xref: /netbsd-src/external/bsd/libpcap/dist/pcap-linux.c (revision 8bda04910f1f2c366c51d4cbd40df19f5b57f1dd)
1 /*	$NetBSD: pcap-linux.c,v 1.8 2024/09/02 15:33:37 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 #define _GNU_SOURCE
74 
75 #include <sys/cdefs.h>
76 __RCSID("$NetBSD: pcap-linux.c,v 1.8 2024/09/02 15:33:37 christos Exp $");
77 
78 #include <config.h>
79 
80 #include <errno.h>
81 #include <stdio.h>
82 #include <stdlib.h>
83 #include <unistd.h>
84 #include <fcntl.h>
85 #include <string.h>
86 #include <limits.h>
87 #include <endian.h>
88 #include <sys/stat.h>
89 #include <sys/socket.h>
90 #include <sys/ioctl.h>
91 #include <sys/utsname.h>
92 #include <sys/mman.h>
93 #include <linux/if.h>
94 #include <linux/if_packet.h>
95 #include <linux/sockios.h>
96 #include <linux/ethtool.h>
97 #include <netinet/in.h>
98 #include <linux/if_ether.h>
99 #include <linux/if_arp.h>
100 #include <poll.h>
101 #include <dirent.h>
102 #include <sys/eventfd.h>
103 
104 #include "pcap-int.h"
105 #include "pcap-util.h"
106 #include "pcap/sll.h"
107 #include "pcap/vlan.h"
108 #include "pcap/can_socketcan.h"
109 
110 #include "diag-control.h"
111 
112 /*
113  * We require TPACKET_V2 support.
114  */
115 #ifndef TPACKET2_HDRLEN
116 #error "Libpcap will only work if TPACKET_V2 is supported; you must build for a 2.6.27 or later kernel"
117 #endif
118 
119 /* check for memory mapped access availability. We assume every needed
120  * struct is defined if the macro TPACKET_HDRLEN is defined, because it
121  * uses many ring related structs and macros */
122 #ifdef TPACKET3_HDRLEN
123 # define HAVE_TPACKET3
124 #endif /* TPACKET3_HDRLEN */
125 
126 /*
127  * Not all compilers that are used to compile code to run on Linux have
128  * these builtins.  For example, older versions of GCC don't, and at
129  * least some people are doing cross-builds for MIPS with older versions
130  * of GCC.
131  */
132 #ifndef HAVE___ATOMIC_LOAD_N
133 #define __atomic_load_n(ptr, memory_model)		(*(ptr))
134 #endif
135 #ifndef HAVE___ATOMIC_STORE_N
136 #define __atomic_store_n(ptr, val, memory_model)	*(ptr) = (val)
137 #endif
138 
139 #define packet_mmap_acquire(pkt) \
140 	(__atomic_load_n(&pkt->tp_status, __ATOMIC_ACQUIRE) != TP_STATUS_KERNEL)
141 #define packet_mmap_release(pkt) \
142 	(__atomic_store_n(&pkt->tp_status, TP_STATUS_KERNEL, __ATOMIC_RELEASE))
143 #define packet_mmap_v3_acquire(pkt) \
144 	(__atomic_load_n(&pkt->hdr.bh1.block_status, __ATOMIC_ACQUIRE) != TP_STATUS_KERNEL)
145 #define packet_mmap_v3_release(pkt) \
146 	(__atomic_store_n(&pkt->hdr.bh1.block_status, TP_STATUS_KERNEL, __ATOMIC_RELEASE))
147 
148 #include <linux/types.h>
149 #include <linux/filter.h>
150 
151 #ifdef HAVE_LINUX_NET_TSTAMP_H
152 #include <linux/net_tstamp.h>
153 #endif
154 
155 /*
156  * For checking whether a device is a bonding device.
157  */
158 #include <linux/if_bonding.h>
159 
160 /*
161  * Got libnl?
162  */
163 #ifdef HAVE_LIBNL
164 #include <linux/nl80211.h>
165 
166 #include <netlink/genl/genl.h>
167 #include <netlink/genl/family.h>
168 #include <netlink/genl/ctrl.h>
169 #include <netlink/msg.h>
170 #include <netlink/attr.h>
171 #endif /* HAVE_LIBNL */
172 
173 #ifndef HAVE_SOCKLEN_T
174 typedef int		socklen_t;
175 #endif
176 
177 #define MAX_LINKHEADER_SIZE	256
178 
179 /*
180  * When capturing on all interfaces we use this as the buffer size.
181  * Should be bigger then all MTUs that occur in real life.
182  * 64kB should be enough for now.
183  */
184 #define BIGGER_THAN_ALL_MTUS	(64*1024)
185 
186 /*
187  * Private data for capturing on Linux PF_PACKET sockets.
188  */
189 struct pcap_linux {
190 	long long sysfs_dropped; /* packets reported dropped by /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors */
191 	struct pcap_stat stat;
192 
193 	char	*device;	/* device name */
194 	int	filter_in_userland; /* must filter in userland */
195 	int	blocks_to_filter_in_userland;
196 	int	must_do_on_close; /* stuff we must do when we close */
197 	int	timeout;	/* timeout for buffering */
198 	int	cooked;		/* using SOCK_DGRAM rather than SOCK_RAW */
199 	int	ifindex;	/* interface index of device we're bound to */
200 	int	lo_ifindex;	/* interface index of the loopback device */
201 	int	netdown;	/* we got an ENETDOWN and haven't resolved it */
202 	bpf_u_int32 oldmode;	/* mode to restore when turning monitor mode off */
203 	char	*mondevice;	/* mac80211 monitor device we created */
204 	u_char	*mmapbuf;	/* memory-mapped region pointer */
205 	size_t	mmapbuflen;	/* size of region */
206 	int	vlan_offset;	/* offset at which to insert vlan tags; if -1, don't insert */
207 	u_int	tp_version;	/* version of tpacket_hdr for mmaped ring */
208 	u_int	tp_hdrlen;	/* hdrlen of tpacket_hdr for mmaped ring */
209 	u_char	*oneshot_buffer; /* buffer for copy of packet */
210 	int	poll_timeout;	/* timeout to use in poll() */
211 #ifdef HAVE_TPACKET3
212 	unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
213 	int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
214 #endif
215 	int poll_breakloop_fd; /* fd to an eventfd to break from blocking operations */
216 };
217 
218 /*
219  * Stuff to do when we close.
220  */
221 #define MUST_CLEAR_RFMON	0x00000001	/* clear rfmon (monitor) mode */
222 #define MUST_DELETE_MONIF	0x00000002	/* delete monitor-mode interface */
223 
224 /*
225  * Prototypes for internal functions and methods.
226  */
227 static int get_if_flags(const char *, bpf_u_int32 *, char *);
228 static int is_wifi(const char *);
229 static int map_arphrd_to_dlt(pcap_t *, int, const char *, int);
230 static int pcap_activate_linux(pcap_t *);
231 static int setup_socket(pcap_t *, int);
232 static int setup_mmapped(pcap_t *);
233 static int pcap_can_set_rfmon_linux(pcap_t *);
234 static int pcap_inject_linux(pcap_t *, const void *, int);
235 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
236 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
237 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
238 static int pcap_set_datalink_linux(pcap_t *, int);
239 static void pcap_cleanup_linux(pcap_t *);
240 
241 union thdr {
242 	struct tpacket2_hdr		*h2;
243 #ifdef HAVE_TPACKET3
244 	struct tpacket_block_desc	*h3;
245 #endif
246 	u_char				*raw;
247 };
248 
249 #define RING_GET_FRAME_AT(h, offset) (((u_char **)h->buffer)[(offset)])
250 #define RING_GET_CURRENT_FRAME(h) RING_GET_FRAME_AT(h, h->offset)
251 
252 static void destroy_ring(pcap_t *handle);
253 static int create_ring(pcap_t *handle);
254 static int prepare_tpacket_socket(pcap_t *handle);
255 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
256 #ifdef HAVE_TPACKET3
257 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
258 #endif
259 static int pcap_setnonblock_linux(pcap_t *p, int nonblock);
260 static int pcap_getnonblock_linux(pcap_t *p);
261 static void pcapint_oneshot_linux(u_char *user, const struct pcap_pkthdr *h,
262     const u_char *bytes);
263 
264 /*
265  * In pre-3.0 kernels, the tp_vlan_tci field is set to whatever the
266  * vlan_tci field in the skbuff is.  0 can either mean "not on a VLAN"
267  * or "on VLAN 0".  There is no flag set in the tp_status field to
268  * distinguish between them.
269  *
270  * In 3.0 and later kernels, if there's a VLAN tag present, the tp_vlan_tci
271  * field is set to the VLAN tag, and the TP_STATUS_VLAN_VALID flag is set
272  * in the tp_status field, otherwise the tp_vlan_tci field is set to 0 and
273  * the TP_STATUS_VLAN_VALID flag isn't set in the tp_status field.
