1 /* 2 * Copyright (c) 1990, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from the Stanford/CMU enet packet filter, 6 * (net/enet.c) distributed as part of 4.3BSD, and code contributed 7 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence 8 * Berkeley Laboratory. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the University of 21 * California, Berkeley and its contributors. 22 * 4. Neither the name of the University nor the names of its contributors 23 * may be used to endorse or promote products derived from this software 24 * without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 36 * SUCH DAMAGE. 37 * 38 * @(#)bpf.c 8.2 (Berkeley) 3/28/94 39 * 40 * $FreeBSD: src/sys/net/bpf.c,v 1.59.2.12 2002/04/14 21:41:48 luigi Exp $ 41 * $DragonFly: src/sys/net/bpf.c,v 1.30 2006/06/13 08:12:03 dillon Exp $ 42 */ 43 44 #include "use_bpf.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/conf.h> 49 #include <sys/malloc.h> 50 #include <sys/mbuf.h> 51 #include <sys/time.h> 52 #include <sys/proc.h> 53 #include <sys/signalvar.h> 54 #include <sys/filio.h> 55 #include <sys/sockio.h> 56 #include <sys/ttycom.h> 57 #include <sys/filedesc.h> 58 59 #include <sys/poll.h> 60 61 #include <sys/socket.h> 62 #include <sys/vnode.h> 63 64 #include <sys/thread2.h> 65 66 #include <net/if.h> 67 #include <net/bpf.h> 68 #include <net/bpfdesc.h> 69 70 #include <netinet/in.h> 71 #include <netinet/if_ether.h> 72 #include <sys/kernel.h> 73 #include <sys/sysctl.h> 74 75 MALLOC_DEFINE(M_BPF, "BPF", "BPF data"); 76 77 #if NBPF > 0 78 79 /* 80 * The default read buffer size is patchable. 81 */ 82 static int bpf_bufsize = BPF_DEFAULTBUFSIZE; 83 SYSCTL_INT(_debug, OID_AUTO, bpf_bufsize, CTLFLAG_RW, 84 &bpf_bufsize, 0, ""); 85 static int bpf_maxbufsize = BPF_MAXBUFSIZE; 86 SYSCTL_INT(_debug, OID_AUTO, bpf_maxbufsize, CTLFLAG_RW, 87 &bpf_maxbufsize, 0, ""); 88 89 /* 90 * bpf_iflist is the list of interfaces; each corresponds to an ifnet 91 */ 92 static struct bpf_if *bpf_iflist; 93 94 static int bpf_allocbufs(struct bpf_d *); 95 static void bpf_attachd(struct bpf_d *d, struct bpf_if *bp); 96 static void bpf_detachd(struct bpf_d *d); 97 static void bpf_freed(struct bpf_d *); 98 static void bpf_mcopy(const void *, void *, size_t); 99 static int bpf_movein(struct uio *, int, struct mbuf **, 100 struct sockaddr *, int *); 101 static int bpf_setif(struct bpf_d *, struct ifreq *); 102 static void bpf_timed_out(void *); 103 static void bpf_wakeup(struct bpf_d *); 104 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 105 void (*)(const void *, void *, size_t)); 106 static void reset_d(struct bpf_d *); 107 static int bpf_setf(struct bpf_d *, struct bpf_program *); 108 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 109 static int bpf_setdlt(struct bpf_d *, u_int); 110 static void bpf_drvinit(void *unused); 111 112 static d_open_t bpfopen; 113 static d_close_t bpfclose; 114 static d_read_t bpfread; 115 static d_write_t bpfwrite; 116 static d_ioctl_t bpfioctl; 117 static d_poll_t bpfpoll; 118 119 #define CDEV_MAJOR 23 120 static struct cdevsw bpf_cdevsw = { 121 /* name */ "bpf", 122 /* maj */ CDEV_MAJOR, 123 /* flags */ 0, 124 /* port */ NULL, 125 /* clone */ NULL, 126 127 /* open */ bpfopen, 128 /* close */ bpfclose, 129 /* read */ bpfread, 130 /* write */ bpfwrite, 131 /* ioctl */ bpfioctl, 132 /* poll */ bpfpoll, 133 /* mmap */ nommap, 134 /* strategy */ nostrategy, 135 /* dump */ nodump, 136 /* psize */ nopsize 137 }; 138 139 140 static int 141 bpf_movein(struct uio *uio, int linktype, struct mbuf **mp, 142 struct sockaddr *sockp, int *datlen) 143 { 144 struct mbuf *m; 145 int error; 146 int len; 147 int hlen; 148 149 /* 150 * Build a sockaddr based on the data link layer type. 151 * We do this at this level because the ethernet header 152 * is copied directly into the data field of the sockaddr. 153 * In the case of SLIP, there is no header and the packet 154 * is forwarded as is. 155 * Also, we are careful to leave room at the front of the mbuf 156 * for the link level header. 157 */ 158 switch (linktype) { 159 160 case DLT_SLIP: 161 sockp->sa_family = AF_INET; 162 hlen = 0; 163 break; 164 165 case DLT_EN10MB: 166 sockp->sa_family = AF_UNSPEC; 167 /* XXX Would MAXLINKHDR be better? */ 168 hlen = sizeof(struct ether_header); 169 break; 170 171 case DLT_FDDI: 172 sockp->sa_family = AF_IMPLINK; 173 hlen = 0; 174 break; 175 176 case DLT_RAW: 177 case DLT_NULL: 178 sockp->sa_family = AF_UNSPEC; 179 hlen = 0; 180 break; 181 182 case DLT_ATM_RFC1483: 183 /* 184 * en atm driver requires 4-byte atm pseudo header. 