The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/net/bpf.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1990, 1991, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from the Stanford/CMU enet packet filter,
    6  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
    7  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
    8  * Berkeley Laboratory.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)bpf.c       8.4 (Berkeley) 1/9/95
   35  */
   36 
   37 #include <sys/cdefs.h>
   38 __FBSDID("$FreeBSD$");
   39 
   40 #include "opt_bpf.h"
   41 #include "opt_compat.h"
   42 #include "opt_ddb.h"
   43 #include "opt_netgraph.h"
   44 
   45 #include <sys/types.h>
   46 #include <sys/param.h>
   47 #include <sys/lock.h>
   48 #include <sys/rwlock.h>
   49 #include <sys/systm.h>
   50 #include <sys/conf.h>
   51 #include <sys/fcntl.h>
   52 #include <sys/jail.h>
   53 #include <sys/malloc.h>
   54 #include <sys/mbuf.h>
   55 #include <sys/time.h>
   56 #include <sys/priv.h>
   57 #include <sys/proc.h>
   58 #include <sys/signalvar.h>
   59 #include <sys/filio.h>
   60 #include <sys/sockio.h>
   61 #include <sys/ttycom.h>
   62 #include <sys/uio.h>
   63 
   64 #include <sys/event.h>
   65 #include <sys/file.h>
   66 #include <sys/poll.h>
   67 #include <sys/proc.h>
   68 
   69 #include <sys/socket.h>
   70 
   71 #ifdef DDB
   72 #include <ddb/ddb.h>
   73 #endif
   74 
   75 #include <net/if.h>
   76 #include <net/if_var.h>
   77 #include <net/if_dl.h>
   78 #include <net/bpf.h>
   79 #include <net/bpf_buffer.h>
   80 #ifdef BPF_JITTER
   81 #include <net/bpf_jitter.h>
   82 #endif
   83 #include <net/bpf_zerocopy.h>
   84 #include <net/bpfdesc.h>
   85 #include <net/route.h>
   86 #include <net/vnet.h>
   87 
   88 #include <netinet/in.h>
   89 #include <netinet/if_ether.h>
   90 #include <sys/kernel.h>
   91 #include <sys/sysctl.h>
   92 
   93 #include <net80211/ieee80211_freebsd.h>
   94 
   95 #include <security/mac/mac_framework.h>
   96 
   97 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
   98 
   99 static struct bpf_if_ext dead_bpf_if = {
  100         .bif_dlist = LIST_HEAD_INITIALIZER()
  101 };
  102 
  103 struct bpf_if {
  104 #define bif_next        bif_ext.bif_next
  105 #define bif_dlist       bif_ext.bif_dlist
  106         struct bpf_if_ext bif_ext;      /* public members */
  107         u_int           bif_dlt;        /* link layer type */
  108         u_int           bif_hdrlen;     /* length of link header */
  109         struct ifnet    *bif_ifp;       /* corresponding interface */
  110         struct rwlock   bif_lock;       /* interface lock */
  111         LIST_HEAD(, bpf_d) bif_wlist;   /* writer-only list */
  112         int             bif_flags;      /* Interface flags */
  113         struct bpf_if   **bif_bpf;      /* Pointer to pointer to us */
  114 };
  115 
  116 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
  117 
  118 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
  119 
  120 #define PRINET  26                      /* interruptible */
  121 
  122 #define SIZEOF_BPF_HDR(type)    \
  123     (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
  124 
  125 #ifdef COMPAT_FREEBSD32
  126 #include <sys/mount.h>
  127 #include <compat/freebsd32/freebsd32.h>
  128 #define BPF_ALIGNMENT32 sizeof(int32_t)
  129 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
  130 
  131 #ifndef BURN_BRIDGES
  132 /*
  133  * 32-bit version of structure prepended to each packet.  We use this header
  134  * instead of the standard one for 32-bit streams.  We mark the a stream as
  135  * 32-bit the first time we see a 32-bit compat ioctl request.
  136  */
  137 struct bpf_hdr32 {
  138         struct timeval32 bh_tstamp;     /* time stamp */
  139         uint32_t        bh_caplen;      /* length of captured portion */
  140         uint32_t        bh_datalen;     /* original length of packet */
  141         uint16_t        bh_hdrlen;      /* length of bpf header (this struct
  142                                            plus alignment padding) */
  143 };
  144 #endif
  145 
  146 struct bpf_program32 {
  147         u_int bf_len;
  148         uint32_t bf_insns;
  149 };
  150 
  151 struct bpf_dltlist32 {
  152         u_int   bfl_len;
  153         u_int   bfl_list;
  154 };
  155 
  156 #define BIOCSETF32      _IOW('B', 103, struct bpf_program32)
  157 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
  158 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
  159 #define BIOCGDLTLIST32  _IOWR('B', 121, struct bpf_dltlist32)
  160 #define BIOCSETWF32     _IOW('B', 123, struct bpf_program32)
  161 #define BIOCSETFNR32    _IOW('B', 130, struct bpf_program32)
  162 #endif
  163 
  164 #define BPF_LOCK()         sx_xlock(&bpf_sx)
  165 #define BPF_UNLOCK()            sx_xunlock(&bpf_sx)
  166 #define BPF_LOCK_ASSERT()       sx_assert(&bpf_sx, SA_XLOCKED)
  167 /*
  168  * bpf_iflist is a list of BPF interface structures, each corresponding to a
  169  * specific DLT.  The same network interface might have several BPF interface
  170  * structures registered by different layers in the stack (i.e., 802.11
  171  * frames, ethernet frames, etc).
  172  */
  173 static LIST_HEAD(, bpf_if)      bpf_iflist, bpf_freelist;
  174 static struct sx        bpf_sx;         /* bpf global lock */
  175 static int              bpf_bpfd_cnt;
  176 
  177 static void     bpf_attachd(struct bpf_d *, struct bpf_if *);
  178 static void     bpf_detachd(struct bpf_d *);
  179 static void     bpf_detachd_locked(struct bpf_d *);
  180 static void     bpf_freed(struct bpf_d *);
  181 static int      bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
  182                     struct sockaddr *, int *, struct bpf_d *);
  183 static int      bpf_setif(struct bpf_d *, struct ifreq *);
  184 static void     bpf_timed_out(void *);
  185 static __inline void
  186                 bpf_wakeup(struct bpf_d *);
  187 static void     catchpacket(struct bpf_d *, u_char *, u_int, u_int,
  188                     void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
  189                     struct bintime *);
  190 static void     reset_d(struct bpf_d *);
  191 static int      bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
  192 static int      bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
  193 static int      bpf_setdlt(struct bpf_d *, u_int);
  194 static void     filt_bpfdetach(struct knote *);
  195 static int      filt_bpfread(struct knote *, long);
  196 static void     bpf_drvinit(void *);
  197 static int      bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
  198 
  199 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
  200 int bpf_maxinsns = BPF_MAXINSNS;
  201 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
  202     &bpf_maxinsns, 0, "Maximum bpf program instructions");
  203 static int bpf_zerocopy_enable = 0;
  204 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
  205     &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
  206 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
  207     bpf_stats_sysctl, "bpf statistics portal");
  208 
  209 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
  210 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
  211 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
  212     &VNET_NAME(bpf_optimize_writers), 0,
  213     "Do not send packets until BPF program is set");
  214 
  215 static  d_open_t        bpfopen;
  216 static  d_read_t        bpfread;
  217 static  d_write_t       bpfwrite;
  218 static  d_ioctl_t       bpfioctl;
  219 static  d_poll_t        bpfpoll;
  220 static  d_kqfilter_t    bpfkqfilter;
  221 
  222 static struct cdevsw bpf_cdevsw = {
  223         .d_version =    D_VERSION,
  224         .d_open =       bpfopen,
  225         .d_read =       bpfread,
  226         .d_write =      bpfwrite,
  227         .d_ioctl =      bpfioctl,
  228         .d_poll =       bpfpoll,
  229         .d_name =       "bpf",
  230         .d_kqfilter =   bpfkqfilter,
  231 };
  232 
  233 static struct filterops bpfread_filtops = {
  234         .f_isfd = 1,
  235         .f_detach = filt_bpfdetach,
  236         .f_event = filt_bpfread,
  237 };
  238 
  239 eventhandler_tag        bpf_ifdetach_cookie = NULL;
  240 
  241 /*
  242  * LOCKING MODEL USED BY BPF:
  243  * Locks:
  244  * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
  245  * some global counters and every bpf_if reference.
  246  * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
  247  * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
  248  *   used by bpf_mtap code.
  249  *
  250  * Lock order:
  251  *
  252  * Global lock, interface lock, descriptor lock
  253  *
  254  * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
  255  * working model. In many places (like bpf_detachd) we start with BPF descriptor
  256  * (and we need to at least rlock it to get reliable interface pointer). This
  257  * gives us potential LOR. As a result, we use global lock to protect from bpf_if
  258  * change in every such place.
  259  *
  260  * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
  261  * 3) descriptor main wlock.
  262  * Reading bd_bif can be protected by any of these locks, typically global lock.
  263  *
  264  * Changing read/write BPF filter is protected by the same three locks,
  265  * the same applies for reading.
  266  *
  267  * Sleeping in global lock is not allowed due to bpfdetach() using it.
  268  */
  269 
  270 /*
  271  * Wrapper functions for various buffering methods.  If the set of buffer
  272  * modes expands, we will probably want to introduce a switch data structure
  273  * similar to protosw, et.
  274  */
  275 static void
  276 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
  277     u_int len)
  278 {
  279 
  280         BPFD_LOCK_ASSERT(d);
  281 
  282         switch (d->bd_bufmode) {
  283         case BPF_BUFMODE_BUFFER:
  284                 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
  285 
  286         case BPF_BUFMODE_ZBUF:
  287                 d->bd_zcopy++;
  288                 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
  289 
  290         default:
  291                 panic("bpf_buf_append_bytes");
  292         }
  293 }
  294 
  295 static void
  296 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
  297     u_int len)
  298 {
  299 
  300         BPFD_LOCK_ASSERT(d);
  301 
  302         switch (d->bd_bufmode) {
  303         case BPF_BUFMODE_BUFFER:
  304                 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
  305 
  306         case BPF_BUFMODE_ZBUF:
  307                 d->bd_zcopy++;
  308                 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
  309 
  310         default:
  311                 panic("bpf_buf_append_mbuf");
  312         }
  313 }
  314 
  315 /*
  316  * This function gets called when the free buffer is re-assigned.
  317  */
  318 static void
  319 bpf_buf_reclaimed(struct bpf_d *d)
  320 {
  321 
  322         BPFD_LOCK_ASSERT(d);
  323 
  324         switch (d->bd_bufmode) {
  325         case BPF_BUFMODE_BUFFER:
  326                 return;
  327 
  328         case BPF_BUFMODE_ZBUF:
  329                 bpf_zerocopy_buf_reclaimed(d);
  330                 return;
  331 
  332         default:
  333                 panic("bpf_buf_reclaimed");
  334         }
  335 }
  336 
  337 /*
  338  * If the buffer mechanism has a way to decide that a held buffer can be made
  339  * free, then it is exposed via the bpf_canfreebuf() interface.  (1) is
  340  * returned if the buffer can be discarded, (0) is returned if it cannot.
  341  */
  342 static int
  343 bpf_canfreebuf(struct bpf_d *d)
  344 {
  345 
  346         BPFD_LOCK_ASSERT(d);
  347 
  348         switch (d->bd_bufmode) {
  349         case BPF_BUFMODE_ZBUF:
  350                 return (bpf_zerocopy_canfreebuf(d));
  351         }
  352         return (0);
  353 }
  354 
  355 /*
  356  * Allow the buffer model to indicate that the current store buffer is
  357  * immutable, regardless of the appearance of space.  Return (1) if the
  358  * buffer is writable, and (0) if not.
  359  */
  360 static int
  361 bpf_canwritebuf(struct bpf_d *d)
  362 {
  363         BPFD_LOCK_ASSERT(d);
  364 
  365         switch (d->bd_bufmode) {
  366         case BPF_BUFMODE_ZBUF:
  367                 return (bpf_zerocopy_canwritebuf(d));
  368         }
  369         return (1);
  370 }
  371 
  372 /*
  373  * Notify buffer model that an attempt to write to the store buffer has
  374  * resulted in a dropped packet, in which case the buffer may be considered
  375  * full.
  376  */
  377 static void
  378 bpf_buffull(struct bpf_d *d)
  379 {
  380 
  381         BPFD_LOCK_ASSERT(d);
  382 
  383         switch (d->bd_bufmode) {
  384         case BPF_BUFMODE_ZBUF:
  385                 bpf_zerocopy_buffull(d);
  386                 break;
  387         }
  388 }
  389 
  390 /*
  391  * Notify the buffer model that a buffer has moved into the hold position.
