]> The Tcpdump Group git mirrors - libpcap/blobdiff - pcap-linux.c
Fix typo.
[libpcap] / pcap-linux.c
index cb805c88cff6017d63cff1ab42ffb8fc0ef5a0fe..de14a7ecd8c3cc41dfb30d1874e37c6626c4799e 100644 (file)
  *                     based on previous works of:
  *                     Simon Patarin <[email protected]>
  *                     Phil Wood <[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_ =
  */
 
 #ifndef lint
 static const char rcsid[] _U_ =
-    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.140 2008-04-04 19:37:45 guy Exp $ (LBL)";
+    "@(#) $Header: /tcpdump/master/libpcap/pcap-linux.c,v 1.164 2008-12-14 22:00:57 guy Exp $ (LBL)";
 #endif
 
 /*
 #endif
 
 /*
@@ -80,20 +113,24 @@ static const char rcsid[] _U_ =
  */
 
 
  */
 
 
+#define _GNU_SOURCE
+
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 
 #include <errno.h>
 #ifdef HAVE_CONFIG_H
 #include "config.h"
 #endif
 
 #include <errno.h>
+#include <stdio.h>
 #include <stdlib.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <string.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 <sys/mman.h>
 #include <sys/socket.h>
 #include <sys/ioctl.h>
 #include <sys/utsname.h>
 #include <sys/mman.h>
-#include <net/if.h>
+#include <linux/if.h>
 #include <netinet/in.h>
 #include <linux/if_ether.h>
 #include <net/if_arp.h>
 #include <netinet/in.h>
 #include <linux/if_ether.h>
 #include <net/if_arp.h>
@@ -104,10 +141,24 @@ static const char rcsid[] _U_ =
  */
 #ifdef HAVE_LINUX_WIRELESS_H
 #include <linux/wireless.h>
  */
 #ifdef HAVE_LINUX_WIRELESS_H
 #include <linux/wireless.h>
-#endif
+#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-int.h"
 #include "pcap/sll.h"
+#include "pcap/vlan.h"
 
 #ifdef HAVE_DAG_API
 #include "pcap-dag.h"
 
 #ifdef HAVE_DAG_API
 #include "pcap-dag.h"
@@ -165,6 +216,9 @@ static const char rcsid[] _U_ =
   */
 # ifdef PACKET_HOST
 #  define HAVE_PF_PACKET_SOCKETS
   */
 # ifdef PACKET_HOST
 #  define HAVE_PF_PACKET_SOCKETS
+#  ifdef PACKET_AUXDATA
+#   define HAVE_PACKET_AUXDATA
+#  endif /* PACKET_AUXDATA */
 # endif /* PACKET_HOST */
 
 
 # endif /* PACKET_HOST */
 
 
@@ -173,6 +227,11 @@ static const char rcsid[] _U_ =
   * uses many ring related structs and macros */
 # ifdef TPACKET_HDRLEN
 #  define HAVE_PACKET_RING
   * 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 */
 
 # endif /* TPACKET_HDRLEN */
 #endif /* PF_PACKET */
 
@@ -234,18 +293,27 @@ 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 int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
-static void pcap_close_linux(pcap_t *);
+static void pcap_cleanup_linux(pcap_t *);
+
+union thdr {
+       struct tpacket_hdr      *h1;
+       struct tpacket2_hdr     *h2;
+       void                    *raw;
+};
 
 #ifdef HAVE_PACKET_RING
 
 #ifdef HAVE_PACKET_RING
-#define RING_GET_FRAME(h) (((struct tpacket_hdr**)h->buffer)[h->offset])
+#define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
 
 static void destroy_ring(pcap_t *handle);
 static int create_ring(pcap_t *handle);
 
 static void destroy_ring(pcap_t *handle);
 static int create_ring(pcap_t *handle);
-static void pcap_close_linux_mmap(pcap_t *);
+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 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
 
 /*
 #endif
 
 /*
@@ -258,14 +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);
 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);
-static int     has_wext(int sock_fd, const char *device);
-static int     enter_rfmon_mode_wext(pcap_t *handle, int sock_fd,
+#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);
     const char *device);
-#endif
+#endif /* HAVE_PF_PACKET_SOCKETS */
 static int     iface_bind_old(int fd, const char *device, char *ebuf);
 
 #ifdef SO_ATTACH_FILTER
 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);
 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);
@@ -281,6 +352,12 @@ pcap_create(const char *device, char *ebuf)
 {
        pcap_t *handle;
 
 {
        pcap_t *handle;
 
+       /*
+        * A null device name is equivalent to the "any" device.
+        */
+       if (device == NULL)
+               device = "any";
+
 #ifdef HAVE_DAG_API
        if (strstr(device, "dag")) {
                return dag_create(device, ebuf);
 #ifdef HAVE_DAG_API
        if (strstr(device, "dag")) {
                return dag_create(device, ebuf);
@@ -300,7 +377,7 @@ pcap_create(const char *device, char *ebuf)
 #endif
 
 #ifdef PCAP_SUPPORT_USB
 #endif
 
 #ifdef PCAP_SUPPORT_USB
-       if (strstr(device, "usb")) {
+       if (strstr(device, "usbmon")) {
                return usb_create(device, ebuf);
        }
 #endif
                return usb_create(device, ebuf);
        }
 #endif
@@ -314,22 +391,462 @@ pcap_create(const char *device, char *ebuf)
        return handle;
 }
 
        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.
+*/
+
+/*
+ * 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;
+
+       /*
+        * Generate the path string for the symlink to the physical 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;
+}
+
+struct nl80211_state {
+       struct nl_handle *nl_handle;
+       struct nl_cache *nl_cache;
+       struct genl_family *nl80211;
+};
+
+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;
+       }
+
+       if (genl_connect(state->nl_handle)) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "%s: failed to connect to generic netlink", device);
+               goto out_handle_destroy;
+       }
+
+       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;
+       }
+
+       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;
+       }
+
+       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;
+       }
+
+       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);
+
+       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;
+               }
+       }
+
+       /*
+        * Success.
+        */
+       nlmsg_free(msg);
+       return 1;
+
+nla_put_failure:
+       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+           "%s: nl_put failed adding %s interface",
+           device, mondevice);
+       nlmsg_free(msg);
+       return PCAP_ERROR;
+}
+
+static int
+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;
+               }
+       }
+
+       /*
+        * Success.
+        */
+       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;
+}
+
+static int
+enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device)
+{
+       int ret;
+       char phydev_path[PATH_MAX+1];
+       struct nl80211_state nlstate;
+       struct ifreq ifr;
+       u_int n;
+
+       /*
+        * 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
 static int
-pcap_can_set_rfmon_linux(pcap_t *p)
+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
 
