1 /* $NetBSD: pcap-linux.c,v 1.3 2015/03/31 21:39:42 christos Exp $ */ 2 3 /* 4 * pcap-linux.c: Packet capture interface to the Linux kernel 5 * 6 * Copyright (c) 2000 Torsten Landschoff <torsten@debian.org> 7 * Sebastian Krahmer <krahmer@cs.uni-potsdam.de> 8 * 9 * License: BSD 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 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 19 * the documentation and/or other materials provided with the 20 * distribution. 21 * 3. The names of the authors may not be used to endorse or promote 22 * products derived from this software without specific prior 23 * written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR 26 * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED 27 * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. 28 * 29 * Modifications: Added PACKET_MMAP support 30 * Paolo Abeni <paolo.abeni@email.it> 31 * Added TPACKET_V3 support 32 * Gabor Tatarka <gabor.tatarka@ericsson.com> 33 * 34 * based on previous works of: 35 * Simon Patarin <patarin@cs.unibo.it> 36 * Phil Wood <cpw@lanl.gov> 37 * 38 * Monitor-mode support for mac80211 includes code taken from the iw 39 * command; the copyright notice for that code is 40 * 41 * Copyright (c) 2007, 2008 Johannes Berg 42 * Copyright (c) 2007 Andy Lutomirski 43 * Copyright (c) 2007 Mike Kershaw 44 * Copyright (c) 2008 Gábor Stefanik 45 * 46 * All rights reserved. 47 * 48 * Redistribution and use in source and binary forms, with or without 49 * modification, are permitted provided that the following conditions 50 * are met: 51 * 1. Redistributions of source code must retain the above copyright 52 * notice, this list of conditions and the following disclaimer. 53 * 2. Redistributions in binary form must reproduce the above copyright 54 * notice, this list of conditions and the following disclaimer in the 55 * documentation and/or other materials provided with the distribution. 56 * 3. The name of the author may not be used to endorse or promote products 57 * derived from this software without specific prior written permission. 58 * 59 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 60 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 61 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 62 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 63 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 64 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 65 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 66 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 67 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 69 * SUCH DAMAGE. 70 */ 71 72 /* 73 * Known problems with 2.0[.x] kernels: 74 * 75 * - The loopback device gives every packet twice; on 2.2[.x] kernels, 76 * if we use PF_PACKET, we can filter out the transmitted version 77 * of the packet by using data in the "sockaddr_ll" returned by 78 * "recvfrom()", but, on 2.0[.x] kernels, we have to use 79 * PF_INET/SOCK_PACKET, which means "recvfrom()" supplies a 80 * "sockaddr_pkt" which doesn't give us enough information to let 81 * us do that. 82 * 83 * - We have to set the interface's IFF_PROMISC flag ourselves, if 84 * we're to run in promiscuous mode, which means we have to turn 85 * it off ourselves when we're done; the kernel doesn't keep track 86 * of how many sockets are listening promiscuously, which means 87 * it won't get turned off automatically when no sockets are 88 * listening promiscuously. We catch "pcap_close()" and, for 89 * interfaces we put into promiscuous mode, take them out of 90 * promiscuous mode - which isn't necessarily the right thing to 91 * do, if another socket also requested promiscuous mode between 92 * the time when we opened the socket and the time when we close 93 * the socket. 94 * 95 * - MSG_TRUNC isn't supported, so you can't specify that "recvfrom()" 96 * return the amount of data that you could have read, rather than 97 * the amount that was returned, so we can't just allocate a buffer 98 * whose size is the snapshot length and pass the snapshot length 99 * as the byte count, and also pass MSG_TRUNC, so that the return 100 * value tells us how long the packet was on the wire. 101 * 102 * This means that, if we want to get the actual size of the packet, 103 * so we can return it in the "len" field of the packet header, 104 * we have to read the entire packet, not just the part that fits 105 * within the snapshot length, and thus waste CPU time copying data 106 * from the kernel that our caller won't see. 107 * 108 * We have to get the actual size, and supply it in "len", because 109 * otherwise, the IP dissector in tcpdump, for example, will complain 110 * about "truncated-ip", as the packet will appear to have been 111 * shorter, on the wire, than the IP header said it should have been. 112 */ 113 114 115 #define _GNU_SOURCE 116 117 #include <sys/cdefs.h> 118 __RCSID("$NetBSD: pcap-linux.c,v 1.3 2015/03/31 21:39:42 christos Exp $"); 119 120 #ifdef HAVE_CONFIG_H 121 #include "config.h" 122 #endif 123 124 #include <errno.h> 125 #include <stdio.h> 126 #include <stdlib.h> 127 #include <ctype.h> 128 #include <unistd.h> 129 #include <fcntl.h> 130 #include <string.h> 131 #include <limits.h> 132 #include <sys/stat.h> 133 #include <sys/socket.h> 134 #include <sys/ioctl.h> 135 #include <sys/utsname.h> 136 #include <sys/mman.h> 137 #include <linux/if.h> 138 #include <linux/if_packet.h> 139 #include <linux/sockios.h> 140 #include <netinet/in.h> 141 #include <linux/if_ether.h> 142 #include <net/if_arp.h> 143 #include <poll.h> 144 #include <dirent.h> 145 146 #include "pcap-int.h" 147 #include "pcap/sll.h" 148 #include "pcap/vlan.h" 149 150 /* 151 * If PF_PACKET is defined, we can use {SOCK_RAW,SOCK_DGRAM}/PF_PACKET 152 * sockets rather than SOCK_PACKET sockets. 153 * 154 * To use them, we include <linux/if_packet.h> rather than 155 * <netpacket/packet.h>; we do so because 156 * 157 * some Linux distributions (e.g., Slackware 4.0) have 2.2 or 158 * later kernels and libc5, and don't provide a <netpacket/packet.h> 159 * file; 160 * 161 * not all versions of glibc2 have a <netpacket/packet.h> file 162 * that defines stuff needed for some of the 2.4-or-later-kernel 163 * features, so if the system has a 2.4 or later kernel, we 164 * still can't use those features. 165 * 166 * We're already including a number of other <linux/XXX.h> headers, and 167 * this code is Linux-specific (no other OS has PF_PACKET sockets as 168 * a raw packet capture mechanism), so it's not as if you gain any 169 * useful portability by using <netpacket/packet.h> 170 * 171 * XXX - should we just include <linux/if_packet.h> even if PF_PACKET 172 * isn't defined? It only defines one data structure in 2.0.x, so 173 * it shouldn't cause any problems. 174 */ 175 #ifdef PF_PACKET 176 # include <linux/if_packet.h> 177 178 /* 179 * On at least some Linux distributions (for example, Red Hat 5.2), 180 * there's no <netpacket/packet.h> file, but PF_PACKET is defined if 181 * you include <sys/socket.h>, but <linux/if_packet.h> doesn't define 182 * any of the PF_PACKET stuff such as "struct sockaddr_ll" or any of 183 * the PACKET_xxx stuff. 184 * 185 * So we check whether PACKET_HOST is defined, and assume that we have 186 * PF_PACKET sockets only if it is defined. 187 */ 188 # ifdef PACKET_HOST 189 # define HAVE_PF_PACKET_SOCKETS 190 # ifdef PACKET_AUXDATA 191 # define HAVE_PACKET_AUXDATA 192 # endif /* PACKET_AUXDATA */ 193 # endif /* PACKET_HOST */ 194 195 196 /* check for memory mapped access avaibility. We assume every needed 197 * struct is defined if the macro TPACKET_HDRLEN is defined, because it 198 * uses many ring related structs and macros */ 199 # ifdef PCAP_SUPPORT_PACKET_RING 200 # ifdef TPACKET_HDRLEN 201 # define HAVE_PACKET_RING 202 # ifdef TPACKET3_HDRLEN 203 # define HAVE_TPACKET3 204 # endif /* TPACKET3_HDRLEN */ 205 # ifdef TPACKET2_HDRLEN 206 # define HAVE_TPACKET2 207 # else /* TPACKET2_HDRLEN */ 208 # define TPACKET_V1 0 /* Old kernel with only V1, so no TPACKET_Vn defined */ 209 # endif /* TPACKET2_HDRLEN */ 210 # endif /* TPACKET_HDRLEN */ 211 # endif /* PCAP_SUPPORT_PACKET_RING */ 212 #endif /* PF_PACKET */ 213 214 #ifdef SO_ATTACH_FILTER 215 #include <linux/types.h> 216 #include <linux/filter.h> 217 #endif 218 219 #ifdef HAVE_LINUX_NET_TSTAMP_H 220 #include <linux/net_tstamp.h> 221 #endif 222 223 /* 224 * Got Wireless Extensions? 225 */ 226 #ifdef HAVE_LINUX_WIRELESS_H 227 #include <linux/wireless.h> 228 #endif /* HAVE_LINUX_WIRELESS_H */ 229 230 /* 231 * Got libnl? 232 */ 233 #ifdef HAVE_LIBNL 234 #include <linux/nl80211.h> 235 236 #include <netlink/genl/genl.h> 237 #include <netlink/genl/family.h> 238 #include <netlink/genl/ctrl.h> 239 #include <netlink/msg.h> 240 #include <netlink/attr.h> 241 #endif /* HAVE_LIBNL */ 242 243 /* 244 * Got ethtool support? 245 */ 246 #ifdef HAVE_LINUX_ETHTOOL_H 247 #include <linux/ethtool.h> 248 #endif 249 250 #ifndef HAVE_SOCKLEN_T 251 typedef int socklen_t; 252 #endif 253 254 #ifndef MSG_TRUNC 255 /* 256 * This is being compiled on a system that lacks MSG_TRUNC; define it 257 * with the value it has in the 2.2 and later kernels, so that, on 258 * those kernels, when we pass it in the flags argument to "recvfrom()" 259 * we're passing the right value and thus get the MSG_TRUNC behavior 260 * we want. (We don't get that behavior on 2.0[.x] kernels, because 261 * they didn't support MSG_TRUNC.) 262 */ 263 #define MSG_TRUNC 0x20 264 #endif 265 266 #ifndef SOL_PACKET 267 /* 268 * This is being compiled on a system that lacks SOL_PACKET; define it 269 * with the value it has in the 2.2 and later kernels, so that we can 270 * set promiscuous mode in the good modern way rather than the old 271 * 2.0-kernel crappy way. 272 */ 273 #define SOL_PACKET 263 274 #endif 275 276 #define MAX_LINKHEADER_SIZE 256 277 278 /* 279 * When capturing on all interfaces we use this as the buffer size. 280 * Should be bigger then all MTUs that occur in real life. 281 * 64kB should be enough for now. 282 */ 283 #define BIGGER_THAN_ALL_MTUS (64*1024) 284 285 /* 286 * Private data for capturing on Linux SOCK_PACKET or PF_PACKET sockets. 287 */ 288 struct pcap_linux { 289 u_int packets_read; /* count of packets read with recvfrom() */ 290 long proc_dropped; /* packets reported dropped by /proc/net/dev */ 291 struct pcap_stat stat; 292 293 char *device; /* device name */ 294 int filter_in_userland; /* must filter in userland */ 295 int blocks_to_filter_in_userland; 296 int must_do_on_close; /* stuff we must do when we close */ 297 int timeout; /* timeout for buffering */ 298 int sock_packet; /* using Linux 2.0 compatible interface */ 299 int cooked; /* using SOCK_DGRAM rather than SOCK_RAW */ 300 int ifindex; /* interface index of device we're bound to */ 301 int lo_ifindex; /* interface index of the loopback device */ 302 bpf_u_int32 oldmode; /* mode to restore when turning monitor mode off */ 303 char *mondevice; /* mac80211 monitor device we created */ 304 u_char *mmapbuf; /* memory-mapped region pointer */ 305 size_t mmapbuflen; /* size of region */ 306 int vlan_offset; /* offset at which to insert vlan tags; if -1, don't insert */ 307 u_int tp_version; /* version of tpacket_hdr for mmaped ring */ 308 u_int tp_hdrlen; /* hdrlen of tpacket_hdr for mmaped ring */ 309 u_char *oneshot_buffer; /* buffer for copy of packet */ 310 #ifdef HAVE_TPACKET3 311 unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */ 312 int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */ 313 #endif 314 }; 315 316 /* 317 * Stuff to do when we close. 318 */ 319 #define MUST_CLEAR_PROMISC 0x00000001 /* clear promiscuous mode */ 320 #define MUST_CLEAR_RFMON 0x00000002 /* clear rfmon (monitor) mode */ 321 #define MUST_DELETE_MONIF 0x00000004 /* delete monitor-mode interface */ 322 323 /* 324 * Prototypes for internal functions and methods. 325 */ 326 static void map_arphrd_to_dlt(pcap_t *, int, int, const char *, int); 327 #ifdef HAVE_PF_PACKET_SOCKETS 328 static short int map_packet_type_to_sll_type(short int); 329 #endif 330 static int pcap_activate_linux(pcap_t *); 331 static int activate_old(pcap_t *); 332 static int activate_new(pcap_t *); 333 static int activate_mmap(pcap_t *, int *); 334 static int pcap_can_set_rfmon_linux(pcap_t *); 335 static int pcap_read_linux(pcap_t *, int, pcap_handler, u_char *); 336 static int pcap_read_packet(pcap_t *, pcap_handler, u_char *); 337 static int pcap_inject_linux(pcap_t *, const void *, size_t); 338 static int pcap_stats_linux(pcap_t *, struct pcap_stat *); 339 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *); 340 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t); 341 static int pcap_set_datalink_linux(pcap_t *, int); 342 static void pcap_cleanup_linux(pcap_t *); 343 344 /* 345 * This is what the header structure looks like in a 64-bit kernel; 346 * we use this, rather than struct tpacket_hdr, if we're using 347 * TPACKET_V1 in 32-bit code running on a 64-bit kernel. 348 */ 349 struct tpacket_hdr_64 { 350 uint64_t tp_status; 351 unsigned int tp_len; 352 unsigned int tp_snaplen; 353 unsigned short tp_mac; 354 unsigned short tp_net; 355 unsigned int tp_sec; 356 unsigned int tp_usec; 357 }; 358 359 /* 360 * We use this internally as the tpacket version for TPACKET_V1 in 361 * 32-bit code on a 64-bit kernel. 362 */ 363 #define TPACKET_V1_64 99 364 365 union thdr { 366 struct tpacket_hdr *h1; 367 struct tpacket_hdr_64 *h1_64; 368 #ifdef HAVE_TPACKET2 369 struct tpacket2_hdr *h2; 370 #endif 371 #ifdef HAVE_TPACKET3 372 struct tpacket_block_desc *h3; 373 #endif 374 void *raw; 375 }; 376 377 #ifdef HAVE_PACKET_RING 378 #define RING_GET_FRAME(h) (((union thdr **)h->buffer)[h->offset]) 379 380 static void destroy_ring(pcap_t *handle); 381 static int create_ring(pcap_t *handle, int *status); 382 static int prepare_tpacket_socket(pcap_t *handle); 383 static void pcap_cleanup_linux_mmap(pcap_t *); 384 static int pcap_read_linux_mmap_v1(pcap_t *, int, pcap_handler , u_char *); 385 static int pcap_read_linux_mmap_v1_64(pcap_t *, int, pcap_handler , u_char *); 386 #ifdef HAVE_TPACKET2 387 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *); 388 #endif 389 #ifdef HAVE_TPACKET3 390 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *); 391 #endif 392 static int pcap_setfilter_linux_mmap(pcap_t *, struct bpf_program *); 393 static int pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf); 394 static int pcap_getnonblock_mmap(pcap_t *p, char *errbuf); 395 static void pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h, 396 const u_char *bytes); 397 #endif 398 399 #ifdef TP_STATUS_VLAN_TPID_VALID 400 # define VLAN_TPID(hdr, hv) (((hv)->tp_vlan_tpid || ((hdr)->tp_status & TP_STATUS_VLAN_TPID_VALID)) ? (hv)->tp_vlan_tpid : ETH_P_8021Q) 401 #else 402 # define VLAN_TPID(hdr, hv) ETH_P_8021Q 403 #endif 404 405 /* 406 * Wrap some ioctl calls 407 */ 408 #ifdef HAVE_PF_PACKET_SOCKETS 409 static int iface_get_id(int fd, const char *device, char *ebuf); 410 #endif /* HAVE_PF_PACKET_SOCKETS */ 411 static int iface_get_mtu(int fd, const char *device, char *ebuf); 412 static int iface_get_arptype(int fd, const char *device, char *ebuf); 413 #ifdef HAVE_PF_PACKET_SOCKETS 414 static int iface_bind(int fd, int ifindex, char *ebuf); 415 #ifdef IW_MODE_MONITOR 416 static int has_wext(int sock_fd, const char *device, char *ebuf); 417 #endif /* IW_MODE_MONITOR */ 418 static int enter_rfmon_mode(pcap_t *handle, int sock_fd, 419 const char *device); 420 #endif /* HAVE_PF_PACKET_SOCKETS */ 421 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) 422 static int iface_ethtool_get_ts_info(pcap_t *handle, char *ebuf); 423 #endif 424 #ifdef HAVE_PACKET_RING 425 static int iface_get_offload(pcap_t *handle); 426 #endif 427 static int iface_bind_old(int fd, const char *device, char *ebuf); 428 429 #ifdef SO_ATTACH_FILTER 430 static int fix_program(pcap_t *handle, struct sock_fprog *fcode, 431 int is_mapped); 432 static int fix_offset(struct bpf_insn *p); 433 static int set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode); 434 static int reset_kernel_filter(pcap_t *handle); 435 436 static struct sock_filter total_insn 437 = BPF_STMT(BPF_RET | BPF_K, 0); 438 static struct sock_fprog total_fcode 439 = { 1, &total_insn }; 440 #endif /* SO_ATTACH_FILTER */ 441 442 pcap_t * 443 pcap_create_interface(const char *device, char *ebuf) 444 { 445 pcap_t *handle; 446 447 handle = pcap_create_common(device, ebuf, sizeof (struct pcap_linux)); 448 if (handle == NULL) 449 return NULL; 450 451 handle->activate_op = pcap_activate_linux; 452 handle->can_set_rfmon_op = pcap_can_set_rfmon_linux; 453 454 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) 455 /* 456 * See what time stamp types we support. 457 */ 458 if (iface_ethtool_get_ts_info(handle, ebuf) == -1) { 459 free(handle); 460 return NULL; 461 } 462 #endif 463 464 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) 465 /* 466 * We claim that we support microsecond and nanosecond time 467 * stamps. 468 * 469 * XXX - with adapter-supplied time stamps, can we choose 470 * microsecond or nanosecond time stamps on arbitrary 471 * adapters? 472 */ 473 handle->tstamp_precision_count = 2; 474 handle->tstamp_precision_list = malloc(2 * sizeof(u_int)); 475 if (handle->tstamp_precision_list == NULL) { 476 snprintf(ebuf, PCAP_ERRBUF_SIZE, "malloc: %s", 477 pcap_strerror(errno)); 478 if (handle->tstamp_type_list != NULL) 479 free(handle->tstamp_type_list); 480 free(handle); 481 return NULL; 482 } 483 handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO; 484 handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO; 485 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */ 486 487 return handle; 488 } 489 490 #ifdef HAVE_LIBNL 491 /* 492 * If interface {if} is a mac80211 driver, the file 493 * /sys/class/net/{if}/phy80211 is a symlink to 494 * /sys/class/ieee80211/{phydev}, for some {phydev}. 495 * 496 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at 497 * least, has a "wmaster0" device and a "wlan0" device; the 498 * latter is the one with the IP address. Both show up in 499 * "tcpdump -D" output. Capturing on the wmaster0 device 500 * captures with 802.11 headers. 501 * 502 * airmon-ng searches through /sys/class/net for devices named 503 * monN, starting with mon0; as soon as one *doesn't* exist, 504 * it chooses that as the monitor device name. If the "iw" 505 * command exists, it does "iw dev {if} interface add {monif} 506 * type monitor", where {monif} is the monitor device. It 507 * then (sigh) sleeps .1 second, and then configures the 508 * device up. Otherwise, if /sys/class/ieee80211/{phydev}/add_iface 509 * is a file, it writes {mondev}, without a newline, to that file, 510 * and again (sigh) sleeps .1 second, and then iwconfig's that 511 * device into monitor mode and configures it up. Otherwise, 512 * you can't do monitor mode. 513 * 514 * All these devices are "glued" together by having the 515 * /sys/class/net/{device}/phy80211 links pointing to the same 516 * place, so, given a wmaster, wlan, or mon device, you can 517 * find the other devices by looking for devices with 518 * the same phy80211 link. 519 * 520 * To turn monitor mode off, delete the monitor interface, 521 * either with "iw dev {monif} interface del" or by sending 522 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface 523 * 524 * Note: if you try to create a monitor device named "monN", and 525 * there's already a "monN" device, it fails, as least with 526 * the netlink interface (which is what iw uses), with a return 527 * value of -ENFILE. (Return values are negative errnos.) We 528 * could probably use that to find an unused device. 529 * 530 * Yes, you can have multiple monitor devices for a given 531 * physical device. 532 */ 533 534 /* 535 * Is this a mac80211 device? If so, fill in the physical device path and 536 * return 1; if not, return 0. On an error, fill in handle->errbuf and 537 * return PCAP_ERROR. 538 */ 539 static int 540 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path, 541 size_t phydev_max_pathlen) 542 { 543 char *pathstr; 544 ssize_t bytes_read; 545 546 /* 547 * Generate the path string for the symlink to the physical device. 548 */ 549 if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) { 550 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 551 "%s: Can't generate path name string for /sys/class/net device", 552 device); 553 return PCAP_ERROR; 554 } 555 bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen); 556 if (bytes_read == -1) { 557 if (errno == ENOENT || errno == EINVAL) { 558 /* 559 * Doesn't exist, or not a symlink; assume that 560 * means it's not a mac80211 device. 561 */ 562 free(pathstr); 563 return 0; 564 } 565 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 566 "%s: Can't readlink %s: %s", device, pathstr, 567 strerror(errno)); 568 free(pathstr); 569 return PCAP_ERROR; 570 } 571 free(pathstr); 572 phydev_path[bytes_read] = '\0'; 573 return 1; 574 } 575 576 #ifdef HAVE_LIBNL_SOCKETS 577 #define get_nl_errmsg nl_geterror 578 #else 579 /* libnl 2.x compatibility code */ 580 581 #define nl_sock nl_handle 582 583 static inline struct nl_handle * 584 nl_socket_alloc(void) 585 { 586 return nl_handle_alloc(); 587 } 588 589 static inline void 590 nl_socket_free(struct nl_handle *h) 591 { 592 nl_handle_destroy(h); 593 } 594 595 #define get_nl_errmsg strerror 596 597 static inline int 598 __genl_ctrl_alloc_cache(struct nl_handle *h, struct nl_cache **cache) 599 { 600 struct nl_cache *tmp = genl_ctrl_alloc_cache(h); 601 if (!tmp) 602 return -ENOMEM; 603 *cache = tmp; 604 return 0; 605 } 606 #define genl_ctrl_alloc_cache __genl_ctrl_alloc_cache 607 #endif /* !HAVE_LIBNL_SOCKETS */ 608 609 struct nl80211_state { 610 struct nl_sock *nl_sock; 611 struct nl_cache *nl_cache; 612 struct genl_family *nl80211; 613 }; 614 615 static int 616 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device) 617 { 618 int err; 619 620 state->nl_sock = nl_socket_alloc(); 621 if (!state->nl_sock) { 622 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 623 "%s: failed to allocate netlink handle", device); 624 return PCAP_ERROR; 625 } 626 627 if (genl_connect(state->nl_sock)) { 628 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 629 "%s: failed to connect to generic netlink", device); 630 goto out_handle_destroy; 631 } 632 633 err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache); 634 if (err < 0) { 635 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 636 "%s: failed to allocate generic netlink cache: %s", 637 device, get_nl_errmsg(-err)); 638 goto out_handle_destroy; 639 } 640 641 state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211"); 642 if (!state->nl80211) { 643 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 644 "%s: nl80211 not found", device); 645 goto out_cache_free; 646 } 647 648 return 0; 649 650 out_cache_free: 651 nl_cache_free(state->nl_cache); 652 out_handle_destroy: 653 nl_socket_free(state->nl_sock); 654 return PCAP_ERROR; 655 } 656 657 static void 658 nl80211_cleanup(struct nl80211_state *state) 659 { 660 genl_family_put(state->nl80211); 661 nl_cache_free(state->nl_cache); 662 nl_socket_free(state->nl_sock); 663 } 664 665 static int 666 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state, 667 const char *device, const char *mondevice) 668 { 669 int ifindex; 670 struct nl_msg *msg; 671 int err; 672 673 ifindex = iface_get_id(sock_fd, device, handle->errbuf); 674 if (ifindex == -1) 675 return PCAP_ERROR; 676 677 msg = nlmsg_alloc(); 678 if (!msg) { 679 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 680 "%s: failed to allocate netlink msg", device); 681 return PCAP_ERROR; 682 } 683 684 genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0, 685 0, NL80211_CMD_NEW_INTERFACE, 0); 686 NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex); 687 NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice); 688 NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR); 689 690 err = nl_send_auto_complete(state->nl_sock, msg); 691 if (err < 0) { 692 #if defined HAVE_LIBNL_NLE 693 if (err == -NLE_FAILURE) { 694 #else 695 if (err == -ENFILE) { 696 #endif 697 /* 698 * Device not available; our caller should just 699 * keep trying. (libnl 2.x maps ENFILE to 700 * NLE_FAILURE; it can also map other errors 701 * to that, but there's not much we can do 702 * about that.) 703 */ 704 nlmsg_free(msg); 705 return 0; 706 } else { 707 /* 708 * Real failure, not just "that device is not 709 * available. 710 */ 711 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 712 "%s: nl_send_auto_complete failed adding %s interface: %s", 713 device, mondevice, get_nl_errmsg(-err)); 714 nlmsg_free(msg); 715 return PCAP_ERROR; 716 } 717 } 718 err = nl_wait_for_ack(state->nl_sock); 719 if (err < 0) { 720 #if defined HAVE_LIBNL_NLE 721 if (err == -NLE_FAILURE) { 722 #else 723 if (err == -ENFILE) { 724 #endif 725 /* 726 * Device not available; our caller should just 727 * keep trying. (libnl 2.x maps ENFILE to 728 * NLE_FAILURE; it can also map other errors 729 * to that, but there's not much we can do 730 * about that.) 731 */ 732 nlmsg_free(msg); 733 return 0; 734 } else { 735 /* 736 * Real failure, not just "that device is not 737 * available. 