185 * though it isn't standard, vpi:vci needs to be 186 * specified anyway. 187 */ 188 sockp->sa_family = AF_UNSPEC; 189 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ 190 break; 191 192 case DLT_PPP: 193 sockp->sa_family = AF_UNSPEC; 194 hlen = 4; /* This should match PPP_HDRLEN */ 195 break; 196 197 default: 198 return(EIO); 199 } 200 201 len = uio->uio_resid; 202 *datlen = len - hlen; 203 if ((unsigned)len > MCLBYTES) 204 return(EIO); 205 206 MGETHDR(m, MB_WAIT, MT_DATA); 207 if (m == NULL) 208 return(ENOBUFS); 209 if (len > MHLEN) { 210 MCLGET(m, MB_WAIT); 211 if (!(m->m_flags & M_EXT)) { 212 error = ENOBUFS; 213 goto bad; 214 } 215 } 216 m->m_pkthdr.len = m->m_len = len; 217 m->m_pkthdr.rcvif = NULL; 218 *mp = m; 219 /* 220 * Make room for link header. 221 */ 222 if (hlen != 0) { 223 m->m_pkthdr.len -= hlen; 224 m->m_len -= hlen; 225 m->m_data += hlen; /* XXX */ 226 error = uiomove(sockp->sa_data, hlen, uio); 227 if (error) 228 goto bad; 229 } 230 error = uiomove(mtod(m, caddr_t), len - hlen, uio); 231 if (!error) 232 return(0); 233 bad: 234 m_freem(m); 235 return(error); 236 } 237 238 /* 239 * Attach file to the bpf interface, i.e. make d listen on bp. 240 * Must be called at splimp. 241 */ 242 static void 243 bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 244 { 245 /* 246 * Point d at bp, and add d to the interface's list of listeners. 247 * Finally, point the driver's bpf cookie at the interface so 248 * it will divert packets to bpf. 249 */ 250 d->bd_bif = bp; 251 SLIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 252 *bp->bif_driverp = bp; 253 } 254 255 /* 256 * Detach a file from its interface. 257 */ 258 static void 259 bpf_detachd(struct bpf_d *d) 260 { 261 int error; 262 struct bpf_if *bp; 263 struct ifnet *ifp; 264 265 bp = d->bd_bif; 266 ifp = bp->bif_ifp; 267 268 /* Remove d from the interface's descriptor list. */ 269 SLIST_REMOVE(&bp->bif_dlist, d, bpf_d, bd_next); 270 271 if (SLIST_EMPTY(&bp->bif_dlist)) { 272 /* 273 * Let the driver know that there are no more listeners. 274 */ 275 *bp->bif_driverp = NULL; 276 } 277 d->bd_bif = NULL; 278 /* 279 * Check if this descriptor had requested promiscuous mode. 280 * If so, turn it off. 281 */ 282 if (d->bd_promisc) { 283 d->bd_promisc = 0; 284 error = ifpromisc(ifp, 0); 285 if (error != 0 && error != ENXIO) { 286 /* 287 * ENXIO can happen if a pccard is unplugged, 288 * Something is really wrong if we were able to put 289 * the driver into promiscuous mode, but can't 290 * take it out. 291 */ 292 if_printf(ifp, "bpf_detach: ifpromisc failed(%d)\n", 293 error); 294 } 295 } 296 } 297 298 /* 299 * Open ethernet device. Returns ENXIO for illegal minor device number, 300 * EBUSY if file is open by another process. 301 */ 302 /* ARGSUSED */ 303 static int 304 bpfopen(dev_t dev, int flags, int fmt, struct thread *td) 305 { 306 struct bpf_d *d; 307 struct proc *p = td->td_proc; 308 309 KKASSERT(p != NULL); 310 311 if (p->p_ucred->cr_prison) 312 return(EPERM); 313 314 d = dev->si_drv1; 315 /* 316 * Each minor can be opened by only one process. If the requested 317 * minor is in use, return EBUSY. 318 */ 319 if (d != NULL) 320 return(EBUSY); 321 make_dev(&bpf_cdevsw, minor(dev), 0, 0, 0600, "bpf%d", lminor(dev)); 322 MALLOC(d, struct bpf_d *, sizeof *d, M_BPF, M_WAITOK | M_ZERO); 323 dev->si_drv1 = d; 324 d->bd_bufsize = bpf_bufsize; 325 d->bd_sig = SIGIO; 326 d->bd_seesent = 1; 327 callout_init(&d->bd_callout); 328 return(0); 329 } 330 331 /* 332 * Close the descriptor by detaching it from its interface, 333 * deallocating its buffers, and marking it free. 334 */ 335 /* ARGSUSED */ 336 static int 337 bpfclose(dev_t dev, int flags, int fmt, struct thread *td) 338 { 339 struct bpf_d *d = dev->si_drv1; 340 341 funsetown(d->bd_sigio); 342 crit_enter(); 343 if (d->bd_state == BPF_WAITING) 344 callout_stop(&d->bd_callout); 345 d->bd_state = BPF_IDLE; 346 if (d->bd_bif != NULL) 347 bpf_detachd(d); 348 crit_exit(); 349 bpf_freed(d); 350 dev->si_drv1 = NULL; 351 free(d, M_BPF); 352 353 return(0); 354 } 355 356 /* 357 * Rotate the packet buffers in descriptor d. Move the store buffer 358 * into the hold slot, and the free buffer into the store slot. 359 * Zero the length of the new store buffer. 