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#ifdef HAVE_CONFIG_H 26#include "config.h" 27#endif 28 29#include <sys/types.h> 30#include <sys/time.h> 31#include <sys/timeb.h> 32#include <sys/socket.h> 33#include <sys/file.h> 34#include <sys/ioctl.h> 35#include <net/pfilt.h> 36 37struct mbuf; 38struct rtentry; 39#include <net/if.h> 40 41#include <netinet/in.h> 42#include <netinet/in_systm.h> 43#include <netinet/ip.h> 44#include <netinet/if_ether.h> 45#include <netinet/ip_var.h> 46#include <netinet/udp.h> 47#include <netinet/udp_var.h> 48#include <netinet/tcp.h> 49#include <netinet/tcpip.h> 50 51#include <ctype.h> 52#include <errno.h> 53#include <netdb.h> 54#include <stdio.h> 55#include <stdlib.h> 56#include <string.h> 57#include <unistd.h> 58 59/* 60 * Make "pcap.h" not include "pcap/bpf.h"; we are going to include the 61 * native OS version, as we need various BPF ioctls from it. 62 */ 63#define PCAP_DONT_INCLUDE_PCAP_BPF_H 64#include <net/bpf.h> 65 66#include "pcap-int.h" 67 68#ifdef HAVE_OS_PROTO_H 69#include "os-proto.h" 70#endif 71 72/* 73 * FDDI packets are padded to make everything line up on a nice boundary. 74 */ 75#define PCAP_FDDIPAD 3 76 77/* 78 * Private data for capturing on Ultrix and DEC OSF/1^WDigital UNIX^W^W 79 * Tru64 UNIX packetfilter devices. 80 */ 81struct pcap_pf { 82 int filtering_in_kernel; /* using kernel filter */ 83 u_long TotPkts; /* can't oflow for 79 hrs on ether */ 84 u_long TotAccepted; /* count accepted by filter */ 85 u_long TotDrops; /* count of dropped packets */ 86 long TotMissed; /* missed by i/f during this run */ 87 long OrigMissed; /* missed by i/f before this run */ 88}; 89 90static int pcap_setfilter_pf(pcap_t *, struct bpf_program *); 91 92/* 93 * BUFSPACE is the size in bytes of the packet read buffer. Most tcpdump 94 * applications aren't going to need more than 200 bytes of packet header 95 * and the read shouldn't return more packets than packetfilter's internal 96 * queue limit (bounded at 256). 97 */ 98#define BUFSPACE (200 * 256) 99 100static int 101pcap_read_pf(pcap_t *pc, int cnt, pcap_handler callback, u_char *user) 102{ 103 struct pcap_pf *pf = pc->priv; 104 register u_char *p, *bp; 105 register int cc, n, buflen, inc; 106 register struct enstamp *sp; 107#ifdef LBL_ALIGN 108 struct enstamp stamp; 109#endif 110 register int pad; 111 112 again: 113 cc = pc->cc; 114 if (cc == 0) { 115 cc = read(pc->fd, (char *)pc->buffer + pc->offset, pc->bufsize); 116 if (cc < 0) { 117 if (errno == EWOULDBLOCK) 118 return (0); 119 if (errno == EINVAL && 120 lseek(pc->fd, 0L, SEEK_CUR) + pc->bufsize < 0) { 121 /* 122 * Due to a kernel bug, after 2^31 bytes, 123 * the kernel file offset overflows and 124 * read fails with EINVAL. The lseek() 125 * to 0 will fix things. 126 */ 127 (void)lseek(pc->fd, 0L, SEEK_SET); 128 goto again; 129 } 130 snprintf(pc->errbuf, sizeof(pc->errbuf), "pf read: %s", 131 pcap_strerror(errno)); 132 return (-1); 133 } 134 bp = pc->buffer + pc->offset; 135 } else 136 bp = pc->bp; 137 /* 138 * Loop through each packet. 139 */ 140 n = 0; 141 pad = pc->fddipad; 142 while (cc > 0) { 143 /* 144 * Has "pcap_breakloop()" been called? 