 #ifdef IW_MODE_MONITOR
        int sock_fd;
        struct iwreq ireq;
 #endif
 
-       if (p->opt.source == NULL) {
+       if (strcmp(handle->opt.source, "any") == 0) {
                /*
                /*
-                * This is equivalent to the "any" device, and we don't
-                * support monitor mode on it.
+                * Monitor mode makes no sense on the "any" device.
                 */
                return 0;
        }
 
                 */
                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
 #ifdef IW_MODE_MONITOR
        /*
         * Bleah.  There doesn't appear to be an ioctl to use to ask
@@ -343,7 +860,7 @@ pcap_can_set_rfmon_linux(pcap_t *p)
         */
        sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
        if (sock_fd == -1) {
         */
        sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
        if (sock_fd == -1) {
-               (void)snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
+               (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                    "socket: %s", pcap_strerror(errno));
                return PCAP_ERROR;
        }
                    "socket: %s", pcap_strerror(errno));
                return PCAP_ERROR;
        }
@@ -351,7 +868,7 @@ pcap_can_set_rfmon_linux(pcap_t *p)
        /*
         * Attempt to get the current mode.
         */
        /*
         * Attempt to get the current mode.
         */
-       strncpy(ireq.ifr_ifrn.ifrn_name, p->opt.source,
+       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) {
            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) {
@@ -383,19 +900,23 @@ pcap_can_set_rfmon_linux(pcap_t *p)
  * Even with newer kernels, we have the same issue with rfmon mode.
  */
 
  * Even with newer kernels, we have the same issue with rfmon mode.
  */
 
-static void    pcap_close_linux( pcap_t *handle )
+static void    pcap_cleanup_linux( pcap_t *handle )
 {
        struct ifreq    ifr;
 {
        struct ifreq    ifr;
+#ifdef HAVE_LIBNL
+       struct nl80211_state nlstate;
+       int ret;
+#endif /* HAVE_LIBNL */
 #ifdef IW_MODE_MONITOR
        struct iwreq ireq;
 #ifdef IW_MODE_MONITOR
        struct iwreq ireq;
-#endif
+#endif /* IW_MODE_MONITOR */
 
 
-       if (handle->md.must_clear != 0) {
+       if (handle->md.must_do_on_close != 0) {
                /*
                 * There's something we have to do when closing this
                 * pcap_t.
                 */
                /*
                 * There's something we have to do when closing this
                 * pcap_t.
                 */
-               if (handle->md.must_clear & MUST_CLEAR_PROMISC) {
+               if (handle->md.must_do_on_close & MUST_CLEAR_PROMISC) {
                        /*
                         * We put the interface into promiscuous mode;
                         * take it out of promiscuous mode.
                        /*
                         * We put the interface into promiscuous mode;
                         * take it out of promiscuous mode.
@@ -433,8 +954,25 @@ static void        pcap_close_linux( pcap_t *handle )
                        }
                }
 
                        }
                }
 
+#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
 #ifdef IW_MODE_MONITOR
-               if (handle->md.must_clear & MUST_CLEAR_RFMON) {
+               if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
                        /*
                         * We put the interface into rfmon mode;
                         * take it out of rfmon mode.
                        /*
                         * We put the interface into rfmon mode;
                         * take it out of rfmon mode.
@@ -458,7 +996,7 @@ static void pcap_close_linux( pcap_t *handle )
                                    strerror(errno));
                        }
                }
                                    strerror(errno));
                        }
                }
-#endif
+#endif /* IW_MODE_MONITOR */
 
                /*
                 * Take this pcap out of the list of pcaps for which we
 
                /*
                 * Take this pcap out of the list of pcaps for which we
@@ -467,11 +1005,15 @@ static void      pcap_close_linux( pcap_t *handle )
                pcap_remove_from_pcaps_to_close(handle);
        }
 
                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;
        }
        if (handle->md.device != NULL) {
                free(handle->md.device);
                handle->md.device = NULL;
        }
-       pcap_close_common(handle);
+       pcap_cleanup_live_common(handle);
 }
 
 /*
 }
 
 /*
@@ -486,8 +1028,7 @@ static int
 pcap_activate_linux(pcap_t *handle)
 {
        const char      *device;
 pcap_activate_linux(pcap_t *handle)
 {
        const char      *device;
-       int             err;
-       int             activate_ok = 0;
+       int             status = 0;
 
        device = handle->opt.source;
 
 
        device = handle->opt.source;
 
@@ -497,27 +1038,26 @@ pcap_activate_linux(pcap_t *handle)
        handle->set_datalink_op = NULL; /* can't change data link type */
        handle->getnonblock_op = pcap_getnonblock_fd;
        handle->setnonblock_op = pcap_setnonblock_fd;
        handle->set_datalink_op = NULL; /* can't change data link type */
        handle->getnonblock_op = pcap_getnonblock_fd;
        handle->setnonblock_op = pcap_setnonblock_fd;
-       handle->close_op = pcap_close_linux;
+       handle->cleanup_op = pcap_cleanup_linux;
        handle->read_op = pcap_read_linux;
        handle->stats_op = pcap_stats_linux;
 
        /*
        handle->read_op = pcap_read_linux;
        handle->stats_op = pcap_stats_linux;
 
        /*
-        * NULL and "any" are special devices which give us the hint to
-        * monitor all devices.
+        * The "any" device is a special device which causes us not
+        * to bind to a particular device and thus to look at all
+        * devices.
         */
         */
-       if (!device || strcmp(device, "any") == 0) {
-               device                  = NULL;
-               handle->md.device       = strdup("any");
+       if (strcmp(device, "any") == 0) {
                if (handle->opt.promisc) {
                        handle->opt.promisc = 0;
                        /* Just a warning. */
                        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                            "Promiscuous mode not supported on the \"any\" device");
                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;
                }
                }
+       }
 