738 */ 739 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 740 "%s: nl_wait_for_ack failed adding %s interface: %s", 741 device, mondevice, get_nl_errmsg(-err)); 742 nlmsg_free(msg); 743 return PCAP_ERROR; 744 } 745 } 746 747 /* 748 * Success. 749 */ 750 nlmsg_free(msg); 751 return 1; 752 753 nla_put_failure: 754 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 755 "%s: nl_put failed adding %s interface", 756 device, mondevice); 757 nlmsg_free(msg); 758 return PCAP_ERROR; 759 } 760 761 static int 762 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state, 763 const char *device, const char *mondevice) 764 { 765 int ifindex; 766 struct nl_msg *msg; 767 int err; 768 769 ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf); 770 if (ifindex == -1) 771 return PCAP_ERROR; 772 773 msg = nlmsg_alloc(); 774 if (!msg) { 775 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 776 "%s: failed to allocate netlink msg", device); 777 return PCAP_ERROR; 778 } 779 780 genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0, 781 0, NL80211_CMD_DEL_INTERFACE, 0); 782 NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex); 783 784 err = nl_send_auto_complete(state->nl_sock, msg); 785 if (err < 0) { 786 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 787 "%s: nl_send_auto_complete failed deleting %s interface: %s", 788 device, mondevice, get_nl_errmsg(-err)); 789 nlmsg_free(msg); 790 return PCAP_ERROR; 791 } 792 err = nl_wait_for_ack(state->nl_sock); 793 if (err < 0) { 794 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 795 "%s: nl_wait_for_ack failed adding %s interface: %s", 796 device, mondevice, get_nl_errmsg(-err)); 797 nlmsg_free(msg); 798 return PCAP_ERROR; 799 } 800 801 /* 802 * Success. 803 */ 804 nlmsg_free(msg); 805 return 1; 806 807 nla_put_failure: 808 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 809 "%s: nl_put failed deleting %s interface", 810 device, mondevice); 811 nlmsg_free(msg); 812 return PCAP_ERROR; 813 } 814 815 static int 816 enter_rfmon_mode_mac80211(pcap_t *handle, int sock_fd, const char *device) 817 { 818 struct pcap_linux *handlep = handle->priv; 819 int ret; 820 char phydev_path[PATH_MAX+1]; 821 struct nl80211_state nlstate; 822 struct ifreq ifr; 823 u_int n; 824 825 /* 826 * Is this a mac80211 device? 827 */ 828 ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX); 829 if (ret < 0) 830 return ret; /* error */ 831 if (ret == 0) 832 return 0; /* no error, but not mac80211 device */ 833 834 /* 835 * XXX - is this already a monN device? 836 * If so, we're done. 837 * Is that determined by old Wireless Extensions ioctls? 838 */ 839 840 /* 841 * OK, it's apparently a mac80211 device. 842 * Try to find an unused monN device for it. 843 */ 844 ret = nl80211_init(handle, &nlstate, device); 845 if (ret != 0) 846 return ret; 847 for (n = 0; n < UINT_MAX; n++) { 848 /* 849 * Try mon{n}. 850 */ 851 char mondevice[3+10+1]; /* mon{UINT_MAX}\0 */ 852 853 snprintf(mondevice, sizeof mondevice, "mon%u", n); 854 ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice); 855 if (ret == 1) { 856 handlep->mondevice = strdup(mondevice); 857 goto added; 858 } 859 if (ret < 0) { 860 /* 861 * Hard failure. Just return ret; handle->errbuf 862 * has already been set. 863 */ 864 nl80211_cleanup(&nlstate); 865 return ret; 866 } 867 } 868 869 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 870 "%s: No free monN interfaces", device); 871 nl80211_cleanup(&nlstate); 872 return PCAP_ERROR; 873 874 added: 875 876 #if 0 877 /* 878 * Sleep for .1 seconds. 879 */ 880 delay.tv_sec = 0; 881 delay.tv_nsec = 500000000; 882 nanosleep(&delay, NULL); 883 #endif 884 885 /* 886 * If we haven't already done so, arrange to have 887 * "pcap_close_all()" called when we exit. 888 */ 889 if (!pcap_do_addexit(handle)) { 890 /* 891 * "atexit()" failed; don't put the interface 892 * in rfmon mode, just give up. 893 */ 894 return PCAP_ERROR_RFMON_NOTSUP; 895 } 896 897 /* 898 * Now configure the monitor interface up. 899 */ 900 memset(&ifr, 0, sizeof(ifr)); 901 strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name)); 902 if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) { 903 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 904 "%s: Can't get flags for %s: %s", device, 905 handlep->mondevice, strerror(errno)); 906 del_mon_if(handle, sock_fd, &nlstate, device, 907 handlep->mondevice); 908 nl80211_cleanup(&nlstate); 909 return PCAP_ERROR; 910 } 911 ifr.ifr_flags |= IFF_UP|IFF_RUNNING; 912 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) { 913 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 914 "%s: Can't set flags for %s: %s", device, 915 handlep->mondevice, strerror(errno)); 916 del_mon_if(handle, sock_fd, &nlstate, device, 917 handlep->mondevice); 918 nl80211_cleanup(&nlstate); 919 return PCAP_ERROR; 920 } 921 922 /* 923 * Success. Clean up the libnl state. 924 */ 925 nl80211_cleanup(&nlstate); 926 927 /* 928 * Note that we have to delete the monitor device when we close 929 * the handle. 930 */ 931 handlep->must_do_on_close |= MUST_DELETE_MONIF; 932 933 /* 934 * Add this to the list of pcaps to close when we exit. 935 */ 936 pcap_add_to_pcaps_to_close(handle); 937 938 return 1; 939 } 940 #endif /* HAVE_LIBNL */ 941 942 static int 943 pcap_can_set_rfmon_linux(pcap_t *handle) 944 { 945 #ifdef HAVE_LIBNL 946 char phydev_path[PATH_MAX+1]; 947 int ret; 948 #endif 949 #ifdef IW_MODE_MONITOR 950 int sock_fd; 951 struct iwreq ireq; 952 #endif 953 954 if (strcmp(handle->opt.source, "any") == 0) { 955 /* 956 * Monitor mode makes no sense on the "any" device. 957 */ 958 return 0; 959 } 960 961 #ifdef HAVE_LIBNL 962 /* 963 * Bleah. There doesn't seem to be a way to ask a mac80211 964 * device, through libnl, whether it supports monitor mode; 965 * we'll just check whether the device appears to be a 966 * mac80211 device and, if so, assume the device supports 967 * monitor mode. 968 * 969 * wmaster devices don't appear to support the Wireless 970 * Extensions, but we can create a mon device for a 971 * wmaster device, so we don't bother checking whether 972 * a mac80211 device supports the Wireless Extensions. 973 */ 974 ret = get_mac80211_phydev(handle, handle->opt.source, phydev_path, 975 PATH_MAX); 976 if (ret < 0) 977 return ret; /* error */ 978 if (ret == 1) 979 return 1; /* mac80211 device */ 980 #endif 981 982 #ifdef IW_MODE_MONITOR 983 /* 984 * Bleah. There doesn't appear to be an ioctl to use to ask 985 * whether a device supports monitor mode; we'll just do 986 * SIOCGIWMODE and, if it succeeds, assume the device supports 987 * monitor mode. 988 * 989 * Open a socket on which to attempt to get the mode. 990 * (We assume that if we have Wireless Extensions support 991 * we also have PF_PACKET support.) 992 */ 993 sock_fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL)); 994 if (sock_fd == -1) { 995 (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 996 "socket: %s", pcap_strerror(errno)); 997 return PCAP_ERROR; 998 } 999 1000 /* 1001 * Attempt to get the current mode. 1002 */ 1003 strlcpy(ireq.ifr_ifrn.ifrn_name, handle->opt.source, 1004 sizeof ireq.ifr_ifrn.ifrn_name); 1005 if (ioctl(sock_fd, SIOCGIWMODE, &ireq) != -1) { 1006 /* 1007 * Well, we got the mode; assume we can set it. 1008 */ 1009 close(sock_fd); 1010 return 1; 1011 } 1012 if (errno == ENODEV) { 1013 /* The device doesn't even exist. */ 1014 (void)snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1015 "SIOCGIWMODE failed: %s", pcap_strerror(errno)); 1016 close(sock_fd); 1017 return PCAP_ERROR_NO_SUCH_DEVICE; 1018 } 1019 close(sock_fd); 1020 #endif 1021 return 0; 1022 } 1023 1024 /* 1025 * Grabs the number of dropped packets by the interface from /proc/net/dev. 1026 * 1027 * XXX - what about /sys/class/net/{interface name}/rx_*? There are 1028 * individual devices giving, in ASCII, various rx_ and tx_ statistics. 1029 * 1030 * Or can we get them in binary form from netlink? 1031 */ 1032 static long int 1033 linux_if_drops(const char * if_name) 1034 { 1035 char buffer[512]; 1036 char * bufptr; 1037 FILE * file; 1038 int field_to_convert = 3, if_name_sz = strlen(if_name); 1039 long int dropped_pkts = 0; 1040 1041 file = fopen("/proc/net/dev", "r"); 1042 if (!file) 1043 return 0; 1044 1045 while (!dropped_pkts && fgets( buffer, sizeof(buffer), file )) 1046 { 1047 /* search for 'bytes' -- if its in there, then 1048 that means we need to grab the fourth field. otherwise 1049 grab the third field. */ 1050 if (field_to_convert != 4 && strstr(buffer, "bytes")) 1051 { 1052 field_to_convert = 4; 1053 continue; 1054 } 1055 1056 /* find iface and make sure it actually matches -- space before the name and : after it */ 1057 if ((bufptr = strstr(buffer, if_name)) && 1058 (bufptr == buffer || *(bufptr-1) == ' ') && 1059 *(bufptr + if_name_sz) == ':') 1060 { 1061 bufptr = bufptr + if_name_sz + 1; 1062 1063 /* grab the nth field from it */ 1064 while( --field_to_convert && *bufptr != '\0') 1065 { 1066 while (*bufptr != '\0' && *(bufptr++) == ' '); 1067 while (*bufptr != '\0' && *(bufptr++) != ' '); 1068 } 1069 1070 /* get rid of any final spaces */ 1071 while (*bufptr != '\0' && *bufptr == ' ') bufptr++; 1072 1073 if (*bufptr != '\0') 1074 dropped_pkts = strtol(bufptr, NULL, 10); 1075 1076 break; 1077 } 1078 } 1079 1080 fclose(file); 1081 return dropped_pkts; 1082 } 1083 1084 1085 /* 1086 * With older kernels promiscuous mode is kind of interesting because we 1087 * have to reset the interface before exiting. The problem can't really 1088 * be solved without some daemon taking care of managing usage counts. 1089 * If we put the interface into promiscuous mode, we set a flag indicating 1090 * that we must take it out of that mode when the interface is closed, 1091 * and, when closing the interface, if that flag is set we take it out 1092 * of promiscuous mode. 1093 * 1094 * Even with newer kernels, we have the same issue with rfmon mode. 1095 */ 1096 1097 static void pcap_cleanup_linux( pcap_t *handle ) 1098 { 1099 struct pcap_linux *handlep = handle->priv; 1100 struct ifreq ifr; 1101 #ifdef HAVE_LIBNL 1102 struct nl80211_state nlstate; 1103 int ret; 1104 #endif /* HAVE_LIBNL */ 1105 #ifdef IW_MODE_MONITOR 1106 int oldflags; 1107 struct iwreq ireq; 1108 #endif /* IW_MODE_MONITOR */ 1109 1110 if (handlep->must_do_on_close != 0) { 1111 /* 1112 * There's something we have to do when closing this 1113 * pcap_t. 1114 */ 1115 if (handlep->must_do_on_close & MUST_CLEAR_PROMISC) { 1116 /* 1117 * We put the interface into promiscuous mode; 1118 * take it out of promiscuous mode. 1119 * 1120 * XXX - if somebody else wants it in promiscuous 1121 * mode, this code cannot know that, so it'll take 1122 * it out of promiscuous mode. That's not fixable 1123 * in 2.0[.x] kernels. 1124 */ 1125 memset(&ifr, 0, sizeof(ifr)); 1126 strlcpy(ifr.ifr_name, handlep->device, 1127 sizeof(ifr.ifr_name)); 1128 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) { 1129 fprintf(stderr, 1130 "Can't restore interface %s flags (SIOCGIFFLAGS failed: %s).\n" 1131 "Please adjust manually.\n" 1132 "Hint: This can't happen with Linux >= 2.2.0.\n", 1133 handlep->device, strerror(errno)); 1134 } else { 1135 if (ifr.ifr_flags & IFF_PROMISC) { 1136 /* 1137 * Promiscuous mode is currently on; 1138 * turn it off. 1139 */ 1140 ifr.ifr_flags &= ~IFF_PROMISC; 1141 if (ioctl(handle->fd, SIOCSIFFLAGS, 1142 &ifr) == -1) { 1143 fprintf(stderr, 1144 "Can't restore interface %s flags (SIOCSIFFLAGS failed: %s).\n" 1145 "Please adjust manually.\n" 1146 "Hint: This can't happen with Linux >= 2.2.0.\n", 1147 handlep->device, 1148 strerror(errno)); 1149 } 1150 } 1151 } 1152 } 1153 1154 #ifdef HAVE_LIBNL 1155 if (handlep->must_do_on_close & MUST_DELETE_MONIF) { 1156 ret = nl80211_init(handle, &nlstate, handlep->device); 1157 if (ret >= 0) { 1158 ret = del_mon_if(handle, handle->fd, &nlstate, 1159 handlep->device, handlep->mondevice); 1160 nl80211_cleanup(&nlstate); 1161 } 1162 if (ret < 0) { 1163 fprintf(stderr, 1164 "Can't delete monitor interface %s (%s).\n" 1165 "Please delete manually.\n", 1166 handlep->mondevice, handle->errbuf); 1167 } 1168 } 1169 #endif /* HAVE_LIBNL */ 1170 1171 #ifdef IW_MODE_MONITOR 1172 if (handlep->must_do_on_close & MUST_CLEAR_RFMON) { 1173 /* 1174 * We put the interface into rfmon mode; 1175 * take it out of rfmon mode. 1176 * 1177 * XXX - if somebody else wants it in rfmon 1178 * mode, this code cannot know that, so it'll take 1179 * it out of rfmon mode. 1180 */ 1181 1182 /* 1183 * First, take the interface down if it's up; 1184 * otherwise, we might get EBUSY. 1185 * If we get errors, just drive on and print 1186 * a warning if we can't restore the mode. 1187 */ 1188 oldflags = 0; 1189 memset(&ifr, 0, sizeof(ifr)); 1190 strlcpy(ifr.ifr_name, handlep->device, 1191 sizeof(ifr.ifr_name)); 1192 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) != -1) { 1193 if (ifr.ifr_flags & IFF_UP) { 1194 oldflags = ifr.ifr_flags; 1195 ifr.ifr_flags &= ~IFF_UP; 1196 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) 1197 oldflags = 0; /* didn't set, don't restore */ 1198 } 1199 } 1200 1201 /* 1202 * Now restore the mode. 1203 */ 1204 strlcpy(ireq.ifr_ifrn.ifrn_name, handlep->device, 1205 sizeof ireq.ifr_ifrn.ifrn_name); 1206 ireq.u.mode = handlep->oldmode; 1207 if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) { 1208 /* 1209 * Scientist, you've failed. 1210 */ 1211 fprintf(stderr, 1212 "Can't restore interface %s wireless mode (SIOCSIWMODE failed: %s).\n" 1213 "Please adjust manually.\n", 1214 handlep->device, strerror(errno)); 1215 } 1216 1217 /* 1218 * Now bring the interface back up if we brought 1219 * it down. 1220 */ 1221 if (oldflags != 0) { 1222 ifr.ifr_flags = oldflags; 1223 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) { 1224 fprintf(stderr, 1225 "Can't bring interface %s back up (SIOCSIFFLAGS failed: %s).\n" 1226 "Please adjust manually.\n", 1227 handlep->device, strerror(errno)); 1228 } 1229 } 1230 } 1231 #endif /* IW_MODE_MONITOR */ 1232 1233 /* 1234 * Take this pcap out of the list of pcaps for which we 1235 * have to take the interface out of some mode. 1236 */ 1237 pcap_remove_from_pcaps_to_close(handle); 1238 } 1239 1240 if (handlep->mondevice != NULL) { 1241 free(handlep->mondevice); 1242 handlep->mondevice = NULL; 1243 } 1244 if (handlep->device != NULL) { 1245 free(handlep->device); 1246 handlep->device = NULL; 1247 } 1248 pcap_cleanup_live_common(handle); 1249 } 1250 1251 /* 1252 * Get a handle for a live capture from the given device. You can 1253 * pass NULL as device to get all packages (without link level 1254 * information of course). If you pass 1 as promisc the interface 1255 * will be set to promiscous mode (XXX: I think this usage should 1256 * be deprecated and functions be added to select that later allow 1257 * modification of that values -- Torsten). 1258 */ 1259 static int 1260 pcap_activate_linux(pcap_t *handle) 1261 { 1262 struct pcap_linux *handlep = handle->priv; 1263 const char *device; 1264 struct ifreq ifr; 1265 int status = 0; 1266 int ret; 1267 1268 device = handle->opt.source; 1269 1270 /* 1271 * Make sure the name we were handed will fit into the ioctls we 1272 * might perform on the device; if not, return a "No such device" 1273 * indication, as the Linux kernel shouldn't support creating 1274 * a device whose name won't fit into those ioctls. 1275 * 1276 * "Will fit" means "will fit, complete with a null terminator", 1277 * so if the length, which does *not* include the null terminator, 1278 * is greater than *or equal to* the size of the field into which 1279 * we'll be copying it, that won't fit. 1280 */ 1281 if (strlen(device) >= sizeof(ifr.ifr_name)) { 1282 status = PCAP_ERROR_NO_SUCH_DEVICE; 1283 goto fail; 1284 } 1285 1286 handle->inject_op = pcap_inject_linux; 1287 handle->setfilter_op = pcap_setfilter_linux; 1288 handle->setdirection_op = pcap_setdirection_linux; 1289 handle->set_datalink_op = pcap_set_datalink_linux; 1290 handle->getnonblock_op = pcap_getnonblock_fd; 1291 handle->setnonblock_op = pcap_setnonblock_fd; 1292 handle->cleanup_op = pcap_cleanup_linux; 1293 handle->read_op = pcap_read_linux; 1294 handle->stats_op = pcap_stats_linux; 1295 1296 /* 1297 * The "any" device is a special device which causes us not 1298 * to bind to a particular device and thus to look at all 1299 * devices. 1300 */ 1301 if (strcmp(device, "any") == 0) { 1302 if (handle->opt.promisc) { 1303 handle->opt.promisc = 0; 1304 /* Just a warning. */ 1305 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1306 "Promiscuous mode not supported on the \"any\" device"); 1307 status = PCAP_WARNING_PROMISC_NOTSUP; 1308 } 1309 } 1310 1311 handlep->device = strdup(device); 1312 if (handlep->device == NULL) { 1313 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "strdup: %s", 1314 pcap_strerror(errno) ); 1315 return PCAP_ERROR; 1316 } 1317 1318 /* copy timeout value */ 1319 handlep->timeout = handle->opt.timeout; 1320 1321 /* 1322 * If we're in promiscuous mode, then we probably want 1323 * to see when the interface drops packets too, so get an 1324 * initial count from /proc/net/dev 1325 */ 1326 if (handle->opt.promisc) 1327 handlep->proc_dropped = linux_if_drops(handlep->device); 1328 1329 /* 1330 * Current Linux kernels use the protocol family PF_PACKET to 1331 * allow direct access to all packets on the network while 1332 * older kernels had a special socket type SOCK_PACKET to 1333 * implement this feature. 1334 * While this old implementation is kind of obsolete we need 1335 * to be compatible with older kernels for a while so we are 1336 * trying both methods with the newer method preferred. 1337 */ 1338 ret = activate_new(handle); 1339 if (ret < 0) { 1340 /* 1341 * Fatal error with the new way; just fail. 1342 * ret has the error return; if it's PCAP_ERROR, 1343 * handle->errbuf has been set appropriately. 1344 */ 1345 status = ret; 1346 goto fail; 1347 } 1348 if (ret == 1) { 1349 /* 1350 * Success. 1351 * Try to use memory-mapped access. 1352 */ 1353 switch (activate_mmap(handle, &status)) { 1354 1355 case 1: 1356 /* 1357 * We succeeded. status has been 1358 * set to the status to return, 1359 * which might be 0, or might be 1360 * a PCAP_WARNING_ value. 1361 */ 1362 return status; 1363 1364 case 0: 1365 /* 1366 * Kernel doesn't support it - just continue 1367 * with non-memory-mapped access. 1368 */ 1369 break; 1370 1371 case -1: 1372 /* 1373 * We failed to set up to use it, or the kernel 1374 * supports it, but we failed to enable it. 1375 * ret has been set to the error status to 1376 * return and, if it's PCAP_ERROR, handle->errbuf 1377 * contains the error message. 1378 */ 1379 status = ret; 1380 goto fail; 1381 } 1382 } 1383 else if (ret == 0) { 1384 /* Non-fatal error; try old way */ 1385 if ((ret = activate_old(handle)) != 1) { 1386 /* 1387 * Both methods to open the packet socket failed. 1388 * Tidy up and report our failure (handle->errbuf 1389 * is expected to be set by the functions above). 1390 */ 1391 status = ret; 1392 goto fail; 1393 } 1394 } 1395 1396 /* 1397 * We set up the socket, but not with memory-mapped access. 1398 */ 1399 if (handle->opt.buffer_size != 0) { 1400 /* 1401 * Set the socket buffer size to the specified value. 1402 */ 1403 if (setsockopt(handle->fd, SOL_SOCKET, SO_RCVBUF, 1404 &handle->opt.buffer_size, 1405 sizeof(handle->opt.buffer_size)) == -1) { 1406 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1407 "SO_RCVBUF: %s", pcap_strerror(errno)); 1408 status = PCAP_ERROR; 1409 goto fail; 1410 } 1411 } 1412 1413 /* Allocate the buffer */ 1414 1415 handle->buffer = malloc(handle->bufsize + handle->offset); 1416 if (!handle->buffer) { 1417 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1418 "malloc: %s", pcap_strerror(errno)); 1419 status = PCAP_ERROR; 1420 goto fail; 1421 } 1422 1423 /* 1424 * "handle->fd" is a socket, so "select()" and "poll()" 1425 * should work on it. 1426 */ 1427 handle->selectable_fd = handle->fd; 1428 1429 return status; 1430 1431 fail: 1432 pcap_cleanup_linux(handle); 1433 return status; 1434 } 1435 1436 /* 1437 * Read at most max_packets from the capture stream and call the callback 1438 * for each of them. Returns the number of packets handled or -1 if an 1439 * error occured. 1440 */ 1441 static int 1442 pcap_read_linux(pcap_t *handle, int max_packets, pcap_handler callback, u_char *user) 1443 { 1444 /* 1445 * Currently, on Linux only one packet is delivered per read, 1446 * so we don't loop. 1447 */ 1448 return pcap_read_packet(handle, callback, user); 1449 } 1450 1451 static int 1452 pcap_set_datalink_linux(pcap_t *handle, int dlt) 1453 { 1454 handle->linktype = dlt; 1455 return 0; 1456 } 1457 1458 /* 1459 * linux_check_direction() 1460 * 1461 * Do checks based on packet direction. 1462 */ 1463 static inline int 1464 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll) 1465 { 1466 struct pcap_linux *handlep = handle->priv; 1467 1468 if (sll->sll_pkttype == PACKET_OUTGOING) { 1469 /* 1470 * Outgoing packet. 1471 * If this is from the loopback device, reject it; 1472 * we'll see the packet as an incoming packet as well, 1473 * and we don't want to see it twice. 1474 */ 1475 if (sll->sll_ifindex == handlep->lo_ifindex) 1476 return 0; 1477 1478 /* 1479 * If the user only wants incoming packets, reject it. 1480 */ 1481 if (handle->direction == PCAP_D_IN) 1482 return 0; 1483 } else { 1484 /* 1485 * Incoming packet. 1486 * If the user only wants outgoing packets, reject it. 1487 */ 1488 if (handle->direction == PCAP_D_OUT) 1489 return 0; 1490 } 1491 return 1; 1492 } 1493 1494 /* 1495 * Read a packet from the socket calling the handler provided by 1496 * the user. Returns the number of packets received or -1 if an 1497 * error occured. 1498 */ 1499 static int 1500 pcap_read_packet(pcap_t *handle, pcap_handler callback, u_char *userdata) 1501 { 1502 struct pcap_linux *handlep = handle->priv; 1503 u_char *bp; 1504 int offset; 1505 #ifdef HAVE_PF_PACKET_SOCKETS 1506 struct sockaddr_ll from; 1507 struct sll_header *hdrp; 1508 #else 1509 struct sockaddr from; 1510 #endif 1511 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) 1512 struct iovec iov; 1513 struct msghdr msg; 1514 struct cmsghdr *cmsg; 1515 union { 1516 struct cmsghdr cmsg; 1517 char buf[CMSG_SPACE(sizeof(struct tpacket_auxdata))]; 1518 } cmsg_buf; 1519 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1520 socklen_t fromlen; 1521 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1522 int packet_len, caplen; 1523 struct pcap_pkthdr pcap_header; 1524 1525 struct bpf_aux_data aux_data; 1526 #ifdef HAVE_PF_PACKET_SOCKETS 1527 /* 1528 * If this is a cooked device, leave extra room for a 1529 * fake packet header. 1530 */ 1531 if (handlep->cooked) 1532 offset = SLL_HDR_LEN; 1533 else 1534 offset = 0; 1535 #else 1536 /* 1537 * This system doesn't have PF_PACKET sockets, so it doesn't 1538 * support cooked devices. 1539 */ 1540 offset = 0; 1541 #endif 1542 1543 /* 1544 * Receive a single packet from the kernel. 1545 * We ignore EINTR, as that might just be due to a signal 1546 * being delivered - if the signal should interrupt the 1547 * loop, the signal handler should call pcap_breakloop() 1548 * to set handle->break_loop (we ignore it on other 1549 * platforms as well). 1550 * We also ignore ENETDOWN, so that we can continue to 1551 * capture traffic if the interface goes down and comes 1552 * back up again; comments in the kernel indicate that 1553 * we'll just block waiting for packets if we try to 1554 * receive from a socket that delivered ENETDOWN, and, 1555 * if we're using a memory-mapped buffer, we won't even 1556 * get notified of "network down" events. 1557 */ 1558 bp = handle->buffer + handle->offset; 1559 1560 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) 1561 msg.msg_name = &from; 1562 msg.msg_namelen = sizeof(from); 1563 msg.msg_iov = &iov; 1564 msg.msg_iovlen = 1; 1565 msg.msg_control = &cmsg_buf; 1566 msg.msg_controllen = sizeof(cmsg_buf); 1567 msg.msg_flags = 0; 1568 1569 iov.iov_len = handle->bufsize - offset; 1570 iov.iov_base = bp + offset; 1571 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1572 1573 do { 1574 /* 1575 * Has "pcap_breakloop()" been called? 