145 * If so, return immediately - if we haven't read any 146 * packets, clear the flag and return -2 to indicate 147 * that we were told to break out of the loop, otherwise 148 * leave the flag set, so that the *next* call will break 149 * out of the loop without having read any packets, and 150 * return the number of packets we've processed so far. 151 */ 152 if (pc->break_loop) { 153 if (n == 0) { 154 pc->break_loop = 0; 155 return (-2); 156 } else { 157 pc->cc = cc; 158 pc->bp = bp; 159 return (n); 160 } 161 } 162 if (cc < sizeof(*sp)) { 163 snprintf(pc->errbuf, sizeof(pc->errbuf), 164 "pf short read (%d)", cc); 165 return (-1); 166 } 167#ifdef LBL_ALIGN 168 if ((long)bp & 3) { 169 sp = &stamp; 170 memcpy((char *)sp, (char *)bp, sizeof(*sp)); 171 } else 172#endif 173 sp = (struct enstamp *)bp; 174 if (sp->ens_stamplen != sizeof(*sp)) { 175 snprintf(pc->errbuf, sizeof(pc->errbuf), 176 "pf short stamplen (%d)", 177 sp->ens_stamplen); 178 return (-1); 179 } 180 181 p = bp + sp->ens_stamplen; 182 buflen = sp->ens_count; 183 if (buflen > pc->snapshot) 184 buflen = pc->snapshot; 185 186 /* Calculate inc before possible pad update */ 187 inc = ENALIGN(buflen + sp->ens_stamplen); 188 cc -= inc; 189 bp += inc; 190 pf->TotPkts++; 191 pf->TotDrops += sp->ens_dropped; 192 pf->TotMissed = sp->ens_ifoverflows; 193 if (pf->OrigMissed < 0) 194 pf->OrigMissed = pf->TotMissed; 195 196 /* 197 * Short-circuit evaluation: if using BPF filter 198 * in kernel, no need to do it now - we already know 199 * the packet passed the filter. 200 * 201 * Note: the filter code was generated assuming 202 * that pc->fddipad was the amount of padding 203 * before the header, as that's what's required 204 * in the kernel, so we run the filter before 205 * skipping that padding. 206 */ 207 if (pf->filtering_in_kernel || 208 bpf_filter(pc->fcode.bf_insns, p, sp->ens_count, buflen)) { 209 struct pcap_pkthdr h; 210 pf->TotAccepted++; 211 h.ts = sp->ens_tstamp; 212 h.len = sp->ens_count - pad; 213 p += pad; 214 buflen -= pad; 215 h.caplen = buflen; 216 (*callback)(user, &h, p); 217 if (++n >= cnt && !PACKET_COUNT_IS_UNLIMITED(cnt)) { 218 pc->cc = cc; 219 pc->bp = bp; 220 return (n); 221 } 222 } 223 } 224 pc->cc = 0; 225 return (n); 226} 227 228static int 229pcap_inject_pf(pcap_t *p, const void *buf, size_t size) 230{ 231 int ret; 232 233 ret = write(p->fd, buf, size); 234 if (ret == -1) { 235 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "send: %s", 236 pcap_strerror(errno)); 237 return (-1); 238 } 239 return (ret); 240} 241 242static int 243pcap_stats_pf(pcap_t *p, struct pcap_stat *ps) 244{ 245 struct pcap_pf *pf = p->priv; 246 247 /* 248 * If packet filtering is being done in the kernel: 249 * 250 * "ps_recv" counts only packets that passed the filter. 251 * This does not include packets dropped because we 252 * ran out of buffer space. (XXX - perhaps it should, 253 * by adding "ps_drop" to "ps_recv", for compatibility 254 * with some other platforms. On the other hand, on 255 * some platforms "ps_recv" counts only packets that 256 * passed the filter, and on others it counts packets 257 * that didn't pass the filter....) 258 * 259 * "ps_drop" counts packets that passed the kernel filter 260 * (if any) but were dropped because the input queue was 261 * full. 262 * 263 * "ps_ifdrop" counts packets dropped by the network 264 * inteface (regardless of whether they would have passed 265 * the input filter, of course). 266 * 267 * If packet filtering is not being done in the kernel: 268 * 269 * "ps_recv" counts only packets that passed the filter. 270 * 271 * "ps_drop" counts packets that were dropped because the 272 * input queue was full, regardless of whether they passed 273 * the userland filter. 