bpf.c revision 1.39 1 /* $NetBSD: bpf.c,v 1.39 1998/03/01 02:25:04 fvdl 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 8.4 (Berkeley) 1/9/95
41 * static char rcsid[] =
42 * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp ";
43 */
44
45 #include "bpfilter.h"
46
47 #include <sys/param.h>
48 #include <sys/systm.h>
49 #include <sys/mbuf.h>
50 #include <sys/buf.h>
51 #include <sys/time.h>
52 #include <sys/proc.h>
53 #include <sys/user.h>
54 #include <sys/ioctl.h>
55 #include <sys/map.h>
56 #include <sys/conf.h>
57
58 #include <sys/file.h>
59 #if defined(sparc) && BSD < 199103
60 #include <sys/stream.h>
61 #endif
62 #include <sys/tty.h>
63 #include <sys/uio.h>
64
65 #include <sys/protosw.h>
66 #include <sys/socket.h>
67 #include <sys/errno.h>
68 #include <sys/kernel.h>
69 #include <sys/poll.h>
70
71 #include <net/if.h>
72
73 #include <net/bpf.h>
74 #include <net/bpfdesc.h>
75
76 #include <net/if_arc.h>
77 #include <net/if_ether.h>
78
79 #include <netinet/in.h>
80 #include <netinet/if_inarp.h>
81
82 /*
83 * Older BSDs don't have kernel malloc.
84 */
85 #if BSD < 199103
86 extern bcopy();
87 static caddr_t bpf_alloc();
88 #include <net/bpf_compat.h>
89 #define BPF_BUFSIZE (MCLBYTES-8)
90 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, code, uio)
91 #else
92 #define BPF_BUFSIZE 4096
93 #define UIOMOVE(cp, len, code, uio) uiomove(cp, len, uio)
94 #endif
95
96 #define PRINET 26 /* interruptible */
97
98 /*
99 * The default read buffer size is patchable.
100 */
101 int bpf_bufsize = BPF_BUFSIZE;
102
103 /*
104 * bpf_iflist is the list of interfaces; each corresponds to an ifnet
105 * bpf_dtab holds the descriptors, indexed by minor device #
106 */
107 struct bpf_if *bpf_iflist;
108 struct bpf_d bpf_dtab[NBPFILTER];
109
110 #if BSD >= 199207 || NetBSD0_9 >= 2
111 /*
112 * bpfilterattach() is called at boot time in new systems. We do
113 * nothing here since old systems will not call this.
114 */
115 /* ARGSUSED */
116 void
117 bpfilterattach(n)
118 int n;
119 {
120 }
121 #endif
122
123 static int bpf_allocbufs __P((struct bpf_d *));
124 static int bpf_allocbufs __P((struct bpf_d *));
125 static void bpf_freed __P((struct bpf_d *));
126 static void bpf_freed __P((struct bpf_d *));
127 static void bpf_ifname __P((struct ifnet *, struct ifreq *));
128 static void bpf_ifname __P((struct ifnet *, struct ifreq *));
129 static void bpf_mcopy __P((const void *, void *, size_t));
130 static int bpf_movein __P((struct uio *, int, int,
131 struct mbuf **, struct sockaddr *));
132 static void bpf_attachd __P((struct bpf_d *, struct bpf_if *));
133 static void bpf_detachd __P((struct bpf_d *));
134 static int bpf_setif __P((struct bpf_d *, struct ifreq *));
135 int bpfpoll __P((dev_t, int, struct proc *));
136 static __inline void
137 bpf_wakeup __P((struct bpf_d *));
138 static void catchpacket __P((struct bpf_d *, u_char *, u_int, u_int,
139 void (*)(const void *, void *, size_t)));
140 static void reset_d __P((struct bpf_d *));
141
142 static int
143 bpf_movein(uio, linktype, mtu, mp, sockp)
144 register struct uio *uio;
145 int linktype;
146 int mtu;
147 register struct mbuf **mp;
148 register 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_UNSPEC;
195 /* XXX 4(FORMAC)+6(dst)+6(src)+3(LLC)+5(SNAP) */
196 hlen = 24;
197 align = 0;
198 break;
199
200 case DLT_NULL:
201 sockp->sa_family = AF_UNSPEC;
202 hlen = 0;
203 align = 0;
204 break;
205
206 default:
207 return (EIO);
208 }
209
210 len = uio->uio_resid;
211 /*
212 * If there aren't enough bytes for a link level header or the
213 * packet length exceeds the interface mtu, return an error.
214 */
215 if (len < hlen || len - hlen > mtu)
216 return (EMSGSIZE);
217
218 /*
219 * XXX Avoid complicated buffer chaining ---
220 * bail if it won't fit in a single mbuf.