1576 */ 1577 if (handle->break_loop) { 1578 /* 1579 * Yes - clear the flag that indicates that it has, 1580 * and return PCAP_ERROR_BREAK as an indication that 1581 * we were told to break out of the loop. 1582 */ 1583 handle->break_loop = 0; 1584 return PCAP_ERROR_BREAK; 1585 } 1586 1587 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) 1588 packet_len = recvmsg(handle->fd, &msg, MSG_TRUNC); 1589 #else /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1590 fromlen = sizeof(from); 1591 packet_len = recvfrom( 1592 handle->fd, bp + offset, 1593 handle->bufsize - offset, MSG_TRUNC, 1594 (struct sockaddr *) &from, &fromlen); 1595 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1596 } while (packet_len == -1 && errno == EINTR); 1597 1598 /* Check if an error occured */ 1599 1600 if (packet_len == -1) { 1601 switch (errno) { 1602 1603 case EAGAIN: 1604 return 0; /* no packet there */ 1605 1606 case ENETDOWN: 1607 /* 1608 * The device on which we're capturing went away. 1609 * 1610 * XXX - we should really return 1611 * PCAP_ERROR_IFACE_NOT_UP, but pcap_dispatch() 1612 * etc. aren't defined to return that. 1613 */ 1614 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1615 "The interface went down"); 1616 return PCAP_ERROR; 1617 1618 default: 1619 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1620 "recvfrom: %s", pcap_strerror(errno)); 1621 return PCAP_ERROR; 1622 } 1623 } 1624 1625 #ifdef HAVE_PF_PACKET_SOCKETS 1626 if (!handlep->sock_packet) { 1627 /* 1628 * Unfortunately, there is a window between socket() and 1629 * bind() where the kernel may queue packets from any 1630 * interface. If we're bound to a particular interface, 1631 * discard packets not from that interface. 1632 * 1633 * (If socket filters are supported, we could do the 1634 * same thing we do when changing the filter; however, 1635 * that won't handle packet sockets without socket 1636 * filter support, and it's a bit more complicated. 1637 * It would save some instructions per packet, however.) 1638 */ 1639 if (handlep->ifindex != -1 && 1640 from.sll_ifindex != handlep->ifindex) 1641 return 0; 1642 1643 /* 1644 * Do checks based on packet direction. 1645 * We can only do this if we're using PF_PACKET; the 1646 * address returned for SOCK_PACKET is a "sockaddr_pkt" 1647 * which lacks the relevant packet type information. 1648 */ 1649 if (!linux_check_direction(handle, &from)) 1650 return 0; 1651 } 1652 #endif 1653 1654 #ifdef HAVE_PF_PACKET_SOCKETS 1655 /* 1656 * If this is a cooked device, fill in the fake packet header. 1657 */ 1658 if (handlep->cooked) { 1659 /* 1660 * Add the length of the fake header to the length 1661 * of packet data we read. 1662 */ 1663 packet_len += SLL_HDR_LEN; 1664 1665 hdrp = (struct sll_header *)bp; 1666 hdrp->sll_pkttype = map_packet_type_to_sll_type(from.sll_pkttype); 1667 hdrp->sll_hatype = htons(from.sll_hatype); 1668 hdrp->sll_halen = htons(from.sll_halen); 1669 memcpy(hdrp->sll_addr, from.sll_addr, 1670 (from.sll_halen > SLL_ADDRLEN) ? 1671 SLL_ADDRLEN : 1672 from.sll_halen); 1673 hdrp->sll_protocol = from.sll_protocol; 1674 } 1675 1676 #if defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) 1677 if (handlep->vlan_offset != -1) { 1678 for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) { 1679 struct tpacket_auxdata *aux; 1680 unsigned int len; 1681 struct vlan_tag *tag; 1682 1683 if (cmsg->cmsg_len < CMSG_LEN(sizeof(struct tpacket_auxdata)) || 1684 cmsg->cmsg_level != SOL_PACKET || 1685 cmsg->cmsg_type != PACKET_AUXDATA) 1686 continue; 1687 1688 aux = (struct tpacket_auxdata *)CMSG_DATA(cmsg); 1689 #if defined(TP_STATUS_VLAN_VALID) 1690 if ((aux->tp_vlan_tci == 0) && !(aux->tp_status & TP_STATUS_VLAN_VALID)) 1691 #else 1692 if (aux->tp_vlan_tci == 0) /* this is ambigious but without the 1693 TP_STATUS_VLAN_VALID flag, there is 1694 nothing that we can do */ 1695 #endif 1696 continue; 1697 1698 len = packet_len > iov.iov_len ? iov.iov_len : packet_len; 1699 if (len < (unsigned int) handlep->vlan_offset) 1700 break; 1701 1702 bp -= VLAN_TAG_LEN; 1703 memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset); 1704 1705 tag = (struct vlan_tag *)(bp + handlep->vlan_offset); 1706 tag->vlan_tpid = htons(VLAN_TPID(aux, aux)); 1707 tag->vlan_tci = htons(aux->tp_vlan_tci); 1708 1709 /* store vlan tci to bpf_aux_data struct for userland bpf filter */ 1710 #if defined(TP_STATUS_VLAN_VALID) 1711 aux_data.vlan_tag = htons(aux->tp_vlan_tci) & 0x0fff; 1712 aux_data.vlan_tag_present = (aux->tp_status & TP_STATUS_VLAN_VALID); 1713 #endif 1714 packet_len += VLAN_TAG_LEN; 1715 } 1716 } 1717 #endif /* defined(HAVE_PACKET_AUXDATA) && defined(HAVE_LINUX_TPACKET_AUXDATA_TP_VLAN_TCI) */ 1718 #endif /* HAVE_PF_PACKET_SOCKETS */ 1719 1720 /* 1721 * XXX: According to the kernel source we should get the real 1722 * packet len if calling recvfrom with MSG_TRUNC set. It does 1723 * not seem to work here :(, but it is supported by this code 1724 * anyway. 1725 * To be honest the code RELIES on that feature so this is really 1726 * broken with 2.2.x kernels. 1727 * I spend a day to figure out what's going on and I found out 1728 * that the following is happening: 1729 * 1730 * The packet comes from a random interface and the packet_rcv 1731 * hook is called with a clone of the packet. That code inserts 1732 * the packet into the receive queue of the packet socket. 1733 * If a filter is attached to that socket that filter is run 1734 * first - and there lies the problem. The default filter always 1735 * cuts the packet at the snaplen: 1736 * 1737 * # tcpdump -d 1738 * (000) ret #68 1739 * 1740 * So the packet filter cuts down the packet. The recvfrom call 1741 * says "hey, it's only 68 bytes, it fits into the buffer" with 1742 * the result that we don't get the real packet length. This 1743 * is valid at least until kernel 2.2.17pre6. 1744 * 1745 * We currently handle this by making a copy of the filter 1746 * program, fixing all "ret" instructions with non-zero 1747 * operands to have an operand of MAXIMUM_SNAPLEN so that the 1748 * filter doesn't truncate the packet, and supplying that modified 1749 * filter to the kernel. 1750 */ 1751 1752 caplen = packet_len; 1753 if (caplen > handle->snapshot) 1754 caplen = handle->snapshot; 1755 1756 /* Run the packet filter if not using kernel filter */ 1757 if (handlep->filter_in_userland && handle->fcode.bf_insns) { 1758 if (bpf_filter_with_aux_data(handle->fcode.bf_insns, bp, 1759 packet_len, caplen, &aux_data) == 0) { 1760 /* rejected by filter */ 1761 return 0; 1762 } 1763 } 1764 1765 /* Fill in our own header data */ 1766 1767 /* get timestamp for this packet */ 1768 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) 1769 if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) { 1770 if (ioctl(handle->fd, SIOCGSTAMPNS, &pcap_header.ts) == -1) { 1771 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1772 "SIOCGSTAMPNS: %s", pcap_strerror(errno)); 1773 return PCAP_ERROR; 1774 } 1775 } else 1776 #endif 1777 { 1778 if (ioctl(handle->fd, SIOCGSTAMP, &pcap_header.ts) == -1) { 1779 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 1780 "SIOCGSTAMP: %s", pcap_strerror(errno)); 1781 return PCAP_ERROR; 1782 } 1783 } 1784 1785 pcap_header.caplen = caplen; 1786 pcap_header.len = packet_len; 1787 1788 /* 1789 * Count the packet. 1790 * 1791 * Arguably, we should count them before we check the filter, 1792 * as on many other platforms "ps_recv" counts packets 1793 * handed to the filter rather than packets that passed 1794 * the filter, but if filtering is done in the kernel, we 1795 * can't get a count of packets that passed the filter, 1796 * and that would mean the meaning of "ps_recv" wouldn't 1797 * be the same on all Linux systems. 1798 * 1799 * XXX - it's not the same on all systems in any case; 1800 * ideally, we should have a "get the statistics" call 1801 * that supplies more counts and indicates which of them 1802 * it supplies, so that we supply a count of packets 1803 * handed to the filter only on platforms where that 1804 * information is available. 1805 * 1806 * We count them here even if we can get the packet count 1807 * from the kernel, as we can only determine at run time 1808 * whether we'll be able to get it from the kernel (if 1809 * HAVE_TPACKET_STATS isn't defined, we can't get it from 1810 * the kernel, but if it is defined, the library might 1811 * have been built with a 2.4 or later kernel, but we 1812 * might be running on a 2.2[.x] kernel without Alexey 1813 * Kuznetzov's turbopacket patches, and thus the kernel 1814 * might not be able to supply those statistics). We 1815 * could, I guess, try, when opening the socket, to get 1816 * the statistics, and if we can not increment the count 1817 * here, but it's not clear that always incrementing 1818 * the count is more expensive than always testing a flag 1819 * in memory. 1820 * 1821 * We keep the count in "handlep->packets_read", and use that 1822 * for "ps_recv" if we can't get the statistics from the kernel. 1823 * We do that because, if we *can* get the statistics from 1824 * the kernel, we use "handlep->stat.ps_recv" and 1825 * "handlep->stat.ps_drop" as running counts, as reading the 1826 * statistics from the kernel resets the kernel statistics, 1827 * and if we directly increment "handlep->stat.ps_recv" here, 1828 * that means it will count packets *twice* on systems where 1829 * we can get kernel statistics - once here, and once in 1830 * pcap_stats_linux(). 1831 */ 1832 handlep->packets_read++; 1833 1834 /* Call the user supplied callback function */ 1835 callback(userdata, &pcap_header, bp); 1836 1837 return 1; 1838 } 1839 1840 static int 1841 pcap_inject_linux(pcap_t *handle, const void *buf, size_t size) 1842 { 1843 struct pcap_linux *handlep = handle->priv; 1844 int ret; 1845 1846 #ifdef HAVE_PF_PACKET_SOCKETS 1847 if (!handlep->sock_packet) { 1848 /* PF_PACKET socket */ 1849 if (handlep->ifindex == -1) { 1850 /* 1851 * We don't support sending on the "any" device. 1852 */ 1853 strlcpy(handle->errbuf, 1854 "Sending packets isn't supported on the \"any\" device", 1855 PCAP_ERRBUF_SIZE); 1856 return (-1); 1857 } 1858 1859 if (handlep->cooked) { 1860 /* 1861 * We don't support sending on the "any" device. 1862 * 1863 * XXX - how do you send on a bound cooked-mode 1864 * socket? 1865 * Is a "sendto()" required there? 1866 */ 1867 strlcpy(handle->errbuf, 1868 "Sending packets isn't supported in cooked mode", 1869 PCAP_ERRBUF_SIZE); 1870 return (-1); 1871 } 1872 } 1873 #endif 1874 1875 ret = send(handle->fd, buf, size, 0); 1876 if (ret == -1) { 1877 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "send: %s", 1878 pcap_strerror(errno)); 1879 return (-1); 1880 } 1881 return (ret); 1882 } 1883 1884 /* 1885 * Get the statistics for the given packet capture handle. 1886 * Reports the number of dropped packets iff the kernel supports 1887 * the PACKET_STATISTICS "getsockopt()" argument (2.4 and later 1888 * kernels, and 2.2[.x] kernels with Alexey Kuznetzov's turbopacket 1889 * patches); otherwise, that information isn't available, and we lie 1890 * and report 0 as the count of dropped packets. 1891 */ 1892 static int 1893 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats) 1894 { 1895 struct pcap_linux *handlep = handle->priv; 1896 #ifdef HAVE_TPACKET_STATS 1897 #ifdef HAVE_TPACKET3 1898 /* 1899 * For sockets using TPACKET_V1 or TPACKET_V2, the extra 1900 * stuff at the end of a struct tpacket_stats_v3 will not 1901 * be filled in, and we don't look at it so this is OK even 1902 * for those sockets. In addition, the PF_PACKET socket 1903 * code in the kernel only uses the length parameter to 1904 * compute how much data to copy out and to indicate how 1905 * much data was copied out, so it's OK to base it on the 1906 * size of a struct tpacket_stats. 1907 * 1908 * XXX - it's probably OK, in fact, to just use a 1909 * struct tpacket_stats for V3 sockets, as we don't 1910 * care about the tp_freeze_q_cnt stat. 1911 */ 1912 struct tpacket_stats_v3 kstats; 1913 #else /* HAVE_TPACKET3 */ 1914 struct tpacket_stats kstats; 1915 #endif /* HAVE_TPACKET3 */ 1916 socklen_t len = sizeof (struct tpacket_stats); 1917 #endif /* HAVE_TPACKET_STATS */ 1918 1919 long if_dropped = 0; 1920 1921 /* 1922 * To fill in ps_ifdrop, we parse /proc/net/dev for the number 1923 */ 1924 if (handle->opt.promisc) 1925 { 1926 if_dropped = handlep->proc_dropped; 1927 handlep->proc_dropped = linux_if_drops(handlep->device); 1928 handlep->stat.ps_ifdrop += (handlep->proc_dropped - if_dropped); 1929 } 1930 1931 #ifdef HAVE_TPACKET_STATS 1932 /* 1933 * Try to get the packet counts from the kernel. 1934 */ 1935 if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, 1936 &kstats, &len) > -1) { 1937 /* 1938 * On systems where the PACKET_STATISTICS "getsockopt()" 1939 * argument is supported on PF_PACKET sockets: 1940 * 1941 * "ps_recv" counts only packets that *passed* the 1942 * filter, not packets that didn't pass the filter. 1943 * This includes packets later dropped because we 1944 * ran out of buffer space. 1945 * 1946 * "ps_drop" counts packets dropped because we ran 1947 * out of buffer space. It doesn't count packets 1948 * dropped by the interface driver. It counts only 1949 * packets that passed the filter. 1950 * 1951 * See above for ps_ifdrop. 1952 * 1953 * Both statistics include packets not yet read from 1954 * the kernel by libpcap, and thus not yet seen by 1955 * the application. 1956 * 1957 * In "linux/net/packet/af_packet.c", at least in the 1958 * 2.4.9 kernel, "tp_packets" is incremented for every 1959 * packet that passes the packet filter *and* is 1960 * successfully queued on the socket; "tp_drops" is 1961 * incremented for every packet dropped because there's 1962 * not enough free space in the socket buffer. 1963 * 1964 * When the statistics are returned for a PACKET_STATISTICS 1965 * "getsockopt()" call, "tp_drops" is added to "tp_packets", 1966 * so that "tp_packets" counts all packets handed to 1967 * the PF_PACKET socket, including packets dropped because 1968 * there wasn't room on the socket buffer - but not 1969 * including packets that didn't pass the filter. 1970 * 1971 * In the BSD BPF, the count of received packets is 1972 * incremented for every packet handed to BPF, regardless 1973 * of whether it passed the filter. 1974 * 1975 * We can't make "pcap_stats()" work the same on both 1976 * platforms, but the best approximation is to return 1977 * "tp_packets" as the count of packets and "tp_drops" 1978 * as the count of drops. 1979 * 1980 * Keep a running total because each call to 1981 * getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, .... 1982 * resets the counters to zero. 1983 */ 1984 handlep->stat.ps_recv += kstats.tp_packets; 1985 handlep->stat.ps_drop += kstats.tp_drops; 1986 *stats = handlep->stat; 1987 return 0; 1988 } 1989 else 1990 { 1991 /* 1992 * If the error was EOPNOTSUPP, fall through, so that 1993 * if you build the library on a system with 1994 * "struct tpacket_stats" and run it on a system 1995 * that doesn't, it works as it does if the library 1996 * is built on a system without "struct tpacket_stats". 1997 */ 1998 if (errno != EOPNOTSUPP) { 1999 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 2000 "pcap_stats: %s", pcap_strerror(errno)); 2001 return -1; 2002 } 2003 } 2004 #endif 2005 /* 2006 * On systems where the PACKET_STATISTICS "getsockopt()" argument 2007 * is not supported on PF_PACKET sockets: 2008 * 2009 * "ps_recv" counts only packets that *passed* the filter, 2010 * not packets that didn't pass the filter. It does not 2011 * count packets dropped because we ran out of buffer 2012 * space. 2013 * 2014 * "ps_drop" is not supported. 2015 * 2016 * "ps_ifdrop" is supported. It will return the number 2017 * of drops the interface reports in /proc/net/dev, 2018 * if that is available. 2019 * 2020 * "ps_recv" doesn't include packets not yet read from 2021 * the kernel by libpcap. 2022 * 2023 * We maintain the count of packets processed by libpcap in 2024 * "handlep->packets_read", for reasons described in the comment 2025 * at the end of pcap_read_packet(). We have no idea how many 2026 * packets were dropped by the kernel buffers -- but we know 2027 * how many the interface dropped, so we can return that. 2028 */ 2029 2030 stats->ps_recv = handlep->packets_read; 2031 stats->ps_drop = 0; 2032 stats->ps_ifdrop = handlep->stat.ps_ifdrop; 2033 return 0; 2034 } 2035 2036 static int 2037 add_linux_if(pcap_if_t **devlistp, const char *ifname, int fd, char *errbuf) 2038 { 2039 const char *p; 2040 char name[512]; /* XXX - pick a size */ 2041 char *q, *saveq; 2042 struct ifreq ifrflags; 2043 2044 /* 2045 * Get the interface name. 2046 */ 2047 p = ifname; 2048 q = &name[0]; 2049 while (*p != '\0' && isascii(*p) && !isspace(*p)) { 2050 if (*p == ':') { 2051 /* 2052 * This could be the separator between a 2053 * name and an alias number, or it could be 2054 * the separator between a name with no 2055 * alias number and the next field. 2056 * 2057 * If there's a colon after digits, it 2058 * separates the name and the alias number, 2059 * otherwise it separates the name and the 2060 * next field. 2061 */ 2062 saveq = q; 2063 while (isascii(*p) && isdigit(*p)) 2064 *q++ = *p++; 2065 if (*p != ':') { 2066 /* 2067 * That was the next field, 2068 * not the alias number. 2069 */ 2070 q = saveq; 2071 } 2072 break; 2073 } else 2074 *q++ = *p++; 2075 } 2076 *q = '\0'; 2077 2078 /* 2079 * Get the flags for this interface. 2080 */ 2081 strlcpy(ifrflags.ifr_name, name, sizeof(ifrflags.ifr_name)); 2082 if (ioctl(fd, SIOCGIFFLAGS, (char *)&ifrflags) < 0) { 2083 if (errno == ENXIO || errno == ENODEV) 2084 return (0); /* device doesn't actually exist - ignore it */ 2085 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2086 "SIOCGIFFLAGS: %.*s: %s", 2087 (int)sizeof(ifrflags.ifr_name), 2088 ifrflags.ifr_name, 2089 pcap_strerror(errno)); 2090 return (-1); 2091 } 2092 2093 /* 2094 * Add an entry for this interface, with no addresses. 2095 */ 2096 if (pcap_add_if(devlistp, name, ifrflags.ifr_flags, NULL, 2097 errbuf) == -1) { 2098 /* 2099 * Failure. 2100 */ 2101 return (-1); 2102 } 2103 2104 return (0); 2105 } 2106 2107 /* 2108 * Get from "/sys/class/net" all interfaces listed there; if they're 2109 * already in the list of interfaces we have, that won't add another 2110 * instance, but if they're not, that'll add them. 2111 * 2112 * We don't bother getting any addresses for them; it appears you can't 2113 * use SIOCGIFADDR on Linux to get IPv6 addresses for interfaces, and, 2114 * although some other types of addresses can be fetched with SIOCGIFADDR, 2115 * we don't bother with them for now. 2116 * 2117 * We also don't fail if we couldn't open "/sys/class/net"; we just leave 2118 * the list of interfaces as is, and return 0, so that we can try 2119 * scanning /proc/net/dev. 2120 * 2121 * Otherwise, we return 1 if we don't get an error and -1 if we do. 2122 */ 2123 static int 2124 scan_sys_class_net(pcap_if_t **devlistp, char *errbuf) 2125 { 2126 DIR *sys_class_net_d; 2127 int fd; 2128 struct dirent *ent; 2129 char subsystem_path[PATH_MAX+1]; 2130 struct stat statb; 2131 int ret = 1; 2132 2133 sys_class_net_d = opendir("/sys/class/net"); 2134 if (sys_class_net_d == NULL) { 2135 /* 2136 * Don't fail if it doesn't exist at all. 2137 */ 2138 if (errno == ENOENT) 2139 return (0); 2140 2141 /* 2142 * Fail if we got some other error. 2143 */ 2144 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2145 "Can't open /sys/class/net: %s", pcap_strerror(errno)); 2146 return (-1); 2147 } 2148 2149 /* 2150 * Create a socket from which to fetch interface information. 2151 */ 2152 fd = socket(AF_INET, SOCK_DGRAM, 0); 2153 if (fd < 0) { 2154 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2155 "socket: %s", pcap_strerror(errno)); 2156 (void)closedir(sys_class_net_d); 2157 return (-1); 2158 } 2159 2160 for (;;) { 2161 errno = 0; 2162 ent = readdir(sys_class_net_d); 2163 if (ent == NULL) { 2164 /* 2165 * Error or EOF; if errno != 0, it's an error. 2166 */ 2167 break; 2168 } 2169 2170 /* 2171 * Ignore "." and "..". 2172 */ 2173 if (strcmp(ent->d_name, ".") == 0 || 2174 strcmp(ent->d_name, "..") == 0) 2175 continue; 2176 2177 /* 2178 * Ignore plain files; they do not have subdirectories 2179 * and thus have no attributes. 2180 */ 2181 if (ent->d_type == DT_REG) 2182 continue; 2183 2184 /* 2185 * Is there an "ifindex" file under that name? 2186 * (We don't care whether it's a directory or 2187 * a symlink; older kernels have directories 2188 * for devices, newer kernels have symlinks to 2189 * directories.) 2190 */ 2191 snprintf(subsystem_path, sizeof subsystem_path, 2192 "/sys/class/net/%s/ifindex", ent->d_name); 2193 if (lstat(subsystem_path, &statb) != 0) { 2194 /* 2195 * Stat failed. Either there was an error 2196 * other than ENOENT, and we don't know if 2197 * this is an interface, or it's ENOENT, 2198 * and either some part of "/sys/class/net/{if}" 2199 * disappeared, in which case it probably means 2200 * the interface disappeared, or there's no 2201 * "ifindex" file, which means it's not a 2202 * network interface. 2203 */ 2204 continue; 2205 } 2206 2207 /* 2208 * Attempt to add the interface. 2209 */ 2210 if (add_linux_if(devlistp, &ent->d_name[0], fd, errbuf) == -1) { 2211 /* Fail. */ 2212 ret = -1; 2213 break; 2214 } 2215 } 2216 if (ret != -1) { 2217 /* 2218 * Well, we didn't fail for any other reason; did we 2219 * fail due to an error reading the directory? 2220 */ 2221 if (errno != 0) { 2222 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2223 "Error reading /sys/class/net: %s", 2224 pcap_strerror(errno)); 2225 ret = -1; 2226 } 2227 } 2228 2229 (void)close(fd); 2230 (void)closedir(sys_class_net_d); 2231 return (ret); 2232 } 2233 2234 /* 2235 * Get from "/proc/net/dev" all interfaces listed there; if they're 2236 * already in the list of interfaces we have, that won't add another 2237 * instance, but if they're not, that'll add them. 2238 * 2239 * See comments from scan_sys_class_net(). 2240 */ 2241 static int 2242 scan_proc_net_dev(pcap_if_t **devlistp, char *errbuf) 2243 { 2244 FILE *proc_net_f; 2245 int fd; 2246 char linebuf[512]; 2247 int linenum; 2248 char *p; 2249 int ret = 0; 2250 2251 proc_net_f = fopen("/proc/net/dev", "r"); 2252 if (proc_net_f == NULL) { 2253 /* 2254 * Don't fail if it doesn't exist at all. 2255 */ 2256 if (errno == ENOENT) 2257 return (0); 2258 2259 /* 2260 * Fail if we got some other error. 2261 */ 2262 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2263 "Can't open /proc/net/dev: %s", pcap_strerror(errno)); 2264 return (-1); 2265 } 2266 2267 /* 2268 * Create a socket from which to fetch interface information. 2269 */ 2270 fd = socket(AF_INET, SOCK_DGRAM, 0); 2271 if (fd < 0) { 2272 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2273 "socket: %s", pcap_strerror(errno)); 2274 (void)fclose(proc_net_f); 2275 return (-1); 2276 } 2277 2278 for (linenum = 1; 2279 fgets(linebuf, sizeof linebuf, proc_net_f) != NULL; linenum++) { 2280 /* 2281 * Skip the first two lines - they're headers. 2282 */ 2283 if (linenum <= 2) 2284 continue; 2285 2286 p = &linebuf[0]; 2287 2288 /* 2289 * Skip leading white space. 2290 */ 2291 while (*p != '\0' && isascii(*p) && isspace(*p)) 2292 p++; 2293 if (*p == '\0' || *p == '\n') 2294 continue; /* blank line */ 2295 2296 /* 2297 * Attempt to add the interface. 2298 */ 2299 if (add_linux_if(devlistp, p, fd, errbuf) == -1) { 2300 /* Fail. */ 2301 ret = -1; 2302 break; 2303 } 2304 } 2305 if (ret != -1) { 2306 /* 2307 * Well, we didn't fail for any other reason; did we 2308 * fail due to an error reading the file? 2309 */ 2310 if (ferror(proc_net_f)) { 2311 (void)snprintf(errbuf, PCAP_ERRBUF_SIZE, 2312 "Error reading /proc/net/dev: %s", 2313 pcap_strerror(errno)); 2314 ret = -1; 2315 } 2316 } 2317 2318 (void)close(fd); 2319 (void)fclose(proc_net_f); 2320 return (ret); 2321 } 2322 2323 /* 2324 * Description string for the "any" device. 2325 */ 2326 static const char any_descr[] = "Pseudo-device that captures on all interfaces"; 2327 2328 int 2329 pcap_platform_finddevs(pcap_if_t **alldevsp, char *errbuf) 2330 { 2331 int ret; 2332 2333 /* 2334 * Read "/sys/class/net", and add to the list of interfaces all 2335 * interfaces listed there that we don't already have, because, 2336 * on Linux, SIOCGIFCONF reports only interfaces with IPv4 addresses, 2337 * and even getifaddrs() won't return information about 2338 * interfaces with no addresses, so you need to read "/sys/class/net" 2339 * to get the names of the rest of the interfaces. 2340 */ 2341 ret = scan_sys_class_net(alldevsp, errbuf); 2342 if (ret == -1) 2343 return (-1); /* failed */ 2344 if (ret == 0) { 2345 /* 2346 * No /sys/class/net; try reading /proc/net/dev instead. 2347 */ 2348 if (scan_proc_net_dev(alldevsp, errbuf) == -1) 2349 return (-1); 2350 } 2351 2352 /* 2353 * Add the "any" device. 2354 */ 2355 if (pcap_add_if(alldevsp, "any", IFF_UP|IFF_RUNNING, 2356 any_descr, errbuf) < 0) 2357 return (-1); 2358 2359 return (0); 2360 } 2361 2362 /* 2363 * Attach the given BPF code to the packet capture device. 2364 */ 2365 static int 2366 pcap_setfilter_linux_common(pcap_t *handle, struct bpf_program *filter, 2367 int is_mmapped) 2368 { 2369 struct pcap_linux *handlep; 2370 #ifdef SO_ATTACH_FILTER 2371 struct sock_fprog fcode; 2372 int can_filter_in_kernel; 2373 int err = 0; 2374 #endif 2375 2376 if (!handle) 2377 return -1; 2378 if (!filter) { 2379 strlcpy(handle->errbuf, "setfilter: No filter specified", 2380 PCAP_ERRBUF_SIZE); 2381 return -1; 2382 } 2383 2384 handlep = handle->priv; 2385 2386 /* Make our private copy of the filter */ 2387 2388 if (install_bpf_program(handle, filter) < 0) 2389 /* install_bpf_program() filled in errbuf */ 2390 return -1; 2391 2392 /* 2393 * Run user level packet filter by default. Will be overriden if 2394 * installing a kernel filter succeeds. 