          Mike Kershaw
+ * Copyright (c) 2008          Gábor Stefanik
+ *
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer. 
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. The name of the author may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
+ * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
+ * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  
+ * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
+ * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 
+ * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
  */
 
 #ifndef lint
 static const char rcsid[] _U_ =
-    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.102 2003-11-21 10:19:34 guy Exp $ (LBL)";
+    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.164 2008-12-14 22:00:57 guy Exp $ (LBL)";
 #endif
 
 /*
@@ -73,29 +113,68 @@ static const char rcsid[] _U_ =
  */
 
 
+#define _GNU_SOURCE
+
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 
-#include "pcap-int.h"
-#include "sll.h"
-
-#ifdef HAVE_DAG_API
-#include "pcap-dag.h"
-#endif /* HAVE_DAG_API */
-         
 #include <errno.h>
+#include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <string.h>
+#include <limits.h>
 #include <sys/socket.h>
 #include <sys/ioctl.h>
 #include <sys/utsname.h>
-#include <net/if.h>
+#include <sys/mman.h>
+#include <linux/if.h>
 #include <netinet/in.h>
 #include <linux/if_ether.h>
 #include <net/if_arp.h>
+#include <poll.h>
+
+/*
+ * Got Wireless Extensions?
+ */
+#ifdef HAVE_LINUX_WIRELESS_H
+#include <linux/wireless.h>
+#endif /* HAVE_LINUX_WIRELESS_H */
+
+/*
+ * Got libnl?
+ */
+#ifdef HAVE_LIBNL
+#include <linux/nl80211.h>
+
+#include <netlink/genl/genl.h>
+#include <netlink/genl/family.h>
+#include <netlink/genl/ctrl.h>
+#include <netlink/msg.h>
+#include <netlink/attr.h>
+#endif /* HAVE_LIBNL */
+
+#include "pcap-int.h"
+#include "pcap/sll.h"
+#include "pcap/vlan.h"
+
+#ifdef HAVE_DAG_API
+#include "pcap-dag.h"
+#endif /* HAVE_DAG_API */
+
+#ifdef HAVE_SEPTEL_API
+#include "pcap-septel.h"
+#endif /* HAVE_SEPTEL_API */
+
+#ifdef PCAP_SUPPORT_USB
+#include "pcap-usb-linux.h"
+#endif
+
+#ifdef PCAP_SUPPORT_BT
+#include "pcap-bt-linux.h"
+#endif
 
 /*
  * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET
@@ -137,7 +216,23 @@ static const char rcsid[] _U_ =
   */
 # ifdef PACKET_HOST
 #  define HAVE_PF_PACKET_SOCKETS
+#  ifdef PACKET_AUXDATA
+#   define HAVE_PACKET_AUXDATA
+#  endif /* PACKET_AUXDATA */
 # endif /* PACKET_HOST */
+
+
+ /* check for memory mapped access avaibility. We assume every needed 
+  * struct is defined if the macro TPACKET_HDRLEN is defined, because it
+  * 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 */
 
 #ifdef SO_ATTACH_FILTER
@@ -145,7 +240,7 @@ static const char rcsid[] _U_ =
 #include <linux/filter.h>
 #endif
 
-#ifndef __GLIBC__
+#ifndef HAVE_SOCKLEN_T
 typedef int            socklen_t;
 #endif
 
@@ -181,16 +276,45 @@ typedef int               socklen_t;
 #define BIGGER_THAN_ALL_MTUS   (64*1024)
 
 /*
- * Prototypes for internal functions
+ * Prototypes for internal functions and methods.
  */
 static void map_arphrd_to_dlt(pcap_t *, int, int);
-static int live_open_old(pcap_t *, const char *, int, int, char *);
-static int live_open_new(pcap_t *, const char *, int, int, char *);
+#ifdef HAVE_PF_PACKET_SOCKETS
+static short int map_packet_type_to_sll_type(short int);
+#endif
+static int pcap_activate_linux(pcap_t *);
+static int activate_old(pcap_t *);
+static int activate_new(pcap_t *);
+static int activate_mmap(pcap_t *);
+static int pcap_can_set_rfmon_linux(pcap_t *);
 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *);
 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *);
+static int pcap_inject_linux(pcap_t *, const void *, size_t);
 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
-static void pcap_close_linux(pcap_t *);
+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) (((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 *);
+static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf);
+static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf);
+static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
+    const u_char *bytes);
+#endif
 
 /*
  * Wrap some ioctl calls
@@ -202,11 +326,17 @@ static int        iface_get_mtu(int fd, const char *device, char *ebuf);
 static int     iface_get_arptype(int fd, const char *device, char *ebuf);
 #ifdef HAVE_PF_PACKET_SOCKETS
 static int     iface_bind(int fd, int ifindex, char *ebuf);
-#endif
+#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(pcap_t *handle, int sock_fd,
+    const char *device);
+#endif /* HAVE_PF_PACKET_SOCKETS */
 static int     iface_bind_old(int fd, const char *device, char *ebuf);
 
 #ifdef SO_ATTACH_FILTER
-static int     fix_program(pcap_t *handle, struct sock_fprog *fcode);
+static int     fix_program(pcap_t *handle, struct sock_fprog *fcode,
+    int is_mapped);
 static int     fix_offset(struct bpf_insn *p);
 static int     set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
 static int     reset_kernel_filter(pcap_t *handle);
@@ -217,239 +347,868 @@ static struct sock_fprog        total_fcode
        = { 1, &total_insn };
 #endif
 
-/*
- *  Get a handle for a live capture from the given device. You can
- *  pass NULL as device to get all packages (without link level
- *  information of course). If you pass 1 as promisc the interface
- *  will be set to promiscous mode (XXX: I think this usage should
- *  be deprecated and functions be added to select that later allow
- *  modification of that values -- Torsten).
- *
- *  See also pcap(3).
- */
 pcap_t *
-pcap_open_live(const char *device, int snaplen, int promisc, int to_ms,
-    char *ebuf)
+pcap_create(const char *device, char *ebuf)
 {
-       pcap_t          *handle;
-       int             mtu;
-       int             err;
-       int             live_open_ok = 0;
-       struct utsname  utsname;
+       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_open_live(device, snaplen, promisc, to_ms, ebuf);
+               return dag_create(device, ebuf);
        }
 #endif /* HAVE_DAG_API */
 
-        /* Allocate a handle for this session. */
+#ifdef HAVE_SEPTEL_API
+       if (strstr(device, "septel")) {
+               return septel_create(device, ebuf);
+       }
+#endif /* HAVE_SEPTEL_API */
 
-       handle = malloc(sizeof(*handle));
-       if (handle == NULL) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s",
-                        pcap_strerror(errno));
-               return NULL;
+#ifdef PCAP_SUPPORT_BT
+       if (strstr(device, "bluetooth")) {
+               return bt_create(device, ebuf);
+       }
+#endif
+
+#ifdef PCAP_SUPPORT_USB
+       if (strstr(device, "usbmon")) {
+               return usb_create(device, ebuf);
        }
+#endif
+
+       handle = pcap_create_common(device, ebuf);
+       if (handle == NULL)
+               return NULL;
 
-       /* Initialize some components of the pcap structure. */
+       handle->activate_op = pcap_activate_linux;
+       handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
+       return handle;
+}
+
+#ifdef HAVE_LIBNL
+/*
+        *
+        * 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.
+        *
+        * Yes, you can have multiple monitor devices for a given
+        * physical device.
+*/
 
-       memset(handle, 0, sizeof(*handle));
-       handle->snapshot        = snaplen;
-       handle->md.timeout      = to_ms;
+/*
+ * Is this a mac80211 device?  If so, fill in the physical device path and
+ * return 1; if not, return 0.  On an error, fill in handle->errbuf and
+ * return PCAP_ERROR.
+ */
+static int
+get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
+    size_t phydev_max_pathlen)
+{
+       char *pathstr;
+       ssize_t bytes_read;
 
        /*
-        * NULL and "any" are special devices which give us the hint to
-        * monitor all devices.
+        * Generate the path string for the symlink to the physical device.
         */
-       if (!device || strcmp(device, "any") == 0) {
-               device                  = NULL;
-               handle->md.device       = strdup("any");
-               if (promisc) {
-                       promisc = 0;
-                       /* Just a warning. */
-                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                           "Promiscuous mode not supported on the \"any\" device");
+       if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: Can't generate path name string for /sys/class/net device",
+                   device);
+               return PCAP_ERROR;
+       }
+       bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
+       if (bytes_read == -1) {
+               if (errno == ENOENT || errno == EINVAL) {
+                       /*
+                        * Doesn't exist, or not a symlink; assume that
+                        * means it's not a mac80211 device.
+                        */
+                       free(pathstr);
+                       return 0;
                }
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: Can't readlink %s: %s", device, pathstr,
+                   strerror(errno));
+               free(pathstr);
+               return PCAP_ERROR;
+       }
+       free(pathstr);
+       phydev_path[bytes_read] = '\0';
+       return 1;
+}
 
-       } else
-               handle->md.device       = strdup(device);
+struct nl80211_state {
+       struct nl_handle *nl_handle;
+       struct nl_cache *nl_cache;
+       struct genl_family *nl80211;
+};
 
-       if (handle->md.device == NULL) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE, "strdup: %s",
-                        pcap_strerror(errno) );
-               free(handle);
-               return NULL;
+static int
+nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
+{
+       state->nl_handle = nl_handle_alloc();
+       if (!state->nl_handle) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to allocate netlink handle", device);
+               return PCAP_ERROR;
        }
 
-       /*
-        * Current Linux kernels use the protocol family PF_PACKET to
-        * allow direct access to all packets on the network while
-        * older kernels had a special socket type SOCK_PACKET to
-        * implement this feature.
-        * While this old implementation is kind of obsolete we need
-        * to be compatible with older kernels for a while so we are
-        * trying both methods with the newer method preferred.
-        */
+       if (genl_connect(state->nl_handle)) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to connect to generic netlink", device);
+               goto out_handle_destroy;
+       }
 
-       if ((err = live_open_new(handle, device, promisc, to_ms, ebuf)) == 1)
-               live_open_ok = 1;
-       else if (err == 0) {
-               /* Non-fatal error; try old way */
-               if (live_open_old(handle, device, promisc, to_ms, ebuf))
-                       live_open_ok = 1;
+       state->nl_cache = genl_ctrl_alloc_cache(state->nl_handle);
+       if (!state->nl_cache) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to allocate generic netlink cache", device);
+               goto out_handle_destroy;
        }
-       if (!live_open_ok) {
-               /*
-                * Both methods to open the packet socket failed. Tidy
-                * up and report our failure (ebuf is expected to be
-                * set by the functions above).
-                */
 
-               if (handle->md.device != NULL)
-                       free(handle->md.device);
-               free(handle);
-               return NULL;
+       state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
+       if (!state->nl80211) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: nl80211 not found", device);
+               goto out_cache_free;
        }
 
-       /*
-        * Compute the buffer size.
-        *
-        * If we're using SOCK_PACKET, this might be a 2.0[.x] kernel,
-        * and might require special handling - check.
-        */
-       if (handle->md.sock_packet && (uname(&utsname) < 0 ||
-           strncmp(utsname.release, "2.0", 3) == 0)) {
-               /*
-                * We're using a SOCK_PACKET structure, and either
-                * we couldn't find out what kernel release this is,
-                * or it's a 2.0[.x] kernel.
-                *
-                * In the 2.0[.x] kernel, a "recvfrom()" on
-                * a SOCK_PACKET socket, with MSG_TRUNC set, will
-                * return the number of bytes read, so if we pass
-                * a length based on the snapshot length, it'll
-                * return the number of bytes from the packet
-                * copied to userland, not the actual length
-                * of the packet.
-                *
-                * This means that, for example, the IP dissector
-                * in tcpdump will get handed a packet length less
-                * than the length in the IP header, and will
-                * complain about "truncated-ip".
-                *
-                * So we don't bother trying to copy from the
-                * kernel only the bytes in which we're interested,
-                * but instead copy them all, just as the older
-                * versions of libpcap for Linux did.
-                *
-                * The buffer therefore needs to be big enough to
-                * hold the largest packet we can get from this
-                * device.  Unfortunately, we can't get the MRU
-                * of the network; we can only get the MTU.  The
-                * MTU may be too small, in which case a packet larger
-                * than the buffer size will be truncated *and* we
-                * won't get the actual packet size.
-                *
-                * However, if the snapshot length is larger than
-                * the buffer size based on the MTU, we use the
-                * snapshot length as the buffer size, instead;
-                * this means that with a sufficiently large snapshot
-                * length we won't artificially truncate packets
-                * to the MTU-based size.
-                *
-                * This mess just one of many problems with packet
-                * capture on 2.0[.x] kernels; you really want a
-                * 2.2[.x] or later kernel if you want packet capture
-                * to work well.
-                */
-               mtu = iface_get_mtu(handle->fd, device, ebuf);
-               if (mtu == -1) {
-                       pcap_close_linux(handle);
-                       free(handle);
-                       return NULL;
-               }
-               handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
-               if (handle->bufsize < handle->snapshot)
-                       handle->bufsize = handle->snapshot;
-       } else {
-               /*
-                * This is a 2.2[.x] or later kernel (we know that
-                * either because we're not using a SOCK_PACKET
-                * socket - PF_PACKET is supported only in 2.2
-                * and later kernels - or because we checked the
-                * kernel version).
-                *
-                * We can safely pass "recvfrom()" a byte count
-                * based on the snapshot length.
-                */
-               handle->bufsize = handle->snapshot;
+       return 0;
+
+out_cache_free:
+       nl_cache_free(state->nl_cache);
+out_handle_destroy:
+       nl_handle_destroy(state->nl_handle);
+       return PCAP_ERROR;
+}
+
+static void
+nl80211_cleanup(struct nl80211_state *state)
+{
+       genl_family_put(state->nl80211);
+       nl_cache_free(state->nl_cache);
+       nl_handle_destroy(state->nl_handle);
+}
+
+static int
+add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
+    const char *device, const char *mondevice)
+{
+       int ifindex;
+       struct nl_msg *msg;
+       int err;
+
+       ifindex = iface_get_id(sock_fd, device, handle->errbuf);
+       if (ifindex == -1)
+               return PCAP_ERROR;
+
+       msg = nlmsg_alloc();
+       if (!msg) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to allocate netlink msg", device);
+               return PCAP_ERROR;
        }
 
