Home | History | Annotate | Line # | Download | only in net
bpf.c revision 1.19
      1 /*	$NetBSD: bpf.c,v 1.19 1995/04/22 13:26:20 cgd 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. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)bpf.c	8.2 (Berkeley) 3/28/94
     41  */
     42 
     43 #include "bpfilter.h"
     44 
     45 #include <sys/param.h>
     46 #include <sys/systm.h>
     47 #include <sys/mbuf.h>
     48 #include <sys/buf.h>
     49 #include <sys/time.h>
     50 #include <sys/proc.h>
     51 #include <sys/user.h>
     52 #include <sys/ioctl.h>
     53 #include <sys/map.h>
     54 
     55 #include <sys/file.h>
     56 #if defined(sparc) && BSD < 199103
     57 #include <sys/stream.h>
     58 #endif
     59 #include <sys/tty.h>
     60 #include <sys/uio.h>
     61 
     62 #include <sys/protosw.h>
     63 #include <sys/socket.h>
     64 #include <net/if.h>
     65 
     66 #include <net/bpf.h>
     67 #include <net/bpfdesc.h>
     68 
     69 #include <sys/errno.h>
     70 
     71 #include <netinet/in.h>
     72 #include <netinet/if_arc.h>
     73 #include <netinet/if_ether.h>
     74 #include <sys/kernel.h>
     75 
     76 /*
     77  * Older BSDs don't have kernel malloc.
     78  */
     79 #if BSD < 199103
     80 extern bcopy();
     81 static caddr_t bpf_alloc();
     82 #include <net/bpf_compat.h>
     83 #define BPF_BUFSIZE (MCLBYTES-8)
     84 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, code, uio)
     85 #else
     86 #define BPF_BUFSIZE 4096
     87 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, uio)
     88 #endif
     89 
     90 #define PRINET  26			/* interruptible */
     91 
     92 /*
     93  * The default read buffer size is patchable.
     94  */
     95 int bpf_bufsize = BPF_BUFSIZE;
     96 
     97 /*
     98  *  bpf_iflist is the list of interfaces; each corresponds to an ifnet
     99  *  bpf_dtab holds the descriptors, indexed by minor device #
    100  */
    101 struct bpf_if	*bpf_iflist;
    102 struct bpf_d	bpf_dtab[NBPFILTER];
    103 
    104 #if BSD >= 199207 || NetBSD0_9 >= 2
    105 /*
    106  * bpfilterattach() is called at boot time in new systems.  We do
    107  * nothing here since old systems will not call this.
    108  */
    109 /* ARGSUSED */
    110 void
    111 bpfilterattach(n)
    112 	int n;
    113 {
    114 }
    115 #endif
    116 
    117 static int	bpf_allocbufs __P((struct bpf_d *));
    118 static int	bpf_allocbufs __P((struct bpf_d *));
    119 static void	bpf_freed __P((struct bpf_d *));
    120 static void	bpf_freed __P((struct bpf_d *));
    121 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
    122 static void	bpf_ifname __P((struct ifnet *, struct ifreq *));
    123 static void	bpf_mcopy __P((const void *, void *, size_t));
    124 static int	bpf_movein __P((struct uio *, int,
    125 		    struct mbuf **, struct sockaddr *, int *));
    126 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
    127 static int	bpf_setif __P((struct bpf_d *, struct ifreq *));
    128 static __inline void
    129 		bpf_wakeup __P((struct bpf_d *));
    130 static void	catchpacket __P((struct bpf_d *, u_char *, size_t,
    131 		    size_t, void (*)(const void *, void *, size_t)));
    132 static void	reset_d __P((struct bpf_d *));
    133 
    134 static int
    135 bpf_movein(uio, linktype, mp, sockp, datlen)
    136 	register struct uio *uio;
    137 	int linktype, *datlen;
    138 	register struct mbuf **mp;
    139 	register struct sockaddr *sockp;
    140 {
    141 	struct mbuf *m;
    142 	int error;
    143 	int len;
    144 	int hlen;
    145 
    146 	/*
    147 	 * Build a sockaddr based on the data link layer type.
    148 	 * We do this at this level because the ethernet header
    149 	 * is copied directly into the data field of the sockaddr.
    150 	 * In the case of SLIP, there is no header and the packet
    151 	 * is forwarded as is.
    152 	 * Also, we are careful to leave room at the front of the mbuf
    153 	 * for the link level header.
    154 	 */
    155 	switch (linktype) {
    156 
    157 	case DLT_SLIP:
    158 		sockp->sa_family = AF_INET;
    159 		hlen = 0;
    160 		break;
    161 
    162 	case DLT_PPP:
    163 		sockp->sa_family = AF_UNSPEC;
    164 		hlen = 0;
    165 		break;
    166 
    167 	case DLT_EN10MB:
    168 		sockp->sa_family = AF_UNSPEC;
    169 		/* XXX Would MAXLINKHDR be better? */
    170 		hlen = sizeof(struct ether_header);
    171 		break;
    172 
    173 	case DLT_ARCNET:
    174 		sockp->sa_family = AF_UNSPEC;
    175 		hlen = ARC_HDRLEN;
    176 		break;
    177 
    178 	case DLT_FDDI:
    179 		sockp->sa_family = AF_UNSPEC;
    180 		/* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
    181 		hlen = 24;
    182 		break;
    183 
    184 	case DLT_NULL:
    185 		sockp->sa_family = AF_UNSPEC;
    186 		hlen = 0;
    187 		break;
    188 
    189 	default:
    190 		return (EIO);
    191 	}
    192 
    193 	len = uio->uio_resid;
    194 	*datlen = len - hlen;
    195 	if ((unsigned)len > MCLBYTES)
    196 		return (EIO);
    197 
    198 	MGET(m, M_WAIT, MT_DATA);
    199 	if (m == 0)
    200 		return (ENOBUFS);
    201 	if (len > MLEN) {
    202 #if BSD >= 199103
    203 		MCLGET(m, M_WAIT);
    204 		if ((m->m_flags & M_EXT) == 0) {
    205 #else
    206 		MCLGET(m);
    207 		if (m->m_len != MCLBYTES) {
    208 #endif
    209 			error = ENOBUFS;
    210 			goto bad;
    211 		}
    212 	}
    213 	m->m_len = len;
    214 	*mp = m;
    215 	/*
    216 	 * Make room for link header.