 
-       } else
-               handle->md.device       = strdup(device);
-
+       handle->md.device       = strdup(device);
        if (handle->md.device == NULL) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
                         pcap_strerror(errno) );
        if (handle->md.device == NULL) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s",
                         pcap_strerror(errno) );
@@ -534,31 +1074,70 @@ pcap_activate_linux(pcap_t *handle)
         * trying both methods with the newer method preferred.
         */
 
         * trying both methods with the newer method preferred.
         */
 
-       if ((err = activate_new(handle)) == 1) {
-               activate_ok = 1;
+       if ((status = activate_new(handle)) == 1) {
                /*
                /*
+                * Success.
                 * Try to use memory-mapped access.
                 */
                 * Try to use memory-mapped access.
                 */
-               if (activate_mmap(handle) == 1)
-                       return 0;       /* we succeeded; nothing more to do */
+               switch (activate_mmap(handle)) {
+
+               case 1:
+                       /* we succeeded; nothing more to do */
+                       return 0;
+
+               case 0:
+                       /*
+                        * Kernel doesn't support it - just continue
+                        * with non-memory-mapped access.
+                        */
+                       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 (err == 0) {
+       else if (status == 0) {
                /* Non-fatal error; try old way */
                /* Non-fatal error; try old way */
-               if ((err = activate_old(handle)) == 1)
-                       activate_ok = 1;
-       }
-       if (!activate_ok) {
+               if ((status = activate_old(handle)) != 1) {
+                       /*
+                        * Both methods to open the packet socket failed.
+                        * Tidy up and report our failure (handle->errbuf
+                        * is expected to be set by the functions above).
+                        */
+                       goto fail;
+               }
+       } else {
                /*
                /*
-                * Both methods to open the packet socket failed. Tidy
-                * up and report our failure (ebuf is expected to be
-                * set by the functions above).
+                * Fatal error with the new way; just fail.
+                * status has the error return; if it's PCAP_ERROR,
+                * handle->errbuf has been set appropriately.
                 */
                 */
+               goto fail;
+       }
 
 
-               if (handle->md.device != NULL) {
-                       free(handle->md.device);
-                       handle->md.device = NULL;
+       /*
+        * 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;
                }
                }
-               return err;
        }
 
        /* Allocate the buffer */
        }
 
        /* Allocate the buffer */
@@ -567,7 +1146,8 @@ pcap_activate_linux(pcap_t *handle)
        if (!handle->buffer) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "malloc: %s", pcap_strerror(errno));
        if (!handle->buffer) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "malloc: %s", pcap_strerror(errno));
-               return PCAP_ERROR;
+               status = PCAP_ERROR;
+               goto fail;
        }
 
        /*
        }
 
        /*
@@ -576,7 +1156,11 @@ pcap_activate_linux(pcap_t *handle)
         */
        handle->selectable_fd = handle->fd;
 
         */
        handle->selectable_fd = handle->fd;
 
-       return 0;
+       return status;
+
+fail:
+       pcap_cleanup_linux(handle);
+       return status;
 }
 
 /*
 }
 
 /*
@@ -610,7 +1194,17 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
 #else
        struct sockaddr         from;
 #endif
 #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;
        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;
 
        int                     packet_len, caplen;
        struct pcap_pkthdr      pcap_header;
 
@@ -631,9 +1225,36 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
        offset = 0;
 #endif
 
        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;
        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?
        do {
                /*
                 * Has "pcap_breakloop()" been called?
@@ -647,12 +1268,17 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                        handle->break_loop = 0;
                        return -2;
                }
                        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);
                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 */
 
 
        /* Check if an error occured */
 
@@ -737,7 +1363,37 @@ pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata)
                      from.sll_halen);
                hdrp->sll_protocol = from.sll_protocol;
        }
                      from.sll_halen);
                hdrp->sll_protocol = from.sll_protocol;
        }
-#endif
+
+#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;
+
+               if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) ||
+                   cmsg->cmsg_level != SOL_PACKET ||
+                   cmsg->cmsg_type != PACKET_AUXDATA)
+                       continue;
+
+               aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg);
+               if (aux->tp_vlan_tci == 0)
+                       continue;
+
+               len = packet_len > iov.iov_len ? iov.iov_len : packet_len;
+               if (len < 2 * ETH_ALEN)
+                       break;
+
+               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(aux->tp_vlan_tci);
+
+               packet_len += VLAN_TAG_LEN;
+       }
+#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
 
        /*
         * XXX: According to the kernel source we should get the real
@@ -1039,7 +1695,8 @@ pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
  *  Attach the given BPF code to the packet capture device.
  */
 static int
  *  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;
 {
 #ifdef SO_ATTACH_FILTER
        struct sock_fprog       fcode;
@@ -1092,13 +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
                 *
                 * 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:
 
                case -1:
                default:
@@ -1172,6 +1829,13 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
        return 0;
 }
 
        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?
 /*
  * Set direction flag: Which packets do we accept on a forwarding
  * single device? IN, OUT or both?
@@ -1524,6 +2188,17 @@ static void map_arphrd_to_dlt(pcap_t *handle, int arptype, int cooked_ok)
                handle->linktype = DLT_LINUX_LAPD;
                break;
 
                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;
        default:
                handle->linktype = -1;
                break;
@@ -1543,19 +2218,24 @@ static int
 activate_new(pcap_t *handle)
 {
 #ifdef HAVE_PF_PACKET_SOCKETS
 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;
        int                     sock_fd = -1, arptype;
+#ifdef HAVE_PACKET_AUXDATA
+       int                     val;
+#endif
        int                     err = 0;
        struct packet_mreq      mr;
        int                     err = 0;
        struct packet_mreq      mr;
-       const char* device = handle->opt.source;
 