-       /* Allocate the buffer */
+       genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
+                   0, NL80211_CMD_NEW_INTERFACE, 0);
+       NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
+       NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
+       NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
 
-       handle->buffer   = malloc(handle->bufsize + handle->offset);
-       if (!handle->buffer) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "malloc: %s", pcap_strerror(errno));
-               pcap_close_linux(handle);
-               free(handle);
-               return NULL;
+       err = nl_send_auto_complete(state->nl_handle, msg);
+       if (err < 0) {
+               if (err == -ENFILE) {
+                       /*
+                        * Device not available; our caller should just
+                        * keep trying.
+                        */
+                       nlmsg_free(msg);
+                       return 0;
+               } else {
+                       /*
+                        * Real failure, not just "that device is not
+                        * available.
+                        */
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "%s: nl_send_auto_complete failed adding %s interface: %s",
+                           device, mondevice, strerror(-err));
+                       nlmsg_free(msg);
+                       return PCAP_ERROR;
+               }
+       }
+       err = nl_wait_for_ack(state->nl_handle);
+       if (err < 0) {
+               if (err == -ENFILE) {
+                       /*
+                        * Device not available; our caller should just
+                        * keep trying.
+                        */
+                       nlmsg_free(msg);
+                       return 0;
+               } else {
+                       /*
+                        * Real failure, not just "that device is not
+                        * available.
+                        */
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "%s: nl_wait_for_ack failed adding %s interface: %s",
+                           device, mondevice, strerror(-err));
+                       nlmsg_free(msg);
+                       return PCAP_ERROR;
+               }
        }
 
        /*
-        * "handle->fd" is a socket, so "select()" and "poll()"
-        * should work on it.
+        * Success.
         */
-       handle->selectable_fd = handle->fd;
-
-       handle->read_op = pcap_read_linux;
-       handle->setfilter_op = pcap_setfilter_linux;
-       handle->set_datalink_op = NULL; /* can't change data link type */
-       handle->getnonblock_op = pcap_getnonblock_fd;
-       handle->setnonblock_op = pcap_setnonblock_fd;
-       handle->stats_op = pcap_stats_linux;
-       handle->close_op = pcap_close_linux;
+       nlmsg_free(msg);
+       return 1;
 
-       return handle;
+nla_put_failure:
+       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+           "%s: nl_put failed adding %s interface",
+           device, mondevice);
+       nlmsg_free(msg);
+       return PCAP_ERROR;
 }
 
-/*
- *  Read at most max_packets from the capture stream and call the callback
- *  for each of them. Returns the number of packets handled or -1 if an
- *  error occured.
- */
 static int
-pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
+del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
+    const char *device, const char *mondevice)
 {
+       int ifindex;
+       struct nl_msg *msg;
+       int err;
+
+       ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
+       if (ifindex == -1)
+               return PCAP_ERROR;
+
+       msg = nlmsg_alloc();
+       if (!msg) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to allocate netlink msg", device);
+               return PCAP_ERROR;
+       }
+
+       genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
+                   0, NL80211_CMD_DEL_INTERFACE, 0);
+       NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
+
+       err = nl_send_auto_complete(state->nl_handle, msg);
+       if (err < 0) {
+               if (err == -ENFILE) {
+                       /*
+                        * Device not available; our caller should just
+                        * keep trying.
+                        */
+                       nlmsg_free(msg);
+                       return 0;
+               } else {
+                       /*
+                        * Real failure, not just "that device is not
+                        * available.
+                        */
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "%s: nl_send_auto_complete failed deleting %s interface: %s",
+                           device, mondevice, strerror(-err));
+                       nlmsg_free(msg);
+                       return PCAP_ERROR;
+               }
+       }
+       err = nl_wait_for_ack(state->nl_handle);
+       if (err < 0) {
+               if (err == -ENFILE) {
+                       /*
+                        * Device not available; our caller should just
+                        * keep trying.
+                        */
+                       nlmsg_free(msg);
+                       return 0;
+               } else {
+                       /*
+                        * Real failure, not just "that device is not
+                        * available.
+                        */
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "%s: nl_wait_for_ack failed adding %s interface: %s",
+                           device, mondevice, strerror(-err));
+                       nlmsg_free(msg);
+                       return PCAP_ERROR;
+               }
+       }
+
        /*
-        * Currently, on Linux only one packet is delivered per read,
-        * so we don't loop.
+        * Success.
         */
-       return pcap_read_packet(handle, callback, user);
+       nlmsg_free(msg);
+       return 1;
+
+nla_put_failure:
+       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+           "%s: nl_put failed deleting %s interface",
+           device, mondevice);
+       nlmsg_free(msg);
+       return PCAP_ERROR;
 }
 
-/*
- *  Read a packet from the socket calling the handler provided by
- *  the user. Returns the number of packets received or -1 if an
- *  error occured.
- */
 static int
-pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
+enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
 {
-       u_char                  *bp;
-       int                     offset;
-#ifdef HAVE_PF_PACKET_SOCKETS
-       struct sockaddr_ll      from;
-       struct sll_header       *hdrp;
-#else
-       struct sockaddr         from;
-#endif
-       socklen_t               fromlen;
-       int                     packet_len, caplen;
-       struct pcap_pkthdr      pcap_header;
+       int ret;
+       char phydev_path[PATH_MAX+1];
+       struct nl80211_state nlstate;
+       struct ifreq ifr;
+       u_int n;
 
-#ifdef HAVE_PF_PACKET_SOCKETS
+       /*
+        * Is this a mac80211 device?
+        */
+       ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
+       if (ret < 0)
+               return ret;     /* error */
+       if (ret == 0)
+               return 0;       /* no error, but not mac80211 device */
+
+       /*
+        * XXX - is this already a monN device?
+        * If so, we're done.
+        * Is that determined by old Wireless Extensions ioctls?
+        */
+
+       /*
+        * OK, it's apparently a mac80211 device.
+        * Try to find an unused monN device for it.
+        */
+       ret = nl80211_init(handle, &nlstate, device);
+       if (ret != 0)
+               return ret;
+       for (n = 0; n < UINT_MAX; n++) {
+               /*
+                * Try mon{n}.
+                */
+               char mondevice[3+10+1]; /* mon{UINT_MAX}\0 */
+
+               snprintf(mondevice, sizeof mondevice, "mon%u", n);
+               ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
+               if (ret == 1) {
+                       handle->md.mondevice = strdup(mondevice);
+                       goto added;
+               }
+               if (ret < 0) {
+                       /*
+                        * Hard failure.  Just return ret; handle->errbuf
+                        * has already been set.
+                        */
+                       nl80211_cleanup(&nlstate);
+                       return ret;
+               }
+       }
+
+       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+           "%s: No free monN interfaces", device);
+       nl80211_cleanup(&nlstate);
+       return PCAP_ERROR;
+
+added:
+
+#if 0
+       /*
+        * Sleep for .1 seconds.
+        */
+       delay.tv_sec = 0;
+       delay.tv_nsec = 500000000;
+       nanosleep(&delay, NULL);
+#endif
+
+       /*
+        * Now configure the monitor interface up.
+        */
+       memset(&ifr, 0, sizeof(ifr));
+       strncpy(ifr.ifr_name, handle->md.mondevice, sizeof(ifr.ifr_name));
+       if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: Can't get flags for %s: %s", device,
+                   handle->md.mondevice, strerror(errno));
+               del_mon_if(handle, sock_fd, &nlstate, device,
+                   handle->md.mondevice);
+               nl80211_cleanup(&nlstate);
+               return PCAP_ERROR;
+       }
+       ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
+       if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: Can't set flags for %s: %s", device,
+                   handle->md.mondevice, strerror(errno));
+               del_mon_if(handle, sock_fd, &nlstate, device,
+                   handle->md.mondevice);
+               nl80211_cleanup(&nlstate);
+               return PCAP_ERROR;
+       }
+
+       /*
+        * Success.  Clean up the libnl state.
+        */
+       nl80211_cleanup(&nlstate);
+
+       /*
+        * Note that we have to delete the monitor device when we close
+        * the handle.
+        */
+       handle->md.must_do_on_close |= MUST_DELETE_MONIF;
+
+       /*
+        * Add this to the list of pcaps to close when we exit.
+        */
+       pcap_add_to_pcaps_to_close(handle);
+
+       return 1;
+}
+#endif /* HAVE_LIBNL */
+
+static int
+pcap_can_set_rfmon_linux(pcap_t *handle)
+{
+#ifdef HAVE_LIBNL
+       char phydev_path[PATH_MAX+1];
+       int ret;
+#endif
+#ifdef IW_MODE_MONITOR
+       int sock_fd;
+       struct iwreq ireq;
+#endif
+
+       if (strcmp(handle->opt.source, "any") == 0) {
+               /*
+                * Monitor mode makes no sense on the "any" device.
+                */
+               return 0;
+       }
+
+#ifdef HAVE_LIBNL
+       /*
+        * Bleah.  There doesn't seem to be a way to ask a mac80211
+        * device, through libnl, whether it supports monitor mode;
+        * we'll just check whether the device appears to be a
+        * mac80211 device and, if so, assume the device supports
+        * monitor mode.
+        *
+        * wmaster devices don't appear to support the Wireless
+        * Extensions, but we can create a mon device for a
+        * wmaster device, so we don't bother checking whether
+        * a mac80211 device supports the Wireless Extensions.
+        */
+       ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
+           PATH_MAX);
+       if (ret < 0)
+               return ret;     /* error */
+       if (ret == 1)
+               return 1;       /* mac80211 device */
+#endif
+
+#ifdef IW_MODE_MONITOR
+       /*
+        * Bleah.  There doesn't appear to be an ioctl to use to ask
+        * whether a device supports monitor mode; we'll just do
+        * SIOCGIWMODE and, if it succeeds, assume the device supports
+        * monitor mode.
+        *
+        * Open a socket on which to attempt to get the mode.
+        * (We assume that if we have Wireless Extensions support
+        * we also have PF_PACKET support.)
+        */
+       sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+       if (sock_fd == -1) {
+               (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "socket: %s", pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+
+       /*
+        * Attempt to get the current mode.
+        */
+       strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
+               /*
+                * Well, we got the mode; assume we can set it.
+                */
+               close(sock_fd);
+               return 1;
+       }
+       if (errno == ENODEV) {
+               /* The device doesn't even exist. */
+               close(sock_fd);
+               return PCAP_ERROR_NO_SUCH_DEVICE;
+       }
+       close(sock_fd);
+#endif
+       return 0;
+}
+
+/*
+ * With older kernels promiscuous mode is kind of interesting because we
+ * have to reset the interface before exiting. The problem can't really
+ * be solved without some daemon taking care of managing usage counts.
+ * If we put the interface into promiscuous mode, we set a flag indicating
+ * that we must take it out of that mode when the interface is closed,
+ * and, when closing the interface, if that flag is set we take it out
+ * of promiscuous mode.
+ *
+ * Even with newer kernels, we have the same issue with rfmon mode.
+ */
+
+static void    pcap_cleanup_linux( pcap_t *handle )
+{
+       struct ifreq    ifr;
+#ifdef HAVE_LIBNL
+       struct nl80211_state nlstate;
+       int ret;
+#endif /* HAVE_LIBNL */
+#ifdef IW_MODE_MONITOR
+       struct iwreq ireq;
+#endif /* IW_MODE_MONITOR */
+
+       if (handle->md.must_do_on_close != 0) {
+               /*
+                * There's something we have to do when closing this
+                * pcap_t.
+                */
+               if (handle->md.must_do_on_close & MUST_CLEAR_PROMISC) {
+                       /*
+                        * We put the interface into promiscuous mode;
+                        * take it out of promiscuous mode.
+                        *
+                        * XXX - if somebody else wants it in promiscuous
+                        * mode, this code cannot know that, so it'll take
+                        * it out of promiscuous mode.  That's not fixable
+                        * in 2.0[.x] kernels.
+                        */
+                       memset(&ifr, 0, sizeof(ifr));
+                       strncpy(ifr.ifr_name, handle->md.device,
+                           sizeof(ifr.ifr_name));
+                       if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
+                               fprintf(stderr,
+                                   "Can't restore interface flags (SIOCGIFFLAGS failed: %s).\n"
+                                   "Please adjust manually.\n"
+                                   "Hint: This can't happen with Linux >= 2.2.0.\n",
+                                   strerror(errno));
+                       } else {
+                               if (ifr.ifr_flags & IFF_PROMISC) {
+                                       /*
+                                        * Promiscuous mode is currently on;
+                                        * turn it off.
+                                        */
+                                       ifr.ifr_flags &= ~IFF_PROMISC;
+                                       if (ioctl(handle->fd, SIOCSIFFLAGS,
+                                           &ifr) == -1) {
+                                               fprintf(stderr,
+                                                   "Can't restore interface flags (SIOCSIFFLAGS failed: %s).\n"
+                                                   "Please adjust manually.\n"
+                                                   "Hint: This can't happen with Linux >= 2.2.0.\n",
+                                                   strerror(errno));
+                                       }
+                               }
+                       }
+               }
+
+#ifdef HAVE_LIBNL
+               if (handle->md.must_do_on_close & MUST_DELETE_MONIF) {
+                       ret = nl80211_init(handle, &nlstate, handle->md.device);
+                       if (ret >= 0) {
+                               ret = del_mon_if(handle, handle->fd, &nlstate,
+                                   handle->md.device, handle->md.mondevice);
+                               nl80211_cleanup(&nlstate);
+                       }
+                       if (ret < 0) {
+                               fprintf(stderr,
+                                   "Can't delete monitor interface %s (%s).\n"
+                                   "Please delete manually.\n",
+                                   handle->md.mondevice, handle->errbuf);
+                       }
+               }
+#endif /* HAVE_LIBNL */
+
+#ifdef IW_MODE_MONITOR
+               if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
+                       /*
+                        * We put the interface into rfmon mode;
+                        * take it out of rfmon mode.
+                        *
+                        * XXX - if somebody else wants it in rfmon
+                        * mode, this code cannot know that, so it'll take
+                        * it out of rfmon mode.
+                        */
+                       strncpy(ireq.ifr_ifrn.ifrn_name, handle->md.device,
+                           sizeof ireq.ifr_ifrn.ifrn_name);
+                       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1]
+                           = 0;
+                       ireq.u.mode = handle->md.oldmode;
+                       if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) {
+                               /*
+                                * Scientist, you've failed.
+                                */
+                               fprintf(stderr,
+                                   "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n"
+                                   "Please adjust manually.\n",
+                                   strerror(errno));
+                       }
+               }
+#endif /* IW_MODE_MONITOR */
+
+               /*
+                * Take this pcap out of the list of pcaps for which we
+                * have to take the interface out of some mode.
+                */
+               pcap_remove_from_pcaps_to_close(handle);
+       }
+
+       if (handle->md.mondevice != NULL) {
+               free(handle->md.mondevice);
+               handle->md.mondevice = NULL;
+       }
+       if (handle->md.device != NULL) {
+               free(handle->md.device);
+               handle->md.device = NULL;
+       }
+       pcap_cleanup_live_common(handle);
+}
+
+/*
+ *  Get a handle for a live capture from the given device. You can
+ *  pass NULL as device to get all packages (without link level
+ *  information of course). If you pass 1 as promisc the interface
+ *  will be set to promiscous mode (XXX: I think this usage should
+ *  be deprecated and functions be added to select that later allow
+ *  modification of that values -- Torsten).
+ */
+static int
+pcap_activate_linux(pcap_t *handle)
+{
+       const char      *device;
+       int             status = 0;
+
+       device = handle->opt.source;
+
+       handle->inject_op = pcap_inject_linux;
+       handle->setfilter_op = pcap_setfilter_linux;
+       handle->setdirection_op = pcap_setdirection_linux;
+       handle->set_datalink_op = NULL; /* can't change data link type */
+       handle->getnonblock_op = pcap_getnonblock_fd;
+       handle->setnonblock_op = pcap_setnonblock_fd;
+       handle->cleanup_op = pcap_cleanup_linux;
+       handle->read_op = pcap_read_linux;
+       handle->stats_op = pcap_stats_linux;
+
+       /*
+        * 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 (strcmp(device, "any") == 0) {
+               if (handle->opt.promisc) {
+                       handle->opt.promisc = 0;
+                       /* Just a warning. */
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "Promiscuous mode not supported on the \"any\" device");
+                       status = PCAP_WARNING_PROMISC_NOTSUP;
+               }
+       }
+
+       handle->md.device       = strdup(device);
+       if (handle->md.device == NULL) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
+                        pcap_strerror(errno) );
+               return PCAP_ERROR;
+       }
+
+       /*
+        * Current Linux kernels use the protocol family PF_PACKET to
+        * allow direct access to all packets on the network while
+        * older kernels had a special socket type SOCK_PACKET to
+        * implement this feature.
+        * While this old implementation is kind of obsolete we need
+        * to be compatible with older kernels for a while so we are
+        * trying both methods with the newer method preferred.
+        */
+
+       if ((status = activate_new(handle)) == 1) {
+               /*
+                * Success.
+                * Try to use memory-mapped access.
+                */
+               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.
+                        */
+                       status = 0;
+                       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) {
+                       /*
+                        * 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 {
+               /*
+                * 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;
+       }
+
+       /*
+        * 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.
+                */
+               if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF,
+                   &handle->opt.buffer_size,
+                   sizeof(handle->opt.buffer_size)) == -1) {
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                "SO_RCVBUF: %s", pcap_strerror(errno));
+                       status = PCAP_ERROR;
+                       goto fail;
+               }
+       }
+
+       /* Allocate the buffer */
+
+       handle->buffer   = malloc(handle->bufsize + handle->offset);
+       if (!handle->buffer) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "malloc: %s", pcap_strerror(errno));
+               status = PCAP_ERROR;
+               goto fail;
+       }
+
+       /*
+        * "handle->fd" is a socket, so "select()" and "poll()"
+        * should work on it.
+        */
+       handle->selectable_fd = handle->fd;
+
+       return status;
+
+fail:
+       pcap_cleanup_linux(handle);
+       return status;
+}
+
+/*
+ *  Read at most max_packets from the capture stream and call the callback
+ *  for each of them. Returns the number of packets handled or -1 if an
+ *  error occured.
+ */
+static int
+pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user)
+{
+       /*
+        * Currently, on Linux only one packet is delivered per read,
+        * so we don't loop.
+        */
+       return pcap_read_packet(handle, callback, user);
+}
+
+/*
+ *  Read a packet from the socket calling the handler provided by
+ *  the user. Returns the number of packets received or -1 if an
+ *  error occured.
+ */
+static int
+pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
+{
+       u_char                  *bp;
+       int                     offset;
+#ifdef HAVE_PF_PACKET_SOCKETS
+       struct sockaddr_ll      from;
+       struct sll_header       *hdrp;
+#else
+       struct sockaddr         from;
+#endif
+#if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI)
+       struct iovec            iov;
+       struct msghdr           msg;
+       struct cmsghdr          *cmsg;
+       union {
+               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
         * fake packet header.
@@ -466,9 +1225,36 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
        offset = 0;
 #endif
 
