]> The Tcpdump Group git mirrors - libpcap/blobdiff - pcap-linux.c
Fix typo.
[libpcap] / pcap-linux.c
index 91afa80751a3415387926244be2b02d5534d8372..de14a7ecd8c3cc41dfb30d1874e37c6626c4799e 100644 (file)
@@ -141,6 +141,7 @@ static const char rcsid[] _U_ =
  */
 #ifdef HAVE_LINUX_WIRELESS_H
 #include <linux/wireless.h>
+#endif /* HAVE_LINUX_WIRELESS_H */
 
 /*
  * Got libnl?
@@ -155,8 +156,6 @@ static const char rcsid[] _U_ =
 #include <netlink/attr.h>
 #endif /* HAVE_LIBNL */
 
-#endif /* HAVE_LINUX_WIRELESS_H */
-
 #include "pcap-int.h"
 #include "pcap/sll.h"
 #include "pcap/vlan.h"
@@ -296,6 +295,12 @@ static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
 static void pcap_cleanup_linux(pcap_t *);
 
+union thdr {
+       struct tpacket_hdr      *h1;
+       struct tpacket2_hdr     *h2;
+       void                    *raw;
+};
+
 #ifdef HAVE_PACKET_RING
 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset])
 
@@ -307,6 +312,8 @@ 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
 
 /*
@@ -328,7 +335,8 @@ static int  enter_rfmon_mode(pcap_t *handle, int sock_fd,
 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);
@@ -383,68 +391,93 @@ pcap_create(const char *device, char *ebuf)
        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
-pcap_can_set_rfmon_linux(pcap_t *p)
+get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
+    size_t phydev_max_pathlen)
 {
-#ifdef IW_MODE_MONITOR
-       int sock_fd;
-       struct iwreq ireq;
-#endif
-
-       if (strcmp(p->opt.source, "any") == 0) {
-               /*
-                * Monitor mode makes no sense on the "any" device.
-                */
-               return 0;
-       }
+       char *pathstr;
+       ssize_t bytes_read;
 
-#ifdef IW_MODE_MONITOR
        /*
-        * Bleah.  There doesn't appear to be an ioctl to use to ask
-        * whether a device supports monitor mode; we'll just do
-        * SIOCGIWMODE and, if it succeeds, assume the device supports
-        * monitor mode.
-        *
-        * Open a socket on which to attempt to get the mode.
-        * (We assume that if we have Wireless Extensions support
-        * we also have PF_PACKET support.)
-        *
-        * This also presumes that the mac80211 framework supports the
-        * Wireless Extensions, which appears to be the case at least
-        * as far back as the 2.6.22.6 kernel.
+        * Generate the path string for the symlink to the physical device.
         */
-       sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
-       if (sock_fd == -1) {
-               (void)snprintf(p->errbuf, PCAP_ERRBUF_SIZE,
-                   "socket: %s", pcap_strerror(errno));
+       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;
        }
-
-       /*
-        * Attempt to get the current mode.
-        */
-       strncpy(ireq.ifr_ifrn.ifrn_name, p->opt.source,
-           sizeof ireq.ifr_ifrn.ifrn_name);
-       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
-       if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
-               /*
-                * Well, we got the mode; assume we can set it.
-                */
-               close(sock_fd);
-               return 1;
-       }
-       if (errno == ENODEV) {
-               /* The device doesn't even exist. */
-               close(sock_fd);
-               return PCAP_ERROR_NO_SUCH_DEVICE;
+       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;
        }
-       close(sock_fd);
-#endif
-       return 0;
+       free(pathstr);
+       phydev_path[bytes_read] = '\0';
+       return 1;
 }
 
-#if defined(IW_MODE_MONITOR) && defined(HAVE_LIBNL)
-
 struct nl80211_state {
        struct nl_handle *nl_handle;
        struct nl_cache *nl_cache;
@@ -660,7 +693,200 @@ nla_put_failure:
        return PCAP_ERROR;
 }
 