        /*
 
        /*
-        * Open a socket with protocol family packet. If a device is
-        * given we try to open it in raw mode otherwise we use
-        * the cooked interface.
+        * Open a socket with protocol family packet. If the
+        * "any" device was specified, we open a SOCK_DGRAM
+        * socket for the cooked interface, otherwise we first
+        * try a SOCK_RAW socket for the raw interface.
         */
         */
-       sock_fd = device ?
-               socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))
-             : socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL));
+       sock_fd = is_any_device ?
+               socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) :
+               socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
 
        if (sock_fd == -1) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
 
        if (sock_fd == -1) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s",
@@ -1590,7 +2270,7 @@ activate_new(pcap_t *handle)
         * to cooked mode if we have an unknown interface type
         * or a type we know doesn't work well in raw mode.
         */
         * to cooked mode if we have an unknown interface type
         * or a type we know doesn't work well in raw mode.
         */
-       if (device) {
+       if (!is_any_device) {
                /* Assume for now we don't need cooked mode. */
                handle->md.cooked = 0;
 
                /* Assume for now we don't need cooked mode. */
                handle->md.cooked = 0;
 
@@ -1601,7 +2281,7 @@ activate_new(pcap_t *handle)
                         * because entering monitor mode could change
                         * the link-layer type.
                         */
                         * because entering monitor mode could change
                         * the link-layer type.
                         */
-                       err = enter_rfmon_mode_wext(handle, sock_fd, device);
+                       err = enter_rfmon_mode(handle, sock_fd, device);
                        if (err < 0) {
                                /* Hard failure */
                                close(sock_fd);
                        if (err < 0) {
                                /* Hard failure */
                                close(sock_fd);
@@ -1615,6 +2295,15 @@ activate_new(pcap_t *handle)
                                close(sock_fd);
                                return PCAP_ERROR_RFMON_NOTSUP;
                        }
                                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) {
                }
                arptype = iface_get_arptype(sock_fd, device, handle->errbuf);
                if (arptype < 0) {
@@ -1696,24 +2385,32 @@ activate_new(pcap_t *handle)
                }
 
                if ((err = iface_bind(sock_fd, handle->md.ifindex,
                }
 
                if ((err = iface_bind(sock_fd, handle->md.ifindex,
-                   handle->errbuf)) < 0) {
+                   handle->errbuf)) != 1) {
                        close(sock_fd);
                        close(sock_fd);
-                       if (err == -2)
-                               return PCAP_ERROR;
+                       if (err < 0)
+                               return err;
                        else
                                return 0;       /* try old mechanism */
                }
        } else {
                /*
                        else
                                return 0;       /* try old mechanism */
                }
        } else {
                /*
-                * This is cooked mode.
+                * The "any" device.
+                */
+               if (handle->opt.rfmon) {
+                       /*
+                        * It doesn't support monitor mode.
+                        */
+                       return PCAP_ERROR_RFMON_NOTSUP;
+               }
+
+               /*
+                * It uses cooked mode.
                 */
                handle->md.cooked = 1;
                handle->linktype = DLT_LINUX_SLL;
 
                /*
                 * We're not bound to a device.
                 */
                handle->md.cooked = 1;
                handle->linktype = DLT_LINUX_SLL;
 
                /*
                 * We're not bound to a device.
-                * XXX - true?  Or true only if we're using
-                * the "any" device?
                 * For now, we're using this as an indication
                 * that we can't transmit; stop doing that only
                 * if we figure out how to transmit in cooked
                 * 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
@@ -1738,10 +2435,13 @@ activate_new(pcap_t *handle)
 
        /*
         * Hmm, how can we set promiscuous mode on all interfaces?
 
        /*
         * Hmm, how can we set promiscuous mode on all interfaces?
-        * I am not sure if that is possible at all.
+        * I am not sure if that is possible at all.  For now, we
+        * silently ignore attempts to turn promiscuous mode on
+        * for the "any" device (so you don't have to explicitly
+        * disable it in programs such as tcpdump).
         */
 
         */
 
-       if (device && handle->opt.promisc) {
+       if (!is_any_device && handle->opt.promisc) {
                memset(&mr, 0, sizeof(mr));
                mr.mr_ifindex = handle->md.ifindex;
                mr.mr_type    = PACKET_MR_PROMISC;
                memset(&mr, 0, sizeof(mr));
                mr.mr_ifindex = handle->md.ifindex;
                mr.mr_type    = PACKET_MR_PROMISC;
@@ -1754,6 +2454,20 @@ activate_new(pcap_t *handle)
                }
        }
 
                }
        }
 
+       /* 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 -
        /*
         * This is a 2.2[.x] or later kernel (we know that
         * because we're not using a SOCK_PACKET socket -
@@ -1775,7 +2489,7 @@ activate_new(pcap_t *handle)
        handle->bufsize = handle->snapshot;
 
        /* Save the socket FD in the pcap structure */
        handle->bufsize = handle->snapshot;
 
        /* Save the socket FD in the pcap structure */
-       handle->fd       = sock_fd;
+       handle->fd = sock_fd;
 
        return 1;
 #else
 
        return 1;
 #else
@@ -1792,23 +2506,43 @@ activate_mmap(pcap_t *handle)
 #ifdef HAVE_PACKET_RING
        int ret;
 
 #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;
        }
        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);
        ret = create_ring(handle);
-       if (ret == 0)
+       if (ret != 1) {
+               free(handle->md.oneshot_buffer);
                return ret;
                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;
 
        /* 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->close_op = pcap_close_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->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 */
        handle->selectable_fd = handle->fd;
        return 1;
 #else /* HAVE_PACKET_RING */
@@ -1817,25 +2551,61 @@ activate_mmap(pcap_t *handle)
 }
 
 #ifdef HAVE_PACKET_RING
 }
 
 #ifdef HAVE_PACKET_RING
-
-static void
-compute_ring_block(int frame_size, unsigned *block_size, unsigned *frames_per_block)
+static int
+prepare_tpacket_socket(pcap_t *handle)
 {
 {
-       /* 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. */
-       *block_size = getpagesize();
-       while (*block_size < frame_size) 
-               *block_size <<= 1;
+#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;
 
 
-       *frames_per_block = *block_size/frame_size;
+       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)
 {
 }
 
 static int
 create_ring(pcap_t *handle)
 {
-       unsigned i, j, ringsize, frames_per_block;
+       unsigned i, j, frames_per_block;
        struct tpacket_req req;
 