-       /* Receive a single packet from the kernel */
-
+       /*
+        * Receive a single packet from the kernel.
+        * We ignore EINTR, as that might just be due to a signal
+        * being delivered - if the signal should interrupt the
+        * loop, the signal handler should call pcap_breakloop()
+        * to set handle->break_loop (we ignore it on other
+        * platforms as well).
+        * We also ignore ENETDOWN, so that we can continue to
+        * capture traffic if the interface goes down and comes
+        * back up again; comments in the kernel indicate that
+        * we'll just block waiting for packets if we try to
+        * receive from a socket that delivered ENETDOWN, and,
+        * if we're using a memory-mapped buffer, we won't even
+        * get notified of "network down" events.
+        */
        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;
+       msg.msg_iovlen          = 1;
+       msg.msg_control         = &cmsg_buf;
+       msg.msg_controllen      = sizeof(cmsg_buf);
+       msg.msg_flags           = 0;
+
+       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 {
                /*
                 * Has "pcap_breakloop()" been called?
@@ -482,12 +1268,17 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                        handle->break_loop = 0;
                        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);
-       } while (packet_len == -1 && errno == EINTR);
+#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 */
 
@@ -495,26 +1286,60 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                if (errno == EAGAIN)
                        return 0;       /* no packet there */
                else {
-                       snprintf(handle->errbuf, sizeof(handle->errbuf),
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                                 "recvfrom: %s", pcap_strerror(errno));
                        return -1;
                }
        }
 
 #ifdef HAVE_PF_PACKET_SOCKETS
-       /*
-        * If this is from the loopback device, reject outgoing packets;
-        * we'll see the packet as an incoming packet as well, and
-        * we don't want to see it twice.
-        *
-        * We can only do this if we're using PF_PACKET; the address
-        * returned for SOCK_PACKET is a "sockaddr_pkt" which lacks
-        * the relevant packet type information.
-        */
-       if (!handle->md.sock_packet &&
-           from.sll_ifindex == handle->md.lo_ifindex &&
-           from.sll_pkttype == PACKET_OUTGOING)
-               return 0;
+       if (!handle->md.sock_packet) {
+               /*
+                * Unfortunately, there is a window between socket() and
+                * bind() where the kernel may queue packets from any
+                * interface.  If we're bound to a particular interface,
+                * discard packets not from that interface.
+                *
+                * (If socket filters are supported, we could do the
+                * same thing we do when changing the filter; however,
+                * that won't handle packet sockets without socket
+                * filter support, and it's a bit more complicated.
+                * It would save some instructions per packet, however.)
+                */
+               if (handle->md.ifindex != -1 &&
+                   from.sll_ifindex != handle->md.ifindex)
+                       return 0;
+
+               /*
+                * Do checks based on packet direction.
+                * We can only do this if we're using PF_PACKET; the
+                * address returned for SOCK_PACKET is a "sockaddr_pkt"
+                * which lacks the relevant packet type information.
+                */
+               if (from.sll_pkttype == PACKET_OUTGOING) {
+                       /*
+                        * Outgoing packet.
+                        * If this is from the loopback device, reject it;
+                        * we'll see the packet as an incoming packet as well,
+                        * and we don't want to see it twice.
+                        */
+                       if (from.sll_ifindex == handle->md.lo_ifindex)
+                               return 0;
+
+                       /*
+                        * If the user only wants incoming packets, reject it.
+                        */
+                       if (handle->direction == PCAP_D_IN)
+                               return 0;
+               } else {
+                       /*
+                        * Incoming packet.
+                        * If the user only wants outgoing packets, reject it.
+                        */
+                       if (handle->direction == PCAP_D_OUT)
+                               return 0;
+               }
+       }
 #endif
 
 #ifdef HAVE_PF_PACKET_SOCKETS
@@ -529,51 +1354,46 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                packet_len += SLL_HDR_LEN;
 
                hdrp = (struct sll_header *)bp;
+               hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype);
+               hdrp->sll_hatype = htons(from.sll_hatype);
+               hdrp->sll_halen = htons(from.sll_halen);
+               memcpy(hdrp->sll_addr, from.sll_addr,
+                   (from.sll_halen > SLL_ADDRLEN) ?
+                     SLL_ADDRLEN :
+                     from.sll_halen);
+               hdrp->sll_protocol = from.sll_protocol;
+       }
 
-               /*
-                * Map the PACKET_ value to a LINUX_SLL_ value; we
-                * want the same numerical value to be used in
-                * the link-layer header even if the numerical values
-                * for the PACKET_ #defines change, so that programs
-                * that look at the packet type field will always be
-                * able to handle DLT_LINUX_SLL captures.
-                */
-               switch (from.sll_pkttype) {
-
-               case PACKET_HOST:
-                       hdrp->sll_pkttype = htons(LINUX_SLL_HOST);
-                       break;
+#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;
+               struct vlan_tag *tag;
 
-               case PACKET_BROADCAST:
-                       hdrp->sll_pkttype = htons(LINUX_SLL_BROADCAST);
-                       break;
+               if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
+                   cmsg->cmsg_level != SOL_PACKET ||
+                   cmsg->cmsg_type != PACKET_AUXDATA)
+                       continue;
 
-               case PACKET_MULTICAST:
-                       hdrp->sll_pkttype = htons(LINUX_SLL_MULTICAST);
-                       break;
+               aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
+               if (aux->tp_vlan_tci == 0)
+                       continue;
 
-               case PACKET_OTHERHOST:
-                       hdrp->sll_pkttype = htons(LINUX_SLL_OTHERHOST);
+               len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
+               if (len < 2 * ETH_ALEN)
                        break;
 
-               case PACKET_OUTGOING:
-                       hdrp->sll_pkttype = htons(LINUX_SLL_OUTGOING);
-                       break;
+               bp -= VLAN_TAG_LEN;
+               memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
 
-               default:
-                       hdrp->sll_pkttype = -1;
-                       break;
-               }
+               tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
+               tag->vlan_tpid = htons(ETH_P_8021Q);
+               tag->vlan_tci = htons(aux->tp_vlan_tci);
 
-               hdrp->sll_hatype = htons(from.sll_hatype);
-               hdrp->sll_halen = htons(from.sll_halen);
-               memcpy(hdrp->sll_addr, from.sll_addr,
-                   (from.sll_halen > SLL_ADDRLEN) ?
-                     SLL_ADDRLEN :
-                     from.sll_halen);
-               hdrp->sll_protocol = from.sll_protocol;
+               packet_len += VLAN_TAG_LEN;
        }
-#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
@@ -624,8 +1444,8 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
        /* Fill in our own header data */
 
        if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) {
-               snprintf(handle->errbuf, sizeof(handle->errbuf),
-                        "ioctl: %s", pcap_strerror(errno));
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "SIOCGSTAMP: %s", pcap_strerror(errno));
                return -1;
        }
        pcap_header.caplen      = caplen;
@@ -663,8 +1483,18 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
         * here, but it's not clear that always incrementing
         * the count is more expensive than always testing a flag
         * in memory.
+        *
+        * We keep the count in "md.packets_read", and use that for
+        * "ps_recv" if we can't get the statistics from the kernel.
+        * We do that because, if we *can* get the statistics from
+        * the kernel, we use "md.stat.ps_recv" and "md.stat.ps_drop"
+        * as running counts, as reading the statistics from the
+        * kernel resets the kernel statistics, and if we directly
+        * increment "md.stat.ps_recv" here, that means it will
+        * count packets *twice* on systems where we can get kernel
+        * statistics - once here, and once in pcap_stats_linux().
         */
-       handle->md.stat.ps_recv++;
+       handle->md.packets_read++;
 
        /* Call the user supplied callback function */
        callback(userdata, &pcap_header, bp);
@@ -672,6 +1502,49 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
        return 1;
 }
 
