Home | History | Annotate | Line # | Download | only in net
bpf.c revision 1.99
      1 /*	$NetBSD: bpf.c,v 1.99 2004/05/29 08:56:19 darrenr Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1990, 1991, 1993
      5  *	The Regents of the University of California.  All rights reserved.
      6  *
      7  * This code is derived from the Stanford/CMU enet packet filter,
      8  * (net/enet.c) distributed as part of 4.3BSD, and code contributed
      9  * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
     10  * Berkeley Laboratory.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. Neither the name of the University nor the names of its contributors
     21  *    may be used to endorse or promote products derived from this software
     22  *    without specific prior written permission.
     23  *
     24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     34  * SUCH DAMAGE.
     35  *
     36  *	@(#)bpf.c	8.4 (Berkeley) 1/9/95
     37  * static char rcsid[] =
     38  * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp ";
     39  */
     40 
     41 #include <sys/cdefs.h>
     42 __KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.99 2004/05/29 08:56:19 darrenr Exp $");
     43 
     44 #include "bpfilter.h"
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/mbuf.h>
     49 #include <sys/buf.h>
     50 #include <sys/time.h>
     51 #include <sys/proc.h>
     52 #include <sys/user.h>
     53 #include <sys/ioctl.h>
     54 #include <sys/conf.h>
     55 #include <sys/vnode.h>
     56 
     57 #include <sys/file.h>
     58 #include <sys/tty.h>
     59 #include <sys/uio.h>
     60 
     61 #include <sys/protosw.h>
     62 #include <sys/socket.h>
     63 #include <sys/errno.h>
     64 #include <sys/kernel.h>
     65 #include <sys/poll.h>
     66 #include <sys/sysctl.h>
     67 
     68 #include <net/if.h>
     69 
     70 #include <net/bpf.h>
     71 #include <net/bpfdesc.h>
     72 
     73 #include <net/if_arc.h>
     74 #include <net/if_ether.h>
     75 
     76 #include <netinet/in.h>
     77 #include <netinet/if_inarp.h>
     78 
     79 #if defined(_KERNEL_OPT)
     80 #include "opt_bpf.h"
     81 #endif
     82 
     83 #ifndef BPF_BUFSIZE
     84 /*
     85  * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet
     86  * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k).
     87  */
     88 # define BPF_BUFSIZE 32768
     89 #endif
     90 
     91 #define PRINET  26			/* interruptible */
     92 
     93 /*
     94  * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able.
     95  * XXX the default values should be computed dynamically based
     96  * on available memory size and available mbuf clusters.
     97  */
     98 int bpf_bufsize = BPF_BUFSIZE;
     99 int bpf_maxbufsize = BPF_DFLTBUFSIZE;	/* XXX set dynamically, see above */
    100 
    101 /*
    102  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
    103  *  bpf_dtab holds the descriptors, indexed by minor device #
    104  */
    105 struct bpf_if	*bpf_iflist;
    106 struct bpf_d	bpf_dtab[NBPFILTER];
    107 
    108 static int	bpf_allocbufs __P((struct bpf_d *));
    109 static void	bpf_freed __P((struct bpf_d *));
    110 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
    111 static void	*bpf_mcpy __P((void *, const void *, size_t));
    112 static int	bpf_movein __P((struct uio *, int, int,
    113 			        struct mbuf **, struct sockaddr *));
    114 static void	bpf_attachd __P((struct bpf_d *, struct bpf_if *));
    115 static void	bpf_detachd __P((struct bpf_d *));
    116 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
    117 static void	bpf_timed_out __P((void *));
    118 static __inline void
    119 		bpf_wakeup __P((struct bpf_d *));
    120 static void	catchpacket __P((struct bpf_d *, u_char *, u_int, u_int,
    121 				 void *(*)(void *, const void *, size_t)));
    122 static void	reset_d __P((struct bpf_d *));
    123 static int	bpf_getdltlist __P((struct bpf_d *, struct bpf_dltlist *));
    124 static int	bpf_setdlt __P((struct bpf_d *, u_int));
    125 static int	bpf_mmapinfo __P((struct bpf_d *, struct bpf_mmapinfo *));
    126 static int	bpf_waitfordata __P((struct bpf_d *));
    127 
    128 dev_type_open(bpfopen);
    129 dev_type_close(bpfclose);
    130 dev_type_read(bpfread);
    131 dev_type_write(bpfwrite);
    132 dev_type_ioctl(bpfioctl);
    133 dev_type_poll(bpfpoll);
    134 dev_type_kqfilter(bpfkqfilter);
    135 dev_type_mmap(bpfmmap);
    136 
    137 const struct cdevsw bpf_cdevsw = {
    138 	bpfopen, bpfclose, bpfread, bpfwrite, bpfioctl,
    139 	nostop, notty, bpfpoll, bpfmmap, bpfkqfilter,
    140 };
    141 
    142 static int
    143 bpf_movein(uio, linktype, mtu, mp, sockp)
    144 	struct uio *uio;
    145 	int linktype;
    146 	int mtu;
    147 	struct mbuf **mp;
    148 	struct sockaddr *sockp;
    149 {
    150 	struct mbuf *m;
    151 	int error;
    152 	int len;
    153 	int hlen;
    154 	int align;
    155 
    156 	/*
    157 	 * Build a sockaddr based on the data link layer type.
    158 	 * We do this at this level because the ethernet header
    159 	 * is copied directly into the data field of the sockaddr.
    160 	 * In the case of SLIP, there is no header and the packet
    161 	 * is forwarded as is.
    162 	 * Also, we are careful to leave room at the front of the mbuf
    163 	 * for the link level header.
    164 	 */
    165 	switch (linktype) {
    166 
    167 	case DLT_SLIP:
    168 		sockp->sa_family = AF_INET;
    169 		hlen = 0;
    170 		align = 0;
    171 		break;
    172 
    173 	case DLT_PPP:
    174 		sockp->sa_family = AF_UNSPEC;
    175 		hlen = 0;
    176 		align = 0;
    177 		break;
    178 
    179 	case DLT_EN10MB:
    180 		sockp->sa_family = AF_UNSPEC;
    181 		/* XXX Would MAXLINKHDR be better? */
    182  		/* 6(dst)+6(src)+2(type) */
    183 		hlen = sizeof(struct ether_header);
    184 		align = 2;
    185 		break;
    186 
    187 	case DLT_ARCNET:
    188 		sockp->sa_family = AF_UNSPEC;
    189 		hlen = ARC_HDRLEN;
    190 		align = 5;
    191 		break;
    192 
    193 	case DLT_FDDI:
    194 		sockp->sa_family = AF_LINK;
    195 		/* XXX 4(FORMAC)+6(dst)+6(src) */
    196 		hlen = 16;
    197 		align = 0;
    198 		break;
    199 
    200 	case DLT_ECONET:
    201 		sockp->sa_family = AF_UNSPEC;
    202 		hlen = 6;
    203 		align = 2;
    204 		break;
    205 
    206 	case DLT_NULL:
    207 		sockp->sa_family = AF_UNSPEC;
    208 		hlen = 0;
    209 		align = 0;
    210 		break;
    211 
    212 	default:
    213 		return (EIO);
    214 	}
    215 
    216 	len = uio->uio_resid;
    217 	/*
    218 	 * If there aren't enough bytes for a link level header or the
    219 	 * packet length exceeds the interface mtu, return an error.
    220 	 */
    221 	if (len < hlen || len - hlen > mtu)
    222 		return (EMSGSIZE);
    223 
    224 	/*
    225 	 * XXX Avoid complicated buffer chaining ---
    226 	 * bail if it won't fit in a single mbuf.
