1/* 2 * Copyright (c) 1990, 1991, 1992, 1993, 1994, 1995, 1996 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that: (1) source code distributions 7 * retain the above copyright notice and this paragraph in its entirety, (2) 8 * distributions including binary code include the above copyright notice and 9 * this paragraph in its entirety in the documentation or other materials 10 * provided with the distribution, and (3) all advertising materials mentioning 11 * features or use of this software display the following acknowledgement: 12 * ``This product includes software developed by the University of California, 13 * Lawrence Berkeley Laboratory and its contributors.'' Neither the name of 14 * the University nor the names of its contributors may be used to endorse 15 * or promote products derived from this software without specific prior 16 * written permission. 17 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR IMPLIED 18 * WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF 19 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. 20 * 21 * packet filter subroutines for tcpdump 22 * Extraction/creation by Jeffrey Mogul, DECWRL 23 */ 24 25#ifndef lint 26static const char rcsid[] _U_ = 27 "@(#) $Header: /tcpdump/master/libpcap/pcap-pf.c,v 1.97 2008-04-14 20:40:58 guy Exp $ (LBL)"; 28#endif 29 30#ifdef HAVE_CONFIG_H 31#include "config.h" 32#endif 33 34#include <sys/types.h> 35#include <sys/time.h> 36#include <sys/timeb.h> 37#include <sys/socket.h> 38#include <sys/file.h> 39#include <sys/ioctl.h> 40#include <net/pfilt.h> 41 42struct mbuf; 43struct rtentry; 44#include <net/if.h> 45 46#include <netinet/in.h> 47#include <netinet/in_systm.h> 48#include <netinet/ip.h> 49#include <netinet/if_ether.h> 50#include <netinet/ip_var.h> 51#include <netinet/udp.h> 52#include <netinet/udp_var.h> 53#include <netinet/tcp.h> 54#include <netinet/tcpip.h> 55 56#include <ctype.h> 57#include <errno.h> 58#include <netdb.h> 59#include <stdio.h> 60#include <stdlib.h> 61#include <string.h> 62#include <unistd.h> 63 64/* 65 * Make "pcap.h" not include "pcap/bpf.h"; we are going to include the 66 * native OS version, as we need various BPF ioctls from it. 67 */ 68#define PCAP_DONT_INCLUDE_PCAP_BPF_H 69#include <net/bpf.h> 70 71#include "pcap-int.h" 72 73#ifdef HAVE_OS_PROTO_H 74#include "os-proto.h" 75#endif 76 77static int pcap_setfilter_pf(pcap_t *, struct bpf_program *); 78 79/* 80 * BUFSPACE is the size in bytes of the packet read buffer. Most tcpdump 81 * applications aren't going to need more than 200 bytes of packet header 82 * and the read shouldn't return more packets than packetfilter's internal 83 * queue limit (bounded at 256). 84 */ 85#define BUFSPACE (200 * 256) 86 87static int 88pcap_read_pf(pcap_t *pc, int cnt, pcap_handler callback, u_char *user) 89{ 90 register u_char *p, *bp; 91 register int cc, n, buflen, inc; 92 register struct enstamp *sp; 93#ifdef LBL_ALIGN 94 struct enstamp stamp; 95#endif 96#ifdef PCAP_FDDIPAD 97 register int pad; 98#endif 99 100 again: 101 cc = pc->cc; 102 if (cc == 0) { 103 cc = read(pc->fd, (char *)pc->buffer + pc->offset, pc->bufsize); 104 if (cc < 0) { 105 if (errno == EWOULDBLOCK) 106 return (0); 107 if (errno == EINVAL && 108 lseek(pc->fd, 0L, SEEK_CUR) + pc->bufsize < 0) { 109 /* 110 * Due to a kernel bug, after 2^31 bytes, 111 * the kernel file offset overflows and 112 * read fails with EINVAL. The lseek() 113 * to 0 will fix things. 