2395 */ 2396 handlep->filter_in_userland = 1; 2397 2398 /* Install kernel level filter if possible */ 2399 2400 #ifdef SO_ATTACH_FILTER 2401 #ifdef USHRT_MAX 2402 if (handle->fcode.bf_len > USHRT_MAX) { 2403 /* 2404 * fcode.len is an unsigned short for current kernel. 2405 * I have yet to see BPF-Code with that much 2406 * instructions but still it is possible. So for the 2407 * sake of correctness I added this check. 2408 */ 2409 fprintf(stderr, "Warning: Filter too complex for kernel\n"); 2410 fcode.len = 0; 2411 fcode.filter = NULL; 2412 can_filter_in_kernel = 0; 2413 } else 2414 #endif /* USHRT_MAX */ 2415 { 2416 /* 2417 * Oh joy, the Linux kernel uses struct sock_fprog instead 2418 * of struct bpf_program and of course the length field is 2419 * of different size. Pointed out by Sebastian 2420 * 2421 * Oh, and we also need to fix it up so that all "ret" 2422 * instructions with non-zero operands have MAXIMUM_SNAPLEN 2423 * as the operand if we're not capturing in memory-mapped 2424 * mode, and so that, if we're in cooked mode, all memory- 2425 * reference instructions use special magic offsets in 2426 * references to the link-layer header and assume that the 2427 * link-layer payload begins at 0; "fix_program()" will do 2428 * that. 2429 */ 2430 switch (fix_program(handle, &fcode, is_mmapped)) { 2431 2432 case -1: 2433 default: 2434 /* 2435 * Fatal error; just quit. 2436 * (The "default" case shouldn't happen; we 2437 * return -1 for that reason.) 2438 */ 2439 return -1; 2440 2441 case 0: 2442 /* 2443 * The program performed checks that we can't make 2444 * work in the kernel. 2445 */ 2446 can_filter_in_kernel = 0; 2447 break; 2448 2449 case 1: 2450 /* 2451 * We have a filter that'll work in the kernel. 2452 */ 2453 can_filter_in_kernel = 1; 2454 break; 2455 } 2456 } 2457 2458 /* 2459 * NOTE: at this point, we've set both the "len" and "filter" 2460 * fields of "fcode". As of the 2.6.32.4 kernel, at least, 2461 * those are the only members of the "sock_fprog" structure, 2462 * so we initialize every member of that structure. 2463 * 2464 * If there is anything in "fcode" that is not initialized, 2465 * it is either a field added in a later kernel, or it's 2466 * padding. 2467 * 2468 * If a new field is added, this code needs to be updated 2469 * to set it correctly. 2470 * 2471 * If there are no other fields, then: 2472 * 2473 * if the Linux kernel looks at the padding, it's 2474 * buggy; 2475 * 2476 * if the Linux kernel doesn't look at the padding, 2477 * then if some tool complains that we're passing 2478 * uninitialized data to the kernel, then the tool 2479 * is buggy and needs to understand that it's just 2480 * padding. 2481 */ 2482 if (can_filter_in_kernel) { 2483 if ((err = set_kernel_filter(handle, &fcode)) == 0) 2484 { 2485 /* 2486 * Installation succeded - using kernel filter, 2487 * so userland filtering not needed. 2488 */ 2489 handlep->filter_in_userland = 0; 2490 } 2491 else if (err == -1) /* Non-fatal error */ 2492 { 2493 /* 2494 * Print a warning if we weren't able to install 2495 * the filter for a reason other than "this kernel 2496 * isn't configured to support socket filters. 2497 */ 2498 if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) { 2499 fprintf(stderr, 2500 "Warning: Kernel filter failed: %s\n", 2501 pcap_strerror(errno)); 2502 } 2503 } 2504 } 2505 2506 /* 2507 * If we're not using the kernel filter, get rid of any kernel 2508 * filter that might've been there before, e.g. because the 2509 * previous filter could work in the kernel, or because some other 2510 * code attached a filter to the socket by some means other than 2511 * calling "pcap_setfilter()". Otherwise, the kernel filter may 2512 * filter out packets that would pass the new userland filter. 2513 */ 2514 if (handlep->filter_in_userland) { 2515 if (reset_kernel_filter(handle) == -1) { 2516 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 2517 "can't remove kernel filter: %s", 2518 pcap_strerror(errno)); 2519 err = -2; /* fatal error */ 2520 } 2521 } 2522 2523 /* 2524 * Free up the copy of the filter that was made by "fix_program()". 2525 */ 2526 if (fcode.filter != NULL) 2527 free(fcode.filter); 2528 2529 if (err == -2) 2530 /* Fatal error */ 2531 return -1; 2532 #endif /* SO_ATTACH_FILTER */ 2533 2534 return 0; 2535 } 2536 2537 static int 2538 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter) 2539 { 2540 return pcap_setfilter_linux_common(handle, filter, 0); 2541 } 2542 2543 2544 /* 2545 * Set direction flag: Which packets do we accept on a forwarding 2546 * single device? IN, OUT or both? 2547 */ 2548 static int 2549 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d) 2550 { 2551 #ifdef HAVE_PF_PACKET_SOCKETS 2552 struct pcap_linux *handlep = handle->priv; 2553 2554 if (!handlep->sock_packet) { 2555 handle->direction = d; 2556 return 0; 2557 } 2558 #endif 2559 /* 2560 * We're not using PF_PACKET sockets, so we can't determine 2561 * the direction of the packet. 2562 */ 2563 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 2564 "Setting direction is not supported on SOCK_PACKET sockets"); 2565 return -1; 2566 } 2567 2568 #ifdef HAVE_PF_PACKET_SOCKETS 2569 /* 2570 * Map the PACKET_ value to a LINUX_SLL_ value; we 2571 * want the same numerical value to be used in 2572 * the link-layer header even if the numerical values 2573 * for the PACKET_ #defines change, so that programs 2574 * that look at the packet type field will always be 2575 * able to handle DLT_LINUX_SLL captures. 2576 */ 2577 static short int 2578 map_packet_type_to_sll_type(short int sll_pkttype) 2579 { 2580 switch (sll_pkttype) { 2581 2582 case PACKET_HOST: 2583 return htons(LINUX_SLL_HOST); 2584 2585 case PACKET_BROADCAST: 2586 return htons(LINUX_SLL_BROADCAST); 2587 2588 case PACKET_MULTICAST: 2589 return htons(LINUX_SLL_MULTICAST); 2590 2591 case PACKET_OTHERHOST: 2592 return htons(LINUX_SLL_OTHERHOST); 2593 2594 case PACKET_OUTGOING: 2595 return htons(LINUX_SLL_OUTGOING); 2596 2597 default: 2598 return -1; 2599 } 2600 } 2601 #endif 2602 2603 static int 2604 is_wifi(int sock_fd 2605 #ifndef IW_MODE_MONITOR 2606 _U_ 2607 #endif 2608 , const char *device) 2609 { 2610 char *pathstr; 2611 struct stat statb; 2612 #ifdef IW_MODE_MONITOR 2613 char errbuf[PCAP_ERRBUF_SIZE]; 2614 #endif 2615 2616 /* 2617 * See if there's a sysfs wireless directory for it. 2618 * If so, it's a wireless interface. 2619 */ 2620 if (asprintf(&pathstr, "/sys/class/net/%s/wireless", device) == -1) { 2621 /* 2622 * Just give up here. 2623 */ 2624 return 0; 2625 } 2626 if (stat(pathstr, &statb) == 0) { 2627 free(pathstr); 2628 return 1; 2629 } 2630 free(pathstr); 2631 2632 #ifdef IW_MODE_MONITOR 2633 /* 2634 * OK, maybe it's not wireless, or maybe this kernel doesn't 2635 * support sysfs. Try the wireless extensions. 2636 */ 2637 if (has_wext(sock_fd, device, errbuf) == 1) { 2638 /* 2639 * It supports the wireless extensions, so it's a Wi-Fi 2640 * device. 2641 */ 2642 return 1; 2643 } 2644 #endif 2645 return 0; 2646 } 2647 2648 /* 2649 * Linux uses the ARP hardware type to identify the type of an 2650 * interface. pcap uses the DLT_xxx constants for this. This 2651 * function takes a pointer to a "pcap_t", and an ARPHRD_xxx 2652 * constant, as arguments, and sets "handle->linktype" to the 2653 * appropriate DLT_XXX constant and sets "handle->offset" to 2654 * the appropriate value (to make "handle->offset" plus link-layer 2655 * header length be a multiple of 4, so that the link-layer payload 2656 * will be aligned on a 4-byte boundary when capturing packets). 2657 * (If the offset isn't set here, it'll be 0; add code as appropriate 2658 * for cases where it shouldn't be 0.) 2659 * 2660 * If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture 2661 * in cooked mode; otherwise, we can't use cooked mode, so we have 2662 * to pick some type that works in raw mode, or fail. 2663 * 2664 * Sets the link type to -1 if unable to map the type. 2665 */ 2666 static void map_arphrd_to_dlt(pcap_t *handle, int sock_fd, int arptype, 2667 const char *device, int cooked_ok) 2668 { 2669 static const char cdma_rmnet[] = "cdma_rmnet"; 2670 2671 switch (arptype) { 2672 2673 case ARPHRD_ETHER: 2674 /* 2675 * For various annoying reasons having to do with DHCP 2676 * software, some versions of Android give the mobile- 2677 * phone-network interface an ARPHRD_ value of 2678 * ARPHRD_ETHER, even though the packets supplied by 2679 * that interface have no link-layer header, and begin 2680 * with an IP header, so that the ARPHRD_ value should 2681 * be ARPHRD_NONE. 2682 * 2683 * Detect those devices by checking the device name, and 2684 * use DLT_RAW for them. 2685 */ 2686 if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) { 2687 handle->linktype = DLT_RAW; 2688 return; 2689 } 2690 2691 /* 2692 * Is this a real Ethernet device? If so, give it a 2693 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so 2694 * that an application can let you choose it, in case you're 2695 * capturing DOCSIS traffic that a Cisco Cable Modem 2696 * Termination System is putting out onto an Ethernet (it 2697 * doesn't put an Ethernet header onto the wire, it puts raw 2698 * DOCSIS frames out on the wire inside the low-level 2699 * Ethernet framing). 2700 * 2701 * XXX - are there any other sorts of "fake Ethernet" that 2702 * have ARPHRD_ETHER but that shouldn't offer DLT_DOCSIS as 2703 * a Cisco CMTS won't put traffic onto it or get traffic 2704 * bridged onto it? ISDN is handled in "activate_new()", 2705 * as we fall back on cooked mode there, and we use 2706 * is_wifi() to check for 802.11 devices; are there any 2707 * others? 2708 */ 2709 if (!is_wifi(sock_fd, device)) { 2710 /* 2711 * It's not a Wi-Fi device; offer DOCSIS. 2712 */ 2713 handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2); 2714 /* 2715 * If that fails, just leave the list empty. 2716 */ 2717 if (handle->dlt_list != NULL) { 2718 handle->dlt_list[0] = DLT_EN10MB; 2719 handle->dlt_list[1] = DLT_DOCSIS; 2720 handle->dlt_count = 2; 2721 } 2722 } 2723 /* FALLTHROUGH */ 2724 2725 case ARPHRD_METRICOM: 2726 case ARPHRD_LOOPBACK: 2727 handle->linktype = DLT_EN10MB; 2728 handle->offset = 2; 2729 break; 2730 2731 case ARPHRD_EETHER: 2732 handle->linktype = DLT_EN3MB; 2733 break; 2734 2735 case ARPHRD_AX25: 2736 handle->linktype = DLT_AX25_KISS; 2737 break; 2738 2739 case ARPHRD_PRONET: 2740 handle->linktype = DLT_PRONET; 2741 break; 2742 2743 case ARPHRD_CHAOS: 2744 handle->linktype = DLT_CHAOS; 2745 break; 2746 #ifndef ARPHRD_CAN 2747 #define ARPHRD_CAN 280 2748 #endif 2749 case ARPHRD_CAN: 2750 handle->linktype = DLT_CAN_SOCKETCAN; 2751 break; 2752 2753 #ifndef ARPHRD_IEEE802_TR 2754 #define ARPHRD_IEEE802_TR 800 /* From Linux 2.4 */ 2755 #endif 2756 case ARPHRD_IEEE802_TR: 2757 case ARPHRD_IEEE802: 2758 handle->linktype = DLT_IEEE802; 2759 handle->offset = 2; 2760 break; 2761 2762 case ARPHRD_ARCNET: 2763 handle->linktype = DLT_ARCNET_LINUX; 2764 break; 2765 2766 #ifndef ARPHRD_FDDI /* From Linux 2.2.13 */ 2767 #define ARPHRD_FDDI 774 2768 #endif 2769 case ARPHRD_FDDI: 2770 handle->linktype = DLT_FDDI; 2771 handle->offset = 3; 2772 break; 2773 2774 #ifndef ARPHRD_ATM /* FIXME: How to #include this? */ 2775 #define ARPHRD_ATM 19 2776 #endif 2777 case ARPHRD_ATM: 2778 /* 2779 * The Classical IP implementation in ATM for Linux 2780 * supports both what RFC 1483 calls "LLC Encapsulation", 2781 * in which each packet has an LLC header, possibly 2782 * with a SNAP header as well, prepended to it, and 2783 * what RFC 1483 calls "VC Based Multiplexing", in which 2784 * different virtual circuits carry different network 2785 * layer protocols, and no header is prepended to packets. 2786 * 2787 * They both have an ARPHRD_ type of ARPHRD_ATM, so 2788 * you can't use the ARPHRD_ type to find out whether 2789 * captured packets will have an LLC header, and, 2790 * while there's a socket ioctl to *set* the encapsulation 2791 * type, there's no ioctl to *get* the encapsulation type. 2792 * 2793 * This means that 2794 * 2795 * programs that dissect Linux Classical IP frames 2796 * would have to check for an LLC header and, 2797 * depending on whether they see one or not, dissect 2798 * the frame as LLC-encapsulated or as raw IP (I 2799 * don't know whether there's any traffic other than 2800 * IP that would show up on the socket, or whether 2801 * there's any support for IPv6 in the Linux 2802 * Classical IP code); 2803 * 2804 * filter expressions would have to compile into 2805 * code that checks for an LLC header and does 2806 * the right thing. 2807 * 2808 * Both of those are a nuisance - and, at least on systems 2809 * that support PF_PACKET sockets, we don't have to put 2810 * up with those nuisances; instead, we can just capture 2811 * in cooked mode. That's what we'll do, if we can. 2812 * Otherwise, we'll just fail. 2813 */ 2814 if (cooked_ok) 2815 handle->linktype = DLT_LINUX_SLL; 2816 else 2817 handle->linktype = -1; 2818 break; 2819 2820 #ifndef ARPHRD_IEEE80211 /* From Linux 2.4.6 */ 2821 #define ARPHRD_IEEE80211 801 2822 #endif 2823 case ARPHRD_IEEE80211: 2824 handle->linktype = DLT_IEEE802_11; 2825 break; 2826 2827 #ifndef ARPHRD_IEEE80211_PRISM /* From Linux 2.4.18 */ 2828 #define ARPHRD_IEEE80211_PRISM 802 2829 #endif 2830 case ARPHRD_IEEE80211_PRISM: 2831 handle->linktype = DLT_PRISM_HEADER; 2832 break; 2833 2834 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */ 2835 #define ARPHRD_IEEE80211_RADIOTAP 803 2836 #endif 2837 case ARPHRD_IEEE80211_RADIOTAP: 2838 handle->linktype = DLT_IEEE802_11_RADIO; 2839 break; 2840 2841 case ARPHRD_PPP: 2842 /* 2843 * Some PPP code in the kernel supplies no link-layer 2844 * header whatsoever to PF_PACKET sockets; other PPP 2845 * code supplies PPP link-layer headers ("syncppp.c"); 2846 * some PPP code might supply random link-layer 2847 * headers (PPP over ISDN - there's code in Ethereal, 2848 * for example, to cope with PPP-over-ISDN captures 2849 * with which the Ethereal developers have had to cope, 2850 * heuristically trying to determine which of the 2851 * oddball link-layer headers particular packets have). 2852 * 2853 * As such, we just punt, and run all PPP interfaces 2854 * in cooked mode, if we can; otherwise, we just treat 2855 * it as DLT_RAW, for now - if somebody needs to capture, 2856 * on a 2.0[.x] kernel, on PPP devices that supply a 2857 * link-layer header, they'll have to add code here to 2858 * map to the appropriate DLT_ type (possibly adding a 2859 * new DLT_ type, if necessary). 2860 */ 2861 if (cooked_ok) 2862 handle->linktype = DLT_LINUX_SLL; 2863 else { 2864 /* 2865 * XXX - handle ISDN types here? We can't fall 2866 * back on cooked sockets, so we'd have to 2867 * figure out from the device name what type of 2868 * link-layer encapsulation it's using, and map 2869 * that to an appropriate DLT_ value, meaning 2870 * we'd map "isdnN" devices to DLT_RAW (they 2871 * supply raw IP packets with no link-layer 2872 * header) and "isdY" devices to a new DLT_I4L_IP 2873 * type that has only an Ethernet packet type as 2874 * a link-layer header. 2875 * 2876 * But sometimes we seem to get random crap 2877 * in the link-layer header when capturing on 2878 * ISDN devices.... 2879 */ 2880 handle->linktype = DLT_RAW; 2881 } 2882 break; 2883 2884 #ifndef ARPHRD_CISCO 2885 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */ 2886 #endif 2887 case ARPHRD_CISCO: 2888 handle->linktype = DLT_C_HDLC; 2889 break; 2890 2891 /* Not sure if this is correct for all tunnels, but it 2892 * works for CIPE */ 2893 case ARPHRD_TUNNEL: 2894 #ifndef ARPHRD_SIT 2895 #define ARPHRD_SIT 776 /* From Linux 2.2.13 */ 2896 #endif 2897 case ARPHRD_SIT: 2898 case ARPHRD_CSLIP: 2899 case ARPHRD_SLIP6: 2900 case ARPHRD_CSLIP6: 2901 case ARPHRD_ADAPT: 2902 case ARPHRD_SLIP: 2903 #ifndef ARPHRD_RAWHDLC 2904 #define ARPHRD_RAWHDLC 518 2905 #endif 2906 case ARPHRD_RAWHDLC: 2907 #ifndef ARPHRD_DLCI 2908 #define ARPHRD_DLCI 15 2909 #endif 2910 case ARPHRD_DLCI: 2911 /* 2912 * XXX - should some of those be mapped to DLT_LINUX_SLL 2913 * instead? Should we just map all of them to DLT_LINUX_SLL? 2914 */ 2915 handle->linktype = DLT_RAW; 2916 break; 2917 2918 #ifndef ARPHRD_FRAD 2919 #define ARPHRD_FRAD 770 2920 #endif 2921 case ARPHRD_FRAD: 2922 handle->linktype = DLT_FRELAY; 2923 break; 2924 2925 case ARPHRD_LOCALTLK: 2926 handle->linktype = DLT_LTALK; 2927 break; 2928 2929 case 18: 2930 /* 2931 * RFC 4338 defines an encapsulation for IP and ARP 2932 * packets that's compatible with the RFC 2625 2933 * encapsulation, but that uses a different ARP 2934 * hardware type and hardware addresses. That 2935 * ARP hardware type is 18; Linux doesn't define 2936 * any ARPHRD_ value as 18, but if it ever officially 2937 * supports RFC 4338-style IP-over-FC, it should define 2938 * one. 2939 * 2940 * For now, we map it to DLT_IP_OVER_FC, in the hopes 2941 * that this will encourage its use in the future, 2942 * should Linux ever officially support RFC 4338-style 2943 * IP-over-FC. 2944 */ 2945 handle->linktype = DLT_IP_OVER_FC; 2946 break; 2947 2948 #ifndef ARPHRD_FCPP 2949 #define ARPHRD_FCPP 784 2950 #endif 2951 case ARPHRD_FCPP: 2952 #ifndef ARPHRD_FCAL 2953 #define ARPHRD_FCAL 785 2954 #endif 2955 case ARPHRD_FCAL: 2956 #ifndef ARPHRD_FCPL 2957 #define ARPHRD_FCPL 786 2958 #endif 2959 case ARPHRD_FCPL: 2960 #ifndef ARPHRD_FCFABRIC 2961 #define ARPHRD_FCFABRIC 787 2962 #endif 2963 case ARPHRD_FCFABRIC: 2964 /* 2965 * Back in 2002, Donald Lee at Cray wanted a DLT_ for 2966 * IP-over-FC: 2967 * 2968 * http://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html 2969 * 2970 * and one was assigned. 2971 * 2972 * In a later private discussion (spun off from a message 2973 * on the ethereal-users list) on how to get that DLT_ 2974 * value in libpcap on Linux, I ended up deciding that 2975 * the best thing to do would be to have him tweak the 2976 * driver to set the ARPHRD_ value to some ARPHRD_FCxx 2977 * type, and map all those types to DLT_IP_OVER_FC: 2978 * 2979 * I've checked into the libpcap and tcpdump CVS tree 2980 * support for DLT_IP_OVER_FC. In order to use that, 2981 * you'd have to modify your modified driver to return 2982 * one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" - 2983 * change it to set "dev->type" to ARPHRD_FCFABRIC, for 2984 * example (the exact value doesn't matter, it can be 2985 * any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or 2986 * ARPHRD_FCFABRIC). 2987 * 2988 * 11 years later, Christian Svensson wanted to map 2989 * various ARPHRD_ values to DLT_FC_2 and 2990 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel 2991 * frames: 2992 * 2993 * https://github.com/mcr/libpcap/pull/29 2994 * 2995 * There doesn't seem to be any network drivers that uses 2996 * any of the ARPHRD_FC* values for IP-over-FC, and 2997 * it's not exactly clear what the "Dummy types for non 2998 * ARP hardware" are supposed to mean (link-layer 2999 * header type? Physical network type?), so it's 3000 * not exactly clear why the ARPHRD_FC* types exist 3001 * in the first place. 3002 * 3003 * For now, we map them to DLT_FC_2, and provide an 3004 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as 3005 * DLT_IP_OVER_FC just in case there's some old 3006 * driver out there that uses one of those types for 3007 * IP-over-FC on which somebody wants to capture 3008 * packets. 3009 */ 3010 handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2); 3011 /* 3012 * If that fails, just leave the list empty. 3013 */ 3014 if (handle->dlt_list != NULL) { 3015 handle->dlt_list[0] = DLT_FC_2; 3016 handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS; 3017 handle->dlt_list[2] = DLT_IP_OVER_FC; 3018 handle->dlt_count = 3; 3019 } 3020 handle->linktype = DLT_FC_2; 3021 break; 3022 3023 #ifndef ARPHRD_IRDA 3024 #define ARPHRD_IRDA 783 3025 #endif 3026 case ARPHRD_IRDA: 3027 /* Don't expect IP packet out of this interfaces... */ 3028 handle->linktype = DLT_LINUX_IRDA; 3029 /* We need to save packet direction for IrDA decoding, 3030 * so let's use "Linux-cooked" mode. Jean II 3031 * 3032 * XXX - this is handled in activate_new(). */ 3033 //handlep->cooked = 1; 3034 break; 3035 3036 /* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation 3037 * is needed, please report it to <daniele@orlandi.com> */ 3038 #ifndef ARPHRD_LAPD 3039 #define ARPHRD_LAPD 8445 3040 #endif 3041 case ARPHRD_LAPD: 3042 /* Don't expect IP packet out of this interfaces... */ 3043 handle->linktype = DLT_LINUX_LAPD; 3044 break; 3045 3046 #ifndef ARPHRD_NONE 3047 #define ARPHRD_NONE 0xFFFE 3048 #endif 3049 case ARPHRD_NONE: 3050 /* 3051 * No link-layer header; packets are just IP 3052 * packets, so use DLT_RAW. 3053 */ 3054 handle->linktype = DLT_RAW; 3055 break; 3056 3057 #ifndef ARPHRD_IEEE802154 3058 #define ARPHRD_IEEE802154 804 3059 #endif 3060 case ARPHRD_IEEE802154: 3061 handle->linktype = DLT_IEEE802_15_4_NOFCS; 3062 break; 3063 3064 #ifndef ARPHRD_NETLINK 3065 #define ARPHRD_NETLINK 824 3066 #endif 3067 case ARPHRD_NETLINK: 3068 handle->linktype = DLT_NETLINK; 3069 /* 3070 * We need to use cooked mode, so that in sll_protocol we 3071 * pick up the netlink protocol type such as NETLINK_ROUTE, 3072 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc. 3073 * 3074 * XXX - this is handled in activate_new(). 3075 */ 3076 //handlep->cooked = 1; 3077 break; 3078 3079 default: 3080 handle->linktype = -1; 3081 break; 3082 } 3083 } 3084 3085 /* ===== Functions to interface to the newer kernels ================== */ 3086 3087 /* 3088 * Try to open a packet socket using the new kernel PF_PACKET interface. 3089 * Returns 1 on success, 0 on an error that means the new interface isn't 3090 * present (so the old SOCK_PACKET interface should be tried), and a 3091 * PCAP_ERROR_ value on an error that means that the old mechanism won't 3092 * work either (so it shouldn't be tried). 3093 */ 3094 static int 3095 activate_new(pcap_t *handle) 3096 { 3097 #ifdef HAVE_PF_PACKET_SOCKETS 3098 struct pcap_linux *handlep = handle->priv; 3099 const char *device = handle->opt.source; 3100 int is_any_device = (strcmp(device, "any") == 0); 3101 int sock_fd = -1, arptype; 3102 #ifdef HAVE_PACKET_AUXDATA 3103 int val; 3104 #endif 3105 int err = 0; 3106 struct packet_mreq mr; 3107 #ifdef SO_BPF_EXTENSIONS 3108 int bpf_extensions; 3109 socklen_t len = sizeof(bpf_extensions); 3110 #endif 3111 3112 /* 3113 * Open a socket with protocol family packet. If the 3114 * "any" device was specified, we open a SOCK_DGRAM 3115 * socket for the cooked interface, otherwise we first 3116 * try a SOCK_RAW socket for the raw interface. 3117 */ 3118 sock_fd = is_any_device ? 3119 socket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_ALL)) : 3120 socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL)); 3121 3122 if (sock_fd == -1) { 3123 if (errno == EINVAL || errno == EAFNOSUPPORT) { 3124 /* 3125 * We don't support PF_PACKET/SOCK_whatever 3126 * sockets; try the old mechanism. 3127 */ 3128 return 0; 3129 } 3130 3131 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "socket: %s", 3132 pcap_strerror(errno) ); 3133 if (errno == EPERM || errno == EACCES) { 3134 /* 3135 * You don't have permission to open the 3136 * socket. 3137 */ 3138 return PCAP_ERROR_PERM_DENIED; 3139 } else { 3140 /* 3141 * Other error. 3142 */ 3143 return PCAP_ERROR; 3144 } 3145 } 3146 3147 /* It seems the kernel supports the new interface. */ 3148 handlep->sock_packet = 0; 3149 3150 /* 3151 * Get the interface index of the loopback device. 3152 * If the attempt fails, don't fail, just set the 3153 * "handlep->lo_ifindex" to -1. 3154 * 3155 * XXX - can there be more than one device that loops 3156 * packets back, i.e. devices other than "lo"? If so, 3157 * we'd need to find them all, and have an array of 3158 * indices for them, and check all of them in 3159 * "pcap_read_packet()". 3160 */ 3161 handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf); 3162 3163 /* 3164 * Default value for offset to align link-layer payload 3165 * on a 4-byte boundary. 