221 * (Take into account possible alignment bytes)
222 */
223 if ((unsigned)len > MCLBYTES - align)
224 return (EIO);
225
226 MGETHDR(m, M_WAIT, MT_DATA);
227 if (m == 0)
228 return (ENOBUFS);
229 m->m_pkthdr.rcvif = 0;
230 m->m_pkthdr.len = len - hlen;
231 if (len > MHLEN - align) {
232 #if BSD >= 199103
233 MCLGET(m, M_WAIT);
234 if ((m->m_flags & M_EXT) == 0) {
235 #else
236 MCLGET(m);
237 if (m->m_len != MCLBYTES) {
238 #endif
239 error = ENOBUFS;
240 goto bad;
241 }
242 }
243
244 /* Insure the data is properly aligned */
245 if (align > 0) {
246 #if BSD >= 199103
247 m->m_data += align;
248 #else
249 m->m_off += align;
250 #endif
251 m->m_len -= align;
252 }
253
254 error = UIOMOVE(mtod(m, caddr_t), len, UIO_WRITE, uio);
255 if (error)
256 goto bad;
257 if (hlen != 0) {
258 bcopy(mtod(m, caddr_t), sockp->sa_data, hlen);
259 #if BSD >= 199103
260 m->m_data += hlen; /* XXX */
261 #else
262 m->m_off += hlen;
263 #endif
264 len -= hlen;
265 }
266 m->m_len = len;
267 *mp = m;
268 return (0);
269
270 bad:
271 m_freem(m);
272 return (error);
273 }
274
275 /*
276 * Attach file to the bpf interface, i.e. make d listen on bp.
277 * Must be called at splimp.
278 */
279 static void
280 bpf_attachd(d, bp)
281 struct bpf_d *d;
282 struct bpf_if *bp;
283 {
284 /*
285 * Point d at bp, and add d to the interface's list of listeners.
286 * Finally, point the driver's bpf cookie at the interface so
287 * it will divert packets to bpf.
288 */
289 d->bd_bif = bp;
290 d->bd_next = bp->bif_dlist;
291 bp->bif_dlist = d;
292
293 *bp->bif_driverp = bp;
294 }
295
296 /*
297 * Detach a file from its interface.
298 */
299 static void
300 bpf_detachd(d)
301 struct bpf_d *d;
302 {
303 struct bpf_d **p;
304 struct bpf_if *bp;
305
306 bp = d->bd_bif;
307 /*
308 * Check if this descriptor had requested promiscuous mode.
309 * If so, turn it off.
310 */
311 if (d->bd_promisc) {
312 int error;
313
314 d->bd_promisc = 0;
315 /*
316 * Take device out of promiscuous mode. Since we were
317 * able to enter promiscuous mode, we should be able
318 * to turn it off. But we can get an error if
319 * the interface was configured down, so only panic
320 * if we don't get an unexpected error.
321 */
322 error = ifpromisc(bp->bif_ifp, 0);
323 if (error && error != EINVAL)
324 panic("bpf: ifpromisc failed");
325 }
326 /* Remove d from the interface's descriptor list. */
327 p = &bp->bif_dlist;
328 while (*p != d) {
329 p = &(*p)->bd_next;
330 if (*p == 0)
331 panic("bpf_detachd: descriptor not in list");
332 }
333 *p = (*p)->bd_next;
334 if (bp->bif_dlist == 0)
335 /*
336 * Let the driver know that there are no more listeners.
337 */
338 *d->bd_bif->bif_driverp = 0;
339 d->bd_bif = 0;
340 }
341
342
343 /*
344 * Mark a descriptor free by making it point to itself.
345 * This is probably cheaper than marking with a constant since
346 * the address should be in a register anyway.
347 */
348 #define D_ISFREE(d) ((d) == (d)->bd_next)
349 #define D_MARKFREE(d) ((d)->bd_next = (d))
350 #define D_MARKUSED(d) ((d)->bd_next = 0)
351
352 /*
353 * Open ethernet device. Returns ENXIO for illegal minor device number,
354 * EBUSY if file is open by another process.
355 */
356 /* ARGSUSED */
357 int
358 bpfopen(dev, flag, mode, p)
359 dev_t dev;
360 int flag;
361 int mode;
362 struct proc *p;
363 {
364 register struct bpf_d *d;
365
366 if (minor(dev) >= NBPFILTER)
367 return (ENXIO);
368 /*
369 * Each minor can be opened by only one process. If the requested
370 * minor is in use, return EBUSY.
371 */
372 d = &bpf_dtab[minor(dev)];
373 if (!D_ISFREE(d))
374 return (EBUSY);
375
376 /* Mark "free" and do most initialization. */
377 bzero((char *)d, sizeof(*d));
378 d->bd_bufsize = bpf_bufsize;
379
380 return (0);
381 }
382
383 /*
384 * Close the descriptor by detaching it from its interface,
385 * deallocating its buffers, and marking it free.
386 */
387 /* ARGSUSED */
388 int
389 bpfclose(dev, flag, mode, p)
390 dev_t dev;
391 int flag;
392 int mode;
393 struct proc *p;
394 {
395 register struct bpf_d *d = &bpf_dtab[minor(dev)];
396 register int s;
397
398 s = splimp();
399 if (d->bd_bif)
400 bpf_detachd(d);
401 splx(s);
402 bpf_freed(d);
403
404 return (0);
405 }
406
407 /*
408 * Support for SunOS, which does not have tsleep.