    217 	 */
    218 	if (hlen != 0) {
    219 		m->m_len -= hlen;
    220 #if BSD >= 199103
    221 		m->m_data += hlen; /* XXX */
    222 #else
    223 		m->m_off += hlen;
    224 #endif
    225 		error = UIOMOVE((caddr_t)sockp->sa_data, hlen, UIO_WRITE, uio);
    226 		if (error)
    227 			goto bad;
    228 	}
    229 	error = UIOMOVE(mtod(m, caddr_t), len - hlen, UIO_WRITE, uio);
    230 	if (!error)
    231 		return (0);
    232  bad:
    233 	m_freem(m);
    234 	return (error);
    235 }
    236 
    237 /*
    238  * Attach file to the bpf interface, i.e. make d listen on bp.
    239  * Must be called at splimp.
    240  */
    241 static void
    242 bpf_attachd(d, bp)
    243 	struct bpf_d *d;
    244 	struct bpf_if *bp;
    245 {
    246 	/*
    247 	 * Point d at bp, and add d to the interface's list of listeners.
    248 	 * Finally, point the driver's bpf cookie at the interface so
    249 	 * it will divert packets to bpf.
    250 	 */
    251 	d->bd_bif = bp;
    252 	d->bd_next = bp->bif_dlist;
    253 	bp->bif_dlist = d;
    254 
    255 	*bp->bif_driverp = bp;
    256 }
    257 
    258 /*
    259  * Detach a file from its interface.
    260  */
    261 static void
    262 bpf_detachd(d)
    263 	struct bpf_d *d;
    264 {
    265 	struct bpf_d **p;
    266 	struct bpf_if *bp;
    267 
    268 	bp = d->bd_bif;
    269 	/*
    270 	 * Check if this descriptor had requested promiscuous mode.
    271 	 * If so, turn it off.
    272 	 */
    273 	if (d->bd_promisc) {
    274 		int error;
    275 
    276 		d->bd_promisc = 0;
    277 		error = ifpromisc(bp->bif_ifp, 0);
    278 		if (error && error != EINVAL)
    279 			/*
    280 			 * Something is really wrong if we were able to put
    281 			 * the driver into promiscuous mode, but can't
    282 			 * take it out.
    283 			 */
    284 			panic("bpf: ifpromisc failed");
    285 	}
    286 	/* Remove d from the interface's descriptor list. */
    287 	p = &bp->bif_dlist;
    288 	while (*p != d) {
    289 		p = &(*p)->bd_next;
    290 		if (*p == 0)
    291 			panic("bpf_detachd: descriptor not in list");
    292 	}
    293 	*p = (*p)->bd_next;
    294 	if (bp->bif_dlist == 0)
    295 		/*
    296 		 * Let the driver know that there are no more listeners.
    297 		 */
    298 		*d->bd_bif->bif_driverp = 0;
    299 	d->bd_bif = 0;
    300 }
    301 
    302 
    303 /*
    304  * Mark a descriptor free by making it point to itself.
    305  * This is probably cheaper than marking with a constant since
    306  * the address should be in a register anyway.
    307  */
    308 #define D_ISFREE(d) ((d) == (d)->bd_next)
    309 #define D_MARKFREE(d) ((d)->bd_next = (d))
    310 #define D_MARKUSED(d) ((d)->bd_next = 0)
    311 
    312 /*
    313  * Open ethernet device.  Returns ENXIO for illegal minor device number,
    314  * EBUSY if file is open by another process.
    315  */
    316 /* ARGSUSED */
    317 int
    318 bpfopen(dev, flag)
    319 	dev_t dev;
    320 	int flag;
    321 {
    322 	register struct bpf_d *d;
    323 
    324 	if (minor(dev) >= NBPFILTER)
    325 		return (ENXIO);
    326 	/*
    327 	 * Each minor can be opened by only one process.  If the requested
    328 	 * minor is in use, return EBUSY.
    329 	 */
    330 	d = &bpf_dtab[minor(dev)];
    331 	if (!D_ISFREE(d))
    332 		return (EBUSY);
    333 
    334 	/* Mark "free" and do most initialization. */
    335 	bzero((char *)d, sizeof(*d));
    336 	d->bd_bufsize = bpf_bufsize;
    337 
    338 	return (0);
    339 }
    340 
    341 /*
    342  * Close the descriptor by detaching it from its interface,
    343  * deallocating its buffers, and marking it free.