        /* Note that with large snapshot (say 64K) only a few frames 
        struct tpacket_req req;
 
        /* Note that with large snapshot (say 64K) only a few frames 
@@ -1843,57 +2613,89 @@ create_ring(pcap_t *handle)
         * (and a lot of memory will be unused). 
         * The snap len should be carefully chosen to achive best
         * performance */
         * (and a lot of memory will be unused). 
         * The snap len should be carefully chosen to achive best
         * performance */
-       req.tp_frame_size = TPACKET_ALIGN(handle->snapshot+TPACKET_HDRLEN);
+       req.tp_frame_size = TPACKET_ALIGN(handle->snapshot +
+                                         TPACKET_ALIGN(handle->md.tp_hdrlen) +
+                                         sizeof(struct sockaddr_ll));
        req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
        req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
-       compute_ring_block(req.tp_frame_size, &req.tp_block_size, &frames_per_block);
-       req.tp_block_nr = req.tp_frame_nr / frames_per_block;
 
 
-       /* 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;
+       /* 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:
 
        /* 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 (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
                                        (void *) &req, sizeof(req))) {
-               /* try to reduce requested ring size to prevent memory failure */
                if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
                if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
-                       req.tp_frame_nr >>= 1;
-                       req.tp_block_nr = req.tp_frame_nr/frames_per_block;
+                       /*
+                        * 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;
                }
                        goto retry;
                }
-               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "can't create rx ring on "
-                               "packet socket %d: %d-%s", handle->fd, errno, 
-                               pcap_strerror(errno));
-               return 0;
+               if (errno == ENOPROTOOPT) {
+                       /*
+                        * We don't have ring buffer support in this kernel.
+                        */
+                       return 0;
+               }
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't create rx ring on packet socket: %s",
+                   pcap_strerror(errno));
+               return -1;
        }
 
        /* memory map the rx ring */
        }
 
        /* memory map the rx ring */
-       ringsize = req.tp_block_nr * req.tp_block_size;
-       handle->bp = mmap(0, ringsize, PROT_READ| PROT_WRITE, MAP_SHARED, 
-                                       handle->fd, 0);
-       if (handle->bp == MAP_FAILED) {
-               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "can't mmap rx ring: %d-%s",
-                       errno, pcap_strerror(errno));
+       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);
 
                /* clear the allocated ring on error*/
                destroy_ring(handle);
-               return 0;
+               return -1;
        }
 
        /* allocate a ring for each frame header pointer*/
        handle->cc = req.tp_frame_nr;
        }
 
        /* allocate a ring for each frame header pointer*/
        handle->cc = req.tp_frame_nr;
-       handle->buffer = malloc(handle->cc * sizeof(struct tpacket_hdr*));
+       handle->buffer = malloc(handle->cc * sizeof(union thdr *));
        if (!handle->buffer) {
        if (!handle->buffer) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "can't allocate ring of frame headers: %s",
+                   pcap_strerror(errno));
+
                destroy_ring(handle);
                destroy_ring(handle);
-               return 0;
+               return -1;
        }
 
        /* fill the header ring with proper frame ptr*/
        handle->offset = 0;
        for (i=0; i<req.tp_block_nr; ++i) {
        }
 
        /* fill the header ring with proper frame ptr*/
        handle->offset = 0;
        for (i=0; i<req.tp_block_nr; ++i) {
-               u_char *base = &handle->bp[i*req.tp_block_size];
+               void *base = &handle->md.mmapbuf[i*req.tp_block_size];
                for (j=0; j<frames_per_block; ++j, ++handle->offset) {
                for (j=0; j<frames_per_block; ++j, ++handle->offset) {
-                       RING_GET_FRAME(handle) = (struct tpacket_hdr*) base;
+                       RING_GET_FRAME(handle) = base;
                        base += req.tp_frame_size;
                }
        }
                        base += req.tp_frame_size;
                }
        }
@@ -1914,28 +2716,51 @@ destroy_ring(pcap_t *handle)
                                (void *) &req, sizeof(req));
 
        /* if ring is mapped, unmap it*/
                                (void *) &req, sizeof(req));
 
        /* if ring is mapped, unmap it*/
-       if (handle->bp) {
-               /* need to re-compute the ring size */
-               unsigned frames_per_block, block_size;
-               compute_ring_block(handle->bufsize, &block_size, &frames_per_block);
-
-               /* do not perform sanity check here: we can't recover any error */
-               munmap(handle->bp, block_size * handle->cc / frames_per_block);
-               handle->bp = 0;
-       }
-
-       /* if the header ring is allocated, clear it*/
-       if (handle->buffer) {
-               free(handle->buffer);
-               handle->buffer = 0;
+       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
 static void
-pcap_close_linux_mmap( pcap_t *handle )
+pcap_cleanup_linux_mmap( pcap_t *handle )
 {
        destroy_ring(handle);
 {
        destroy_ring(handle);
-       pcap_close_linux(handle);
+       if (handle->md.oneshot_buffer != NULL) {
+               free(handle->md.oneshot_buffer);
+               handle->md.oneshot_buffer = NULL;
+       }
+       pcap_cleanup_linux(handle);
 }
 
 
 }
 
 
@@ -1952,14 +2777,52 @@ 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) {
        /* 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)
+               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;
                        p->md.timeout = p->md.timeout*-1 - 1;
-       } else 
-               if (p->md.timeout < 0)
+               }
+       } 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;
                        p->md.timeout = (p->md.timeout+1)*-1;
+               }
+       }
        return 0;
 }
 
        return 0;
 }
 
+static inline union thdr *
+pcap_get_ring_frame(pcap_t *handle, int status)
+{
+       union thdr h;
+
+       h.raw = RING_GET_FRAME(handle);
+       switch (handle->md.tp_version) {
+       case TPACKET_V1:
+               if (status != (h.h1->tp_status ? TP_STATUS_USER :
+                                               TP_STATUS_KERNEL))
+                       return NULL;
+               break;
+#ifdef HAVE_TPACKET2
+       case TPACKET_V2:
+               if (status != (h.h2->tp_status ? TP_STATUS_USER :
+                                               TP_STATUS_KERNEL))
+                       return NULL;
+               break;
+#endif
+       }
+       return h.raw;
+}
+
 static int
 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback, 
                u_char *user)
 static int
 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback, 
                u_char *user)
@@ -1967,7 +2830,8 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
        int pkts = 0;
 