409 */
410 #if BSD < 199103
411 static
412 bpf_timeout(arg)
413 caddr_t arg;
414 {
415 struct bpf_d *d = (struct bpf_d *)arg;
416 d->bd_timedout = 1;
417 wakeup(arg);
418 }
419
420 #define BPF_SLEEP(chan, pri, s, t) bpf_sleep((struct bpf_d *)chan)
421
422 int
423 bpf_sleep(d)
424 register struct bpf_d *d;
425 {
426 register int rto = d->bd_rtout;
427 register int st;
428
429 if (rto != 0) {
430 d->bd_timedout = 0;
431 timeout(bpf_timeout, (caddr_t)d, rto);
432 }
433 st = sleep((caddr_t)d, PRINET|PCATCH);
434 if (rto != 0) {
435 if (d->bd_timedout == 0)
436 untimeout(bpf_timeout, (caddr_t)d);
437 else if (st == 0)
438 return EWOULDBLOCK;
439 }
440 return (st != 0) ? EINTR : 0;
441 }
442 #else
443 #define BPF_SLEEP tsleep
444 #endif
445
446 /*
447 * Rotate the packet buffers in descriptor d. Move the store buffer
448 * into the hold slot, and the free buffer into the store slot.
449 * Zero the length of the new store buffer.
450 */
451 #define ROTATE_BUFFERS(d) \
452 (d)->bd_hbuf = (d)->bd_sbuf; \
453 (d)->bd_hlen = (d)->bd_slen; \
454 (d)->bd_sbuf = (d)->bd_fbuf; \
455 (d)->bd_slen = 0; \
456 (d)->bd_fbuf = 0;
457 /*
458 * bpfread - read next chunk of packets from buffers
459 */
460 int
461 bpfread(dev, uio, ioflag)
462 dev_t dev;
463 register struct uio *uio;
464 int ioflag;
465 {
466 register struct bpf_d *d = &bpf_dtab[minor(dev)];
467 int error;
468 int s;
469
470 /*
471 * Restrict application to use a buffer the same size as
472 * as kernel buffers.
473 */
474 if (uio->uio_resid != d->bd_bufsize)
475 return (EINVAL);
476
477 s = splimp();
478 /*
479 * If the hold buffer is empty, then do a timed sleep, which
480 * ends when the timeout expires or when enough packets
481 * have arrived to fill the store buffer.
482 */
483 while (d->bd_hbuf == 0) {
484 if (d->bd_immediate) {
485 if (d->bd_slen == 0) {
486 splx(s);
487 return (EWOULDBLOCK);
488 }
489 /*
490 * A packet(s) either arrived since the previous
491 * read or arrived while we were asleep.
492 * Rotate the buffers and return what's here.
493 */
494 ROTATE_BUFFERS(d);
495 break;
496 }
497 if (d->bd_rtout != -1)
498 error = BPF_SLEEP((caddr_t)d, PRINET|PCATCH, "bpf",
499 d->bd_rtout);
500 else
501 error = EWOULDBLOCK; /* User requested non-blocking I/O */
502 if (error == EINTR || error == ERESTART) {
503 splx(s);
504 return (error);
505 }
506 if (error == EWOULDBLOCK) {
507 /*
508 * On a timeout, return what's in the buffer,
509 * which may be nothing. If there is something
510 * in the store buffer, we can rotate the buffers.
511 */
512 if (d->bd_hbuf)
513 /*
514 * We filled up the buffer in between
515 * getting the timeout and arriving
516 * here, so we don't need to rotate.
517 */
518 break;
519
520 if (d->bd_slen == 0) {
521 splx(s);
522 return (0);
523 }
524 ROTATE_BUFFERS(d);
525 break;
526 }
527 if (error != 0)
528 goto done;
529 }
530 /*
531 * At this point, we know we have something in the hold slot.
532 */
533 splx(s);
534
535 /*
536 * Move data from hold buffer into user space.
537 * We know the entire buffer is transferred since
538 * we checked above that the read buffer is bpf_bufsize bytes.
539 */
540 error = UIOMOVE(d->bd_hbuf, d->bd_hlen, UIO_READ, uio);
541
542 s = splimp();
543 d->bd_fbuf = d->bd_hbuf;
544 d->bd_hbuf = 0;
545 d->bd_hlen = 0;
546 done:
547 splx(s);
548 return (error);
549 }
550
551
552 /*
553 * If there are processes sleeping on this descriptor, wake them up.
554 */
555 static __inline void
556 bpf_wakeup(d)
557 register struct bpf_d *d;
558 {
559 struct proc *p;
560
561 wakeup((caddr_t)d);
562 if (d->bd_async)
563 if (d->bd_pgid > 0)
564 gsignal (d->bd_pgid, SIGIO);
565 else if ((p = pfind (-d->bd_pgid)) != NULL)
566 psignal (p, SIGIO);
567
568 #if BSD >= 199103
569 selwakeup(&d->bd_sel);
570 /* XXX */
571 d->bd_sel.si_pid = 0;
572 #else
573 if (d->bd_selproc) {
574 selwakeup(d->bd_selproc, (int)d->bd_selcoll);
575 d->bd_selcoll = 0;
576 d->bd_selproc = 0;
577 }
578 #endif
579 }
580
581 int
582 bpfwrite(dev, uio, ioflag)
583 dev_t dev;
584 struct uio *uio;
585 int ioflag;
586 {
587 register struct bpf_d *d = &bpf_dtab[minor(dev)];
588 struct ifnet *ifp;
589 struct mbuf *m;
590 int error, s;
591 static struct sockaddr dst;
592
593 if (d->bd_bif == 0)
594 return (ENXIO);
595
596 ifp = d->bd_bif->bif_ifp;
597
598 if (uio->uio_resid == 0)
599 return (0);
600
601 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp->if_mtu, &m, &dst);
602 if (error)
603 return (error);
604
605 if (m->m_pkthdr.len > ifp->if_mtu)
606 return (EMSGSIZE);
607
608 s = splsoftnet();
609 #if BSD >= 199103
610 error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)0);
611 #else
612 error = (*ifp->if_output)(ifp, m, &dst);
613 #endif
614 splx(s);
615 /*
616 * The driver frees the mbuf.