  392  */
  393 void
  394 bpf_bufheld(struct bpf_d *d)
  395 {
  396 
  397         BPFD_LOCK_ASSERT(d);
  398 
  399         switch (d->bd_bufmode) {
  400         case BPF_BUFMODE_ZBUF:
  401                 bpf_zerocopy_bufheld(d);
  402                 break;
  403         }
  404 }
  405 
  406 static void
  407 bpf_free(struct bpf_d *d)
  408 {
  409 
  410         switch (d->bd_bufmode) {
  411         case BPF_BUFMODE_BUFFER:
  412                 return (bpf_buffer_free(d));
  413 
  414         case BPF_BUFMODE_ZBUF:
  415                 return (bpf_zerocopy_free(d));
  416 
  417         default:
  418                 panic("bpf_buf_free");
  419         }
  420 }
  421 
  422 static int
  423 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
  424 {
  425 
  426         if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
  427                 return (EOPNOTSUPP);
  428         return (bpf_buffer_uiomove(d, buf, len, uio));
  429 }
  430 
  431 static int
  432 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
  433 {
  434 
  435         if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
  436                 return (EOPNOTSUPP);
  437         return (bpf_buffer_ioctl_sblen(d, i));
  438 }
  439 
  440 static int
  441 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
  442 {
  443 
  444         if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
  445                 return (EOPNOTSUPP);
  446         return (bpf_zerocopy_ioctl_getzmax(td, d, i));
  447 }
  448 
  449 static int
  450 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
  451 {
  452 
  453         if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
  454                 return (EOPNOTSUPP);
  455         return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
  456 }
  457 
  458 static int
  459 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
  460 {
  461 
  462         if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
  463                 return (EOPNOTSUPP);
  464         return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
  465 }
  466 
  467 /*
  468  * General BPF functions.
  469  */
  470 static int
  471 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
  472     struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
  473 {
  474         const struct ieee80211_bpf_params *p;
  475         struct ether_header *eh;
  476         struct mbuf *m;
  477         int error;
  478         int len;
  479         int hlen;
  480         int slen;
  481 
  482         /*
  483          * Build a sockaddr based on the data link layer type.
  484          * We do this at this level because the ethernet header
  485          * is copied directly into the data field of the sockaddr.
  486          * In the case of SLIP, there is no header and the packet
  487          * is forwarded as is.
  488          * Also, we are careful to leave room at the front of the mbuf
  489          * for the link level header.
  490          */
  491         switch (linktype) {
  492 
  493         case DLT_SLIP:
  494                 sockp->sa_family = AF_INET;
  495                 hlen = 0;
  496                 break;
  497 
  498         case DLT_EN10MB:
  499                 sockp->sa_family = AF_UNSPEC;
  500                 /* XXX Would MAXLINKHDR be better? */
  501                 hlen = ETHER_HDR_LEN;
  502                 break;
  503 
  504         case DLT_FDDI:
  505                 sockp->sa_family = AF_IMPLINK;
  506                 hlen = 0;
  507                 break;
  508 
  509         case DLT_RAW:
  510                 sockp->sa_family = AF_UNSPEC;
  511                 hlen = 0;
  512                 break;
  513 
  514         case DLT_NULL:
  515                 /*
  516                  * null interface types require a 4 byte pseudo header which
  517                  * corresponds to the address family of the packet.
  518                  */
  519                 sockp->sa_family = AF_UNSPEC;
  520                 hlen = 4;
  521                 break;
  522 
  523         case DLT_ATM_RFC1483:
  524                 /*
  525                  * en atm driver requires 4-byte atm pseudo header.
  526                  * though it isn't standard, vpi:vci needs to be
  527                  * specified anyway.
  528                  */
  529                 sockp->sa_family = AF_UNSPEC;
  530                 hlen = 12;      /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
  531                 break;
  532 
  533         case DLT_PPP:
  534                 sockp->sa_family = AF_UNSPEC;
  535                 hlen = 4;       /* This should match PPP_HDRLEN */
  536                 break;
  537 
  538         case DLT_IEEE802_11:            /* IEEE 802.11 wireless */
  539                 sockp->sa_family = AF_IEEE80211;
  540                 hlen = 0;
  541                 break;
  542 
  543         case DLT_IEEE802_11_RADIO:      /* IEEE 802.11 wireless w/ phy params */
  544                 sockp->sa_family = AF_IEEE80211;
  545                 sockp->sa_len = 12;     /* XXX != 0 */
  546                 hlen = sizeof(struct ieee80211_bpf_params);
  547                 break;
  548 
  549         default:
  550                 return (EIO);
  551         }
  552 
  553         len = uio->uio_resid;
  554         if (len < hlen || len - hlen > ifp->if_mtu)
  555                 return (EMSGSIZE);
  556 
  557         m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
  558         if (m == NULL)
  559                 return (EIO);
  560         m->m_pkthdr.len = m->m_len = len;
  561         *mp = m;
  562 
  563         error = uiomove(mtod(m, u_char *), len, uio);
  564         if (error)
  565                 goto bad;
  566 
  567         slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
  568         if (slen == 0) {
  569                 error = EPERM;
  570                 goto bad;
  571         }
  572 
  573         /* Check for multicast destination */
  574         switch (linktype) {
  575         case DLT_EN10MB:
  576                 eh = mtod(m, struct ether_header *);
  577                 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
  578                         if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
  579                             ETHER_ADDR_LEN) == 0)
  580                                 m->m_flags |= M_BCAST;
  581                         else
  582                                 m->m_flags |= M_MCAST;
  583                 }
  584                 if (d->bd_hdrcmplt == 0) {
  585                         memcpy(eh->ether_shost, IF_LLADDR(ifp),
  586                             sizeof(eh->ether_shost));
  587                 }
  588                 break;
  589         }
  590 
  591         /*
  592          * Make room for link header, and copy it to sockaddr
  593          */
  594         if (hlen != 0) {
  595                 if (sockp->sa_family == AF_IEEE80211) {
  596                         /*
  597                          * Collect true length from the parameter header
  598                          * NB: sockp is known to be zero'd so if we do a
  599                          *     short copy unspecified parameters will be
  600                          *     zero.
  601                          * NB: packet may not be aligned after stripping
  602                          *     bpf params
  603                          * XXX check ibp_vers
  604                          */
  605                         p = mtod(m, const struct ieee80211_bpf_params *);
  606                         hlen = p->ibp_len;
  607                         if (hlen > sizeof(sockp->sa_data)) {
  608                                 error = EINVAL;
  609                                 goto bad;
  610                         }
  611                 }
  612                 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
  613         }
  614         *hdrlen = hlen;
  615 
  616         return (0);
  617 bad:
  618         m_freem(m);
  619         return (error);
  620 }
  621 
  622 /*
  623  * Attach file to the bpf interface, i.e. make d listen on bp.
  624  */
  625 static void
  626 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
  627 {
  628         int op_w;
  629 
  630         BPF_LOCK_ASSERT();
  631 
  632         /*
  633          * Save sysctl value to protect from sysctl change
  634          * between reads
  635          */
  636         op_w = V_bpf_optimize_writers || d->bd_writer;
  637 
  638         if (d->bd_bif != NULL)
  639                 bpf_detachd_locked(d);
  640         /*
  641          * Point d at bp, and add d to the interface's list.
  642          * Since there are many applications using BPF for
  643          * sending raw packets only (dhcpd, cdpd are good examples)
  644          * we can delay adding d to the list of active listeners until
  645          * some filter is configured.
  646          */
  647 
  648         BPFIF_WLOCK(bp);
  649         BPFD_LOCK(d);
  650 
  651         d->bd_bif = bp;
  652 
  653         if (op_w != 0) {
  654                 /* Add to writers-only list */
  655                 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
  656                 /*
  657                  * We decrement bd_writer on every filter set operation.
  658                  * First BIOCSETF is done by pcap_open_live() to set up
  659                  * snap length. After that appliation usually sets its own filter
  660                  */
  661                 d->bd_writer = 2;
  662         } else
  663                 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
  664 
  665         BPFD_UNLOCK(d);
  666         BPFIF_WUNLOCK(bp);
  667 
  668         bpf_bpfd_cnt++;
  669 
  670         CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
  671             __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
  672 
  673         if (op_w == 0)
  674                 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
  675 }
  676 
  677 /*
  678  * Check if we need to upgrade our descriptor @d from write-only mode.
  679  */
  680 static int
  681 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
  682 {
  683         int is_snap, need_upgrade;
  684 
  685         /*
  686          * Check if we've already upgraded or new filter is empty.
  687          */
  688         if (d->bd_writer == 0 || fcode == NULL)
  689                 return (0);
  690 
  691         need_upgrade = 0;
  692 
  693         /*
  694          * Check if cmd looks like snaplen setting from
  695          * pcap_bpf.c:pcap_open_live().
  696          * Note we're not checking .k value here:
  697          * while pcap_open_live() definitely sets to non-zero value,
  698          * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
  699          * do not consider upgrading immediately
  700          */
  701         if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
  702                 is_snap = 1;
  703         else
  704                 is_snap = 0;
  705 
  706         if (is_snap == 0) {
  707                 /*
  708                  * We're setting first filter and it doesn't look like
  709                  * setting snaplen.  We're probably using bpf directly.
  710                  * Upgrade immediately.
  711                  */
  712                 need_upgrade = 1;
  713         } else {
  714                 /*
  715                  * Do not require upgrade by first BIOCSETF
  716                  * (used to set snaplen) by pcap_open_live().
  717                  */
  718 
  719                 if (--d->bd_writer == 0) {
  720                         /*
  721                          * First snaplen filter has already
  722                          * been set. This is probably catch-all
  723                          * filter
  724                          */
  725                         need_upgrade = 1;
  726                 }
  727         }
  728 
  729         CTR5(KTR_NET,
  730             "%s: filter function set by pid %d, "
  731             "bd_writer counter %d, snap %d upgrade %d",
  732             __func__, d->bd_pid, d->bd_writer,
  733             is_snap, need_upgrade);
  734 
  735         return (need_upgrade);
  736 }
  737 
  738 /*
  739  * Add d to the list of active bp filters.
  740  * Requires bpf_attachd() to be called before.
  741  */
  742 static void
  743 bpf_upgraded(struct bpf_d *d)
  744 {
  745         struct bpf_if *bp;
  746 
  747         BPF_LOCK_ASSERT();
  748 
  749         bp = d->bd_bif;
  750 
  751         /*
  752          * Filter can be set several times without specifying interface.
  753          * Mark d as reader and exit.
  754          */
  755         if (bp == NULL) {
  756                 BPFD_LOCK(d);
  757                 d->bd_writer = 0;
  758                 BPFD_UNLOCK(d);
  759                 return;
  760         }
  761 
  762         BPFIF_WLOCK(bp);
  763         BPFD_LOCK(d);
  764 
  765         /* Remove from writers-only list */
  766         LIST_REMOVE(d, bd_next);
  767         LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
  768         /* Mark d as reader */
  769         d->bd_writer = 0;
  770 
  771         BPFD_UNLOCK(d);
  772         BPFIF_WUNLOCK(bp);
  773 
  774         CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
  775 
  776         EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
  777 }
  778 
  779 /*
  780  * Detach a file from its interface.
  781  */
  782 static void
  783 bpf_detachd(struct bpf_d *d)
  784 {
  785         BPF_LOCK();
  786         bpf_detachd_locked(d);
  787         BPF_UNLOCK();
  788 }
  789 
  790 static void
  791 bpf_detachd_locked(struct bpf_d *d)
  792 {
  793         int error;
  794         struct bpf_if *bp;
  795         struct ifnet *ifp;
  796 
  797         CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
  798 
  799         BPF_LOCK_ASSERT();
  800 
  801         /* Check if descriptor is attached */
  802         if ((bp = d->bd_bif) == NULL)
  803                 return;
  804 
  805         BPFIF_WLOCK(bp);
  806         BPFD_LOCK(d);
  807 
  808         /* Save bd_writer value */
  809         error = d->bd_writer;
  810 
  811         /*
  812          * Remove d from the interface's descriptor list.
  813          */
  814         LIST_REMOVE(d, bd_next);
  815 
  816         ifp = bp->bif_ifp;
  817         d->bd_bif = NULL;
  818         BPFD_UNLOCK(d);
  819         BPFIF_WUNLOCK(bp);
  820 
  821         bpf_bpfd_cnt--;
  822 
  823         /* Call event handler iff d is attached */
  824         if (error == 0)
  825                 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
  826 
  827         /*
  828          * Check if this descriptor had requested promiscuous mode.
  829          * If so, turn it off.
  830          */
  831         if (d->bd_promisc) {
  832                 d->bd_promisc = 0;
  833                 CURVNET_SET(ifp->if_vnet);
  834                 error = ifpromisc(ifp, 0);
  835                 CURVNET_RESTORE();
  836                 if (error != 0 && error != ENXIO) {
  837                         /*
  838                          * ENXIO can happen if a pccard is unplugged
  839                          * Something is really wrong if we were able to put
  840                          * the driver into promiscuous mode, but can't
  841                          * take it out.