360 */ 361 #define ROTATE_BUFFERS(d) \ 362 (d)->bd_hbuf = (d)->bd_sbuf; \ 363 (d)->bd_hlen = (d)->bd_slen; \ 364 (d)->bd_sbuf = (d)->bd_fbuf; \ 365 (d)->bd_slen = 0; \ 366 (d)->bd_fbuf = NULL; 367 /* 368 * bpfread - read next chunk of packets from buffers 369 */ 370 static int 371 bpfread(dev_t dev, struct uio *uio, int ioflag) 372 { 373 struct bpf_d *d = dev->si_drv1; 374 int timed_out; 375 int error; 376 377 /* 378 * Restrict application to use a buffer the same size as 379 * as kernel buffers. 380 */ 381 if (uio->uio_resid != d->bd_bufsize) 382 return(EINVAL); 383 384 crit_enter(); 385 if (d->bd_state == BPF_WAITING) 386 callout_stop(&d->bd_callout); 387 timed_out = (d->bd_state == BPF_TIMED_OUT); 388 d->bd_state = BPF_IDLE; 389 /* 390 * If the hold buffer is empty, then do a timed sleep, which 391 * ends when the timeout expires or when enough packets 392 * have arrived to fill the store buffer. 393 */ 394 while (d->bd_hbuf == NULL) { 395 if ((d->bd_immediate || timed_out) && d->bd_slen != 0) { 396 /* 397 * A packet(s) either arrived since the previous 398 * read or arrived while we were asleep. 399 * Rotate the buffers and return what's here. 400 */ 401 ROTATE_BUFFERS(d); 402 break; 403 } 404 405 /* 406 * No data is available, check to see if the bpf device 407 * is still pointed at a real interface. If not, return 408 * ENXIO so that the userland process knows to rebind 409 * it before using it again. 410 */ 411 if (d->bd_bif == NULL) { 412 crit_exit(); 413 return(ENXIO); 414 } 415 416 if (ioflag & IO_NDELAY) { 417 crit_exit(); 418 return(EWOULDBLOCK); 419 } 420 error = tsleep(d, PCATCH, "bpf", d->bd_rtout); 421 if (error == EINTR || error == ERESTART) { 422 crit_exit(); 423 return(error); 424 } 425 if (error == EWOULDBLOCK) { 426 /* 427 * On a timeout, return what's in the buffer, 428 * which may be nothing. If there is something 429 * in the store buffer, we can rotate the buffers. 430 */ 431 if (d->bd_hbuf) 432 /* 433 * We filled up the buffer in between 434 * getting the timeout and arriving 435 * here, so we don't need to rotate. 436 */ 437 break; 438 439 if (d->bd_slen == 0) { 440 crit_exit(); 441 return(0); 442 } 443 ROTATE_BUFFERS(d); 444 break; 445 } 446 } 447 /* 448 * At this point, we know we have something in the hold slot. 449 */ 450 crit_exit(); 451 452 /* 453 * Move data from hold buffer into user space. 454 * We know the entire buffer is transferred since 455 * we checked above that the read buffer is bpf_bufsize bytes. 456 */ 457 error = uiomove(d->bd_hbuf, d->bd_hlen, uio); 458 459 crit_enter(); 460 d->bd_fbuf = d->bd_hbuf; 461 d->bd_hbuf = NULL; 462 d->bd_hlen = 0; 463 crit_exit(); 464 465 return(error); 466 } 467 468 469 /* 470 * If there are processes sleeping on this descriptor, wake them up. 471 */ 472 static void 473 bpf_wakeup(struct bpf_d *d) 474 { 475 if (d->bd_state == BPF_WAITING) { 476 callout_stop(&d->bd_callout); 477 d->bd_state = BPF_IDLE; 478 } 479 wakeup(d); 480 if (d->bd_async && d->bd_sig && d->bd_sigio) 481 pgsigio(d->bd_sigio, d->bd_sig, 0); 482 483 get_mplock(); 484 selwakeup(&d->bd_sel); 485 rel_mplock(); 486 /* XXX */ 487 d->bd_sel.si_pid = 0; 488 } 489 490 static void 491 bpf_timed_out(void *arg) 492 { 493 struct bpf_d *d = (struct bpf_d *)arg; 494 495 crit_enter(); 496 if (d->bd_state == BPF_WAITING) { 497 d->bd_state = BPF_TIMED_OUT; 498 if (d->bd_slen != 0) 499 bpf_wakeup(d); 500 } 501 crit_exit(); 502 } 503 504 static int 505 bpfwrite(dev_t dev, struct uio *uio, int ioflag) 506 { 507 struct bpf_d *d = dev->si_drv1; 508 struct ifnet *ifp; 509 struct mbuf *m; 510 int error; 511 static struct sockaddr dst; 512 int datlen; 513 514 if (d->bd_bif == NULL) 515 return(ENXIO); 516 517 ifp = d->bd_bif->bif_ifp; 518 519 if (uio->uio_resid == 0) 520 return(0); 521 522 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, &m, &dst, &datlen); 523 if (error) 524 return(error); 525 526 if (datlen > ifp->if_mtu) { 527 m_freem(m); 528 return(EMSGSIZE); 529 } 530 531 if (d->bd_hdrcmplt) 532 dst.sa_family = pseudo_AF_HDRCMPLT; 533 534 crit_enter(); 535 lwkt_serialize_enter(ifp->if_serializer); 536 error = (*ifp->if_output)(ifp, m, &dst, (struct rtentry *)NULL); 537 lwkt_serialize_exit(ifp->if_serializer); 538 crit_exit(); 539 /* 540 * The driver frees the mbuf. 541 */ 542 return(error); 543 } 544 545 /* 546 * Reset a descriptor by flushing its packet buffer and clearing the 547 * receive and drop counts. Should be called at splimp. 