617 */
618 return (error);
619 }
620
621 /*
622 * Reset a descriptor by flushing its packet buffer and clearing the
623 * receive and drop counts. Should be called at splimp.
624 */
625 static void
626 reset_d(d)
627 struct bpf_d *d;
628 {
629 if (d->bd_hbuf) {
630 /* Free the hold buffer. */
631 d->bd_fbuf = d->bd_hbuf;
632 d->bd_hbuf = 0;
633 }
634 d->bd_slen = 0;
635 d->bd_hlen = 0;
636 d->bd_rcount = 0;
637 d->bd_dcount = 0;
638 }
639
640 #ifdef BPF_KERN_FILTER
641 extern struct bpf_insn *bpf_tcp_filter;
642 extern struct bpf_insn *bpf_udp_filter;
643 #endif
644
645 /*
646 * FIONREAD Check for read packet available.
647 * BIOCGBLEN Get buffer len [for read()].
648 * BIOCSETF Set ethernet read filter.
649 * BIOCFLUSH Flush read packet buffer.
650 * BIOCPROMISC Put interface into promiscuous mode.
651 * BIOCGDLT Get link layer type.
652 * BIOCGETIF Get interface name.
653 * BIOCSETIF Set interface.
654 * BIOCSRTIMEOUT Set read timeout.
655 * BIOCGRTIMEOUT Get read timeout.
656 * BIOCGSTATS Get packet stats.
657 * BIOCIMMEDIATE Set immediate mode.
658 * BIOCVERSION Get filter language version.
659 */
660 /* ARGSUSED */
661 int
662 bpfioctl(dev, cmd, addr, flag, p)
663 dev_t dev;
664 u_long cmd;
665 caddr_t addr;
666 int flag;
667 struct proc *p;
668 {
669 register struct bpf_d *d = &bpf_dtab[minor(dev)];
670 int s, error = 0;
671 #ifdef BPF_KERN_FILTER
672 register struct bpf_insn **p;
673 #endif
674
675 switch (cmd) {
676
677 default:
678 error = EINVAL;
679 break;
680
681 /*
682 * Check for read packet available.
683 */
684 case FIONREAD:
685 {
686 int n;
687
688 s = splimp();
689 n = d->bd_slen;
690 if (d->bd_hbuf)
691 n += d->bd_hlen;
692 splx(s);
693
694 *(int *)addr = n;
695 break;
696 }
697
698 /*
699 * Get buffer len [for read()].
700 */
701 case BIOCGBLEN:
702 *(u_int *)addr = d->bd_bufsize;
703 break;
704
705 /*
706 * Set buffer length.
707 */
708 case BIOCSBLEN:
709 #if BSD < 199103
710 error = EINVAL;
711 #else
712 if (d->bd_bif != 0)
713 error = EINVAL;
714 else {
715 register u_int size = *(u_int *)addr;
716
717 if (size > BPF_MAXBUFSIZE)
718 *(u_int *)addr = size = BPF_MAXBUFSIZE;
719 else if (size < BPF_MINBUFSIZE)
720 *(u_int *)addr = size = BPF_MINBUFSIZE;
721 d->bd_bufsize = size;
722 }
723 #endif
724 break;
725
726 /*
727 * Set link layer read filter.
728 */
729 case BIOCSETF:
730 error = bpf_setf(d, (struct bpf_program *)addr);
731 break;
732
733 #ifdef BPF_KERN_FILTER
734 /*
735 * Set TCP or UDP reject filter.
736 */
737 case BIOCSTCPF:
738 case BIOCSUDPF:
739 if (!suser()) {
740 error = EPERM;
741 break;
742 }
743
744 /* Validate and store filter */
745 error = bpf_setf(d, (struct bpf_program *)addr);
746
747 /* Free possible old filter */
748 if (cmd == BIOCSTCPF)
749 p = &bpf_tcp_filter;
750 else
751 p = &bpf_udp_filter;
752 if (*p != NULL)
753 free((caddr_t)*p, M_DEVBUF);
754
755 /* Steal new filter (noop if error) */
756 s = splimp();
757 *p = d->bd_filter;
758 d->bd_filter = NULL;
759 splx(s);
760 break;
761 #endif
762
763 /*
764 * Flush read packet buffer.
765 */
766 case BIOCFLUSH:
767 s = splimp();
768 reset_d(d);
769 splx(s);
770 break;
771
772 /*
773 * Put interface into promiscuous mode.
774 */
775 case BIOCPROMISC:
776 if (d->bd_bif == 0) {
777 /*
778 * No interface attached yet.