274  *
275  * With a pre-3.0 kernel, we cannot distinguish between packets with no
276  * VLAN tag and packets on VLAN 0, so we will mishandle some packets, and
277  * there's nothing we can do about that.
278  *
279  * So, on those systems, which never set the TP_STATUS_VLAN_VALID flag, we
280  * continue the behavior of earlier libpcaps, wherein we treated packets
281  * with a VLAN tag of 0 as being packets without a VLAN tag rather than packets
282  * on VLAN 0.  We do this by treating packets with a tp_vlan_tci of 0 and
283  * with the TP_STATUS_VLAN_VALID flag not set in tp_status as not having
284  * VLAN tags.  This does the right thing on 3.0 and later kernels, and
285  * continues the old unfixably-imperfect behavior on pre-3.0 kernels.
286  *
287  * If TP_STATUS_VLAN_VALID isn't defined, we test it as the 0x10 bit; it
288  * has that value in 3.0 and later kernels.
289  */
290 #ifdef TP_STATUS_VLAN_VALID
291   #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & TP_STATUS_VLAN_VALID))
292 #else
293   /*
294    * This is being compiled on a system that lacks TP_STATUS_VLAN_VALID,
295    * so we test with the value it has in the 3.0 and later kernels, so
296    * we can test it if we're running on a system that has it.  (If we're
297    * running on a system that doesn't have it, it won't be set in the
298    * tp_status field, so the tests of it will always fail; that means
299    * we behave the way we did before we introduced this macro.)
300    */
301   #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & 0x10))
302 #endif
303 
304 #ifdef TP_STATUS_VLAN_TPID_VALID
305 # define VLAN_TPID(hdr, hv)	(((hv)->tp_vlan_tpid || ((hdr)->tp_status & TP_STATUS_VLAN_TPID_VALID)) ? (hv)->tp_vlan_tpid : ETH_P_8021Q)
306 #else
307 # define VLAN_TPID(hdr, hv)	ETH_P_8021Q
308 #endif
309 
310 /*
311  * Required select timeout if we're polling for an "interface disappeared"
312  * indication - 1 millisecond.
313  */
314 static const struct timeval netdown_timeout = {
315 	0, 1000		/* 1000 microseconds = 1 millisecond */
316 };
317 
318 /*
319  * Wrap some ioctl calls
320  */
321 static int	iface_get_id(int fd, const char *device, char *ebuf);
322 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
323 static int	iface_get_arptype(int fd, const char *device, char *ebuf);
324 static int	iface_bind(int fd, int ifindex, char *ebuf, int protocol);
325 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
326     const char *device);
327 static int	iface_get_ts_types(const char *device, pcap_t *handle,
328     char *ebuf);
329 static int	iface_get_offload(pcap_t *handle);
330 
331 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode);
332 static int	fix_offset(pcap_t *handle, struct bpf_insn *p);
333 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
334 static int	reset_kernel_filter(pcap_t *handle);
335 
336 static struct sock_filter	total_insn
337 	= BPF_STMT(BPF_RET | BPF_K, 0);
338 static struct sock_fprog	total_fcode
339 	= { 1, &total_insn };
340 
341 static int	iface_dsa_get_proto_info(const char *device, pcap_t *handle);
342 
343 pcap_t *
344 pcapint_create_interface(const char *device, char *ebuf)
345 {
346 	pcap_t *handle;
347 
348 	handle = PCAP_CREATE_COMMON(ebuf, struct pcap_linux);
349 	if (handle == NULL)
350 		return NULL;
351 
352 	handle->activate_op = pcap_activate_linux;
353 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
354 
355 	/*
356 	 * See what time stamp types we support.
357 	 */
358 	if (iface_get_ts_types(device, handle, ebuf) == -1) {
359 		pcap_close(handle);
360 		return NULL;
361 	}
362 
363 	/*
364 	 * We claim that we support microsecond and nanosecond time
365 	 * stamps.
366 	 *
367 	 * XXX - with adapter-supplied time stamps, can we choose
368 	 * microsecond or nanosecond time stamps on arbitrary
369 	 * adapters?
370 	 */
371 	handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
372 	if (handle->tstamp_precision_list == NULL) {
373 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
374 		    errno, "malloc");
375 		pcap_close(handle);
376 		return NULL;
377 	}
378 	handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
379 	handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
380 	handle->tstamp_precision_count = 2;
381 
382 	/*
383 	 * Start out with the breakloop handle not open; we don't
384 	 * need it until we're activated and ready to capture.
385 	 */
386 	struct pcap_linux *handlep = handle->priv;
387 	handlep->poll_breakloop_fd = -1;
388 
389 	return handle;
390 }
391 
392 #ifdef HAVE_LIBNL
393 /*
394  * If interface {if_name} is a mac80211 driver, the file
395  * /sys/class/net/{if_name}/phy80211 is a symlink to
396  * /sys/class/ieee80211/{phydev_name}, for some {phydev_name}.
397  *
398  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
399  * least, has a "wmaster0" device and a "wlan0" device; the
400  * latter is the one with the IP address.  Both show up in
401  * "tcpdump -D" output.  Capturing on the wmaster0 device
402  * captures with 802.11 headers.
403  *
404  * airmon-ng searches through /sys/class/net for devices named
405  * monN, starting with mon0; as soon as one *doesn't* exist,
406  * it chooses that as the monitor device name.  If the "iw"
407  * command exists, it does
408  *
409  *    iw dev {if_name} interface add {monif_name} type monitor
410  *
411  * where {monif_name} is the monitor device.  It then (sigh) sleeps
412  * .1 second, and then configures the device up.  Otherwise, if
413  * /sys/class/ieee80211/{phydev_name}/add_iface is a file, it writes
414  * {mondev_name}, without a newline, to that file, and again (sigh)
415  * sleeps .1 second, and then iwconfig's that device into monitor
416  * mode and configures it up.  Otherwise, you can't do monitor mode.
417  *
418  * All these devices are "glued" together by having the
419  * /sys/class/net/{if_name}/phy80211 links pointing to the same
420  * place, so, given a wmaster, wlan, or mon device, you can
421  * find the other devices by looking for devices with
422  * the same phy80211 link.
423  *
424  * To turn monitor mode off, delete the monitor interface,
425  * either with
426  *
427  *    iw dev {monif_name} interface del
428  *
429  * or by sending {monif_name}, with no NL, down
430  * /sys/class/ieee80211/{phydev_name}/remove_iface
431  *
432  * Note: if you try to create a monitor device named "monN", and
433  * there's already a "monN" device, it fails, as least with
434  * the netlink interface (which is what iw uses), with a return
435  * value of -ENFILE.  (Return values are negative errnos.)  We
436  * could probably use that to find an unused device.
437  *
438  * Yes, you can have multiple monitor devices for a given
439  * physical device.
440  */
441 
442 /*
443  * Is this a mac80211 device?  If so, fill in the physical device path and
444  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
445  * return PCAP_ERROR.
446  */
447 static int
448 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
449     size_t phydev_max_pathlen)
450 {
451 	char *pathstr;
452 	ssize_t bytes_read;
453 
454 	/*
455 	 * Generate the path string for the symlink to the physical device.
456 	 */
457 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
458 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
459 		    "%s: Can't generate path name string for /sys/class/net device",
460 		    device);
461 		return PCAP_ERROR;
462 	}
463 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
464 	if (bytes_read == -1) {
465 		if (errno == ENOENT || errno == EINVAL) {
466 			/*
467 			 * Doesn't exist, or not a symlink; assume that
468 			 * means it's not a mac80211 device.