  842                          */
  843                         if_printf(bp->bif_ifp,
  844                                 "bpf_detach: ifpromisc failed (%d)\n", error);
  845                 }
  846         }
  847 }
  848 
  849 /*
  850  * Close the descriptor by detaching it from its interface,
  851  * deallocating its buffers, and marking it free.
  852  */
  853 static void
  854 bpf_dtor(void *data)
  855 {
  856         struct bpf_d *d = data;
  857 
  858         BPFD_LOCK(d);
  859         if (d->bd_state == BPF_WAITING)
  860                 callout_stop(&d->bd_callout);
  861         d->bd_state = BPF_IDLE;
  862         BPFD_UNLOCK(d);
  863         funsetown(&d->bd_sigio);
  864         bpf_detachd(d);
  865 #ifdef MAC
  866         mac_bpfdesc_destroy(d);
  867 #endif /* MAC */
  868         seldrain(&d->bd_sel);
  869         knlist_destroy(&d->bd_sel.si_note);
  870         callout_drain(&d->bd_callout);
  871         bpf_freed(d);
  872         free(d, M_BPF);
  873 }
  874 
  875 /*
  876  * Open ethernet device.  Returns ENXIO for illegal minor device number,
  877  * EBUSY if file is open by another process.
  878  */
  879 /* ARGSUSED */
  880 static  int
  881 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
  882 {
  883         struct bpf_d *d;
  884         int error;
  885 
  886         d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
  887         error = devfs_set_cdevpriv(d, bpf_dtor);
  888         if (error != 0) {
  889                 free(d, M_BPF);
  890                 return (error);
  891         }
  892 
  893         /*
  894          * For historical reasons, perform a one-time initialization call to
  895          * the buffer routines, even though we're not yet committed to a
  896          * particular buffer method.
  897          */
  898         bpf_buffer_init(d);
  899         if ((flags & FREAD) == 0)
  900                 d->bd_writer = 2;
  901         d->bd_hbuf_in_use = 0;
  902         d->bd_bufmode = BPF_BUFMODE_BUFFER;
  903         d->bd_sig = SIGIO;
  904         d->bd_direction = BPF_D_INOUT;
  905         BPF_PID_REFRESH(d, td);
  906 #ifdef MAC
  907         mac_bpfdesc_init(d);
  908         mac_bpfdesc_create(td->td_ucred, d);
  909 #endif
  910         mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
  911         callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
  912         knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
  913 
  914         return (0);
  915 }
  916 
  917 /*
  918  *  bpfread - read next chunk of packets from buffers
  919  */
  920 static  int
  921 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
  922 {
  923         struct bpf_d *d;
  924         int error;
  925         int non_block;
  926         int timed_out;
  927 
  928         error = devfs_get_cdevpriv((void **)&d);
  929         if (error != 0)
  930                 return (error);
  931 
  932         /*
  933          * Restrict application to use a buffer the same size as
  934          * as kernel buffers.
  935          */
  936         if (uio->uio_resid != d->bd_bufsize)
  937                 return (EINVAL);
  938 
  939         non_block = ((ioflag & O_NONBLOCK) != 0);
  940 
  941         BPFD_LOCK(d);
  942         BPF_PID_REFRESH_CUR(d);
  943         if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
  944                 BPFD_UNLOCK(d);
  945                 return (EOPNOTSUPP);
  946         }
  947         if (d->bd_state == BPF_WAITING)
  948                 callout_stop(&d->bd_callout);
  949         timed_out = (d->bd_state == BPF_TIMED_OUT);
  950         d->bd_state = BPF_IDLE;
  951         while (d->bd_hbuf_in_use) {
  952                 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
  953                     PRINET|PCATCH, "bd_hbuf", 0);
  954                 if (error != 0) {
  955                         BPFD_UNLOCK(d);
  956                         return (error);
  957                 }
  958         }
  959         /*
  960          * If the hold buffer is empty, then do a timed sleep, which
  961          * ends when the timeout expires or when enough packets
  962          * have arrived to fill the store buffer.
  963          */
  964         while (d->bd_hbuf == NULL) {
  965                 if (d->bd_slen != 0) {
  966                         /*
  967                          * A packet(s) either arrived since the previous
  968                          * read or arrived while we were asleep.
  969                          */
  970                         if (d->bd_immediate || non_block || timed_out) {
  971                                 /*
  972                                  * Rotate the buffers and return what's here
  973                                  * if we are in immediate mode, non-blocking
  974                                  * flag is set, or this descriptor timed out.
  975                                  */
  976                                 ROTATE_BUFFERS(d);
  977                                 break;
  978                         }
  979                 }
  980 
  981                 /*
  982                  * No data is available, check to see if the bpf device
  983                  * is still pointed at a real interface.  If not, return
  984                  * ENXIO so that the userland process knows to rebind
  985                  * it before using it again.
  986                  */
  987                 if (d->bd_bif == NULL) {
  988                         BPFD_UNLOCK(d);
  989                         return (ENXIO);
  990                 }
  991 
  992                 if (non_block) {
  993                         BPFD_UNLOCK(d);
  994                         return (EWOULDBLOCK);
  995                 }
  996                 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
  997                      "bpf", d->bd_rtout);
  998                 if (error == EINTR || error == ERESTART) {
  999                         BPFD_UNLOCK(d);
 1000                         return (error);
 1001                 }
 1002                 if (error == EWOULDBLOCK) {
 1003                         /*
 1004                          * On a timeout, return what's in the buffer,
 1005                          * which may be nothing.  If there is something
 1006                          * in the store buffer, we can rotate the buffers.
 1007                          */
 1008                         if (d->bd_hbuf)
 1009                                 /*
 1010                                  * We filled up the buffer in between
 1011                                  * getting the timeout and arriving
 1012                                  * here, so we don't need to rotate.
 1013                                  */
 1014                                 break;
 1015 
 1016                         if (d->bd_slen == 0) {
 1017                                 BPFD_UNLOCK(d);
 1018                                 return (0);
 1019                         }
 1020                         ROTATE_BUFFERS(d);
 1021                         break;
 1022                 }
 1023         }
 1024         /*
 1025          * At this point, we know we have something in the hold slot.
 1026          */
 1027         d->bd_hbuf_in_use = 1;
 1028         BPFD_UNLOCK(d);
 1029 
 1030         /*
 1031          * Move data from hold buffer into user space.
 1032          * We know the entire buffer is transferred since
 1033          * we checked above that the read buffer is bpf_bufsize bytes.
 1034          *
 1035          * We do not have to worry about simultaneous reads because
 1036          * we waited for sole access to the hold buffer above.
 1037          */
 1038         error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
 1039 
 1040         BPFD_LOCK(d);
 1041         KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
 1042         d->bd_fbuf = d->bd_hbuf;
 1043         d->bd_hbuf = NULL;
 1044         d->bd_hlen = 0;
 1045         bpf_buf_reclaimed(d);
 1046         d->bd_hbuf_in_use = 0;
 1047         wakeup(&d->bd_hbuf_in_use);
 1048         BPFD_UNLOCK(d);
 1049 
 1050         return (error);
 1051 }
 1052 
 1053 /*
 1054  * If there are processes sleeping on this descriptor, wake them up.
 1055  */
 1056 static __inline void
 1057 bpf_wakeup(struct bpf_d *d)
 1058 {
 1059 
 1060         BPFD_LOCK_ASSERT(d);
 1061         if (d->bd_state == BPF_WAITING) {
 1062                 callout_stop(&d->bd_callout);
 1063                 d->bd_state = BPF_IDLE;
 1064         }
 1065         wakeup(d);
 1066         if (d->bd_async && d->bd_sig && d->bd_sigio)
 1067                 pgsigio(&d->bd_sigio, d->bd_sig, 0);
 1068 
 1069         selwakeuppri(&d->bd_sel, PRINET);
 1070         KNOTE_LOCKED(&d->bd_sel.si_note, 0);
 1071 }
 1072 
 1073 static void
 1074 bpf_timed_out(void *arg)
 1075 {
 1076         struct bpf_d *d = (struct bpf_d *)arg;
 1077 
 1078         BPFD_LOCK_ASSERT(d);
 1079 
 1080         if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
 1081                 return;
 1082         if (d->bd_state == BPF_WAITING) {
 1083                 d->bd_state = BPF_TIMED_OUT;
 1084                 if (d->bd_slen != 0)
 1085                         bpf_wakeup(d);
 1086         }
 1087 }
 1088 
 1089 static int
 1090 bpf_ready(struct bpf_d *d)
 1091 {
 1092 
 1093         BPFD_LOCK_ASSERT(d);
 1094 
 1095         if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
 1096                 return (1);
 1097         if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
 1098             d->bd_slen != 0)
 1099                 return (1);
 1100         return (0);
 1101 }
 1102 
 1103 static int
 1104 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
 1105 {
 1106         struct bpf_d *d;
 1107         struct ifnet *ifp;
 1108         struct mbuf *m, *mc;
 1109         struct sockaddr dst;
 1110         struct route ro;
 1111         int error, hlen;
 1112 
 1113         error = devfs_get_cdevpriv((void **)&d);
 1114         if (error != 0)
 1115                 return (error);
 1116 
 1117         BPF_PID_REFRESH_CUR(d);
 1118         d->bd_wcount++;
 1119         /* XXX: locking required */
 1120         if (d->bd_bif == NULL) {
 1121                 d->bd_wdcount++;
 1122                 return (ENXIO);
 1123         }
 1124 
 1125         ifp = d->bd_bif->bif_ifp;
 1126 
 1127         if ((ifp->if_flags & IFF_UP) == 0) {
 1128                 d->bd_wdcount++;
 1129                 return (ENETDOWN);
 1130         }
 1131 
 1132         if (uio->uio_resid == 0) {
 1133                 d->bd_wdcount++;
 1134                 return (0);
 1135         }
 1136 
 1137         bzero(&dst, sizeof(dst));
 1138         m = NULL;
 1139         hlen = 0;
 1140         /* XXX: bpf_movein() can sleep */
 1141         error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
 1142             &m, &dst, &hlen, d);
 1143         if (error) {
 1144                 d->bd_wdcount++;
 1145                 return (error);
 1146         }
 1147         d->bd_wfcount++;
 1148         if (d->bd_hdrcmplt)
 1149                 dst.sa_family = pseudo_AF_HDRCMPLT;
 1150 
 1151         if (d->bd_feedback) {
 1152                 mc = m_dup(m, M_NOWAIT);
 1153                 if (mc != NULL)
 1154                         mc->m_pkthdr.rcvif = ifp;
 1155                 /* Set M_PROMISC for outgoing packets to be discarded. */
 1156                 if (d->bd_direction == BPF_D_INOUT)
 1157                         m->m_flags |= M_PROMISC;
 1158         } else
 1159                 mc = NULL;
 1160 
 1161         m->m_pkthdr.len -= hlen;
 1162         m->m_len -= hlen;
 1163         m->m_data += hlen;      /* XXX */
 1164 
 1165         CURVNET_SET(ifp->if_vnet);
 1166 #ifdef MAC
 1167         BPFD_LOCK(d);
 1168         mac_bpfdesc_create_mbuf(d, m);
 1169         if (mc != NULL)
 1170                 mac_bpfdesc_create_mbuf(d, mc);
 1171         BPFD_UNLOCK(d);
 1172 #endif
 1173 
 1174         bzero(&ro, sizeof(ro));
 1175         if (hlen != 0) {
 1176                 ro.ro_prepend = (u_char *)&dst.sa_data;
 1177                 ro.ro_plen = hlen;
 1178                 ro.ro_flags = RT_HAS_HEADER;
 1179         }
 1180 
 1181         error = (*ifp->if_output)(ifp, m, &dst, &ro);
 1182         if (error)
 1183                 d->bd_wdcount++;
 1184 
 1185         if (mc != NULL) {
 1186                 if (error == 0)
 1187                         (*ifp->if_input)(ifp, mc);
 1188                 else
 1189                         m_freem(mc);
 1190         }
 1191         CURVNET_RESTORE();
 1192 
 1193         return (error);
 1194 }
 1195 
 1196 /*
 1197  * Reset a descriptor by flushing its packet buffer and clearing the receive
 1198  * and drop counts.  This is doable for kernel-only buffers, but with
 1199  * zero-copy buffers, we can't write to (or rotate) buffers that are
 1200  * currently owned by userspace.  It would be nice if we could encapsulate
 1201  * this logic in the buffer code rather than here.