779 */
780 error = EINVAL;
781 break;
782 }
783 s = splimp();
784 if (d->bd_promisc == 0) {
785 error = ifpromisc(d->bd_bif->bif_ifp, 1);
786 if (error == 0)
787 d->bd_promisc = 1;
788 }
789 splx(s);
790 break;
791
792 /*
793 * Get device parameters.
794 */
795 case BIOCGDLT:
796 if (d->bd_bif == 0)
797 error = EINVAL;
798 else
799 *(u_int *)addr = d->bd_bif->bif_dlt;
800 break;
801
802 /*
803 * Set interface name.
804 */
805 case BIOCGETIF:
806 if (d->bd_bif == 0)
807 error = EINVAL;
808 else
809 bpf_ifname(d->bd_bif->bif_ifp, (struct ifreq *)addr);
810 break;
811
812 /*
813 * Set interface.
814 */
815 case BIOCSETIF:
816 error = bpf_setif(d, (struct ifreq *)addr);
817 break;
818
819 /*
820 * Set read timeout.
821 */
822 case BIOCSRTIMEOUT:
823 {
824 struct timeval *tv = (struct timeval *)addr;
825
826 /* Compute number of ticks. */
827 d->bd_rtout = tv->tv_sec * hz + tv->tv_usec / tick;
828 if ((d->bd_rtout == 0) && (tv->tv_usec != 0))
829 d->bd_rtout = 1;
830 break;
831 }
832
833 /*
834 * Get read timeout.
835 */
836 case BIOCGRTIMEOUT:
837 {
838 struct timeval *tv = (struct timeval *)addr;
839
840 tv->tv_sec = d->bd_rtout / hz;
841 tv->tv_usec = (d->bd_rtout % hz) * tick;
842 break;
843 }
844
845 /*
846 * Get packet stats.
847 */
848 case BIOCGSTATS:
849 {
850 struct bpf_stat *bs = (struct bpf_stat *)addr;
851
852 bs->bs_recv = d->bd_rcount;
853 bs->bs_drop = d->bd_dcount;
854 break;
855 }
856
857 /*
858 * Set immediate mode.
859 */
860 case BIOCIMMEDIATE:
861 d->bd_immediate = *(u_int *)addr;
862 break;
863
864 case BIOCVERSION:
865 {
866 struct bpf_version *bv = (struct bpf_version *)addr;
867
868 bv->bv_major = BPF_MAJOR_VERSION;
869 bv->bv_minor = BPF_MINOR_VERSION;
870 break;
871 }
872
873
874 case FIONBIO: /* Non-blocking I/O */
875 if (*(int *)addr)
876 d->bd_rtout = -1;
877 else
878 d->bd_rtout = 0;
879 break;
880
881 case FIOASYNC: /* Send signal on receive packets */
882 d->bd_async = *(int *)addr;
883 break;
884
885 /*
886 * N.B. ioctl (FIOSETOWN) and fcntl (F_SETOWN) both end up doing
887 * the equivalent of a TIOCSPGRP and hence end up here. *However*
888 * TIOCSPGRP's arg is a process group if it's positive and a process
889 * id if it's negative. This is exactly the opposite of what the
890 * other two functions want! Therefore there is code in ioctl and
891 * fcntl to negate the arg before calling here.
892 */
893 case TIOCSPGRP: /* Process or group to send signals to */
894 d->bd_pgid = *(int *)addr;
895 break;
896
897 case TIOCGPGRP:
898 *(int *)addr = d->bd_pgid;
899 break;
900 }
901 return (error);
902 }
903
904 /*
905 * Set d's packet filter program to fp. If this file already has a filter,
906 * free it and replace it. Returns EINVAL for bogus requests.
907 */
908 int
909 bpf_setf(d, fp)
910 struct bpf_d *d;
911 struct bpf_program *fp;
912 {
913 struct bpf_insn *fcode, *old;
914 u_int flen, size;
915 int s;
916
917 old = d->bd_filter;
918 if (fp->bf_insns == 0) {
919 if (fp->bf_len != 0)
920 return (EINVAL);
921 s = splimp();
922 d->bd_filter = 0;
923 reset_d(d);
924 splx(s);
925 if (old != 0)
926 free((caddr_t)old, M_DEVBUF);
927 return (0);
928 }
929 flen = fp->bf_len;
930 if (flen > BPF_MAXINSNS)
931 return (EINVAL);
932
933 size = flen * sizeof(*fp->bf_insns);
934 fcode = (struct bpf_insn *)malloc(size, M_DEVBUF, M_WAITOK);
935 if (fcode == 0)
936 return (ENOMEM);
937 if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
938 bpf_validate(fcode, (int)flen)) {
939 s = splimp();
940 d->bd_filter = fcode;
941 reset_d(d);
942 splx(s);
943 if (old != 0)
944 free((caddr_t)old, M_DEVBUF);
945
946 return (0);
947 }
948 free((caddr_t)fcode, M_DEVBUF);
949 return (EINVAL);
950 }
951
952 /*
953 * Detach a file from its current interface (if attached at all) and attach
954 * to the interface indicated by the name stored in ifr.
955 * Return an errno or 0.