114 */ 115 (void)lseek(pc->fd, 0L, SEEK_SET); 116 goto again; 117 } 118 snprintf(pc->errbuf, sizeof(pc->errbuf), "pf read: %s", 119 pcap_strerror(errno)); 120 return (-1); 121 } 122 bp = pc->buffer + pc->offset; 123 } else 124 bp = pc->bp; 125 /* 126 * Loop through each packet. 127 */ 128 n = 0; 129#ifdef PCAP_FDDIPAD 130 pad = pc->fddipad; 131#endif 132 while (cc > 0) { 133 /* 134 * Has "pcap_breakloop()" been called? 135 * If so, return immediately - if we haven't read any 136 * packets, clear the flag and return -2 to indicate 137 * that we were told to break out of the loop, otherwise 138 * leave the flag set, so that the *next* call will break 139 * out of the loop without having read any packets, and 140 * return the number of packets we've processed so far. 141 */ 142 if (pc->break_loop) { 143 if (n == 0) { 144 pc->break_loop = 0; 145 return (-2); 146 } else { 147 pc->cc = cc; 148 pc->bp = bp; 149 return (n); 150 } 151 } 152 if (cc < sizeof(*sp)) { 153 snprintf(pc->errbuf, sizeof(pc->errbuf), 154 "pf short read (%d)", cc); 155 return (-1); 156 } 157#ifdef LBL_ALIGN 158 if ((long)bp & 3) { 159 sp = &stamp; 160 memcpy((char *)sp, (char *)bp, sizeof(*sp)); 161 } else 162#endif 163 sp = (struct enstamp *)bp; 164 if (sp->ens_stamplen != sizeof(*sp)) { 165 snprintf(pc->errbuf, sizeof(pc->errbuf), 166 "pf short stamplen (%d)", 167 sp->ens_stamplen); 168 return (-1); 169 } 170 171 p = bp + sp->ens_stamplen; 172 buflen = sp->ens_count; 173 if (buflen > pc->snapshot) 174 buflen = pc->snapshot; 175 176 /* Calculate inc before possible pad update */ 177 inc = ENALIGN(buflen + sp->ens_stamplen); 178 cc -= inc; 179 bp += inc; 180 pc->md.TotPkts++; 181 pc->md.TotDrops += sp->ens_dropped; 182 pc->md.TotMissed = sp->ens_ifoverflows; 183 if (pc->md.OrigMissed < 0) 184 pc->md.OrigMissed = pc->md.TotMissed; 185 186 /* 187 * Short-circuit evaluation: if using BPF filter 188 * in kernel, no need to do it now - we already know 189 * the packet passed the filter. 190 * 191#ifdef PCAP_FDDIPAD 192 * Note: the filter code was generated assuming 193 * that pc->fddipad was the amount of padding 194 * before the header, as that's what's required 195 * in the kernel, so we run the filter before 196 * skipping that padding. 197#endif 198 */ 199 if (pc->md.use_bpf || 200 bpf_filter(pc->fcode.bf_insns, p, sp->ens_count, buflen)) { 201 struct pcap_pkthdr h; 202 pc->md.TotAccepted++; 203 h.ts = sp->ens_tstamp; 204#ifdef PCAP_FDDIPAD 205 h.len = sp->ens_count - pad; 206#else 207 h.len = sp->ens_count; 208#endif 209#ifdef PCAP_FDDIPAD 210 p += pad; 211 buflen -= pad; 212#endif 213 h.caplen = buflen; 214 (*callback)(user, &h, p); 215 if (++n >= cnt && cnt > 0) { 216 pc->cc = cc; 217 pc->bp = bp; 218 return (n); 219 } 220 } 221 } 222 pc->cc = 0; 223 return (n); 224} 225 226static int 227pcap_inject_pf(pcap_t *p, const void *buf, size_t size) 228{ 229 int ret; 230 231 ret = write(p->fd, buf, size); 232 if (ret == -1) { 233 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "send: %s", 234 pcap_strerror(errno)); 235 return (-1); 236 } 237 return (ret); 238} 239 240static int 241pcap_stats_pf(pcap_t *p, struct pcap_stat *ps) 242{ 243 244 /* 245 * If packet filtering is being done in the kernel: 246 * 247 * "ps_recv" counts only packets that passed the filter. 248 * This does not include packets dropped because we 249 * ran out of buffer space. (XXX - perhaps it should, 250 * by adding "ps_drop" to "ps_recv", for compatibility 251 * with some other platforms. On the other hand, on 252 * some platforms "ps_recv" counts only packets that 253 * passed the filter, and on others it counts packets 254 * that didn't pass the filter....) 