    344  */
    345 /* ARGSUSED */
    346 int
    347 bpfclose(dev, flag)
    348 	dev_t dev;
    349 	int flag;
    350 {
    351 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    352 	register int s;
    353 
    354 	s = splimp();
    355 	if (d->bd_bif)
    356 		bpf_detachd(d);
    357 	splx(s);
    358 	bpf_freed(d);
    359 
    360 	return (0);
    361 }
    362 
    363 /*
    364  * Support for SunOS, which does not have tsleep.
    365  */
    366 #if BSD < 199103
    367 static
    368 bpf_timeout(arg)
    369 	caddr_t arg;
    370 {
    371 	struct bpf_d *d = (struct bpf_d *)arg;
    372 	d->bd_timedout = 1;
    373 	wakeup(arg);
    374 }
    375 
    376 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
    377 
    378 int
    379 bpf_sleep(d)
    380 	register struct bpf_d *d;
    381 {
    382 	register int rto = d->bd_rtout;
    383 	register int st;
    384 
    385 	if (rto != 0) {
    386 		d->bd_timedout = 0;
    387 		timeout(bpf_timeout, (caddr_t)d, rto);
    388 	}
    389 	st = sleep((caddr_t)d, PRINET|PCATCH);
    390 	if (rto != 0) {
    391 		if (d->bd_timedout == 0)
    392 			untimeout(bpf_timeout, (caddr_t)d);
    393 		else if (st == 0)
    394 			return EWOULDBLOCK;
    395 	}
    396 	return (st != 0) ? EINTR : 0;
    397 }
    398 #else
    399 #define BPF_SLEEP tsleep
    400 #endif
    401 
    402 /*
    403  * Rotate the packet buffers in descriptor d.  Move the store buffer
    404  * into the hold slot, and the free buffer into the store slot.
    405  * Zero the length of the new store buffer.
    406  */
    407 #define ROTATE_BUFFERS(d) \
    408 	(d)->bd_hbuf = (d)->bd_sbuf; \
    409 	(d)->bd_hlen = (d)->bd_slen; \
    410 	(d)->bd_sbuf = (d)->bd_fbuf; \
    411 	(d)->bd_slen = 0; \
    412 	(d)->bd_fbuf = 0;
    413 /*
    414  *  bpfread - read next chunk of packets from buffers
    415  */
    416 int
    417 bpfread(dev, uio)
    418 	dev_t dev;
    419 	register struct uio *uio;
    420 {
    421 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    422 	int error;
    423 	int s;
    424 
    425 	/*
    426 	 * Restrict application to use a buffer the same size as
    427 	 * as kernel buffers.
    428 	 */
    429 	if (uio->uio_resid != d->bd_bufsize)
    430 		return (EINVAL);
    431 
    432 	s = splimp();
    433 	/*
    434 	 * If the hold buffer is empty, then do a timed sleep, which
    435 	 * ends when the timeout expires or when enough packets
    436 	 * have arrived to fill the store buffer.
    437 	 */
    438 	while (d->bd_hbuf == 0) {
    439 		if (d->bd_immediate && d->bd_slen != 0) {
    440 			/*
    441 			 * A packet(s) either arrived since the previous
    442 			 * read or arrived while we were asleep.
    443 			 * Rotate the buffers and return what's here.
    444 			 */
    445 			ROTATE_BUFFERS(d);
    446 			break;
    447 		}
    448 		error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
    449 				  d->bd_rtout);
    450 		if (error == EINTR || error == ERESTART) {
    451 			splx(s);
    452 			return (error);
    453 		}
    454 		if (error == EWOULDBLOCK) {
    455 			/*
    456 			 * On a timeout, return what's in the buffer,
    457 			 * which may be nothing.  If there is something
    458 			 * in the store buffer, we can rotate the buffers.
    459 			 */
    460 			if (d->bd_hbuf)
    461 				/*
    462 				 * We filled up the buffer in between
    463 				 * getting the timeout and arriving
    464 				 * here, so we don't need to rotate.
    465 				 */
    466 				break;
    467 
    468 			if (d->bd_slen == 0) {
    469 				splx(s);
    470 				return (0);
    471 			}
    472 			ROTATE_BUFFERS(d);
    473 			break;
    474 		}
    475 	}
    476 	/*
    477 	 * At this point, we know we have something in the hold slot.
    478 	 */
    479 	splx(s);
    480 
    481 	/*
    482 	 * Move data from hold buffer into user space.
    483 	 * We know the entire buffer is transferred since
    484 	 * we checked above that the read buffer is bpf_bufsize bytes.
    485 	 */
    486 	error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
    487 
    488 	s = splimp();
    489 	d->bd_fbuf = d->bd_hbuf;
    490 	d->bd_hbuf = 0;
    491 	d->bd_hlen = 0;
    492 	splx(s);
    493 
    494 	return (error);
    495 }
    496 
    497 
    498 /*
    499  * If there are processes sleeping on this descriptor, wake them up.