956 */
957 static int
958 bpf_setif(d, ifr)
959 struct bpf_d *d;
960 struct ifreq *ifr;
961 {
962 struct bpf_if *bp;
963 char *cp;
964 int unit_seen, i, s, error;
965
966 /*
967 * Make sure the provided name has a unit number, and default
968 * it to '0' if not specified.
969 * XXX This is ugly ... do this differently?
970 */
971 unit_seen = 0;
972 cp = ifr->ifr_name;
973 cp[sizeof(ifr->ifr_name) - 1] = '\0'; /* sanity */
974 while (*cp++)
975 if (*cp >= '0' && *cp <= '9')
976 unit_seen = 1;
977 if (!unit_seen) {
978 /* Make sure to leave room for the '\0'. */
979 for (i = 0; i < (IFNAMSIZ - 1); ++i) {
980 if ((ifr->ifr_name[i] >= 'a' &&
981 ifr->ifr_name[i] <= 'z') ||
982 (ifr->ifr_name[i] >= 'A' &&
983 ifr->ifr_name[i] <= 'Z'))
984 continue;
985 ifr->ifr_name[i] = '0';
986 }
987 }
988
989 /*
990 * Look through attached interfaces for the named one.
991 */
992 for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) {
993 struct ifnet *ifp = bp->bif_ifp;
994
995 if (ifp == 0 ||
996 strcmp(ifp->if_xname, ifr->ifr_name) != 0)
997 continue;
998 /*
999 * We found the requested interface.
1000 * If it's not up, return an error.
1001 * Allocate the packet buffers if we need to.
1002 * If we're already attached to requested interface,
1003 * just flush the buffer.
1004 */
1005 if ((ifp->if_flags & IFF_UP) == 0)
1006 return (ENETDOWN);
1007
1008 if (d->bd_sbuf == 0) {
1009 error = bpf_allocbufs(d);
1010 if (error != 0)
1011 return (error);
1012 }
1013 s = splimp();
1014 if (bp != d->bd_bif) {
1015 if (d->bd_bif)
1016 /*
1017 * Detach if attached to something else.
1018 */
1019 bpf_detachd(d);
1020
1021 bpf_attachd(d, bp);
1022 }
1023 reset_d(d);
1024 splx(s);
1025 return (0);
1026 }
1027 /* Not found. */
1028 return (ENXIO);
1029 }
1030
1031 /*
1032 * Copy the interface name to the ifreq.
1033 */
1034 static void
1035 bpf_ifname(ifp, ifr)
1036 struct ifnet *ifp;
1037 struct ifreq *ifr;
1038 {
1039
1040 bcopy(ifp->if_xname, ifr->ifr_name, IFNAMSIZ);
1041 }
1042
1043 /*
1044 * Support for poll() system call
1045 *
1046 * Return true iff the specific operation will not block indefinitely.
1047 * Otherwise, return false but make a note that a selwakeup() must be done.
1048 */
1049 int
1050 bpfpoll(dev, events, p)
1051 register dev_t dev;
1052 int events;
1053 struct proc *p;
1054 {
1055 register struct bpf_d *d = &bpf_dtab[minor(dev)];
1056 int revents = 0;
1057 register int s = splimp();
1058
1059 /*
1060 * An imitation of the FIONREAD ioctl code.
1061 */
1062 if (events & (POLLIN | POLLRDNORM))
1063 if (d->bd_hlen != 0 || (d->bd_immediate && d->bd_slen != 0))
1064 revents |= events & (POLLIN | POLLRDNORM);
1065 else
1066 selrecord(p, &d->bd_sel);
1067
1068 splx(s);
1069 return (revents);
1070 }
1071
1072 /*
1073 * Incoming linkage from device drivers. Process the packet pkt, of length
1074 * pktlen, which is stored in a contiguous buffer. The packet is parsed
1075 * by each process' filter, and if accepted, stashed into the corresponding
1076 * buffer.
1077 */
1078 void
1079 bpf_tap(arg, pkt, pktlen)
1080 caddr_t arg;
1081 register u_char *pkt;
1082 register u_int pktlen;
1083 {
1084 struct bpf_if *bp;
1085 register struct bpf_d *d;
1086 register u_int slen;
1087 /*
1088 * Note that the ipl does not have to be raised at this point.
1089 * The only problem that could arise here is that if two different
1090 * interfaces shared any data. This is not the case.
1091 */
1092 bp = (struct bpf_if *)arg;
1093 for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1094 ++d->bd_rcount;
1095 slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen);
1096 if (slen != 0)
1097 catchpacket(d, pkt, pktlen, slen, bcopy);
1098 }
1099 }
1100
1101 /*
1102 * Copy data from an mbuf chain into a buffer. This code is derived
1103 * from m_copydata in sys/uipc_mbuf.c.
1104 */
1105 static void
1106 bpf_mcopy(src_arg, dst_arg, len)
1107 const void *src_arg;
1108 void *dst_arg;
1109 register size_t len;
1110 {
1111 register const struct mbuf *m;
1112 register u_int count;
1113 u_char *dst;
1114
1115 m = src_arg;
1116 dst = dst_arg;
1117 while (len > 0) {
1118 if (m == 0)
1119 panic("bpf_mcopy");
1120 count = min(m->m_len, len);
1121 bcopy(mtod(m, caddr_t), (caddr_t)dst, count);
1122 m = m->m_next;
1123 dst += count;
1124 len -= count;
1125 }
1126 }
1127
1128 /*
1129 * Incoming linkage from device drivers, when packet is in an mbuf chain.