469 			 */
470 			free(pathstr);
471 			return 0;
472 		}
473 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
474 		    errno, "%s: Can't readlink %s", device, pathstr);
475 		free(pathstr);
476 		return PCAP_ERROR;
477 	}
478 	free(pathstr);
479 	phydev_path[bytes_read] = '\0';
480 	return 1;
481 }
482 
483 struct nl80211_state {
484 	struct nl_sock *nl_sock;
485 	struct nl_cache *nl_cache;
486 	struct genl_family *nl80211;
487 };
488 
489 static int
490 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
491 {
492 	int err;
493 
494 	state->nl_sock = nl_socket_alloc();
495 	if (!state->nl_sock) {
496 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
497 		    "%s: failed to allocate netlink handle", device);
498 		return PCAP_ERROR;
499 	}
500 
501 	if (genl_connect(state->nl_sock)) {
502 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
503 		    "%s: failed to connect to generic netlink", device);
504 		goto out_handle_destroy;
505 	}
506 
507 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
508 	if (err < 0) {
509 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
510 		    "%s: failed to allocate generic netlink cache: %s",
511 		    device, nl_geterror(-err));
512 		goto out_handle_destroy;
513 	}
514 
515 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
516 	if (!state->nl80211) {
517 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
518 		    "%s: nl80211 not found", device);
519 		goto out_cache_free;
520 	}
521 
522 	return 0;
523 
524 out_cache_free:
525 	nl_cache_free(state->nl_cache);
526 out_handle_destroy:
527 	nl_socket_free(state->nl_sock);
528 	return PCAP_ERROR;
529 }
530 
531 static void
532 nl80211_cleanup(struct nl80211_state *state)
533 {
534 	genl_family_put(state->nl80211);
535 	nl_cache_free(state->nl_cache);
536 	nl_socket_free(state->nl_sock);
537 }
538 
539 static int
540 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
541     const char *device, const char *mondevice);
542 
543 static int
544 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
545     const char *device, const char *mondevice)
546 {
547 	struct pcap_linux *handlep = handle->priv;
548 	int ifindex;
549 	struct nl_msg *msg;
550 	int err;
551 
552 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
553 	if (ifindex == -1)
554 		return PCAP_ERROR;
555 
556 	msg = nlmsg_alloc();
557 	if (!msg) {
558 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
559 		    "%s: failed to allocate netlink msg", device);
560 		return PCAP_ERROR;
561 	}
562 
563 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
564 		    0, NL80211_CMD_NEW_INTERFACE, 0);
565 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
566 DIAG_OFF_NARROWING
567 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
568 DIAG_ON_NARROWING
569 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
570 
571 	err = nl_send_auto_complete(state->nl_sock, msg);
572 	if (err < 0) {
573 		if (err == -NLE_FAILURE) {
574 			/*
575 			 * Device not available; our caller should just
576 			 * keep trying.  (libnl 2.x maps ENFILE to
577 			 * NLE_FAILURE; it can also map other errors
578 			 * to that, but there's not much we can do
579 			 * about that.)
580 			 */
581 			nlmsg_free(msg);
582 			return 0;
583 		} else {
584 			/*
585 			 * Real failure, not just "that device is not
586 			 * available.
587 			 */
588 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
589 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
590 			    device, mondevice, nl_geterror(-err));
591 			nlmsg_free(msg);
592 			return PCAP_ERROR;
593 		}
594 	}
595 	err = nl_wait_for_ack(state->nl_sock);
596 	if (err < 0) {
597 		if (err == -NLE_FAILURE) {
598 			/*
599 			 * Device not available; our caller should just
600 			 * keep trying.  (libnl 2.x maps ENFILE to
601 			 * NLE_FAILURE; it can also map other errors
602 			 * to that, but there's not much we can do
603 			 * about that.)
604 			 */
605 			nlmsg_free(msg);
606 			return 0;
607 		} else {
608 			/*
609 			 * Real failure, not just "that device is not
610 			 * available.
611 			 */
612 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
613 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
614 			    device, mondevice, nl_geterror(-err));
615 			nlmsg_free(msg);
616 			return PCAP_ERROR;
617 		}
618 	}
619 
620 	/*
621 	 * Success.
622 	 */
623 	nlmsg_free(msg);
624 
625 	/*
626 	 * Try to remember the monitor device.
627 	 */
628 	handlep->mondevice = strdup(mondevice);
629 	if (handlep->mondevice == NULL) {
630 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
631 		    errno, "strdup");
632 		/*
633 		 * Get rid of the monitor device.
634 		 */
635 		del_mon_if(handle, sock_fd, state, device, mondevice);
636 		return PCAP_ERROR;
637 	}
638 	return 1;
639 
640 nla_put_failure:
641 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
642 	    "%s: nl_put failed adding %s interface",
643 	    device, mondevice);
644 	nlmsg_free(msg);
645 	return PCAP_ERROR;
646 }
647 
648 static int
649 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
650     const char *device, const char *mondevice)
651 {
652 	int ifindex;
653 	struct nl_msg *msg;
654 	int err;
655 
656 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
657 	if (ifindex == -1)
658 		return PCAP_ERROR;
659 
660 	msg = nlmsg_alloc();
661 	if (!msg) {
662 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
663 		    "%s: failed to allocate netlink msg", device);
664 		return PCAP_ERROR;
665 	}
666 
667 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
668 		    0, NL80211_CMD_DEL_INTERFACE, 0);
669 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
670 
671 	err = nl_send_auto_complete(state->nl_sock, msg);
672 	if (err < 0) {
673 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
674 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
675 		    device, mondevice, nl_geterror(-err));
676 		nlmsg_free(msg);
677 		return PCAP_ERROR;
678 	}
679 	err = nl_wait_for_ack(state->nl_sock);
680 	if (err < 0) {
681 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
682 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
683 		    device, mondevice, nl_geterror(-err));
684 		nlmsg_free(msg);
685 		return PCAP_ERROR;
686 	}
687 
688 	/*
689 	 * Success.
690 	 */
691 	nlmsg_free(msg);
692 	return 1;
693 
694 nla_put_failure:
695 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
696 	    "%s: nl_put failed deleting %s interface",
697 	    device, mondevice);
698 	nlmsg_free(msg);
699 	return PCAP_ERROR;
700 }
701 #endif /* HAVE_LIBNL */
702 
703 static int pcap_protocol(pcap_t *handle)
704 {
705 	int protocol;
706 
707 	protocol = handle->opt.protocol;
708 	if (protocol == 0)
709 		protocol = ETH_P_ALL;
710 
711 	return htons(protocol);
712 }
713 
714 static int
715 pcap_can_set_rfmon_linux(pcap_t *handle)
716 {
717 #ifdef HAVE_LIBNL
718 	char phydev_path[PATH_MAX+1];
719 	int ret;
720 #endif
721 
722 	if (strcmp(handle->opt.device, "any") == 0) {
723 		/*
724 		 * Monitor mode makes no sense on the "any" device.
725 		 */
726 		return 0;
727 	}
728 
729 #ifdef HAVE_LIBNL
730 	/*
731 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
732 	 * device, through libnl, whether it supports monitor mode;
733 	 * we'll just check whether the device appears to be a
734 	 * mac80211 device and, if so, assume the device supports
735 	 * monitor mode.
736 	 */
737 	ret = get_mac80211_phydev(handle, handle->opt.device, phydev_path,
738 	    PATH_MAX);
739 	if (ret < 0)
740 		return ret;	/* error */
741 	if (ret == 1)
742 		return 1;	/* mac80211 device */
743 #endif
744 
745 	return 0;
746 }
747 
748 /*
749  * Grabs the number of missed packets by the interface from
750  * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors.
751  *
752  * Compared to /proc/net/dev this avoids counting software drops,
753  * but may be unimplemented and just return 0.
754  * The author has found no straightforward way to check for support.
755  */
756 static long long int
757 linux_get_stat(const char * if_name, const char * stat) {
758 	ssize_t bytes_read;
759 	int fd;
760 	char buffer[PATH_MAX];
761 
762 	snprintf(buffer, sizeof(buffer), "/sys/class/net/%s/statistics/%s", if_name, stat);
763 	fd = open(buffer, O_RDONLY);
764 	if (fd == -1)
765 		return 0;
766 
767 	bytes_read = read(fd, buffer, sizeof(buffer) - 1);
768 	close(fd);
769 	if (bytes_read == -1)
770 		return 0;
771 	buffer[bytes_read] = '\0';
772 
773 	return strtoll(buffer, NULL, 10);
774 }
775 
776 static long long int
777 linux_if_drops(const char * if_name)
778 {
779 	long long int missed = linux_get_stat(if_name, "rx_missed_errors");
780 	long long int fifo = linux_get_stat(if_name, "rx_fifo_errors");
781 	return missed + fifo;
782 }
783 
784 
785 /*
786  * Monitor mode is kind of interesting because we have to reset the
787  * interface before exiting. The problem can't really be solved without
788  * some daemon taking care of managing usage counts.  If we put the
789  * interface into monitor mode, we set a flag indicating that we must
790  * take it out of that mode when the interface is closed, and, when
791  * closing the interface, if that flag is set we take it out of monitor
792  * mode.
793  */
794 
795 static void	pcap_cleanup_linux( pcap_t *handle )
796 {
797 	struct pcap_linux *handlep = handle->priv;
798 #ifdef HAVE_LIBNL
799 	struct nl80211_state nlstate;
800 	int ret;
801 #endif /* HAVE_LIBNL */
802 
803 	if (handlep->must_do_on_close != 0) {
804 		/*
805 		 * There's something we have to do when closing this
806 		 * pcap_t.