    500  */
    501 static __inline void
    502 bpf_wakeup(d)
    503 	register struct bpf_d *d;
    504 {
    505 	wakeup((caddr_t)d);
    506 #if BSD >= 199103
    507 	selwakeup(&d->bd_sel);
    508 	/* XXX */
    509 	d->bd_sel.si_pid = 0;
    510 #else
    511 	if (d->bd_selproc) {
    512 		selwakeup(d->bd_selproc, (int)d->bd_selcoll);
    513 		d->bd_selcoll = 0;
    514 		d->bd_selproc = 0;
    515 	}
    516 #endif
    517 }
    518 
    519 int
    520 bpfwrite(dev, uio)
    521 	dev_t dev;
    522 	struct uio *uio;
    523 {
    524 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    525 	struct ifnet *ifp;
    526 	struct mbuf *m;
    527 	int error, s;
    528 	static struct sockaddr dst;
    529 	int datlen;
    530 
    531 	if (d->bd_bif == 0)
    532 		return (ENXIO);
    533 
    534 	ifp = d->bd_bif->bif_ifp;
    535 
    536 	if (uio->uio_resid == 0)
    537 		return (0);
    538 
    539 	error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen);
    540 	if (error)
    541 		return (error);
    542 
    543 	if (datlen > ifp->if_mtu)
    544 		return (EMSGSIZE);
    545 
    546 	s = splnet();
    547 #if BSD >= 199103
    548 	error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
    549 #else
    550 	error = (*ifp->if_output)(ifp, m, &dst);
    551 #endif
    552 	splx(s);
    553 	/*
    554 	 * The driver frees the mbuf.
    555 	 */
    556 	return (error);
    557 }
    558 
    559 /*
    560  * Reset a descriptor by flushing its packet buffer and clearing the
    561  * receive and drop counts.  Should be called at splimp.
    562  */
    563 static void
    564 reset_d(d)
    565 	struct bpf_d *d;
    566 {
    567 	if (d->bd_hbuf) {
    568 		/* Free the hold buffer. */
    569 		d->bd_fbuf = d->bd_hbuf;
    570 		d->bd_hbuf = 0;
    571 	}
    572 	d->bd_slen = 0;
    573 	d->bd_hlen = 0;
    574 	d->bd_rcount = 0;
    575 	d->bd_dcount = 0;
    576 }
    577 
    578 /*
    579  *  FIONREAD		Check for read packet available.
    580  *  SIOCGIFADDR		Get interface address - convenient hook to driver.
    581  *  BIOCGBLEN		Get buffer len [for read()].
    582  *  BIOCSETF		Set ethernet read filter.
    583  *  BIOCFLUSH		Flush read packet buffer.
    584  *  BIOCPROMISC		Put interface into promiscuous mode.
    585  *  BIOCGDLT		Get link layer type.
    586  *  BIOCGETIF		Get interface name.
    587  *  BIOCSETIF		Set interface.
    588  *  BIOCSRTIMEOUT	Set read timeout.
    589  *  BIOCGRTIMEOUT	Get read timeout.
    590  *  BIOCGSTATS		Get packet stats.
    591  *  BIOCIMMEDIATE	Set immediate mode.
    592  *  BIOCVERSION		Get filter language version.
    593  */
    594 /* ARGSUSED */
    595 int
    596 bpfioctl(dev, cmd, addr, flag)
    597 	dev_t dev;
    598 	u_long cmd;
    599 	caddr_t addr;
    600 	int flag;
    601 {
    602 	register struct bpf_d *d = &bpf_dtab[minor(dev)];
    603 	int s, error = 0;
    604 
    605 	switch (cmd) {
    606 
    607 	default:
    608 		error = EINVAL;
    609 		break;
    610 
    611 	/*
    612 	 * Check for read packet available.
    613 	 */
    614 	case FIONREAD:
    615 		{
    616 			int n;
    617 
    618 			s = splimp();
    619 			n = d->bd_slen;
    620 			if (d->bd_hbuf)
    621 				n += d->bd_hlen;
    622 			splx(s);
    623 
    624 			*(int *)addr = n;
    625 			break;
    626 		}
    627 
    628 	case SIOCGIFADDR:
    629 		{
    630 			struct ifnet *ifp;
    631 
    632 			if (d->bd_bif == 0)
    633 				error = EINVAL;
    634 			else {
    635 				ifp = d->bd_bif->bif_ifp;
    636 				error = (*ifp->if_ioctl)(ifp, cmd, addr);
    637 			}
    638 			break;
    639 		}
    640 
    641 	/*
    642 	 * Get buffer len [for read()].
    643 	 */
    644 	case BIOCGBLEN:
    645 		*(u_int *)addr = d->bd_bufsize;
    646 		break;
    647 
    648 	/*
    649 	 * Set buffer length.
    650 	 */
    651 	case BIOCSBLEN:
    652 #if BSD < 199103
    653 		error = EINVAL;
    654 #else
    655 		if (d->bd_bif != 0)
    656 			error = EINVAL;
    657 		else {
    658 			register u_int size = *(u_int *)addr;
    659 
    660 			if (size > BPF_MAXBUFSIZE)
    661 				*(u_int *)addr = size = BPF_MAXBUFSIZE;
    662 			else if (size < BPF_MINBUFSIZE)
    663 				*(u_int *)addr = size = BPF_MINBUFSIZE;
    664 			d->bd_bufsize = size;
    665 		}
    666 #endif
    667 		break;
    668 
    669 	/*
    670 	 * Set link layer read filter.
    671 	 */
    672 	case BIOCSETF:
    673 		error = bpf_setf(d, (struct bpf_program *)addr);
    674 		break;
    675 
    676 	/*
    677 	 * Flush read packet buffer.
    678 	 */
    679 	case BIOCFLUSH:
    680 		s = splimp();
    681 		reset_d(d);
    682 		splx(s);
    683 		break;
    684 
    685 	/*
    686 	 * Put interface into promiscuous mode.
    687 	 */
    688 	case BIOCPROMISC:
    689 		if (d->bd_bif == 0) {
    690 			/*
    691 			 * No interface attached yet.