    227 	 * (Take into account possible alignment bytes)
    228 	 */
    229 	if ((unsigned)len > MCLBYTES - align)
    230 		return (EIO);
    231 
    232 	m = m_gethdr(M_WAIT, MT_DATA);
    233 	m->m_pkthdr.rcvif = 0;
    234 	m->m_pkthdr.len = len - hlen;
    235 	if (len > MHLEN - align) {
    236 		m_clget(m, M_WAIT);
    237 		if ((m->m_flags & M_EXT) == 0) {
    238 			error = ENOBUFS;
    239 			goto bad;
    240 		}
    241 	}
    242 
    243 	/* Insure the data is properly aligned */
    244 	if (align > 0) {
    245 		m->m_data += align;
    246 		m->m_len -= align;
    247 	}
    248 
    249 	error = uiomove(mtod(m, caddr_t), len, uio);
    250 	if (error)
    251 		goto bad;
    252 	if (hlen != 0) {
    253 		memcpy(sockp->sa_data, mtod(m, caddr_t), hlen);
    254 		m->m_data += hlen; /* XXX */
    255 		len -= hlen;
    256 	}
    257 	m->m_len = len;
    258 	*mp = m;
    259 	return (0);
    260 
    261 bad:
    262 	m_freem(m);
    263 	return (error);
    264 }
    265 
    266 /*
    267  * Attach file to the bpf interface, i.e. make d listen on bp.
    268  * Must be called at splnet.
    269  */
    270 static void
    271 bpf_attachd(d, bp)
    272 	struct bpf_d *d;
    273 	struct bpf_if *bp;
    274 {
    275 	/*
    276 	 * Point d at bp, and add d to the interface's list of listeners.
    277 	 * Finally, point the driver's bpf cookie at the interface so
    278 	 * it will divert packets to bpf.
    279 	 */
    280 	d->bd_bif = bp;
    281 	d->bd_next = bp->bif_dlist;
    282 	bp->bif_dlist = d;
    283 
    284 	*bp->bif_driverp = bp;
    285 }
    286 
    287 /*
    288  * Detach a file from its interface.
    289  */
    290 static void
    291 bpf_detachd(d)
    292 	struct bpf_d *d;
    293 {
    294 	struct bpf_d **p;
    295 	struct bpf_if *bp;
    296 
    297 	bp = d->bd_bif;
    298 	/*
    299 	 * Check if this descriptor had requested promiscuous mode.
    300 	 * If so, turn it off.
    301 	 */
    302 	if (d->bd_promisc) {
    303 		int error;
    304 
    305 		d->bd_promisc = 0;
    306 		/*
    307 		 * Take device out of promiscuous mode.  Since we were
    308 		 * able to enter promiscuous mode, we should be able
    309 		 * to turn it off.  But we can get an error if
    310 		 * the interface was configured down, so only panic
    311 		 * if we don't get an unexpected error.
    312 		 */
    313   		error = ifpromisc(bp->bif_ifp, 0);
    314 		if (error && error != EINVAL)
    315 			panic("bpf: ifpromisc failed");
    316 	}
    317 	/* Remove d from the interface's descriptor list. */
    318 	p = &bp->bif_dlist;
    319 	while (*p != d) {
    320 		p = &(*p)->bd_next;
    321 		if (*p == 0)
    322 			panic("bpf_detachd: descriptor not in list");
    323 	}
    324 	*p = (*p)->bd_next;
    325 	if (bp->bif_dlist == 0)
    326 		/*
    327 		 * Let the driver know that there are no more listeners.
    328 		 */
    329 		*d->bd_bif->bif_driverp = 0;
    330 	d->bd_bif = 0;
    331 }
    332 
    333 
    334 /*
    335  * Mark a descriptor free by making it point to itself.
    336  * This is probably cheaper than marking with a constant since
    337  * the address should be in a register anyway.
    338  */
    339 #define D_ISFREE(d) ((d) == (d)->bd_next)
    340 #define D_MARKFREE(d) ((d)->bd_next = (d))
    341 #define D_MARKUSED(d) ((d)->bd_next = 0)
    342 
    343 /*
    344  * bpfilterattach() is called at boot time.
    345  */
    346 /* ARGSUSED */
    347 void
    348 bpfilterattach(n)
    349 	int n;
    350 {
    351 	int i;
    352 	/*
    353 	 * Mark all the descriptors free.
    354 	 */
    355 	for (i = 0; i < NBPFILTER; ++i)
    356 		D_MARKFREE(&bpf_dtab[i]);
    357 
    358 }
    359 
    360 /*
    361  * Open ethernet device.  Returns ENXIO for illegal minor device number,
    362  * EBUSY if file is open by another process.
    363  */
    364 /* ARGSUSED */
    365 int
    366 bpfopen(dev, flag, mode, p)
    367 	dev_t dev;
    368 	int flag;
    369 	int mode;
    370 	struct proc *p;
    371 {
    372 	struct bpf_d *d;
    373 
    374 	if (minor(dev) >= NBPFILTER)
    375 		return (ENXIO);
    376 	/*
    377 	 * Each minor can be opened by only one process.  If the requested
    378 	 * minor is in use, return EBUSY.
    379 	 */
    380 	d = &bpf_dtab[minor(dev)];
    381 	if (!D_ISFREE(d))
    382 		return (EBUSY);
    383 
    384 	/* Mark "free" and do most initialization. */
    385 	memset((char *)d, 0, sizeof(*d));
    386 	d->bd_bufsize = bpf_bufsize;
    387 	d->bd_seesent = 1;
    388 	callout_init(&d->bd_callout);
    389 
    390 	return (0);
    391 }
    392 
    393 /*
    394  * Close the descriptor by detaching it from its interface,
    395  * deallocating its buffers, and marking it free.
    396  */
    397 /* ARGSUSED */
    398 int
    399 bpfclose(dev, flag, mode, p)
    400 	dev_t dev;
    401 	int flag;
    402 	int mode;
    403 	struct proc *p;
    404 {
    405 	struct bpf_d *d = &bpf_dtab[minor(dev)];
    406 	int s;
    407 
    408 	s = splnet();
    409 	if (d->bd_state == BPF_WAITING)
    410 		callout_stop(&d->bd_callout);
    411 	d->bd_state = BPF_IDLE;
    412 	if (d->bd_bif)
    413 		bpf_detachd(d);
    414 	splx(s);
    415 	bpf_freed(d);
    416 
    417 	return (0);
    418 }
    419 
    420 /*
    421  * Rotate the packet buffers in descriptor d.  Move the store buffer
    422  * into the hold slot, and the free buffer into the store slot.
    423  * Zero the length of the new store buffer.
    424  */
    425 #define ROTATE_BUFFERS(d) \
    426 	do { \
    427 		(d)->bd_hbuf = (d)->bd_sbuf; \
    428 		(d)->bd_hlen = (d)->bd_slen; \
    429 		(d)->bd_sbuf = (d)->bd_fbuf; \
    430 		(d)->bd_slen = 0; \
    431 		(d)->bd_fbuf = 0; \
    432 	} while (0)
    433 
    434 /*
    435  *  bpfread - read next chunk of packets from buffers
    436  */
    437 int
    438 bpfread(dev, uio, ioflag)
    439 	dev_t dev;
    440 	struct uio *uio;
    441 	int ioflag;
    442 {
    443 	struct bpf_d *d = &bpf_dtab[minor(dev)];
    444 	int timed_out;
    445 	int error;
    446 	int s;
    447 
    448 	/*
    449 	 * Restrict application to use a buffer the same size as
    450 	 * as kernel buffers.
    451 	 */
    452 	if (uio->uio_resid != d->bd_bufsize)
    453 		return (EINVAL);
    454 
    455 	s = splnet();
    456 	if (d->bd_state == BPF_WAITING)
    457 		callout_stop(&d->bd_callout);
    458 	timed_out = (d->bd_state == BPF_TIMED_OUT);
    459 	d->bd_state = BPF_IDLE;
    460 	error = bpf_waitfordata(d);
    461 	if (error != 0)
    462 		goto done;
    463 	/*
    464 	 * At this point, we know we have something in the hold slot.