807 		 */
808 #ifdef HAVE_LIBNL
809 		if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
810 			ret = nl80211_init(handle, &nlstate, handlep->device);
811 			if (ret >= 0) {
812 				ret = del_mon_if(handle, handle->fd, &nlstate,
813 				    handlep->device, handlep->mondevice);
814 				nl80211_cleanup(&nlstate);
815 			}
816 			if (ret < 0) {
817 				fprintf(stderr,
818 				    "Can't delete monitor interface %s (%s).\n"
819 				    "Please delete manually.\n",
820 				    handlep->mondevice, handle->errbuf);
821 			}
822 		}
823 #endif /* HAVE_LIBNL */
824 
825 		/*
826 		 * Take this pcap out of the list of pcaps for which we
827 		 * have to take the interface out of some mode.
828 		 */
829 		pcapint_remove_from_pcaps_to_close(handle);
830 	}
831 
832 	if (handle->fd != -1) {
833 		/*
834 		 * Destroy the ring buffer (assuming we've set it up),
835 		 * and unmap it if it's mapped.
836 		 */
837 		destroy_ring(handle);
838 	}
839 
840 	if (handlep->oneshot_buffer != NULL) {
841 		free(handlep->oneshot_buffer);
842 		handlep->oneshot_buffer = NULL;
843 	}
844 
845 	if (handlep->mondevice != NULL) {
846 		free(handlep->mondevice);
847 		handlep->mondevice = NULL;
848 	}
849 	if (handlep->device != NULL) {
850 		free(handlep->device);
851 		handlep->device = NULL;
852 	}
853 
854 	if (handlep->poll_breakloop_fd != -1) {
855 		close(handlep->poll_breakloop_fd);
856 		handlep->poll_breakloop_fd = -1;
857 	}
858 	pcapint_cleanup_live_common(handle);
859 }
860 
861 #ifdef HAVE_TPACKET3
862 /*
863  * Some versions of TPACKET_V3 have annoying bugs/misfeatures
864  * around which we have to work.  Determine if we have those
865  * problems or not.
866  * 3.19 is the first release with a fixed version of
867  * TPACKET_V3.  We treat anything before that as
868  * not having a fixed version; that may really mean
869  * it has *no* version.
870  */
871 static int has_broken_tpacket_v3(void)
872 {
873 	struct utsname utsname;
874 	const char *release;
875 	long major, minor;
876 	int matches, verlen;
877 
878 	/* No version information, assume broken. */
879 	if (uname(&utsname) == -1)
880 		return 1;
881 	release = utsname.release;
882 
883 	/* A malformed version, ditto. */
884 	matches = sscanf(release, "%ld.%ld%n", &major, &minor, &verlen);
885 	if (matches != 2)
886 		return 1;
887 	if (release[verlen] != '.' && release[verlen] != '\0')
888 		return 1;
889 
890 	/* OK, a fixed version. */
891 	if (major > 3 || (major == 3 && minor >= 19))
892 		return 0;
893 
894 	/* Too old :( */
895 	return 1;
896 }
897 #endif
898 
899 /*
900  * Set the timeout to be used in poll() with memory-mapped packet capture.
901  */
902 static void
903 set_poll_timeout(struct pcap_linux *handlep)
904 {
905 #ifdef HAVE_TPACKET3
906 	int broken_tpacket_v3 = has_broken_tpacket_v3();
907 #endif
908 	if (handlep->timeout == 0) {
909 #ifdef HAVE_TPACKET3
910 		/*
911 		 * XXX - due to a set of (mis)features in the TPACKET_V3
912 		 * kernel code prior to the 3.19 kernel, blocking forever
913 		 * with a TPACKET_V3 socket can, if few packets are
914 		 * arriving and passing the socket filter, cause most
915 		 * packets to be dropped.  See libpcap issue #335 for the
916 		 * full painful story.
917 		 *
918 		 * The workaround is to have poll() time out very quickly,
919 		 * so we grab the frames handed to us, and return them to
920 		 * the kernel, ASAP.
921 		 */
922 		if (handlep->tp_version == TPACKET_V3 && broken_tpacket_v3)
923 			handlep->poll_timeout = 1;	/* don't block for very long */
924 		else
925 #endif
926 			handlep->poll_timeout = -1;	/* block forever */
927 	} else if (handlep->timeout > 0) {
928 #ifdef HAVE_TPACKET3
929 		/*
930 		 * For TPACKET_V3, the timeout is handled by the kernel,
931 		 * so block forever; that way, we don't get extra timeouts.
932 		 * Don't do that if we have a broken TPACKET_V3, though.
933 		 */
934 		if (handlep->tp_version == TPACKET_V3 && !broken_tpacket_v3)
935 			handlep->poll_timeout = -1;	/* block forever, let TPACKET_V3 wake us up */
936 		else
937 #endif
938 			handlep->poll_timeout = handlep->timeout;	/* block for that amount of time */
939 	} else {
940 		/*
941 		 * Non-blocking mode; we call poll() to pick up error
942 		 * indications, but we don't want it to wait for
943 		 * anything.
944 		 */
945 		handlep->poll_timeout = 0;
946 	}
947 }
948 
949 static void pcap_breakloop_linux(pcap_t *handle)
950 {
951 	pcapint_breakloop_common(handle);
952 	struct pcap_linux *handlep = handle->priv;
953 
954 	uint64_t value = 1;
955 
956 	if (handlep->poll_breakloop_fd != -1) {
957 		/*
958 		 * XXX - pcap_breakloop() doesn't have a return value,
959 		 * so we can't indicate an error.
960 		 */
961 DIAG_OFF_WARN_UNUSED_RESULT
962 		(void)write(handlep->poll_breakloop_fd, &value, sizeof(value));
963 DIAG_ON_WARN_UNUSED_RESULT
964 	}
965 }
966 
967 /*
968  * Set the offset at which to insert VLAN tags.
969  * That should be the offset of the type field.
970  */
971 static void
972 set_vlan_offset(pcap_t *handle)
973 {
974 	struct pcap_linux *handlep = handle->priv;
975 
976 	switch (handle->linktype) {
977 
978 	case DLT_EN10MB:
979 		/*
980 		 * The type field is after the destination and source
981 		 * MAC address.
982 		 */
983 		handlep->vlan_offset = 2 * ETH_ALEN;
984 		break;
985 
986 	case DLT_LINUX_SLL:
987 		/*
988 		 * The type field is in the last 2 bytes of the
989 		 * DLT_LINUX_SLL header.
990 		 */
991 		handlep->vlan_offset = SLL_HDR_LEN - 2;
992 		break;
993 
994 	default:
995 		handlep->vlan_offset = -1; /* unknown */
996 		break;
997 	}
998 }
999 
1000 /*
1001  *  Get a handle for a live capture from the given device. You can
1002  *  pass NULL as device to get all packages (without link level
1003  *  information of course). If you pass 1 as promisc the interface
1004  *  will be set to promiscuous mode (XXX: I think this usage should
1005  *  be deprecated and functions be added to select that later allow
1006  *  modification of that values -- Torsten).
1007  */
1008 static int
1009 pcap_activate_linux(pcap_t *handle)
1010 {
1011 	struct pcap_linux *handlep = handle->priv;
1012 	const char	*device;
1013 	int		is_any_device;
1014 	struct ifreq	ifr;
1015 	int		status;
1016 	int		ret;
1017 
1018 	device = handle->opt.device;
1019 
1020 	/*
1021 	 * Start out assuming no warnings.
1022 	 */
1023 	status = 0;
1024 
1025 	/*
1026 	 * Make sure the name we were handed will fit into the ioctls we
1027 	 * might perform on the device; if not, return a "No such device"
1028 	 * indication, as the Linux kernel shouldn't support creating
1029 	 * a device whose name won't fit into those ioctls.
1030 	 *
1031 	 * "Will fit" means "will fit, complete with a null terminator",
1032 	 * so if the length, which does *not* include the null terminator,
1033 	 * is greater than *or equal to* the size of the field into which
1034 	 * we'll be copying it, that won't fit.
1035 	 */
1036 	if (strlen(device) >= sizeof(ifr.ifr_name)) {
1037 		/*
1038 		 * There's nothing more to say, so clear the error
1039 		 * message.
1040 		 */
1041 		handle->errbuf[0] = '\0';
1042 		status = PCAP_ERROR_NO_SUCH_DEVICE;
1043 		goto fail;
1044 	}
1045 
1046 	/*
1047 	 * Turn a negative snapshot value (invalid), a snapshot value of
1048 	 * 0 (unspecified), or a value bigger than the normal maximum
1049 	 * value, into the maximum allowed value.
1050 	 *
1051 	 * If some application really *needs* a bigger snapshot
1052 	 * length, we should just increase MAXIMUM_SNAPLEN.
1053 	 */
1054 	if (handle->snapshot <= 0 || handle->snapshot > MAXIMUM_SNAPLEN)
1055 		handle->snapshot = MAXIMUM_SNAPLEN;
1056 
1057 	handlep->device	= strdup(device);
1058 	if (handlep->device == NULL) {
1059 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1060 		    errno, "strdup");
1061 		status = PCAP_ERROR;
1062 		goto fail;
1063 	}
1064 
1065 	/*
1066 	 * The "any" device is a special device which causes us not
1067 	 * to bind to a particular device and thus to look at all
1068 	 * devices.