1130 */
1131 void
1132 bpf_mtap(arg, m)
1133 caddr_t arg;
1134 struct mbuf *m;
1135 {
1136 struct bpf_if *bp = (struct bpf_if *)arg;
1137 struct bpf_d *d;
1138 u_int pktlen, slen;
1139 struct mbuf *m0;
1140
1141 pktlen = 0;
1142 for (m0 = m; m0 != 0; m0 = m0->m_next)
1143 pktlen += m0->m_len;
1144
1145 for (d = bp->bif_dlist; d != 0; d = d->bd_next) {
1146 ++d->bd_rcount;
1147 slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0);
1148 if (slen != 0)
1149 catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy);
1150 }
1151 }
1152
1153 /*
1154 * Move the packet data from interface memory (pkt) into the
1155 * store buffer. Return 1 if it's time to wakeup a listener (buffer full),
1156 * otherwise 0. "copy" is the routine called to do the actual data
1157 * transfer. bcopy is passed in to copy contiguous chunks, while
1158 * bpf_mcopy is passed in to copy mbuf chains. In the latter case,
1159 * pkt is really an mbuf.
1160 */
1161 static void
1162 catchpacket(d, pkt, pktlen, snaplen, cpfn)
1163 register struct bpf_d *d;
1164 register u_char *pkt;
1165 register u_int pktlen, snaplen;
1166 register void (*cpfn) __P((const void *, void *, size_t));
1167 {
1168 register struct bpf_hdr *hp;
1169 register int totlen, curlen;
1170 register int hdrlen = d->bd_bif->bif_hdrlen;
1171 /*
1172 * Figure out how many bytes to move. If the packet is
1173 * greater or equal to the snapshot length, transfer that
1174 * much. Otherwise, transfer the whole packet (unless
1175 * we hit the buffer size limit).
1176 */
1177 totlen = hdrlen + min(snaplen, pktlen);
1178 if (totlen > d->bd_bufsize)
1179 totlen = d->bd_bufsize;
1180
1181 /*
1182 * Round up the end of the previous packet to the next longword.
1183 */
1184 curlen = BPF_WORDALIGN(d->bd_slen);
1185 if (curlen + totlen > d->bd_bufsize) {
1186 /*
1187 * This packet will overflow the storage buffer.
1188 * Rotate the buffers if we can, then wakeup any
1189 * pending reads.
1190 */
1191 if (d->bd_fbuf == 0) {
1192 /*
1193 * We haven't completed the previous read yet,
1194 * so drop the packet.
1195 */
1196 ++d->bd_dcount;
1197 return;
1198 }
1199 ROTATE_BUFFERS(d);
1200 bpf_wakeup(d);
1201 curlen = 0;
1202 }
1203 else if (d->bd_immediate)
1204 /*
1205 * Immediate mode is set. A packet arrived so any
1206 * reads should be woken up.
1207 */
1208 bpf_wakeup(d);
1209
1210 /*
1211 * Append the bpf header.
1212 */
1213 hp = (struct bpf_hdr *)(d->bd_sbuf + curlen);
1214 #if BSD >= 199103
1215 microtime(&hp->bh_tstamp);
1216 #elif defined(sun)
1217 uniqtime(&hp->bh_tstamp);
1218 #else
1219 hp->bh_tstamp = time;
1220 #endif
1221 hp->bh_datalen = pktlen;
1222 hp->bh_hdrlen = hdrlen;
1223 /*
1224 * Copy the packet data into the store buffer and update its length.
1225 */
1226 (*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen));
1227 d->bd_slen = curlen + totlen;
1228 }
1229
1230 /*
1231 * Initialize all nonzero fields of a descriptor.
1232 */
1233 static int
1234 bpf_allocbufs(d)
1235 register struct bpf_d *d;
1236 {
1237 d->bd_fbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1238 if (d->bd_fbuf == 0)
1239 return (ENOBUFS);
1240
1241 d->bd_sbuf = (caddr_t)malloc(d->bd_bufsize, M_DEVBUF, M_WAITOK);
1242 if (d->bd_sbuf == 0) {
1243 free(d->bd_fbuf, M_DEVBUF);
1244 return (ENOBUFS);
1245 }
1246 d->bd_slen = 0;
1247 d->bd_hlen = 0;
1248 return (0);
1249 }
1250
1251 /*
1252 * Free buffers currently in use by a descriptor.
1253 * Called on close.
1254 */
1255 static void
1256 bpf_freed(d)
1257 register struct bpf_d *d;
1258 {
1259 /*
1260 * We don't need to lock out interrupts since this descriptor has
1261 * been detached from its interface and it yet hasn't been marked
1262 * free.