255 * 256 * "ps_drop" counts packets that passed the kernel filter 257 * (if any) but were dropped because the input queue was 258 * full. 259 * 260 * "ps_ifdrop" counts packets dropped by the network 261 * inteface (regardless of whether they would have passed 262 * the input filter, of course). 263 * 264 * If packet filtering is not being done in the kernel: 265 * 266 * "ps_recv" counts only packets that passed the filter. 267 * 268 * "ps_drop" counts packets that were dropped because the 269 * input queue was full, regardless of whether they passed 270 * the userland filter. 271 * 272 * "ps_ifdrop" counts packets dropped by the network 273 * inteface (regardless of whether they would have passed 274 * the input filter, of course). 275 * 276 * These statistics don't include packets not yet read from 277 * the kernel by libpcap, but they may include packets not 278 * yet read from libpcap by the application. 279 */ 280 ps->ps_recv = p->md.TotAccepted; 281 ps->ps_drop = p->md.TotDrops; 282 ps->ps_ifdrop = p->md.TotMissed - p->md.OrigMissed; 283 return (0); 284} 285 286/* 287 * We include the OS's <net/bpf.h>, not our "pcap/bpf.h", so we probably 288 * don't get DLT_DOCSIS defined. 289 */ 290#ifndef DLT_DOCSIS 291#define DLT_DOCSIS 143 292#endif 293 294static int 295pcap_activate_pf(pcap_t *p) 296{ 297 short enmode; 298 int backlog = -1; /* request the most */ 299 struct enfilter Filter; 300 struct endevp devparams; 301 302 /* 303 * Initially try a read/write open (to allow the inject 304 * method to work). If that fails due to permission 305 * issues, fall back to read-only. This allows a 306 * non-root user to be granted specific access to pcap 307 * capabilities via file permissions. 308 * 309 * XXX - we should have an API that has a flag that 310 * controls whether to open read-only or read-write, 311 * so that denial of permission to send (or inability 312 * to send, if sending packets isn't supported on 313 * the device in question) can be indicated at open 314 * time. 315 * 316 * XXX - we assume here that "pfopen()" does not, in fact, modify 317 * its argument, even though it takes a "char *" rather than a 318 * "const char *" as its first argument. That appears to be 319 * the case, at least on Digital UNIX 4.0. 320 */ 321 p->fd = pfopen(p->opt.source, O_RDWR); 322 if (p->fd == -1 && errno == EACCES) 323 p->fd = pfopen(p->opt.source, O_RDONLY); 324 if (p->fd < 0) { 325 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "pf open: %s: %s\n\ 326your system may not be properly configured; see the packetfilter(4) man page\n", 327 p->opt.source, pcap_strerror(errno)); 328 goto bad; 329 } 330 p->md.OrigMissed = -1; 331 enmode = ENTSTAMP|ENBATCH|ENNONEXCL; 332 if (p->opt.promisc) 333 enmode |= ENPROMISC; 334 if (ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode) < 0) { 335 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCMBIS: %s", 336 pcap_strerror(errno)); 337 goto bad; 338 } 339#ifdef ENCOPYALL 340 /* Try to set COPYALL mode so that we see packets to ourself */ 341 enmode = ENCOPYALL; 342 (void)ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode);/* OK if this fails */ 343#endif 344 /* set the backlog */ 345 if (ioctl(p->fd, EIOCSETW, (caddr_t)&backlog) < 0) { 346 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETW: %s", 347 pcap_strerror(errno)); 348 goto bad; 349 } 350 /* discover interface type */ 351 if (ioctl(p->fd, EIOCDEVP, (caddr_t)&devparams) < 0) { 352 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCDEVP: %s", 353 pcap_strerror(errno)); 354 goto bad; 355 } 356 /* HACK: to compile prior to Ultrix 4.2 */ 357#ifndef ENDT_FDDI 358#define ENDT_FDDI 4 359#endif 360 switch (devparams.end_dev_type) { 361 362 case ENDT_10MB: 363 p->linktype = DLT_EN10MB; 364 p->offset = 2; 365 /* 366 * This is (presumably) a real Ethernet capture; give it a 367 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so 368 * that an application can let you choose it, in case you're 369 * capturing DOCSIS traffic that a Cisco Cable Modem 370 * Termination System is putting out onto an Ethernet (it 371 * doesn't put an Ethernet header onto the wire, it puts raw 372 * DOCSIS frames out on the wire inside the low-level 373 * Ethernet framing). 374 */ 375 p->dlt_list = (u_int *) malloc(sizeof(u_int) * 2); 376 /* 377 * If that fails, just leave the list empty. 378 */ 379 if (p->dlt_list != NULL) { 380 p->dlt_list[0] = DLT_EN10MB; 381 p->dlt_list[1] = DLT_DOCSIS; 382 p->dlt_count = 2; 383 } 384 break; 385 386 case ENDT_FDDI: 387 p->linktype = DLT_FDDI; 388 break; 389 390#ifdef ENDT_SLIP 391 case ENDT_SLIP: 392 p->linktype = DLT_SLIP; 393 break; 394#endif 395 396#ifdef ENDT_PPP 397 case ENDT_PPP: 398 p->linktype = DLT_PPP; 399 break; 400#endif 401 402#ifdef ENDT_LOOPBACK 403 case ENDT_LOOPBACK: 404 /* 405 * It appears to use Ethernet framing, at least on 406 * Digital UNIX 4.0. 407 */ 408 p->linktype = DLT_EN10MB; 409 p->offset = 2; 410 break; 411#endif 412 413#ifdef ENDT_TRN 414 case ENDT_TRN: 415 p->linktype = DLT_IEEE802; 416 break; 417#endif 418 419 default: 420 /* 421 * XXX - what about ENDT_IEEE802? The pfilt.h header 422 * file calls this "IEEE 802 networks (non-Ethernet)", 423 * but that doesn't specify a specific link layer type; 424 * it could be 802.4, or 802.5 (except that 802.5 is 425 * ENDT_TRN), or 802.6, or 802.11, or.... That's why 426 * DLT_IEEE802 was hijacked to mean Token Ring in various 427 * BSDs, and why we went along with that hijacking. 428 * 429 * XXX - what about ENDT_HDLC and ENDT_NULL? 430 * Presumably, as ENDT_OTHER is just "Miscellaneous 431 * framing", there's not much we can do, as that 432 * doesn't specify a particular type of header. 433 */ 434 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, 435 "unknown data-link type %u", devparams.end_dev_type); 436 goto bad; 437 } 438 /* set truncation */ 439#ifdef PCAP_FDDIPAD 440 if (p->linktype == DLT_FDDI) { 441 p->fddipad = PCAP_FDDIPAD; 442 443 /* packetfilter includes the padding in the snapshot */ 444 p->snapshot += PCAP_FDDIPAD; 445 } else 446 p->fddipad = 0; 447#endif 448 if (ioctl(p->fd, EIOCTRUNCATE, (caddr_t)&p->snapshot) < 0) { 449 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCTRUNCATE: %s", 450 pcap_strerror(errno)); 451 goto bad; 452 } 453 /* accept all packets */ 454 memset(&Filter, 0, sizeof(Filter)); 455 Filter.enf_Priority = 37; /* anything > 2 */ 456 Filter.enf_FilterLen = 0; /* means "always true" */ 457 if (ioctl(p->fd, EIOCSETF, (caddr_t)&Filter) < 0) { 458 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETF: %s", 459 pcap_strerror(errno)); 460 goto bad; 461 } 462 463 if (p->md.timeout != 0) { 464 struct timeval timeout; 465 timeout.tv_sec = p->md.timeout / 1000; 466 timeout.tv_usec = (p->md.timeout * 1000) % 1000000; 467 if (ioctl(p->fd, EIOCSRTIMEOUT, (caddr_t)&timeout) < 0) { 468 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSRTIMEOUT: %s", 469 pcap_strerror(errno)); 470 goto bad; 471 } 472 } 473 474 p->bufsize = BUFSPACE; 475 p->buffer = (u_char*)malloc(p->bufsize + p->offset); 476 if (p->buffer == NULL) { 477 strlcpy(p->errbuf, pcap_strerror(errno), PCAP_ERRBUF_SIZE); 478 goto bad; 479 } 480 481 /* 482 * "select()" and "poll()" work on packetfilter devices. 