    692 			 */
    693 			error = EINVAL;
    694 			break;
    695 		}
    696 		s = splimp();
    697 		if (d->bd_promisc == 0) {
    698 			error = ifpromisc(d->bd_bif->bif_ifp, 1);
    699 			if (error == 0)
    700 				d->bd_promisc = 1;
    701 		}
    702 		splx(s);
    703 		break;
    704 
    705 	/*
    706 	 * Get device parameters.
    707 	 */
    708 	case BIOCGDLT:
    709 		if (d->bd_bif == 0)
    710 			error = EINVAL;
    711 		else
    712 			*(u_int *)addr = d->bd_bif->bif_dlt;
    713 		break;
    714 
    715 	/*
    716 	 * Set interface name.
    717 	 */
    718 	case BIOCGETIF:
    719 		if (d->bd_bif == 0)
    720 			error = EINVAL;
    721 		else
    722 			bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
    723 		break;
    724 
    725 	/*
    726 	 * Set interface.
    727 	 */
    728 	case BIOCSETIF:
    729 		error = bpf_setif(d, (struct ifreq *)addr);
    730 		break;
    731 
    732 	/*
    733 	 * Set read timeout.
    734 	 */
    735 	case BIOCSRTIMEOUT:
    736 		{
    737 			struct timeval *tv = (struct timeval *)addr;
    738 
    739 			/* Compute number of ticks. */
    740 			d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
    741 			break;
    742 		}
    743 
    744 	/*
    745 	 * Get read timeout.
    746 	 */
    747 	case BIOCGRTIMEOUT:
    748 		{
    749 			struct timeval *tv = (struct timeval *)addr;
    750 
    751 			tv->tv_sec = d->bd_rtout / hz;
    752 			tv->tv_usec = (d->bd_rtout % hz) * tick;
    753 			break;
    754 		}
    755 
    756 	/*
    757 	 * Get packet stats.
    758 	 */
    759 	case BIOCGSTATS:
    760 		{
    761 			struct bpf_stat *bs = (struct bpf_stat *)addr;
    762 
    763 			bs->bs_recv = d->bd_rcount;
    764 			bs->bs_drop = d->bd_dcount;
    765 			break;
    766 		}
    767 
    768 	/*
    769 	 * Set immediate mode.
    770 	 */
    771 	case BIOCIMMEDIATE:
    772 		d->bd_immediate = *(u_int *)addr;
    773 		break;
    774 
    775 	case BIOCVERSION:
    776 		{
    777 			struct bpf_version *bv = (struct bpf_version *)addr;
    778 
    779 			bv->bv_major = BPF_MAJOR_VERSION;
    780 			bv->bv_minor = BPF_MINOR_VERSION;
    781 			break;
    782 		}
    783 	}
    784 	return (error);
    785 }
    786 
    787 /*
    788  * Set d's packet filter program to fp.  If this file already has a filter,
    789  * free it and replace it.  Returns EINVAL for bogus requests.
    790  */
    791 int
    792 bpf_setf(d, fp)
    793 	struct bpf_d *d;
    794 	struct bpf_program *fp;
    795 {
    796 	struct bpf_insn *fcode, *old;
    797 	u_int flen, size;
    798 	int s;
    799 
    800 	old = d->bd_filter;
    801 	if (fp->bf_insns == 0) {
    802 		if (fp->bf_len != 0)
    803 			return (EINVAL);
    804 		s = splimp();
    805 		d->bd_filter = 0;
    806 		reset_d(d);
    807 		splx(s);
    808 		if (old != 0)
    809 			free((caddr_t)old, M_DEVBUF);
    810 		return (0);
    811 	}
    812 	flen = fp->bf_len;
    813 	if (flen > BPF_MAXINSNS)
    814 		return (EINVAL);
    815 
    816 	size = flen * sizeof(*fp->bf_insns);
    817 	fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
    818 	if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
    819 	    bpf_validate(fcode, (int)flen)) {
    820 		s = splimp();
    821 		d->bd_filter = fcode;
    822 		reset_d(d);
    823 		splx(s);
    824 		if (old != 0)
    825 			free((caddr_t)old, M_DEVBUF);
    826 
    827 		return (0);
    828 	}
    829 	free((caddr_t)fcode, M_DEVBUF);
    830 	return (EINVAL);
    831 }
    832 
    833 /*
    834  * Detach a file from its current interface (if attached at all) and attach
    835  * to the interface indicated by the name stored in ifr.
    836  * Return an errno or 0.
    837  */
    838 static int
    839 bpf_setif(d, ifr)
    840 	struct bpf_d *d;
    841 	struct ifreq *ifr;
    842 {
    843 	struct bpf_if *bp;
    844 	char *cp;
    845 	int unit, s, error;
    846 
    847 	/*
    848 	 * Separate string into name part and unit number.  Put a null
    849 	 * byte at the end of the name part, and compute the number.
    850 	 * If the a unit number is unspecified, the default is 0,
    851 	 * as initialized above.  XXX This should be common code.
    852 	 */
    853 	unit = 0;
    854 	cp = ifr->ifr_name;
    855 	cp[sizeof(ifr->ifr_name) - 1] = '\0';
    856 	while (*cp++) {
    857 		if (*cp >= '0' && *cp <= '9') {
    858 			unit = *cp - '0';
    859 			*cp++ = '\0';
    860 			while (*cp)
    861 				unit = 10 * unit + *cp++ - '0';
    862 			break;
    863 		}
    864 	}
    865 	/*
    866 	 * Look through attached interfaces for the named one.