    465 	 */
    466 	splx(s);
    467 
    468 	/*
    469 	 * Move data from hold buffer into user space.
    470 	 * We know the entire buffer is transferred since
    471 	 * we checked above that the read buffer is bpf_bufsize bytes.
    472 	 */
    473 	error = uiomove(d->bd_hbuf, d->bd_hlen, uio);
    474 
    475 	s = splnet();
    476 	d->bd_fbuf = d->bd_hbuf;
    477 	d->bd_hbuf = 0;
    478 	d->bd_hlen = 0;
    479 done:
    480 	splx(s);
    481 	if (error == -1)
    482 		error = 0;
    483 	return (error);
    484 }
    485 
    486 
    487 /*
    488  * NOTE: splnet() is assumed to be held when calling this function.
    489  * It is left to the caller to drop the spl.
    490  */
    491 static int
    492 bpf_waitfordata(d)
    493 	struct bpf_d *d;
    494 {
    495 	int error;
    496 
    497 	/*
    498 	 * If the hold buffer is empty, then do a timed sleep, which
    499 	 * ends when the timeout expires or when enough packets
    500 	 * have arrived to fill the store buffer.
    501 	 */
    502 	while (d->bd_hbuf == 0) {
    503 		if (ioflag & IO_NDELAY) {
    504 			if (d->bd_slen == 0) {
    505 				return (EWOULDBLOCK);
    506 			}
    507 			ROTATE_BUFFERS(d);
    508 			break;
    509 		}
    510 
    511 		if ((d->bd_immediate || timed_out) && d->bd_slen != 0) {
    512 			/*
    513 			 * A packet(s) either arrived since the previous
    514 			 * read or arrived while we were asleep.
    515 			 * Rotate the buffers and return what's here.
    516 			 */
    517 			ROTATE_BUFFERS(d);
    518 			break;
    519 		}
    520 		error = tsleep((caddr_t)d, PRINET|PCATCH, "bpf",
    521 				d->bd_rtout);
    522 		if (error == EINTR || error == ERESTART) {
    523 			splx(s);
    524 			return (error);
    525 		}
    526 		if (error == EWOULDBLOCK) {
    527 			/*
    528 			 * On a timeout, return what's in the buffer,
    529 			 * which may be nothing.  If there is something
    530 			 * in the store buffer, we can rotate the buffers.
    531 			 */
    532 			if (d->bd_hbuf)
    533 				/*
    534 				 * We filled up the buffer in between
    535 				 * getting the timeout and arriving
    536 				 * here, so we don't need to rotate.
    537 				 */
    538 				break;
    539 
    540 			if (d->bd_slen == 0)
    541 				return -1;
    542 
    543 			if (d->bd_mapbuf == -1) {
    544 				ROTATE_BUFFERS(d);
    545 				break;
    546 			}
    547 		}
    548 		if (error != 0)
    549 			return error;
    550 	}
    551 
    552 	return 0;
    553 }
    554 
    555 
    556 /*
    557  * If there are processes sleeping on this descriptor, wake them up.
    558  */
    559 static __inline void
    560 bpf_wakeup(d)
    561 	struct bpf_d *d;
    562 {
    563 	wakeup((caddr_t)d);
    564 	if (d->bd_async)
    565 		fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL);
    566 
    567 	selnotify(&d->bd_sel, 0);
    568 	/* XXX */
    569 	d->bd_sel.sel_pid = 0;
    570 }
    571 
    572 
    573 static void
    574 bpf_timed_out(arg)
    575 	void *arg;
    576 {
    577 	struct bpf_d *d = (struct bpf_d *)arg;
    578 	int s;
    579 
    580 	s = splnet();
    581 	if (d->bd_state == BPF_WAITING) {
    582 		d->bd_state = BPF_TIMED_OUT;
    583 		if (d->bd_slen != 0)
    584 			bpf_wakeup(d);
    585 	}
    586 	splx(s);
    587 }
    588 
    589 
    590 int
    591 bpfwrite(dev, uio, ioflag)
    592 	dev_t dev;
    593 	struct uio *uio;
    594 	int ioflag;
    595 {
    596 	struct bpf_d *d = &bpf_dtab[minor(dev)];
    597 	struct ifnet *ifp;
    598 	struct mbuf *m;
    599 	int error, s;
    600 	static struct sockaddr_storage dst;
    601 
    602 	if (d->bd_bif == 0)
    603 		return (ENXIO);
    604 
    605 	ifp = d->bd_bif->bif_ifp;
    606 
    607 	if (uio->uio_resid == 0)
    608 		return (0);
    609 
    610 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m,
    611 		(struct sockaddr *) &dst);
    612 	if (error)
    613 		return (error);
    614 
    615 	if (m->m_pkthdr.len > ifp->if_mtu)
    616 		return (EMSGSIZE);
    617 
    618 	if (d->bd_hdrcmplt)
    619 		dst.ss_family = pseudo_AF_HDRCMPLT;
    620 
    621 	s = splsoftnet();
    622 	error = (*ifp->if_output)(ifp, m, (struct sockaddr *) &dst, NULL);
    623 	splx(s);
    624 	/*
    625 	 * The driver frees the mbuf.
    626 	 */
    627 	return (error);
    628 }
    629 
    630 /*
    631  * Reset a descriptor by flushing its packet buffer and clearing the
    632  * receive and drop counts.  Should be called at splnet.
    633  */
    634 static void
    635 reset_d(d)
    636 	struct bpf_d *d;
    637 {
    638 	if (d->bd_hbuf) {
    639 		/* Free the hold buffer. */
    640 		d->bd_fbuf = d->bd_hbuf;
    641 		d->bd_hbuf = 0;
    642 	}
    643 	d->bd_slen = 0;
    644 	d->bd_hlen = 0;
    645 	d->bd_rcount = 0;
    646 	d->bd_dcount = 0;
    647 	d->bd_ccount = 0;
    648 }
    649 
    650 #ifdef BPF_KERN_FILTER
    651 extern struct bpf_insn *bpf_tcp_filter;
    652 extern struct bpf_insn *bpf_udp_filter;
    653 #endif
    654 
    655 /*
    656  *  FIONREAD		Check for read packet available.
    657  *  BIOCGBLEN		Get buffer len [for read()].
    658  *  BIOCSETF		Set ethernet read filter.
    659  *  BIOCFLUSH		Flush read packet buffer.
    660  *  BIOCPROMISC		Put interface into promiscuous mode.
    661  *  BIOCGDLT		Get link layer type.
    662  *  BIOCGETIF		Get interface name.
    663  *  BIOCSETIF		Set interface.
    664  *  BIOCSRTIMEOUT	Set read timeout.
    665  *  BIOCGRTIMEOUT	Get read timeout.
    666  *  BIOCGSTATS		Get packet stats.
    667  *  BIOCIMMEDIATE	Set immediate mode.
    668  *  BIOCVERSION		Get filter language version.
    669  *  BIOGHDRCMPLT	Get "header already complete" flag.
    670  *  BIOSHDRCMPLT	Set "header already complete" flag.
    671  */
    672 /* ARGSUSED */
    673 int
    674 bpfioctl(dev, cmd, addr, flag, p)
    675 	dev_t dev;
    676 	u_long cmd;
    677 	caddr_t addr;
    678 	int flag;
    679 	struct proc *p;
    680 {
    681 	struct bpf_d *d = &bpf_dtab[minor(dev)];
    682 	int s, error = 0;
    683 #ifdef BPF_KERN_FILTER
    684 	struct bpf_insn **p;
    685 #endif
    686 
    687 	s = splnet();
    688 	if (d->bd_state == BPF_WAITING)
    689 		callout_stop(&d->bd_callout);
    690 	d->bd_state = BPF_IDLE;
    691 	splx(s);
    692 
    693 	switch (cmd) {
    694 
    695 	default:
    696 		error = EINVAL;
    697 		break;
    698 
    699 	/*
    700 	 * Check for read packet available.