+static int
+pcap_inject_linux(pcap_t *handle, const void *buf, size_t size)
+{
+       int ret;
+
+#ifdef HAVE_PF_PACKET_SOCKETS
+       if (!handle->md.sock_packet) {
+               /* PF_PACKET socket */
+               if (handle->md.ifindex == -1) {
+                       /*
+                        * We don't support sending on the "any" device.
+                        */
+                       strlcpy(handle->errbuf,
+                           "Sending packets isn't supported on the \"any\" device",
+                           PCAP_ERRBUF_SIZE);
+                       return (-1);
+               }
+
+               if (handle->md.cooked) {
+                       /*
+                        * We don't support sending on the "any" device.
+                        *
+                        * XXX - how do you send on a bound cooked-mode
+                        * socket?
+                        * Is a "sendto()" required there?
+                        */
+                       strlcpy(handle->errbuf,
+                           "Sending packets isn't supported in cooked mode",
+                           PCAP_ERRBUF_SIZE);
+                       return (-1);
+               }
+       }
+#endif
+
+       ret = send(handle->fd, buf, size, 0);
+       if (ret == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s",
+                   pcap_strerror(errno));
+               return (-1);
+       }
+       return (ret);
+}                           
+
 /*
  *  Get the statistics for the given packet capture handle.
  *  Reports the number of dropped packets iff the kernel supports
@@ -695,6 +1568,23 @@ pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
        if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
                        &kstats, &len) > -1) {
                /*
+                * On systems where the PACKET_STATISTICS "getsockopt()"
+                * argument is supported on PF_PACKET sockets:
+                *
+                *      "ps_recv" counts only packets that *passed* the
+                *      filter, not packets that didn't pass the filter.
+                *      This includes packets later dropped because we
+                *      ran out of buffer space.
+                *
+                *      "ps_drop" counts packets dropped because we ran
+                *      out of buffer space.  It doesn't count packets
+                *      dropped by the interface driver.  It counts only
+                *      packets that passed the filter.
+                *
+                *      Both statistics include packets not yet read from
+                *      the kernel by libpcap, and thus not yet seen by
+                *      the application.
+                *
                 * In "linux/net/packet/af_packet.c", at least in the
                 * 2.4.9 kernel, "tp_packets" is incremented for every
                 * packet that passes the packet filter *and* is
@@ -717,9 +1607,15 @@ pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
                 * platforms, but the best approximation is to return
                 * "tp_packets" as the count of packets and "tp_drops"
                 * as the count of drops.
+                *
+                * Keep a running total because each call to 
+                *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
+                * resets the counters to zero.
                 */
-               handle->md.stat.ps_recv = kstats.tp_packets;
-               handle->md.stat.ps_drop = kstats.tp_drops;
+               handle->md.stat.ps_recv += kstats.tp_packets;
+               handle->md.stat.ps_drop += kstats.tp_drops;
+               *stats = handle->md.stat;
+               return 0;
        }
        else
        {
@@ -738,21 +1634,6 @@ pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
        }
 #endif
        /*
-        * On systems where the PACKET_STATISTICS "getsockopt()" argument
-        * is supported on PF_PACKET sockets:
-        *
-        *      "ps_recv" counts only packets that *passed* the filter,
-        *      not packets that didn't pass the filter.  This includes
-        *      packets later dropped because we ran out of buffer space.
-        *
-        *      "ps_drop" counts packets dropped because we ran out of
-        *      buffer space.  It doesn't count packets dropped by the
-        *      interface driver.  It counts only packets that passed
-        *      the filter.
-        *
-        *      Both statistics include packets not yet read from the
-        *      kernel by libpcap, and thus not yet seen by the application.
-        *
         * On systems where the PACKET_STATISTICS "getsockopt()" argument
         * is not supported on PF_PACKET sockets:
         *
@@ -765,8 +1646,14 @@ pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
         *
         *      "ps_recv" doesn't include packets not yet read from
         *      the kernel by libpcap.
+        *
+        * We maintain the count of packets processed by libpcap in
+        * "md.packets_read", for reasons described in the comment
+        * at the end of pcap_read_packet().  We have no idea how many
+        * packets were dropped.
         */
-       *stats = handle->md.stat;
+       stats->ps_recv = handle->md.packets_read;
+       stats->ps_drop = 0;
        return 0;
 }
 
@@ -786,6 +1673,21 @@ pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
                return (-1);
 #endif /* HAVE_DAG_API */
 
+#ifdef HAVE_SEPTEL_API
+       if (septel_platform_finddevs(alldevsp, errbuf) < 0)
+               return (-1);
+#endif /* HAVE_SEPTEL_API */
+
+#ifdef PCAP_SUPPORT_BT
+       if (bt_platform_finddevs(alldevsp, errbuf) < 0)
+               return (-1);
+#endif
+
+#ifdef PCAP_SUPPORT_USB
+       if (usb_platform_finddevs(alldevsp, errbuf) < 0)
+               return (-1);
+#endif
+
        return (0);
 }
 
@@ -793,7 +1695,8 @@ pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
  *  Attach the given BPF code to the packet capture device.
  */
 static int
-pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
+pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
+    int is_mmapped)
 {
 #ifdef SO_ATTACH_FILTER
        struct sock_fprog       fcode;
@@ -805,7 +1708,7 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
                return -1;
        if (!filter) {
                strncpy(handle->errbuf, "setfilter: No filter specified",
-                       sizeof(handle->errbuf));
+                       PCAP_ERRBUF_SIZE);
                return -1;
        }
 
@@ -833,6 +1736,7 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
                 * sake of correctness I added this check.
                 */
                fprintf(stderr, "Warning: Filter too complex for kernel\n");
+               fcode.len = 0;
                fcode.filter = NULL;
                can_filter_in_kernel = 0;
        } else
@@ -845,13 +1749,13 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
                 *
                 * Oh, and we also need to fix it up so that all "ret"
                 * instructions with non-zero operands have 65535 as the
-                * operand, and so that, if we're in cooked mode, all
-                * memory-reference instructions use special magic offsets
-                * in references to the link-layer header and assume that
-                * the link-layer payload begins at 0; "fix_program()"
-                * will do that.
+                * operand if we're not capturing in memory-mapped modee,
+                * and so that, if we're in cooked mode, all memory-reference
+                * instructions use special magic offsets in references to
+                * the link-layer header and assume that the link-layer
+                * payload begins at 0; "fix_program()" will do that.
                 */
-               switch (fix_program(handle, &fcode)) {
+               switch (fix_program(handle, &fcode, is_mmapped)) {
 
                case -1:
                default:
@@ -912,18 +1816,83 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
                reset_kernel_filter(handle);
 
        /*
-        * Free up the copy of the filter that was made by "fix_program()".
+        * Free up the copy of the filter that was made by "fix_program()".
+        */
+       if (fcode.filter != NULL)
+               free(fcode.filter);
+
+       if (err == -2)
+               /* Fatal error */
+               return -1;
+#endif /* SO_ATTACH_FILTER */
+
+       return 0;
+}
+
+static int
+pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
+{
+       return pcap_setfilter_linux_common(handle, filter, 0);
+}
+
+
+/*
+ * Set direction flag: Which packets do we accept on a forwarding
+ * single device? IN, OUT or both?
+ */
+static int
+pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
+{
+#ifdef HAVE_PF_PACKET_SOCKETS
+       if (!handle->md.sock_packet) {
+               handle->direction = d;
+               return 0;
+       }
+#endif
+       /*
+        * We're not using PF_PACKET sockets, so we can't determine
+        * the direction of the packet.
         */
-       if (fcode.filter != NULL)
-               free(fcode.filter);
+       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+           "Setting direction is not supported on SOCK_PACKET sockets");
+       return -1;
+}
 
-       if (err == -2)
-               /* Fatal error */
-               return -1;
-#endif /* SO_ATTACH_FILTER */
 
-       return 0;
+#ifdef HAVE_PF_PACKET_SOCKETS
+/*
+ * Map the PACKET_ value to a LINUX_SLL_ value; we
+ * want the same numerical value to be used in
+ * the link-layer header even if the numerical values
+ * for the PACKET_ #defines change, so that programs
+ * that look at the packet type field will always be
+ * able to handle DLT_LINUX_SLL captures.
+ */
+static short int
+map_packet_type_to_sll_type(short int sll_pkttype)
+{
+       switch (sll_pkttype) {
+
+       case PACKET_HOST:
+               return htons(LINUX_SLL_HOST);
+
+       case PACKET_BROADCAST:
+               return htons(LINUX_SLL_BROADCAST);
+
+       case PACKET_MULTICAST:
+               return  htons(LINUX_SLL_MULTICAST);
+
+       case PACKET_OTHERHOST:
+               return htons(LINUX_SLL_OTHERHOST);
+
+       case PACKET_OUTGOING:
+               return htons(LINUX_SLL_OUTGOING);
+
+       default:
+               return -1;
+       }
 }
+#endif
 
 /*
  *  Linux uses the ARP hardware type to identify the type of an
@@ -948,6 +1917,34 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
        switch (arptype) {
 
        case ARPHRD_ETHER:
+               /*
+                * This is (presumably) a real Ethernet capture; give it a
+                * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
+                * that an application can let you choose it, in case you're
+                * capturing DOCSIS traffic that a Cisco Cable Modem
+                * Termination System is putting out onto an Ethernet (it
+                * doesn't put an Ethernet header onto the wire, it puts raw
+                * DOCSIS frames out on the wire inside the low-level
+                * Ethernet framing).
+                *
+                * XXX - are there any sorts of "fake Ethernet" that have
+                * ARPHRD_ETHER but that *shouldn't offer DLT_DOCSIS as
+                * a Cisco CMTS won't put traffic onto it or get traffic
+                * bridged onto it?  ISDN is handled in "activate_new()",
+                * as we fall back on cooked mode there; are there any
+                * others?
+                */
+               handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
+               /*
+                * If that fails, just leave the list empty.
+                */
+               if (handle->dlt_list != NULL) {
+                       handle->dlt_list[0] = DLT_EN10MB;
+                       handle->dlt_list[1] = DLT_DOCSIS;
+                       handle->dlt_count = 2;
+               }
+               /* FALLTHROUGH */
+
        case ARPHRD_METRICOM:
        case ARPHRD_LOOPBACK:
                handle->linktype = DLT_EN10MB;
@@ -959,7 +1956,7 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
                break;
 
        case ARPHRD_AX25:
-               handle->linktype = DLT_AX25;
+               handle->linktype = DLT_AX25_KISS;
                break;
 
        case ARPHRD_PRONET:
@@ -1051,6 +2048,13 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
                handle->linktype = DLT_PRISM_HEADER;
                break;
 
+#ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
+#define ARPHRD_IEEE80211_RADIOTAP 803
+#endif
+       case ARPHRD_IEEE80211_RADIOTAP:
+               handle->linktype = DLT_IEEE802_11_RADIO;
+               break;
+
        case ARPHRD_PPP:
                /*
                 * Some PPP code in the kernel supplies no link-layer
@@ -1163,6 +2167,9 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
                handle->linktype = DLT_IP_OVER_FC;
                break;
 
+#ifndef ARPHRD_IRDA
+#define ARPHRD_IRDA    783
+#endif
        case ARPHRD_IRDA:
                /* Don't expect IP packet out of this interfaces... */
                handle->linktype = DLT_LINUX_IRDA;
@@ -1171,215 +2178,931 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
                //handle->md.cooked = 1;
                break;
 
+       /* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
+        * is needed, please report it to <[email protected]> */
+#ifndef ARPHRD_LAPD
+#define ARPHRD_LAPD    8445
+#endif
+       case ARPHRD_LAPD:
+               /* Don't expect IP packet out of this interfaces... */
+               handle->linktype = DLT_LINUX_LAPD;
+               break;
+
+#ifndef ARPHRD_NONE
+#define ARPHRD_NONE    0xFFFE
+#endif
+       case ARPHRD_NONE:
+               /*
+                * No link-layer header; packets are just IP
+                * packets, so use DLT_RAW.
+                */
+               handle->linktype = DLT_RAW;
+               break;
+
        default:
                handle->linktype = -1;
                break;
        }
 }
 