-#endif /* defined(IW_MODE_MONITOR) && defined(HAVE_LIBNL) */
+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
+pcap_can_set_rfmon_linux(pcap_t *handle)
+{
+#ifdef HAVE_LIBNL
+       char phydev_path[PATH_MAX+1];
+       int ret;
+#endif
+#ifdef IW_MODE_MONITOR
+       int sock_fd;
+       struct iwreq ireq;
+#endif
+
+       if (strcmp(handle->opt.source, "any") == 0) {
+               /*
+                * Monitor mode makes no sense on the "any" device.
+                */
+               return 0;
+       }
+
+#ifdef HAVE_LIBNL
+       /*
+        * Bleah.  There doesn't seem to be a way to ask a mac80211
+        * device, through libnl, whether it supports monitor mode;
+        * we'll just check whether the device appears to be a
+        * mac80211 device and, if so, assume the device supports
+        * monitor mode.
+        *
+        * wmaster devices don't appear to support the Wireless
+        * Extensions, but we can create a mon device for a
+        * wmaster device, so we don't bother checking whether
+        * a mac80211 device supports the Wireless Extensions.
+        */
+       ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path,
+           PATH_MAX);
+       if (ret < 0)
+               return ret;     /* error */
+       if (ret == 1)
+               return 1;       /* mac80211 device */
+#endif
+
+#ifdef IW_MODE_MONITOR
+       /*
+        * Bleah.  There doesn't appear to be an ioctl to use to ask
+        * whether a device supports monitor mode; we'll just do
+        * SIOCGIWMODE and, if it succeeds, assume the device supports
+        * monitor mode.
+        *
+        * Open a socket on which to attempt to get the mode.
+        * (We assume that if we have Wireless Extensions support
+        * we also have PF_PACKET support.)
+        */
+       sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+       if (sock_fd == -1) {
+               (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                   "socket: %s", pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+
+       /*
+        * Attempt to get the current mode.
+        */
+       strncpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source,
+           sizeof ireq.ifr_ifrn.ifrn_name);
+       ireq.ifr_ifrn.ifrn_name[sizeof ireq.ifr_ifrn.ifrn_name - 1] = 0;
+       if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) {
+               /*
+                * Well, we got the mode; assume we can set it.
+                */
+               close(sock_fd);
+               return 1;
+       }
+       if (errno == ENODEV) {
+               /* The device doesn't even exist. */
+               close(sock_fd);
+               return PCAP_ERROR_NO_SUCH_DEVICE;
+       }
+       close(sock_fd);
+#endif
+       return 0;
+}
 
 /*
  * With older kernels promiscuous mode is kind of interesting because we
@@ -677,11 +903,11 @@ nla_put_failure:
 static void    pcap_cleanup_linux( pcap_t *handle )
 {
        struct ifreq    ifr;
-#ifdef IW_MODE_MONITOR
 #ifdef HAVE_LIBNL
        struct nl80211_state nlstate;
        int ret;
 #endif /* HAVE_LIBNL */
+#ifdef IW_MODE_MONITOR
        struct iwreq ireq;
 #endif /* IW_MODE_MONITOR */
 
@@ -728,7 +954,6 @@ static void pcap_cleanup_linux( pcap_t *handle )
                        }
                }
 
-#ifdef IW_MODE_MONITOR
 #ifdef HAVE_LIBNL
                if (handle->md.must_do_on_close & MUST_DELETE_MONIF) {
                        ret = nl80211_init(handle, &nlstate, handle->md.device);
@@ -746,6 +971,7 @@ static void pcap_cleanup_linux( pcap_t *handle )
                }
 #endif /* HAVE_LIBNL */
 
+#ifdef IW_MODE_MONITOR
                if (handle->md.must_do_on_close & MUST_CLEAR_RFMON) {
                        /*
                         * We put the interface into rfmon mode;
@@ -1469,7 +1695,8 @@ pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf)
  *  Attach the given BPF code to the packet capture device.
  */
 static int
-pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
+pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter,
+    int is_mmapped)
 {
 #ifdef SO_ATTACH_FILTER
        struct sock_fprog       fcode;
@@ -1522,13 +1749,13 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
                 *
                 * Oh, and we also need to fix it up so that all "ret"
                 * instructions with non-zero operands have 65535 as the
-                * operand, and so that, if we're in cooked mode, all
-                * memory-reference instructions use special magic offsets
-                * in references to the link-layer header and assume that
-                * the link-layer payload begins at 0; "fix_program()"
-                * will do that.
+                * operand if we're not capturing in memory-mapped modee,
+                * and so that, if we're in cooked mode, all memory-reference
+                * instructions use special magic offsets in references to
+                * the link-layer header and assume that the link-layer
+                * payload begins at 0; "fix_program()" will do that.
                 */
-               switch (fix_program(handle, &fcode)) {
+               switch (fix_program(handle, &fcode, is_mmapped)) {
 
                case -1:
                default:
@@ -1602,6 +1829,13 @@ pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
        return 0;
 }
 