1069 	 */
1070 	is_any_device = (strcmp(device, "any") == 0);
1071 	if (is_any_device) {
1072 		if (handle->opt.promisc) {
1073 			handle->opt.promisc = 0;
1074 			/* Just a warning. */
1075 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1076 			    "Promiscuous mode not supported on the \"any\" device");
1077 			status = PCAP_WARNING_PROMISC_NOTSUP;
1078 		}
1079 	}
1080 
1081 	/* copy timeout value */
1082 	handlep->timeout = handle->opt.timeout;
1083 
1084 	/*
1085 	 * If we're in promiscuous mode, then we probably want
1086 	 * to see when the interface drops packets too, so get an
1087 	 * initial count from
1088 	 * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors
1089 	 */
1090 	if (handle->opt.promisc)
1091 		handlep->sysfs_dropped = linux_if_drops(handlep->device);
1092 
1093 	/*
1094 	 * If the "any" device is specified, try to open a SOCK_DGRAM.
1095 	 * Otherwise, open a SOCK_RAW.
1096 	 */
1097 	ret = setup_socket(handle, is_any_device);
1098 	if (ret < 0) {
1099 		/*
1100 		 * Fatal error; the return value is the error code,
1101 		 * and handle->errbuf has been set to an appropriate
1102 		 * error message.
1103 		 */
1104 		status = ret;
1105 		goto fail;
1106 	}
1107 	if (ret > 0) {
1108 		/*
1109 		 * We got a warning; return that, as handle->errbuf
1110 		 * might have been overwritten by this warning.
1111 		 */
1112 		status = ret;
1113 	}
1114 
1115 	/*
1116 	 * Success (possibly with a warning).
1117 	 *
1118 	 * First, try to allocate an event FD for breakloop, if
1119 	 * we're not going to start in non-blocking mode.
1120 	 */
1121 	if (!handle->opt.nonblock) {
1122 		handlep->poll_breakloop_fd = eventfd(0, EFD_NONBLOCK);
1123 		if (handlep->poll_breakloop_fd == -1) {
1124 			/*
1125 			 * Failed.
1126 			 */
1127 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
1128 			    PCAP_ERRBUF_SIZE, errno, "could not open eventfd");
1129 			status = PCAP_ERROR;
1130 			goto fail;
1131 		}
1132 	}
1133 
1134 	/*
1135 	 * Succeeded.
1136 	 * Try to set up memory-mapped access.
1137 	 */
1138 	ret = setup_mmapped(handle);
1139 	if (ret < 0) {
1140 		/*
1141 		 * We failed to set up to use it, or the
1142 		 * kernel supports it, but we failed to
1143 		 * enable it.  The return value is the
1144 		 * error status to return and, if it's
1145 		 * PCAP_ERROR, handle->errbuf contains
1146 		 * the error message.
1147 		 */
1148 		status = ret;
1149 		goto fail;
1150 	}
1151 	if (ret > 0) {
1152 		/*
1153 		 * We got a warning; return that, as handle->errbuf
1154 		 * might have been overwritten by this warning.
1155 		 */
1156 		status = ret;
1157 	}
1158 
1159 	/*
1160 	 * We succeeded.  status has been set to the status to return,
1161 	 * which might be 0, or might be a PCAP_WARNING_ value.
1162 	 */
1163 	/*
1164 	 * Now that we have activated the mmap ring, we can
1165 	 * set the correct protocol.
1166 	 */
1167 	if ((ret = iface_bind(handle->fd, handlep->ifindex,
1168 	    handle->errbuf, pcap_protocol(handle))) != 0) {
1169 		status = ret;
1170 		goto fail;
1171 	}
1172 
1173 	handle->inject_op = pcap_inject_linux;
1174 	handle->setfilter_op = pcap_setfilter_linux;
1175 	handle->setdirection_op = pcap_setdirection_linux;
1176 	handle->set_datalink_op = pcap_set_datalink_linux;
1177 	handle->setnonblock_op = pcap_setnonblock_linux;
1178 	handle->getnonblock_op = pcap_getnonblock_linux;
1179 	handle->cleanup_op = pcap_cleanup_linux;
1180 	handle->stats_op = pcap_stats_linux;
1181 	handle->breakloop_op = pcap_breakloop_linux;
1182 
1183 	switch (handlep->tp_version) {
1184 
1185 	case TPACKET_V2:
1186 		handle->read_op = pcap_read_linux_mmap_v2;
1187 		break;
1188 #ifdef HAVE_TPACKET3
1189 	case TPACKET_V3:
1190 		handle->read_op = pcap_read_linux_mmap_v3;
1191 		break;
1192 #endif
1193 	}
1194 	handle->oneshot_callback = pcapint_oneshot_linux;
1195 	handle->selectable_fd = handle->fd;
1196 
1197 	return status;
1198 
1199 fail:
1200 	pcap_cleanup_linux(handle);
1201 	return status;
1202 }
1203 
1204 static int
1205 pcap_set_datalink_linux(pcap_t *handle, int dlt)
1206 {
1207 	handle->linktype = dlt;
1208 
1209 	/*
1210 	 * Update the offset at which to insert VLAN tags for the
1211 	 * new link-layer type.
1212 	 */
1213 	set_vlan_offset(handle);
1214 
1215 	return 0;
1216 }
1217 
1218 /*
1219  * linux_check_direction()
1220  *
1221  * Do checks based on packet direction.
1222  */
1223 static inline int
1224 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1225 {
1226 	struct pcap_linux	*handlep = handle->priv;
1227 
1228 	if (sll->sll_pkttype == PACKET_OUTGOING) {
1229 		/*
1230 		 * Outgoing packet.
1231 		 * If this is from the loopback device, reject it;
1232 		 * we'll see the packet as an incoming packet as well,
1233 		 * and we don't want to see it twice.
1234 		 */
1235 		if (sll->sll_ifindex == handlep->lo_ifindex)
1236 			return 0;
1237 
1238 		/*
1239 		 * If this is an outgoing CAN or CAN FD frame, and
1240 		 * the user doesn't only want outgoing packets,
1241 		 * reject it; CAN devices and drivers, and the CAN
1242 		 * stack, always arrange to loop back transmitted
1243 		 * packets, so they also appear as incoming packets.
1244 		 * We don't want duplicate packets, and we can't
1245 		 * easily distinguish packets looped back by the CAN
1246 		 * layer than those received by the CAN layer, so we
1247 		 * eliminate this packet instead.
1248 		 *
1249 		 * We check whether this is a CAN or CAN FD frame
1250 		 * by checking whether the device's hardware type
1251 		 * is ARPHRD_CAN.
1252 		 */
1253 		if (sll->sll_hatype == ARPHRD_CAN &&
1254 		     handle->direction != PCAP_D_OUT)
1255 			return 0;
1256 
1257 		/*
1258 		 * If the user only wants incoming packets, reject it.
1259 		 */
1260 		if (handle->direction == PCAP_D_IN)
1261 			return 0;
1262 	} else {
1263 		/*
1264 		 * Incoming packet.
1265 		 * If the user only wants outgoing packets, reject it.
1266 		 */
1267 		if (handle->direction == PCAP_D_OUT)
1268 			return 0;
1269 	}
1270 	return 1;
1271 }
1272 
1273 /*
1274  * Check whether the device to which the pcap_t is bound still exists.
1275  * We do so by asking what address the socket is bound to, and checking
1276  * whether the ifindex in the address is -1, meaning "that device is gone",
1277  * or some other value, meaning "that device still exists".
1278  */
1279 static int
1280 device_still_exists(pcap_t *handle)
1281 {
1282 	struct pcap_linux *handlep = handle->priv;
1283 	struct sockaddr_ll addr;
1284 	socklen_t addr_len;
1285 
1286 	/*
1287 	 * If handlep->ifindex is -1, the socket isn't bound, meaning
1288 	 * we're capturing on the "any" device; that device never
1289 	 * disappears.  (It should also never be configured down, so
1290 	 * we shouldn't even get here, but let's make sure.)
1291 	 */
1292 	if (handlep->ifindex == -1)
1293 		return (1);	/* it's still here */
1294 
1295 	/*
1296 	 * OK, now try to get the address for the socket.
1297 	 */
1298 	addr_len = sizeof (addr);
1299 	if (getsockname(handle->fd, (struct sockaddr *) &addr, &addr_len) == -1) {
1300 		/*
1301 		 * Error - report an error and return -1.
1302 		 */
1303 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1304 		    errno, "getsockname failed");
1305 		return (-1);
1306 	}
1307 	if (addr.sll_ifindex == -1) {
1308 		/*
1309 		 * This means the device went away.
1310 		 */
1311 		return (0);
1312 	}
1313 
1314 	/*
1315 	 * The device presumably just went down.