        /* wait for frames availability.*/
        int pkts = 0;
 
        /* wait for frames availability.*/
-       if ((handle->md.timeout >= 0) && !(RING_GET_FRAME(handle)->tp_status)) {
+       if ((handle->md.timeout >= 0) &&
+           !pcap_get_ring_frame(handle, TP_STATUS_USER)) {
                struct pollfd pollinfo;
                int ret;
 
                struct pollfd pollinfo;
                int ret;
 
@@ -1999,16 +2863,46 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                struct sockaddr_ll *sll;
                struct pcap_pkthdr pcaphdr;
                unsigned char *bp;
                struct sockaddr_ll *sll;
                struct pcap_pkthdr pcaphdr;
                unsigned char *bp;
-               struct tpacket_hdr* thdr = RING_GET_FRAME(handle);
-               if (thdr->tp_status == TP_STATUS_KERNEL)
+               union thdr h;
+               unsigned int tp_len;
+               unsigned int tp_mac;
+               unsigned int tp_snaplen;
+               unsigned int tp_sec;
+               unsigned int tp_usec;
+
+               h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
+               if (!h.raw)
                        break;
 
                        break;
 
+               switch (handle->md.tp_version) {
+               case TPACKET_V1:
+                       tp_len     = h.h1->tp_len;
+                       tp_mac     = h.h1->tp_mac;
+                       tp_snaplen = h.h1->tp_snaplen;
+                       tp_sec     = h.h1->tp_sec;
+                       tp_usec    = h.h1->tp_usec;
+                       break;
+#ifdef HAVE_TPACKET2
+               case TPACKET_V2:
+                       tp_len     = h.h2->tp_len;
+                       tp_mac     = h.h2->tp_mac;
+                       tp_snaplen = h.h2->tp_snaplen;
+                       tp_sec     = h.h2->tp_sec;
+                       tp_usec    = h.h2->tp_nsec / 1000;
+                       break;
+#endif
+               default:
+                       snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
+                               "unsupported tpacket version %d",
+                               handle->md.tp_version);
+                       return -1;
+               }
                /* perform sanity check on internal offset. */
                /* perform sanity check on internal offset. */
-               if (thdr->tp_mac+thdr->tp_snaplen > handle->bufsize) {
+               if (tp_mac + tp_snaplen > handle->bufsize) {
                        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                "corrupted frame on kernel ring mac "
                                "offset %d + caplen %d > frame len %d", 
                        snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                "corrupted frame on kernel ring mac "
                                "offset %d + caplen %d > frame len %d", 
-                               thdr->tp_mac, thdr->tp_snaplen, handle->bufsize);
+                               tp_mac, tp_snaplen, handle->bufsize);
                        return -1;
                }
 
                        return -1;
                }
 
@@ -2021,25 +2915,47 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                 * Note: alternatively it could be possible to stop applying 
                 * the filter when the ring became empty, but it can possibly
                 * happen a lot later... */
                 * Note: alternatively it could be possible to stop applying 
                 * the filter when the ring became empty, but it can possibly
                 * happen a lot later... */
-               bp = (unsigned char*)thdr + thdr->tp_mac;
+               bp = (unsigned char*)h.raw + tp_mac;
                run_bpf = (!handle->md.use_bpf) || 
                        ((handle->md.use_bpf>1) && handle->md.use_bpf--);
                if (run_bpf && handle->fcode.bf_insns && 
                                (bpf_filter(handle->fcode.bf_insns, bp,
                run_bpf = (!handle->md.use_bpf) || 
                        ((handle->md.use_bpf>1) && handle->md.use_bpf--);
                if (run_bpf && handle->fcode.bf_insns && 
                                (bpf_filter(handle->fcode.bf_insns, bp,
-                                       thdr->tp_len, thdr->tp_snaplen) == 0))
+                                       tp_len, tp_snaplen) == 0))
                        goto skip;
 
                        goto skip;
 
-               /* check direction and interface index */
-               sll = (void*)thdr + TPACKET_ALIGN(sizeof(*thdr));
-               if ((sll->sll_ifindex == handle->md.lo_ifindex) &&
-                                       (sll->sll_pkttype == PACKET_OUTGOING))
-                       goto skip;
+               /*
+                * 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 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;
+               }
 
                /* get required packet info from ring header */
 
                /* get required packet info from ring header */
-               pcaphdr.ts.tv_sec = thdr->tp_sec;
-               pcaphdr.ts.tv_usec = thdr->tp_usec;
-               pcaphdr.caplen = thdr->tp_snaplen;
-               pcaphdr.len = thdr->tp_len;
+               pcaphdr.ts.tv_sec = tp_sec;
+               pcaphdr.ts.tv_usec = tp_usec;
+               pcaphdr.caplen = tp_snaplen;
+               pcaphdr.len = tp_len;
 
                /* if required build in place the sll header*/
                if (handle->md.cooked) {
 
                /* if required build in place the sll header*/
                if (handle->md.cooked) {
@@ -2061,7 +2977,9 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                         * don't step on the header when we construct
                         * the sll header.
                         */
                         * don't step on the header when we construct
                         * the sll header.
                         */
-                       if (bp < (u_char *)thdr + TPACKET_HDRLEN) {
+                       if (bp < (u_char *)h.raw +
+                                          TPACKET_ALIGN(handle->md.tp_hdrlen) +
+                                          sizeof(struct sockaddr_ll)) {
                                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                        "cooked-mode frame doesn't have room for sll header");
                                return -1;
                                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 
                                        "cooked-mode frame doesn't have room for sll header");
                                return -1;
@@ -2083,6 +3001,35 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                        pcaphdr.len += SLL_HDR_LEN;
                }
 