274 * 275 * "ps_ifdrop" counts packets dropped by the network 276 * inteface (regardless of whether they would have passed 277 * the input filter, of course). 278 * 279 * These statistics don't include packets not yet read from 280 * the kernel by libpcap, but they may include packets not 281 * yet read from libpcap by the application. 282 */ 283 ps->ps_recv = pf->TotAccepted; 284 ps->ps_drop = pf->TotDrops; 285 ps->ps_ifdrop = pf->TotMissed - pf->OrigMissed; 286 return (0); 287} 288 289/* 290 * We include the OS's <net/bpf.h>, not our "pcap/bpf.h", so we probably 291 * don't get DLT_DOCSIS defined. 292 */ 293#ifndef DLT_DOCSIS 294#define DLT_DOCSIS 143 295#endif 296 297static int 298pcap_activate_pf(pcap_t *p) 299{ 300 struct pcap_pf *pf = p->priv; 301 short enmode; 302 int backlog = -1; /* request the most */ 303 struct enfilter Filter; 304 struct endevp devparams; 305 306 /* 307 * Initially try a read/write open (to allow the inject 308 * method to work). If that fails due to permission 309 * issues, fall back to read-only. This allows a 310 * non-root user to be granted specific access to pcap 311 * capabilities via file permissions. 312 * 313 * XXX - we should have an API that has a flag that 314 * controls whether to open read-only or read-write, 315 * so that denial of permission to send (or inability 316 * to send, if sending packets isn't supported on 317 * the device in question) can be indicated at open 318 * time. 319 * 320 * XXX - we assume here that "pfopen()" does not, in fact, modify 321 * its argument, even though it takes a "char *" rather than a 322 * "const char *" as its first argument. That appears to be 323 * the case, at least on Digital UNIX 4.0. 324 */ 325 p->fd = pfopen(p->opt.source, O_RDWR); 326 if (p->fd == -1 && errno == EACCES) 327 p->fd = pfopen(p->opt.source, O_RDONLY); 328 if (p->fd < 0) { 329 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "pf open: %s: %s\n\ 330your system may not be properly configured; see the packetfilter(4) man page\n", 331 p->opt.source, pcap_strerror(errno)); 332 goto bad; 333 } 334 pf->OrigMissed = -1; 335 enmode = ENTSTAMP|ENNONEXCL; 336 if (!p->opt.immediate) 337 enmode |= ENBATCH; 338 if (p->opt.promisc) 339 enmode |= ENPROMISC; 340 if (ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode) < 0) { 341 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCMBIS: %s", 342 pcap_strerror(errno)); 343 goto bad; 344 } 345#ifdef ENCOPYALL 346 /* Try to set COPYALL mode so that we see packets to ourself */ 347 enmode = ENCOPYALL; 348 (void)ioctl(p->fd, EIOCMBIS, (caddr_t)&enmode);/* OK if this fails */ 349#endif 350 /* set the backlog */ 351 if (ioctl(p->fd, EIOCSETW, (caddr_t)&backlog) < 0) { 352 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETW: %s", 353 pcap_strerror(errno)); 354 goto bad; 355 } 356 /* discover interface type */ 357 if (ioctl(p->fd, EIOCDEVP, (caddr_t)&devparams) < 0) { 358 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCDEVP: %s", 359 pcap_strerror(errno)); 360 goto bad; 361 } 362 /* HACK: to compile prior to Ultrix 4.2 */ 363#ifndef ENDT_FDDI 364#define ENDT_FDDI 4 365#endif 366 switch (devparams.end_dev_type) { 367 368 case ENDT_10MB: 369 p->linktype = DLT_EN10MB; 370 p->offset = 2; 371 /* 372 * This is (presumably) a real Ethernet capture; give it a 373 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so 374 * that an application can let you choose it, in case you're 375 * capturing DOCSIS traffic that a Cisco Cable Modem 376 * Termination System is putting out onto an Ethernet (it 377 * doesn't put an Ethernet header onto the wire, it puts raw 378 * DOCSIS frames out on the wire inside the low-level 379 * Ethernet framing). 