1316 	 */
1317 	return (1);
1318 }
1319 
1320 static int
1321 pcap_inject_linux(pcap_t *handle, const void *buf, int size)
1322 {
1323 	struct pcap_linux *handlep = handle->priv;
1324 	int ret;
1325 
1326 	if (handlep->ifindex == -1) {
1327 		/*
1328 		 * We don't support sending on the "any" device.
1329 		 */
1330 		pcapint_strlcpy(handle->errbuf,
1331 		    "Sending packets isn't supported on the \"any\" device",
1332 		    PCAP_ERRBUF_SIZE);
1333 		return (-1);
1334 	}
1335 
1336 	if (handlep->cooked) {
1337 		/*
1338 		 * We don't support sending on cooked-mode sockets.
1339 		 *
1340 		 * XXX - how do you send on a bound cooked-mode
1341 		 * socket?
1342 		 * Is a "sendto()" required there?
1343 		 */
1344 		pcapint_strlcpy(handle->errbuf,
1345 		    "Sending packets isn't supported in cooked mode",
1346 		    PCAP_ERRBUF_SIZE);
1347 		return (-1);
1348 	}
1349 
1350 	ret = (int)send(handle->fd, buf, size, 0);
1351 	if (ret == -1) {
1352 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1353 		    errno, "send");
1354 		return (-1);
1355 	}
1356 	return (ret);
1357 }
1358 
1359 /*
1360  *  Get the statistics for the given packet capture handle.
1361  */
1362 static int
1363 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1364 {
1365 	struct pcap_linux *handlep = handle->priv;
1366 #ifdef HAVE_TPACKET3
1367 	/*
1368 	 * For sockets using TPACKET_V2, the extra stuff at the end
1369 	 * of a struct tpacket_stats_v3 will not be filled in, and
1370 	 * we don't look at it so this is OK even for those sockets.
1371 	 * In addition, the PF_PACKET socket code in the kernel only
1372 	 * uses the length parameter to compute how much data to
1373 	 * copy out and to indicate how much data was copied out, so
1374 	 * it's OK to base it on the size of a struct tpacket_stats.
1375 	 *
1376 	 * XXX - it's probably OK, in fact, to just use a
1377 	 * struct tpacket_stats for V3 sockets, as we don't
1378 	 * care about the tp_freeze_q_cnt stat.
1379 	 */
1380 	struct tpacket_stats_v3 kstats;
1381 #else /* HAVE_TPACKET3 */
1382 	struct tpacket_stats kstats;
1383 #endif /* HAVE_TPACKET3 */
1384 	socklen_t len = sizeof (struct tpacket_stats);
1385 
1386 	long long if_dropped = 0;
1387 
1388 	/*
1389 	 * To fill in ps_ifdrop, we parse
1390 	 * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors
1391 	 * for the numbers
1392 	 */
1393 	if (handle->opt.promisc)
1394 	{
1395 		/*
1396 		 * XXX - is there any reason to do this by remembering
1397 		 * the last counts value, subtracting it from the
1398 		 * current counts value, and adding that to stat.ps_ifdrop,
1399 		 * maintaining stat.ps_ifdrop as a count, rather than just
1400 		 * saving the *initial* counts value and setting
1401 		 * stat.ps_ifdrop to the difference between the current
1402 		 * value and the initial value?
1403 		 *
1404 		 * One reason might be to handle the count wrapping
1405 		 * around, on platforms where the count is 32 bits
1406 		 * and where you might get more than 2^32 dropped
1407 		 * packets; is there any other reason?
1408 		 *
1409 		 * (We maintain the count as a long long int so that,
1410 		 * if the kernel maintains the counts as 64-bit even
1411 		 * on 32-bit platforms, we can handle the real count.
1412 		 *
1413 		 * Unfortunately, we can't report 64-bit counts; we
1414 		 * need a better API for reporting statistics, such as
1415 		 * one that reports them in a style similar to the
1416 		 * pcapng Interface Statistics Block, so that 1) the
1417 		 * counts are 64-bit, 2) it's easier to add new statistics
1418 		 * without breaking the ABI, and 3) it's easier to
1419 		 * indicate to a caller that wants one particular
1420 		 * statistic that it's not available by just not supplying
1421 		 * it.)
1422 		 */
1423 		if_dropped = handlep->sysfs_dropped;
1424 		handlep->sysfs_dropped = linux_if_drops(handlep->device);
1425 		handlep->stat.ps_ifdrop += (u_int)(handlep->sysfs_dropped - if_dropped);
1426 	}
1427 
1428 	/*
1429 	 * Try to get the packet counts from the kernel.
1430 	 */
1431 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1432 			&kstats, &len) > -1) {
1433 		/*
1434 		 * "ps_recv" counts only packets that *passed* the
1435 		 * filter, not packets that didn't pass the filter.
1436 		 * This includes packets later dropped because we
1437 		 * ran out of buffer space.
1438 		 *
1439 		 * "ps_drop" counts packets dropped because we ran
1440 		 * out of buffer space.  It doesn't count packets
1441 		 * dropped by the interface driver.  It counts only
1442 		 * packets that passed the filter.
1443 		 *
1444 		 * See above for ps_ifdrop.
1445 		 *
1446 		 * Both statistics include packets not yet read from
1447 		 * the kernel by libpcap, and thus not yet seen by
1448 		 * the application.
1449 		 *
1450 		 * In "linux/net/packet/af_packet.c", at least in 2.6.27
1451 		 * through 5.6 kernels, "tp_packets" is incremented for
1452 		 * every packet that passes the packet filter *and* is
1453 		 * successfully copied to the ring buffer; "tp_drops" is
1454 		 * incremented for every packet dropped because there's
1455 		 * not enough free space in the ring buffer.
1456 		 *
1457 		 * When the statistics are returned for a PACKET_STATISTICS
1458 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1459 		 * so that "tp_packets" counts all packets handed to
1460 		 * the PF_PACKET socket, including packets dropped because
1461 		 * there wasn't room on the socket buffer - but not
1462 		 * including packets that didn't pass the filter.
1463 		 *
1464 		 * In the BSD BPF, the count of received packets is
1465 		 * incremented for every packet handed to BPF, regardless
1466 		 * of whether it passed the filter.
1467 		 *
1468 		 * We can't make "pcap_stats()" work the same on both
1469 		 * platforms, but the best approximation is to return
1470 		 * "tp_packets" as the count of packets and "tp_drops"
1471 		 * as the count of drops.
1472 		 *
1473 		 * Keep a running total because each call to
1474 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1475 		 * resets the counters to zero.
1476 		 */
1477 		handlep->stat.ps_recv += kstats.tp_packets;
1478 		handlep->stat.ps_drop += kstats.tp_drops;
1479 		*stats = handlep->stat;
1480 		return 0;
1481 	}
1482 
1483 	pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE, errno,
1484 	    "failed to get statistics from socket");
1485 	return -1;
1486 }
1487 
1488 /*
1489  * A PF_PACKET socket can be bound to any network interface.
1490  */
1491 static int
1492 can_be_bound(const char *name _U_)
1493 {
1494 	return (1);
1495 }
1496 
1497 /*
1498  * Get a socket to use with various interface ioctls.
1499  */
1500 static int
1501 get_if_ioctl_socket(void)
1502 {
1503 	int fd;
1504 
1505 	/*
1506 	 * This is a bit ugly.
1507 	 *
1508 	 * There isn't a socket type that's guaranteed to work.
1509 	 *
1510 	 * AF_NETLINK will work *if* you have Netlink configured into the
1511 	 * kernel (can it be configured out if you have any networking
1512 	 * support at all?) *and* if you're running a sufficiently recent
1513 	 * kernel, but not all the kernels we support are sufficiently
1514 	 * recent - that feature was introduced in Linux 4.6.
1515 	 *
1516 	 * AF_UNIX will work *if* you have UNIX-domain sockets configured
1517 	 * into the kernel and *if* you're not on a system that doesn't
1518 	 * allow them - some SELinux systems don't allow you create them.
1519 	 * Most systems probably have them configured in, but not all systems
1520 	 * have them configured in and allow them to be created.
1521 	 *
1522 	 * AF_INET will work *if* you have IPv4 configured into the kernel,
1523 	 * but, apparently, some systems have network adapters but have
1524 	 * kernels without IPv4 support.
1525 	 *
1526 	 * AF_INET6 will work *if* you have IPv6 configured into the
1527 	 * kernel, but if you don't have AF_INET, you might not have
1528 	 * AF_INET6, either (that is, independently on its own grounds).