    701 	 */
    702 	case FIONREAD:
    703 		{
    704 			int n;
    705 
    706 			s = splnet();
    707 			n = d->bd_slen;
    708 			if (d->bd_hbuf)
    709 				n += d->bd_hlen;
    710 			splx(s);
    711 
    712 			*(int *)addr = n;
    713 			break;
    714 		}
    715 
    716 	/*
    717 	 * Get buffer len [for read()].
    718 	 */
    719 	case BIOCGBLEN:
    720 		*(u_int *)addr = d->bd_bufsize;
    721 		break;
    722 
    723 	/*
    724 	 * Set buffer length.
    725 	 */
    726 	case BIOCSBLEN:
    727 		if (d->bd_bif != 0)
    728 			error = EINVAL;
    729 		else {
    730 			u_int size = *(u_int *)addr;
    731 
    732 			if (size > bpf_maxbufsize)
    733 				*(u_int *)addr = size = bpf_maxbufsize;
    734 			else if (size < BPF_MINBUFSIZE)
    735 				*(u_int *)addr = size = BPF_MINBUFSIZE;
    736 			d->bd_bufsize = size;
    737 		}
    738 		break;
    739 
    740 	/*
    741 	 * Set link layer read filter.
    742 	 */
    743 	case BIOCSETF:
    744 		error = bpf_setf(d, (struct bpf_program *)addr);
    745 		break;
    746 
    747 #ifdef BPF_KERN_FILTER
    748 	/*
    749 	 * Set TCP or UDP reject filter.
    750 	 */
    751 	case BIOCSTCPF:
    752 	case BIOCSUDPF:
    753 		if (!suser()) {
    754 			error = EPERM;
    755 			break;
    756 		}
    757 
    758 		/* Validate and store filter */
    759 		error = bpf_setf(d, (struct bpf_program *)addr);
    760 
    761 		/* Free possible old filter */
    762 		if (cmd == BIOCSTCPF)
    763 			p = &bpf_tcp_filter;
    764 		else
    765 			p = &bpf_udp_filter;
    766 		if (*p != NULL)
    767 			free((caddr_t)*p, M_DEVBUF);
    768 
    769 		/* Steal new filter (noop if error) */
    770 		s = splnet();
    771 		*p = d->bd_filter;
    772 		d->bd_filter = NULL;
    773 		splx(s);
    774 		break;
    775 #endif
    776 
    777 	/*
    778 	 * Flush read packet buffer.
    779 	 */
    780 	case BIOCFLUSH:
    781 		s = splnet();
    782 		reset_d(d);
    783 		splx(s);
    784 		break;
    785 
    786 	/*
    787 	 * Put interface into promiscuous mode.
    788 	 */
    789 	case BIOCPROMISC:
    790 		if (d->bd_bif == 0) {
    791 			/*
    792 			 * No interface attached yet.
    793 			 */
    794 			error = EINVAL;
    795 			break;
    796 		}
    797 		s = splnet();
    798 		if (d->bd_promisc == 0) {
    799 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
    800 			if (error == 0)
    801 				d->bd_promisc = 1;
    802 		}
    803 		splx(s);
    804 		break;
    805 
    806 	/*
    807 	 * Get device parameters.
    808 	 */
    809 	case BIOCGDLT:
    810 		if (d->bd_bif == 0)
    811 			error = EINVAL;
    812 		else
    813 			*(u_int *)addr = d->bd_bif->bif_dlt;
    814 		break;
    815 
    816 	/*
    817 	 * Get a list of supported device parameters.
    818 	 */
    819 	case BIOCGDLTLIST:
    820 		if (d->bd_bif == 0)
    821 			error = EINVAL;
    822 		else
    823 			error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
    824 		break;
    825 
    826 	/*
    827 	 * Set device parameters.
    828 	 */
    829 	case BIOCSDLT:
    830 		if (d->bd_bif == 0)
    831 			error = EINVAL;
    832 		else
    833 			error = bpf_setdlt(d, *(u_int *)addr);
    834 		break;
    835 
    836 	/*
    837 	 * Set interface name.
    838 	 */
    839 	case BIOCGETIF:
    840 		if (d->bd_bif == 0)
    841 			error = EINVAL;
    842 		else
    843 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
    844 		break;
    845 
    846 	/*
    847 	 * Set interface.
    848 	 */
    849 	case BIOCSETIF:
    850 		error = bpf_setif(d, (struct ifreq *)addr);
    851 		break;
    852 
    853 	/*
    854 	 * Set read timeout.
    855 	 */
    856 	case BIOCSRTIMEOUT:
    857 		{
    858 			struct timeval *tv = (struct timeval *)addr;
    859 
    860 			/* Compute number of ticks. */
    861 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
    862 			if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
    863 				d->bd_rtout = 1;
    864 			break;
    865 		}
    866 
    867 	/*
    868 	 * Get read timeout.
    869 	 */
    870 	case BIOCGRTIMEOUT:
    871 		{
    872 			struct timeval *tv = (struct timeval *)addr;
    873 
    874 			tv->tv_sec = d->bd_rtout / hz;
    875 			tv->tv_usec = (d->bd_rtout % hz) * tick;
    876 			break;
    877 		}
    878 
    879 	/*
    880 	 * Get packet stats.
    881 	 */
    882 	case BIOCGSTATS:
    883 		{
    884 			struct bpf_stat *bs = (struct bpf_stat *)addr;
    885 
    886 			bs->bs_recv = d->bd_rcount;
    887 			bs->bs_drop = d->bd_dcount;
    888 			bs->bs_capt = d->bd_ccount;
    889 			break;
    890 		}
    891 
    892 	case BIOCGSTATSOLD:
    893 		{
    894 			struct bpf_stat_old *bs = (struct bpf_stat_old *)addr;
    895 
    896 			bs->bs_recv = d->bd_rcount;
    897 			bs->bs_drop = d->bd_dcount;
    898 			break;
    899 		}
    900 
    901 	/*
    902 	 * Set immediate mode.
    903 	 */
    904 	case BIOCIMMEDIATE:
    905 		d->bd_immediate = *(u_int *)addr;
    906 		break;
    907 
    908 	case BIOCVERSION:
    909 		{
    910 			struct bpf_version *bv = (struct bpf_version *)addr;
    911 
    912 			bv->bv_major = BPF_MAJOR_VERSION;
    913 			bv->bv_minor = BPF_MINOR_VERSION;
    914 			break;
    915 		}
    916 
    917 	case BIOCGHDRCMPLT:	/* get "header already complete" flag */
    918 		*(u_int *)addr = d->bd_hdrcmplt;
    919 		break;
    920 
    921 	case BIOCSHDRCMPLT:	/* set "header already complete" flag */
    922 		d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
    923 		break;
    924 
    925 	/*
    926 	 * Get "see sent packets" flag
    927 	 */
    928 	case BIOCGSEESENT:
    929 		*(u_int *)addr = d->bd_seesent;
    930 		break;
    931 
    932 	/*
    933 	 * Set "see sent" packets flag
    934 	 */
    935 	case BIOCSSEESENT:
    936 		d->bd_seesent = *(u_int *)addr;
    937 		break;
    938 
    939 	case BIOCMMAPINFO:
    940 		error = bpf_mmapinfo(d, (struct bpf_mmapinfo *)addr);
    941 		break;
    942 
    943 	case FIONBIO:		/* Non-blocking I/O */
    944 		/*
    945 		 * No need to do anything special as we use IO_NDELAY in
    946 		 * bpfread() as an indication of whether or not to block
    947 		 * the read.