    867 	 */
    868 	for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
    869 		struct ifnet *ifp = bp->bif_ifp;
    870 
    871 		if (ifp == 0 || unit != ifp->if_unit
    872 		    || strcmp(ifp->if_name, ifr->ifr_name) != 0)
    873 			continue;
    874 		/*
    875 		 * We found the requested interface.
    876 		 * If it's not up, return an error.
    877 		 * Allocate the packet buffers if we need to.
    878 		 * If we're already attached to requested interface,
    879 		 * just flush the buffer.
    880 		 */
    881 		if ((ifp->if_flags & IFF_UP) == 0)
    882 			return (ENETDOWN);
    883 
    884 		if (d->bd_sbuf == 0) {
    885 			error = bpf_allocbufs(d);
    886 			if (error != 0)
    887 				return (error);
    888 		}
    889 		s = splimp();
    890 		if (bp != d->bd_bif) {
    891 			if (d->bd_bif)
    892 				/*
    893 				 * Detach if attached to something else.
    894 				 */
    895 				bpf_detachd(d);
    896 
    897 			bpf_attachd(d, bp);
    898 		}
    899 		reset_d(d);
    900 		splx(s);
    901 		return (0);
    902 	}
    903 	/* Not found. */
    904 	return (ENXIO);
    905 }
    906 
    907 /*
    908  * Convert an interface name plus unit number of an ifp to a single
    909  * name which is returned in the ifr.
    910  */
    911 static void
    912 bpf_ifname(ifp, ifr)
    913 	struct ifnet *ifp;
    914 	struct ifreq *ifr;
    915 {
    916 	char *s = ifp->if_name;
    917 	char *d = ifr->ifr_name;
    918 
    919 	while (*d++ = *s++)
    920 		continue;
    921 	/* XXX Assume that unit number is less than 10. */
    922 	*d++ = ifp->if_unit + '0';
    923 	*d = '\0';
    924 }
    925 
    926 /*
    927  * The new select interface passes down the proc pointer; the old select
    928  * stubs had to grab it out of the user struct.  This glue allows either case.
    929  */
    930 #if BSD >= 199103
    931 #define bpf_select bpfselect
    932 #else
    933 int
    934 bpfselect(dev, rw)
    935 	register dev_t dev;
    936 	int rw;
    937 {
    938 	return (bpf_select(dev, rw, u.u_procp));
    939 }
    940 #endif
    941 
    942 /*
    943  * Support for select() system call
    944  *
    945  * Return true iff the specific operation will not block indefinitely.
    946  * Otherwise, return false but make a note that a selwakeup() must be done.
    947  */
    948 int
    949 bpf_select(dev, rw, p)
    950 	register dev_t dev;
    951 	int rw;
    952 	struct proc *p;
    953 {
    954 	register struct bpf_d *d;
    955 	register int s;
    956 
    957 	if (rw != FREAD)
    958 		return (0);
    959 	/*
    960 	 * An imitation of the FIONREAD ioctl code.
    961 	 */
    962 	d = &bpf_dtab[minor(dev)];
    963 
    964 	s = splimp();
    965 	if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0)) {
    966 		/*
    967 		 * There is data waiting.
    968 		 */
    969 		splx(s);
    970 		return (1);
    971 	}
    972 #if BSD >= 199103
    973 	selrecord(p, &d->bd_sel);
    974 #else
    975 	/*
    976 	 * No data ready.  If there's already a select() waiting on this
    977 	 * minor device then this is a collision.  This shouldn't happen
    978 	 * because minors really should not be shared, but if a process
    979 	 * forks while one of these is open, it is possible that both
    980 	 * processes could select on the same descriptor.
    981 	 */
    982 	if (d->bd_selproc && d->bd_selproc->p_wchan == (caddr_t)&selwait)
    983 		d->bd_selcoll = 1;
    984 	else
    985 		d->bd_selproc = p;
    986 #endif
    987 	splx(s);
    988 	return (0);
    989 }
    990 
    991 /*
    992  * Incoming linkage from device drivers.  Process the packet pkt, of length
    993  * pktlen, which is stored in a contiguous buffer.  The packet is parsed
    994  * by each process' filter, and if accepted, stashed into the corresponding
    995  * buffer.
    996  */
    997 void
    998 bpf_tap(arg, pkt, pktlen)
    999 	caddr_t arg;
   1000 	register u_char *pkt;
   1001 	register u_int pktlen;
   1002 {
   1003 	struct bpf_if *bp;
   1004 	register struct bpf_d *d;
   1005 	register size_t slen;
   1006 	/*
   1007 	 * Note that the ipl does not have to be raised at this point.
   1008 	 * The only problem that could arise here is that if two different
   1009 	 * interfaces shared any data.  This is not the case.
   1010 	 */
   1011 	bp = (struct bpf_if *)arg;
   1012 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
   1013 		++d->bd_rcount;
   1014 		slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
   1015 		if (slen != 0)
   1016 			catchpacket(d, pkt, pktlen, slen, bcopy);
   1017 	}
   1018 }
   1019 
   1020 /*
   1021  * Copy data from an mbuf chain into a buffer.  This code is derived
   1022  * from m_copydata in sys/uipc_mbuf.c.