380 */ 381 p->dlt_list = (u_int *) malloc(sizeof(u_int) * 2); 382 /* 383 * If that fails, just leave the list empty. 384 */ 385 if (p->dlt_list != NULL) { 386 p->dlt_list[0] = DLT_EN10MB; 387 p->dlt_list[1] = DLT_DOCSIS; 388 p->dlt_count = 2; 389 } 390 break; 391 392 case ENDT_FDDI: 393 p->linktype = DLT_FDDI; 394 break; 395 396#ifdef ENDT_SLIP 397 case ENDT_SLIP: 398 p->linktype = DLT_SLIP; 399 break; 400#endif 401 402#ifdef ENDT_PPP 403 case ENDT_PPP: 404 p->linktype = DLT_PPP; 405 break; 406#endif 407 408#ifdef ENDT_LOOPBACK 409 case ENDT_LOOPBACK: 410 /* 411 * It appears to use Ethernet framing, at least on 412 * Digital UNIX 4.0. 413 */ 414 p->linktype = DLT_EN10MB; 415 p->offset = 2; 416 break; 417#endif 418 419#ifdef ENDT_TRN 420 case ENDT_TRN: 421 p->linktype = DLT_IEEE802; 422 break; 423#endif 424 425 default: 426 /* 427 * XXX - what about ENDT_IEEE802? The pfilt.h header 428 * file calls this "IEEE 802 networks (non-Ethernet)", 429 * but that doesn't specify a specific link layer type; 430 * it could be 802.4, or 802.5 (except that 802.5 is 431 * ENDT_TRN), or 802.6, or 802.11, or.... That's why 432 * DLT_IEEE802 was hijacked to mean Token Ring in various 433 * BSDs, and why we went along with that hijacking. 434 * 435 * XXX - what about ENDT_HDLC and ENDT_NULL? 436 * Presumably, as ENDT_OTHER is just "Miscellaneous 437 * framing", there's not much we can do, as that 438 * doesn't specify a particular type of header. 439 */ 440 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, 441 "unknown data-link type %u", devparams.end_dev_type); 442 goto bad; 443 } 444 /* set truncation */ 445 if (p->linktype == DLT_FDDI) { 446 p->fddipad = PCAP_FDDIPAD; 447 448 /* packetfilter includes the padding in the snapshot */ 449 p->snapshot += PCAP_FDDIPAD; 450 } else 451 p->fddipad = 0; 452 if (ioctl(p->fd, EIOCTRUNCATE, (caddr_t)&p->snapshot) < 0) { 453 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCTRUNCATE: %s", 454 pcap_strerror(errno)); 455 goto bad; 456 } 457 /* accept all packets */ 458 memset(&Filter, 0, sizeof(Filter)); 459 Filter.enf_Priority = 37; /* anything > 2 */ 460 Filter.enf_FilterLen = 0; /* means "always true" */ 461 if (ioctl(p->fd, EIOCSETF, (caddr_t)&Filter) < 0) { 462 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSETF: %s", 463 pcap_strerror(errno)); 464 goto bad; 465 } 466 467 if (p->opt.timeout != 0) { 468 struct timeval timeout; 469 timeout.tv_sec = p->opt.timeout / 1000; 470 timeout.tv_usec = (p->opt.timeout * 1000) % 1000000; 471 if (ioctl(p->fd, EIOCSRTIMEOUT, (caddr_t)&timeout) < 0) { 472 snprintf(p->errbuf, PCAP_ERRBUF_SIZE, "EIOCSRTIMEOUT: %s", 473 pcap_strerror(errno)); 474 goto bad; 475 } 476 } 477 478 p->bufsize = BUFSPACE; 479 p->buffer = (u_char*)malloc(p->bufsize + p->offset); 480 if (p->buffer == NULL) { 481 strlcpy(p->errbuf, pcap_strerror(errno), PCAP_ERRBUF_SIZE); 482 goto bad; 483 } 484 485 /* 486 * "select()" and "poll()" work on packetfilter devices. 