    948 		 */
    949 		break;
    950 
    951 	case FIOASYNC:		/* Send signal on receive packets */
    952 		d->bd_async = *(int *)addr;
    953 		break;
    954 
    955 	case TIOCSPGRP:		/* Process or group to send signals to */
    956 	case FIOSETOWN:
    957 		error = fsetown(p, &d->bd_pgid, cmd, addr);
    958 		break;
    959 
    960 	case TIOCGPGRP:
    961 	case FIOGETOWN:
    962 		error = fgetown(p, d->bd_pgid, cmd, addr);
    963 		break;
    964 	}
    965 	return (error);
    966 }
    967 
    968 /*
    969  * Set d's packet filter program to fp.  If this file already has a filter,
    970  * free it and replace it.  Returns EINVAL for bogus requests.
    971  */
    972 int
    973 bpf_setf(d, fp)
    974 	struct bpf_d *d;
    975 	struct bpf_program *fp;
    976 {
    977 	struct bpf_insn *fcode, *old;
    978 	u_int flen, size;
    979 	int s;
    980 
    981 	old = d->bd_filter;
    982 	if (fp->bf_insns == 0) {
    983 		if (fp->bf_len != 0)
    984 			return (EINVAL);
    985 		s = splnet();
    986 		d->bd_filter = 0;
    987 		reset_d(d);
    988 		splx(s);
    989 		if (old != 0)
    990 			free((caddr_t)old, M_DEVBUF);
    991 		return (0);
    992 	}
    993 	flen = fp->bf_len;
    994 	if (flen > BPF_MAXINSNS)
    995 		return (EINVAL);
    996 
    997 	size = flen * sizeof(*fp->bf_insns);
    998 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
    999 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
   1000 	    bpf_validate(fcode, (int)flen)) {
   1001 		s = splnet();
   1002 		d->bd_filter = fcode;
   1003 		reset_d(d);
   1004 		splx(s);
   1005 		if (old != 0)
   1006 			free((caddr_t)old, M_DEVBUF);
   1007 
   1008 		return (0);
   1009 	}
   1010 	free((caddr_t)fcode, M_DEVBUF);
   1011 	return (EINVAL);
   1012 }
   1013 
   1014 /*
   1015  * Detach a file from its current interface (if attached at all) and attach
   1016  * to the interface indicated by the name stored in ifr.
   1017  * Return an errno or 0.
   1018  */
   1019 static int
   1020 bpf_setif(d, ifr)
   1021 	struct bpf_d *d;
   1022 	struct ifreq *ifr;
   1023 {
   1024 	struct bpf_if *bp;
   1025 	char *cp;
   1026 	int unit_seen, i, s, error;
   1027 
   1028 	/*
   1029 	 * Make sure the provided name has a unit number, and default
   1030 	 * it to '0' if not specified.
   1031 	 * XXX This is ugly ... do this differently?
   1032 	 */
   1033 	unit_seen = 0;
   1034 	cp = ifr->ifr_name;
   1035 	cp[sizeof(ifr->ifr_name) - 1] = '\0';	/* sanity */
   1036 	while (*cp++)
   1037 		if (*cp >= '0' && *cp <= '9')
   1038 			unit_seen = 1;
   1039 	if (!unit_seen) {
   1040 		/* Make sure to leave room for the '\0'. */
   1041 		for (i = 0; i < (IFNAMSIZ - 1); ++i) {
   1042 			if ((ifr->ifr_name[i] >= 'a' &&
   1043 			     ifr->ifr_name[i] <= 'z') ||
   1044 			    (ifr->ifr_name[i] >= 'A' &&
   1045 			     ifr->ifr_name[i] <= 'Z'))
   1046 				continue;
   1047 			ifr->ifr_name[i] = '0';
   1048 		}
   1049 	}
   1050 
   1051 	/*
   1052 	 * Look through attached interfaces for the named one.
   1053 	 */
   1054 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
   1055 		struct ifnet *ifp = bp->bif_ifp;
   1056 
   1057 		if (ifp == 0 ||
   1058 		    strcmp(ifp->if_xname, ifr->ifr_name) != 0)
   1059 			continue;
   1060 		/* skip additional entry */
   1061 		if (bp->bif_driverp != (struct bpf_if **)&ifp->if_bpf)
   1062 			continue;
   1063 		/*
   1064 		 * We found the requested interface.
   1065 		 * If it's not up, return an error.
   1066 		 * Allocate the packet buffers if we need to.
   1067 		 * If we're already attached to requested interface,
   1068 		 * just flush the buffer.
   1069 		 */
   1070 		if ((ifp->if_flags & IFF_UP) == 0)
   1071 			return (ENETDOWN);
   1072 
   1073 		if (d->bd_sbuf == 0) {
   1074 			error = bpf_allocbufs(d);
   1075 			if (error != 0)
   1076 				return (error);
   1077 		}
   1078 		s = splnet();
   1079 		if (bp != d->bd_bif) {
   1080 			if (d->bd_bif)
   1081 				/*
   1082 				 * Detach if attached to something else.
   1083 				 */
   1084 				bpf_detachd(d);
   1085 
   1086 			bpf_attachd(d, bp);
   1087 		}
   1088 		reset_d(d);
   1089 		splx(s);
   1090 		return (0);
   1091 	}
   1092 	/* Not found. */
   1093 	return (ENXIO);
   1094 }
   1095 
   1096 /*
   1097  * Copy the interface name to the ifreq.
   1098  */
   1099 static void
   1100 bpf_ifname(ifp, ifr)
   1101 	struct ifnet *ifp;
   1102 	struct ifreq *ifr;
   1103 {
   1104 
   1105 	memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
   1106 }
   1107 
   1108 /*
   1109  * Support for poll() system call
   1110  *
   1111  * Return true iff the specific operation will not block indefinitely - with
   1112  * the assumption that it is safe to positively acknowledge a request for the
   1113  * ability to write to the BPF device.
   1114  * Otherwise, return false but make a note that a selwakeup() must be done.
   1115  */
   1116 int
   1117 bpfpoll(dev, events, p)
   1118 	dev_t dev;
   1119 	int events;
   1120 	struct proc *p;
   1121 {
   1122 	struct bpf_d *d = &bpf_dtab[minor(dev)];
   1123 	int s = splnet();
   1124 	int revents;
   1125 
   1126 	revents = events & (POLLOUT | POLLWRNORM);
   1127 	if (events & (POLLIN | POLLRDNORM)) {
   1128 		/*
   1129 		 * An imitation of the FIONREAD ioctl code.
   1130 		 */
   1131 		if ((d->bd_hlen != 0) ||
   1132 		    (d->bd_immediate && d->bd_slen != 0)) {
   1133 			revents |= events & (POLLIN | POLLRDNORM);
   1134 		} else if (d->bd_state == BPF_TIMED_OUT) {
   1135 			if (d->bd_slen != 0)
   1136 				revents |= events & (POLLIN | POLLRDNORM);
   1137 			else
   1138 				revents |= events & POLLIN;
   1139 		} else {
   1140 			selrecord(p, &d->bd_sel);
   1141 			/* Start the read timeout if necessary */
   1142 			if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
   1143 				callout_reset(&d->bd_callout, d->bd_rtout,
   1144 					      bpf_timed_out, d);
   1145 				d->bd_state = BPF_WAITING;
   1146 			}
   1147 		}
   1148 	}
   1149 
   1150 	splx(s);
   1151 	return (revents);
   1152 }
   1153 
   1154 static void
   1155 filt_bpfrdetach(struct knote *kn)
   1156 {
   1157 	struct bpf_d *d = kn->kn_hook;
   1158 	int s;
   1159 
   1160 	s = splnet();
   1161 	SLIST_REMOVE(&d->bd_sel.sel_klist, kn, knote, kn_selnext);
   1162 	splx(s);
   1163 }
   1164 
   1165 static int
   1166 filt_bpfread(struct knote *kn, long hint)
   1167 {
   1168 	struct bpf_d *d = kn->kn_hook;
   1169 
   1170 	kn->kn_data = d->bd_hlen;
   1171 	if (d->bd_immediate)
   1172 		kn->kn_data += d->bd_slen;
   1173 	return (kn->kn_data > 0);
   1174 }
   1175 
   1176 static const struct filterops bpfread_filtops =
   1177 	{ 1, NULL, filt_bpfrdetach, filt_bpfread };
   1178 
   1179 int
   1180 bpfkqfilter(dev, kn)
   1181 	dev_t dev;
   1182 	struct knote *kn;
   1183 {
   1184 	struct bpf_d *d = &bpf_dtab[minor(dev)];
   1185 	struct klist *klist;
   1186 	int s;
   1187 
   1188 	switch (kn->kn_filter) {
   1189 	case EVFILT_READ:
   1190 		klist = &d->bd_sel.sel_klist;
   1191 		kn->kn_fop = &bpfread_filtops;
   1192 		break;
   1193 
   1194 	default:
   1195 		return (1);
   1196 	}
   1197 
   1198 	kn->kn_hook = d;
   1199 
   1200 	s = splnet();
   1201 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
   1202 	splx(s);
   1203 
   1204 	return (0);
   1205 }
   1206 
   1207 /*
   1208  * Incoming linkage from device drivers.  Process the packet pkt, of length
   1209  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
   1210  * by each process' filter, and if accepted, stashed into the corresponding
   1211  * buffer.