+static int
+pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
+{
+       return pcap_setfilter_linux_common(handle, filter, 0);
+}
+
+
 /*
  * Set direction flag: Which packets do we accept on a forwarding
  * single device? IN, OUT or both?
@@ -1986,7 +2220,10 @@ activate_new(pcap_t *handle)
 #ifdef HAVE_PF_PACKET_SOCKETS
        const char              *device = handle->opt.source;
        int                     is_any_device = (strcmp(device, "any") == 0);
-       int                     sock_fd = -1, arptype, val;
+       int                     sock_fd = -1, arptype;
+#ifdef HAVE_PACKET_AUXDATA
+       int                     val;
+#endif
        int                     err = 0;
        struct packet_mreq      mr;
 
@@ -2269,16 +2506,32 @@ activate_mmap(pcap_t *handle)
 #ifdef HAVE_PACKET_RING
        int ret;
 
+       /*
+        * Attempt to allocate a buffer to hold the contents of one
+        * packet, for use by the oneshot callback.
+        */
+       handle->md.oneshot_buffer = malloc(handle->snapshot);
+       if (handle->md.oneshot_buffer == NULL) {
+               snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
+                        "can't allocate oneshot buffer: %s",
+                        pcap_strerror(errno));
+               return PCAP_ERROR;
+       }
+
        if (handle->opt.buffer_size == 0) {
                /* by default request 2M for the ring buffer */
                handle->opt.buffer_size = 2*1024*1024;
        }
        ret = prepare_tpacket_socket(handle);
-       if (ret != 1)
+       if (ret != 1) {
+               free(handle->md.oneshot_buffer);
                return ret;
+       }
        ret = create_ring(handle);
-       if (ret != 1)
+       if (ret != 1) {
+               free(handle->md.oneshot_buffer);
                return ret;
+       }
 
        /* override some defaults and inherit the other fields from
         * activate_new
@@ -2289,6 +2542,7 @@ activate_mmap(pcap_t *handle)
        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 */
@@ -2348,24 +2602,10 @@ prepare_tpacket_socket(pcap_t *handle)
        return 1;
 }
 
-static void
-compute_ring_block(int frame_size, unsigned *block_size, unsigned *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. */
-       *block_size = getpagesize();
-       while (*block_size < frame_size) 
-               *block_size <<= 1;
-
-       *frames_per_block = *block_size/frame_size;
-}
-
 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 
@@ -2377,20 +2617,40 @@ create_ring(pcap_t *handle)
                                          TPACKET_ALIGN(handle->md.tp_hdrlen) +
                                          sizeof(struct sockaddr_ll));
        req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size;
-       compute_ring_block(req.tp_frame_size, &req.tp_block_size, &frames_per_block);
-       req.tp_block_nr = req.tp_frame_nr / frames_per_block;
 
-       /* 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:
+       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))) {
-               /* try to reduce requested ring size to prevent memory failure */
                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;
                }
                if (errno == ENOPROTOOPT) {
@@ -2406,10 +2666,10 @@ retry:
        }
 
        /* 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) {
+       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));
 
@@ -2433,7 +2693,7 @@ retry:
        /* fill the header ring with proper frame ptr*/
        handle->offset = 0;
        for (i=0; i<req.tp_block_nr; ++i) {
-               void *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) {
                        RING_GET_FRAME(handle) = base;
                        base += req.tp_frame_size;
@@ -2456,21 +2716,50 @@ destroy_ring(pcap_t *handle)
                                (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 (handle->md.mmapbuf) {
+               /* do not test for mmap failure, as we can't recover from any error */
+               munmap(handle->md.mmapbuf, handle->md.mmapbuflen);
+               handle->md.mmapbuf = NULL;
        }
 }
 