483 */ 484 p->selectable_fd = p->fd; 485 486 p->read_op = pcap_read_pf; 487 p->inject_op = pcap_inject_pf; 488 p->setfilter_op = pcap_setfilter_pf; 489 p->setdirection_op = NULL; /* Not implemented. */ 490 p->set_datalink_op = NULL; /* can't change data link type */ 491 p->getnonblock_op = pcap_getnonblock_fd; 492 p->setnonblock_op = pcap_setnonblock_fd; 493 p->stats_op = pcap_stats_pf; 494 495 return (0); 496 bad: 497 pcap_cleanup_live_common(p); 498 return (PCAP_ERROR); 499} 500 501pcap_t * 502pcap_create_interface(const char *device, char *ebuf) 503{ 504 pcap_t *p; 505 506 p = pcap_create_common(device, ebuf); 507 if (p == NULL) 508 return (NULL); 509 510 p->activate_op = pcap_activate_pf; 511 return (p); 512} 513 514int 515pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf) 516{ 517 return (0); 518} 519 520static int 521pcap_setfilter_pf(pcap_t *p, struct bpf_program *fp) 522{ 523 struct bpf_version bv; 524 525 /* 526 * See if BIOCVERSION works. If not, we assume the kernel doesn't 527 * support BPF-style filters (it's not documented in the bpf(7) 528 * or packetfiler(7) man pages, but the code used to fail if 529 * BIOCSETF worked but BIOCVERSION didn't, and I've seen it do 530 * kernel filtering in DU 4.0, so presumably BIOCVERSION works 531 * there, at least). 532 */ 533 if (ioctl(p->fd, BIOCVERSION, (caddr_t)&bv) >= 0) { 534 /* 535 * OK, we have the version of the BPF interpreter; 536 * is it the same major version as us, and the same 537 * or better minor version? 538 */ 539 if (bv.bv_major == BPF_MAJOR_VERSION && 540 bv.bv_minor >= BPF_MINOR_VERSION) { 541 /* 542 * Yes. Try to install the filter. 543 */ 544 if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) { 545 snprintf(p->errbuf, sizeof(p->errbuf), 546 "BIOCSETF: %s", pcap_strerror(errno)); 547 return (-1); 548 } 549 550 /* 551 * OK, that succeeded. We're doing filtering in 552 * the kernel. (We assume we don't have a 553 * userland filter installed - that'd require 554 * a previous version check to have failed but 555 * this one to succeed.) 556 * 557 * XXX - this message should be supplied to the 558 * application as a warning of some sort, 559 * except that if it's a GUI application, it's 560 * not clear that it should be displayed in 561 * a window to annoy the user. 562 */ 563 fprintf(stderr, "tcpdump: Using kernel BPF filter\n"); 564 p->md.use_bpf = 1; 565 566 /* 567 * Discard any previously-received packets, 568 * as they might have passed whatever filter 569 * was formerly in effect, but might not pass 570 * this filter (BIOCSETF discards packets buffered 571 * in the kernel, so you can lose packets in any 572 * case). 573 */ 574 p->cc = 0; 575 return (0); 576 } 577 578 /* 579 * We can't use the kernel's BPF interpreter; don't give 580 * up, just log a message and be inefficient. 581 * 582 * XXX - this should really be supplied to the application 583 * as a warning of some sort. 584 */ 585 fprintf(stderr, 586 "tcpdump: Requires BPF language %d.%d or higher; kernel is %d.%d\n", 587 BPF_MAJOR_VERSION, BPF_MINOR_VERSION, 588 bv.bv_major, bv.bv_minor); 589 } 590 591 /* 592 * We couldn't do filtering in the kernel; do it in userland. 593 */ 594 if (install_bpf_program(p, fp) < 0) 595 return (-1); 596 597 /* 598 * XXX - this message should be supplied by the application as 599 * a warning of some sort. 600 */ 601 fprintf(stderr, "tcpdump: Filtering in user process\n"); 602 p->md.use_bpf = 0; 603 return (0); 604} 605