   1212  */
   1213 void
   1214 bpf_tap(arg, pkt, pktlen)
   1215 	caddr_t arg;
   1216 	u_char *pkt;
   1217 	u_int pktlen;
   1218 {
   1219 	struct bpf_if *bp;
   1220 	struct bpf_d *d;
   1221 	u_int slen;
   1222 	/*
   1223 	 * Note that the ipl does not have to be raised at this point.
   1224 	 * The only problem that could arise here is that if two different
   1225 	 * interfaces shared any data.  This is not the case.
   1226 	 */
   1227 	bp = (struct bpf_if *)arg;
   1228 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
   1229 		++d->bd_rcount;
   1230 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
   1231 		if (slen != 0)
   1232 			catchpacket(d, pkt, pktlen, slen, memcpy);
   1233 	}
   1234 }
   1235 
   1236 /*
   1237  * Copy data from an mbuf chain into a buffer.  This code is derived
   1238  * from m_copydata in sys/uipc_mbuf.c.
   1239  */
   1240 static void *
   1241 bpf_mcpy(dst_arg, src_arg, len)
   1242 	void *dst_arg;
   1243 	const void *src_arg;
   1244 	size_t len;
   1245 {
   1246 	const struct mbuf *m;
   1247 	u_int count;
   1248 	u_char *dst;
   1249 
   1250 	m = src_arg;
   1251 	dst = dst_arg;
   1252 	while (len > 0) {
   1253 		if (m == 0)
   1254 			panic("bpf_mcpy");
   1255 		count = min(m->m_len, len);
   1256 		memcpy((caddr_t)dst, mtod(m, caddr_t), count);
   1257 		m = m->m_next;
   1258 		dst += count;
   1259 		len -= count;
   1260 	}
   1261 	return (dst_arg);
   1262 }
   1263 
   1264 /*
   1265  * Dispatch a packet to all the listeners on interface bp.
   1266  *
   1267  * marg    pointer to the packet, either a data buffer or an mbuf chain
   1268  * buflen  buffer length, if marg is a data buffer
   1269  * cpfn    a function that can copy marg into the listener's buffer
   1270  * pktlen  length of the packet
   1271  * rcvif   either NULL or the interface the packet came in on.
   1272  */
   1273 static __inline void
   1274 bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t),
   1275     void *marg, u_int pktlen, u_int buflen, struct ifnet *rcvif)
   1276 {
   1277 	u_int slen;
   1278 	struct bpf_d *d;
   1279 
   1280 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
   1281 		if (!d->bd_seesent && (rcvif == NULL))
   1282 			continue;
   1283 		++d->bd_rcount;
   1284 		slen = bpf_filter(d->bd_filter, marg, pktlen, buflen);
   1285 		if (slen != 0)
   1286 			catchpacket(d, marg, pktlen, slen, cpfn);
   1287 	}
   1288 }
   1289 
   1290 static __inline u_int
   1291 bpf_measure(struct mbuf *m)
   1292 {
   1293 	struct mbuf *m0;
   1294 	u_int pktlen;
   1295 
   1296 	pktlen = 0;
   1297 	for (m0 = m; m0 != 0; m0 = m0->m_next)
   1298 		pktlen += m0->m_len;
   1299 	return pktlen;
   1300 }
   1301 
   1302 /*
   1303  * Incoming linkage from device drivers, when the head of the packet is in
   1304  * a buffer, and the tail is in an mbuf chain.
   1305  */
   1306 void
   1307 bpf_mtap2(arg, data, dlen, m)
   1308 	caddr_t arg;
   1309 	void *data;
   1310 	u_int dlen;
   1311 	struct mbuf *m;
   1312 {
   1313 	struct bpf_if *bp = (struct bpf_if *)arg;
   1314 	u_int pktlen;
   1315 	struct mbuf mb;
   1316 
   1317 	pktlen = bpf_measure(m) + dlen;
   1318 
   1319 	/*
   1320 	 * Craft on-stack mbuf suitable for passing to bpf_filter.
   1321 	 * Note that we cut corners here; we only setup what's
   1322 	 * absolutely needed--this mbuf should never go anywhere else.
   1323 	 */
   1324 	mb.m_next = m;
   1325 	mb.m_data = data;
   1326 	mb.m_len = dlen;
   1327 
   1328 	bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, m->m_pkthdr.rcvif);
   1329 }
   1330 
   1331 /*
   1332  * Incoming linkage from device drivers, when packet is in an mbuf chain.
   1333  */
   1334 void
   1335 bpf_mtap(arg, m)
   1336 	caddr_t arg;
   1337 	struct mbuf *m;
   1338 {
   1339 	void *(*cpfn) __P((void *, const void *, size_t));
   1340 	struct bpf_if *bp = (struct bpf_if *)arg;
   1341 	u_int pktlen, buflen;
   1342 	void *marg;
   1343 
   1344 	pktlen = bpf_measure(m);
   1345 
   1346 	if (pktlen == m->m_len) {
   1347 		cpfn = memcpy;
   1348 		marg = mtod(m, void *);
   1349 		buflen = pktlen;
   1350 	} else {
   1351 		cpfn = bpf_mcpy;
   1352 		marg = m;
   1353 		buflen = 0;
   1354 	}
   1355 
   1356 	bpf_deliver(bp, cpfn, marg, pktlen, buflen, m->m_pkthdr.rcvif);
   1357 }
   1358 
   1359 /*
   1360  * Move the packet data from interface memory (pkt) into the
   1361  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
   1362  * otherwise 0.  "copy" is the routine called to do the actual data
   1363  * transfer.  memcpy is passed in to copy contiguous chunks, while
   1364  * bpf_mcpy is passed in to copy mbuf chains.  In the latter case,
   1365  * pkt is really an mbuf.
   1366  */
   1367 static void
   1368 catchpacket(d, pkt, pktlen, snaplen, cpfn)
   1369 	struct bpf_d *d;
   1370 	u_char *pkt;
   1371 	u_int pktlen, snaplen;
   1372 	void *(*cpfn) __P((void *, const void *, size_t));
   1373 {
   1374 	struct bpf_hdr *hp;
   1375 	int totlen, curlen;
   1376 	int hdrlen = d->bd_bif->bif_hdrlen;
   1377 
   1378 	++d->bd_ccount;
   1379 	/*
   1380 	 * Figure out how many bytes to move.  If the packet is
   1381 	 * greater or equal to the snapshot length, transfer that
   1382 	 * much.  Otherwise, transfer the whole packet (unless
   1383 	 * we hit the buffer size limit).
   1384 	 */
   1385 	totlen = hdrlen + min(snaplen, pktlen);
   1386 	if (totlen > d->bd_bufsize)
   1387 		totlen = d->bd_bufsize;
   1388 
   1389 	/*
   1390 	 * Round up the end of the previous packet to the next longword.