   1023  */
   1024 static void
   1025 bpf_mcopy(src_arg, dst_arg, len)
   1026 	const void *src_arg;
   1027 	void *dst_arg;
   1028 	register size_t len;
   1029 {
   1030 	register const struct mbuf *m;
   1031 	register u_int count;
   1032 	u_char *dst;
   1033 
   1034 	m = src_arg;
   1035 	dst = dst_arg;
   1036 	while (len > 0) {
   1037 		if (m == 0)
   1038 			panic("bpf_mcopy");
   1039 		count = min(m->m_len, len);
   1040 		bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
   1041 		m = m->m_next;
   1042 		dst += count;
   1043 		len -= count;
   1044 	}
   1045 }
   1046 
   1047 /*
   1048  * Incoming linkage from device drivers, when packet is in an mbuf chain.
   1049  */
   1050 void
   1051 bpf_mtap(arg, m)
   1052 	caddr_t arg;
   1053 	struct mbuf *m;
   1054 {
   1055 	struct bpf_if *bp = (struct bpf_if *)arg;
   1056 	struct bpf_d *d;
   1057 	size_t pktlen, slen;
   1058 	struct mbuf *m0;
   1059 
   1060 	pktlen = 0;
   1061 	for (m0 = m; m0 != 0; m0 = m0->m_next)
   1062 		pktlen += m0->m_len;
   1063 
   1064 	for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
   1065 		++d->bd_rcount;
   1066 		slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
   1067 		if (slen != 0)
   1068 			catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
   1069 	}
   1070 }
   1071 
   1072 /*
   1073  * Move the packet data from interface memory (pkt) into the
   1074  * store buffer.  Return 1 if it's time to wakeup a listener (buffer full),
   1075  * otherwise 0.  "copy" is the routine called to do the actual data
   1076  * transfer.  bcopy is passed in to copy contiguous chunks, while
   1077  * bpf_mcopy is passed in to copy mbuf chains.  In the latter case,
   1078  * pkt is really an mbuf.
   1079  */
   1080 static void
   1081 catchpacket(d, pkt, pktlen, snaplen, cpfn)
   1082 	register struct bpf_d *d;
   1083 	register u_char *pkt;
   1084 	register size_t pktlen, snaplen;
   1085 	register void (*cpfn) __P((const void *, void *, size_t));
   1086 {
   1087 	register struct bpf_hdr *hp;
   1088 	register int totlen, curlen;
   1089 	register int hdrlen = d->bd_bif->bif_hdrlen;
   1090 	/*
   1091 	 * Figure out how many bytes to move.  If the packet is
   1092 	 * greater or equal to the snapshot length, transfer that
   1093 	 * much.  Otherwise, transfer the whole packet (unless
   1094 	 * we hit the buffer size limit).
   1095 	 */
   1096 	totlen = hdrlen + min(snaplen, pktlen);
   1097 	if (totlen > d->bd_bufsize)
   1098 		totlen = d->bd_bufsize;
   1099 
   1100 	/*
   1101 	 * Round up the end of the previous packet to the next longword.
   1102 	 */
   1103 	curlen = BPF_WORDALIGN(d->bd_slen);
   1104 	if (curlen + totlen > d->bd_bufsize) {
   1105 		/*
   1106 		 * This packet will overflow the storage buffer.
   1107 		 * Rotate the buffers if we can, then wakeup any
   1108 		 * pending reads.
   1109 		 */
   1110 		if (d->bd_fbuf == 0) {
   1111 			/*
   1112 			 * We haven't completed the previous read yet,
   1113 			 * so drop the packet.
   1114 			 */
   1115 			++d->bd_dcount;
   1116 			return;
   1117 		}
   1118 		ROTATE_BUFFERS(d);
   1119 		bpf_wakeup(d);
   1120 		curlen = 0;
   1121 	}
   1122 	else if (d->bd_immediate)
   1123 		/*
   1124 		 * Immediate mode is set.  A packet arrived so any
   1125 		 * reads should be woken up.
   1126 		 */
   1127 		bpf_wakeup(d);
   1128 
   1129 	/*
   1130 	 * Append the bpf header.
   1131 	 */
   1132 	hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
   1133 #if BSD >= 199103
   1134 	microtime(&hp->bh_tstamp);
   1135 #elif defined(sun)
   1136 	uniqtime(&hp->bh_tstamp);
   1137 #else
   1138 	hp->bh_tstamp = time;
   1139 #endif
   1140 	hp->bh_datalen = pktlen;
   1141 	hp->bh_hdrlen = hdrlen;
   1142 	/*
   1143 	 * Copy the packet data into the store buffer and update its length.
   1144 	 */
   1145 	(*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
   1146 	d->bd_slen = curlen + totlen;
   1147 }
   1148 
   1149 /*
   1150  * Initialize all nonzero fields of a descriptor.
   1151  */
   1152 static int
   1153 bpf_allocbufs(d)
   1154 	register struct bpf_d *d;
   1155 {
   1156 	d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
   1157 	if (d->bd_fbuf == 0)
   1158 		return (ENOBUFS);
   1159 
   1160 	d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
   1161 	if (d->bd_sbuf == 0) {
   1162 		free(d->bd_fbuf, M_DEVBUF);
   1163 		return (ENOBUFS);
   1164 	}
   1165 	d->bd_slen = 0;
   1166 	d->bd_hlen = 0;
   1167 	return (0);
   1168 }
   1169 
   1170 /*
   1171  * Free buffers currently in use by a descriptor.
   1172  * Called on close.
   1173  */
   1174 static void
   1175 bpf_freed(d)
   1176 	register struct bpf_d *d;
   1177 {
   1178 	/*
   1179 	 * We don't need to lock out interrupts since this descriptor has
   1180 	 * been detached from its interface and it yet hasn't been marked
   1181 	 * free.