 1202  */
 1203 static void
 1204 reset_d(struct bpf_d *d)
 1205 {
 1206 
 1207         BPFD_LOCK_ASSERT(d);
 1208 
 1209         while (d->bd_hbuf_in_use)
 1210                 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
 1211                     "bd_hbuf", 0);
 1212         if ((d->bd_hbuf != NULL) &&
 1213             (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
 1214                 /* Free the hold buffer. */
 1215                 d->bd_fbuf = d->bd_hbuf;
 1216                 d->bd_hbuf = NULL;
 1217                 d->bd_hlen = 0;
 1218                 bpf_buf_reclaimed(d);
 1219         }
 1220         if (bpf_canwritebuf(d))
 1221                 d->bd_slen = 0;
 1222         d->bd_rcount = 0;
 1223         d->bd_dcount = 0;
 1224         d->bd_fcount = 0;
 1225         d->bd_wcount = 0;
 1226         d->bd_wfcount = 0;
 1227         d->bd_wdcount = 0;
 1228         d->bd_zcopy = 0;
 1229 }
 1230 
 1231 /*
 1232  *  FIONREAD            Check for read packet available.
 1233  *  BIOCGBLEN           Get buffer len [for read()].
 1234  *  BIOCSETF            Set read filter.
 1235  *  BIOCSETFNR          Set read filter without resetting descriptor.
 1236  *  BIOCSETWF           Set write filter.
 1237  *  BIOCFLUSH           Flush read packet buffer.
 1238  *  BIOCPROMISC         Put interface into promiscuous mode.
 1239  *  BIOCGDLT            Get link layer type.
 1240  *  BIOCGETIF           Get interface name.
 1241  *  BIOCSETIF           Set interface.
 1242  *  BIOCSRTIMEOUT       Set read timeout.
 1243  *  BIOCGRTIMEOUT       Get read timeout.
 1244  *  BIOCGSTATS          Get packet stats.
 1245  *  BIOCIMMEDIATE       Set immediate mode.
 1246  *  BIOCVERSION         Get filter language version.
 1247  *  BIOCGHDRCMPLT       Get "header already complete" flag
 1248  *  BIOCSHDRCMPLT       Set "header already complete" flag
 1249  *  BIOCGDIRECTION      Get packet direction flag
 1250  *  BIOCSDIRECTION      Set packet direction flag
 1251  *  BIOCGTSTAMP         Get time stamp format and resolution.
 1252  *  BIOCSTSTAMP         Set time stamp format and resolution.
 1253  *  BIOCLOCK            Set "locked" flag
 1254  *  BIOCFEEDBACK        Set packet feedback mode.
 1255  *  BIOCSETZBUF         Set current zero-copy buffer locations.
 1256  *  BIOCGETZMAX         Get maximum zero-copy buffer size.
 1257  *  BIOCROTZBUF         Force rotation of zero-copy buffer
 1258  *  BIOCSETBUFMODE      Set buffer mode.
 1259  *  BIOCGETBUFMODE      Get current buffer mode.
 1260  */
 1261 /* ARGSUSED */
 1262 static  int
 1263 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
 1264     struct thread *td)
 1265 {
 1266         struct bpf_d *d;
 1267         int error;
 1268 
 1269         error = devfs_get_cdevpriv((void **)&d);
 1270         if (error != 0)
 1271                 return (error);
 1272 
 1273         /*
 1274          * Refresh PID associated with this descriptor.
 1275          */
 1276         BPFD_LOCK(d);
 1277         BPF_PID_REFRESH(d, td);
 1278         if (d->bd_state == BPF_WAITING)
 1279                 callout_stop(&d->bd_callout);
 1280         d->bd_state = BPF_IDLE;
 1281         BPFD_UNLOCK(d);
 1282 
 1283         if (d->bd_locked == 1) {
 1284                 switch (cmd) {
 1285                 case BIOCGBLEN:
 1286                 case BIOCFLUSH:
 1287                 case BIOCGDLT:
 1288                 case BIOCGDLTLIST:
 1289 #ifdef COMPAT_FREEBSD32
 1290                 case BIOCGDLTLIST32:
 1291 #endif
 1292                 case BIOCGETIF:
 1293                 case BIOCGRTIMEOUT:
 1294 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1295                 case BIOCGRTIMEOUT32:
 1296 #endif
 1297                 case BIOCGSTATS:
 1298                 case BIOCVERSION:
 1299                 case BIOCGRSIG:
 1300                 case BIOCGHDRCMPLT:
 1301                 case BIOCSTSTAMP:
 1302                 case BIOCFEEDBACK:
 1303                 case FIONREAD:
 1304                 case BIOCLOCK:
 1305                 case BIOCSRTIMEOUT:
 1306 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1307                 case BIOCSRTIMEOUT32:
 1308 #endif
 1309                 case BIOCIMMEDIATE:
 1310                 case TIOCGPGRP:
 1311                 case BIOCROTZBUF:
 1312                         break;
 1313                 default:
 1314                         return (EPERM);
 1315                 }
 1316         }
 1317 #ifdef COMPAT_FREEBSD32
 1318         /*
 1319          * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
 1320          * that it will get 32-bit packet headers.
 1321          */
 1322         switch (cmd) {
 1323         case BIOCSETF32:
 1324         case BIOCSETFNR32:
 1325         case BIOCSETWF32:
 1326         case BIOCGDLTLIST32:
 1327         case BIOCGRTIMEOUT32:
 1328         case BIOCSRTIMEOUT32:
 1329                 BPFD_LOCK(d);
 1330                 d->bd_compat32 = 1;
 1331                 BPFD_UNLOCK(d);
 1332         }
 1333 #endif
 1334 
 1335         CURVNET_SET(TD_TO_VNET(td));
 1336         switch (cmd) {
 1337 
 1338         default:
 1339                 error = EINVAL;
 1340                 break;
 1341 
 1342         /*
 1343          * Check for read packet available.
 1344          */
 1345         case FIONREAD:
 1346                 {
 1347                         int n;
 1348 
 1349                         BPFD_LOCK(d);
 1350                         n = d->bd_slen;
 1351                         while (d->bd_hbuf_in_use)
 1352                                 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
 1353                                     PRINET, "bd_hbuf", 0);
 1354                         if (d->bd_hbuf)
 1355                                 n += d->bd_hlen;
 1356                         BPFD_UNLOCK(d);
 1357 
 1358                         *(int *)addr = n;
 1359                         break;
 1360                 }
 1361 
 1362         /*
 1363          * Get buffer len [for read()].
 1364          */
 1365         case BIOCGBLEN:
 1366                 BPFD_LOCK(d);
 1367                 *(u_int *)addr = d->bd_bufsize;
 1368                 BPFD_UNLOCK(d);
 1369                 break;
 1370 
 1371         /*
 1372          * Set buffer length.
 1373          */
 1374         case BIOCSBLEN:
 1375                 error = bpf_ioctl_sblen(d, (u_int *)addr);
 1376                 break;
 1377 
 1378         /*
 1379          * Set link layer read filter.
 1380          */
 1381         case BIOCSETF:
 1382         case BIOCSETFNR:
 1383         case BIOCSETWF:
 1384 #ifdef COMPAT_FREEBSD32
 1385         case BIOCSETF32:
 1386         case BIOCSETFNR32:
 1387         case BIOCSETWF32:
 1388 #endif
 1389                 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
 1390                 break;
 1391 
 1392         /*
 1393          * Flush read packet buffer.
 1394          */
 1395         case BIOCFLUSH:
 1396                 BPFD_LOCK(d);
 1397                 reset_d(d);
 1398                 BPFD_UNLOCK(d);
 1399                 break;
 1400 
 1401         /*
 1402          * Put interface into promiscuous mode.
 1403          */
 1404         case BIOCPROMISC:
 1405                 if (d->bd_bif == NULL) {
 1406                         /*
 1407                          * No interface attached yet.
 1408                          */
 1409                         error = EINVAL;
 1410                         break;
 1411                 }
 1412                 if (d->bd_promisc == 0) {
 1413                         error = ifpromisc(d->bd_bif->bif_ifp, 1);
 1414                         if (error == 0)
 1415                                 d->bd_promisc = 1;
 1416                 }
 1417                 break;
 1418 
 1419         /*
 1420          * Get current data link type.
 1421          */
 1422         case BIOCGDLT:
 1423                 BPF_LOCK();
 1424                 if (d->bd_bif == NULL)
 1425                         error = EINVAL;
 1426                 else
 1427                         *(u_int *)addr = d->bd_bif->bif_dlt;
 1428                 BPF_UNLOCK();
 1429                 break;
 1430 
 1431         /*
 1432          * Get a list of supported data link types.
 1433          */
 1434 #ifdef COMPAT_FREEBSD32
 1435         case BIOCGDLTLIST32:
 1436                 {
 1437                         struct bpf_dltlist32 *list32;
 1438                         struct bpf_dltlist dltlist;
 1439 
 1440                         list32 = (struct bpf_dltlist32 *)addr;
 1441                         dltlist.bfl_len = list32->bfl_len;
 1442                         dltlist.bfl_list = PTRIN(list32->bfl_list);
 1443                         BPF_LOCK();
 1444                         if (d->bd_bif == NULL)
 1445                                 error = EINVAL;
 1446                         else {
 1447                                 error = bpf_getdltlist(d, &dltlist);
 1448                                 if (error == 0)
 1449                                         list32->bfl_len = dltlist.bfl_len;
 1450                         }
 1451                         BPF_UNLOCK();
 1452                         break;
 1453                 }
 1454 #endif
 1455 
 1456         case BIOCGDLTLIST:
 1457                 BPF_LOCK();
 1458                 if (d->bd_bif == NULL)
 1459                         error = EINVAL;
 1460                 else
 1461                         error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
 1462                 BPF_UNLOCK();
 1463                 break;
 1464 
 1465         /*
 1466          * Set data link type.
 1467          */
 1468         case BIOCSDLT:
 1469                 BPF_LOCK();
 1470                 if (d->bd_bif == NULL)
 1471                         error = EINVAL;
 1472                 else
 1473                         error = bpf_setdlt(d, *(u_int *)addr);
 1474                 BPF_UNLOCK();
 1475                 break;
 1476 
 1477         /*
 1478          * Get interface name.
 1479          */
 1480         case BIOCGETIF:
 1481                 BPF_LOCK();
 1482                 if (d->bd_bif == NULL)
 1483                         error = EINVAL;
 1484                 else {
 1485                         struct ifnet *const ifp = d->bd_bif->bif_ifp;
 1486                         struct ifreq *const ifr = (struct ifreq *)addr;
 1487 
 1488                         strlcpy(ifr->ifr_name, ifp->if_xname,
 1489                             sizeof(ifr->ifr_name));
 1490                 }
 1491                 BPF_UNLOCK();
 1492                 break;
 1493 
 1494         /*
 1495          * Set interface.
 1496          */
 1497         case BIOCSETIF:
 1498                 {
 1499                         int alloc_buf, size;
 1500 
 1501                         /*
 1502                          * Behavior here depends on the buffering model.  If
 1503                          * we're using kernel memory buffers, then we can
 1504                          * allocate them here.  If we're using zero-copy,
 1505                          * then the user process must have registered buffers
 1506                          * by the time we get here.
 1507                          */
 1508                         alloc_buf = 0;
 1509                         BPFD_LOCK(d);
 1510                         if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
 1511                             d->bd_sbuf == NULL)
 1512                                 alloc_buf = 1;
 1513                         BPFD_UNLOCK(d);
 1514                         if (alloc_buf) {
 1515                                 size = d->bd_bufsize;
 1516                                 error = bpf_buffer_ioctl_sblen(d, &size);
 1517                                 if (error != 0)
 1518                                         break;
 1519                         }
 1520                         BPF_LOCK();
 1521                         error = bpf_setif(d, (struct ifreq *)addr);
 1522                         BPF_UNLOCK();
 1523                         break;
 1524                 }
 1525 
 1526         /*
 1527          * Set read timeout.
 1528          */
 1529         case BIOCSRTIMEOUT:
 1530 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1531         case BIOCSRTIMEOUT32:
 1532 #endif
 1533                 {
 1534                         struct timeval *tv = (struct timeval *)addr;
 1535 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1536                         struct timeval32 *tv32;
 1537                         struct timeval tv64;
 1538 
 1539                         if (cmd == BIOCSRTIMEOUT32) {
 1540                                 tv32 = (struct timeval32 *)addr;
 1541                                 tv = &tv64;
 1542                                 tv->tv_sec = tv32->tv_sec;
 1543                                 tv->tv_usec = tv32->tv_usec;
 1544                         } else
 1545 #endif
 1546                                 tv = (struct timeval *)addr;
 1547 
 1548                         /*
 1549                          * Subtract 1 tick from tvtohz() since this isn't
 1550                          * a one-shot timer.
 1551                          */
 1552                         if ((error = itimerfix(tv)) == 0)
 1553                                 d->bd_rtout = tvtohz(tv) - 1;
 1554                         break;
 1555                 }
 1556 
 1557         /*
 1558          * Get read timeout.