+/*
+ * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
+ * for Linux mmapped capture.
+ *
+ * The problem is that pcap_next() and pcap_next_ex() expect the packet
+ * data handed to the callback to be valid after the callback returns,
+ * but pcap_read_linux_mmap() has to release that packet as soon as
+ * the callback returns (otherwise, the kernel thinks there's still
+ * at least one unprocessed packet available in the ring, so a select()
+ * will immediately return indicating that there's data to process), so,
+ * in the callback, we have to make a copy of the packet.
+ *
+ * Yes, this means that, if the capture is using the ring buffer, using
+ * pcap_next() or pcap_next_ex() requires more copies than using
+ * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
+ * pcap_next() or pcap_next_ex().
+ */
+static void
+pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h,
+    const u_char *bytes)
+{
+       struct pkt_for_oneshot *sp = (struct pkt_for_oneshot *)user;
+       bpf_u_int32 copylen;
+
+       *sp->hdr = *h;
+       memcpy(sp->pd->md.oneshot_buffer, bytes, h->caplen);
+       *sp->pkt = sp->pd->md.oneshot_buffer;
+}
+    
 static void
 pcap_cleanup_linux_mmap( pcap_t *handle )
 {
        destroy_ring(handle);
+       if (handle->md.oneshot_buffer != NULL) {
+               free(handle->md.oneshot_buffer);
+               handle->md.oneshot_buffer = NULL;
+       }
        pcap_cleanup_linux(handle);
 }
 
@@ -2488,11 +2777,26 @@ pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf)
        /* map each value to the corresponding 2's complement, to 
         * preserve the timeout value provided with pcap_set_timeout */
        if (nonblock) {
-               if (p->md.timeout > 0)
+               if (p->md.timeout >= 0) {
+                       /*
+                        * Timeout is non-negative, so we're not already
+                        * in non-blocking mode; set it to the 2's
+                        * complement, to make it negative, as an
+                        * indication that we're in non-blocking mode.
+                        */
                        p->md.timeout = p->md.timeout*-1 - 1;
-       } else 
-               if (p->md.timeout < 0)
+               }
+       } else {
+               if (p->md.timeout < 0) {
+                       /*
+                        * Timeout is negative, so we're not already
+                        * in blocking mode; reverse the previous
+                        * operation, to make the timeout non-negative
+                        * again.
+                        */
                        p->md.timeout = (p->md.timeout+1)*-1;
+               }
+       }
        return 0;
 }
 
@@ -2519,24 +2823,6 @@ pcap_get_ring_frame(pcap_t *handle, int status)
        return h.raw;
 }
 
-static inline void
-pcap_release_previous_ring_frame(pcap_t *handle)
-{
-       if (handle->md.prev_pkt.raw != NULL) {
-               switch (handle->md.tp_version) {
-               case TPACKET_V1:
-                       handle->md.prev_pkt.h1->tp_status = TP_STATUS_KERNEL;
-                       break;
-#ifdef HAVE_TPACKET2
-               case TPACKET_V2:
-                       handle->md.prev_pkt.h2->tp_status = TP_STATUS_KERNEL;
-                       break;
-#endif
-               }
-               handle->md.prev_pkt.raw = NULL;
-       }
-}
-
 static int
 pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback, 
                u_char *user)
@@ -2584,44 +2870,10 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                unsigned int tp_sec;
                unsigned int tp_usec;
 
-               /*
-                * Check for break loop condition; a callback might have
-                * set it.
-                */
-               if (handle->break_loop) {
-                       handle->break_loop = 0;
-                       return -2;
-               }
-
                h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER);
                if (!h.raw)
                        break;
 
-               /*
-                * We have a packet; release the previous packet,
-                * if any.
-                *
-                * Libpcap has never guaranteed that, if we get a
-                * packet from the underlying packet capture
-                * mechanism, the data passed to callbacks for
-                * any previous packets is still valid.  It did
-                * implicitly guarantee that the data will still
-                * be available after the callback returns, by
-                * virtue of implementing pcap_next() by calling
-                * pcap_dispatch() with a count of 1 and a callback
-                * that fills in a structure with a pointer to
-                * the packet data, meaning that pointer is
-                * expected to point to valid data after the
-                * callback returns and pcap_next() returns,
-                * so we can't release the packet when the
-                * callback returns.
-                *
-                * Therefore, we remember the packet that
-                * needs to be released after handing it
-                * to the callback, and release it up here.
-                */
-               pcap_release_previous_ring_frame(handle);
-
                switch (handle->md.tp_version) {
                case TPACKET_V1:
                        tp_len     = h.h1->tp_len;
@@ -2766,20 +3018,35 @@ pcap_read_linux_mmap(pcap_t *handle, int max_packets, pcap_handler callback,
                }
 #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);
                handle->md.packets_read++;
 