-/* ===== Functions to interface to the newer kernels ================== */
-
-/*
- *  Try to open a packet socket using the new kernel interface.
- *  Returns 0 on failure.
- *  FIXME: 0 uses to mean success (Sebastian)
- */
+/* ===== Functions to interface to the newer kernels ================== */
+
+/*
+ * Try to open a packet socket using the new kernel PF_PACKET interface.
+ * Returns 1 on success, 0 on an error that means the new interface isn't
+ * present (so the old SOCK_PACKET interface should be tried), and a
+ * PCAP_ERROR_ value on an error that means that the old mechanism won't
+ * work either (so it shouldn't be tried).
+ */
+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;
+#ifdef HAVE_PACKET_AUXDATA
+       int                     val;
+#endif
+       int                     err = 0;
+       struct packet_mreq      mr;
+
+       /*
+        * 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 = 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",
+                        pcap_strerror(errno) );
+               return 0;       /* try old mechanism */
+       }
+
+       /* It seems the kernel supports the new interface. */
+       handle->md.sock_packet = 0;
+
+       /*
+        * Get the interface index of the loopback device.
+        * If the attempt fails, don't fail, just set the
+        * "md.lo_ifindex" to -1.
+        *
+        * XXX - can there be more than one device that loops
+        * packets back, i.e. devices other than "lo"?  If so,
+        * we'd need to find them all, and have an array of
+        * indices for them, and check all of them in
+        * "pcap_read_packet()".
+        */
+       handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
+
+       /*
+        * Default value for offset to align link-layer payload
+        * on a 4-byte boundary.
+        */
+       handle->offset   = 0;
+
+       /*
+        * What kind of frames do we have to deal with? Fall back
+        * to cooked mode if we have an unknown interface type
+        * or a type we know doesn't work well in raw mode.
+        */
+       if (!is_any_device) {
+               /* Assume for now we don't need cooked mode. */
+               handle->md.cooked = 0;
+
+               if (handle->opt.rfmon) {
+                       /*
+                        * We were asked to turn on monitor mode.
+                        * Do so before we get the link-layer type,
+                        * because entering monitor mode could change
+                        * the link-layer type.
+                        */
+                       err = enter_rfmon_mode(handle, sock_fd, device);
+                       if (err < 0) {
+                               /* Hard failure */
+                               close(sock_fd);
+                               return err;
+                       }
+                       if (err == 0) {
+                               /*
+                                * Nothing worked for turning monitor mode
+                                * on.
+                                */
+                               close(sock_fd);
+                               return PCAP_ERROR_RFMON_NOTSUP;
+                       }
+
+                       /*
+                        * Either monitor mode has been turned on for
+                        * the device, or we've been given a different
+                        * device to open for monitor mode.  If we've
+                        * been given a different device, use it.
+                        */
+                       if (handle->md.mondevice != NULL)
+                               device = handle->md.mondevice;
+               }
+               arptype = iface_get_arptype(sock_fd, device, handle->errbuf);
+               if (arptype < 0) {
+                       close(sock_fd);
+                       return arptype;
+               }
+               map_arphrd_to_dlt(handle, arptype, 1);
+               if (handle->linktype == -1 ||
+                   handle->linktype == DLT_LINUX_SLL ||
+                   handle->linktype == DLT_LINUX_IRDA ||
+                   handle->linktype == DLT_LINUX_LAPD ||
+                   (handle->linktype == DLT_EN10MB &&
+                    (strncmp("isdn", device, 4) == 0 ||
+                     strncmp("isdY", device, 4) == 0))) {
+                       /*
+                        * Unknown interface type (-1), or a
+                        * device we explicitly chose to run
+                        * in cooked mode (e.g., PPP devices),
+                        * or an ISDN device (whose link-layer
+                        * type we can only determine by using
+                        * APIs that may be different on different
+                        * kernels) - reopen in cooked mode.
+                        */
+                       if (close(sock_fd) == -1) {
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                        "close: %s", pcap_strerror(errno));
+                               return PCAP_ERROR;
+                       }
+                       sock_fd = socket(PF_PACKET, SOCK_DGRAM,
+                           htons(ETH_P_ALL));
+                       if (sock_fd == -1) {
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                   "socket: %s", pcap_strerror(errno));
+                               return PCAP_ERROR;
+                       }
+                       handle->md.cooked = 1;
+
+                       /*
+                        * Get rid of any link-layer type list
+                        * we allocated - this only supports cooked
+                        * capture.
+                        */
+                       if (handle->dlt_list != NULL) {
+                               free(handle->dlt_list);
+                               handle->dlt_list = NULL;
+                               handle->dlt_count = 0;
+                       }
+
+                       if (handle->linktype == -1) {
+                               /*
+                                * Warn that we're falling back on
+                                * cooked mode; we may want to
+                                * update "map_arphrd_to_dlt()"
+                                * to handle the new type.
+                                */
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                       "arptype %d not "
+                                       "supported by libpcap - "
+                                       "falling back to cooked "
+                                       "socket",
+                                       arptype);
+                       }
+
+                       /*
+                        * IrDA capture is not a real "cooked" capture,
+                        * it's IrLAP frames, not IP packets.  The
+                        * same applies to LAPD capture.
+                        */
+                       if (handle->linktype != DLT_LINUX_IRDA &&
+                           handle->linktype != DLT_LINUX_LAPD)
+                               handle->linktype = DLT_LINUX_SLL;
+               }
+
+               handle->md.ifindex = iface_get_id(sock_fd, device,
+                   handle->errbuf);
+               if (handle->md.ifindex == -1) {
+                       close(sock_fd);
+                       return PCAP_ERROR;
+               }
+
+               if ((err = iface_bind(sock_fd, handle->md.ifindex,
+                   handle->errbuf)) != 1) {
+                       close(sock_fd);
+                       if (err < 0)
+                               return err;
+                       else
+                               return 0;       /* try old mechanism */
+               }
+       } else {
+               /*
+                * 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.
+                * 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
+                * mode.
+                */
+               handle->md.ifindex = -1;
+       }
+
+       /*
+        * Select promiscuous mode on if "promisc" is set.
+        *
+        * Do not turn allmulti mode on if we don't select
+        * promiscuous mode - on some devices (e.g., Orinoco
+        * wireless interfaces), allmulti mode isn't supported
+        * and the driver implements it by turning promiscuous
+        * mode on, and that screws up the operation of the
+        * card as a normal networking interface, and on no
+        * other platform I know of does starting a non-
+        * promiscuous capture affect which multicast packets
+        * are received by the interface.
+        */
+
+       /*
+        * Hmm, how can we set promiscuous mode on all interfaces?
+        * 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 (!is_any_device && handle->opt.promisc) {
+               memset(&mr, 0, sizeof(mr));
+               mr.mr_ifindex = handle->md.ifindex;
+               mr.mr_type    = PACKET_MR_PROMISC;
+               if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
+                   &mr, sizeof(mr)) == -1) {
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                               "setsockopt: %s", pcap_strerror(errno));
+                       close(sock_fd);
+                       return PCAP_ERROR;
+               }
+       }
+
+       /* Enable auxillary data if supported and reserve room for
+        * reconstructing VLAN headers. */
+#ifdef HAVE_PACKET_AUXDATA
+       val = 1;
+       if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
+                      sizeof(val)) == -1 && errno != ENOPROTOOPT) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "setsockopt: %s", pcap_strerror(errno));
+               close(sock_fd);
+               return PCAP_ERROR;
+       }
+       handle->offset += VLAN_TAG_LEN;
+#endif /* HAVE_PACKET_AUXDATA */
+
+       /*
+        * This is a 2.2[.x] or later kernel (we know that
+        * because we're not using a SOCK_PACKET socket -
+        * PF_PACKET is supported only in 2.2 and later
+        * kernels).
+        *
+        * We can safely pass "recvfrom()" a byte count
+        * based on the snapshot length.
+        *
+        * If we're in cooked mode, make the snapshot length
+        * large enough to hold a "cooked mode" header plus
+        * 1 byte of packet data (so we don't pass a byte
+        * count of 0 to "recvfrom()").
+        */
+       if (handle->md.cooked) {
+               if (handle->snapshot < SLL_HDR_LEN + 1)
+                       handle->snapshot = SLL_HDR_LEN + 1;
+       }
+       handle->bufsize = handle->snapshot;
+
+       /* Save the socket FD in the pcap structure */
+       handle->fd = sock_fd;
+
+       return 1;
+#else
+       strncpy(ebuf,
+               "New packet capturing interface not supported by build "
+               "environment", PCAP_ERRBUF_SIZE);
+       return 0;
+#endif
+}
+
+static int 
+activate_mmap(pcap_t *handle)
+{
+#ifdef HAVE_PACKET_RING
+       int ret;
+
+       /*
+        * Attempt to allocate a buffer to hold the contents of one
+        * packet, for use by the oneshot callback.
+        */
+       handle->md.oneshot_buffer = malloc(handle->snapshot);
+       if (handle->md.oneshot_buffer == NULL) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "can't allocate oneshot buffer: %s",
+                        pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+
+       if (handle->opt.buffer_size == 0) {
+               /* by default request 2M for the ring buffer */
+               handle->opt.buffer_size = 2*1024*1024;
+       }
+       ret = prepare_tpacket_socket(handle);
+       if (ret != 1) {
+               free(handle->md.oneshot_buffer);
+               return ret;
+       }
+       ret = create_ring(handle);
+       if (ret != 1) {
+               free(handle->md.oneshot_buffer);
+               return ret;
+       }
+
+       /* override some defaults and inherit the other fields from
+        * activate_new
+        * handle->offset is used to get the current position into the rx ring 
+        * handle->cc is used to store the ring size */
+       handle->read_op = pcap_read_linux_mmap;
+       handle->cleanup_op = pcap_cleanup_linux_mmap;
+       handle->setfilter_op = pcap_setfilter_linux_mmap;
+       handle->setnonblock_op = pcap_setnonblock_mmap;
+       handle->getnonblock_op = pcap_getnonblock_mmap;
+       handle->oneshot_callback = pcap_oneshot_mmap;
+       handle->selectable_fd = handle->fd;
+       return 1;
+#else /* HAVE_PACKET_RING */
+       return 0;
+#endif /* HAVE_PACKET_RING */
+}
+
+#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 int
+create_ring(pcap_t *handle)
+{
+       unsigned i, j, frames_per_block;
+       struct tpacket_req req;
+
+       /* Note that with large snapshot (say 64K) only a few frames 
+        * will be available in the ring even with pretty large ring size
+        * (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_ALIGN(handle->md.tp_hdrlen) +
+                                         sizeof(struct sockaddr_ll));
+       req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
+
+       /* compute the minumum block size that will handle this frame. 
+        * The block has to be page size aligned. 
+        * The max block size allowed by the kernel is arch-dependent and 
+        * it's not explicitly checked here. */
+       req.tp_block_size = getpagesize();
+       while (req.tp_block_size < req.tp_frame_size) 
+               req.tp_block_size <<= 1;
+
+       frames_per_block = req.tp_block_size/req.tp_frame_size;
+
+       /* ask the kernel to create the ring */
+retry:
+       req.tp_block_nr = req.tp_frame_nr / frames_per_block;
+
+       /* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
+       req.tp_frame_nr = req.tp_block_nr * frames_per_block;
+       
+       if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
+                                       (void *) &req, sizeof(req))) {
+               if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
+                       /*
+                        * Memory failure; try to reduce the requested ring
+                        * size.
+                        *
+                        * We used to reduce this by half -- do 5% instead.
+                        * That may result in more iterations and a longer
+                        * startup, but the user will be much happier with
+                        * the resulting buffer size.
+                        */
+                       if (req.tp_frame_nr < 20)
+                               req.tp_frame_nr -= 1;
+                       else
+                               req.tp_frame_nr -= req.tp_frame_nr/20;
+                       goto retry;
+               }
+               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 */
+       handle->md.mmapbuflen = req.tp_block_nr * req.tp_block_size;
+       handle->md.mmapbuf = mmap(0, handle->md.mmapbuflen,
+           PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
+       if (handle->md.mmapbuf == MAP_FAILED) {
+               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 -1;
+       }
+
+       /* allocate a ring for each frame header pointer*/
+       handle->cc = req.tp_frame_nr;
+       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 -1;
+       }
+
+       /* fill the header ring with proper frame ptr*/
+       handle->offset = 0;
+       for (i=0; i<req.tp_block_nr; ++i) {
+               void *base = &handle->md.mmapbuf[i*req.tp_block_size];
+               for (j=0; j<frames_per_block; ++j, ++handle->offset) {
+                       RING_GET_FRAME(handle) = base;
+                       base += req.tp_frame_size;
+               }
+       }
+
+       handle->bufsize = req.tp_frame_size;
+       handle->offset = 0;
+       return 1;
+}
+
+/* free all ring related resources*/
+static void
+destroy_ring(pcap_t *handle)
+{
+       /* tell the kernel to destroy the ring*/
+       struct tpacket_req req;
+       memset(&req, 0, sizeof(req));
+       setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
+                               (void *) &req, sizeof(req));
+
+       /* if ring is mapped, unmap it*/
+       if (handle->md.mmapbuf) {
+               /* do not test for mmap failure, as we can't recover from any error */
+               munmap(handle->md.mmapbuf, handle->md.mmapbuflen);
+               handle->md.mmapbuf = NULL;
+       }
+}
+
+/*
+ * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
+ * for Linux mmapped capture.
+ *
+ * The problem is that pcap_next() and pcap_next_ex() expect the packet
+ * data handed to the callback to be valid after the callback returns,
+ * but pcap_read_linux_mmap() has to release that packet as soon as
+ * the callback returns (otherwise, the kernel thinks there's still
+ * at least one unprocessed packet available in the ring, so a select()
+ * will immediately return indicating that there's data to process), so,
+ * in the callback, we have to make a copy of the packet.
+ *
+ * Yes, this means that, if the capture is using the ring buffer, using
+ * pcap_next() or pcap_next_ex() requires more copies than using
+ * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
+ * pcap_next() or pcap_next_ex().
+ */
+static void
+pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
+    const u_char *bytes)
+{
+       struct pkt_for_oneshot *sp = (struct pkt_for_oneshot *)user;
+       bpf_u_int32 copylen;
+
+       *sp->hdr = *h;
+       memcpy(sp->pd->md.oneshot_buffer, bytes, h->caplen);
+       *sp->pkt = sp->pd->md.oneshot_buffer;
+}
+    
+static void
+pcap_cleanup_linux_mmap( pcap_t *handle )
+{
+       destroy_ring(handle);
+       if (handle->md.oneshot_buffer != NULL) {
+               free(handle->md.oneshot_buffer);
+               handle->md.oneshot_buffer = NULL;
+       }
+       pcap_cleanup_linux(handle);
+}
+
+
+static int
+pcap_getnonblock_mmap(pcap_t *p, char *errbuf)
+{
+       /* use negative value of timeout to indicate non blocking ops */
+       return (p->md.timeout<0);
+}
+
+static int
+pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
+{
+       /* map each value to the corresponding 2's complement, to 
+        * preserve the timeout value provided with pcap_set_timeout */
+       if (nonblock) {
+               if (p->md.timeout >= 0) {
+                       /*
+                        * Timeout is non-negative, so we're not already
+                        * in non-blocking mode; set it to the 2's
+                        * complement, to make it negative, as an
+                        * indication that we're in non-blocking mode.
+                        */
+                       p->md.timeout = p->md.timeout*-1 - 1;
+               }
+       } else {
+               if (p->md.timeout < 0) {
+                       /*
+                        * Timeout is negative, so we're not already
+                        * in blocking mode; reverse the previous
+                        * operation, to make the timeout non-negative
+                        * again.
+                        */
+                       p->md.timeout = (p->md.timeout+1)*-1;
+               }
+       }
+       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
-live_open_new(pcap_t *handle, const char *device, int promisc,
-             int to_ms, char *ebuf)
+pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback, 
+               u_char *user)
 {
-#ifdef HAVE_PF_PACKET_SOCKETS
-       int                     sock_fd = -1, device_id, arptype;
-       int                     err;
-       int                     fatal_err = 0;
-       struct packet_mreq      mr;
-
-       /* One shot loop used for error handling - bail out with break */
+       int pkts = 0;
+
+       /* wait for frames availability.*/
+       if ((handle->md.timeout >= 0) &&
+           !pcap_get_ring_frame(handle, TP_STATUS_USER)) {
+               struct pollfd pollinfo;
+               int ret;
+
+               pollinfo.fd = handle->fd;
+               pollinfo.events = POLLIN;
+
+               do {
+                       /* poll() requires a negative timeout to wait forever */
+                       ret = poll(&pollinfo, 1, (handle->md.timeout > 0)?
+                                               handle->md.timeout: -1);
+                       if ((ret < 0) && (errno != EINTR)) {
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
+                                        "can't poll on packet socket fd %d: %d-%s",
+                                       handle->fd, errno, pcap_strerror(errno));
+                               return -1;
+                       }
+                       /* check for break loop condition on interrupted syscall*/
+                       if (handle->break_loop) {
+                               handle->break_loop = 0;
+                               return -2;
+                       }
+               } while (ret < 0);
+       }
 
-       do {
-               /*
-                * Open a socket with protocol family packet. If a device is
-                * given we try to open it in raw mode otherwise we use
-                * the cooked interface.
-                */
-               sock_fd = device ?
-                       socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))
-                     : socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL));
+       /* non-positive values of max_packets are used to require all 
+        * packets currently available in the ring */
+       while ((pkts < max_packets) || (max_packets <= 0)) {
+               int run_bpf;
+               struct sockaddr_ll *sll;
+               struct pcap_pkthdr pcaphdr;
+               unsigned char *bp;
+               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;
 