   1182 	 */
   1183 	if (d->bd_sbuf != 0) {
   1184 		free(d->bd_sbuf, M_DEVBUF);
   1185 		if (d->bd_hbuf != 0)
   1186 			free(d->bd_hbuf, M_DEVBUF);
   1187 		if (d->bd_fbuf != 0)
   1188 			free(d->bd_fbuf, M_DEVBUF);
   1189 	}
   1190 	if (d->bd_filter)
   1191 		free((caddr_t)d->bd_filter, M_DEVBUF);
   1192 
   1193 	D_MARKFREE(d);
   1194 }
   1195 
   1196 /*
   1197  * Attach an interface to bpf.  driverp is a pointer to a (struct bpf_if *)
   1198  * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
   1199  * size of the link header (variable length headers not yet supported).
   1200  */
   1201 void
   1202 bpfattach(driverp, ifp, dlt, hdrlen)
   1203 	caddr_t *driverp;
   1204 	struct ifnet *ifp;
   1205 	u_int dlt, hdrlen;
   1206 {
   1207 	struct bpf_if *bp;
   1208 	int i;
   1209 #if BSD < 199103
   1210 	static struct bpf_if bpf_ifs[NBPFILTER];
   1211 	static int bpfifno;
   1212 
   1213 	bp = (bpfifno < NBPFILTER) ? &bpf_ifs[bpfifno++] : 0;
   1214 #else
   1215 	bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
   1216 #endif
   1217 	if (bp == 0)
   1218 		panic("bpfattach");
   1219 
   1220 	bp->bif_dlist = 0;
   1221 	bp->bif_driverp = (struct bpf_if **)driverp;
   1222 	bp->bif_ifp = ifp;
   1223 	bp->bif_dlt = dlt;
   1224 
   1225 	bp->bif_next = bpf_iflist;
   1226 	bpf_iflist = bp;
   1227 
   1228 	*bp->bif_driverp = 0;
   1229 
   1230 	/*
   1231 	 * Compute the length of the bpf header.  This is not necessarily
   1232 	 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
   1233 	 * that the network layer header begins on a longword boundary (for
   1234 	 * performance reasons and to alleviate alignment restrictions).
   1235 	 */
   1236 	bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
   1237 
   1238 	/*
   1239 	 * Mark all the descriptors free if this hasn't been done.
   1240 	 */
   1241 	if (!D_ISFREE(&bpf_dtab[0]))
   1242 		for (i = 0; i < NBPFILTER; ++i)
   1243 			D_MARKFREE(&bpf_dtab[i]);
   1244 
   1245 #if 0
   1246 	printf("bpf: %s%d attached\n", ifp->if_name, ifp->if_unit);
   1247 #endif
   1248 }
   1249 
   1250 #if BSD >= 199103
   1251 /* XXX This routine belongs in net/if.c. */
   1252 /*
   1253  * Set/clear promiscuous mode on interface ifp based on the truth value
   1254  * of pswitch.  The calls are reference counted so that only the first
   1255  * "on" request actually has an effect, as does the final "off" request.
   1256  * Results are undefined if the "off" and "on" requests are not matched.
   1257  */
   1258 int
   1259 ifpromisc(ifp, pswitch)
   1260 	struct ifnet *ifp;
   1261 	int pswitch;
   1262 {
   1263 	struct ifreq ifr;
   1264 
   1265 	if (pswitch) {
   1266 		/*
   1267 		 * If the device is not configured up, we cannot put it in
   1268 		 * promiscuous mode.
   1269 		 */
   1270 		if ((ifp->if_flags & IFF_UP) == 0)
   1271 			return (ENETDOWN);
   1272 		if (ifp->if_pcount++ != 0)
   1273 			return (0);
   1274 		ifp->if_flags |= IFF_PROMISC;
   1275 	} else {
   1276 		if (--ifp->if_pcount > 0)
   1277 			return (0);
   1278 		ifp->if_flags &= ~IFF_PROMISC;
   1279 		/*
   1280 		 * If the device is not configured up, we should not need to
   1281 		 * turn off promiscuous mode (device should have turned it
   1282 		 * off when interface went down; and will look at IFF_PROMISC
   1283 		 * again next time interface comes up).
   1284 		 */
   1285 		if ((ifp->if_flags & IFF_UP) == 0)
   1286 			return (0);
   1287 	}
   1288 	ifr.ifr_flags = ifp->if_flags;
   1289 	return ((*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr));
   1290 }
   1291 #endif
   1292 
   1293 #if BSD < 199103
   1294 /*
   1295  * Allocate some memory for bpf.  This is temporary SunOS support, and
   1296  * is admittedly a hack.
   1297  * If resources unavaiable, return 0.
   1298  */
   1299 static caddr_t
   1300 bpf_alloc(size, canwait)
   1301 	register int size;
   1302 	register int canwait;
   1303 {
   1304 	register struct mbuf *m;
   1305 
   1306 	if ((unsigned)size > (MCLBYTES-8))
   1307 		return 0;
   1308 
   1309 	MGET(m, canwait, MT_DATA);
   1310 	if (m == 0)
   1311 		return 0;
   1312 	if ((unsigned)size > (MLEN-8)) {
   1313 		MCLGET(m);
   1314 		if (m->m_len != MCLBYTES) {
   1315 			m_freem(m);
   1316 			return 0;
   1317 		}
   1318 	}
   1319 	*mtod(m, struct mbuf **) = m;
   1320 	return mtod(m, caddr_t) + 8;
   1321 }
   1322 #endif
   1323