3166 */ 3167 handle->offset = 0; 3168 3169 /* 3170 * What kind of frames do we have to deal with? Fall back 3171 * to cooked mode if we have an unknown interface type 3172 * or a type we know doesn't work well in raw mode. 3173 */ 3174 if (!is_any_device) { 3175 /* Assume for now we don't need cooked mode. */ 3176 handlep->cooked = 0; 3177 3178 if (handle->opt.rfmon) { 3179 /* 3180 * We were asked to turn on monitor mode. 3181 * Do so before we get the link-layer type, 3182 * because entering monitor mode could change 3183 * the link-layer type. 3184 */ 3185 err = enter_rfmon_mode(handle, sock_fd, device); 3186 if (err < 0) { 3187 /* Hard failure */ 3188 close(sock_fd); 3189 return err; 3190 } 3191 if (err == 0) { 3192 /* 3193 * Nothing worked for turning monitor mode 3194 * on. 3195 */ 3196 close(sock_fd); 3197 return PCAP_ERROR_RFMON_NOTSUP; 3198 } 3199 3200 /* 3201 * Either monitor mode has been turned on for 3202 * the device, or we've been given a different 3203 * device to open for monitor mode. If we've 3204 * been given a different device, use it. 3205 */ 3206 if (handlep->mondevice != NULL) 3207 device = handlep->mondevice; 3208 } 3209 arptype = iface_get_arptype(sock_fd, device, handle->errbuf); 3210 if (arptype < 0) { 3211 close(sock_fd); 3212 return arptype; 3213 } 3214 map_arphrd_to_dlt(handle, sock_fd, arptype, device, 1); 3215 if (handle->linktype == -1 || 3216 handle->linktype == DLT_LINUX_SLL || 3217 handle->linktype == DLT_LINUX_IRDA || 3218 handle->linktype == DLT_LINUX_LAPD || 3219 handle->linktype == DLT_NETLINK || 3220 (handle->linktype == DLT_EN10MB && 3221 (strncmp("isdn", device, 4) == 0 || 3222 strncmp("isdY", device, 4) == 0))) { 3223 /* 3224 * Unknown interface type (-1), or a 3225 * device we explicitly chose to run 3226 * in cooked mode (e.g., PPP devices), 3227 * or an ISDN device (whose link-layer 3228 * type we can only determine by using 3229 * APIs that may be different on different 3230 * kernels) - reopen in cooked mode. 3231 */ 3232 if (close(sock_fd) == -1) { 3233 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3234 "close: %s", pcap_strerror(errno)); 3235 return PCAP_ERROR; 3236 } 3237 sock_fd = socket(PF_PACKET, SOCK_DGRAM, 3238 htons(ETH_P_ALL)); 3239 if (sock_fd == -1) { 3240 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3241 "socket: %s", pcap_strerror(errno)); 3242 if (errno == EPERM || errno == EACCES) { 3243 /* 3244 * You don't have permission to 3245 * open the socket. 3246 */ 3247 return PCAP_ERROR_PERM_DENIED; 3248 } else { 3249 /* 3250 * Other error. 3251 */ 3252 return PCAP_ERROR; 3253 } 3254 } 3255 handlep->cooked = 1; 3256 3257 /* 3258 * Get rid of any link-layer type list 3259 * we allocated - this only supports cooked 3260 * capture. 3261 */ 3262 if (handle->dlt_list != NULL) { 3263 free(handle->dlt_list); 3264 handle->dlt_list = NULL; 3265 handle->dlt_count = 0; 3266 } 3267 3268 if (handle->linktype == -1) { 3269 /* 3270 * Warn that we're falling back on 3271 * cooked mode; we may want to 3272 * update "map_arphrd_to_dlt()" 3273 * to handle the new type. 3274 */ 3275 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3276 "arptype %d not " 3277 "supported by libpcap - " 3278 "falling back to cooked " 3279 "socket", 3280 arptype); 3281 } 3282 3283 /* 3284 * IrDA capture is not a real "cooked" capture, 3285 * it's IrLAP frames, not IP packets. The 3286 * same applies to LAPD capture. 3287 */ 3288 if (handle->linktype != DLT_LINUX_IRDA && 3289 handle->linktype != DLT_LINUX_LAPD && 3290 handle->linktype != DLT_NETLINK) 3291 handle->linktype = DLT_LINUX_SLL; 3292 } 3293 3294 handlep->ifindex = iface_get_id(sock_fd, device, 3295 handle->errbuf); 3296 if (handlep->ifindex == -1) { 3297 close(sock_fd); 3298 return PCAP_ERROR; 3299 } 3300 3301 if ((err = iface_bind(sock_fd, handlep->ifindex, 3302 handle->errbuf)) != 1) { 3303 close(sock_fd); 3304 if (err < 0) 3305 return err; 3306 else 3307 return 0; /* try old mechanism */ 3308 } 3309 } else { 3310 /* 3311 * The "any" device. 3312 */ 3313 if (handle->opt.rfmon) { 3314 /* 3315 * It doesn't support monitor mode. 3316 */ 3317 close(sock_fd); 3318 return PCAP_ERROR_RFMON_NOTSUP; 3319 } 3320 3321 /* 3322 * It uses cooked mode. 3323 */ 3324 handlep->cooked = 1; 3325 handle->linktype = DLT_LINUX_SLL; 3326 3327 /* 3328 * We're not bound to a device. 3329 * For now, we're using this as an indication 3330 * that we can't transmit; stop doing that only 3331 * if we figure out how to transmit in cooked 3332 * mode. 3333 */ 3334 handlep->ifindex = -1; 3335 } 3336 3337 /* 3338 * Select promiscuous mode on if "promisc" is set. 3339 * 3340 * Do not turn allmulti mode on if we don't select 3341 * promiscuous mode - on some devices (e.g., Orinoco 3342 * wireless interfaces), allmulti mode isn't supported 3343 * and the driver implements it by turning promiscuous 3344 * mode on, and that screws up the operation of the 3345 * card as a normal networking interface, and on no 3346 * other platform I know of does starting a non- 3347 * promiscuous capture affect which multicast packets 3348 * are received by the interface. 3349 */ 3350 3351 /* 3352 * Hmm, how can we set promiscuous mode on all interfaces? 3353 * I am not sure if that is possible at all. For now, we 3354 * silently ignore attempts to turn promiscuous mode on 3355 * for the "any" device (so you don't have to explicitly 3356 * disable it in programs such as tcpdump). 3357 */ 3358 3359 if (!is_any_device && handle->opt.promisc) { 3360 memset(&mr, 0, sizeof(mr)); 3361 mr.mr_ifindex = handlep->ifindex; 3362 mr.mr_type = PACKET_MR_PROMISC; 3363 if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP, 3364 &mr, sizeof(mr)) == -1) { 3365 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3366 "setsockopt: %s", pcap_strerror(errno)); 3367 close(sock_fd); 3368 return PCAP_ERROR; 3369 } 3370 } 3371 3372 /* Enable auxillary data if supported and reserve room for 3373 * reconstructing VLAN headers. */ 3374 #ifdef HAVE_PACKET_AUXDATA 3375 val = 1; 3376 if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val, 3377 sizeof(val)) == -1 && errno != ENOPROTOOPT) { 3378 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3379 "setsockopt: %s", pcap_strerror(errno)); 3380 close(sock_fd); 3381 return PCAP_ERROR; 3382 } 3383 handle->offset += VLAN_TAG_LEN; 3384 #endif /* HAVE_PACKET_AUXDATA */ 3385 3386 /* 3387 * This is a 2.2[.x] or later kernel (we know that 3388 * because we're not using a SOCK_PACKET socket - 3389 * PF_PACKET is supported only in 2.2 and later 3390 * kernels). 3391 * 3392 * We can safely pass "recvfrom()" a byte count 3393 * based on the snapshot length. 3394 * 3395 * If we're in cooked mode, make the snapshot length 3396 * large enough to hold a "cooked mode" header plus 3397 * 1 byte of packet data (so we don't pass a byte 3398 * count of 0 to "recvfrom()"). 3399 */ 3400 if (handlep->cooked) { 3401 if (handle->snapshot < SLL_HDR_LEN + 1) 3402 handle->snapshot = SLL_HDR_LEN + 1; 3403 } 3404 handle->bufsize = handle->snapshot; 3405 3406 /* 3407 * Set the offset at which to insert VLAN tags. 3408 */ 3409 switch (handle->linktype) { 3410 3411 case DLT_EN10MB: 3412 handlep->vlan_offset = 2 * ETH_ALEN; 3413 break; 3414 3415 case DLT_LINUX_SLL: 3416 handlep->vlan_offset = 14; 3417 break; 3418 3419 default: 3420 handlep->vlan_offset = -1; /* unknown */ 3421 break; 3422 } 3423 3424 #if defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) 3425 if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) { 3426 int nsec_tstamps = 1; 3427 3428 if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) { 3429 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS"); 3430 close(sock_fd); 3431 return PCAP_ERROR; 3432 } 3433 } 3434 #endif /* defined(SIOCGSTAMPNS) && defined(SO_TIMESTAMPNS) */ 3435 3436 /* 3437 * We've succeeded. Save the socket FD in the pcap structure. 3438 */ 3439 handle->fd = sock_fd; 3440 3441 #ifdef SO_BPF_EXTENSIONS 3442 /* 3443 * Can we generate special code for VLAN checks? 3444 * (XXX - what if we need the special code but it's not supported 3445 * by the OS? Is that possible?) 3446 */ 3447 if (getsockopt(sock_fd, SOL_SOCKET, SO_BPF_EXTENSIONS, 3448 &bpf_extensions, &len) == 0) { 3449 if (bpf_extensions >= SKF_AD_VLAN_TAG_PRESENT) { 3450 /* 3451 * Yes, we can. Request that we do so. 3452 */ 3453 handle->bpf_codegen_flags |= BPF_SPECIAL_VLAN_HANDLING; 3454 } 3455 } 3456 #endif /* SO_BPF_EXTENSIONS */ 3457 3458 return 1; 3459 #else /* HAVE_PF_PACKET_SOCKETS */ 3460 strlcpy(ebuf, 3461 "New packet capturing interface not supported by build " 3462 "environment", PCAP_ERRBUF_SIZE); 3463 return 0; 3464 #endif /* HAVE_PF_PACKET_SOCKETS */ 3465 } 3466 3467 #ifdef HAVE_PACKET_RING 3468 /* 3469 * Attempt to activate with memory-mapped access. 3470 * 3471 * On success, returns 1, and sets *status to 0 if there are no warnings 3472 * or to a PCAP_WARNING_ code if there is a warning. 3473 * 3474 * On failure due to lack of support for memory-mapped capture, returns 3475 * 0. 3476 * 3477 * On error, returns -1, and sets *status to the appropriate error code; 3478 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message. 3479 */ 3480 static int 3481 activate_mmap(pcap_t *handle, int *status) 3482 { 3483 struct pcap_linux *handlep = handle->priv; 3484 int ret; 3485 3486 /* 3487 * Attempt to allocate a buffer to hold the contents of one 3488 * packet, for use by the oneshot callback. 3489 */ 3490 handlep->oneshot_buffer = malloc(handle->snapshot); 3491 if (handlep->oneshot_buffer == NULL) { 3492 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3493 "can't allocate oneshot buffer: %s", 3494 pcap_strerror(errno)); 3495 *status = PCAP_ERROR; 3496 return -1; 3497 } 3498 3499 if (handle->opt.buffer_size == 0) { 3500 /* by default request 2M for the ring buffer */ 3501 handle->opt.buffer_size = 2*1024*1024; 3502 } 3503 ret = prepare_tpacket_socket(handle); 3504 if (ret == -1) { 3505 free(handlep->oneshot_buffer); 3506 *status = PCAP_ERROR; 3507 return ret; 3508 } 3509 ret = create_ring(handle, status); 3510 if (ret == 0) { 3511 /* 3512 * We don't support memory-mapped capture; our caller 3513 * will fall back on reading from the socket. 3514 */ 3515 free(handlep->oneshot_buffer); 3516 return 0; 3517 } 3518 if (ret == -1) { 3519 /* 3520 * Error attempting to enable memory-mapped capture; 3521 * fail. create_ring() has set *status. 3522 */ 3523 free(handlep->oneshot_buffer); 3524 return -1; 3525 } 3526 3527 /* 3528 * Success. *status has been set either to 0 if there are no 3529 * warnings or to a PCAP_WARNING_ value if there is a warning. 3530 * 3531 * Override some defaults and inherit the other fields from 3532 * activate_new. 3533 * handle->offset is used to get the current position into the rx ring. 3534 * handle->cc is used to store the ring size. 3535 */ 3536 3537 switch (handlep->tp_version) { 3538 case TPACKET_V1: 3539 handle->read_op = pcap_read_linux_mmap_v1; 3540 break; 3541 case TPACKET_V1_64: 3542 handle->read_op = pcap_read_linux_mmap_v1_64; 3543 break; 3544 #ifdef HAVE_TPACKET2 3545 case TPACKET_V2: 3546 handle->read_op = pcap_read_linux_mmap_v2; 3547 break; 3548 #endif 3549 #ifdef HAVE_TPACKET3 3550 case TPACKET_V3: 3551 handle->read_op = pcap_read_linux_mmap_v3; 3552 break; 3553 #endif 3554 } 3555 handle->cleanup_op = pcap_cleanup_linux_mmap; 3556 handle->setfilter_op = pcap_setfilter_linux_mmap; 3557 handle->setnonblock_op = pcap_setnonblock_mmap; 3558 handle->getnonblock_op = pcap_getnonblock_mmap; 3559 handle->oneshot_callback = pcap_oneshot_mmap; 3560 handle->selectable_fd = handle->fd; 3561 return 1; 3562 } 3563 #else /* HAVE_PACKET_RING */ 3564 static int 3565 activate_mmap(pcap_t *handle _U_, int *status _U_) 3566 { 3567 return 0; 3568 } 3569 #endif /* HAVE_PACKET_RING */ 3570 3571 #ifdef HAVE_PACKET_RING 3572 3573 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3) 3574 /* 3575 * Attempt to set the socket to the specified version of the memory-mapped 3576 * header. 3577 * 3578 * Return 0 if we succeed; return 1 if we fail because that version isn't 3579 * supported; return -1 on any other error, and set handle->errbuf. 3580 */ 3581 static int 3582 init_tpacket(pcap_t *handle, int version, const char *version_str) 3583 { 3584 struct pcap_linux *handlep = handle->priv; 3585 int val = version; 3586 socklen_t len = sizeof(val); 3587 3588 /* 3589 * Probe whether kernel supports the specified TPACKET version; 3590 * this also gets the length of the header for that version. 3591 */ 3592 if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) { 3593 if (errno == ENOPROTOOPT || errno == EINVAL) 3594 return 1; /* no */ 3595 3596 /* Failed to even find out; this is a fatal error. */ 3597 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3598 "can't get %s header len on packet socket: %s", 3599 version_str, 3600 pcap_strerror(errno)); 3601 return -1; 3602 } 3603 handlep->tp_hdrlen = val; 3604 3605 val = version; 3606 if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val, 3607 sizeof(val)) < 0) { 3608 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3609 "can't activate %s on packet socket: %s", 3610 version_str, 3611 pcap_strerror(errno)); 3612 return -1; 3613 } 3614 handlep->tp_version = version; 3615 3616 /* Reserve space for VLAN tag reconstruction */ 3617 val = VLAN_TAG_LEN; 3618 if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, &val, 3619 sizeof(val)) < 0) { 3620 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3621 "can't set up reserve on packet socket: %s", 3622 pcap_strerror(errno)); 3623 return -1; 3624 } 3625 3626 return 0; 3627 } 3628 #endif /* defined HAVE_TPACKET2 || defined HAVE_TPACKET3 */ 3629 3630 /* 3631 * If the instruction set for which we're compiling has both 32-bit 3632 * and 64-bit versions, and Linux support for the 64-bit version 3633 * predates TPACKET_V2, define ISA_64_BIT as the .machine value 3634 * you get from uname() for the 64-bit version. Otherwise, leave 3635 * it undefined. (This includes ARM, which has a 64-bit version, 3636 * but Linux support for it appeared well after TPACKET_V2 support 3637 * did, so there should never be a case where 32-bit ARM code is 3638 * running o a 64-bit kernel that only supports TPACKET_V1.) 3639 * 3640 * If we've omitted your favorite such architecture, please contribute 3641 * a patch. (No patch is needed for architectures that are 32-bit-only 3642 * or for which Linux has no support for 32-bit userland - or for which, 3643 * as noted, 64-bit support appeared in Linux after TPACKET_V2 support 3644 * did.) 3645 */ 3646 #if defined(__i386__) 3647 #define ISA_64_BIT "x86_64" 3648 #elif defined(__ppc__) 3649 #define ISA_64_BIT "ppc64" 3650 #elif defined(__sparc__) 3651 #define ISA_64_BIT "sparc64" 3652 #elif defined(__s390__) 3653 #define ISA_64_BIT "s390x" 3654 #elif defined(__mips__) 3655 #define ISA_64_BIT "mips64" 3656 #elif defined(__hppa__) 3657 #define ISA_64_BIT "parisc64" 3658 #endif 3659 3660 /* 3661 * Attempt to set the socket to version 3 of the memory-mapped header and, 3662 * if that fails because version 3 isn't supported, attempt to fall 3663 * back to version 2. If version 2 isn't supported, just leave it at 3664 * version 1. 3665 * 3666 * Return 1 if we succeed or if we fail because neither version 2 nor 3 is 3667 * supported; return -1 on any other error, and set handle->errbuf. 3668 */ 3669 static int 3670 prepare_tpacket_socket(pcap_t *handle) 3671 { 3672 struct pcap_linux *handlep = handle->priv; 3673 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3) 3674 int ret; 3675 #endif 3676 3677 #ifdef HAVE_TPACKET3 3678 /* 3679 * Try setting the version to TPACKET_V3. 3680 * 3681 * The only mode in which buffering is done on PF_PACKET 3682 * sockets, so that packets might not be delivered 3683 * immediately, is TPACKET_V3 mode. 3684 * 3685 * The buffering cannot be disabled in that mode, so 3686 * if the user has requested immediate mode, we don't 3687 * use TPACKET_V3. 3688 */ 3689 if (!handle->opt.immediate) { 3690 ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3"); 3691 if (ret == 0) { 3692 /* 3693 * Success. 3694 */ 3695 return 1; 3696 } 3697 if (ret == -1) { 3698 /* 3699 * We failed for some reason other than "the 3700 * kernel doesn't support TPACKET_V3". 3701 */ 3702 return -1; 3703 } 3704 } 3705 #endif /* HAVE_TPACKET3 */ 3706 3707 #ifdef HAVE_TPACKET2 3708 /* 3709 * Try setting the version to TPACKET_V2. 3710 */ 3711 ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2"); 3712 if (ret == 0) { 3713 /* 3714 * Success. 3715 */ 3716 return 1; 3717 } 3718 if (ret == -1) { 3719 /* 3720 * We failed for some reason other than "the 3721 * kernel doesn't support TPACKET_V2". 3722 */ 3723 return -1; 3724 } 3725 #endif /* HAVE_TPACKET2 */ 3726 3727 /* 3728 * OK, we're using TPACKET_V1, as that's all the kernel supports. 3729 */ 3730 handlep->tp_version = TPACKET_V1; 3731 handlep->tp_hdrlen = sizeof(struct tpacket_hdr); 3732 3733 #ifdef ISA_64_BIT 3734 /* 3735 * 32-bit userspace + 64-bit kernel + TPACKET_V1 are not compatible with 3736 * each other due to platform-dependent data type size differences. 3737 * 3738 * If we have a 32-bit userland and a 64-bit kernel, use an 3739 * internally-defined TPACKET_V1_64, with which we use a 64-bit 3740 * version of the data structures. 3741 */ 3742 if (sizeof(long) == 4) { 3743 /* 3744 * This is 32-bit code. 3745 */ 3746 struct utsname utsname; 3747 3748 if (uname(&utsname) == -1) { 3749 /* 3750 * Failed. 3751 */ 3752 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3753 "uname failed: %s", pcap_strerror(errno)); 3754 return -1; 3755 } 3756 if (strcmp(utsname.machine, ISA_64_BIT) == 0) { 3757 /* 3758 * uname() tells us the machine is 64-bit, 3759 * so we presumably have a 64-bit kernel. 3760 * 3761 * XXX - this presumes that uname() won't lie 3762 * in 32-bit code and claim that the machine 3763 * has the 32-bit version of the ISA. 3764 */ 3765 handlep->tp_version = TPACKET_V1_64; 3766 handlep->tp_hdrlen = sizeof(struct tpacket_hdr_64); 3767 } 3768 } 3769 #endif 3770 3771 return 1; 3772 } 3773 3774 /* 3775 * Attempt to set up memory-mapped access. 3776 * 3777 * On success, returns 1, and sets *status to 0 if there are no warnings 3778 * or to a PCAP_WARNING_ code if there is a warning. 3779 * 3780 * On failure due to lack of support for memory-mapped capture, returns 3781 * 0. 3782 * 3783 * On error, returns -1, and sets *status to the appropriate error code; 3784 * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message. 3785 */ 3786 static int 3787 create_ring(pcap_t *handle, int *status) 3788 { 3789 struct pcap_linux *handlep = handle->priv; 3790 unsigned i, j, frames_per_block; 3791 #ifdef HAVE_TPACKET3 3792 /* 3793 * For sockets using TPACKET_V1 or TPACKET_V2, the extra 3794 * stuff at the end of a struct tpacket_req3 will be 3795 * ignored, so this is OK even for those sockets. 3796 */ 3797 struct tpacket_req3 req; 3798 #else 3799 struct tpacket_req req; 3800 #endif 3801 socklen_t len; 3802 unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff; 3803 unsigned int frame_size; 3804 3805 /* 3806 * Start out assuming no warnings or errors. 3807 */ 3808 *status = 0; 3809 3810 switch (handlep->tp_version) { 3811 3812 case TPACKET_V1: 3813 case TPACKET_V1_64: 3814 #ifdef HAVE_TPACKET2 3815 case TPACKET_V2: 3816 #endif 3817 /* Note that with large snapshot length (say 64K, which is 3818 * the default for recent versions of tcpdump, the value that 3819 * "-s 0" has given for a long time with tcpdump, and the 3820 * default in Wireshark/TShark/dumpcap), if we use the snapshot 3821 * length to calculate the frame length, only a few frames 3822 * will be available in the ring even with pretty 3823 * large ring size (and a lot of memory will be unused). 3824 * 3825 * Ideally, we should choose a frame length based on the 3826 * minimum of the specified snapshot length and the maximum 3827 * packet size. That's not as easy as it sounds; consider, 3828 * for example, an 802.11 interface in monitor mode, where 3829 * the frame would include a radiotap header, where the 3830 * maximum radiotap header length is device-dependent. 3831 * 3832 * So, for now, we just do this for Ethernet devices, where 3833 * there's no metadata header, and the link-layer header is 3834 * fixed length. We can get the maximum packet size by 3835 * adding 18, the Ethernet header length plus the CRC length 3836 * (just in case we happen to get the CRC in the packet), to 3837 * the MTU of the interface; we fetch the MTU in the hopes 3838 * that it reflects support for jumbo frames. (Even if the 3839 * interface is just being used for passive snooping, the 3840 * driver might set the size of buffers in the receive ring 3841 * based on the MTU, so that the MTU limits the maximum size 3842 * of packets that we can receive.) 3843 * 3844 * We don't do that if segmentation/fragmentation or receive 3845 * offload are enabled, so we don't get rudely surprised by 3846 * "packets" bigger than the MTU. */ 3847 frame_size = handle->snapshot; 3848 if (handle->linktype == DLT_EN10MB) { 3849 int mtu; 3850 int offload; 3851 3852 offload = iface_get_offload(handle); 3853 if (offload == -1) { 3854 *status = PCAP_ERROR; 3855 return -1; 3856 } 3857 if (!offload) { 3858 mtu = iface_get_mtu(handle->fd, handle->opt.source, 3859 handle->errbuf); 3860 if (mtu == -1) { 3861 *status = PCAP_ERROR; 3862 return -1; 3863 } 3864 if (frame_size > mtu + 18) 3865 frame_size = mtu + 18; 3866 } 3867 } 3868 3869 /* NOTE: calculus matching those in tpacket_rcv() 3870 * in linux-2.6/net/packet/af_packet.c 3871 */ 3872 len = sizeof(sk_type); 3873 if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type, 3874 &len) < 0) { 3875 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3876 "getsockopt: %s", pcap_strerror(errno)); 3877 *status = PCAP_ERROR; 3878 return -1; 3879 } 3880 #ifdef PACKET_RESERVE 3881 len = sizeof(tp_reserve); 3882 if (getsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE, 3883 &tp_reserve, &len) < 0) { 3884 if (errno != ENOPROTOOPT) { 3885 /* 3886 * ENOPROTOOPT means "kernel doesn't support 3887 * PACKET_RESERVE", in which case we fall back 3888 * as best we can. 3889 */ 3890 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3891 "getsockopt: %s", pcap_strerror(errno)); 3892 *status = PCAP_ERROR; 3893 return -1; 3894 } 3895 tp_reserve = 0; /* older kernel, reserve not supported */ 3896 } 3897 #else 3898 tp_reserve = 0; /* older kernel, reserve not supported */ 3899 #endif 3900 maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE; 3901 /* XXX: in the kernel maclen is calculated from 3902 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len 3903 * in: packet_snd() in linux-2.6/net/packet/af_packet.c 3904 * then packet_alloc_skb() in linux-2.6/net/packet/af_packet.c 3905 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c 3906 * but I see no way to get those sizes in userspace, 3907 * like for instance with an ifreq ioctl(); 3908 * the best thing I've found so far is MAX_HEADER in 3909 * the kernel part of linux-2.6/include/linux/netdevice.h 3910 * which goes up to 128+48=176; since pcap-linux.c 3911 * defines a MAX_LINKHEADER_SIZE of 256 which is 3912 * greater than that, let's use it.. maybe is it even 3913 * large enough to directly replace macoff.. 3914 */ 3915 tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ; 3916 netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve; 3917 /* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN 3918 * of netoff, which contradicts 3919 * linux-2.6/Documentation/networking/packet_mmap.txt 3920 * documenting that: 3921 * "- Gap, chosen so that packet data (Start+tp_net) 3922 * aligns to TPACKET_ALIGNMENT=16" 3923 */ 3924 /* NOTE: in linux-2.6/include/linux/skbuff.h: 3925 * "CPUs often take a performance hit 3926 * when accessing unaligned memory locations" 3927 */ 3928 macoff = netoff - maclen; 3929 req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size); 3930 req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size; 3931 break; 3932 3933 #ifdef HAVE_TPACKET3 3934 case TPACKET_V3: 3935 /* The "frames" for this are actually buffers that 3936 * contain multiple variable-sized frames. 3937 * 3938 * We pick a "frame" size of 128K to leave enough 3939 * room for at least one reasonably-sized packet 3940 * in the "frame". */ 3941 req.tp_frame_size = MAXIMUM_SNAPLEN; 3942 req.tp_frame_nr = handle->opt.buffer_size/req.tp_frame_size; 3943 break; 3944 #endif 3945 default: 3946 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 3947 "Internal error: unknown TPACKET_ value %u", 3948 handlep->tp_version); 3949 *status = PCAP_ERROR; 3950 return -1; 3951 } 3952 3953 /* compute the minumum block size that will handle this frame. 3954 * The block has to be page size aligned. 3955 * The max block size allowed by the kernel is arch-dependent and 3956 * it's not explicitly checked here. */ 3957 req.tp_block_size = getpagesize(); 3958 while (req.tp_block_size < req.tp_frame_size) 3959 req.tp_block_size <<= 1; 3960 3961 frames_per_block = req.tp_block_size/req.tp_frame_size; 3962 3963 /* 3964 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was, 3965 * so we check for PACKET_TIMESTAMP. We check for 3966 * linux/net_tstamp.h just in case a system somehow has 3967 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might 3968 * be unnecessary. 3969 * 3970 * SIOCSHWTSTAMP was introduced in the patch that introduced 3971 * linux/net_tstamp.h, so we don't bother checking whether 3972 * SIOCSHWTSTAMP is defined (if your Linux system has 3973 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your 3974 * Linux system is badly broken). 3975 */ 3976 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) 3977 /* 3978 * If we were told to do so, ask the kernel and the driver 3979 * to use hardware timestamps. 3980 * 3981 * Hardware timestamps are only supported with mmapped 3982 * captures. 3983 */ 3984 if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER || 3985 handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) { 3986 struct hwtstamp_config hwconfig; 3987 struct ifreq ifr; 3988 int timesource; 3989 3990 /* 3991 * Ask for hardware time stamps on all packets, 3992 * including transmitted packets. 