                        pcaphdr.len += SLL_HDR_LEN;
                }
 
+#ifdef HAVE_TPACKET2
+               if (handle->md.tp_version == TPACKET_V2 && h.h2->tp_vlan_tci &&
+                   tp_snaplen >= 2 * ETH_ALEN) {
+                       struct vlan_tag *tag;
+
+                       bp -= VLAN_TAG_LEN;
+                       memmove(bp, bp + VLAN_TAG_LEN, 2 * ETH_ALEN);
+
+                       tag = (struct vlan_tag *)(bp + 2 * ETH_ALEN);
+                       tag->vlan_tpid = htons(ETH_P_8021Q);
+                       tag->vlan_tci = htons(h.h2->tp_vlan_tci);
+
+                       pcaphdr.caplen += VLAN_TAG_LEN;
+                       pcaphdr.len += VLAN_TAG_LEN;
+               }
+#endif
+
+               /*
+                * 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 (pcaphdr.caplen > handle->snapshot)
+                       pcaphdr.caplen = handle->snapshot;
+
                /* pass the packet to the user */
                pkts++;
                callback(user, &pcaphdr, bp);
                /* pass the packet to the user */
                pkts++;
                callback(user, &pcaphdr, bp);
@@ -2090,7 +3037,16 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
 
 skip:
                /* next packet */
 
 skip:
                /* next packet */
-               thdr->tp_status = TP_STATUS_KERNEL;
+               switch (handle->md.tp_version) {
+               case TPACKET_V1:
+                       h.h1->tp_status = TP_STATUS_KERNEL;
+                       break;
+#ifdef HAVE_TPACKET2
+               case TPACKET_V2:
+                       h.h2->tp_status = TP_STATUS_KERNEL;
+                       break;
+#endif
+               }
                if (++handle->offset >= handle->cc)
                        handle->offset = 0;
 
                if (++handle->offset >= handle->cc)
                        handle->offset = 0;
 
@@ -2107,7 +3063,15 @@ static int
 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
 {
        int n, offset;
 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
 {
        int n, offset;
-       int ret = pcap_setfilter_linux(handle, filter);
+       int ret;
+
+       /*
+        * 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;
 
        if (ret < 0)
                return ret;
 
@@ -2124,7 +3088,7 @@ pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter)
        for (n=0; n < handle->cc; ++n) {
                if (--handle->offset < 0)
                        handle->offset = handle->cc - 1;
        for (n=0; n < handle->cc; ++n) {
                if (--handle->offset < 0)
                        handle->offset = handle->cc - 1;
-               if (RING_GET_FRAME(handle)->tp_status != TP_STATUS_KERNEL)
+               if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL))
                        break;
        }
 
                        break;
        }
 
@@ -2163,6 +3127,8 @@ iface_get_id(int fd, const char *device, char *ebuf)
 
 /*
  *  Bind the socket associated with FD to the given device.
 
 /*
  *  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)
  */
 static int
 iface_bind(int fd, int ifindex, char *ebuf)
@@ -2177,9 +3143,20 @@ iface_bind(int fd, int ifindex, char *ebuf)
        sll.sll_protocol        = htons(ETH_P_ALL);
 
        if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
        sll.sll_protocol        = htons(ETH_P_ALL);
 
        if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "bind: %s", pcap_strerror(errno));
-               return -1;
+               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? */
        }
 
        /* Any pending errors, e.g., network is down? */
@@ -2187,27 +3164,36 @@ iface_bind(int fd, int ifindex, char *ebuf)
        if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                        "getsockopt: %s", pcap_strerror(errno));
        if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                        "getsockopt: %s", pcap_strerror(errno));
-               return -2;
+               return 0;
        }
 
        }
 
-       if (err > 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));
                snprintf(ebuf, PCAP_ERRBUF_SIZE,
                        "bind: %s", pcap_strerror(err));
-               return -2;
+               return 0;
        }
 
        }
 
-       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
 /*
  * 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)
+has_wext(int sock_fd, const char *device, char *ebuf)
 {
 {
-#ifdef IW_MODE_MONITOR
        struct iwreq ireq;
 
        strncpy(ireq.ifr_ifrn.ifrn_name, device,
        struct iwreq ireq;
 
        strncpy(ireq.ifr_ifrn.ifrn_name, device,
@@ -2215,9 +3201,10 @@ has_wext(int sock_fd, const char *device)
        ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
        if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0)
                return 1;       /* yes */
        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;
        if (errno == ENODEV)
                return PCAP_ERROR_NO_SUCH_DEVICE;
-#endif
        return 0;
 }
 
        return 0;
 }
 
@@ -2226,6 +3213,9 @@ has_wext(int sock_fd, const char *device)
  * Per me si va ne l'etterno dolore,
  *     ...
  * Lasciate ogne speranza, voi ch'intrate.
  * 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,
  */
 typedef enum {
        MONITOR_WEXT,
@@ -2250,21 +3240,63 @@ typedef enum {
 static int
 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
 {
 static int
 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
 {
-#ifdef IW_MODE_MONITOR
        /*
         * 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.
        /*
         * 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.
         *
         * 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.
         *
         * 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;
         */
        int err;
        struct iwreq ireq;
@@ -2278,7 +3310,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
        /*
         * Does this device *support* the Wireless Extensions?
         */
        /*
         * Does this device *support* the Wireless Extensions?
         */
-       err = has_wext(sock_fd, device);
+       err = has_wext(sock_fd, device, handle->errbuf);
        if (err <= 0)
                return err;     /* either it doesn't or the device doesn't even exist */
        /*
        if (err <= 0)
                return err;     /* either it doesn't or the device doesn't even exist */
        /*
@@ -2295,7 +3327,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
        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;
        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 = args;
+       ireq.u.data.pointer = (void *)args;
        ireq.u.data.length = 0;
        ireq.u.data.flags = 0;
        if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
        ireq.u.data.length = 0;
        ireq.u.data.flags = 0;
        if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) {
@@ -2319,16 +3351,17 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
                return PCAP_ERROR;
        }
                    "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno));
                return PCAP_ERROR;
        }
-       priv = malloc(ireq.u.data.length);
+       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;
        }
        if (priv == NULL) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "malloc: %s", pcap_strerror(errno));
                return PCAP_ERROR;
        }
-       ireq.u.data.pointer = priv;
+       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));
        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;
        }
 
                return PCAP_ERROR;
        }
 
@@ -2344,8 +3377,10 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                         * Hostap driver, use this one.
                         * Set monitor mode first.
                         * You can set it to 0 to get DLT_IEEE80211,
                         * Hostap driver, use this one.
                         * Set monitor mode first.
                         * You can set it to 0 to get DLT_IEEE80211,
-                        * 1 to get DLT_PRISM, or 2 to get
-                        * DLT_IEEE80211_RADIO_AVS.
+                        * 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_TYPE_MASK) != IW_PRIV_TYPE_INT)
                                break;
@@ -2481,6 +3516,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                        }
                }
        }
                        }
                }
        }
+       free(priv);
 
        /*
         * XXX - ipw3945?  islism?
 