487 */ 488 p->selectable_fd = p->fd; 489 490 p->read_op = pcap_read_pf; 491 p->inject_op = pcap_inject_pf; 492 p->setfilter_op = pcap_setfilter_pf; 493 p->setdirection_op = NULL; /* Not implemented. */ 494 p->set_datalink_op = NULL; /* can't change data link type */ 495 p->getnonblock_op = pcap_getnonblock_fd; 496 p->setnonblock_op = pcap_setnonblock_fd; 497 p->stats_op = pcap_stats_pf; 498 499 return (0); 500 bad: 501 pcap_cleanup_live_common(p); 502 return (PCAP_ERROR); 503} 504 505pcap_t * 506pcap_create_interface(const char *device, char *ebuf) 507{ 508 pcap_t *p; 509 510 p = pcap_create_common(device, ebuf, sizeof (struct pcap_pf)); 511 if (p == NULL) 512 return (NULL); 513 514 p->activate_op = pcap_activate_pf; 515 return (p); 516} 517 518int 519pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf) 520{ 521 return (0); 522} 523 524static int 525pcap_setfilter_pf(pcap_t *p, struct bpf_program *fp) 526{ 527 struct pcap_pf *pf = p->priv; 528 struct bpf_version bv; 529 530 /* 531 * See if BIOCVERSION works. If not, we assume the kernel doesn't 532 * support BPF-style filters (it's not documented in the bpf(7) 533 * or packetfiler(7) man pages, but the code used to fail if 534 * BIOCSETF worked but BIOCVERSION didn't, and I've seen it do 535 * kernel filtering in DU 4.0, so presumably BIOCVERSION works 536 * there, at least). 537 */ 538 if (ioctl(p->fd, BIOCVERSION, (caddr_t)&bv) >= 0) { 539 /* 540 * OK, we have the version of the BPF interpreter; 541 * is it the same major version as us, and the same 542 * or better minor version? 543 */ 544 if (bv.bv_major == BPF_MAJOR_VERSION && 545 bv.bv_minor >= BPF_MINOR_VERSION) { 546 /* 547 * Yes. Try to install the filter. 548 */ 549 if (ioctl(p->fd, BIOCSETF, (caddr_t)fp) < 0) { 550 snprintf(p->errbuf, sizeof(p->errbuf), 551 "BIOCSETF: %s", pcap_strerror(errno)); 552 return (-1); 553 } 554 555 /* 556 * OK, that succeeded. We're doing filtering in 557 * the kernel. (We assume we don't have a 558 * userland filter installed - that'd require 559 * a previous version check to have failed but 560 * this one to succeed.) 561 * 562 * XXX - this message should be supplied to the 563 * application as a warning of some sort, 564 * except that if it's a GUI application, it's 565 * not clear that it should be displayed in 566 * a window to annoy the user. 567 */ 568 fprintf(stderr, "tcpdump: Using kernel BPF filter\n"); 569 pf->filtering_in_kernel = 1; 570 571 /* 572 * Discard any previously-received packets, 573 * as they might have passed whatever filter 574 * was formerly in effect, but might not pass 575 * this filter (BIOCSETF discards packets buffered 576 * in the kernel, so you can lose packets in any 577 * case). 578 */ 579 p->cc = 0; 580 return (0); 581 } 582 583 /* 584 * We can't use the kernel's BPF interpreter; don't give 585 * up, just log a message and be inefficient. 586 * 587 * XXX - this should really be supplied to the application 588 * as a warning of some sort. 589 */ 590 fprintf(stderr, 591 "tcpdump: Requires BPF language %d.%d or higher; kernel is %d.%d\n", 592 BPF_MAJOR_VERSION, BPF_MINOR_VERSION, 593 bv.bv_major, bv.bv_minor); 594 } 595 596 /* 597 * We couldn't do filtering in the kernel; do it in userland. 598 */ 599 if (install_bpf_program(p, fp) < 0) 600 return (-1); 601 602 /* 603 * XXX - this message should be supplied by the application as 604 * a warning of some sort. 605 */ 606 fprintf(stderr, "tcpdump: Filtering in user process\n"); 607 pf->filtering_in_kernel = 0; 608 return (0); 609} 610