548 */ 549 static void 550 reset_d(struct bpf_d *d) 551 { 552 if (d->bd_hbuf) { 553 /* Free the hold buffer. */ 554 d->bd_fbuf = d->bd_hbuf; 555 d->bd_hbuf = NULL; 556 } 557 d->bd_slen = 0; 558 d->bd_hlen = 0; 559 d->bd_rcount = 0; 560 d->bd_dcount = 0; 561 } 562 563 /* 564 * FIONREAD Check for read packet available. 565 * SIOCGIFADDR Get interface address - convenient hook to driver. 566 * BIOCGBLEN Get buffer len [for read()]. 567 * BIOCSETF Set ethernet read filter. 568 * BIOCFLUSH Flush read packet buffer. 569 * BIOCPROMISC Put interface into promiscuous mode. 570 * BIOCGDLT Get link layer type. 571 * BIOCGETIF Get interface name. 572 * BIOCSETIF Set interface. 573 * BIOCSRTIMEOUT Set read timeout. 574 * BIOCGRTIMEOUT Get read timeout. 575 * BIOCGSTATS Get packet stats. 576 * BIOCIMMEDIATE Set immediate mode. 577 * BIOCVERSION Get filter language version. 578 * BIOCGHDRCMPLT Get "header already complete" flag 579 * BIOCSHDRCMPLT Set "header already complete" flag 580 * BIOCGSEESENT Get "see packets sent" flag 581 * BIOCSSEESENT Set "see packets sent" flag 582 */ 583 /* ARGSUSED */ 584 static int 585 bpfioctl(dev_t dev, u_long cmd, caddr_t addr, int flags, struct thread *td) 586 { 587 struct bpf_d *d = dev->si_drv1; 588 int error = 0; 589 590 crit_enter(); 591 if (d->bd_state == BPF_WAITING) 592 callout_stop(&d->bd_callout); 593 d->bd_state = BPF_IDLE; 594 crit_exit(); 595 596 switch (cmd) { 597 598 default: 599 error = EINVAL; 600 break; 601 602 /* 603 * Check for read packet available. 604 */ 605 case FIONREAD: 606 { 607 int n; 608 609 crit_enter(); 610 n = d->bd_slen; 611 if (d->bd_hbuf) 612 n += d->bd_hlen; 613 crit_exit(); 614 615 *(int *)addr = n; 616 break; 617 } 618 619 case SIOCGIFADDR: 620 { 621 struct ifnet *ifp; 622 623 if (d->bd_bif == NULL) 624 error = EINVAL; 625 else { 626 ifp = d->bd_bif->bif_ifp; 627 lwkt_serialize_enter(ifp->if_serializer); 628 error = ifp->if_ioctl(ifp, cmd, addr, 629 td->td_proc->p_ucred); 630 lwkt_serialize_exit(ifp->if_serializer); 631 } 632 break; 633 } 634 635 /* 636 * Get buffer len [for read()]. 637 */ 638 case BIOCGBLEN: 639 *(u_int *)addr = d->bd_bufsize; 640 break; 641 642 /* 643 * Set buffer length. 644 */ 645 case BIOCSBLEN: 646 if (d->bd_bif != 0) 647 error = EINVAL; 648 else { 649 u_int size = *(u_int *)addr; 650 651 if (size > bpf_maxbufsize) 652 *(u_int *)addr = size = bpf_maxbufsize; 653 else if (size < BPF_MINBUFSIZE) 654 *(u_int *)addr = size = BPF_MINBUFSIZE; 655 d->bd_bufsize = size; 656 } 657 break; 658 659 /* 660 * Set link layer read filter. 661 */ 662 case BIOCSETF: 663 error = bpf_setf(d, (struct bpf_program *)addr); 664 break; 665 666 /* 667 * Flush read packet buffer. 668 */ 669 case BIOCFLUSH: 670 crit_enter(); 671 reset_d(d); 672 crit_exit(); 673 break; 674 675 /* 676 * Put interface into promiscuous mode. 677 */ 678 case BIOCPROMISC: 679 if (d->bd_bif == NULL) { 680 /* 681 * No interface attached yet. 682 */ 683 error = EINVAL; 684 break; 685 } 686 crit_enter(); 687 if (d->bd_promisc == 0) { 688 error = ifpromisc(d->bd_bif->bif_ifp, 1); 689 if (error == 0) 690 d->bd_promisc = 1; 691 } 692 crit_exit(); 693 break; 694 695 /* 696 * Get device parameters. 697 */ 698 case BIOCGDLT: 699 if (d->bd_bif == NULL) 700 error = EINVAL; 701 else 702 *(u_int *)addr = d->bd_bif->bif_dlt; 703 break; 704 705 /* 706 * Get a list of supported data link types. 707 */ 708 case BIOCGDLTLIST: 709 if (d->bd_bif == NULL) 710 error = EINVAL; 711 else 712 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 713 break; 714 715 /* 716 * Set data link type. 717 */ 718 case BIOCSDLT: 719 if (d->bd_bif == NULL) 720 error = EINVAL; 721 else 722 error = bpf_setdlt(d, *(u_int *)addr); 723 break; 724 725 /* 726 * Get interface name. 727 */ 728 case BIOCGETIF: 729 if (d->bd_bif == NULL) { 730 error = EINVAL; 731 } else { 732 struct ifnet *const ifp = d->bd_bif->bif_ifp; 733 struct ifreq *const ifr = (struct ifreq *)addr; 734 735 strlcpy(ifr->ifr_name, ifp->if_xname, 736 sizeof ifr->ifr_name); 737 } 738 break; 739 740 /* 741 * Set interface. 