 skip:
-               /*
-                * As per the comment above, we can't yet release this
-                * packet, even though the callback has returned, as
-                * some users of pcap_loop() and pcap_dispatch() - such
-                * as pcap_next() and pcap_next_ex() - expect the packet
-                * to be available until the next pcap_dispatch() call.
-                */
-               handle->md.prev_pkt = h;
+               /* next packet */
+               switch (handle->md.tp_version) {
+               case TPACKET_V1:
+                       h.h1->tp_status = TP_STATUS_KERNEL;
+                       break;
+#ifdef HAVE_TPACKET2
+               case TPACKET_V2:
+                       h.h2->tp_status = TP_STATUS_KERNEL;
+                       break;
+#endif
+               }
                if (++handle->offset >= handle->cc)
                        handle->offset = 0;
 
@@ -2796,7 +3063,15 @@ static int
 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;
 
@@ -2954,207 +3229,6 @@ typedef enum {
        MONITOR_RTL8XXX
 } monitor_type;
 
-/*
-        *
-        * 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.
-*/
-
-#ifdef HAVE_LIBNL
-/*
- * 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;
-}
-
-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 */
-
 /*
  * Use the Wireless Extensions, if we have them, to try to turn monitor mode
  * on if it's not already on.
@@ -3253,7 +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;
-       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) {
@@ -3283,7 +3357,7 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
                         "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));
@@ -3729,8 +3803,9 @@ enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device)
 static int
 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
 {
-#ifdef IW_MODE_MONITOR
+#if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR)
        int ret;
+#endif
 
 #ifdef HAVE_LIBNL
        ret = enter_rfmon_mode_mac80211(handle, sock_fd, device);
@@ -3740,6 +3815,7 @@ enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
                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 */
@@ -4029,7 +4105,7 @@ iface_get_arptype(int fd, const char *device, char *ebuf)
 
 #ifdef SO_ATTACH_FILTER
 static int
-fix_program(pcap_t *handle, struct sock_fprog *fcode)
+fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped)
 {
        size_t prog_size;
        register int i;
@@ -4062,26 +4138,33 @@ fix_program(pcap_t *handle, struct sock_fprog *fcode)
 
                case BPF_RET:
                        /*
-                        * It's a return instruction; is the snapshot
-                        * length a constant, rather than the contents
-                        * of the accumulator?
+                        * It's a return instruction; are we capturing
+                        * in memory-mapped mode?
                         */
-                       if (BPF_MODE(p->code) == BPF_K) {
+                       if (!is_mmapped) {
                                /*
-                                * Yes - if the value to be returned,
-                                * i.e. the snapshot length, is anything
-                                * other than 0, make it 65535, so that
-                                * the packet is truncated by "recvfrom()",
-                                * not by the filter.
-                                *
-                                * XXX - there's nothing we can easily do
-                                * if it's getting the value from the
-                                * accumulator; we'd have to insert
-                                * code to force non-zero values to be
-                                * 65535.
+                                * No; is the snapshot length a constant,
+                                * rather than the contents of the
+                                * accumulator?
                                 */
-                               if (p->k != 0)
-                                       p->k = 65535;
+                               if (BPF_MODE(p->code) == BPF_K) {
+                                       /*
+                                        * Yes - if the value to be returned,
+                                        * i.e. the snapshot length, is
+                                        * anything other than 0, make it
+                                        * 65535, so that the packet is
+                                        * truncated by "recvfrom()",
+                                        * not by the filter.
+                                        *
+                                        * XXX - there's nothing we can
+                                        * easily do if it's getting the
+                                        * value from the accumulator; we'd
+                                        * have to insert code to force
+                                        * non-zero values to be 65535.
+                                        */
+                                       if (p->k != 0)
+                                               p->k = 65535;
+                               }
                        }
                        break;