   1391 	 */
   1392 	curlen = BPF_WORDALIGN(d->bd_slen);
   1393 	if (curlen + totlen > d->bd_bufsize) {
   1394 		/*
   1395 		 * This packet will overflow the storage buffer.
   1396 		 * Rotate the buffers if we can, then wakeup any
   1397 		 * pending reads.
   1398 		 */
   1399 		if (d->bd_fbuf == 0 || d->bd_mapbuf != -1) {
   1400 			/*
   1401 			 * We haven't completed the previous read yet,
   1402 			 * so drop the packet.
   1403 			 */
   1404 			++d->bd_dcount;
   1405 			return;
   1406 		}
   1407 		ROTATE_BUFFERS(d);
   1408 		bpf_wakeup(d);
   1409 		curlen = 0;
   1410 	}
   1411 
   1412 	/*
   1413 	 * Append the bpf header.
   1414 	 */
   1415 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
   1416 	microtime(&hp->bh_tstamp);
   1417 	hp->bh_datalen = pktlen;
   1418 	hp->bh_hdrlen = hdrlen;
   1419 	/*
   1420 	 * Copy the packet data into the store buffer and update its length.
   1421 	 */
   1422 	(*cpfn)((u_char *)hp + hdrlen, pkt, (hp->bh_caplen = totlen - hdrlen));
   1423 	d->bd_slen = curlen + totlen;
   1424 
   1425 	/*
   1426 	 * Call bpf_wakeup after bd_slen has been updated so that kevent(2)
   1427 	 * will cause filt_bpfread() to be called with it adjusted.
   1428 	 */
   1429 	if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
   1430 		/*
   1431 		 * Immediate mode is set, or the read timeout has
   1432 		 * already expired during a select call.  A packet
   1433 		 * arrived, so the reader should be woken up.
   1434 		 */
   1435 		bpf_wakeup(d);
   1436 }
   1437 
   1438 /*
   1439  * Initialize all nonzero fields of a descriptor.
   1440  */
   1441 static int
   1442 bpf_allocbufs(d)
   1443 	struct bpf_d *d;
   1444 {
   1445 
   1446 	d->bd_bufs[1] = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
   1447 	if (d->bd_bufs[1] == NULL)
   1448 		return (ENOBUFS);
   1449 	d->bd_fbuf = d->bd_bufs[1];
   1450 	d->bd_bufs[0] = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_NOWAIT);
   1451 	if (d->bd_bufs[0] == NULL) {
   1452 		free(d->bd_bufs[1], M_DEVBUF);
   1453 		d->bd_bufs[1] = NULL;
   1454 		d->bd_fbuf = NULL;
   1455 		return (ENOBUFS);
   1456 	}
   1457 	d->bd_sbuf = d->bd_bufs[0];
   1458 	d->bd_fbuf = d->bd_bufs[1];
   1459 	d->bd_slen = 0;
   1460 	d->bd_hlen = 0;
   1461 	d->bd_mapbuf = -1;
   1462 	return (0);
   1463 }
   1464 
   1465 /*
   1466  * Free buffers currently in use by a descriptor.
   1467  * Called on close.
   1468  */
   1469 static void
   1470 bpf_freed(d)
   1471 	struct bpf_d *d;
   1472 {
   1473 	/*
   1474 	 * We don't need to lock out interrupts since this descriptor has
   1475 	 * been detached from its interface and it yet hasn't been marked
   1476 	 * free.
   1477 	 */
   1478 	if (d->bd_bufs[0] != 0)
   1479 		free(d->bd_bufs[0], M_DEVBUF);
   1480 	if (d->bd_bufs[1] != 0)
   1481 		free(d->bd_bufs[1], M_DEVBUF);
   1482 	if (d->bd_filter)
   1483 		free((caddr_t)d->bd_filter, M_DEVBUF);
   1484 
   1485 	D_MARKFREE(d);
   1486 }
   1487 
   1488 /*
   1489  * Attach an interface to bpf.  dlt is the link layer type; hdrlen is the
   1490  * fixed size of the link header (variable length headers not yet supported).
   1491  */
   1492 void
   1493 bpfattach(ifp, dlt, hdrlen)
   1494 	struct ifnet *ifp;
   1495 	u_int dlt, hdrlen;
   1496 {
   1497 
   1498 	bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
   1499 }
   1500 
   1501 /*
   1502  * Attach additional dlt for a interface to bpf.  dlt is the link layer type;
   1503  * hdrlen is the fixed size of the link header for the specified dlt
   1504  * (variable length headers not yet supported).
   1505  */
   1506 void
   1507 bpfattach2(ifp, dlt, hdrlen, driverp)
   1508 	struct ifnet *ifp;
   1509 	u_int dlt, hdrlen;
   1510 	caddr_t *driverp;
   1511 {
   1512 	struct bpf_if *bp;
   1513 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
   1514 	if (bp == 0)
   1515 		panic("bpfattach");
   1516 
   1517 	bp->bif_dlist = 0;
   1518 	bp->bif_driverp = (struct bpf_if **)driverp;
   1519 	bp->bif_ifp = ifp;
   1520 	bp->bif_dlt = dlt;
   1521 
   1522 	bp->bif_next = bpf_iflist;
   1523 	bpf_iflist = bp;
   1524 
   1525 	*bp->bif_driverp = 0;
   1526 
   1527 	/*
   1528 	 * Compute the length of the bpf header.  This is not necessarily
   1529 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
   1530 	 * that the network layer header begins on a longword boundary (for
   1531 	 * performance reasons and to alleviate alignment restrictions).
   1532 	 */
   1533 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
   1534 
   1535 #if 0
   1536 	printf("bpf: %s attached\n", ifp->if_xname);
   1537 #endif
   1538 }
   1539 
   1540 /*
   1541  * Remove an interface from bpf.
   1542  */
   1543 void
   1544 bpfdetach(ifp)
   1545 	struct ifnet *ifp;
   1546 {
   1547 	struct bpf_if *bp, **pbp;
   1548 	struct bpf_d *d;
   1549 	int i, s, cmaj;
   1550 
   1551 	/* locate the major number */
   1552 	cmaj = cdevsw_lookup_major(&bpf_cdevsw);
   1553 
   1554 	/* Nuke the vnodes for any open instances */
   1555 	for (i = 0; i < NBPFILTER; ++i) {
   1556 		d = &bpf_dtab[i];
   1557 		if (!D_ISFREE(d) && d->bd_bif != NULL &&
   1558 		    d->bd_bif->bif_ifp == ifp) {
   1559 			/*
   1560 			 * Detach the descriptor from an interface now.
   1561 			 * It will be free'ed later by close routine.
   1562 			 */
   1563 			s = splnet();
   1564 			d->bd_promisc = 0;	/* we can't touch device. */
   1565 			bpf_detachd(d);
   1566 			splx(s);
   1567 			vdevgone(cmaj, i, i, VCHR);
   1568 		}
   1569 	}
   1570 
   1571   again:
   1572 	for (bp = bpf_iflist, pbp = &bpf_iflist;
   1573 	     bp != NULL; pbp = &bp->bif_next, bp = bp->bif_next) {
   1574 		if (bp->bif_ifp == ifp) {
   1575 			*pbp = bp->bif_next;
   1576 			free(bp, M_DEVBUF);
   1577 			goto again;
   1578 		}
   1579 	}
   1580 }
   1581 
   1582 /*
   1583  * Change the data link type of a interface.
   1584  */
   1585 void
   1586 bpf_change_type(ifp, dlt, hdrlen)
   1587 	struct ifnet *ifp;
   1588 	u_int dlt, hdrlen;
   1589 {
   1590 	struct bpf_if *bp;
   1591 
   1592 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
   1593 		if (bp->bif_driverp == (struct bpf_if **)&ifp->if_bpf)
   1594 			break;
   1595 	}
   1596 	if (bp == NULL)
   1597 		panic("bpf_change_type");
   1598 
   1599 	bp->bif_dlt = dlt;
   1600 
   1601 	/*
   1602 	 * Compute the length of the bpf header.  This is not necessarily
   1603 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
   1604 	 * that the network layer header begins on a longword boundary (for
   1605 	 * performance reasons and to alleviate alignment restrictions).