 1559          */
 1560         case BIOCGRTIMEOUT:
 1561 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1562         case BIOCGRTIMEOUT32:
 1563 #endif
 1564                 {
 1565                         struct timeval *tv;
 1566 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1567                         struct timeval32 *tv32;
 1568                         struct timeval tv64;
 1569 
 1570                         if (cmd == BIOCGRTIMEOUT32)
 1571                                 tv = &tv64;
 1572                         else
 1573 #endif
 1574                                 tv = (struct timeval *)addr;
 1575 
 1576                         tv->tv_sec = d->bd_rtout / hz;
 1577                         tv->tv_usec = (d->bd_rtout % hz) * tick;
 1578 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
 1579                         if (cmd == BIOCGRTIMEOUT32) {
 1580                                 tv32 = (struct timeval32 *)addr;
 1581                                 tv32->tv_sec = tv->tv_sec;
 1582                                 tv32->tv_usec = tv->tv_usec;
 1583                         }
 1584 #endif
 1585 
 1586                         break;
 1587                 }
 1588 
 1589         /*
 1590          * Get packet stats.
 1591          */
 1592         case BIOCGSTATS:
 1593                 {
 1594                         struct bpf_stat *bs = (struct bpf_stat *)addr;
 1595 
 1596                         /* XXXCSJP overflow */
 1597                         bs->bs_recv = d->bd_rcount;
 1598                         bs->bs_drop = d->bd_dcount;
 1599                         break;
 1600                 }
 1601 
 1602         /*
 1603          * Set immediate mode.
 1604          */
 1605         case BIOCIMMEDIATE:
 1606                 BPFD_LOCK(d);
 1607                 d->bd_immediate = *(u_int *)addr;
 1608                 BPFD_UNLOCK(d);
 1609                 break;
 1610 
 1611         case BIOCVERSION:
 1612                 {
 1613                         struct bpf_version *bv = (struct bpf_version *)addr;
 1614 
 1615                         bv->bv_major = BPF_MAJOR_VERSION;
 1616                         bv->bv_minor = BPF_MINOR_VERSION;
 1617                         break;
 1618                 }
 1619 
 1620         /*
 1621          * Get "header already complete" flag
 1622          */
 1623         case BIOCGHDRCMPLT:
 1624                 BPFD_LOCK(d);
 1625                 *(u_int *)addr = d->bd_hdrcmplt;
 1626                 BPFD_UNLOCK(d);
 1627                 break;
 1628 
 1629         /*
 1630          * Set "header already complete" flag
 1631          */
 1632         case BIOCSHDRCMPLT:
 1633                 BPFD_LOCK(d);
 1634                 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
 1635                 BPFD_UNLOCK(d);
 1636                 break;
 1637 
 1638         /*
 1639          * Get packet direction flag
 1640          */
 1641         case BIOCGDIRECTION:
 1642                 BPFD_LOCK(d);
 1643                 *(u_int *)addr = d->bd_direction;
 1644                 BPFD_UNLOCK(d);
 1645                 break;
 1646 
 1647         /*
 1648          * Set packet direction flag
 1649          */
 1650         case BIOCSDIRECTION:
 1651                 {
 1652                         u_int   direction;
 1653 
 1654                         direction = *(u_int *)addr;
 1655                         switch (direction) {
 1656                         case BPF_D_IN:
 1657                         case BPF_D_INOUT:
 1658                         case BPF_D_OUT:
 1659                                 BPFD_LOCK(d);
 1660                                 d->bd_direction = direction;
 1661                                 BPFD_UNLOCK(d);
 1662                                 break;
 1663                         default:
 1664                                 error = EINVAL;
 1665                         }
 1666                 }
 1667                 break;
 1668 
 1669         /*
 1670          * Get packet timestamp format and resolution.
 1671          */
 1672         case BIOCGTSTAMP:
 1673                 BPFD_LOCK(d);
 1674                 *(u_int *)addr = d->bd_tstamp;
 1675                 BPFD_UNLOCK(d);
 1676                 break;
 1677 
 1678         /*
 1679          * Set packet timestamp format and resolution.
 1680          */
 1681         case BIOCSTSTAMP:
 1682                 {
 1683                         u_int   func;
 1684 
 1685                         func = *(u_int *)addr;
 1686                         if (BPF_T_VALID(func))
 1687                                 d->bd_tstamp = func;
 1688                         else
 1689                                 error = EINVAL;
 1690                 }
 1691                 break;
 1692 
 1693         case BIOCFEEDBACK:
 1694                 BPFD_LOCK(d);
 1695                 d->bd_feedback = *(u_int *)addr;
 1696                 BPFD_UNLOCK(d);
 1697                 break;
 1698 
 1699         case BIOCLOCK:
 1700                 BPFD_LOCK(d);
 1701                 d->bd_locked = 1;
 1702                 BPFD_UNLOCK(d);
 1703                 break;
 1704 
 1705         case FIONBIO:           /* Non-blocking I/O */
 1706                 break;
 1707 
 1708         case FIOASYNC:          /* Send signal on receive packets */
 1709                 BPFD_LOCK(d);
 1710                 d->bd_async = *(int *)addr;
 1711                 BPFD_UNLOCK(d);
 1712                 break;
 1713 
 1714         case FIOSETOWN:
 1715                 /*
 1716                  * XXX: Add some sort of locking here?
 1717                  * fsetown() can sleep.
 1718                  */
 1719                 error = fsetown(*(int *)addr, &d->bd_sigio);
 1720                 break;
 1721 
 1722         case FIOGETOWN:
 1723                 BPFD_LOCK(d);
 1724                 *(int *)addr = fgetown(&d->bd_sigio);
 1725                 BPFD_UNLOCK(d);
 1726                 break;
 1727 
 1728         /* This is deprecated, FIOSETOWN should be used instead. */
 1729         case TIOCSPGRP:
 1730                 error = fsetown(-(*(int *)addr), &d->bd_sigio);
 1731                 break;
 1732 
 1733         /* This is deprecated, FIOGETOWN should be used instead. */
 1734         case TIOCGPGRP:
 1735                 *(int *)addr = -fgetown(&d->bd_sigio);
 1736                 break;
 1737 
 1738         case BIOCSRSIG:         /* Set receive signal */
 1739                 {
 1740                         u_int sig;
 1741 
 1742                         sig = *(u_int *)addr;
 1743 
 1744                         if (sig >= NSIG)
 1745                                 error = EINVAL;
 1746                         else {
 1747                                 BPFD_LOCK(d);
 1748                                 d->bd_sig = sig;
 1749                                 BPFD_UNLOCK(d);
 1750                         }
 1751                         break;
 1752                 }
 1753         case BIOCGRSIG:
 1754                 BPFD_LOCK(d);
 1755                 *(u_int *)addr = d->bd_sig;
 1756                 BPFD_UNLOCK(d);
 1757                 break;
 1758 
 1759         case BIOCGETBUFMODE:
 1760                 BPFD_LOCK(d);
 1761                 *(u_int *)addr = d->bd_bufmode;
 1762                 BPFD_UNLOCK(d);
 1763                 break;
 1764 
 1765         case BIOCSETBUFMODE:
 1766                 /*
 1767                  * Allow the buffering mode to be changed as long as we
 1768                  * haven't yet committed to a particular mode.  Our
 1769                  * definition of commitment, for now, is whether or not a
 1770                  * buffer has been allocated or an interface attached, since
 1771                  * that's the point where things get tricky.
 1772                  */
 1773                 switch (*(u_int *)addr) {
 1774                 case BPF_BUFMODE_BUFFER:
 1775                         break;
 1776 
 1777                 case BPF_BUFMODE_ZBUF:
 1778                         if (bpf_zerocopy_enable)
 1779                                 break;
 1780                         /* FALLSTHROUGH */
 1781 
 1782                 default:
 1783                         CURVNET_RESTORE();
 1784                         return (EINVAL);
 1785                 }
 1786 
 1787                 BPFD_LOCK(d);
 1788                 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
 1789                     d->bd_fbuf != NULL || d->bd_bif != NULL) {
 1790                         BPFD_UNLOCK(d);
 1791                         CURVNET_RESTORE();
 1792                         return (EBUSY);
 1793                 }
 1794                 d->bd_bufmode = *(u_int *)addr;
 1795                 BPFD_UNLOCK(d);
 1796                 break;
 1797 
 1798         case BIOCGETZMAX:
 1799                 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
 1800                 break;
 1801 
 1802         case BIOCSETZBUF:
 1803                 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
 1804                 break;
 1805 
 1806         case BIOCROTZBUF:
 1807                 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
 1808                 break;
 1809         }
 1810         CURVNET_RESTORE();
 1811         return (error);
 1812 }
 1813 
 1814 /*
 1815  * Set d's packet filter program to fp.  If this file already has a filter,
 1816  * free it and replace it.  Returns EINVAL for bogus requests.
 1817  *
 1818  * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
 1819  * since reading d->bd_bif can't be protected by d or interface lock due to
 1820  * lock order.
 1821  *
 1822  * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
 1823  * interface read lock to read all filers.
 1824  *
 1825  */
 1826 static int
 1827 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
 1828 {
 1829 #ifdef COMPAT_FREEBSD32
 1830         struct bpf_program fp_swab;
 1831         struct bpf_program32 *fp32;
 1832 #endif
 1833         struct bpf_insn *fcode, *old;
 1834 #ifdef BPF_JITTER
 1835         bpf_jit_filter *jfunc, *ofunc;
 1836 #endif
 1837         size_t size;
 1838         u_int flen;
 1839         int need_upgrade;
 1840 
 1841 #ifdef COMPAT_FREEBSD32
 1842         switch (cmd) {
 1843         case BIOCSETF32:
 1844         case BIOCSETWF32:
 1845         case BIOCSETFNR32:
 1846                 fp32 = (struct bpf_program32 *)fp;
 1847                 fp_swab.bf_len = fp32->bf_len;
 1848                 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
 1849                 fp = &fp_swab;
 1850                 switch (cmd) {
 1851                 case BIOCSETF32:
 1852                         cmd = BIOCSETF;
 1853                         break;
 1854                 case BIOCSETWF32:
 1855                         cmd = BIOCSETWF;
 1856                         break;
 1857                 }
 1858                 break;
 1859         }
 1860 #endif
 1861 
 1862         fcode = NULL;
 1863 #ifdef BPF_JITTER
 1864         jfunc = ofunc = NULL;
 1865 #endif
 1866         need_upgrade = 0;
 1867 
 1868         /*
 1869          * Check new filter validness before acquiring any locks.
 1870          * Allocate memory for new filter, if needed.
 1871          */
 1872         flen = fp->bf_len;
 1873         if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
 1874                 return (EINVAL);
 1875         size = flen * sizeof(*fp->bf_insns);
 1876         if (size > 0) {
 1877                 /* We're setting up new filter.  Copy and check actual data. */
 1878                 fcode = malloc(size, M_BPF, M_WAITOK);
 1879                 if (copyin(fp->bf_insns, fcode, size) != 0 ||
 1880                     !bpf_validate(fcode, flen)) {
 1881                         free(fcode, M_BPF);
 1882                         return (EINVAL);
 1883                 }
 1884 #ifdef BPF_JITTER
 1885                 /* Filter is copied inside fcode and is perfectly valid. */
 1886                 jfunc = bpf_jitter(fcode, flen);
 1887 #endif
 1888         }
 1889 
 1890         BPF_LOCK();
 1891 
 1892         /*
 1893          * Set up new filter.
 1894          * Protect filter change by interface lock.
 1895          * Additionally, we are protected by global lock here.
 1896          */
 1897         if (d->bd_bif != NULL)
 1898                 BPFIF_WLOCK(d->bd_bif);
 1899         BPFD_LOCK(d);
 1900         if (cmd == BIOCSETWF) {
 1901                 old = d->bd_wfilter;
 1902                 d->bd_wfilter = fcode;
 1903         } else {
 1904                 old = d->bd_rfilter;
 1905                 d->bd_rfilter = fcode;
 1906 #ifdef BPF_JITTER
 1907                 ofunc = d->bd_bfilter;
 1908                 d->bd_bfilter = jfunc;
 1909 #endif
 1910                 if (cmd == BIOCSETF)
 1911                         reset_d(d);
 1912 
 1913                 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
 1914         }
 1915         BPFD_UNLOCK(d);
 1916         if (d->bd_bif != NULL)
 1917                 BPFIF_WUNLOCK(d->bd_bif);
 1918         if (old != NULL)
 1919                 free(old, M_BPF);
 1920 #ifdef BPF_JITTER
 1921         if (ofunc != NULL)
 1922                 bpf_destroy_jit_filter(ofunc);
 1923 #endif
 1924 
 1925         /* Move d to active readers list. */
 1926         if (need_upgrade != 0)
 1927                 bpf_upgraded(d);
 1928 
 1929         BPF_UNLOCK();
 1930         return (0);
 1931 }
 1932 
 1933 /*
 1934  * Detach a file from its current interface (if attached at all) and attach
 1935  * to the interface indicated by the name stored in ifr.
 1936  * Return an errno or 0.