3993 */ 3994 memset(&hwconfig, 0, sizeof(hwconfig)); 3995 hwconfig.tx_type = HWTSTAMP_TX_ON; 3996 hwconfig.rx_filter = HWTSTAMP_FILTER_ALL; 3997 3998 memset(&ifr, 0, sizeof(ifr)); 3999 strlcpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name)); 4000 ifr.ifr_data = (void *)&hwconfig; 4001 4002 if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) { 4003 switch (errno) { 4004 4005 case EPERM: 4006 /* 4007 * Treat this as an error, as the 4008 * user should try to run this 4009 * with the appropriate privileges - 4010 * and, if they can't, shouldn't 4011 * try requesting hardware time stamps. 4012 */ 4013 *status = PCAP_ERROR_PERM_DENIED; 4014 return -1; 4015 4016 case EOPNOTSUPP: 4017 /* 4018 * Treat this as a warning, as the 4019 * only way to fix the warning is to 4020 * get an adapter that supports hardware 4021 * time stamps. We'll just fall back 4022 * on the standard host time stamps. 4023 */ 4024 *status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP; 4025 break; 4026 4027 default: 4028 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4029 "SIOCSHWTSTAMP failed: %s", 4030 pcap_strerror(errno)); 4031 *status = PCAP_ERROR; 4032 return -1; 4033 } 4034 } else { 4035 /* 4036 * Well, that worked. Now specify the type of 4037 * hardware time stamp we want for this 4038 * socket. 4039 */ 4040 if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) { 4041 /* 4042 * Hardware timestamp, synchronized 4043 * with the system clock. 4044 */ 4045 timesource = SOF_TIMESTAMPING_SYS_HARDWARE; 4046 } else { 4047 /* 4048 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware 4049 * timestamp, not synchronized with the 4050 * system clock. 4051 */ 4052 timesource = SOF_TIMESTAMPING_RAW_HARDWARE; 4053 } 4054 if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP, 4055 (void *)×ource, sizeof(timesource))) { 4056 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4057 "can't set PACKET_TIMESTAMP: %s", 4058 pcap_strerror(errno)); 4059 *status = PCAP_ERROR; 4060 return -1; 4061 } 4062 } 4063 } 4064 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */ 4065 4066 /* ask the kernel to create the ring */ 4067 retry: 4068 req.tp_block_nr = req.tp_frame_nr / frames_per_block; 4069 4070 /* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */ 4071 req.tp_frame_nr = req.tp_block_nr * frames_per_block; 4072 4073 #ifdef HAVE_TPACKET3 4074 /* timeout value to retire block - use the configured buffering timeout, or default if <0. */ 4075 req.tp_retire_blk_tov = (handlep->timeout>=0)?handlep->timeout:0; 4076 /* private data not used */ 4077 req.tp_sizeof_priv = 0; 4078 /* Rx ring - feature request bits - none (rxhash will not be filled) */ 4079 req.tp_feature_req_word = 0; 4080 #endif 4081 4082 if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING, 4083 (void *) &req, sizeof(req))) { 4084 if ((errno == ENOMEM) && (req.tp_block_nr > 1)) { 4085 /* 4086 * Memory failure; try to reduce the requested ring 4087 * size. 4088 * 4089 * We used to reduce this by half -- do 5% instead. 4090 * That may result in more iterations and a longer 4091 * startup, but the user will be much happier with 4092 * the resulting buffer size. 4093 */ 4094 if (req.tp_frame_nr < 20) 4095 req.tp_frame_nr -= 1; 4096 else 4097 req.tp_frame_nr -= req.tp_frame_nr/20; 4098 goto retry; 4099 } 4100 if (errno == ENOPROTOOPT) { 4101 /* 4102 * We don't have ring buffer support in this kernel. 4103 */ 4104 return 0; 4105 } 4106 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4107 "can't create rx ring on packet socket: %s", 4108 pcap_strerror(errno)); 4109 *status = PCAP_ERROR; 4110 return -1; 4111 } 4112 4113 /* memory map the rx ring */ 4114 handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size; 4115 handlep->mmapbuf = mmap(0, handlep->mmapbuflen, 4116 PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0); 4117 if (handlep->mmapbuf == MAP_FAILED) { 4118 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4119 "can't mmap rx ring: %s", pcap_strerror(errno)); 4120 4121 /* clear the allocated ring on error*/ 4122 destroy_ring(handle); 4123 *status = PCAP_ERROR; 4124 return -1; 4125 } 4126 4127 /* allocate a ring for each frame header pointer*/ 4128 handle->cc = req.tp_frame_nr; 4129 handle->buffer = malloc(handle->cc * sizeof(union thdr *)); 4130 if (!handle->buffer) { 4131 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4132 "can't allocate ring of frame headers: %s", 4133 pcap_strerror(errno)); 4134 4135 destroy_ring(handle); 4136 *status = PCAP_ERROR; 4137 return -1; 4138 } 4139 4140 /* fill the header ring with proper frame ptr*/ 4141 handle->offset = 0; 4142 for (i=0; i<req.tp_block_nr; ++i) { 4143 void *base = &handlep->mmapbuf[i*req.tp_block_size]; 4144 for (j=0; j<frames_per_block; ++j, ++handle->offset) { 4145 RING_GET_FRAME(handle) = base; 4146 base += req.tp_frame_size; 4147 } 4148 } 4149 4150 handle->bufsize = req.tp_frame_size; 4151 handle->offset = 0; 4152 return 1; 4153 } 4154 4155 /* free all ring related resources*/ 4156 static void 4157 destroy_ring(pcap_t *handle) 4158 { 4159 struct pcap_linux *handlep = handle->priv; 4160 4161 /* tell the kernel to destroy the ring*/ 4162 struct tpacket_req req; 4163 memset(&req, 0, sizeof(req)); 4164 /* do not test for setsockopt failure, as we can't recover from any error */ 4165 (void)setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING, 4166 (void *) &req, sizeof(req)); 4167 4168 /* if ring is mapped, unmap it*/ 4169 if (handlep->mmapbuf) { 4170 /* do not test for mmap failure, as we can't recover from any error */ 4171 (void)munmap(handlep->mmapbuf, handlep->mmapbuflen); 4172 handlep->mmapbuf = NULL; 4173 } 4174 } 4175 4176 /* 4177 * Special one-shot callback, used for pcap_next() and pcap_next_ex(), 4178 * for Linux mmapped capture. 4179 * 4180 * The problem is that pcap_next() and pcap_next_ex() expect the packet 4181 * data handed to the callback to be valid after the callback returns, 4182 * but pcap_read_linux_mmap() has to release that packet as soon as 4183 * the callback returns (otherwise, the kernel thinks there's still 4184 * at least one unprocessed packet available in the ring, so a select() 4185 * will immediately return indicating that there's data to process), so, 4186 * in the callback, we have to make a copy of the packet. 4187 * 4188 * Yes, this means that, if the capture is using the ring buffer, using 4189 * pcap_next() or pcap_next_ex() requires more copies than using 4190 * pcap_loop() or pcap_dispatch(). If that bothers you, don't use 4191 * pcap_next() or pcap_next_ex(). 4192 */ 4193 static void 4194 pcap_oneshot_mmap(u_char *user, const struct pcap_pkthdr *h, 4195 const u_char *bytes) 4196 { 4197 struct oneshot_userdata *sp = (struct oneshot_userdata *)user; 4198 pcap_t *handle = sp->pd; 4199 struct pcap_linux *handlep = handle->priv; 4200 4201 *sp->hdr = *h; 4202 memcpy(handlep->oneshot_buffer, bytes, h->caplen); 4203 *sp->pkt = handlep->oneshot_buffer; 4204 } 4205 4206 static void 4207 pcap_cleanup_linux_mmap( pcap_t *handle ) 4208 { 4209 struct pcap_linux *handlep = handle->priv; 4210 4211 destroy_ring(handle); 4212 if (handlep->oneshot_buffer != NULL) { 4213 free(handlep->oneshot_buffer); 4214 handlep->oneshot_buffer = NULL; 4215 } 4216 pcap_cleanup_linux(handle); 4217 } 4218 4219 4220 static int 4221 pcap_getnonblock_mmap(pcap_t *p, char *errbuf) 4222 { 4223 struct pcap_linux *handlep = p->priv; 4224 4225 /* use negative value of timeout to indicate non blocking ops */ 4226 return (handlep->timeout<0); 4227 } 4228 4229 static int 4230 pcap_setnonblock_mmap(pcap_t *p, int nonblock, char *errbuf) 4231 { 4232 struct pcap_linux *handlep = p->priv; 4233 4234 /* 4235 * Set the file descriptor to non-blocking mode, as we use 4236 * it for sending packets. 4237 */ 4238 if (pcap_setnonblock_fd(p, nonblock, errbuf) == -1) 4239 return -1; 4240 4241 /* 4242 * Map each value to their corresponding negation to 4243 * preserve the timeout value provided with pcap_set_timeout. 4244 */ 4245 if (nonblock) { 4246 if (handlep->timeout >= 0) { 4247 /* 4248 * Indicate that we're switching to 4249 * non-blocking mode. 4250 */ 4251 handlep->timeout = ~handlep->timeout; 4252 } 4253 } else { 4254 if (handlep->timeout < 0) { 4255 handlep->timeout = ~handlep->timeout; 4256 } 4257 } 4258 return 0; 4259 } 4260 4261 static inline union thdr * 4262 pcap_get_ring_frame(pcap_t *handle, int status) 4263 { 4264 struct pcap_linux *handlep = handle->priv; 4265 union thdr h; 4266 4267 h.raw = RING_GET_FRAME(handle); 4268 switch (handlep->tp_version) { 4269 case TPACKET_V1: 4270 if (status != (h.h1->tp_status ? TP_STATUS_USER : 4271 TP_STATUS_KERNEL)) 4272 return NULL; 4273 break; 4274 case TPACKET_V1_64: 4275 if (status != (h.h1_64->tp_status ? TP_STATUS_USER : 4276 TP_STATUS_KERNEL)) 4277 return NULL; 4278 break; 4279 #ifdef HAVE_TPACKET2 4280 case TPACKET_V2: 4281 if (status != (h.h2->tp_status ? TP_STATUS_USER : 4282 TP_STATUS_KERNEL)) 4283 return NULL; 4284 break; 4285 #endif 4286 #ifdef HAVE_TPACKET3 4287 case TPACKET_V3: 4288 if (status != (h.h3->hdr.bh1.block_status ? TP_STATUS_USER : 4289 TP_STATUS_KERNEL)) 4290 return NULL; 4291 break; 4292 #endif 4293 } 4294 return h.raw; 4295 } 4296 4297 #ifndef POLLRDHUP 4298 #define POLLRDHUP 0 4299 #endif 4300 4301 /* wait for frames availability.*/ 4302 static int pcap_wait_for_frames_mmap(pcap_t *handle) 4303 { 4304 if (!pcap_get_ring_frame(handle, TP_STATUS_USER)) { 4305 struct pcap_linux *handlep = handle->priv; 4306 int timeout; 4307 char c; 4308 struct pollfd pollinfo; 4309 int ret; 4310 4311 pollinfo.fd = handle->fd; 4312 pollinfo.events = POLLIN; 4313 4314 if (handlep->timeout == 0) { 4315 #ifdef HAVE_TPACKET3 4316 /* 4317 * XXX - due to a set of (mis)features in the 4318 * TPACKET_V3 kernel code, blocking forever with 4319 * a TPACKET_V3 socket can, if few packets 4320 * are arriving and passing the socket filter, 4321 * cause most packets to be dropped. See 4322 * libpcap issue #335 for the full painful 4323 * story. The workaround is to have poll() 4324 * time out very quickly, so we grab the 4325 * frames handed to us, and return them to 4326 * the kernel, ASAP. 4327 * 4328 * If those issues are ever fixed, we might 4329 * want to check the kernel version and block 4330 * forever with TPACKET_V3 if we're running 4331 * with a kernel that has the fix. 4332 */ 4333 if (handlep->tp_version == TPACKET_V3) 4334 timeout = 1; /* don't block for very long */ 4335 else 4336 #endif 4337 timeout = -1; /* block forever */ 4338 } else if (handlep->timeout > 0) 4339 timeout = handlep->timeout; /* block for that amount of time */ 4340 else 4341 timeout = 0; /* non-blocking mode - poll to pick up errors */ 4342 do { 4343 ret = poll(&pollinfo, 1, timeout); 4344 if (ret < 0 && errno != EINTR) { 4345 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4346 "can't poll on packet socket: %s", 4347 pcap_strerror(errno)); 4348 return PCAP_ERROR; 4349 } else if (ret > 0 && 4350 (pollinfo.revents & (POLLHUP|POLLRDHUP|POLLERR|POLLNVAL))) { 4351 /* 4352 * There's some indication other than 4353 * "you can read on this descriptor" on 4354 * the descriptor. 4355 */ 4356 if (pollinfo.revents & (POLLHUP | POLLRDHUP)) { 4357 snprintf(handle->errbuf, 4358 PCAP_ERRBUF_SIZE, 4359 "Hangup on packet socket"); 4360 return PCAP_ERROR; 4361 } 4362 if (pollinfo.revents & POLLERR) { 4363 /* 4364 * A recv() will give us the 4365 * actual error code. 4366 * 4367 * XXX - make the socket non-blocking? 4368 */ 4369 if (recv(handle->fd, &c, sizeof c, 4370 MSG_PEEK) != -1) 4371 continue; /* what, no error? */ 4372 if (errno == ENETDOWN) { 4373 /* 4374 * The device on which we're 4375 * capturing went away. 4376 * 4377 * XXX - we should really return 4378 * PCAP_ERROR_IFACE_NOT_UP, 4379 * but pcap_dispatch() etc. 4380 * aren't defined to return 4381 * that. 4382 */ 4383 snprintf(handle->errbuf, 4384 PCAP_ERRBUF_SIZE, 4385 "The interface went down"); 4386 } else { 4387 snprintf(handle->errbuf, 4388 PCAP_ERRBUF_SIZE, 4389 "Error condition on packet socket: %s", 4390 strerror(errno)); 4391 } 4392 return PCAP_ERROR; 4393 } 4394 if (pollinfo.revents & POLLNVAL) { 4395 snprintf(handle->errbuf, 4396 PCAP_ERRBUF_SIZE, 4397 "Invalid polling request on packet socket"); 4398 return PCAP_ERROR; 4399 } 4400 } 4401 /* check for break loop condition on interrupted syscall*/ 4402 if (handle->break_loop) { 4403 handle->break_loop = 0; 4404 return PCAP_ERROR_BREAK; 4405 } 4406 } while (ret < 0); 4407 } 4408 return 0; 4409 } 4410 4411 /* handle a single memory mapped packet */ 4412 static int pcap_handle_packet_mmap( 4413 pcap_t *handle, 4414 pcap_handler callback, 4415 u_char *user, 4416 unsigned char *frame, 4417 unsigned int tp_len, 4418 unsigned int tp_mac, 4419 unsigned int tp_snaplen, 4420 unsigned int tp_sec, 4421 unsigned int tp_usec, 4422 int tp_vlan_tci_valid, 4423 __u16 tp_vlan_tci, 4424 __u16 tp_vlan_tpid) 4425 { 4426 struct pcap_linux *handlep = handle->priv; 4427 unsigned char *bp; 4428 struct sockaddr_ll *sll; 4429 struct pcap_pkthdr pcaphdr; 4430 4431 /* perform sanity check on internal offset. */ 4432 if (tp_mac + tp_snaplen > handle->bufsize) { 4433 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4434 "corrupted frame on kernel ring mac " 4435 "offset %u + caplen %u > frame len %d", 4436 tp_mac, tp_snaplen, handle->bufsize); 4437 return -1; 4438 } 4439 4440 /* run filter on received packet 4441 * If the kernel filtering is enabled we need to run the 4442 * filter until all the frames present into the ring 4443 * at filter creation time are processed. 4444 * In this case, blocks_to_filter_in_userland is used 4445 * as a counter for the packet we need to filter. 4446 * Note: alternatively it could be possible to stop applying 4447 * the filter when the ring became empty, but it can possibly 4448 * happen a lot later... */ 4449 bp = frame + tp_mac; 4450 4451 /* if required build in place the sll header*/ 4452 sll = (void *)frame + TPACKET_ALIGN(handlep->tp_hdrlen); 4453 if (handlep->cooked) { 4454 struct sll_header *hdrp; 4455 4456 /* 4457 * The kernel should have left us with enough 4458 * space for an sll header; back up the packet 4459 * data pointer into that space, as that'll be 4460 * the beginning of the packet we pass to the 4461 * callback. 4462 */ 4463 bp -= SLL_HDR_LEN; 4464 4465 /* 4466 * Let's make sure that's past the end of 4467 * the tpacket header, i.e. >= 4468 * ((u_char *)thdr + TPACKET_HDRLEN), so we 4469 * don't step on the header when we construct 4470 * the sll header. 4471 */ 4472 if (bp < (u_char *)frame + 4473 TPACKET_ALIGN(handlep->tp_hdrlen) + 4474 sizeof(struct sockaddr_ll)) { 4475 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 4476 "cooked-mode frame doesn't have room for sll header"); 4477 return -1; 4478 } 4479 4480 /* 4481 * OK, that worked; construct the sll header. 4482 */ 4483 hdrp = (struct sll_header *)bp; 4484 hdrp->sll_pkttype = map_packet_type_to_sll_type( 4485 sll->sll_pkttype); 4486 hdrp->sll_hatype = htons(sll->sll_hatype); 4487 hdrp->sll_halen = htons(sll->sll_halen); 4488 memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN); 4489 hdrp->sll_protocol = sll->sll_protocol; 4490 } 4491 4492 if (handlep->filter_in_userland && handle->fcode.bf_insns) { 4493 struct bpf_aux_data aux_data; 4494 4495 aux_data.vlan_tag = tp_vlan_tci & 0x0fff; 4496 aux_data.vlan_tag_present = tp_vlan_tci_valid; 4497 4498 if (bpf_filter_with_aux_data(handle->fcode.bf_insns, bp, 4499 tp_len, tp_snaplen, &aux_data) == 0) 4500 return 0; 4501 } 4502 4503 if (!linux_check_direction(handle, sll)) 4504 return 0; 4505 4506 /* get required packet info from ring header */ 4507 pcaphdr.ts.tv_sec = tp_sec; 4508 pcaphdr.ts.tv_usec = tp_usec; 4509 pcaphdr.caplen = tp_snaplen; 4510 pcaphdr.len = tp_len; 4511 4512 /* if required build in place the sll header*/ 4513 if (handlep->cooked) { 4514 /* update packet len */ 4515 pcaphdr.caplen += SLL_HDR_LEN; 4516 pcaphdr.len += SLL_HDR_LEN; 4517 } 4518 4519 #if defined(HAVE_TPACKET2) || defined(HAVE_TPACKET3) 4520 if (tp_vlan_tci_valid && 4521 handlep->vlan_offset != -1 && 4522 tp_snaplen >= (unsigned int) handlep->vlan_offset) 4523 { 4524 struct vlan_tag *tag; 4525 4526 bp -= VLAN_TAG_LEN; 4527 memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset); 4528 4529 tag = (struct vlan_tag *)(bp + handlep->vlan_offset); 4530 tag->vlan_tpid = htons(tp_vlan_tpid); 4531 tag->vlan_tci = htons(tp_vlan_tci); 4532 4533 pcaphdr.caplen += VLAN_TAG_LEN; 4534 pcaphdr.len += VLAN_TAG_LEN; 4535 } 4536 #endif 4537 4538 /* 4539 * The only way to tell the kernel to cut off the 4540 * packet at a snapshot length is with a filter program; 4541 * if there's no filter program, the kernel won't cut 4542 * the packet off. 4543 * 4544 * Trim the snapshot length to be no longer than the 4545 * specified snapshot length. 4546 */ 4547 if (pcaphdr.caplen > handle->snapshot) 4548 pcaphdr.caplen = handle->snapshot; 4549 4550 /* pass the packet to the user */ 4551 callback(user, &pcaphdr, bp); 4552 4553 return 1; 4554 } 4555 4556 static int 4557 pcap_read_linux_mmap_v1(pcap_t *handle, int max_packets, pcap_handler callback, 4558 u_char *user) 4559 { 4560 struct pcap_linux *handlep = handle->priv; 4561 int pkts = 0; 4562 int ret; 4563 4564 /* wait for frames availability.*/ 4565 ret = pcap_wait_for_frames_mmap(handle); 4566 if (ret) { 4567 return ret; 4568 } 4569 4570 /* non-positive values of max_packets are used to require all 4571 * packets currently available in the ring */ 4572 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) { 4573 union thdr h; 4574 4575 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER); 4576 if (!h.raw) 4577 break; 4578 4579 ret = pcap_handle_packet_mmap( 4580 handle, 4581 callback, 4582 user, 4583 h.raw, 4584 h.h1->tp_len, 4585 h.h1->tp_mac, 4586 h.h1->tp_snaplen, 4587 h.h1->tp_sec, 4588 h.h1->tp_usec, 4589 0, 4590 0, 4591 0); 4592 if (ret == 1) { 4593 pkts++; 4594 handlep->packets_read++; 4595 } else if (ret < 0) { 4596 return ret; 4597 } 4598 4599 /* 4600 * Hand this block back to the kernel, and, if we're 4601 * counting blocks that need to be filtered in userland 4602 * after having been filtered by the kernel, count 4603 * the one we've just processed. 4604 */ 4605 h.h1->tp_status = TP_STATUS_KERNEL; 4606 if (handlep->blocks_to_filter_in_userland > 0) { 4607 handlep->blocks_to_filter_in_userland--; 4608 if (handlep->blocks_to_filter_in_userland == 0) { 4609 /* 4610 * No more blocks need to be filtered 4611 * in userland. 4612 */ 4613 handlep->filter_in_userland = 0; 4614 } 4615 } 4616 4617 /* next block */ 4618 if (++handle->offset >= handle->cc) 4619 handle->offset = 0; 4620 4621 /* check for break loop condition*/ 4622 if (handle->break_loop) { 4623 handle->break_loop = 0; 4624 return PCAP_ERROR_BREAK; 4625 } 4626 } 4627 return pkts; 4628 } 4629 4630 static int 4631 pcap_read_linux_mmap_v1_64(pcap_t *handle, int max_packets, pcap_handler callback, 4632 u_char *user) 4633 { 4634 struct pcap_linux *handlep = handle->priv; 4635 int pkts = 0; 4636 int ret; 4637 4638 /* wait for frames availability.*/ 4639 ret = pcap_wait_for_frames_mmap(handle); 4640 if (ret) { 4641 return ret; 4642 } 4643 4644 /* non-positive values of max_packets are used to require all 4645 * packets currently available in the ring */ 4646 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) { 4647 union thdr h; 4648 4649 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER); 4650 if (!h.raw) 4651 break; 4652 4653 ret = pcap_handle_packet_mmap( 4654 handle, 4655 callback, 4656 user, 4657 h.raw, 4658 h.h1_64->tp_len, 4659 h.h1_64->tp_mac, 4660 h.h1_64->tp_snaplen, 4661 h.h1_64->tp_sec, 4662 h.h1_64->tp_usec, 4663 0, 4664 0, 4665 0); 4666 if (ret == 1) { 4667 pkts++; 4668 handlep->packets_read++; 4669 } else if (ret < 0) { 4670 return ret; 4671 } 4672 4673 /* 4674 * Hand this block back to the kernel, and, if we're 4675 * counting blocks that need to be filtered in userland 4676 * after having been filtered by the kernel, count 4677 * the one we've just processed. 4678 */ 4679 h.h1_64->tp_status = TP_STATUS_KERNEL; 4680 if (handlep->blocks_to_filter_in_userland > 0) { 4681 handlep->blocks_to_filter_in_userland--; 4682 if (handlep->blocks_to_filter_in_userland == 0) { 4683 /* 4684 * No more blocks need to be filtered 4685 * in userland. 4686 */ 4687 handlep->filter_in_userland = 0; 4688 } 4689 } 4690 4691 /* next block */ 4692 if (++handle->offset >= handle->cc) 4693 handle->offset = 0; 4694 4695 /* check for break loop condition*/ 4696 if (handle->break_loop) { 4697 handle->break_loop = 0; 4698 return PCAP_ERROR_BREAK; 4699 } 4700 } 4701 return pkts; 4702 } 4703 4704 #ifdef HAVE_TPACKET2 4705 static int 4706 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback, 4707 u_char *user) 4708 { 4709 struct pcap_linux *handlep = handle->priv; 4710 int pkts = 0; 4711 int ret; 4712 4713 /* wait for frames availability.*/ 4714 ret = pcap_wait_for_frames_mmap(handle); 4715 if (ret) { 4716 return ret; 4717 } 4718 4719 /* non-positive values of max_packets are used to require all 4720 * packets currently available in the ring */ 4721 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) { 4722 union thdr h; 4723 4724 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER); 4725 if (!h.raw) 4726 break; 4727 4728 ret = pcap_handle_packet_mmap( 4729 handle, 4730 callback, 4731 user, 4732 h.raw, 4733 h.h2->tp_len, 4734 h.h2->tp_mac, 4735 h.h2->tp_snaplen, 4736 h.h2->tp_sec, 4737 handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000, 4738 #if defined(TP_STATUS_VLAN_VALID) 4739 (h.h2->tp_vlan_tci || (h.h2->tp_status & TP_STATUS_VLAN_VALID)), 4740 #else 4741 h.h2->tp_vlan_tci != 0, 4742 #endif 4743 h.h2->tp_vlan_tci, 4744 VLAN_TPID(h.h2, h.h2)); 4745 if (ret == 1) { 4746 pkts++; 4747 handlep->packets_read++; 4748 } else if (ret < 0) { 4749 return ret; 4750 } 4751 4752 /* 4753 * Hand this block back to the kernel, and, if we're 4754 * counting blocks that need to be filtered in userland 4755 * after having been filtered by the kernel, count 4756 * the one we've just processed. 4757 */ 4758 h.h2->tp_status = TP_STATUS_KERNEL; 4759 if (handlep->blocks_to_filter_in_userland > 0) { 4760 handlep->blocks_to_filter_in_userland--; 4761 if (handlep->blocks_to_filter_in_userland == 0) { 4762 /* 4763 * No more blocks need to be filtered 4764 * in userland. 4765 */ 4766 handlep->filter_in_userland = 0; 4767 } 4768 } 4769 4770 /* next block */ 4771 if (++handle->offset >= handle->cc) 4772 handle->offset = 0; 4773 4774 /* check for break loop condition*/ 4775 if (handle->break_loop) { 4776 handle->break_loop = 0; 4777 return PCAP_ERROR_BREAK; 4778 } 4779 } 4780 return pkts; 4781 } 4782 #endif /* HAVE_TPACKET2 */ 4783 4784 #ifdef HAVE_TPACKET3 4785 static int 4786 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback, 4787 u_char *user) 4788 { 4789 struct pcap_linux *handlep = handle->priv; 4790 union thdr h; 4791 int pkts = 0; 4792 int ret; 4793 4794 again: 4795 if (handlep->current_packet == NULL) { 4796 /* wait for frames availability.*/ 4797 ret = pcap_wait_for_frames_mmap(handle); 4798 if (ret) { 4799 return ret; 4800 } 4801 } 4802 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER); 4803 if (!h.raw) { 4804 if (pkts == 0 && handlep->timeout == 0) { 4805 /* Block until we see a packet. */ 4806 goto again; 4807 } 4808 return pkts; 4809 } 4810 4811 /* non-positive values of max_packets are used to require all 4812 * packets currently available in the ring */ 4813 while ((pkts < max_packets) || PACKET_COUNT_IS_UNLIMITED(max_packets)) { 4814 if (handlep->current_packet == NULL) { 4815 h.raw = pcap_get_ring_frame(handle, TP_STATUS_USER); 4816 if (!h.raw) 4817 break; 4818 4819 handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt; 4820 handlep->packets_left = h.h3->hdr.bh1.num_pkts; 4821 } 4822 int packets_to_read = handlep->packets_left; 4823 4824 if (!PACKET_COUNT_IS_UNLIMITED(max_packets) && packets_to_read > max_packets) { 4825 packets_to_read = max_packets; 4826 } 4827 4828 while(packets_to_read--) { 4829 struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet; 4830 ret = pcap_handle_packet_mmap( 4831 handle, 4832 callback, 4833 user, 4834 handlep->current_packet, 4835 tp3_hdr->tp_len, 4836 tp3_hdr->tp_mac, 4837 tp3_hdr->tp_snaplen, 4838 tp3_hdr->tp_sec, 4839 handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000, 4840 #if defined(TP_STATUS_VLAN_VALID) 4841 (tp3_hdr->hv1.tp_vlan_tci || (tp3_hdr->tp_status & TP_STATUS_VLAN_VALID)), 4842 #else 4843 tp3_hdr->hv1.tp_vlan_tci != 0, 4844 #endif 4845 tp3_hdr->hv1.tp_vlan_tci, 4846 VLAN_TPID(tp3_hdr, &tp3_hdr->hv1)); 4847 if (ret == 1) { 4848 pkts++; 4849 handlep->packets_read++; 4850 } else if (ret < 0) { 4851 handlep->current_packet = NULL; 4852 return ret; 4853 } 4854 handlep->current_packet += tp3_hdr->tp_next_offset; 4855 handlep->packets_left--; 4856 } 4857 4858 if (handlep->packets_left <= 0) { 4859 /* 4860 * Hand this block back to the kernel, and, if 4861 * we're counting blocks that need to be 4862 * filtered in userland after having been 4863 * filtered by the kernel, count the one we've 4864 * just processed. 