   1606 	 */
   1607 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
   1608 }
   1609 
   1610 /*
   1611  * Get a list of available data link type of the interface.
   1612  */
   1613 static int
   1614 bpf_getdltlist(d, bfl)
   1615 	struct bpf_d *d;
   1616 	struct bpf_dltlist *bfl;
   1617 {
   1618 	int n, error;
   1619 	struct ifnet *ifp;
   1620 	struct bpf_if *bp;
   1621 
   1622 	ifp = d->bd_bif->bif_ifp;
   1623 	n = 0;
   1624 	error = 0;
   1625 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
   1626 		if (bp->bif_ifp != ifp)
   1627 			continue;
   1628 		if (bfl->bfl_list != NULL) {
   1629 			if (n >= bfl->bfl_len)
   1630 				return ENOMEM;
   1631 			error = copyout(&bp->bif_dlt,
   1632 			    bfl->bfl_list + n, sizeof(u_int));
   1633 		}
   1634 		n++;
   1635 	}
   1636 	bfl->bfl_len = n;
   1637 	return error;
   1638 }
   1639 
   1640 /*
   1641  * Set the data link type of a BPF instance.
   1642  */
   1643 static int
   1644 bpf_setdlt(d, dlt)
   1645 	struct bpf_d *d;
   1646 	u_int dlt;
   1647 {
   1648 	int s, error, opromisc;
   1649 	struct ifnet *ifp;
   1650 	struct bpf_if *bp;
   1651 
   1652 	if (d->bd_bif->bif_dlt == dlt)
   1653 		return 0;
   1654 	ifp = d->bd_bif->bif_ifp;
   1655 	for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) {
   1656 		if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
   1657 			break;
   1658 	}
   1659 	if (bp == NULL)
   1660 		return EINVAL;
   1661 
   1662 /*
   1663  * Provide a mmap(2) interface to the BPF buffers.
   1664  * Read-only mapping (PROT_READ) is enforced by this driver - an application
   1665  * using this should never write to this buffer and especially not with copy
   1666  * on write as the real buffer contents will then disappear from the process
   1667  * view.
   1668  *
   1669  * An application should create two maps: one for each buffer that bpf has
   1670  * internally and use the information returned from BIOCMMAPINFO to determine
   1671  * which one has valid data in it and how much data is valid.
   1672  */
   1673 paddr_t
   1674 bpfmmap(dev_t dev, off_t off, int prot)
   1675 {
   1676 	struct bpf_d *d;
   1677 	u_int uoff;
   1678 
   1679 	if (prot != VM_PROT_READ)
   1680 		return -1;
   1681 
   1682 	if (off & PAGE_MASK)
   1683 		panic("bpfmmap");
   1684 
   1685 	d = &bpf_dtab[minor(dev)];
   1686 	uoff = (u_int)off;
   1687 
   1688 	if (uoff >= 0 && uoff < d->bd_bufsize)
   1689 		return (atop(d->bd_bufs[0] + uoff));
   1690 
   1691 	if (uoff >= d->bd_bufsize && uoff < (d->bd_bufsize * 2))
   1692 		return (atop(d->bd_bufs[1] + (uoff - d->bd_bufsize)));
   1693 
   1694 	/* Page not found. */
   1695 	return (-1);
   1696 }
   1697 
   1698 static int
   1699 bpf_mmapinfo(d, info)
   1700 	struct bpf_d *d;
   1701 	struct bpf_mmapinfo *info;
   1702 {
   1703 	int which, s, error;
   1704 
   1705 	s = splnet();
   1706 
   1707 	if (info->bpm_op == BPM_RELEASE) {	/* only want to unlock */
   1708 		d->bd_mapbuf = -1;
   1709 		splx(s);
   1710 		return 0;
   1711 	}
   1712 
   1713 	/*
   1714 	 * Currently only two operations are supported, release and acquire.
   1715 	 * If it's not one of these then return an error.
   1716 	 */
   1717 	if (info->bpm_op != BPM_ACQUIRE) {
   1718 		splx(s);
   1719 		return EINVAL;
   1720 	}
   1721 
   1722 	/*
   1723 	 * An incoming call must give up the current buffer locked for use
   1724 	 * with mmap, if it has one, so that bpf has somewhere to write new
   1725 	 * data when this call returns.
   1726 	 */
   1727 	if (d->bd_mapbuf != -1) {
   1728 		d->bd_fbuf = d->bd_hbuf;
   1729 		d->bd_hbuf = NULL;
   1730 		d->bd_hlen = 0;
   1731 		d->bd_mapbuf = -1;
   1732 	}
   1733 
   1734 	error = bpf_waitfordata(d);
   1735 	if (error == 0) {
   1736 		if (d->bd_hbuf == d->bd_bufs[0])
   1737 			which = 0;
   1738 		else if (d->bd_hbuf == d->bd_bufs[1])
   1739 			which = 1;
   1740 		else
   1741 			which = -1;
   1742 		d->bd_mapbuf = which;
   1743 		info->bpm_len = d->bd_hlen;
   1744 		info->bpm_which = which;
   1745 	}
   1746 	splx(s);
   1747 	return 0;
   1748 }
   1749 	s = splnet();
   1750 	opromisc = d->bd_promisc;
   1751 	bpf_detachd(d);
   1752 	bpf_attachd(d, bp);
   1753 	reset_d(d);
   1754 	if (opromisc) {
   1755 		error = ifpromisc(bp->bif_ifp, 1);
   1756 		if (error)
   1757 			printf("%s: bpf_setdlt: ifpromisc failed (%d)\n",
   1758 			    bp->bif_ifp->if_xname, error);
   1759 		else
   1760 			d->bd_promisc = 1;
   1761 	}
   1762 	splx(s);
   1763 	return 0;
   1764 }
   1765 
   1766 static int
   1767 sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS)
   1768 {
   1769 	int newsize, error;
   1770 	struct sysctlnode node;
   1771 
   1772 	node = *rnode;
   1773 	node.sysctl_data = &newsize;
   1774 	newsize = bpf_maxbufsize;
   1775 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
   1776 	if (error || newp == NULL)
   1777 		return (error);
   1778 
   1779 	if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE)
   1780 		return (EINVAL);
   1781 
   1782 	bpf_maxbufsize = newsize;
   1783 
   1784 	return (0);
   1785 }
   1786 
   1787 SYSCTL_SETUP(sysctl_net_bfp_setup, "sysctl net.bpf subtree setup")
   1788 {
   1789 	struct sysctlnode *node;
   1790 
   1791 	sysctl_createv(clog, 0, NULL, NULL,
   1792 		       CTLFLAG_PERMANENT,
   1793 		       CTLTYPE_NODE, "net", NULL,
   1794 		       NULL, 0, NULL, 0,
   1795 		       CTL_NET, CTL_EOL);
   1796 
   1797 	node = NULL;
   1798 	sysctl_createv(clog, 0, NULL, &node,
   1799 		       CTLFLAG_PERMANENT,
   1800 		       CTLTYPE_NODE, "bpf",
   1801 		       SYSCTL_DESCR("BPF options"),
   1802 		       NULL, 0, NULL, 0,
   1803 		       CTL_NET, CTL_CREATE, CTL_EOL);
   1804 	if (node != NULL)
   1805 		sysctl_createv(clog, 0, NULL, NULL,
   1806 			CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
   1807 			CTLTYPE_INT, "maxbufsize",
   1808 			SYSCTL_DESCR("Maximum size for data capture buffer"),
   1809 			sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0,
   1810 			CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL);
   1811 }
   1812 
   1813 
   1814