1529 	 *
1530 	 * AF_PACKET would work, except that some of these calls should
1531 	 * work even if you *don't* have capture permission (you should be
1532 	 * able to enumerate interfaces and get information about them
1533 	 * without capture permission; you shouldn't get a failure until
1534 	 * you try pcap_activate()).  (If you don't allow programs to
1535 	 * get as much information as possible about interfaces if you
1536 	 * don't have permission to capture, you run the risk of users
1537 	 * asking "why isn't it showing XXX" - or, worse, if you don't
1538 	 * show interfaces *at all* if you don't have permission to
1539 	 * capture on them, "why do no interfaces show up?" - when the
1540 	 * real problem is a permissions problem.  Error reports of that
1541 	 * type require a lot more back-and-forth to debug, as evidenced
1542 	 * by many Wireshark bugs/mailing list questions/Q&A questions.)
1543 	 *
1544 	 * So:
1545 	 *
1546 	 * we first try an AF_NETLINK socket, where "try" includes
1547 	 * "try to do a device ioctl on it", as, in the future, once
1548 	 * pre-4.6 kernels are sufficiently rare, that will probably
1549 	 * be the mechanism most likely to work;
1550 	 *
1551 	 * if that fails, we try an AF_UNIX socket, as that's less
1552 	 * likely to be configured out on a networking-capable system
1553 	 * than is IP;
1554 	 *
1555 	 * if that fails, we try an AF_INET6 socket;
1556 	 *
1557 	 * if that fails, we try an AF_INET socket.
1558 	 */
1559 	fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
1560 	if (fd != -1) {
1561 		/*
1562 		 * OK, let's make sure we can do an SIOCGIFNAME
1563 		 * ioctl.
1564 		 */
1565 		struct ifreq ifr;
1566 
1567 		memset(&ifr, 0, sizeof(ifr));
1568 		if (ioctl(fd, SIOCGIFNAME, &ifr) == 0 ||
1569 		    errno != EOPNOTSUPP) {
1570 			/*
1571 			 * It succeeded, or failed for some reason
1572 			 * other than "netlink sockets don't support
1573 			 * device ioctls".  Go with the AF_NETLINK
1574 			 * socket.
1575 			 */
1576 			return (fd);
1577 		}
1578 
1579 		/*
1580 		 * OK, that didn't work, so it's as bad as "netlink
1581 		 * sockets aren't available".  Close the socket and
1582 		 * drive on.
1583 		 */
1584 		close(fd);
1585 	}
1586 
1587 	/*
1588 	 * Now try an AF_UNIX socket.
1589 	 */
1590 	fd = socket(AF_UNIX, SOCK_RAW, 0);
1591 	if (fd != -1) {
1592 		/*
1593 		 * OK, we got it!
1594 		 */
1595 		return (fd);
1596 	}
1597 
1598 	/*
1599 	 * Now try an AF_INET6 socket.
1600 	 */
1601 	fd = socket(AF_INET6, SOCK_DGRAM, 0);
1602 	if (fd != -1) {
1603 		return (fd);
1604 	}
1605 
1606 	/*
1607 	 * Now try an AF_INET socket.
1608 	 *
1609 	 * XXX - if that fails, is there anything else we should try?
1610 	 * AF_CAN, for embedded systems in vehicles, in case they're
1611 	 * built without Internet protocol support?  Any other socket
1612 	 * types popular in non-Internet embedded systems?
1613 	 */
1614 	return (socket(AF_INET, SOCK_DGRAM, 0));
1615 }
1616 
1617 /*
1618  * Get additional flags for a device, using SIOCGIFMEDIA.
1619  */
1620 static int
1621 get_if_flags(const char *name, bpf_u_int32 *flags, char *errbuf)
1622 {
1623 	int sock;
1624 	FILE *fh;
1625 	unsigned int arptype;
1626 	struct ifreq ifr;
1627 	struct ethtool_value info;
1628 
1629 	if (*flags & PCAP_IF_LOOPBACK) {
1630 		/*
1631 		 * Loopback devices aren't wireless, and "connected"/
1632 		 * "disconnected" doesn't apply to them.
1633 		 */
1634 		*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
1635 		return 0;
1636 	}
1637 
1638 	sock = get_if_ioctl_socket();
1639 	if (sock == -1) {
1640 		pcapint_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE, errno,
1641 		    "Can't create socket to get ethtool information for %s",
1642 		    name);
1643 		return -1;
1644 	}
1645 
1646 	/*
1647 	 * OK, what type of network is this?
1648 	 * In particular, is it wired or wireless?
1649 	 */
1650 	if (is_wifi(name)) {
1651 		/*
1652 		 * Wi-Fi, hence wireless.
1653 		 */
1654 		*flags |= PCAP_IF_WIRELESS;
1655 	} else {
1656 		/*
1657 		 * OK, what does /sys/class/net/{if_name}/type contain?
1658 		 * (We don't use that for Wi-Fi, as it'll report
1659 		 * "Ethernet", i.e. ARPHRD_ETHER, for non-monitor-
1660 		 * mode devices.)
1661 		 */
1662 		char *pathstr;
1663 
1664 		if (asprintf(&pathstr, "/sys/class/net/%s/type", name) == -1) {
1665 			snprintf(errbuf, PCAP_ERRBUF_SIZE,
1666 			    "%s: Can't generate path name string for /sys/class/net device",
1667 			    name);
1668 			close(sock);
1669 			return -1;
1670 		}
1671 		fh = fopen(pathstr, "r");
1672 		if (fh != NULL) {
1673 			if (fscanf(fh, "%u", &arptype) == 1) {
1674 				/*
1675 				 * OK, we got an ARPHRD_ type; what is it?
1676 				 */
1677 				switch (arptype) {
1678 
1679 				case ARPHRD_LOOPBACK:
1680 					/*
1681 					 * These are types to which
1682 					 * "connected" and "disconnected"
1683 					 * don't apply, so don't bother
1684 					 * asking about it.
1685 					 *
1686 					 * XXX - add other types?
1687 					 */
1688 					close(sock);
1689 					fclose(fh);
1690 					free(pathstr);
1691 					return 0;
1692 
1693 				case ARPHRD_IRDA:
1694 				case ARPHRD_IEEE80211:
1695 				case ARPHRD_IEEE80211_PRISM:
1696 				case ARPHRD_IEEE80211_RADIOTAP:
1697 #ifdef ARPHRD_IEEE802154
1698 				case ARPHRD_IEEE802154:
1699 #endif
1700 #ifdef ARPHRD_IEEE802154_MONITOR
1701 				case ARPHRD_IEEE802154_MONITOR:
1702 #endif
1703 #ifdef ARPHRD_6LOWPAN
1704 				case ARPHRD_6LOWPAN:
1705 #endif
1706 					/*
1707 					 * Various wireless types.
1708 					 */
1709 					*flags |= PCAP_IF_WIRELESS;
1710 					break;
1711 				}
1712 			}
1713 			fclose(fh);
1714 		}
1715 		free(pathstr);
1716 	}
1717 
1718 #ifdef ETHTOOL_GLINK
1719 	memset(&ifr, 0, sizeof(ifr));
1720 	pcapint_strlcpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
1721 	info.cmd = ETHTOOL_GLINK;
1722 	/*
1723 	 * XXX - while Valgrind handles SIOCETHTOOL and knows that
1724 	 * the ETHTOOL_GLINK command sets the .data member of the
1725 	 * structure, Memory Sanitizer doesn't yet do so:
1726 	 *
1727 	 *    https://bugs.llvm.org/show_bug.cgi?id=45814
1728 	 *
1729 	 * For now, we zero it out to squelch warnings; if the bug
1730 	 * in question is fixed, we can remove this.
1731 	 */
1732 	info.data = 0;
1733 	ifr.ifr_data = (caddr_t)&info;
1734 	if (ioctl(sock, SIOCETHTOOL, &ifr) == -1) {
1735 		int save_errno = errno;
1736 
1737 		switch (save_errno) {
1738 
1739 		case EOPNOTSUPP:
1740 		case EINVAL:
1741 			/*
1742 			 * OK, this OS version or driver doesn't support
1743 			 * asking for this information.
1744 			 * XXX - distinguish between "this doesn't
1745 			 * support ethtool at all because it's not
1746 			 * that type of device" vs. "this doesn't
1747 			 * support ethtool even though it's that
1748 			 * type of device", and return "unknown".
1749 			 */
1750 			*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
1751 			close(sock);
1752 			return 0;
1753 
1754 		case ENODEV:
1755 			/*
1756 			 * OK, no such device.
1757 			 * The user will find that out when they try to
1758 			 * activate the device; just say "OK" and
1759 			 * don't set anything.
1760 			 */
1761 			close(sock);
1762 			return 0;
1763 
1764 		default:
1765 			/*
1766 			 * Other error.
1767 			 */
1768 			pcapint_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
1769 			    save_errno,
1770 			    "%s: SIOCETHTOOL(ETHTOOL_GLINK) ioctl failed",
1771 			    name);
1772 			close(sock);
1773 			return -1;
1774 		}
1775 	}
1776 
1777 	/*
1778 	 * Is it connected?