4865 */ 4866 h.h3->hdr.bh1.block_status = TP_STATUS_KERNEL; 4867 if (handlep->blocks_to_filter_in_userland > 0) { 4868 handlep->blocks_to_filter_in_userland--; 4869 if (handlep->blocks_to_filter_in_userland == 0) { 4870 /* 4871 * No more blocks need to be filtered 4872 * in userland. 4873 */ 4874 handlep->filter_in_userland = 0; 4875 } 4876 } 4877 4878 /* next block */ 4879 if (++handle->offset >= handle->cc) 4880 handle->offset = 0; 4881 4882 handlep->current_packet = NULL; 4883 } 4884 4885 /* check for break loop condition*/ 4886 if (handle->break_loop) { 4887 handle->break_loop = 0; 4888 return PCAP_ERROR_BREAK; 4889 } 4890 } 4891 if (pkts == 0 && handlep->timeout == 0) { 4892 /* Block until we see a packet. */ 4893 goto again; 4894 } 4895 return pkts; 4896 } 4897 #endif /* HAVE_TPACKET3 */ 4898 4899 static int 4900 pcap_setfilter_linux_mmap(pcap_t *handle, struct bpf_program *filter) 4901 { 4902 struct pcap_linux *handlep = handle->priv; 4903 int n, offset; 4904 int ret; 4905 4906 /* 4907 * Don't rewrite "ret" instructions; we don't need to, as 4908 * we're not reading packets with recvmsg(), and we don't 4909 * want to, as, by not rewriting them, the kernel can avoid 4910 * copying extra data. 4911 */ 4912 ret = pcap_setfilter_linux_common(handle, filter, 1); 4913 if (ret < 0) 4914 return ret; 4915 4916 /* 4917 * If we're filtering in userland, there's nothing to do; 4918 * the new filter will be used for the next packet. 4919 */ 4920 if (handlep->filter_in_userland) 4921 return ret; 4922 4923 /* 4924 * We're filtering in the kernel; the packets present in 4925 * all blocks currently in the ring were already filtered 4926 * by the old filter, and so will need to be filtered in 4927 * userland by the new filter. 4928 * 4929 * Get an upper bound for the number of such blocks; first, 4930 * walk the ring backward and count the free blocks. 4931 */ 4932 offset = handle->offset; 4933 if (--handle->offset < 0) 4934 handle->offset = handle->cc - 1; 4935 for (n=0; n < handle->cc; ++n) { 4936 if (--handle->offset < 0) 4937 handle->offset = handle->cc - 1; 4938 if (!pcap_get_ring_frame(handle, TP_STATUS_KERNEL)) 4939 break; 4940 } 4941 4942 /* 4943 * If we found free blocks, decrement the count of free 4944 * blocks by 1, just in case we lost a race with another 4945 * thread of control that was adding a packet while 4946 * we were counting and that had run the filter before 4947 * we changed it. 4948 * 4949 * XXX - could there be more than one block added in 4950 * this fashion? 4951 * 4952 * XXX - is there a way to avoid that race, e.g. somehow 4953 * wait for all packets that passed the old filter to 4954 * be added to the ring? 4955 */ 4956 if (n != 0) 4957 n--; 4958 4959 /* be careful to not change current ring position */ 4960 handle->offset = offset; 4961 4962 /* 4963 * Set the count of blocks worth of packets to filter 4964 * in userland to the total number of blocks in the 4965 * ring minus the number of free blocks we found, and 4966 * turn on userland filtering. (The count of blocks 4967 * worth of packets to filter in userland is guaranteed 4968 * not to be zero - n, above, couldn't be set to a 4969 * value > handle->cc, and if it were equal to 4970 * handle->cc, it wouldn't be zero, and thus would 4971 * be decremented to handle->cc - 1.) 4972 */ 4973 handlep->blocks_to_filter_in_userland = handle->cc - n; 4974 handlep->filter_in_userland = 1; 4975 return ret; 4976 } 4977 4978 #endif /* HAVE_PACKET_RING */ 4979 4980 4981 #ifdef HAVE_PF_PACKET_SOCKETS 4982 /* 4983 * Return the index of the given device name. Fill ebuf and return 4984 * -1 on failure. 4985 */ 4986 static int 4987 iface_get_id(int fd, const char *device, char *ebuf) 4988 { 4989 struct ifreq ifr; 4990 4991 memset(&ifr, 0, sizeof(ifr)); 4992 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); 4993 4994 if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) { 4995 snprintf(ebuf, PCAP_ERRBUF_SIZE, 4996 "SIOCGIFINDEX: %s", pcap_strerror(errno)); 4997 return -1; 4998 } 4999 5000 return ifr.ifr_ifindex; 5001 } 5002 5003 /* 5004 * Bind the socket associated with FD to the given device. 5005 * Return 1 on success, 0 if we should try a SOCK_PACKET socket, 5006 * or a PCAP_ERROR_ value on a hard error. 5007 */ 5008 static int 5009 iface_bind(int fd, int ifindex, char *ebuf) 5010 { 5011 struct sockaddr_ll sll; 5012 int err; 5013 socklen_t errlen = sizeof(err); 5014 5015 memset(&sll, 0, sizeof(sll)); 5016 sll.sll_family = AF_PACKET; 5017 sll.sll_ifindex = ifindex; 5018 sll.sll_protocol = htons(ETH_P_ALL); 5019 5020 if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) { 5021 if (errno == ENETDOWN) { 5022 /* 5023 * Return a "network down" indication, so that 5024 * the application can report that rather than 5025 * saying we had a mysterious failure and 5026 * suggest that they report a problem to the 5027 * libpcap developers. 5028 */ 5029 return PCAP_ERROR_IFACE_NOT_UP; 5030 } else { 5031 snprintf(ebuf, PCAP_ERRBUF_SIZE, 5032 "bind: %s", pcap_strerror(errno)); 5033 return PCAP_ERROR; 5034 } 5035 } 5036 5037 /* Any pending errors, e.g., network is down? */ 5038 5039 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) { 5040 snprintf(ebuf, PCAP_ERRBUF_SIZE, 5041 "getsockopt: %s", pcap_strerror(errno)); 5042 return 0; 5043 } 5044 5045 if (err == ENETDOWN) { 5046 /* 5047 * Return a "network down" indication, so that 5048 * the application can report that rather than 5049 * saying we had a mysterious failure and 5050 * suggest that they report a problem to the 5051 * libpcap developers. 5052 */ 5053 return PCAP_ERROR_IFACE_NOT_UP; 5054 } else if (err > 0) { 5055 snprintf(ebuf, PCAP_ERRBUF_SIZE, 5056 "bind: %s", pcap_strerror(err)); 5057 return 0; 5058 } 5059 5060 return 1; 5061 } 5062 5063 #ifdef IW_MODE_MONITOR 5064 /* 5065 * Check whether the device supports the Wireless Extensions. 5066 * Returns 1 if it does, 0 if it doesn't, PCAP_ERROR_NO_SUCH_DEVICE 5067 * if the device doesn't even exist. 5068 */ 5069 static int 5070 has_wext(int sock_fd, const char *device, char *ebuf) 5071 { 5072 struct iwreq ireq; 5073 5074 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5075 sizeof ireq.ifr_ifrn.ifrn_name); 5076 if (ioctl(sock_fd, SIOCGIWNAME, &ireq) >= 0) 5077 return 1; /* yes */ 5078 snprintf(ebuf, PCAP_ERRBUF_SIZE, 5079 "%s: SIOCGIWPRIV: %s", device, pcap_strerror(errno)); 5080 if (errno == ENODEV) 5081 return PCAP_ERROR_NO_SUCH_DEVICE; 5082 return 0; 5083 } 5084 5085 /* 5086 * Per me si va ne la citta dolente, 5087 * Per me si va ne l'etterno dolore, 5088 * ... 5089 * Lasciate ogne speranza, voi ch'intrate. 5090 * 5091 * XXX - airmon-ng does special stuff with the Orinoco driver and the 5092 * wlan-ng driver. 5093 */ 5094 typedef enum { 5095 MONITOR_WEXT, 5096 MONITOR_HOSTAP, 5097 MONITOR_PRISM, 5098 MONITOR_PRISM54, 5099 MONITOR_ACX100, 5100 MONITOR_RT2500, 5101 MONITOR_RT2570, 5102 MONITOR_RT73, 5103 MONITOR_RTL8XXX 5104 } monitor_type; 5105 5106 /* 5107 * Use the Wireless Extensions, if we have them, to try to turn monitor mode 5108 * on if it's not already on. 5109 * 5110 * Returns 1 on success, 0 if we don't support the Wireless Extensions 5111 * on this device, or a PCAP_ERROR_ value if we do support them but 5112 * we weren't able to turn monitor mode on. 5113 */ 5114 static int 5115 enter_rfmon_mode_wext(pcap_t *handle, int sock_fd, const char *device) 5116 { 5117 /* 5118 * XXX - at least some adapters require non-Wireless Extensions 5119 * mechanisms to turn monitor mode on. 5120 * 5121 * Atheros cards might require that a separate "monitor virtual access 5122 * point" be created, with later versions of the madwifi driver. 5123 * airmon-ng does "wlanconfig ath create wlandev {if} wlanmode 5124 * monitor -bssid", which apparently spits out a line "athN" 5125 * where "athN" is the monitor mode device. To leave monitor 5126 * mode, it destroys the monitor mode device. 5127 * 5128 * Some Intel Centrino adapters might require private ioctls to get 5129 * radio headers; the ipw2200 and ipw3945 drivers allow you to 5130 * configure a separate "rtapN" interface to capture in monitor 5131 * mode without preventing the adapter from operating normally. 5132 * (airmon-ng doesn't appear to use that, though.) 5133 * 5134 * It would be Truly Wonderful if mac80211 and nl80211 cleaned this 5135 * up, and if all drivers were converted to mac80211 drivers. 5136 * 5137 * If interface {if} is a mac80211 driver, the file 5138 * /sys/class/net/{if}/phy80211 is a symlink to 5139 * /sys/class/ieee80211/{phydev}, for some {phydev}. 5140 * 5141 * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at 5142 * least, has a "wmaster0" device and a "wlan0" device; the 5143 * latter is the one with the IP address. Both show up in 5144 * "tcpdump -D" output. Capturing on the wmaster0 device 5145 * captures with 802.11 headers. 5146 * 5147 * airmon-ng searches through /sys/class/net for devices named 5148 * monN, starting with mon0; as soon as one *doesn't* exist, 5149 * it chooses that as the monitor device name. If the "iw" 5150 * command exists, it does "iw dev {if} interface add {monif} 5151 * type monitor", where {monif} is the monitor device. It 5152 * then (sigh) sleeps .1 second, and then configures the 5153 * device up. Otherwise, if /sys/class/ieee80211/{phydev}/add_iface 5154 * is a file, it writes {mondev}, without a newline, to that file, 5155 * and again (sigh) sleeps .1 second, and then iwconfig's that 5156 * device into monitor mode and configures it up. Otherwise, 5157 * you can't do monitor mode. 5158 * 5159 * All these devices are "glued" together by having the 5160 * /sys/class/net/{device}/phy80211 links pointing to the same 5161 * place, so, given a wmaster, wlan, or mon device, you can 5162 * find the other devices by looking for devices with 5163 * the same phy80211 link. 5164 * 5165 * To turn monitor mode off, delete the monitor interface, 5166 * either with "iw dev {monif} interface del" or by sending 5167 * {monif}, with no NL, down /sys/class/ieee80211/{phydev}/remove_iface 5168 * 5169 * Note: if you try to create a monitor device named "monN", and 5170 * there's already a "monN" device, it fails, as least with 5171 * the netlink interface (which is what iw uses), with a return 5172 * value of -ENFILE. (Return values are negative errnos.) We 5173 * could probably use that to find an unused device. 5174 */ 5175 struct pcap_linux *handlep = handle->priv; 5176 int err; 5177 struct iwreq ireq; 5178 struct iw_priv_args *priv; 5179 monitor_type montype; 5180 int i; 5181 __u32 cmd; 5182 struct ifreq ifr; 5183 int oldflags; 5184 int args[2]; 5185 int channel; 5186 5187 /* 5188 * Does this device *support* the Wireless Extensions? 5189 */ 5190 err = has_wext(sock_fd, device, handle->errbuf); 5191 if (err <= 0) 5192 return err; /* either it doesn't or the device doesn't even exist */ 5193 /* 5194 * Start out assuming we have no private extensions to control 5195 * radio metadata. 5196 */ 5197 montype = MONITOR_WEXT; 5198 cmd = 0; 5199 5200 /* 5201 * Try to get all the Wireless Extensions private ioctls 5202 * supported by this device. 5203 * 5204 * First, get the size of the buffer we need, by supplying no 5205 * buffer and a length of 0. If the device supports private 5206 * ioctls, it should return E2BIG, with ireq.u.data.length set 5207 * to the length we need. If it doesn't support them, it should 5208 * return EOPNOTSUPP. 5209 */ 5210 memset(&ireq, 0, sizeof ireq); 5211 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5212 sizeof ireq.ifr_ifrn.ifrn_name); 5213 ireq.u.data.pointer = (void *)args; 5214 ireq.u.data.length = 0; 5215 ireq.u.data.flags = 0; 5216 if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) != -1) { 5217 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5218 "%s: SIOCGIWPRIV with a zero-length buffer didn't fail!", 5219 device); 5220 return PCAP_ERROR; 5221 } 5222 if (errno != EOPNOTSUPP) { 5223 /* 5224 * OK, it's not as if there are no private ioctls. 5225 */ 5226 if (errno != E2BIG) { 5227 /* 5228 * Failed. 5229 */ 5230 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5231 "%s: SIOCGIWPRIV: %s", device, 5232 pcap_strerror(errno)); 5233 return PCAP_ERROR; 5234 } 5235 5236 /* 5237 * OK, try to get the list of private ioctls. 5238 */ 5239 priv = malloc(ireq.u.data.length * sizeof (struct iw_priv_args)); 5240 if (priv == NULL) { 5241 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5242 "malloc: %s", pcap_strerror(errno)); 5243 return PCAP_ERROR; 5244 } 5245 ireq.u.data.pointer = (void *)priv; 5246 if (ioctl(sock_fd, SIOCGIWPRIV, &ireq) == -1) { 5247 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5248 "%s: SIOCGIWPRIV: %s", device, 5249 pcap_strerror(errno)); 5250 free(priv); 5251 return PCAP_ERROR; 5252 } 5253 5254 /* 5255 * Look for private ioctls to turn monitor mode on or, if 5256 * monitor mode is on, to set the header type. 5257 */ 5258 for (i = 0; i < ireq.u.data.length; i++) { 5259 if (strcmp(priv[i].name, "monitor_type") == 0) { 5260 /* 5261 * Hostap driver, use this one. 5262 * Set monitor mode first. 5263 * You can set it to 0 to get DLT_IEEE80211, 5264 * 1 to get DLT_PRISM, 2 to get 5265 * DLT_IEEE80211_RADIO_AVS, and, with more 5266 * recent versions of the driver, 3 to get 5267 * DLT_IEEE80211_RADIO. 5268 */ 5269 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5270 break; 5271 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED)) 5272 break; 5273 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1) 5274 break; 5275 montype = MONITOR_HOSTAP; 5276 cmd = priv[i].cmd; 5277 break; 5278 } 5279 if (strcmp(priv[i].name, "set_prismhdr") == 0) { 5280 /* 5281 * Prism54 driver, use this one. 5282 * Set monitor mode first. 5283 * You can set it to 2 to get DLT_IEEE80211 5284 * or 3 or get DLT_PRISM. 5285 */ 5286 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5287 break; 5288 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED)) 5289 break; 5290 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1) 5291 break; 5292 montype = MONITOR_PRISM54; 5293 cmd = priv[i].cmd; 5294 break; 5295 } 5296 if (strcmp(priv[i].name, "forceprismheader") == 0) { 5297 /* 5298 * RT2570 driver, use this one. 5299 * Do this after turning monitor mode on. 5300 * You can set it to 1 to get DLT_PRISM or 2 5301 * to get DLT_IEEE80211. 5302 */ 5303 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5304 break; 5305 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED)) 5306 break; 5307 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1) 5308 break; 5309 montype = MONITOR_RT2570; 5310 cmd = priv[i].cmd; 5311 break; 5312 } 5313 if (strcmp(priv[i].name, "forceprism") == 0) { 5314 /* 5315 * RT73 driver, use this one. 5316 * Do this after turning monitor mode on. 5317 * Its argument is a *string*; you can 5318 * set it to "1" to get DLT_PRISM or "2" 5319 * to get DLT_IEEE80211. 5320 */ 5321 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_CHAR) 5322 break; 5323 if (priv[i].set_args & IW_PRIV_SIZE_FIXED) 5324 break; 5325 montype = MONITOR_RT73; 5326 cmd = priv[i].cmd; 5327 break; 5328 } 5329 if (strcmp(priv[i].name, "prismhdr") == 0) { 5330 /* 5331 * One of the RTL8xxx drivers, use this one. 5332 * It can only be done after monitor mode 5333 * has been turned on. You can set it to 1 5334 * to get DLT_PRISM or 0 to get DLT_IEEE80211. 5335 */ 5336 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5337 break; 5338 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED)) 5339 break; 5340 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 1) 5341 break; 5342 montype = MONITOR_RTL8XXX; 5343 cmd = priv[i].cmd; 5344 break; 5345 } 5346 if (strcmp(priv[i].name, "rfmontx") == 0) { 5347 /* 5348 * RT2500 or RT61 driver, use this one. 5349 * It has one one-byte parameter; set 5350 * u.data.length to 1 and u.data.pointer to 5351 * point to the parameter. 5352 * It doesn't itself turn monitor mode on. 5353 * You can set it to 1 to allow transmitting 5354 * in monitor mode(?) and get DLT_IEEE80211, 5355 * or set it to 0 to disallow transmitting in 5356 * monitor mode(?) and get DLT_PRISM. 5357 */ 5358 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5359 break; 5360 if ((priv[i].set_args & IW_PRIV_SIZE_MASK) != 2) 5361 break; 5362 montype = MONITOR_RT2500; 5363 cmd = priv[i].cmd; 5364 break; 5365 } 5366 if (strcmp(priv[i].name, "monitor") == 0) { 5367 /* 5368 * Either ACX100 or hostap, use this one. 5369 * It turns monitor mode on. 5370 * If it takes two arguments, it's ACX100; 5371 * the first argument is 1 for DLT_PRISM 5372 * or 2 for DLT_IEEE80211, and the second 5373 * argument is the channel on which to 5374 * run. If it takes one argument, it's 5375 * HostAP, and the argument is 2 for 5376 * DLT_IEEE80211 and 3 for DLT_PRISM. 5377 * 5378 * If we see this, we don't quit, as this 5379 * might be a version of the hostap driver 5380 * that also supports "monitor_type". 5381 */ 5382 if ((priv[i].set_args & IW_PRIV_TYPE_MASK) != IW_PRIV_TYPE_INT) 5383 break; 5384 if (!(priv[i].set_args & IW_PRIV_SIZE_FIXED)) 5385 break; 5386 switch (priv[i].set_args & IW_PRIV_SIZE_MASK) { 5387 5388 case 1: 5389 montype = MONITOR_PRISM; 5390 cmd = priv[i].cmd; 5391 break; 5392 5393 case 2: 5394 montype = MONITOR_ACX100; 5395 cmd = priv[i].cmd; 5396 break; 5397 5398 default: 5399 break; 5400 } 5401 } 5402 } 5403 free(priv); 5404 } 5405 5406 /* 5407 * XXX - ipw3945? islism? 5408 */ 5409 5410 /* 5411 * Get the old mode. 5412 */ 5413 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5414 sizeof ireq.ifr_ifrn.ifrn_name); 5415 if (ioctl(sock_fd, SIOCGIWMODE, &ireq) == -1) { 5416 /* 5417 * We probably won't be able to set the mode, either. 5418 */ 5419 return PCAP_ERROR_RFMON_NOTSUP; 5420 } 5421 5422 /* 5423 * Is it currently in monitor mode? 5424 */ 5425 if (ireq.u.mode == IW_MODE_MONITOR) { 5426 /* 5427 * Yes. Just leave things as they are. 5428 * We don't offer multiple link-layer types, as 5429 * changing the link-layer type out from under 5430 * somebody else capturing in monitor mode would 5431 * be considered rude. 5432 */ 5433 return 1; 5434 } 5435 /* 5436 * No. We have to put the adapter into rfmon mode. 5437 */ 5438 5439 /* 5440 * If we haven't already done so, arrange to have 5441 * "pcap_close_all()" called when we exit. 5442 */ 5443 if (!pcap_do_addexit(handle)) { 5444 /* 5445 * "atexit()" failed; don't put the interface 5446 * in rfmon mode, just give up. 5447 */ 5448 return PCAP_ERROR_RFMON_NOTSUP; 5449 } 5450 5451 /* 5452 * Save the old mode. 5453 */ 5454 handlep->oldmode = ireq.u.mode; 5455 5456 /* 5457 * Put the adapter in rfmon mode. How we do this depends 5458 * on whether we have a special private ioctl or not. 5459 */ 5460 if (montype == MONITOR_PRISM) { 5461 /* 5462 * We have the "monitor" private ioctl, but none of 5463 * the other private ioctls. Use this, and select 5464 * the Prism header. 5465 * 5466 * If it fails, just fall back on SIOCSIWMODE. 5467 */ 5468 memset(&ireq, 0, sizeof ireq); 5469 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5470 sizeof ireq.ifr_ifrn.ifrn_name); 5471 ireq.u.data.length = 1; /* 1 argument */ 5472 args[0] = 3; /* request Prism header */ 5473 memcpy(ireq.u.name, args, sizeof (int)); 5474 if (ioctl(sock_fd, cmd, &ireq) != -1) { 5475 /* 5476 * Success. 5477 * Note that we have to put the old mode back 5478 * when we close the device. 5479 */ 5480 handlep->must_do_on_close |= MUST_CLEAR_RFMON; 5481 5482 /* 5483 * Add this to the list of pcaps to close 5484 * when we exit. 5485 */ 5486 pcap_add_to_pcaps_to_close(handle); 5487 5488 return 1; 5489 } 5490 5491 /* 5492 * Failure. Fall back on SIOCSIWMODE. 5493 */ 5494 } 5495 5496 /* 5497 * First, take the interface down if it's up; otherwise, we 5498 * might get EBUSY. 5499 */ 5500 memset(&ifr, 0, sizeof(ifr)); 5501 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); 5502 if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) { 5503 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5504 "%s: Can't get flags: %s", device, strerror(errno)); 5505 return PCAP_ERROR; 5506 } 5507 oldflags = 0; 5508 if (ifr.ifr_flags & IFF_UP) { 5509 oldflags = ifr.ifr_flags; 5510 ifr.ifr_flags &= ~IFF_UP; 5511 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) { 5512 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5513 "%s: Can't set flags: %s", device, strerror(errno)); 5514 return PCAP_ERROR; 5515 } 5516 } 5517 5518 /* 5519 * Then turn monitor mode on. 5520 */ 5521 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5522 sizeof ireq.ifr_ifrn.ifrn_name); 5523 ireq.u.mode = IW_MODE_MONITOR; 5524 if (ioctl(sock_fd, SIOCSIWMODE, &ireq) == -1) { 5525 /* 5526 * Scientist, you've failed. 5527 * Bring the interface back up if we shut it down. 5528 */ 5529 ifr.ifr_flags = oldflags; 5530 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) { 5531 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5532 "%s: Can't set flags: %s", device, strerror(errno)); 5533 return PCAP_ERROR; 5534 } 5535 return PCAP_ERROR_RFMON_NOTSUP; 5536 } 5537 5538 /* 5539 * XXX - airmon-ng does "iwconfig {if} key off" after setting 5540 * monitor mode and setting the channel, and then does 5541 * "iwconfig up". 5542 */ 5543 5544 /* 5545 * Now select the appropriate radio header. 5546 */ 5547 switch (montype) { 5548 5549 case MONITOR_WEXT: 5550 /* 5551 * We don't have any private ioctl to set the header. 5552 */ 5553 break; 5554 5555 case MONITOR_HOSTAP: 5556 /* 5557 * Try to select the radiotap header. 5558 */ 5559 memset(&ireq, 0, sizeof ireq); 5560 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5561 sizeof ireq.ifr_ifrn.ifrn_name); 5562 args[0] = 3; /* request radiotap header */ 5563 memcpy(ireq.u.name, args, sizeof (int)); 5564 if (ioctl(sock_fd, cmd, &ireq) != -1) 5565 break; /* success */ 5566 5567 /* 5568 * That failed. Try to select the AVS header. 5569 */ 5570 memset(&ireq, 0, sizeof ireq); 5571 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5572 sizeof ireq.ifr_ifrn.ifrn_name); 5573 args[0] = 2; /* request AVS header */ 5574 memcpy(ireq.u.name, args, sizeof (int)); 5575 if (ioctl(sock_fd, cmd, &ireq) != -1) 5576 break; /* success */ 5577 5578 /* 5579 * That failed. Try to select the Prism header. 5580 */ 5581 memset(&ireq, 0, sizeof ireq); 5582 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5583 sizeof ireq.ifr_ifrn.ifrn_name); 5584 args[0] = 1; /* request Prism header */ 5585 memcpy(ireq.u.name, args, sizeof (int)); 5586 ioctl(sock_fd, cmd, &ireq); 5587 break; 5588 5589 case MONITOR_PRISM: 5590 /* 5591 * The private ioctl failed. 5592 */ 5593 break; 5594 5595 case MONITOR_PRISM54: 5596 /* 5597 * Select the Prism header. 5598 */ 5599 memset(&ireq, 0, sizeof ireq); 5600 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5601 sizeof ireq.ifr_ifrn.ifrn_name); 5602 args[0] = 3; /* request Prism header */ 5603 memcpy(ireq.u.name, args, sizeof (int)); 5604 ioctl(sock_fd, cmd, &ireq); 5605 break; 5606 5607 case MONITOR_ACX100: 5608 /* 5609 * Get the current channel. 5610 */ 5611 memset(&ireq, 0, sizeof ireq); 5612 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5613 sizeof ireq.ifr_ifrn.ifrn_name); 5614 if (ioctl(sock_fd, SIOCGIWFREQ, &ireq) == -1) { 5615 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5616 "%s: SIOCGIWFREQ: %s", device, 5617 pcap_strerror(errno)); 5618 return PCAP_ERROR; 5619 } 5620 channel = ireq.u.freq.m; 5621 5622 /* 5623 * Select the Prism header, and set the channel to the 5624 * current value. 5625 */ 5626 memset(&ireq, 0, sizeof ireq); 5627 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5628 sizeof ireq.ifr_ifrn.ifrn_name); 5629 args[0] = 1; /* request Prism header */ 5630 args[1] = channel; /* set channel */ 5631 memcpy(ireq.u.name, args, 2*sizeof (int)); 5632 ioctl(sock_fd, cmd, &ireq); 5633 break; 5634 5635 case MONITOR_RT2500: 5636 /* 5637 * Disallow transmission - that turns on the 5638 * Prism header. 5639 */ 5640 memset(&ireq, 0, sizeof ireq); 5641 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5642 sizeof ireq.ifr_ifrn.ifrn_name); 5643 args[0] = 0; /* disallow transmitting */ 5644 memcpy(ireq.u.name, args, sizeof (int)); 5645 ioctl(sock_fd, cmd, &ireq); 5646 break; 5647 5648 case MONITOR_RT2570: 5649 /* 5650 * Force the Prism header. 5651 */ 5652 memset(&ireq, 0, sizeof ireq); 5653 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5654 sizeof ireq.ifr_ifrn.ifrn_name); 5655 args[0] = 1; /* request Prism header */ 5656 memcpy(ireq.u.name, args, sizeof (int)); 5657 ioctl(sock_fd, cmd, &ireq); 5658 break; 5659 5660 case MONITOR_RT73: 5661 /* 5662 * Force the Prism header. 5663 */ 5664 memset(&ireq, 0, sizeof ireq); 5665 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5666 sizeof ireq.ifr_ifrn.ifrn_name); 5667 ireq.u.data.length = 1; /* 1 argument */ 5668 ireq.u.data.pointer = "1"; 5669 ireq.u.data.flags = 0; 5670 ioctl(sock_fd, cmd, &ireq); 5671 break; 5672 5673 case MONITOR_RTL8XXX: 5674 /* 5675 * Force the Prism header. 