 1937  */
 1938 static int
 1939 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
 1940 {
 1941         struct bpf_if *bp;
 1942         struct ifnet *theywant;
 1943 
 1944         BPF_LOCK_ASSERT();
 1945 
 1946         theywant = ifunit(ifr->ifr_name);
 1947         if (theywant == NULL || theywant->if_bpf == NULL)
 1948                 return (ENXIO);
 1949 
 1950         bp = theywant->if_bpf;
 1951 
 1952         /* Check if interface is not being detached from BPF */
 1953         BPFIF_RLOCK(bp);
 1954         if (bp->bif_flags & BPFIF_FLAG_DYING) {
 1955                 BPFIF_RUNLOCK(bp);
 1956                 return (ENXIO);
 1957         }
 1958         BPFIF_RUNLOCK(bp);
 1959 
 1960         /*
 1961          * At this point, we expect the buffer is already allocated.  If not,
 1962          * return an error.
 1963          */
 1964         switch (d->bd_bufmode) {
 1965         case BPF_BUFMODE_BUFFER:
 1966         case BPF_BUFMODE_ZBUF:
 1967                 if (d->bd_sbuf == NULL)
 1968                         return (EINVAL);
 1969                 break;
 1970 
 1971         default:
 1972                 panic("bpf_setif: bufmode %d", d->bd_bufmode);
 1973         }
 1974         if (bp != d->bd_bif)
 1975                 bpf_attachd(d, bp);
 1976         BPFD_LOCK(d);
 1977         reset_d(d);
 1978         BPFD_UNLOCK(d);
 1979         return (0);
 1980 }
 1981 
 1982 /*
 1983  * Support for select() and poll() system calls
 1984  *
 1985  * Return true iff the specific operation will not block indefinitely.
 1986  * Otherwise, return false but make a note that a selwakeup() must be done.
 1987  */
 1988 static int
 1989 bpfpoll(struct cdev *dev, int events, struct thread *td)
 1990 {
 1991         struct bpf_d *d;
 1992         int revents;
 1993 
 1994         if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
 1995                 return (events &
 1996                     (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
 1997 
 1998         /*
 1999          * Refresh PID associated with this descriptor.
 2000          */
 2001         revents = events & (POLLOUT | POLLWRNORM);
 2002         BPFD_LOCK(d);
 2003         BPF_PID_REFRESH(d, td);
 2004         if (events & (POLLIN | POLLRDNORM)) {
 2005                 if (bpf_ready(d))
 2006                         revents |= events & (POLLIN | POLLRDNORM);
 2007                 else {
 2008                         selrecord(td, &d->bd_sel);
 2009                         /* Start the read timeout if necessary. */
 2010                         if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
 2011                                 callout_reset(&d->bd_callout, d->bd_rtout,
 2012                                     bpf_timed_out, d);
 2013                                 d->bd_state = BPF_WAITING;
 2014                         }
 2015                 }
 2016         }
 2017         BPFD_UNLOCK(d);
 2018         return (revents);
 2019 }
 2020 
 2021 /*
 2022  * Support for kevent() system call.  Register EVFILT_READ filters and
 2023  * reject all others.
 2024  */
 2025 int
 2026 bpfkqfilter(struct cdev *dev, struct knote *kn)
 2027 {
 2028         struct bpf_d *d;
 2029 
 2030         if (devfs_get_cdevpriv((void **)&d) != 0 ||
 2031             kn->kn_filter != EVFILT_READ)
 2032                 return (1);
 2033 
 2034         /*
 2035          * Refresh PID associated with this descriptor.
 2036          */
 2037         BPFD_LOCK(d);
 2038         BPF_PID_REFRESH_CUR(d);
 2039         kn->kn_fop = &bpfread_filtops;
 2040         kn->kn_hook = d;
 2041         knlist_add(&d->bd_sel.si_note, kn, 1);
 2042         BPFD_UNLOCK(d);
 2043 
 2044         return (0);
 2045 }
 2046 
 2047 static void
 2048 filt_bpfdetach(struct knote *kn)
 2049 {
 2050         struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
 2051 
 2052         knlist_remove(&d->bd_sel.si_note, kn, 0);
 2053 }
 2054 
 2055 static int
 2056 filt_bpfread(struct knote *kn, long hint)
 2057 {
 2058         struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
 2059         int ready;
 2060 
 2061         BPFD_LOCK_ASSERT(d);
 2062         ready = bpf_ready(d);
 2063         if (ready) {
 2064                 kn->kn_data = d->bd_slen;
 2065                 /*
 2066                  * Ignore the hold buffer if it is being copied to user space.
 2067                  */
 2068                 if (!d->bd_hbuf_in_use && d->bd_hbuf)
 2069                         kn->kn_data += d->bd_hlen;
 2070         } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
 2071                 callout_reset(&d->bd_callout, d->bd_rtout,
 2072                     bpf_timed_out, d);
 2073                 d->bd_state = BPF_WAITING;
 2074         }
 2075 
 2076         return (ready);
 2077 }
 2078 
 2079 #define BPF_TSTAMP_NONE         0
 2080 #define BPF_TSTAMP_FAST         1
 2081 #define BPF_TSTAMP_NORMAL       2
 2082 #define BPF_TSTAMP_EXTERN       3
 2083 
 2084 static int
 2085 bpf_ts_quality(int tstype)
 2086 {
 2087 
 2088         if (tstype == BPF_T_NONE)
 2089                 return (BPF_TSTAMP_NONE);
 2090         if ((tstype & BPF_T_FAST) != 0)
 2091                 return (BPF_TSTAMP_FAST);
 2092 
 2093         return (BPF_TSTAMP_NORMAL);
 2094 }
 2095 
 2096 static int
 2097 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
 2098 {
 2099         struct m_tag *tag;
 2100         int quality;
 2101 
 2102         quality = bpf_ts_quality(tstype);
 2103         if (quality == BPF_TSTAMP_NONE)
 2104                 return (quality);
 2105 
 2106         if (m != NULL) {
 2107                 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
 2108                 if (tag != NULL) {
 2109                         *bt = *(struct bintime *)(tag + 1);
 2110                         return (BPF_TSTAMP_EXTERN);
 2111                 }
 2112         }
 2113         if (quality == BPF_TSTAMP_NORMAL)
 2114                 binuptime(bt);
 2115         else
 2116                 getbinuptime(bt);
 2117 
 2118         return (quality);
 2119 }
 2120 
 2121 /*
 2122  * Incoming linkage from device drivers.  Process the packet pkt, of length
 2123  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
 2124  * by each process' filter, and if accepted, stashed into the corresponding
 2125  * buffer.
 2126  */
 2127 void
 2128 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
 2129 {
 2130         struct bintime bt;
 2131         struct bpf_d *d;
 2132 #ifdef BPF_JITTER
 2133         bpf_jit_filter *bf;
 2134 #endif
 2135         u_int slen;
 2136         int gottime;
 2137 
 2138         gottime = BPF_TSTAMP_NONE;
 2139 
 2140         BPFIF_RLOCK(bp);
 2141 
 2142         LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
 2143                 /*
 2144                  * We are not using any locks for d here because:
 2145                  * 1) any filter change is protected by interface
 2146                  * write lock
 2147                  * 2) destroying/detaching d is protected by interface
 2148                  * write lock, too
 2149                  */
 2150 
 2151                 /* XXX: Do not protect counter for the sake of performance. */
 2152                 ++d->bd_rcount;
 2153                 /*
 2154                  * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
 2155                  * way for the caller to indiciate to us whether this packet
 2156                  * is inbound or outbound.  In the bpf_mtap() routines, we use
 2157                  * the interface pointers on the mbuf to figure it out.
 2158                  */
 2159 #ifdef BPF_JITTER
 2160                 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
 2161                 if (bf != NULL)
 2162                         slen = (*(bf->func))(pkt, pktlen, pktlen);
 2163                 else
 2164 #endif
 2165                 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
 2166                 if (slen != 0) {
 2167                         /*
 2168                          * Filter matches. Let's to acquire write lock.
 2169                          */
 2170                         BPFD_LOCK(d);
 2171 
 2172                         d->bd_fcount++;
 2173                         if (gottime < bpf_ts_quality(d->bd_tstamp))
 2174                                 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
 2175 #ifdef MAC
 2176                         if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
 2177 #endif
 2178                                 catchpacket(d, pkt, pktlen, slen,
 2179                                     bpf_append_bytes, &bt);
 2180                         BPFD_UNLOCK(d);
 2181                 }
 2182         }
 2183         BPFIF_RUNLOCK(bp);
 2184 }
 2185 
 2186 #define BPF_CHECK_DIRECTION(d, r, i)                            \
 2187             (((d)->bd_direction == BPF_D_IN && (r) != (i)) ||   \
 2188             ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
 2189 
 2190 /*
 2191  * Incoming linkage from device drivers, when packet is in an mbuf chain.
 2192  * Locking model is explained in bpf_tap().
 2193  */
 2194 void
 2195 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
 2196 {
 2197         struct bintime bt;
 2198         struct bpf_d *d;
 2199 #ifdef BPF_JITTER
 2200         bpf_jit_filter *bf;
 2201 #endif
 2202         u_int pktlen, slen;
 2203         int gottime;
 2204 
 2205         /* Skip outgoing duplicate packets. */
 2206         if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
 2207                 m->m_flags &= ~M_PROMISC;
 2208                 return;
 2209         }
 2210 
 2211         pktlen = m_length(m, NULL);
 2212         gottime = BPF_TSTAMP_NONE;
 2213 
 2214         BPFIF_RLOCK(bp);
 2215 
 2216         LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
 2217                 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
 2218                         continue;
 2219                 ++d->bd_rcount;
 2220 #ifdef BPF_JITTER
 2221                 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
 2222                 /* XXX We cannot handle multiple mbufs. */
 2223                 if (bf != NULL && m->m_next == NULL)
 2224                         slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
 2225                 else
 2226 #endif
 2227                 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
 2228                 if (slen != 0) {
 2229                         BPFD_LOCK(d);
 2230 
 2231                         d->bd_fcount++;
 2232                         if (gottime < bpf_ts_quality(d->bd_tstamp))
 2233                                 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
 2234 #ifdef MAC
 2235                         if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
 2236 #endif
 2237                                 catchpacket(d, (u_char *)m, pktlen, slen,
 2238                                     bpf_append_mbuf, &bt);
 2239                         BPFD_UNLOCK(d);
 2240                 }
 2241         }
 2242         BPFIF_RUNLOCK(bp);
 2243 }
 2244 
 2245 /*
 2246  * Incoming linkage from device drivers, when packet is in
 2247  * an mbuf chain and to be prepended by a contiguous header.
 2248  */
 2249 void
 2250 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
 2251 {
 2252         struct bintime bt;
 2253         struct mbuf mb;
 2254         struct bpf_d *d;
 2255         u_int pktlen, slen;
 2256         int gottime;
 2257 
 2258         /* Skip outgoing duplicate packets. */
 2259         if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
 2260                 m->m_flags &= ~M_PROMISC;
 2261                 return;
 2262         }
 2263 
 2264         pktlen = m_length(m, NULL);
 2265         /*
 2266          * Craft on-stack mbuf suitable for passing to bpf_filter.
 2267          * Note that we cut corners here; we only setup what's
 2268          * absolutely needed--this mbuf should never go anywhere else.