-               if (sock_fd == -1) {
-                       snprintf(ebuf, PCAP_ERRBUF_SIZE, "socket: %s",
-                                pcap_strerror(errno) );
+               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 (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", 
+                               tp_mac, tp_snaplen, handle->bufsize);
+                       return -1;
                }
 
-               /* It seems the kernel supports the new interface. */
-               handle->md.sock_packet = 0;
-
-               /*
-                * Get the interface index of the loopback device.
-                * If the attempt fails, don't fail, just set the
-                * "md.lo_ifindex" to -1.
-                *
-                * XXX - can there be more than one device that loops
-                * packets back, i.e. devices other than "lo"?  If so,
-                * we'd need to find them all, and have an array of
-                * indices for them, and check all of them in
-                * "pcap_read_packet()".
-                */
-               handle->md.lo_ifindex = iface_get_id(sock_fd, "lo", ebuf);
-
-               /*
-                * Default value for offset to align link-layer payload
-                * on a 4-byte boundary.
-                */
-               handle->offset   = 0;
+               /* run filter on received packet
+                * If the kernel filtering is enabled we need to run the
+                * filter until all the frames present into the ring 
+                * at filter creation time are processed. 
+                * In such case md.use_bpf is used as a counter for the 
+                * packet we need to filter.
+                * 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*)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,
+                                       tp_len, tp_snaplen) == 0))
+                       goto skip;
 
                /*
-                * What kind of frames do we have to deal with? Fall back
-                * to cooked mode if we have an unknown interface type.
+                * Do checks based on packet direction.
                 */
+               sll = (void *)h.raw + TPACKET_ALIGN(handle->md.tp_hdrlen);
+               if (sll->sll_pkttype == PACKET_OUTGOING) {
+                       /*
+                        * Outgoing packet.
+                        * If this is from the loopback device, reject it;
+                        * we'll see the packet as an incoming packet as well,
+                        * and we don't want to see it twice.
+                        */
+                       if (sll->sll_ifindex == handle->md.lo_ifindex)
+                               goto skip;
 
-               if (device) {
-                       /* Assume for now we don't need cooked mode. */
-                       handle->md.cooked = 0;
+                       /*
+                        * If the user only wants incoming packets, reject it.
+                        */
+                       if (handle->direction == PCAP_D_IN)
+                               goto skip;
+               } else {
+                       /*
+                        * Incoming packet.
+                        * If the user only wants outgoing packets, reject it.
+                        */
+                       if (handle->direction == PCAP_D_OUT)
+                               goto skip;
+               }
 
-                       arptype = iface_get_arptype(sock_fd, device, ebuf);
-                       if (arptype == -1) {
-                               fatal_err = 1;
-                               break;
-                       }
-                       map_arphrd_to_dlt(handle, arptype, 1);
-                       if (handle->linktype == -1 ||
-                           handle->linktype == DLT_LINUX_SLL ||
-                           handle->linktype == DLT_LINUX_IRDA ||
-                           (handle->linktype == DLT_EN10MB &&
-                            (strncmp("isdn", device, 4) == 0 ||
-                             strncmp("isdY", device, 4) == 0))) {
-                               /*
-                                * Unknown interface type (-1), or a
-                                * device we explicitly chose to run
-                                * in cooked mode (e.g., PPP devices),
-                                * or an ISDN device (whose link-layer
-                                * type we can only determine by using
-                                * APIs that may be different on different
-                                * kernels) - reopen in cooked mode.
-                                */
-                               if (close(sock_fd) == -1) {
-                                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                                "close: %s", pcap_strerror(errno));
-                                       break;
-                               }
-                               sock_fd = socket(PF_PACKET, SOCK_DGRAM,
-                                                htons(ETH_P_ALL));
-                               if (sock_fd == -1) {
-                                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                                "socket: %s", pcap_strerror(errno));
-                                       break;
-                               }
-                               handle->md.cooked = 1;
+               /* get required packet info from ring header */
+               pcaphdr.ts.tv_sec = tp_sec;
+               pcaphdr.ts.tv_usec = tp_usec;
+               pcaphdr.caplen = tp_snaplen;
+               pcaphdr.len = tp_len;
 
-                               if (handle->linktype == -1) {
-                                       /*
-                                        * Warn that we're falling back on
-                                        * cooked mode; we may want to
-                                        * update "map_arphrd_to_dlt()"
-                                        * to handle the new type.
-                                        */
-                                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                               "arptype %d not "
-                                               "supported by libpcap - "
-                                               "falling back to cooked "
-                                               "socket",
-                                               arptype);
-                               }
-                               /* IrDA capture is not a real "cooked" capture,
-                                * it's IrLAP frames, not IP packets. */
-                               if(handle->linktype != DLT_LINUX_IRDA)
-                                       handle->linktype = DLT_LINUX_SLL;
-                       }
+               /* if required build in place the sll header*/
+               if (handle->md.cooked) {
+                       struct sll_header *hdrp;
 
-                       device_id = iface_get_id(sock_fd, device, ebuf);
-                       if (device_id == -1)
-                               break;
+                       /*
+                        * The kernel should have left us with enough
+                        * space for an sll header; back up the packet
+                        * data pointer into that space, as that'll be
+                        * the beginning of the packet we pass to the
+                        * callback.
+                        */
+                       bp -= SLL_HDR_LEN;
 
-                       if ((err = iface_bind(sock_fd, device_id, ebuf)) < 0) {
-                               if (err == -2)
-                                       fatal_err = 1;
-                               break;
-                       }
-               } else {
                        /*
-                        * This is cooked mode.
+                        * Let's make sure that's past the end of
+                        * the tpacket header, i.e. >=
+                        * ((u_char *)thdr + TPACKET_HDRLEN), so we
+                        * don't step on the header when we construct
+                        * the sll header.
                         */
-                       handle->md.cooked = 1;
-                       handle->linktype = DLT_LINUX_SLL;
+                       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;
+                       }
 
                        /*
-                        * XXX - squelch GCC complaints about
-                        * uninitialized variables; if we can't
-                        * select promiscuous mode on all interfaces,
-                        * we should move the code below into the
-                        * "if (device)" branch of the "if" and
-                        * get rid of the next statement.
+                        * OK, that worked; construct the sll header.
                         */
-                       device_id = -1;
+                       hdrp = (struct sll_header *)bp;
+                       hdrp->sll_pkttype = map_packet_type_to_sll_type(
+                                                       sll->sll_pkttype);
+                       hdrp->sll_hatype = htons(sll->sll_hatype);
+                       hdrp->sll_halen = htons(sll->sll_halen);
+                       memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
+                       hdrp->sll_protocol = sll->sll_protocol;
+
+                       /* update packet len */
+                       pcaphdr.caplen += SLL_HDR_LEN;
+                       pcaphdr.len += SLL_HDR_LEN;
                }
 
-               /*
-                * Select promiscuous mode on if "promisc" is set.
-                *
-                * Do not turn allmulti mode on if we don't select
-                * promiscuous mode - on some devices (e.g., Orinoco
-                * wireless interfaces), allmulti mode isn't supported
-                * and the driver implements it by turning promiscuous
-                * mode on, and that screws up the operation of the
-                * card as a normal networking interface, and on no
-                * other platform I know of does starting a non-
-                * promiscuous capture affect which multicast packets
-                * are received by the interface.
-                */
+#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
 
                /*
-                * Hmm, how can we set promiscuous mode on all interfaces?
-                * I am not sure if that is possible at all.
+                * The only way to tell the kernel to cut off the
+                * packet at a snapshot length is with a filter program;
+                * if there's no filter program, the kernel won't cut
+                * the packet off.
+                *
+                * Trim the snapshot length to be no longer than the
+                * specified snapshot length.
                 */
-
-               if (device && promisc) {
-                       memset(&mr, 0, sizeof(mr));
-                       mr.mr_ifindex = device_id;
-                       mr.mr_type    = PACKET_MR_PROMISC;
-                       if (setsockopt(sock_fd, SOL_PACKET,
-                               PACKET_ADD_MEMBERSHIP, &mr, sizeof(mr)) == -1)
-                       {
-                               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                       "setsockopt: %s", pcap_strerror(errno));
-                               break;
-                       }
+               if (pcaphdr.caplen > handle->snapshot)
+                       pcaphdr.caplen = handle->snapshot;
+
+               /* pass the packet to the user */
+               pkts++;
+               callback(user, &pcaphdr, bp);
+               handle->md.packets_read++;
+
+skip:
+               /* next packet */
+               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;
 
-               /* Save the socket FD in the pcap structure */
+               /* check for break loop condition*/
+               if (handle->break_loop) {
+                       handle->break_loop = 0;
+                       return -2;
+               }
+       }
+       return pkts;
+}
 
-               handle->fd       = sock_fd;
+static int 
+pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
+{
+       int n, offset;
+       int ret;
 
-               return 1;
+       /*
+        * Don't rewrite "ret" instructions; we don't need to, as
+        * we're not reading packets with recvmsg(), and we don't
+        * want to, as, by not rewriting them, the kernel can avoid
+        * copying extra data.
+        */
+       ret = pcap_setfilter_linux_common(handle, filter, 1);
+       if (ret < 0)
+               return ret;
 
-       } while(0);
+       /* if the kernel filter is enabled, we need to apply the filter on
+        * all packets present into the ring. Get an upper bound of their number
+        */
+       if (!handle->md.use_bpf)
+               return ret;
+
+       /* walk the ring backward and count the free slot */
+       offset = handle->offset;
+       if (--handle->offset < 0)
+               handle->offset = handle->cc - 1;
+       for (n=0; n < handle->cc; ++n) {
+               if (--handle->offset < 0)
+                       handle->offset = handle->cc - 1;
+               if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
+                       break;
+       }
 
-       if (sock_fd != -1)
-               close(sock_fd);
+       /* be careful to not change current ring position */
+       handle->offset = offset;
 
-       if (fatal_err)
-               return -2;
-       else
-               return 0;
-#else
-       strncpy(ebuf,
-               "New packet capturing interface not supported by build "
-               "environment", PCAP_ERRBUF_SIZE);
-       return 0;
-#endif
+       /* store the number of packets currently present in the ring */
+       handle->md.use_bpf = 1 + (handle->cc - n);
+       return ret;
 }
 
+#endif /* HAVE_PACKET_RING */
+
+
 #ifdef HAVE_PF_PACKET_SOCKETS
 /*
  *  Return the index of the given device name. Fill ebuf and return
@@ -1395,7 +3118,7 @@ iface_get_id(int fd, const char *device, char *ebuf)
 
        if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "ioctl: %s", pcap_strerror(errno));
+                        "SIOCGIFINDEX: %s", pcap_strerror(errno));
                return -1;
        }
 
@@ -1404,6 +3127,8 @@ iface_get_id(int fd, const char *device, char *ebuf)
 
 /*
  *  Bind the socket associated with FD to the given device.
+ *  Return 1 on success, 0 if we should try a SOCK_PACKET socket,
+ *  or a PCAP_ERROR_ value on a hard error.
  */
 static int
 iface_bind(int fd, int ifindex, char *ebuf)
@@ -1412,270 +3137,882 @@ iface_bind(int fd, int ifindex, char *ebuf)
        int                     err;
        socklen_t               errlen = sizeof(err);
 