742 */ 743 case BIOCSETIF: 744 error = bpf_setif(d, (struct ifreq *)addr); 745 break; 746 747 /* 748 * Set read timeout. 749 */ 750 case BIOCSRTIMEOUT: 751 { 752 struct timeval *tv = (struct timeval *)addr; 753 754 /* 755 * Subtract 1 tick from tvtohz() since this isn't 756 * a one-shot timer. 757 */ 758 if ((error = itimerfix(tv)) == 0) 759 d->bd_rtout = tvtohz_low(tv); 760 break; 761 } 762 763 /* 764 * Get read timeout. 765 */ 766 case BIOCGRTIMEOUT: 767 { 768 struct timeval *tv = (struct timeval *)addr; 769 770 tv->tv_sec = d->bd_rtout / hz; 771 tv->tv_usec = (d->bd_rtout % hz) * tick; 772 break; 773 } 774 775 /* 776 * Get packet stats. 777 */ 778 case BIOCGSTATS: 779 { 780 struct bpf_stat *bs = (struct bpf_stat *)addr; 781 782 bs->bs_recv = d->bd_rcount; 783 bs->bs_drop = d->bd_dcount; 784 break; 785 } 786 787 /* 788 * Set immediate mode. 789 */ 790 case BIOCIMMEDIATE: 791 d->bd_immediate = *(u_int *)addr; 792 break; 793 794 case BIOCVERSION: 795 { 796 struct bpf_version *bv = (struct bpf_version *)addr; 797 798 bv->bv_major = BPF_MAJOR_VERSION; 799 bv->bv_minor = BPF_MINOR_VERSION; 800 break; 801 } 802 803 /* 804 * Get "header already complete" flag 805 */ 806 case BIOCGHDRCMPLT: 807 *(u_int *)addr = d->bd_hdrcmplt; 808 break; 809 810 /* 811 * Set "header already complete" flag 812 */ 813 case BIOCSHDRCMPLT: 814 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 815 break; 816 817 /* 818 * Get "see sent packets" flag 819 */ 820 case BIOCGSEESENT: 821 *(u_int *)addr = d->bd_seesent; 822 break; 823 824 /* 825 * Set "see sent packets" flag 826 */ 827 case BIOCSSEESENT: 828 d->bd_seesent = *(u_int *)addr; 829 break; 830 831 case FIOASYNC: /* Send signal on receive packets */ 832 d->bd_async = *(int *)addr; 833 break; 834 835 case FIOSETOWN: 836 error = fsetown(*(int *)addr, &d->bd_sigio); 837 break; 838 839 case FIOGETOWN: 840 *(int *)addr = fgetown(d->bd_sigio); 841 break; 842 843 /* This is deprecated, FIOSETOWN should be used instead. */ 844 case TIOCSPGRP: 845 error = fsetown(-(*(int *)addr), &d->bd_sigio); 846 break; 847 848 /* This is deprecated, FIOGETOWN should be used instead. */ 849 case TIOCGPGRP: 850 *(int *)addr = -fgetown(d->bd_sigio); 851 break; 852 853 case BIOCSRSIG: /* Set receive signal */ 854 { 855 u_int sig; 856 857 sig = *(u_int *)addr; 858 859 if (sig >= NSIG) 860 error = EINVAL; 861 else 862 d->bd_sig = sig; 863 break; 864 } 865 case BIOCGRSIG: 866 *(u_int *)addr = d->bd_sig; 867 break; 868 } 869 return(error); 870 } 871 872 /* 873 * Set d's packet filter program to fp. If this file already has a filter, 874 * free it and replace it. Returns EINVAL for bogus requests. 875 */ 876 static int 877 bpf_setf(struct bpf_d *d, struct bpf_program *fp) 878 { 879 struct bpf_insn *fcode, *old; 880 u_int flen, size; 881 882 old = d->bd_filter; 883 if (fp->bf_insns == NULL) { 884 if (fp->bf_len != 0) 885 return(EINVAL); 886 crit_enter(); 887 d->bd_filter = NULL; 888 reset_d(d); 889 crit_exit(); 890 if (old != 0) 891 free(old, M_BPF); 892 return(0); 893 } 894 flen = fp->bf_len; 895 if (flen > BPF_MAXINSNS) 896 return(EINVAL); 897 898 size = flen * sizeof *fp->bf_insns; 899 fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK); 900 if (copyin(fp->bf_insns, fcode, size) == 0 && 901 bpf_validate(fcode, (int)flen)) { 902 crit_enter(); 903 d->bd_filter = fcode; 904 reset_d(d); 905 crit_exit(); 906 if (old != 0) 907 free(old, M_BPF); 908 909 return(0); 910 } 911 free(fcode, M_BPF); 912 return(EINVAL); 913 } 914 915 /* 916 * Detach a file from its current interface (if attached at all) and attach 917 * to the interface indicated by the name stored in ifr. 918 * Return an errno or 0. 919 */ 920 static int 921 bpf_setif(struct bpf_d *d, struct ifreq *ifr) 922 { 923 struct bpf_if *bp; 924 int error; 925 struct ifnet *theywant; 926 927 theywant = ifunit(ifr->ifr_name); 928 if (theywant == NULL) 929 return(ENXIO); 930 931 /* 932 * Look through attached interfaces for the named one. 933 */ 934 for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) { 935 struct ifnet *ifp = bp->bif_ifp; 936 937 if (ifp == NULL || ifp != theywant) 938 continue; 939 /* skip additional entry */ 940 if (bp->bif_driverp != &ifp->if_bpf) 941 continue; 942 /* 943 * We found the requested interface. 