 2269          */
 2270         mb.m_next = m;
 2271         mb.m_data = data;
 2272         mb.m_len = dlen;
 2273         pktlen += dlen;
 2274 
 2275         gottime = BPF_TSTAMP_NONE;
 2276 
 2277         BPFIF_RLOCK(bp);
 2278 
 2279         LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
 2280                 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
 2281                         continue;
 2282                 ++d->bd_rcount;
 2283                 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
 2284                 if (slen != 0) {
 2285                         BPFD_LOCK(d);
 2286 
 2287                         d->bd_fcount++;
 2288                         if (gottime < bpf_ts_quality(d->bd_tstamp))
 2289                                 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
 2290 #ifdef MAC
 2291                         if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
 2292 #endif
 2293                                 catchpacket(d, (u_char *)&mb, pktlen, slen,
 2294                                     bpf_append_mbuf, &bt);
 2295                         BPFD_UNLOCK(d);
 2296                 }
 2297         }
 2298         BPFIF_RUNLOCK(bp);
 2299 }
 2300 
 2301 #undef  BPF_CHECK_DIRECTION
 2302 
 2303 #undef  BPF_TSTAMP_NONE
 2304 #undef  BPF_TSTAMP_FAST
 2305 #undef  BPF_TSTAMP_NORMAL
 2306 #undef  BPF_TSTAMP_EXTERN
 2307 
 2308 static int
 2309 bpf_hdrlen(struct bpf_d *d)
 2310 {
 2311         int hdrlen;
 2312 
 2313         hdrlen = d->bd_bif->bif_hdrlen;
 2314 #ifndef BURN_BRIDGES
 2315         if (d->bd_tstamp == BPF_T_NONE ||
 2316             BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
 2317 #ifdef COMPAT_FREEBSD32
 2318                 if (d->bd_compat32)
 2319                         hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
 2320                 else
 2321 #endif
 2322                         hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
 2323         else
 2324 #endif
 2325                 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
 2326 #ifdef COMPAT_FREEBSD32
 2327         if (d->bd_compat32)
 2328                 hdrlen = BPF_WORDALIGN32(hdrlen);
 2329         else
 2330 #endif
 2331                 hdrlen = BPF_WORDALIGN(hdrlen);
 2332 
 2333         return (hdrlen - d->bd_bif->bif_hdrlen);
 2334 }
 2335 
 2336 static void
 2337 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
 2338 {
 2339         struct bintime bt2, boottimebin;
 2340         struct timeval tsm;
 2341         struct timespec tsn;
 2342 
 2343         if ((tstype & BPF_T_MONOTONIC) == 0) {
 2344                 bt2 = *bt;
 2345                 getboottimebin(&boottimebin);
 2346                 bintime_add(&bt2, &boottimebin);
 2347                 bt = &bt2;
 2348         }
 2349         switch (BPF_T_FORMAT(tstype)) {
 2350         case BPF_T_MICROTIME:
 2351                 bintime2timeval(bt, &tsm);
 2352                 ts->bt_sec = tsm.tv_sec;
 2353                 ts->bt_frac = tsm.tv_usec;
 2354                 break;
 2355         case BPF_T_NANOTIME:
 2356                 bintime2timespec(bt, &tsn);
 2357                 ts->bt_sec = tsn.tv_sec;
 2358                 ts->bt_frac = tsn.tv_nsec;
 2359                 break;
 2360         case BPF_T_BINTIME:
 2361                 ts->bt_sec = bt->sec;
 2362                 ts->bt_frac = bt->frac;
 2363                 break;
 2364         }
 2365 }
 2366 
 2367 /*
 2368  * Move the packet data from interface memory (pkt) into the
 2369  * store buffer.  "cpfn" is the routine called to do the actual data
 2370  * transfer.  bcopy is passed in to copy contiguous chunks, while
 2371  * bpf_append_mbuf is passed in to copy mbuf chains.  In the latter case,
 2372  * pkt is really an mbuf.
 2373  */
 2374 static void
 2375 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
 2376     void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
 2377     struct bintime *bt)
 2378 {
 2379         struct bpf_xhdr hdr;
 2380 #ifndef BURN_BRIDGES
 2381         struct bpf_hdr hdr_old;
 2382 #ifdef COMPAT_FREEBSD32
 2383         struct bpf_hdr32 hdr32_old;
 2384 #endif
 2385 #endif
 2386         int caplen, curlen, hdrlen, totlen;
 2387         int do_wakeup = 0;
 2388         int do_timestamp;
 2389         int tstype;
 2390 
 2391         BPFD_LOCK_ASSERT(d);
 2392 
 2393         /*
 2394          * Detect whether user space has released a buffer back to us, and if
 2395          * so, move it from being a hold buffer to a free buffer.  This may
 2396          * not be the best place to do it (for example, we might only want to
 2397          * run this check if we need the space), but for now it's a reliable
 2398          * spot to do it.
 2399          */
 2400         if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
 2401                 d->bd_fbuf = d->bd_hbuf;
 2402                 d->bd_hbuf = NULL;
 2403                 d->bd_hlen = 0;
 2404                 bpf_buf_reclaimed(d);
 2405         }
 2406 
 2407         /*
 2408          * Figure out how many bytes to move.  If the packet is
 2409          * greater or equal to the snapshot length, transfer that
 2410          * much.  Otherwise, transfer the whole packet (unless
 2411          * we hit the buffer size limit).
 2412          */
 2413         hdrlen = bpf_hdrlen(d);
 2414         totlen = hdrlen + min(snaplen, pktlen);
 2415         if (totlen > d->bd_bufsize)
 2416                 totlen = d->bd_bufsize;
 2417 
 2418         /*
 2419          * Round up the end of the previous packet to the next longword.
 2420          *
 2421          * Drop the packet if there's no room and no hope of room
 2422          * If the packet would overflow the storage buffer or the storage
 2423          * buffer is considered immutable by the buffer model, try to rotate
 2424          * the buffer and wakeup pending processes.
 2425          */
 2426 #ifdef COMPAT_FREEBSD32
 2427         if (d->bd_compat32)
 2428                 curlen = BPF_WORDALIGN32(d->bd_slen);
 2429         else
 2430 #endif
 2431                 curlen = BPF_WORDALIGN(d->bd_slen);
 2432         if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
 2433                 if (d->bd_fbuf == NULL) {
 2434                         /*
 2435                          * There's no room in the store buffer, and no
 2436                          * prospect of room, so drop the packet.  Notify the
 2437                          * buffer model.
 2438                          */
 2439                         bpf_buffull(d);
 2440                         ++d->bd_dcount;
 2441                         return;
 2442                 }
 2443                 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
 2444                 ROTATE_BUFFERS(d);
 2445                 do_wakeup = 1;
 2446                 curlen = 0;
 2447         } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
 2448                 /*
 2449                  * Immediate mode is set, or the read timeout has already
 2450                  * expired during a select call.  A packet arrived, so the
 2451                  * reader should be woken up.
 2452                  */
 2453                 do_wakeup = 1;
 2454         caplen = totlen - hdrlen;
 2455         tstype = d->bd_tstamp;
 2456         do_timestamp = tstype != BPF_T_NONE;
 2457 #ifndef BURN_BRIDGES
 2458         if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
 2459                 struct bpf_ts ts;
 2460                 if (do_timestamp)
 2461                         bpf_bintime2ts(bt, &ts, tstype);
 2462 #ifdef COMPAT_FREEBSD32
 2463                 if (d->bd_compat32) {
 2464                         bzero(&hdr32_old, sizeof(hdr32_old));
 2465                         if (do_timestamp) {
 2466                                 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
 2467                                 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
 2468                         }
 2469                         hdr32_old.bh_datalen = pktlen;
 2470                         hdr32_old.bh_hdrlen = hdrlen;
 2471                         hdr32_old.bh_caplen = caplen;
 2472                         bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
 2473                             sizeof(hdr32_old));
 2474                         goto copy;
 2475                 }
 2476 #endif
 2477                 bzero(&hdr_old, sizeof(hdr_old));
 2478                 if (do_timestamp) {
 2479                         hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
 2480                         hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
 2481                 }
 2482                 hdr_old.bh_datalen = pktlen;
 2483                 hdr_old.bh_hdrlen = hdrlen;
 2484                 hdr_old.bh_caplen = caplen;
 2485                 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
 2486                     sizeof(hdr_old));
 2487                 goto copy;
 2488         }
 2489 #endif
 2490 
 2491         /*
 2492          * Append the bpf header.  Note we append the actual header size, but
 2493          * move forward the length of the header plus padding.
 2494          */
 2495         bzero(&hdr, sizeof(hdr));
 2496         if (do_timestamp)
 2497                 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
 2498         hdr.bh_datalen = pktlen;
 2499         hdr.bh_hdrlen = hdrlen;
 2500         hdr.bh_caplen = caplen;
 2501         bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
 2502 
 2503         /*
 2504          * Copy the packet data into the store buffer and update its length.
 2505          */
 2506 #ifndef BURN_BRIDGES
 2507 copy:
 2508 #endif
 2509         (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
 2510         d->bd_slen = curlen + totlen;
 2511 
 2512         if (do_wakeup)
 2513                 bpf_wakeup(d);
 2514 }
 2515 
 2516 /*
 2517  * Free buffers currently in use by a descriptor.
 2518  * Called on close.
 2519  */
 2520 static void
 2521 bpf_freed(struct bpf_d *d)
 2522 {
 2523 
 2524         /*
 2525          * We don't need to lock out interrupts since this descriptor has
 2526          * been detached from its interface and it yet hasn't been marked
 2527          * free.
 2528          */
 2529         bpf_free(d);
 2530         if (d->bd_rfilter != NULL) {
 2531                 free((caddr_t)d->bd_rfilter, M_BPF);
 2532 #ifdef BPF_JITTER
 2533                 if (d->bd_bfilter != NULL)
 2534                         bpf_destroy_jit_filter(d->bd_bfilter);
 2535 #endif
 2536         }
 2537         if (d->bd_wfilter != NULL)
 2538                 free((caddr_t)d->bd_wfilter, M_BPF);
 2539         mtx_destroy(&d->bd_lock);
 2540 }
 2541 
 2542 /*
 2543  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
 2544  * fixed size of the link header (variable length headers not yet supported).
 2545  */
 2546 void
 2547 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
 2548 {
 2549 
 2550         bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
 2551 }
 2552 
 2553 /*
 2554  * Attach an interface to bpf.  ifp is a pointer to the structure
 2555  * defining the interface to be attached, dlt is the link layer type,
 2556  * and hdrlen is the fixed size of the link header (variable length
 2557  * headers are not yet supporrted).
 2558  */
 2559 void
 2560 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
 2561 {
 2562         struct bpf_if *bp;
 2563 
 2564         KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
 2565 
 2566         bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
 2567 
 2568         rw_init(&bp->bif_lock, "bpf interface lock");
 2569         LIST_INIT(&bp->bif_dlist);
 2570         LIST_INIT(&bp->bif_wlist);
 2571         bp->bif_ifp = ifp;
 2572         bp->bif_dlt = dlt;
 2573         bp->bif_hdrlen = hdrlen;
 2574         bp->bif_bpf = driverp;
 2575         *driverp = bp;
 2576 
 2577         BPF_LOCK();
 2578         LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
 2579         BPF_UNLOCK();
 2580 
 2581         if (bootverbose && IS_DEFAULT_VNET(curvnet))
 2582                 if_printf(ifp, "bpf attached\n");
 2583 }
 2584 
 2585 #ifdef VIMAGE
 2586 /*
 2587  * When moving interfaces between vnet instances we need a way to
 2588  * query the dlt and hdrlen before detach so we can re-attch the if_bpf
 2589  * after the vmove.  We unfortunately have no device driver infrastructure
 2590  * to query the interface for these values after creation/attach, thus
 2591  * add this as a workaround.
 2592  */
 2593 int
 2594 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
 2595 {
 2596 
 2597         if (bp == NULL)
 2598                 return (ENXIO);
 2599         if (bif_dlt == NULL && bif_hdrlen == NULL)
 2600                 return (0);
 2601 
 2602         if (bif_dlt != NULL)
 2603                 *bif_dlt = bp->bif_dlt;
 2604         if (bif_hdrlen != NULL)
 2605                 *bif_hdrlen = bp->bif_hdrlen;
 2606 
 2607         return (0);
 2608 }
 2609 #endif
 2610 
 2611 /*
 2612  * Detach bpf from an interface. This involves detaching each descriptor
 2613  * associated with the interface. Notify each descriptor as it's detached
 2614  * so that any sleepers wake up and get ENXIO.
 2615  */
 2616 void
 2617 bpfdetach(struct ifnet *ifp)
 2618 {
 2619         struct bpf_if   *bp, *bp_temp;
 2620         struct bpf_d    *d;
 2621         int ndetached;
 2622 
 2623         ndetached = 0;
 2624 
 2625         BPF_LOCK();
 2626         /* Find all bpf_if struct's which reference ifp and detach them. */
 2627         LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
 2628                 if (ifp != bp->bif_ifp)
 2629                         continue;
 2630 
 2631                 LIST_REMOVE(bp, bif_next);
 2632                 /* Add to to-be-freed list */
 2633                 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
 2634 
 2635                 ndetached++;
 2636                 /*
 2637                  * Delay freeing bp till interface is detached
 2638                  * and all routes through this interface are removed.
 2639                  * Mark bp as detached to restrict new consumers.
 2640                  */
 2641                 BPFIF_WLOCK(bp);
 2642                 bp->bif_flags |= BPFIF_FLAG_DYING;
 2643                 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
 2644                 BPFIF_WUNLOCK(bp);
 2645 
 2646                 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
 2647                     __func__, bp->bif_dlt, bp, ifp);
 2648 
 2649                 /* Free common descriptors */
 2650                 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
 2651                         bpf_detachd_locked(d);
 2652                         BPFD_LOCK(d);
 2653                         bpf_wakeup(d);
 2654                         BPFD_UNLOCK(d);
 2655                 }
 2656 
 2657                 /* Free writer-only descriptors */
 2658                 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
 2659                         bpf_detachd_locked(d);
 2660                         BPFD_LOCK(d);
 2661                         bpf_wakeup(d);
 2662                         BPFD_UNLOCK(d);
 2663                 }
 2664         }
 2665         BPF_UNLOCK();
 2666 
 2667 #ifdef INVARIANTS
 2668         if (ndetached == 0)
 2669                 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
 2670 #endif
 2671 }
 2672 
 2673 /*
 2674  * Interface departure handler.