-       memset(&sll, 0, sizeof(sll));
-       sll.sll_family          = AF_PACKET;
-       sll.sll_ifindex         = ifindex;
-       sll.sll_protocol        = htons(ETH_P_ALL);
+       memset(&sll, 0, sizeof(sll));
+       sll.sll_family          = AF_PACKET;
+       sll.sll_ifindex         = ifindex;
+       sll.sll_protocol        = htons(ETH_P_ALL);
+
+       if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
+               if (errno == ENETDOWN) {
+                       /*
+                        * Return a "network down" indication, so that
+                        * the application can report that rather than
+                        * saying we had a mysterious failure and
+                        * suggest that they report a problem to the
+                        * libpcap developers.
+                        */
+                       return PCAP_ERROR_IFACE_NOT_UP;
+               } else {
+                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
+                                "bind: %s", pcap_strerror(errno));
+                       return PCAP_ERROR;
+               }
+       }
+
+       /* Any pending errors, e.g., network is down? */
+
+       if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
+               snprintf(ebuf, PCAP_ERRBUF_SIZE,
+                       "getsockopt: %s", pcap_strerror(errno));
+               return 0;
+       }
+
+       if (err == ENETDOWN) {
+               /*
+                * Return a "network down" indication, so that
+                * the application can report that rather than
+                * saying we had a mysterious failure and
+                * suggest that they report a problem to the
+                * libpcap developers.
+                */
+               return PCAP_ERROR_IFACE_NOT_UP;
+       } else if (err > 0) {
+               snprintf(ebuf, PCAP_ERRBUF_SIZE,
+                       "bind: %s", pcap_strerror(err));
+               return 0;
+       }
+
+       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
+ * if the device doesn't even exist.
+ */
+static int
+has_wext(int sock_fd, const char *device, char *ebuf)
+{
+       struct iwreq ireq;
+
+       strncpy(ireq.ifr_ifrn.ifrn_name, device,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
+               return 1;       /* yes */
+       snprintf(ebuf, PCAP_ERRBUF_SIZE,
+           "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
+       if (errno == ENODEV)
+               return PCAP_ERROR_NO_SUCH_DEVICE;
+       return 0;
+}
+
+/*
+ * 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,
+       MONITOR_HOSTAP,
+       MONITOR_PRISM,
+       MONITOR_PRISM54,
+       MONITOR_ACX100,
+       MONITOR_RT2500,
+       MONITOR_RT2570,
+       MONITOR_RT73,
+       MONITOR_RTL8XXX
+} monitor_type;
+
+/*
+ * Use the Wireless Extensions, if we have them, to try to turn monitor mode
+ * on if it's not already on.
+ *
+ * Returns 1 on success, 0 if we don't support the Wireless Extensions
+ * on this device, or a PCAP_ERROR_ value if we do support them but
+ * we weren't able to turn monitor mode on.
+ */
+static int
+enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
+{
+       /*
+        * XXX - at least some adapters require non-Wireless Extensions
+        * mechanisms to turn monitor mode on.
+        *
+        * 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;
+       struct iw_priv_args *priv;
+       monitor_type montype;
+       int i;
+       __u32 cmd;
+       int args[2];
+       int channel;
+
+       /*
+        * Does this device *support* the Wireless Extensions?
+        */
+       err = has_wext(sock_fd, device, handle->errbuf);
+       if (err <= 0)
+               return err;     /* either it doesn't or the device doesn't even exist */
+       /*
+        * Try to get all the Wireless Extensions private ioctls
+        * supported by this device.
+        *
+        * First, get the size of the buffer we need, by supplying no
+        * buffer and a length of 0.  If the device supports private
+        * ioctls, it should return E2BIG, with ireq.u.data.length set
+        * to the length we need.  If it doesn't support them, it should
+        * return EOPNOTSUPP.
+        */
+       memset(&ireq, 0, sizeof ireq);
+       strncpy(ireq.ifr_ifrn.ifrn_name, device,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       ireq.u.data.pointer = (void *)args;
+       ireq.u.data.length = 0;
+       ireq.u.data.flags = 0;
+       if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!",
+                   device);
+               return PCAP_ERROR;
+       }
+       if (errno == EOPNOTSUPP) {
+               /*
+                * No private ioctls, so we assume that there's only one
+                * DLT_ for monitor mode.
+                */
+               return 0;
+       }
+       if (errno != E2BIG) {
+               /*
+                * Failed.
+                */
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+       priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args));
+       if (priv == NULL) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "malloc: %s", pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+       ireq.u.data.pointer = (void *)priv;
+       if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
+               free(priv);
+               return PCAP_ERROR;
+       }
+
+       /*
+        * Look for private ioctls to turn monitor mode on or, if
+        * monitor mode is on, to set the header type.
+        */
+       montype = MONITOR_WEXT;
+       cmd = 0;
+       for (i = 0; i < ireq.u.data.length; i++) {
+               if (strcmp(priv[i].name, "monitor_type") == 0) {
+                       /*
+                        * Hostap driver, use this one.
+                        * Set monitor mode first.
+                        * You can set it to 0 to get DLT_IEEE80211,
+                        * 1 to get DLT_PRISM, 2 to get
+                        * DLT_IEEE80211_RADIO_AVS, and, with more
+                        * recent versions of the driver, 3 to get
+                        * DLT_IEEE80211_RADIO.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
+                               break;
+                       if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
+                               break;
+                       montype = MONITOR_HOSTAP;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "set_prismhdr") == 0) {
+                       /*
+                        * Prism54 driver, use this one.
+                        * Set monitor mode first.
+                        * You can set it to 2 to get DLT_IEEE80211
+                        * or 3 or get DLT_PRISM.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
+                               break;
+                       if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
+                               break;
+                       montype = MONITOR_PRISM54;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "forceprismheader") == 0) {
+                       /*
+                        * RT2570 driver, use this one.
+                        * Do this after turning monitor mode on.
+                        * You can set it to 1 to get DLT_PRISM or 2
+                        * to get DLT_IEEE80211.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
+                               break;
+                       if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
+                               break;
+                       montype = MONITOR_RT2570;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "forceprism") == 0) {
+                       /*
+                        * RT73 driver, use this one.
+                        * Do this after turning monitor mode on.
+                        * Its argument is a *string*; you can
+                        * set it to "1" to get DLT_PRISM or "2"
+                        * to get DLT_IEEE80211.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR)
+                               break;
+                       if (priv[i].set_args & IW_PRIV_SIZE_FIXED)
+                               break;
+                       montype = MONITOR_RT73;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "prismhdr") == 0) {
+                       /*
+                        * One of the RTL8xxx drivers, use this one.
+                        * It can only be done after monitor mode
+                        * has been turned on.  You can set it to 1
+                        * to get DLT_PRISM or 0 to get DLT_IEEE80211.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
+                               break;
+                       if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1)
+                               break;
+                       montype = MONITOR_RTL8XXX;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "rfmontx") == 0) {
+                       /*
+                        * RT2500 or RT61 driver, use this one.
+                        * It has one one-byte parameter; set
+                        * u.data.length to 1 and u.data.pointer to
+                        * point to the parameter.
+                        * It doesn't itself turn monitor mode on.
+                        * You can set it to 1 to allow transmitting
+                        * in monitor mode(?) and get DLT_IEEE80211,
+                        * or set it to 0 to disallow transmitting in
+                        * monitor mode(?) and get DLT_PRISM.
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2)
+                               break;
+                       montype = MONITOR_RT2500;
+                       cmd = priv[i].cmd;
+                       break;
+               }
+               if (strcmp(priv[i].name, "monitor") == 0) {
+                       /*
+                        * Either ACX100 or hostap, use this one.
+                        * It turns monitor mode on.
+                        * If it takes two arguments, it's ACX100;
+                        * the first argument is 1 for DLT_PRISM
+                        * or 2 for DLT_IEEE80211, and the second
+                        * argument is the channel on which to
+                        * run.  If it takes one argument, it's
+                        * HostAP, and the argument is 2 for
+                        * DLT_IEEE80211 and 3 for DLT_PRISM.
+                        *
+                        * If we see this, we don't quit, as this
+                        * might be a version of the hostap driver
+                        * that also supports "monitor_type".
+                        */
+                       if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT)
+                               break;
+                       if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED))
+                               break;
+                       switch (priv[i].set_args & IW_PRIV_SIZE_MASK) {
+
+                       case 1:
+                               montype = MONITOR_PRISM;
+                               cmd = priv[i].cmd;
+                               break;
+
+                       case 2:
+                               montype = MONITOR_ACX100;
+                               cmd = priv[i].cmd;
+                               break;
+
+                       default:
+                               break;
+                       }
+               }
+       }
+       free(priv);
+
+       /*
+        * XXX - ipw3945?  islism?
+        */
+
+       /*
+        * Get the old mode.
+        */
+       strncpy(ireq.ifr_ifrn.ifrn_name, device,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) {
+               /*
+                * We probably won't be able to set the mode, either.
+                */
+               return PCAP_ERROR_RFMON_NOTSUP;
+       }
+
+       /*
+        * Is it currently in monitor mode?
+        */
+       if (ireq.u.mode == IW_MODE_MONITOR) {
+               /*
+                * Yes.  Just leave things as they are.
+                * We don't offer multiple link-layer types, as
+                * changing the link-layer type out from under
+                * somebody else capturing in monitor mode would
+                * be considered rude.
+                */
+               return 1;
+       }
+       /*
+        * No.  We have to put the adapter into rfmon mode.
+        */
+
+       /*
+        * If we haven't already done so, arrange to have
+        * "pcap_close_all()" called when we exit.
+        */
+       if (!pcap_do_addexit(handle)) {
+               /*
+                * "atexit()" failed; don't put the interface
+                * in rfmon mode, just give up.
+                */
+               return PCAP_ERROR_RFMON_NOTSUP;
+       }
+
+       /*
+        * Save the old mode.
+        */
+       handle->md.oldmode = ireq.u.mode;
+
+       /*
+        * Put the adapter in rfmon mode.  How we do this depends
+        * on whether we have a special private ioctl or not.
+        */
+       if (montype == MONITOR_PRISM) {
+               /*
+                * We have the "monitor" private ioctl, but none of
+                * the other private ioctls.  Use this, and select
+                * the Prism header.
+                *
+                * If it fails, just fall back on SIOCSIWMODE.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               ireq.u.data.length = 1; /* 1 argument */
+               args[0] = 3;    /* request Prism header */
+               memcpy(ireq.u.name, args, IFNAMSIZ);
+               if (ioctl(sock_fd, cmd, &ireq) != -1) {
+                       /*
+                        * Success.
+                        * Note that we have to put the old mode back
+                        * when we close the device.
+                        */
+                       handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
 
-       if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "bind: %s", pcap_strerror(errno));
-               return -1;
-       }
+                       /*
+                        * Add this to the list of pcaps to close
+                        * when we exit.
+                        */
+                       pcap_add_to_pcaps_to_close(handle);
 
-       /* Any pending errors, e.g., network is down? */
+                       return 1;
+               }
 
-       if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                       "getsockopt: %s", pcap_strerror(errno));
-               return -2;
+               /*
+                * Failure.  Fall back on SIOCSIWMODE.
+                */
        }
 
-       if (err > 0) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                       "bind: %s", pcap_strerror(err));
-               return -2;
+       /*
+        * First, turn monitor mode on.
+        */
+       strncpy(ireq.ifr_ifrn.ifrn_name, device,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       ireq.u.mode = IW_MODE_MONITOR;
+       if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) {
+               /*
+                * Scientist, you've failed.
+                */
+               return PCAP_ERROR_RFMON_NOTSUP;
        }
 
-       return 0;
-}
+       /*
+        * XXX - airmon-ng does "iwconfig {if} key off" after setting
+        * monitor mode and setting the channel, and then does
+        * "iwconfig up".
+        */
 
-#endif
+       /*
+        * Now select the appropriate radio header.
+        */
+       switch (montype) {
 
+       case MONITOR_WEXT:
+               /*
+                * We don't have any private ioctl to set the header.
+                */
+               break;
 
-/* ===== Functions to interface to the older kernels ================== */
+       case MONITOR_HOSTAP:
+               /*
+                * Try to select the radiotap header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 3;    /* request radiotap header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               if (ioctl(sock_fd, cmd, &ireq) != -1)
+                       break;  /* success */
 
-/*
- * With older kernels promiscuous mode is kind of interesting because we
- * have to reset the interface before exiting. The problem can't really
- * be solved without some daemon taking care of managing usage counts.
- * If we put the interface into promiscuous mode, we set a flag indicating
- * that we must take it out of that mode when the interface is closed,
- * and, when closing the interface, if that flag is set we take it out
- * of promiscuous mode.
- */
+               /*
+                * That failed.  Try to select the AVS header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 2;    /* request AVS header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               if (ioctl(sock_fd, cmd, &ireq) != -1)
+                       break;  /* success */
 
-/*
- * List of pcaps for which we turned promiscuous mode on by hand.
- * If there are any such pcaps, we arrange to call "pcap_close_all()"
- * when we exit, and have it close all of them to turn promiscuous mode
- * off.
- */
-static struct pcap *pcaps_to_close;
+               /*
+                * That failed.  Try to select the Prism header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 1;    /* request Prism header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
 
-/*
- * TRUE if we've already called "atexit()" to cause "pcap_close_all()" to
* be called on exit.
- */
-static int did_atexit;
+       case MONITOR_PRISM:
+               /*
               * The private ioctl failed.
               */
+               break;
 
-static void    pcap_close_all(void)
-{
-       struct pcap *handle;
+       case MONITOR_PRISM54:
+               /*
+                * Select the Prism header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 3;    /* request Prism header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
 
-       while ((handle = pcaps_to_close) != NULL)
-               pcap_close(handle);
-}
+       case MONITOR_ACX100:
+               /*
+                * Get the current channel.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) {
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                           "%s: SIOCGIWFREQ: %s", device,
+                           pcap_strerror(errno));
+                       return PCAP_ERROR;
+               }
+               channel = ireq.u.freq.m;
 
-static void    pcap_close_linux( pcap_t *handle )
-{
-       struct pcap     *p, *prevp;
-       struct ifreq    ifr;
+               /*
+                * Select the Prism header, and set the channel to the
+                * current value.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 1;            /* request Prism header */
+               args[1] = channel;      /* set channel */
+               memcpy(ireq.u.name, args, 2*sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
 
-       if (handle->md.clear_promisc) {
+       case MONITOR_RT2500:
                /*
-                * We put the interface into promiscuous mode; take
-                * it out of promiscuous mode.
-                *
-                * XXX - if somebody else wants it in promiscuous mode,
-                * this code cannot know that, so it'll take it out
-                * of promiscuous mode.  That's not fixable in 2.0[.x]
-                * kernels.
+                * Disallow transmission - that turns on the
+                * Prism header.
                 */
-               memset(&ifr, 0, sizeof(ifr));
-               strncpy(ifr.ifr_name, handle->md.device, sizeof(ifr.ifr_name));
-               if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
-                       fprintf(stderr,
-                           "Can't restore interface flags (SIOCGIFFLAGS failed: %s).\n"
-                           "Please adjust manually.\n"
-                           "Hint: This can't happen with Linux >= 2.2.0.\n",
-                           strerror(errno));
-               } else {
-                       if (ifr.ifr_flags & IFF_PROMISC) {
-                               /*
-                                * Promiscuous mode is currently on; turn it
-                                * off.
-                                */
-                               ifr.ifr_flags &= ~IFF_PROMISC;
-                               if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
-                                       fprintf(stderr,
-                                           "Can't restore interface flags (SIOCSIFFLAGS failed: %s).\n"
-                                           "Please adjust manually.\n"
-                                           "Hint: This can't happen with Linux >= 2.2.0.\n",
-                                           strerror(errno));
-                               }
-                       }
-               }
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 0;    /* disallow transmitting */
+               memcpy(ireq.u.name, args, sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
 
+       case MONITOR_RT2570:
                /*
-                * Take this pcap out of the list of pcaps for which we
-                * have to take the interface out of promiscuous mode.
+                * Force the Prism header.
                 */
-               for (p = pcaps_to_close, prevp = NULL; p != NULL;
-                   prevp = p, p = p->md.next) {
-                       if (p == handle) {
-                               /*
-                                * Found it.  Remove it from the list.
-                                */
-                               if (prevp == NULL) {
-                                       /*
-                                        * It was at the head of the list.
-                                        */
-                                       pcaps_to_close = p->md.next;
-                               } else {
-                                       /*
-                                        * It was in the middle of the list.
-                                        */
-                                       prevp->md.next = p->md.next;
-                               }
-                               break;
-                       }
-               }
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 1;    /* request Prism header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
+
+       case MONITOR_RT73:
+               /*
+                * Force the Prism header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               ireq.u.data.length = 1; /* 1 argument */
+               ireq.u.data.pointer = "1";
+               ireq.u.data.flags = 0;
+               ioctl(sock_fd, cmd, &ireq);
+               break;
+
+       case MONITOR_RTL8XXX:
+               /*
+                * Force the Prism header.
+                */
+               memset(&ireq, 0, sizeof ireq);
+               strncpy(ireq.ifr_ifrn.ifrn_name, device,
+                   sizeof ireq.ifr_ifrn.ifrn_name);
+               ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+               args[0] = 1;    /* request Prism header */
+               memcpy(ireq.u.name, args, sizeof (int));
+               ioctl(sock_fd, cmd, &ireq);
+               break;
        }
 