944 * If it's not up, return an error. 945 * Allocate the packet buffers if we need to. 946 * If we're already attached to requested interface, 947 * just flush the buffer. 948 */ 949 if (!(ifp->if_flags & IFF_UP)) 950 return(ENETDOWN); 951 952 if (d->bd_sbuf == NULL) { 953 error = bpf_allocbufs(d); 954 if (error != 0) 955 return(error); 956 } 957 crit_enter(); 958 if (bp != d->bd_bif) { 959 if (d->bd_bif != NULL) { 960 /* 961 * Detach if attached to something else. 962 */ 963 bpf_detachd(d); 964 } 965 966 bpf_attachd(d, bp); 967 } 968 reset_d(d); 969 crit_exit(); 970 return(0); 971 } 972 973 /* Not found. */ 974 return(ENXIO); 975 } 976 977 /* 978 * Support for select() and poll() system calls 979 * 980 * Return true iff the specific operation will not block indefinitely. 981 * Otherwise, return false but make a note that a selwakeup() must be done. 982 */ 983 int 984 bpfpoll(dev_t dev, int events, struct thread *td) 985 { 986 struct bpf_d *d; 987 int revents; 988 989 d = dev->si_drv1; 990 if (d->bd_bif == NULL) 991 return(ENXIO); 992 993 revents = events & (POLLOUT | POLLWRNORM); 994 crit_enter(); 995 if (events & (POLLIN | POLLRDNORM)) { 996 /* 997 * An imitation of the FIONREAD ioctl code. 998 * XXX not quite. An exact imitation: 999 * if (d->b_slen != 0 || 1000 * (d->bd_hbuf != NULL && d->bd_hlen != 0) 1001 */ 1002 if (d->bd_hlen != 0 || 1003 ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1004 d->bd_slen != 0)) 1005 revents |= events & (POLLIN | POLLRDNORM); 1006 else { 1007 selrecord(td, &d->bd_sel); 1008 /* Start the read timeout if necessary. */ 1009 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1010 callout_reset(&d->bd_callout, d->bd_rtout, 1011 bpf_timed_out, d); 1012 d->bd_state = BPF_WAITING; 1013 } 1014 } 1015 } 1016 crit_exit(); 1017 return(revents); 1018 } 1019 1020 /* 1021 * Process the packet pkt of length pktlen. The packet is parsed 1022 * by each listener's filter, and if accepted, stashed into the 1023 * corresponding buffer. 1024 */ 1025 void 1026 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 1027 { 1028 struct bpf_d *d; 1029 u_int slen; 1030 1031 /* 1032 * Note that the ipl does not have to be raised at this point. 1033 * The only problem that could arise here is that if two different 1034 * interfaces shared any data. This is not the case. 1035 */ 1036 SLIST_FOREACH(d, &bp->bif_dlist, bd_next) { 1037 ++d->bd_rcount; 1038 slen = bpf_filter(d->bd_filter, pkt, pktlen, pktlen); 1039 if (slen != 0) 1040 catchpacket(d, pkt, pktlen, slen, ovbcopy); 1041 } 1042 } 1043 1044 /* 1045 * Copy data from an mbuf chain into a buffer. This code is derived 1046 * from m_copydata in sys/uipc_mbuf.c. 1047 */ 1048 static void 1049 bpf_mcopy(const void *src_arg, void *dst_arg, size_t len) 1050 { 1051 const struct mbuf *m; 1052 u_int count; 1053 u_char *dst; 1054 1055 m = src_arg; 1056 dst = dst_arg; 1057 while (len > 0) { 1058 if (m == NULL) 1059 panic("bpf_mcopy"); 1060 count = min(m->m_len, len); 1061 bcopy(mtod(m, void *), dst, count); 1062 m = m->m_next; 1063 dst += count; 1064 len -= count; 1065 } 1066 } 1067 1068 /* 1069 * Process the packet in the mbuf chain m. The packet is parsed by each 1070 * listener's filter, and if accepted, stashed into the corresponding 1071 * buffer. 1072 */ 1073 void 1074 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 1075 { 1076 struct bpf_d *d; 1077 u_int pktlen, slen; 1078 struct mbuf *m0; 1079 1080 /* Don't compute pktlen, if no descriptor is attached. */ 1081 if (SLIST_EMPTY(&bp->bif_dlist)) 1082 return; 1083 1084 pktlen = 0; 1085 for (m0 = m; m0 != NULL; m0 = m0->m_next) 1086 pktlen += m0->m_len; 1087 1088 SLIST_FOREACH(d, &bp->bif_dlist, bd_next) { 1089 if (!d->bd_seesent && (m->m_pkthdr.rcvif == NULL)) 1090 continue; 1091 ++d->bd_rcount; 1092 slen = bpf_filter(d->bd_filter, (u_char *)m, pktlen, 0); 1093 if (slen != 0) 1094 catchpacket(d, (u_char *)m, pktlen, slen, bpf_mcopy); 1095 } 1096 } 1097 1098 void 1099 bpf_mtap_family(struct bpf_if *bp, struct mbuf *m, sa_family_t family) 1100 { 1101 u_int family4; 1102 1103 KKASSERT(family != AF_UNSPEC); 1104 1105 family4 = (u_int)family; 1106 bpf_ptap(bp, m, &family4, sizeof(family4)); 1107 } 1108 1109 /* 1110 * Process the packet in the mbuf chain m with the header in m prepended. 