1263 */
1264 if (d->bd_sbuf != 0) {
1265 free(d->bd_sbuf, M_DEVBUF);
1266 if (d->bd_hbuf != 0)
1267 free(d->bd_hbuf, M_DEVBUF);
1268 if (d->bd_fbuf != 0)
1269 free(d->bd_fbuf, M_DEVBUF);
1270 }
1271 if (d->bd_filter)
1272 free((caddr_t)d->bd_filter, M_DEVBUF);
1273
1274 D_MARKFREE(d);
1275 }
1276
1277 /*
1278 * Attach an interface to bpf. driverp is a pointer to a (struct bpf_if *)
1279 * in the driver's softc; dlt is the link layer type; hdrlen is the fixed
1280 * size of the link header (variable length headers not yet supported).
1281 */
1282 void
1283 bpfattach(driverp, ifp, dlt, hdrlen)
1284 caddr_t *driverp;
1285 struct ifnet *ifp;
1286 u_int dlt, hdrlen;
1287 {
1288 struct bpf_if *bp;
1289 int i;
1290 #if BSD < 199103
1291 static struct bpf_if bpf_ifs[NBPFILTER];
1292 static int bpfifno;
1293
1294 bp = (bpfifno < NBPFILTER) ? &bpf_ifs[bpfifno++] : 0;
1295 #else
1296 bp = (struct bpf_if *)malloc(sizeof(*bp), M_DEVBUF, M_DONTWAIT);
1297 #endif
1298 if (bp == 0)
1299 panic("bpfattach");
1300
1301 bp->bif_dlist = 0;
1302 bp->bif_driverp = (struct bpf_if **)driverp;
1303 bp->bif_ifp = ifp;
1304 bp->bif_dlt = dlt;
1305
1306 bp->bif_next = bpf_iflist;
1307 bpf_iflist = bp;
1308
1309 *bp->bif_driverp = 0;
1310
1311 /*
1312 * Compute the length of the bpf header. This is not necessarily
1313 * equal to SIZEOF_BPF_HDR because we want to insert spacing such
1314 * that the network layer header begins on a longword boundary (for
1315 * performance reasons and to alleviate alignment restrictions).
1316 */
1317 bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen;
1318
1319 /*
1320 * Mark all the descriptors free if this hasn't been done.
1321 */
1322 if (!D_ISFREE(&bpf_dtab[0]))
1323 for (i = 0; i < NBPFILTER; ++i)
1324 D_MARKFREE(&bpf_dtab[i]);
1325
1326 #if 0
1327 printf("bpf: %s attached\n", ifp->if_xname);
1328 #endif
1329 }
1330
1331 #if BSD >= 199103
1332 /* XXX This routine belongs in net/if.c. */
1333 /*
1334 * Set/clear promiscuous mode on interface ifp based on the truth value
1335 * of pswitch. The calls are reference counted so that only the first
1336 * "on" request actually has an effect, as does the final "off" request.
1337 * Results are undefined if the "off" and "on" requests are not matched.
1338 */
1339 int
1340 ifpromisc(ifp, pswitch)
1341 register struct ifnet *ifp;
1342 register int pswitch;
1343 {
1344 register int pcount, ret;
1345 register short flags;
1346 struct ifreq ifr;
1347
1348 pcount = ifp->if_pcount;
1349 flags = ifp->if_flags;
1350 if (pswitch) {
1351 /*
1352 * If the device is not configured up, we cannot put it in
1353 * promiscuous mode.
1354 */
1355 if ((ifp->if_flags & IFF_UP) == 0)
1356 return (ENETDOWN);
1357 if (ifp->if_pcount++ != 0)
1358 return (0);
1359 ifp->if_flags |= IFF_PROMISC;
1360 } else {
1361 if (--ifp->if_pcount > 0)
1362 return (0);
1363 ifp->if_flags &= ~IFF_PROMISC;
1364 /*
1365 * If the device is not configured up, we should not need to
1366 * turn off promiscuous mode (device should have turned it
1367 * off when interface went down; and will look at IFF_PROMISC
1368 * again next time interface comes up).
1369 */
1370 if ((ifp->if_flags & IFF_UP) == 0)
1371 return (0);
1372 }
1373 bzero((caddr_t)&ifr, sizeof(ifr));
1374 ifr.ifr_flags = ifp->if_flags;
1375 ret = (*ifp->if_ioctl)(ifp, SIOCSIFFLAGS, (caddr_t)&ifr);
1376 /* Restore interface state if not successful */
1377 if (ret != 0) {
1378 ifp->if_pcount = pcount;
1379 ifp->if_flags = flags;
1380 }
1381 return (ret);
1382 }
1383 #endif
1384
1385 #if BSD < 199103
1386 /*
1387 * Allocate some memory for bpf. This is temporary SunOS support, and
1388 * is admittedly a hack.
1389 * If resources unavailable, return 0.
1390 */
1391 static caddr_t
1392 bpf_alloc(size, canwait)
1393 register int size;
1394 register int canwait;
1395 {
1396 register struct mbuf *m;
1397
1398 if ((unsigned)size > (MCLBYTES-8))
1399 return 0;
1400
1401 MGET(m, canwait, MT_DATA);
1402 if (m == 0)
1403 return 0;
1404 if ((unsigned)size > (MLEN-8)) {
1405 MCLGET(m);
1406 if (m->m_len != MCLBYTES) {
1407 m_freem(m);
1408 return 0;
1409 }
1410 }
1411 *mtod(m, struct mbuf **) = m;
1412 return mtod(m, caddr_t) + 8;
1413 }
1414 #endif
1415