 2675  * Note departure event does not guarantee interface is going down.
 2676  * Interface renaming is currently done via departure/arrival event set.
 2677  *
 2678  * Departure handled is called after all routes pointing to
 2679  * given interface are removed and interface is in down state
 2680  * restricting any packets to be sent/received. We assume it is now safe
 2681  * to free data allocated by BPF.
 2682  */
 2683 static void
 2684 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
 2685 {
 2686         struct bpf_if *bp, *bp_temp;
 2687         int nmatched = 0;
 2688 
 2689         /* Ignore ifnet renaming. */
 2690         if (ifp->if_flags & IFF_RENAMING)
 2691                 return;
 2692 
 2693         BPF_LOCK();
 2694         /*
 2695          * Find matching entries in free list.
 2696          * Nothing should be found if bpfdetach() was not called.
 2697          */
 2698         LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
 2699                 if (ifp != bp->bif_ifp)
 2700                         continue;
 2701 
 2702                 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
 2703                     __func__, bp, ifp);
 2704 
 2705                 LIST_REMOVE(bp, bif_next);
 2706 
 2707                 rw_destroy(&bp->bif_lock);
 2708                 free(bp, M_BPF);
 2709 
 2710                 nmatched++;
 2711         }
 2712         BPF_UNLOCK();
 2713 }
 2714 
 2715 /*
 2716  * Get a list of available data link type of the interface.
 2717  */
 2718 static int
 2719 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
 2720 {
 2721         struct ifnet *ifp;
 2722         struct bpf_if *bp;
 2723         u_int *lst;
 2724         int error, n, n1;
 2725 
 2726         BPF_LOCK_ASSERT();
 2727 
 2728         ifp = d->bd_bif->bif_ifp;
 2729 again:
 2730         n1 = 0;
 2731         LIST_FOREACH(bp, &bpf_iflist, bif_next) {
 2732                 if (bp->bif_ifp == ifp)
 2733                         n1++;
 2734         }
 2735         if (bfl->bfl_list == NULL) {
 2736                 bfl->bfl_len = n1;
 2737                 return (0);
 2738         }
 2739         if (n1 > bfl->bfl_len)
 2740                 return (ENOMEM);
 2741         BPF_UNLOCK();
 2742         lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
 2743         n = 0;
 2744         BPF_LOCK();
 2745         LIST_FOREACH(bp, &bpf_iflist, bif_next) {
 2746                 if (bp->bif_ifp != ifp)
 2747                         continue;
 2748                 if (n >= n1) {
 2749                         free(lst, M_TEMP);
 2750                         goto again;
 2751                 }
 2752                 lst[n] = bp->bif_dlt;
 2753                 n++;
 2754         }
 2755         BPF_UNLOCK();
 2756         error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
 2757         free(lst, M_TEMP);
 2758         BPF_LOCK();
 2759         bfl->bfl_len = n;
 2760         return (error);
 2761 }
 2762 
 2763 /*
 2764  * Set the data link type of a BPF instance.
 2765  */
 2766 static int
 2767 bpf_setdlt(struct bpf_d *d, u_int dlt)
 2768 {
 2769         int error, opromisc;
 2770         struct ifnet *ifp;
 2771         struct bpf_if *bp;
 2772 
 2773         BPF_LOCK_ASSERT();
 2774 
 2775         if (d->bd_bif->bif_dlt == dlt)
 2776                 return (0);
 2777         ifp = d->bd_bif->bif_ifp;
 2778 
 2779         LIST_FOREACH(bp, &bpf_iflist, bif_next) {
 2780                 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
 2781                         break;
 2782         }
 2783 
 2784         if (bp != NULL) {
 2785                 opromisc = d->bd_promisc;
 2786                 bpf_attachd(d, bp);
 2787                 BPFD_LOCK(d);
 2788                 reset_d(d);
 2789                 BPFD_UNLOCK(d);
 2790                 if (opromisc) {
 2791                         error = ifpromisc(bp->bif_ifp, 1);
 2792                         if (error)
 2793                                 if_printf(bp->bif_ifp,
 2794                                         "bpf_setdlt: ifpromisc failed (%d)\n",
 2795                                         error);
 2796                         else
 2797                                 d->bd_promisc = 1;
 2798                 }
 2799         }
 2800         return (bp == NULL ? EINVAL : 0);
 2801 }
 2802 
 2803 static void
 2804 bpf_drvinit(void *unused)
 2805 {
 2806         struct cdev *dev;
 2807 
 2808         sx_init(&bpf_sx, "bpf global lock");
 2809         LIST_INIT(&bpf_iflist);
 2810         LIST_INIT(&bpf_freelist);
 2811 
 2812         dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
 2813         /* For compatibility */
 2814         make_dev_alias(dev, "bpf0");
 2815 
 2816         /* Register interface departure handler */
 2817         bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
 2818                     ifnet_departure_event, bpf_ifdetach, NULL,
 2819                     EVENTHANDLER_PRI_ANY);
 2820 }
 2821 
 2822 /*
 2823  * Zero out the various packet counters associated with all of the bpf
 2824  * descriptors.  At some point, we will probably want to get a bit more
 2825  * granular and allow the user to specify descriptors to be zeroed.
 2826  */
 2827 static void
 2828 bpf_zero_counters(void)
 2829 {
 2830         struct bpf_if *bp;
 2831         struct bpf_d *bd;
 2832 
 2833         BPF_LOCK();
 2834         LIST_FOREACH(bp, &bpf_iflist, bif_next) {
 2835                 BPFIF_RLOCK(bp);
 2836                 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
 2837                         BPFD_LOCK(bd);
 2838                         bd->bd_rcount = 0;
 2839                         bd->bd_dcount = 0;
 2840                         bd->bd_fcount = 0;
 2841                         bd->bd_wcount = 0;
 2842                         bd->bd_wfcount = 0;
 2843                         bd->bd_zcopy = 0;
 2844                         BPFD_UNLOCK(bd);
 2845                 }
 2846                 BPFIF_RUNLOCK(bp);
 2847         }
 2848         BPF_UNLOCK();
 2849 }
 2850 
 2851 /*
 2852  * Fill filter statistics
 2853  */
 2854 static void
 2855 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
 2856 {
 2857 
 2858         bzero(d, sizeof(*d));
 2859         BPFD_LOCK_ASSERT(bd);
 2860         d->bd_structsize = sizeof(*d);
 2861         /* XXX: reading should be protected by global lock */
 2862         d->bd_immediate = bd->bd_immediate;
 2863         d->bd_promisc = bd->bd_promisc;
 2864         d->bd_hdrcmplt = bd->bd_hdrcmplt;
 2865         d->bd_direction = bd->bd_direction;
 2866         d->bd_feedback = bd->bd_feedback;
 2867         d->bd_async = bd->bd_async;
 2868         d->bd_rcount = bd->bd_rcount;
 2869         d->bd_dcount = bd->bd_dcount;
 2870         d->bd_fcount = bd->bd_fcount;
 2871         d->bd_sig = bd->bd_sig;
 2872         d->bd_slen = bd->bd_slen;
 2873         d->bd_hlen = bd->bd_hlen;
 2874         d->bd_bufsize = bd->bd_bufsize;
 2875         d->bd_pid = bd->bd_pid;
 2876         strlcpy(d->bd_ifname,
 2877             bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
 2878         d->bd_locked = bd->bd_locked;
 2879         d->bd_wcount = bd->bd_wcount;
 2880         d->bd_wdcount = bd->bd_wdcount;
 2881         d->bd_wfcount = bd->bd_wfcount;
 2882         d->bd_zcopy = bd->bd_zcopy;
 2883         d->bd_bufmode = bd->bd_bufmode;
 2884 }
 2885 
 2886 /*
 2887  * Handle `netstat -B' stats request
 2888  */
 2889 static int
 2890 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
 2891 {
 2892         static const struct xbpf_d zerostats;
 2893         struct xbpf_d *xbdbuf, *xbd, tempstats;
 2894         int index, error;
 2895         struct bpf_if *bp;
 2896         struct bpf_d *bd;
 2897 
 2898         /*
 2899          * XXX This is not technically correct. It is possible for non
 2900          * privileged users to open bpf devices. It would make sense
 2901          * if the users who opened the devices were able to retrieve
 2902          * the statistics for them, too.
 2903          */
 2904         error = priv_check(req->td, PRIV_NET_BPF);
 2905         if (error)
 2906                 return (error);
 2907         /*
 2908          * Check to see if the user is requesting that the counters be
 2909          * zeroed out.  Explicitly check that the supplied data is zeroed,
 2910          * as we aren't allowing the user to set the counters currently.
 2911          */
 2912         if (req->newptr != NULL) {
 2913                 if (req->newlen != sizeof(tempstats))
 2914                         return (EINVAL);
 2915                 memset(&tempstats, 0, sizeof(tempstats));
 2916                 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
 2917                 if (error)
 2918                         return (error);
 2919                 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
 2920                         return (EINVAL);
 2921                 bpf_zero_counters();
 2922                 return (0);
 2923         }
 2924         if (req->oldptr == NULL)
 2925                 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
 2926         if (bpf_bpfd_cnt == 0)
 2927                 return (SYSCTL_OUT(req, 0, 0));
 2928         xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
 2929         BPF_LOCK();
 2930         if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
 2931                 BPF_UNLOCK();
 2932                 free(xbdbuf, M_BPF);
 2933                 return (ENOMEM);
 2934         }
 2935         index = 0;
 2936         LIST_FOREACH(bp, &bpf_iflist, bif_next) {
 2937                 BPFIF_RLOCK(bp);
 2938                 /* Send writers-only first */
 2939                 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
 2940                         xbd = &xbdbuf[index++];
 2941                         BPFD_LOCK(bd);
 2942                         bpfstats_fill_xbpf(xbd, bd);
 2943                         BPFD_UNLOCK(bd);
 2944                 }
 2945                 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
 2946                         xbd = &xbdbuf[index++];
 2947                         BPFD_LOCK(bd);
 2948                         bpfstats_fill_xbpf(xbd, bd);
 2949                         BPFD_UNLOCK(bd);
 2950                 }
 2951                 BPFIF_RUNLOCK(bp);
 2952         }
 2953         BPF_UNLOCK();
 2954         error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
 2955         free(xbdbuf, M_BPF);
 2956         return (error);
 2957 }
 2958 
 2959 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
 2960 
 2961 #else /* !DEV_BPF && !NETGRAPH_BPF */
 2962 
 2963 /*
 2964  * NOP stubs to allow bpf-using drivers to load and function.
 2965  *
 2966  * A 'better' implementation would allow the core bpf functionality
 2967  * to be loaded at runtime.
 2968  */
 2969 
 2970 void
 2971 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
 2972 {
 2973 }
 2974 
 2975 void
 2976 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
 2977 {
 2978 }
 2979 
 2980 void
 2981 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
 2982 {
 2983 }
 2984 
 2985 void
 2986 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
 2987 {
 2988 
 2989         bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
 2990 }
 2991 
 2992 void
 2993 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
 2994 {
 2995 
 2996         *driverp = (struct bpf_if *)&dead_bpf_if;
 2997 }
 2998 
 2999 void
 3000 bpfdetach(struct ifnet *ifp)
 3001 {
 3002 }
 3003 
 3004 u_int
 3005 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
 3006 {
 3007         return -1;      /* "no filter" behaviour */
 3008 }
 3009 
 3010 int
 3011 bpf_validate(const struct bpf_insn *f, int len)
 3012 {
 3013         return 0;               /* false */
 3014 }
 3015 
 3016 #endif /* !DEV_BPF && !NETGRAPH_BPF */
 3017 
 3018 #ifdef DDB
 3019 static void
 3020 bpf_show_bpf_if(struct bpf_if *bpf_if)
 3021 {
 3022 
 3023         if (bpf_if == NULL)
 3024                 return;
 3025         db_printf("%p:\n", bpf_if);
 3026 #define BPF_DB_PRINTF(f, e)     db_printf("   %s = " f "\n", #e, bpf_if->e);
 3027         /* bif_ext.bif_next */
 3028         /* bif_ext.bif_dlist */
 3029         BPF_DB_PRINTF("%#x", bif_dlt);
 3030         BPF_DB_PRINTF("%u", bif_hdrlen);
 3031         BPF_DB_PRINTF("%p", bif_ifp);
 3032         /* bif_lock */
 3033         /* bif_wlist */
 3034         BPF_DB_PRINTF("%#x", bif_flags);
 3035 }
 3036 
 3037 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
 3038 {
 3039 
 3040         if (!have_addr) {
 3041                 db_printf("usage: show bpf_if <struct bpf_if *>\n");
 3042                 return;
 3043         }
 3044 
 3045         bpf_show_bpf_if((struct bpf_if *)addr);
 3046 }
 3047 #endif

Cache object: 2a5b808e59f45841b549a4fabefc7a42


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.