-       if (handle->md.device != NULL)
-               free(handle->md.device);
-       handle->md.device = NULL;
-       if (handle->buffer != NULL)
-               free(handle->buffer);
-       if (handle->fd >= 0)
-               close(handle->fd);
+       /*
+        * Note that we have to put the old mode back when we
+        * close the device.
+        */
+       handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
+
+       /*
+        * Add this to the list of pcaps to close when we exit.
+        */
+       pcap_add_to_pcaps_to_close(handle);
+
+       return 1;
+}
+#endif /* IW_MODE_MONITOR */
+
+/*
+ * Try various mechanisms to enter monitor mode.
+ */
+static int
+enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
+{
+#if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
+       int ret;
+#endif
+
+#ifdef HAVE_LIBNL
+       ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
+       if (ret < 0)
+               return ret;     /* error attempting to do so */
+       if (ret == 1)
+               return 1;       /* success */
+#endif /* HAVE_LIBNL */
+
+#ifdef IW_MODE_MONITOR
+       ret = enter_rfmon_mode_wext(handle, sock_fd, device);
+       if (ret < 0)
+               return ret;     /* error attempting to do so */
+       if (ret == 1)
+               return 1;       /* success */
+#endif /* IW_MODE_MONITOR */
+
+       /*
+        * Either none of the mechanisms we know about work or none
+        * of those mechanisms are available, so we can't do monitor
+        * mode.
+        */
+       return 0;
 }
 
+#endif /* HAVE_PF_PACKET_SOCKETS */
+
+/* ===== Functions to interface to the older kernels ================== */
+
 /*
- *  Try to open a packet socket using the old kernel interface.
- *  Returns 0 on failure.
- *  FIXME: 0 uses to mean success (Sebastian)
+ * Try to open a packet socket using the old kernel interface.
+ * Returns 1 on success and a PCAP_ERROR_ value on an error.
  */
 static int
-live_open_old(pcap_t *handle, const char *device, int promisc,
-             int to_ms, char *ebuf)
+activate_old(pcap_t *handle)
 {
        int             arptype;
        struct ifreq    ifr;
+       const char      *device = handle->opt.source;
+       struct utsname  utsname;
+       int             mtu;
 
-       do {
-               /* Open the socket */
+       /* Open the socket */
 
-               handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
-               if (handle->fd == -1) {
-                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                "socket: %s", pcap_strerror(errno));
-                       break;
-               }
+       handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL));
+       if (handle->fd == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "socket: %s", pcap_strerror(errno));
+               return PCAP_ERROR_PERM_DENIED;
+       }
 
-               /* It worked - we are using the old interface */
-               handle->md.sock_packet = 1;
+       /* It worked - we are using the old interface */
+       handle->md.sock_packet = 1;
 
-               /* ...which means we get the link-layer header. */
-               handle->md.cooked = 0;
+       /* ...which means we get the link-layer header. */
+       handle->md.cooked = 0;
 
-               /* Bind to the given device */
+       /* Bind to the given device */
 
-               if (!device) {
-                       strncpy(ebuf, "pcap_open_live: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
-                               PCAP_ERRBUF_SIZE);
-                       break;
-               }
-               if (iface_bind_old(handle->fd, device, ebuf) == -1)
-                       break;
+       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;
+       }
+       if (iface_bind_old(handle->fd, device, handle->errbuf) == -1)
+               return PCAP_ERROR;
 
-               /*
-                * Try to get the link-layer type.
-                */
-               arptype = iface_get_arptype(handle->fd, device, ebuf);
-               if (arptype == -1)
-                       break;
+       /*
+        * Try to get the link-layer type.
+        */
+       arptype = iface_get_arptype(handle->fd, device, handle->errbuf);
+       if (arptype < 0)
+               return PCAP_ERROR;
 
-               /*
-                * Try to find the DLT_ type corresponding to that
-                * link-layer type.
-                */
-               map_arphrd_to_dlt(handle, arptype, 0);
-               if (handle->linktype == -1) {
-                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                "unknown arptype %d", arptype);
-                       break;
-               }
+       /*
+        * Try to find the DLT_ type corresponding to that
+        * link-layer type.
+        */
+       map_arphrd_to_dlt(handle, arptype, 0);
+       if (handle->linktype == -1) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "unknown arptype %d", arptype);
+               return PCAP_ERROR;
+       }
 
-               /* Go to promisc mode if requested */
+       /* Go to promisc mode if requested */
 
-               if (promisc) {
-                       memset(&ifr, 0, sizeof(ifr));
-                       strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
-                       if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
-                               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                        "ioctl: %s", pcap_strerror(errno));
-                               break;
-                       }
-                       if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
-                               /*
-                                * Promiscuous mode isn't currently on,
-                                * so turn it on, and remember that
-                                * we should turn it off when the
-                                * pcap_t is closed.
-                                */
+       if (handle->opt.promisc) {
+               memset(&ifr, 0, sizeof(ifr));
+               strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
+               if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                "SIOCGIFFLAGS: %s", pcap_strerror(errno));
+                       return PCAP_ERROR;
+               }
+               if ((ifr.ifr_flags & IFF_PROMISC) == 0) {
+                       /*
+                        * Promiscuous mode isn't currently on,
+                        * so turn it on, and remember that
+                        * we should turn it off when the
+                        * pcap_t is closed.
+                        */
 
+                       /*
+                        * If we haven't already done so, arrange
+                        * to have "pcap_close_all()" called when
+                        * we exit.
+                        */
+                       if (!pcap_do_addexit(handle)) {
                                /*
-                                * If we haven't already done so, arrange
-                                * to have "pcap_close_all()" called when
-                                * we exit.
+                                * "atexit()" failed; don't put
+                                * the interface in promiscuous
+                                * mode, just give up.
                                 */
-                               if (!did_atexit) {
-                                       if (atexit(pcap_close_all) == -1) {
-                                               /*
-                                                * "atexit()" failed; don't
-                                                * put the interface in
-                                                * promiscuous mode, just
-                                                * give up.
-                                                */
-                                               strncpy(ebuf, "atexit failed",
-                                                       PCAP_ERRBUF_SIZE);
-                                               break;
-                                       }
-                                       did_atexit = 1;
-                               }
-
-                               ifr.ifr_flags |= IFF_PROMISC;
-                               if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
-                                       snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                                                "ioctl: %s",
-                                                pcap_strerror(errno));
-                                       break;
-                               }
-                               handle->md.clear_promisc = 1;
+                               return PCAP_ERROR;
+                       }
 
-                               /*
-                                * Add this to the list of pcaps
-                                * to close when we exit.
-                                */
-                               handle->md.next = pcaps_to_close;
-                               pcaps_to_close = handle;
+                       ifr.ifr_flags |= IFF_PROMISC;
+                       if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) {
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                                        "SIOCSIFFLAGS: %s",
+                                        pcap_strerror(errno));
+                               return PCAP_ERROR;
                        }
+                       handle->md.must_do_on_close |= MUST_CLEAR_PROMISC;
+
+                       /*
+                        * Add this to the list of pcaps
+                        * to close when we exit.
+                        */
+                       pcap_add_to_pcaps_to_close(handle);
                }
+       }
 
+       /*
+        * Compute the buffer size.
+        *
+        * We're using SOCK_PACKET, so this might be a 2.0[.x]
+        * kernel, and might require special handling - check.
+        */
+       if (uname(&utsname) < 0 ||
+           strncmp(utsname.release, "2.0", 3) == 0) {
                /*
-                * Default value for offset to align link-layer payload
-                * on a 4-byte boundary.
+                * Either we couldn't find out what kernel release
+                * this is, or it's a 2.0[.x] kernel.
+                *
+                * In the 2.0[.x] kernel, a "recvfrom()" on
+                * a SOCK_PACKET socket, with MSG_TRUNC set, will
+                * return the number of bytes read, so if we pass
+                * a length based on the snapshot length, it'll
+                * return the number of bytes from the packet
+                * copied to userland, not the actual length
+                * of the packet.
+                *
+                * This means that, for example, the IP dissector
+                * in tcpdump will get handed a packet length less
+                * than the length in the IP header, and will
+                * complain about "truncated-ip".
+                *
+                * So we don't bother trying to copy from the
+                * kernel only the bytes in which we're interested,
+                * but instead copy them all, just as the older
+                * versions of libpcap for Linux did.
+                *
+                * The buffer therefore needs to be big enough to
+                * hold the largest packet we can get from this
+                * device.  Unfortunately, we can't get the MRU
+                * of the network; we can only get the MTU.  The
+                * MTU may be too small, in which case a packet larger
+                * than the buffer size will be truncated *and* we
+                * won't get the actual packet size.
+                *
+                * However, if the snapshot length is larger than
+                * the buffer size based on the MTU, we use the
+                * snapshot length as the buffer size, instead;
+                * this means that with a sufficiently large snapshot
+                * length we won't artificially truncate packets
+                * to the MTU-based size.
+                *
+                * This mess just one of many problems with packet
+                * capture on 2.0[.x] kernels; you really want a
+                * 2.2[.x] or later kernel if you want packet capture
+                * to work well.
                 */
-               handle->offset   = 0;
-
-               return 1;
+               mtu = iface_get_mtu(handle->fd, device, handle->errbuf);
+               if (mtu == -1)
+                       return PCAP_ERROR;
+               handle->bufsize = MAX_LINKHEADER_SIZE + mtu;
+               if (handle->bufsize < handle->snapshot)
+                       handle->bufsize = handle->snapshot;
+       } else {
+               /*
+                * This is a 2.2[.x] or later kernel.
+                *
+                * We can safely pass "recvfrom()" a byte count
+                * based on the snapshot length.
+                */
+               handle->bufsize = handle->snapshot;
+       }
 
-       } while (0);
+       /*
+        * Default value for offset to align link-layer payload
+        * on a 4-byte boundary.
+        */
+       handle->offset   = 0;
 
-       pcap_close_linux(handle);
-       return 0;
+       return 1;
 }
 
 /*
@@ -1733,7 +4070,7 @@ iface_get_mtu(int fd, const char *device, char *ebuf)
 
        if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "ioctl: %s", pcap_strerror(errno));
+                        "SIOCGIFMTU: %s", pcap_strerror(errno));
                return -1;
        }
 
@@ -1753,8 +4090,14 @@ iface_get_arptype(int fd, const char *device, char *ebuf)
 
        if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "ioctl: %s", pcap_strerror(errno));
-               return -1;
+                        "SIOCGIFHWADDR: %s", pcap_strerror(errno));
+               if (errno == ENODEV) {
+                       /*
+                        * No such device.
+                        */
+                       return PCAP_ERROR_NO_SUCH_DEVICE;
+               }
+               return PCAP_ERROR;
        }
 
        return ifr.ifr_hwaddr.sa_family;
@@ -1762,7 +4105,7 @@ iface_get_arptype(int fd, const char *device, char *ebuf)
 
 #ifdef SO_ATTACH_FILTER
 static int
-fix_program(pcap_t *handle, struct sock_fprog *fcode)
+fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
 {
        size_t prog_size;
        register int i;
@@ -1778,7 +4121,7 @@ fix_program(pcap_t *handle, struct sock_fprog *fcode)
        len = handle->fcode.bf_len;
        f = (struct bpf_insn *)malloc(prog_size);
        if (f == NULL) {
-               snprintf(handle->errbuf, sizeof(handle->errbuf),
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "malloc: %s", pcap_strerror(errno));
                return -1;
        }
@@ -1795,26 +4138,33 @@ fix_program(pcap_t *handle, struct sock_fprog *fcode)
 
                case BPF_RET:
                        /*
-                        * It's a return instruction; is the snapshot
-                        * length a constant, rather than the contents
-                        * of the accumulator?
+                        * It's a return instruction; are we capturing
+                        * in memory-mapped mode?
                         */
-                       if (BPF_MODE(p->code) == BPF_K) {
+                       if (!is_mmapped) {
                                /*
-                                * Yes - if the value to be returned,
-                                * i.e. the snapshot length, is anything
-                                * other than 0, make it 65535, so that
-                                * the packet is truncated by "recvfrom()",
-                                * not by the filter.
-                                *
-                                * XXX - there's nothing we can easily do
-                                * if it's getting the value from the
-                                * accumulator; we'd have to insert
-                                * code to force non-zero values to be
-                                * 65535.
+                                * No; is the snapshot length a constant,
+                                * rather than the contents of the
+                                * accumulator?
                                 */
-                               if (p->k != 0)
-                                       p->k = 65535;
+                               if (BPF_MODE(p->code) == BPF_K) {
+                                       /*
+                                        * Yes - if the value to be returned,
+                                        * i.e. the snapshot length, is
+                                        * anything other than 0, make it
+                                        * 65535, so that the packet is
+                                        * truncated by "recvfrom()",
+                                        * not by the filter.
+                                        *
+                                        * XXX - there's nothing we can
+                                        * easily do if it's getting the
+                                        * value from the accumulator; we'd
+                                        * have to insert code to force
+                                        * non-zero values to be 65535.
+                                        */
+                                       if (p->k != 0)
+                                               p->k = 65535;
+                               }
                        }
                        break;
 
@@ -1950,7 +4300,7 @@ set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
                        if (save_errno != EAGAIN) {
                                /* Fatal error */
                                reset_kernel_filter(handle);
-                               snprintf(handle->errbuf, sizeof(handle->errbuf),
+                               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                                 "recv: %s", pcap_strerror(save_errno));
                                return -2;
                        }
@@ -1990,8 +4340,13 @@ set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
 static int
 reset_kernel_filter(pcap_t *handle)
 {
-       /* setsockopt() barfs unless it get a dummy parameter */
-       int dummy;
+       /*
+        * setsockopt() barfs unless it get a dummy parameter.
+        * valgrind whines unless the value is initialized,
+        * as it has no idea that setsockopt() ignores its
+        * parameter.
+        */
+       int dummy = 0;
 
        return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
                                   &dummy, sizeof(dummy));