5676 */ 5677 memset(&ireq, 0, sizeof ireq); 5678 strlcpy(ireq.ifr_ifrn.ifrn_name, device, 5679 sizeof ireq.ifr_ifrn.ifrn_name); 5680 args[0] = 1; /* request Prism header */ 5681 memcpy(ireq.u.name, args, sizeof (int)); 5682 ioctl(sock_fd, cmd, &ireq); 5683 break; 5684 } 5685 5686 /* 5687 * Now bring the interface back up if we brought it down. 5688 */ 5689 if (oldflags != 0) { 5690 ifr.ifr_flags = oldflags; 5691 if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) { 5692 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5693 "%s: Can't set flags: %s", device, strerror(errno)); 5694 5695 /* 5696 * At least try to restore the old mode on the 5697 * interface. 5698 */ 5699 if (ioctl(handle->fd, SIOCSIWMODE, &ireq) == -1) { 5700 /* 5701 * Scientist, you've failed. 5702 */ 5703 fprintf(stderr, 5704 "Can't restore interface wireless mode (SIOCSIWMODE failed: %s).\n" 5705 "Please adjust manually.\n", 5706 strerror(errno)); 5707 } 5708 return PCAP_ERROR; 5709 } 5710 } 5711 5712 /* 5713 * Note that we have to put the old mode back when we 5714 * close the device. 5715 */ 5716 handlep->must_do_on_close |= MUST_CLEAR_RFMON; 5717 5718 /* 5719 * Add this to the list of pcaps to close when we exit. 5720 */ 5721 pcap_add_to_pcaps_to_close(handle); 5722 5723 return 1; 5724 } 5725 #endif /* IW_MODE_MONITOR */ 5726 5727 /* 5728 * Try various mechanisms to enter monitor mode. 5729 */ 5730 static int 5731 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device) 5732 { 5733 #if defined(HAVE_LIBNL) || defined(IW_MODE_MONITOR) 5734 int ret; 5735 #endif 5736 5737 #ifdef HAVE_LIBNL 5738 ret = enter_rfmon_mode_mac80211(handle, sock_fd, device); 5739 if (ret < 0) 5740 return ret; /* error attempting to do so */ 5741 if (ret == 1) 5742 return 1; /* success */ 5743 #endif /* HAVE_LIBNL */ 5744 5745 #ifdef IW_MODE_MONITOR 5746 ret = enter_rfmon_mode_wext(handle, sock_fd, device); 5747 if (ret < 0) 5748 return ret; /* error attempting to do so */ 5749 if (ret == 1) 5750 return 1; /* success */ 5751 #endif /* IW_MODE_MONITOR */ 5752 5753 /* 5754 * Either none of the mechanisms we know about work or none 5755 * of those mechanisms are available, so we can't do monitor 5756 * mode. 5757 */ 5758 return 0; 5759 } 5760 5761 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) 5762 /* 5763 * Map SOF_TIMESTAMPING_ values to PCAP_TSTAMP_ values. 5764 */ 5765 static const struct { 5766 int soft_timestamping_val; 5767 int pcap_tstamp_val; 5768 } sof_ts_type_map[3] = { 5769 { SOF_TIMESTAMPING_SOFTWARE, PCAP_TSTAMP_HOST }, 5770 { SOF_TIMESTAMPING_SYS_HARDWARE, PCAP_TSTAMP_ADAPTER }, 5771 { SOF_TIMESTAMPING_RAW_HARDWARE, PCAP_TSTAMP_ADAPTER_UNSYNCED } 5772 }; 5773 #define NUM_SOF_TIMESTAMPING_TYPES (sizeof sof_ts_type_map / sizeof sof_ts_type_map[0]) 5774 5775 static void 5776 iface_set_default_ts_types(pcap_t *handle) 5777 { 5778 int i; 5779 5780 handle->tstamp_type_count = NUM_SOF_TIMESTAMPING_TYPES; 5781 handle->tstamp_type_list = malloc(NUM_SOF_TIMESTAMPING_TYPES * sizeof(u_int)); 5782 for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) 5783 handle->tstamp_type_list[i] = sof_ts_type_map[i].pcap_tstamp_val; 5784 } 5785 5786 #ifdef ETHTOOL_GET_TS_INFO 5787 /* 5788 * Get a list of time stamping capabilities. 5789 */ 5790 static int 5791 iface_ethtool_get_ts_info(pcap_t *handle, char *ebuf) 5792 { 5793 int fd; 5794 struct ifreq ifr; 5795 struct ethtool_ts_info info; 5796 int num_ts_types; 5797 int i, j; 5798 5799 /* 5800 * This doesn't apply to the "any" device; you have to ask 5801 * specific devices for their capabilities, so just default 5802 * to saying we support all of them. 5803 */ 5804 if (strcmp(handle->opt.source, "any") == 0) { 5805 iface_set_default_ts_types(handle); 5806 return 0; 5807 } 5808 5809 /* 5810 * Create a socket from which to fetch time stamping capabilities. 5811 */ 5812 fd = socket(AF_INET, SOCK_DGRAM, 0); 5813 if (fd < 0) { 5814 (void)snprintf(ebuf, PCAP_ERRBUF_SIZE, 5815 "socket for SIOCETHTOOL(ETHTOOL_GET_TS_INFO): %s", pcap_strerror(errno)); 5816 return -1; 5817 } 5818 5819 memset(&ifr, 0, sizeof(ifr)); 5820 strlcpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name)); 5821 memset(&info, 0, sizeof(info)); 5822 info.cmd = ETHTOOL_GET_TS_INFO; 5823 ifr.ifr_data = (caddr_t)&info; 5824 if (ioctl(fd, SIOCETHTOOL, &ifr) == -1) { 5825 close(fd); 5826 if (errno == EOPNOTSUPP || errno == EINVAL) { 5827 /* 5828 * OK, let's just return all the possible time 5829 * stamping types. 5830 */ 5831 iface_set_default_ts_types(handle); 5832 return 0; 5833 } 5834 snprintf(ebuf, PCAP_ERRBUF_SIZE, 5835 "%s: SIOCETHTOOL(ETHTOOL_GET_TS_INFO) ioctl failed: %s", handle->opt.source, 5836 strerror(errno)); 5837 return -1; 5838 } 5839 close(fd); 5840 5841 num_ts_types = 0; 5842 for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) { 5843 if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) 5844 num_ts_types++; 5845 } 5846 handle->tstamp_type_count = num_ts_types; 5847 if (num_ts_types != 0) { 5848 handle->tstamp_type_list = malloc(num_ts_types * sizeof(u_int)); 5849 for (i = 0, j = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) { 5850 if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) { 5851 handle->tstamp_type_list[j] = sof_ts_type_map[i].pcap_tstamp_val; 5852 j++; 5853 } 5854 } 5855 } else 5856 handle->tstamp_type_list = NULL; 5857 5858 return 0; 5859 } 5860 #else /* ETHTOOL_GET_TS_INFO */ 5861 static int 5862 iface_ethtool_get_ts_info(pcap_t *handle, char *ebuf _U_) 5863 { 5864 /* 5865 * We don't have an ioctl to use to ask what's supported, 5866 * so say we support everything. 5867 */ 5868 iface_set_default_ts_types(handle); 5869 return 0; 5870 } 5871 #endif /* ETHTOOL_GET_TS_INFO */ 5872 5873 #endif /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */ 5874 5875 #ifdef HAVE_PACKET_RING 5876 /* 5877 * Find out if we have any form of fragmentation/reassembly offloading. 5878 * 5879 * We do so using SIOCETHTOOL checking for various types of offloading; 5880 * if SIOCETHTOOL isn't defined, or we don't have any #defines for any 5881 * of the types of offloading, there's nothing we can do to check, so 5882 * we just say "no, we don't". 5883 */ 5884 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO)) 5885 static int 5886 iface_ethtool_flag_ioctl(pcap_t *handle, int cmd, const char *cmdname) 5887 { 5888 struct ifreq ifr; 5889 struct ethtool_value eval; 5890 5891 memset(&ifr, 0, sizeof(ifr)); 5892 strlcpy(ifr.ifr_name, handle->opt.source, sizeof(ifr.ifr_name)); 5893 eval.cmd = cmd; 5894 eval.data = 0; 5895 ifr.ifr_data = (caddr_t)&eval; 5896 if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) { 5897 if (errno == EOPNOTSUPP || errno == EINVAL) { 5898 /* 5899 * OK, let's just return 0, which, in our 5900 * case, either means "no, what we're asking 5901 * about is not enabled" or "all the flags 5902 * are clear (i.e., nothing is enabled)". 5903 */ 5904 return 0; 5905 } 5906 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 5907 "%s: SIOCETHTOOL(%s) ioctl failed: %s", handle->opt.source, 5908 cmdname, strerror(errno)); 5909 return -1; 5910 } 5911 return eval.data; 5912 } 5913 5914 static int 5915 iface_get_offload(pcap_t *handle) 5916 { 5917 int ret; 5918 5919 #ifdef ETHTOOL_GTSO 5920 ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO"); 5921 if (ret == -1) 5922 return -1; 5923 if (ret) 5924 return 1; /* TCP segmentation offloading on */ 5925 #endif 5926 5927 #ifdef ETHTOOL_GUFO 5928 ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO"); 5929 if (ret == -1) 5930 return -1; 5931 if (ret) 5932 return 1; /* UDP fragmentation offloading on */ 5933 #endif 5934 5935 #ifdef ETHTOOL_GGSO 5936 /* 5937 * XXX - will this cause large unsegmented packets to be 5938 * handed to PF_PACKET sockets on transmission? If not, 5939 * this need not be checked. 5940 */ 5941 ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO"); 5942 if (ret == -1) 5943 return -1; 5944 if (ret) 5945 return 1; /* generic segmentation offloading on */ 5946 #endif 5947 5948 #ifdef ETHTOOL_GFLAGS 5949 ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS"); 5950 if (ret == -1) 5951 return -1; 5952 if (ret & ETH_FLAG_LRO) 5953 return 1; /* large receive offloading on */ 5954 #endif 5955 5956 #ifdef ETHTOOL_GGRO 5957 /* 5958 * XXX - will this cause large reassembled packets to be 5959 * handed to PF_PACKET sockets on receipt? If not, 5960 * this need not be checked. 5961 */ 5962 ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO"); 5963 if (ret == -1) 5964 return -1; 5965 if (ret) 5966 return 1; /* generic (large) receive offloading on */ 5967 #endif 5968 5969 return 0; 5970 } 5971 #else /* SIOCETHTOOL */ 5972 static int 5973 iface_get_offload(pcap_t *handle _U_) 5974 { 5975 /* 5976 * XXX - do we need to get this information if we don't 5977 * have the ethtool ioctls? If so, how do we do that? 5978 */ 5979 return 0; 5980 } 5981 #endif /* SIOCETHTOOL */ 5982 5983 #endif /* HAVE_PACKET_RING */ 5984 5985 #endif /* HAVE_PF_PACKET_SOCKETS */ 5986 5987 /* ===== Functions to interface to the older kernels ================== */ 5988 5989 /* 5990 * Try to open a packet socket using the old kernel interface. 5991 * Returns 1 on success and a PCAP_ERROR_ value on an error. 5992 */ 5993 static int 5994 activate_old(pcap_t *handle) 5995 { 5996 struct pcap_linux *handlep = handle->priv; 5997 int arptype; 5998 struct ifreq ifr; 5999 const char *device = handle->opt.source; 6000 struct utsname utsname; 6001 int mtu; 6002 6003 /* Open the socket */ 6004 6005 handle->fd = socket(PF_INET, SOCK_PACKET, htons(ETH_P_ALL)); 6006 if (handle->fd == -1) { 6007 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6008 "socket: %s", pcap_strerror(errno)); 6009 if (errno == EPERM || errno == EACCES) { 6010 /* 6011 * You don't have permission to open the 6012 * socket. 6013 */ 6014 return PCAP_ERROR_PERM_DENIED; 6015 } else { 6016 /* 6017 * Other error. 6018 */ 6019 return PCAP_ERROR; 6020 } 6021 } 6022 6023 /* It worked - we are using the old interface */ 6024 handlep->sock_packet = 1; 6025 6026 /* ...which means we get the link-layer header. */ 6027 handlep->cooked = 0; 6028 6029 /* Bind to the given device */ 6030 6031 if (strcmp(device, "any") == 0) { 6032 strlcpy(handle->errbuf, "pcap_activate: The \"any\" device isn't supported on 2.0[.x]-kernel systems", 6033 PCAP_ERRBUF_SIZE); 6034 return PCAP_ERROR; 6035 } 6036 if (iface_bind_old(handle->fd, device, handle->errbuf) == -1) 6037 return PCAP_ERROR; 6038 6039 /* 6040 * Try to get the link-layer type. 6041 */ 6042 arptype = iface_get_arptype(handle->fd, device, handle->errbuf); 6043 if (arptype < 0) 6044 return PCAP_ERROR; 6045 6046 /* 6047 * Try to find the DLT_ type corresponding to that 6048 * link-layer type. 6049 */ 6050 map_arphrd_to_dlt(handle, handle->fd, arptype, device, 0); 6051 if (handle->linktype == -1) { 6052 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6053 "unknown arptype %d", arptype); 6054 return PCAP_ERROR; 6055 } 6056 6057 /* Go to promisc mode if requested */ 6058 6059 if (handle->opt.promisc) { 6060 memset(&ifr, 0, sizeof(ifr)); 6061 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); 6062 if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) { 6063 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6064 "SIOCGIFFLAGS: %s", pcap_strerror(errno)); 6065 return PCAP_ERROR; 6066 } 6067 if ((ifr.ifr_flags & IFF_PROMISC) == 0) { 6068 /* 6069 * Promiscuous mode isn't currently on, 6070 * so turn it on, and remember that 6071 * we should turn it off when the 6072 * pcap_t is closed. 6073 */ 6074 6075 /* 6076 * If we haven't already done so, arrange 6077 * to have "pcap_close_all()" called when 6078 * we exit. 6079 */ 6080 if (!pcap_do_addexit(handle)) { 6081 /* 6082 * "atexit()" failed; don't put 6083 * the interface in promiscuous 6084 * mode, just give up. 6085 */ 6086 return PCAP_ERROR; 6087 } 6088 6089 ifr.ifr_flags |= IFF_PROMISC; 6090 if (ioctl(handle->fd, SIOCSIFFLAGS, &ifr) == -1) { 6091 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6092 "SIOCSIFFLAGS: %s", 6093 pcap_strerror(errno)); 6094 return PCAP_ERROR; 6095 } 6096 handlep->must_do_on_close |= MUST_CLEAR_PROMISC; 6097 6098 /* 6099 * Add this to the list of pcaps 6100 * to close when we exit. 6101 */ 6102 pcap_add_to_pcaps_to_close(handle); 6103 } 6104 } 6105 6106 /* 6107 * Compute the buffer size. 6108 * 6109 * We're using SOCK_PACKET, so this might be a 2.0[.x] 6110 * kernel, and might require special handling - check. 6111 */ 6112 if (uname(&utsname) < 0 || 6113 strncmp(utsname.release, "2.0", 3) == 0) { 6114 /* 6115 * Either we couldn't find out what kernel release 6116 * this is, or it's a 2.0[.x] kernel. 6117 * 6118 * In the 2.0[.x] kernel, a "recvfrom()" on 6119 * a SOCK_PACKET socket, with MSG_TRUNC set, will 6120 * return the number of bytes read, so if we pass 6121 * a length based on the snapshot length, it'll 6122 * return the number of bytes from the packet 6123 * copied to userland, not the actual length 6124 * of the packet. 6125 * 6126 * This means that, for example, the IP dissector 6127 * in tcpdump will get handed a packet length less 6128 * than the length in the IP header, and will 6129 * complain about "truncated-ip". 6130 * 6131 * So we don't bother trying to copy from the 6132 * kernel only the bytes in which we're interested, 6133 * but instead copy them all, just as the older 6134 * versions of libpcap for Linux did. 6135 * 6136 * The buffer therefore needs to be big enough to 6137 * hold the largest packet we can get from this 6138 * device. Unfortunately, we can't get the MRU 6139 * of the network; we can only get the MTU. The 6140 * MTU may be too small, in which case a packet larger 6141 * than the buffer size will be truncated *and* we 6142 * won't get the actual packet size. 6143 * 6144 * However, if the snapshot length is larger than 6145 * the buffer size based on the MTU, we use the 6146 * snapshot length as the buffer size, instead; 6147 * this means that with a sufficiently large snapshot 6148 * length we won't artificially truncate packets 6149 * to the MTU-based size. 6150 * 6151 * This mess just one of many problems with packet 6152 * capture on 2.0[.x] kernels; you really want a 6153 * 2.2[.x] or later kernel if you want packet capture 6154 * to work well. 6155 */ 6156 mtu = iface_get_mtu(handle->fd, device, handle->errbuf); 6157 if (mtu == -1) 6158 return PCAP_ERROR; 6159 handle->bufsize = MAX_LINKHEADER_SIZE + mtu; 6160 if (handle->bufsize < handle->snapshot) 6161 handle->bufsize = handle->snapshot; 6162 } else { 6163 /* 6164 * This is a 2.2[.x] or later kernel. 6165 * 6166 * We can safely pass "recvfrom()" a byte count 6167 * based on the snapshot length. 6168 */ 6169 handle->bufsize = handle->snapshot; 6170 } 6171 6172 /* 6173 * Default value for offset to align link-layer payload 6174 * on a 4-byte boundary. 6175 */ 6176 handle->offset = 0; 6177 6178 /* 6179 * SOCK_PACKET sockets don't supply information from 6180 * stripped VLAN tags. 6181 */ 6182 handlep->vlan_offset = -1; /* unknown */ 6183 6184 return 1; 6185 } 6186 6187 /* 6188 * Bind the socket associated with FD to the given device using the 6189 * interface of the old kernels. 6190 */ 6191 static int 6192 iface_bind_old(int fd, const char *device, char *ebuf) 6193 { 6194 struct sockaddr saddr; 6195 int err; 6196 socklen_t errlen = sizeof(err); 6197 6198 memset(&saddr, 0, sizeof(saddr)); 6199 strlcpy(saddr.sa_data, device, sizeof(saddr.sa_data)); 6200 if (bind(fd, &saddr, sizeof(saddr)) == -1) { 6201 snprintf(ebuf, PCAP_ERRBUF_SIZE, 6202 "bind: %s", pcap_strerror(errno)); 6203 return -1; 6204 } 6205 6206 /* Any pending errors, e.g., network is down? */ 6207 6208 if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) { 6209 snprintf(ebuf, PCAP_ERRBUF_SIZE, 6210 "getsockopt: %s", pcap_strerror(errno)); 6211 return -1; 6212 } 6213 6214 if (err > 0) { 6215 snprintf(ebuf, PCAP_ERRBUF_SIZE, 6216 "bind: %s", pcap_strerror(err)); 6217 return -1; 6218 } 6219 6220 return 0; 6221 } 6222 6223 6224 /* ===== System calls available on all supported kernels ============== */ 6225 6226 /* 6227 * Query the kernel for the MTU of the given interface. 6228 */ 6229 static int 6230 iface_get_mtu(int fd, const char *device, char *ebuf) 6231 { 6232 struct ifreq ifr; 6233 6234 if (!device) 6235 return BIGGER_THAN_ALL_MTUS; 6236 6237 memset(&ifr, 0, sizeof(ifr)); 6238 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); 6239 6240 if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) { 6241 snprintf(ebuf, PCAP_ERRBUF_SIZE, 6242 "SIOCGIFMTU: %s", pcap_strerror(errno)); 6243 return -1; 6244 } 6245 6246 return ifr.ifr_mtu; 6247 } 6248 6249 /* 6250 * Get the hardware type of the given interface as ARPHRD_xxx constant. 6251 */ 6252 static int 6253 iface_get_arptype(int fd, const char *device, char *ebuf) 6254 { 6255 struct ifreq ifr; 6256 6257 memset(&ifr, 0, sizeof(ifr)); 6258 strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name)); 6259 6260 if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) { 6261 snprintf(ebuf, PCAP_ERRBUF_SIZE, 6262 "SIOCGIFHWADDR: %s", pcap_strerror(errno)); 6263 if (errno == ENODEV) { 6264 /* 6265 * No such device. 6266 */ 6267 return PCAP_ERROR_NO_SUCH_DEVICE; 6268 } 6269 return PCAP_ERROR; 6270 } 6271 6272 return ifr.ifr_hwaddr.sa_family; 6273 } 6274 6275 #ifdef SO_ATTACH_FILTER 6276 static int 6277 fix_program(pcap_t *handle, struct sock_fprog *fcode, int is_mmapped) 6278 { 6279 struct pcap_linux *handlep = handle->priv; 6280 size_t prog_size; 6281 register int i; 6282 register struct bpf_insn *p; 6283 struct bpf_insn *f; 6284 int len; 6285 6286 /* 6287 * Make a copy of the filter, and modify that copy if 6288 * necessary. 6289 */ 6290 prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len; 6291 len = handle->fcode.bf_len; 6292 f = (struct bpf_insn *)malloc(prog_size); 6293 if (f == NULL) { 6294 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6295 "malloc: %s", pcap_strerror(errno)); 6296 return -1; 6297 } 6298 memcpy(f, handle->fcode.bf_insns, prog_size); 6299 fcode->len = len; 6300 fcode->filter = (struct sock_filter *) f; 6301 6302 for (i = 0; i < len; ++i) { 6303 p = &f[i]; 6304 /* 6305 * What type of instruction is this? 6306 */ 6307 switch (BPF_CLASS(p->code)) { 6308 6309 case BPF_RET: 6310 /* 6311 * It's a return instruction; are we capturing 6312 * in memory-mapped mode? 6313 */ 6314 if (!is_mmapped) { 6315 /* 6316 * No; is the snapshot length a constant, 6317 * rather than the contents of the 6318 * accumulator? 6319 */ 6320 if (BPF_MODE(p->code) == BPF_K) { 6321 /* 6322 * Yes - if the value to be returned, 6323 * i.e. the snapshot length, is 6324 * anything other than 0, make it 6325 * MAXIMUM_SNAPLEN, so that the packet 6326 * is truncated by "recvfrom()", 6327 * not by the filter. 6328 * 6329 * XXX - there's nothing we can 6330 * easily do if it's getting the 6331 * value from the accumulator; we'd 6332 * have to insert code to force 6333 * non-zero values to be 6334 * MAXIMUM_SNAPLEN. 6335 */ 6336 if (p->k != 0) 6337 p->k = MAXIMUM_SNAPLEN; 6338 } 6339 } 6340 break; 6341 6342 case BPF_LD: 6343 case BPF_LDX: 6344 /* 6345 * It's a load instruction; is it loading 6346 * from the packet? 6347 */ 6348 switch (BPF_MODE(p->code)) { 6349 6350 case BPF_ABS: 6351 case BPF_IND: 6352 case BPF_MSH: 6353 /* 6354 * Yes; are we in cooked mode? 6355 */ 6356 if (handlep->cooked) { 6357 /* 6358 * Yes, so we need to fix this 6359 * instruction. 6360 */ 6361 if (fix_offset(p) < 0) { 6362 /* 6363 * We failed to do so. 6364 * Return 0, so our caller 6365 * knows to punt to userland. 6366 */ 6367 return 0; 6368 } 6369 } 6370 break; 6371 } 6372 break; 6373 } 6374 } 6375 return 1; /* we succeeded */ 6376 } 6377 6378 static int 6379 fix_offset(struct bpf_insn *p) 6380 { 6381 /* 6382 * What's the offset? 6383 */ 6384 if (p->k >= SLL_HDR_LEN) { 6385 /* 6386 * It's within the link-layer payload; that starts at an 6387 * offset of 0, as far as the kernel packet filter is 6388 * concerned, so subtract the length of the link-layer 6389 * header. 6390 */ 6391 p->k -= SLL_HDR_LEN; 6392 } else if (p->k == 0) { 6393 /* 6394 * It's the packet type field; map it to the special magic 6395 * kernel offset for that field. 6396 */ 6397 p->k = SKF_AD_OFF + SKF_AD_PKTTYPE; 6398 } else if (p->k == 14) { 6399 /* 6400 * It's the protocol field; map it to the special magic 6401 * kernel offset for that field. 6402 */ 6403 p->k = SKF_AD_OFF + SKF_AD_PROTOCOL; 6404 } else if ((bpf_int32)(p->k) > 0) { 6405 /* 6406 * It's within the header, but it's not one of those 6407 * fields; we can't do that in the kernel, so punt 6408 * to userland. 6409 */ 6410 return -1; 6411 } 6412 return 0; 6413 } 6414 6415 static int 6416 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode) 6417 { 6418 int total_filter_on = 0; 6419 int save_mode; 6420 int ret; 6421 int save_errno; 6422 6423 /* 6424 * The socket filter code doesn't discard all packets queued 6425 * up on the socket when the filter is changed; this means 6426 * that packets that don't match the new filter may show up 6427 * after the new filter is put onto the socket, if those 6428 * packets haven't yet been read. 6429 * 6430 * This means, for example, that if you do a tcpdump capture 6431 * with a filter, the first few packets in the capture might 6432 * be packets that wouldn't have passed the filter. 6433 * 6434 * We therefore discard all packets queued up on the socket 6435 * when setting a kernel filter. (This isn't an issue for 6436 * userland filters, as the userland filtering is done after 6437 * packets are queued up.) 6438 * 6439 * To flush those packets, we put the socket in read-only mode, 6440 * and read packets from the socket until there are no more to 6441 * read. 6442 * 6443 * In order to keep that from being an infinite loop - i.e., 6444 * to keep more packets from arriving while we're draining 6445 * the queue - we put the "total filter", which is a filter 6446 * that rejects all packets, onto the socket before draining 6447 * the queue. 6448 * 6449 * This code deliberately ignores any errors, so that you may 6450 * get bogus packets if an error occurs, rather than having 6451 * the filtering done in userland even if it could have been 6452 * done in the kernel. 6453 */ 6454 if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER, 6455 &total_fcode, sizeof(total_fcode)) == 0) { 6456 char drain[1]; 6457 6458 /* 6459 * Note that we've put the total filter onto the socket. 6460 */ 6461 total_filter_on = 1; 6462 6463 /* 6464 * Save the socket's current mode, and put it in 6465 * non-blocking mode; we drain it by reading packets 6466 * until we get an error (which is normally a 6467 * "nothing more to be read" error). 6468 */ 6469 save_mode = fcntl(handle->fd, F_GETFL, 0); 6470 if (save_mode == -1) { 6471 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6472 "can't get FD flags when changing filter: %s", 6473 pcap_strerror(errno)); 6474 return -2; 6475 } 6476 if (fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) < 0) { 6477 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6478 "can't set nonblocking mode when changing filter: %s", 6479 pcap_strerror(errno)); 6480 return -2; 6481 } 6482 while (recv(handle->fd, &drain, sizeof drain, MSG_TRUNC) >= 0) 6483 ; 6484 save_errno = errno; 6485 if (save_errno != EAGAIN) { 6486 /* 6487 * Fatal error. 6488 * 6489 * If we can't restore the mode or reset the 6490 * kernel filter, there's nothing we can do. 6491 */ 6492 (void)fcntl(handle->fd, F_SETFL, save_mode); 6493 (void)reset_kernel_filter(handle); 6494 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6495 "recv failed when changing filter: %s", 6496 pcap_strerror(save_errno)); 6497 return -2; 6498 } 6499 if (fcntl(handle->fd, F_SETFL, save_mode) == -1) { 6500 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6501 "can't restore FD flags when changing filter: %s", 6502 pcap_strerror(save_errno)); 6503 return -2; 6504 } 6505 } 6506 6507 /* 6508 * Now attach the new filter. 6509 */ 6510 ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER, 6511 fcode, sizeof(*fcode)); 6512 if (ret == -1 && total_filter_on) { 6513 /* 6514 * Well, we couldn't set that filter on the socket, 6515 * but we could set the total filter on the socket. 6516 * 6517 * This could, for example, mean that the filter was 6518 * too big to put into the kernel, so we'll have to 6519 * filter in userland; in any case, we'll be doing 6520 * filtering in userland, so we need to remove the 6521 * total filter so we see packets. 6522 */ 6523 save_errno = errno; 6524 6525 /* 6526 * If this fails, we're really screwed; we have the 6527 * total filter on the socket, and it won't come off. 6528 * Report it as a fatal error. 6529 */ 6530 if (reset_kernel_filter(handle) == -1) { 6531 snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, 6532 "can't remove kernel total filter: %s", 6533 pcap_strerror(errno)); 6534 return -2; /* fatal error */ 6535 } 6536 6537 errno = save_errno; 6538 } 6539 return ret; 6540 } 6541 6542 static int 6543 reset_kernel_filter(pcap_t *handle) 6544 { 6545 /* 6546 * setsockopt() barfs unless it get a dummy parameter. 6547 * valgrind whines unless the value is initialized, 6548 * as it has no idea that setsockopt() ignores its 6549 * parameter. 6550 */ 6551 int dummy = 0; 6552 6553 return setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER, 6554 &dummy, sizeof(dummy)); 6555 } 6556 #endif 6557