1111 * The packet is parsed by each listener's filter, and if accepted, 1112 * stashed into the corresponding buffer. 1113 */ 1114 void 1115 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen) 1116 { 1117 struct mbuf mb; 1118 1119 /* 1120 * Craft on-stack mbuf suitable for passing to bpf_mtap. 1121 * Note that we cut corners here; we only setup what's 1122 * absolutely needed--this mbuf should never go anywhere else. 1123 */ 1124 mb.m_next = m; 1125 mb.m_data = __DECONST(void *, data); /* LINTED */ 1126 mb.m_len = dlen; 1127 1128 bpf_mtap(bp, &mb); 1129 } 1130 1131 /* 1132 * Move the packet data from interface memory (pkt) into the 1133 * store buffer. Return 1 if it's time to wakeup a listener (buffer full), 1134 * otherwise 0. "copy" is the routine called to do the actual data 1135 * transfer. bcopy is passed in to copy contiguous chunks, while 1136 * bpf_mcopy is passed in to copy mbuf chains. In the latter case, 1137 * pkt is really an mbuf. 1138 */ 1139 static void 1140 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 1141 void (*cpfn)(const void *, void *, size_t)) 1142 { 1143 struct bpf_hdr *hp; 1144 int totlen, curlen; 1145 int hdrlen = d->bd_bif->bif_hdrlen; 1146 /* 1147 * Figure out how many bytes to move. If the packet is 1148 * greater or equal to the snapshot length, transfer that 1149 * much. Otherwise, transfer the whole packet (unless 1150 * we hit the buffer size limit). 1151 */ 1152 totlen = hdrlen + min(snaplen, pktlen); 1153 if (totlen > d->bd_bufsize) 1154 totlen = d->bd_bufsize; 1155 1156 /* 1157 * Round up the end of the previous packet to the next longword. 1158 */ 1159 curlen = BPF_WORDALIGN(d->bd_slen); 1160 if (curlen + totlen > d->bd_bufsize) { 1161 /* 1162 * This packet will overflow the storage buffer. 1163 * Rotate the buffers if we can, then wakeup any 1164 * pending reads. 1165 */ 1166 if (d->bd_fbuf == NULL) { 1167 /* 1168 * We haven't completed the previous read yet, 1169 * so drop the packet. 1170 */ 1171 ++d->bd_dcount; 1172 return; 1173 } 1174 ROTATE_BUFFERS(d); 1175 bpf_wakeup(d); 1176 curlen = 0; 1177 } 1178 else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 1179 /* 1180 * Immediate mode is set, or the read timeout has 1181 * already expired during a select call. A packet 1182 * arrived, so the reader should be woken up. 1183 */ 1184 bpf_wakeup(d); 1185 1186 /* 1187 * Append the bpf header. 1188 */ 1189 hp = (struct bpf_hdr *)(d->bd_sbuf + curlen); 1190 microtime(&hp->bh_tstamp); 1191 hp->bh_datalen = pktlen; 1192 hp->bh_hdrlen = hdrlen; 1193 /* 1194 * Copy the packet data into the store buffer and update its length. 1195 */ 1196 (*cpfn)(pkt, (u_char *)hp + hdrlen, (hp->bh_caplen = totlen - hdrlen)); 1197 d->bd_slen = curlen + totlen; 1198 } 1199 1200 /* 1201 * Initialize all nonzero fields of a descriptor. 1202 */ 1203 static int 1204 bpf_allocbufs(struct bpf_d *d) 1205 { 1206 d->bd_fbuf = malloc(d->bd_bufsize, M_BPF, M_WAITOK); 1207 if (d->bd_fbuf == NULL) 1208 return(ENOBUFS); 1209 1210 d->bd_sbuf = malloc(d->bd_bufsize, M_BPF, M_WAITOK); 1211 if (d->bd_sbuf == NULL) { 1212 free(d->bd_fbuf, M_BPF); 1213 return(ENOBUFS); 1214 } 1215 d->bd_slen = 0; 1216 d->bd_hlen = 0; 1217 return(0); 1218 } 1219 1220 /* 1221 * Free buffers currently in use by a descriptor. 1222 * Called on close. 1223 */ 1224 static void 1225 bpf_freed(struct bpf_d *d) 1226 { 1227 /* 1228 * We don't need to lock out interrupts since this descriptor has 1229 * been detached from its interface and it yet hasn't been marked 1230 * free. 1231 */ 1232 if (d->bd_sbuf != NULL) { 1233 free(d->bd_sbuf, M_BPF); 1234 if (d->bd_hbuf != NULL) 1235 free(d->bd_hbuf, M_BPF); 1236 if (d->bd_fbuf != NULL) 1237 free(d->bd_fbuf, M_BPF); 1238 } 1239 if (d->bd_filter) 1240 free(d->bd_filter, M_BPF); 1241 } 1242 1243 /* 1244 * Attach an interface to bpf. ifp is a pointer to the structure 1245 * defining the interface to be attached, dlt is the link layer type, 1246 * and hdrlen is the fixed size of the link header (variable length 1247 * headers are not yet supported). 