        /*
         * XXX - ipw3945?  islism?
@@ -2558,7 +3594,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                         * Note that we have to put the old mode back
                         * when we close the device.
                         */
                         * Note that we have to put the old mode back
                         * when we close the device.
                         */
-                       handle->md.must_clear |= MUST_CLEAR_RFMON;
+                       handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
 
                        /*
                         * Add this to the list of pcaps to close
 
                        /*
                         * Add this to the list of pcaps to close
@@ -2588,6 +3624,12 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                return PCAP_ERROR_RFMON_NOTSUP;
        }
 
                return PCAP_ERROR_RFMON_NOTSUP;
        }
 
+       /*
+        * XXX - airmon-ng does "iwconfig {if} key off" after setting
+        * monitor mode and setting the channel, and then does
+        * "iwconfig up".
+        */
+
        /*
         * Now select the appropriate radio header.
         */
        /*
         * Now select the appropriate radio header.
         */
@@ -2601,8 +3643,19 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
 
        case MONITOR_HOSTAP:
                /*
 
        case MONITOR_HOSTAP:
                /*
-                * Select the AVS header if we can, otherwise
-                * select the Prism header.
+                * 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 */
+
+               /*
+                * That failed.  Try to select the AVS header.
                 */
                memset(&ireq, 0, sizeof ireq);
                strncpy(ireq.ifr_ifrn.ifrn_name, device,
                 */
                memset(&ireq, 0, sizeof ireq);
                strncpy(ireq.ifr_ifrn.ifrn_name, device,
@@ -2610,18 +3663,19 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                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));
                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) {
-                       /*
-                        * Failure - try 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);
-               }
+               if (ioctl(sock_fd, cmd, &ireq) != -1)
+                       break;  /* success */
+
+               /*
+                * 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;
 
        case MONITOR_PRISM:
                break;
 
        case MONITOR_PRISM:
@@ -2732,7 +3786,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
         * Note that we have to put the old mode back when we
         * close the device.
         */
         * Note that we have to put the old mode back when we
         * close the device.
         */
-       handle->md.must_clear |= MUST_CLEAR_RFMON;
+       handle->md.must_do_on_close |= MUST_CLEAR_RFMON;
 
        /*
         * Add this to the list of pcaps to close when we exit.
 
        /*
         * Add this to the list of pcaps to close when we exit.
@@ -2740,13 +3794,41 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
        pcap_add_to_pcaps_to_close(handle);
 
        return 1;
        pcap_add_to_pcaps_to_close(handle);
 
        return 1;
-#else
+}
+#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 */
+
        /*
        /*
-        * We don't have the Wireless Extensions available, so we can't
-        * do monitor mode.
+        * 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;
         */
        return 0;
-#endif
 }
 
 #endif /* HAVE_PF_PACKET_SOCKETS */
 }
 
 #endif /* HAVE_PF_PACKET_SOCKETS */
@@ -2772,7 +3854,7 @@ activate_old(pcap_t *handle)
        if (handle->fd == -1) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "socket: %s", pcap_strerror(errno));
        if (handle->fd == -1) {
                snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
                         "socket: %s", pcap_strerror(errno));
-               return PCAP_ERROR;
+               return PCAP_ERROR_PERM_DENIED;
        }
 
        /* It worked - we are using the old interface */
        }
 
        /* It worked - we are using the old interface */
@@ -2783,7 +3865,7 @@ activate_old(pcap_t *handle)
 
        /* Bind to the given device */
 
 
        /* Bind to the given device */
 
-       if (!device) {
+       if (strcmp(device, "any") == 0) {
                strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
                        PCAP_ERRBUF_SIZE);
                return PCAP_ERROR;
                strncpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems",
                        PCAP_ERRBUF_SIZE);
                return PCAP_ERROR;
@@ -2848,7 +3930,7 @@ activate_old(pcap_t *handle)
                                         pcap_strerror(errno));
                                return PCAP_ERROR;
                        }
                                         pcap_strerror(errno));
                                return PCAP_ERROR;
                        }
-                       handle->md.must_clear |= MUST_CLEAR_PROMISC;
+                       handle->md.must_do_on_close |= MUST_CLEAR_PROMISC;
 
                        /*
                         * Add this to the list of pcaps
 
                        /*
                         * Add this to the list of pcaps
@@ -3007,14 +4089,14 @@ iface_get_arptype(int fd, const char *device, char *ebuf)
        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
 
        if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
        strncpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
 
        if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
+               snprintf(ebuf, PCAP_ERRBUF_SIZE,
+                        "SIOCGIFHWADDR: %s", pcap_strerror(errno));
                if (errno == ENODEV) {
                        /*
                         * No such device.
                         */
                        return PCAP_ERROR_NO_SUCH_DEVICE;
                }
                if (errno == ENODEV) {
                        /*
                         * No such device.
                         */
                        return PCAP_ERROR_NO_SUCH_DEVICE;
                }
-               snprintf(ebuf, PCAP_ERRBUF_SIZE,
-                        "SIOCGIFHWADDR: %s", pcap_strerror(errno));
                return PCAP_ERROR;
        }
 
                return PCAP_ERROR;
        }
 
@@ -3023,7 +4105,7 @@ iface_get_arptype(int fd, const char *device, char *ebuf)
 
 #ifdef SO_ATTACH_FILTER
 static int
 
 #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;
 {
        size_t prog_size;
        register int i;
@@ -3056,26 +4138,33 @@ fix_program(pcap_t *handle, struct sock_fprog *fcode)
 
                case BPF_RET:
                        /*
 
                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;
 
                        }
                        break;