1248 */ 1249 void 1250 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 1251 { 1252 bpfattach_dlt(ifp, dlt, hdrlen, &ifp->if_bpf); 1253 } 1254 1255 void 1256 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 1257 { 1258 struct bpf_if *bp; 1259 1260 bp = malloc(sizeof *bp, M_BPF, M_WAITOK | M_ZERO); 1261 1262 SLIST_INIT(&bp->bif_dlist); 1263 bp->bif_ifp = ifp; 1264 bp->bif_dlt = dlt; 1265 bp->bif_driverp = driverp; 1266 *bp->bif_driverp = NULL; 1267 1268 bp->bif_next = bpf_iflist; 1269 bpf_iflist = bp; 1270 1271 /* 1272 * Compute the length of the bpf header. This is not necessarily 1273 * equal to SIZEOF_BPF_HDR because we want to insert spacing such 1274 * that the network layer header begins on a longword boundary (for 1275 * performance reasons and to alleviate alignment restrictions). 1276 */ 1277 bp->bif_hdrlen = BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen; 1278 1279 if (bootverbose) 1280 if_printf(ifp, "bpf attached\n"); 1281 } 1282 1283 /* 1284 * Detach bpf from an interface. This involves detaching each descriptor 1285 * associated with the interface, and leaving bd_bif NULL. Notify each 1286 * descriptor as it's detached so that any sleepers wake up and get 1287 * ENXIO. 1288 */ 1289 void 1290 bpfdetach(struct ifnet *ifp) 1291 { 1292 struct bpf_if *bp, *bp_prev; 1293 struct bpf_d *d; 1294 1295 crit_enter(); 1296 1297 /* Locate BPF interface information */ 1298 bp_prev = NULL; 1299 for (bp = bpf_iflist; bp != NULL; bp = bp->bif_next) { 1300 if (ifp == bp->bif_ifp) 1301 break; 1302 bp_prev = bp; 1303 } 1304 1305 /* Interface wasn't attached */ 1306 if (bp->bif_ifp == NULL) { 1307 crit_exit(); 1308 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 1309 return; 1310 } 1311 1312 while ((d = SLIST_FIRST(&bp->bif_dlist)) != NULL) { 1313 bpf_detachd(d); 1314 bpf_wakeup(d); 1315 } 1316 1317 if (bp_prev != NULL) 1318 bp_prev->bif_next = bp->bif_next; 1319 else 1320 bpf_iflist = bp->bif_next; 1321 1322 free(bp, M_BPF); 1323 1324 crit_exit(); 1325 } 1326 1327 /* 1328 * Get a list of available data link type of the interface. 1329 */ 1330 static int 1331 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 1332 { 1333 int n, error; 1334 struct ifnet *ifp; 1335 struct bpf_if *bp; 1336 1337 ifp = d->bd_bif->bif_ifp; 1338 n = 0; 1339 error = 0; 1340 for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) { 1341 if (bp->bif_ifp != ifp) 1342 continue; 1343 if (bfl->bfl_list != NULL) { 1344 if (n >= bfl->bfl_len) { 1345 return (ENOMEM); 1346 } 1347 error = copyout(&bp->bif_dlt, 1348 bfl->bfl_list + n, sizeof(u_int)); 1349 } 1350 n++; 1351 } 1352 bfl->bfl_len = n; 1353 return(error); 1354 } 1355 1356 /* 1357 * Set the data link type of a BPF instance. 1358 */ 1359 static int 1360 bpf_setdlt(struct bpf_d *d, u_int dlt) 1361 { 1362 int error, opromisc; 1363 struct ifnet *ifp; 1364 struct bpf_if *bp; 1365 1366 if (d->bd_bif->bif_dlt == dlt) 1367 return (0); 1368 ifp = d->bd_bif->bif_ifp; 1369 for (bp = bpf_iflist; bp != 0; bp = bp->bif_next) { 1370 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 1371 break; 1372 } 1373 if (bp != NULL) { 1374 opromisc = d->bd_promisc; 1375 crit_enter(); 1376 bpf_detachd(d); 1377 bpf_attachd(d, bp); 1378 reset_d(d); 1379 if (opromisc) { 1380 error = ifpromisc(bp->bif_ifp, 1); 1381 if (error) 1382 if_printf(bp->bif_ifp, 1383 "bpf_setdlt: ifpromisc failed (%d)\n", 1384 error); 1385 else 1386 d->bd_promisc = 1; 1387 } 1388 crit_exit(); 1389 } 1390 return(bp == NULL ? EINVAL : 0); 1391 } 1392 1393 static void 1394 bpf_drvinit(void *unused) 1395 { 1396 cdevsw_add(&bpf_cdevsw, 0, 0); 1397 } 1398 1399 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE+CDEV_MAJOR,bpf_drvinit,NULL) 1400 1401 #else /* !BPF */ 1402 /* 1403 * NOP stubs to allow bpf-using drivers to load and function. 1404 * 1405 * A 'better' implementation would allow the core bpf functionality 1406 * to be loaded at runtime. 1407 */ 1408 1409 void 1410 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 1411 { 1412 } 1413 1414 void 1415 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 1416 { 1417 } 1418 1419 void 1420 bpf_ptap(struct bpf_if *bp, struct mbuf *m, const void *data, u_int dlen) 1421 { 1422 } 1423 1424 void 1425 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 1426 { 1427 } 1428 1429 void 1430 bpfattach_dlt(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 1431 { 1432 } 1433 1434 void 1435 bpfdetach(struct ifnet *ifp) 1436 { 1437 } 1438 1439 u_int 1440 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 1441 { 1442 return -1; /* "no filter" behaviour */ 1443 } 1444 1445 #endif /* !BPF */ 1446