1779 	 */
1780 	if (info.data) {
1781 		/*
1782 		 * It's connected.
1783 		 */
1784 		*flags |= PCAP_IF_CONNECTION_STATUS_CONNECTED;
1785 	} else {
1786 		/*
1787 		 * It's disconnected.
1788 		 */
1789 		*flags |= PCAP_IF_CONNECTION_STATUS_DISCONNECTED;
1790 	}
1791 #endif
1792 
1793 	close(sock);
1794 	return 0;
1795 }
1796 
1797 int
1798 pcapint_platform_finddevs(pcap_if_list_t *devlistp, char *errbuf)
1799 {
1800 	/*
1801 	 * Get the list of regular interfaces first.
1802 	 */
1803 	if (pcapint_findalldevs_interfaces(devlistp, errbuf, can_be_bound,
1804 	    get_if_flags) == -1)
1805 		return (-1);	/* failure */
1806 
1807 	/*
1808 	 * Add the "any" device.
1809 	 */
1810 	if (pcap_add_any_dev(devlistp, errbuf) == NULL)
1811 		return (-1);
1812 
1813 	return (0);
1814 }
1815 
1816 /*
1817  * Set direction flag: Which packets do we accept on a forwarding
1818  * single device? IN, OUT or both?
1819  */
1820 static int
1821 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
1822 {
1823 	/*
1824 	 * It's guaranteed, at this point, that d is a valid
1825 	 * direction value.
1826 	 */
1827 	handle->direction = d;
1828 	return 0;
1829 }
1830 
1831 static int
1832 is_wifi(const char *device)
1833 {
1834 	char *pathstr;
1835 	struct stat statb;
1836 
1837 	/*
1838 	 * See if there's a sysfs wireless directory for it.
1839 	 * If so, it's a wireless interface.
1840 	 */
1841 	if (asprintf(&pathstr, "/sys/class/net/%s/wireless", device) == -1) {
1842 		/*
1843 		 * Just give up here.
1844 		 */
1845 		return 0;
1846 	}
1847 	if (stat(pathstr, &statb) == 0) {
1848 		free(pathstr);
1849 		return 1;
1850 	}
1851 	free(pathstr);
1852 
1853 	return 0;
1854 }
1855 
1856 /*
1857  *  Linux uses the ARP hardware type to identify the type of an
1858  *  interface. pcap uses the DLT_xxx constants for this. This
1859  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
1860  *  constant, as arguments, and sets "handle->linktype" to the
1861  *  appropriate DLT_XXX constant and sets "handle->offset" to
1862  *  the appropriate value (to make "handle->offset" plus link-layer
1863  *  header length be a multiple of 4, so that the link-layer payload
1864  *  will be aligned on a 4-byte boundary when capturing packets).
1865  *  (If the offset isn't set here, it'll be 0; add code as appropriate
1866  *  for cases where it shouldn't be 0.)
1867  *
1868  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
1869  *  in cooked mode; otherwise, we can't use cooked mode, so we have
1870  *  to pick some type that works in raw mode, or fail.
1871  *
1872  *  Sets the link type to -1 if unable to map the type.
1873  *
1874  *  Returns 0 on success or a PCAP_ERROR_ value on error.
1875  */
1876 static int map_arphrd_to_dlt(pcap_t *handle, int arptype,
1877 			     const char *device, int cooked_ok)
1878 {
1879 	static const char cdma_rmnet[] = "cdma_rmnet";
1880 
1881 	switch (arptype) {
1882 
1883 	case ARPHRD_ETHER:
1884 		/*
1885 		 * For various annoying reasons having to do with DHCP
1886 		 * software, some versions of Android give the mobile-
1887 		 * phone-network interface an ARPHRD_ value of
1888 		 * ARPHRD_ETHER, even though the packets supplied by
1889 		 * that interface have no link-layer header, and begin
1890 		 * with an IP header, so that the ARPHRD_ value should
1891 		 * be ARPHRD_NONE.
1892 		 *
1893 		 * Detect those devices by checking the device name, and
1894 		 * use DLT_RAW for them.
1895 		 */
1896 		if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) {
1897 			handle->linktype = DLT_RAW;
1898 			return 0;
1899 		}
1900 
1901 		/*
1902 		 * Is this a real Ethernet device?  If so, give it a
1903 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
1904 		 * that an application can let you choose it, in case you're
1905 		 * capturing DOCSIS traffic that a Cisco Cable Modem
1906 		 * Termination System is putting out onto an Ethernet (it
1907 		 * doesn't put an Ethernet header onto the wire, it puts raw
1908 		 * DOCSIS frames out on the wire inside the low-level
1909 		 * Ethernet framing).
1910 		 *
1911 		 * XXX - are there any other sorts of "fake Ethernet" that
1912 		 * have ARPHRD_ETHER but that shouldn't offer DLT_DOCSIS as
1913 		 * a Cisco CMTS won't put traffic onto it or get traffic
1914 		 * bridged onto it?  ISDN is handled in "setup_socket()",
1915 		 * as we fall back on cooked mode there, and we use
1916 		 * is_wifi() to check for 802.11 devices; are there any
1917 		 * others?
1918 		 */
1919 		if (!is_wifi(device)) {
1920 			int ret;
1921 
1922 			/*
1923 			 * This is not a Wi-Fi device but it could be
1924 			 * a DSA master/management network device.
1925 			 */
1926 			ret = iface_dsa_get_proto_info(device, handle);
1927 			if (ret < 0)
1928 				return ret;
1929 
1930 			if (ret == 1) {
1931 				/*
1932 				 * This is a DSA master/management network
1933 				 * device linktype is already set by
1934 				 * iface_dsa_get_proto_info() set an
1935 				 * appropriate offset here.
1936 				 */
1937 				handle->offset = 2;
1938 				break;
1939 			}
1940 
1941 			/*
1942 			 * It's not a Wi-Fi device; offer DOCSIS.
1943 			 */
1944 			handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
1945 			if (handle->dlt_list == NULL) {
1946 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
1947 				    PCAP_ERRBUF_SIZE, errno, "malloc");
1948 				return (PCAP_ERROR);
1949 			}
1950 			handle->dlt_list[0] = DLT_EN10MB;
1951 			handle->dlt_list[1] = DLT_DOCSIS;
1952 			handle->dlt_count = 2;
1953 		}
1954 		/* FALLTHROUGH */
1955 
1956 	case ARPHRD_METRICOM:
1957 	case ARPHRD_LOOPBACK:
1958 		handle->linktype = DLT_EN10MB;
1959 		handle->offset = 2;
1960 		break;
1961 
1962 	case ARPHRD_EETHER:
1963 		handle->linktype = DLT_EN3MB;
1964 		break;
1965 
1966 	case ARPHRD_AX25:
1967 		handle->linktype = DLT_AX25_KISS;
1968 		break;
1969 
1970 	case ARPHRD_PRONET:
1971 		handle->linktype = DLT_PRONET;
1972 		break;
1973 
1974 	case ARPHRD_CHAOS:
1975 		handle->linktype = DLT_CHAOS;
1976 		break;
1977 #ifndef ARPHRD_CAN
1978 #define ARPHRD_CAN 280
1979 #endif
1980 	case ARPHRD_CAN:
1981 		handle->linktype = DLT_CAN_SOCKETCAN;
1982 		break;
1983 
1984 #ifndef ARPHRD_IEEE802_TR
1985 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
1986 #endif
1987 	case ARPHRD_IEEE802_TR:
1988 	case ARPHRD_IEEE802:
1989 		handle->linktype = DLT_IEEE802;
1990 		handle->offset = 2;
1991 		break;
1992 
1993 	case ARPHRD_ARCNET:
1994 		handle->linktype = DLT_ARCNET_LINUX;
1995 		break;
1996 
1997 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
1998 #define ARPHRD_FDDI	774
1999 #endif
2000 	case ARPHRD_FDDI:
2001 		handle->linktype = DLT_FDDI;
2002 		handle->offset = 3;
2003 		break;
2004 
2005 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
2006 #define ARPHRD_ATM 19
2007 #endif
2008 	case ARPHRD_ATM:
2009 		/*
2010 		 * The Classical IP implementation in ATM for Linux
2011 		 * supports both what RFC 1483 calls "LLC Encapsulation",
2012 		 * in which each packet has an LLC header, possibly
2013 		 * with a SNAP header as well, prepended to it, and
2014 		 * what RFC 1483 calls "VC Based Multiplexing", in which
2015 		 * different virtual circuits carry different network
2016 		 * layer protocols, and no header is prepended to packets.
2017 		 *
2018 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
2019 		 * you can't use the ARPHRD_ type to find out whether
2020 		 * captured packets will have an LLC header, and,
2021 		 * while there's a socket ioctl to *set* the encapsulation
2022 		 * type, there's no ioctl to *get* the encapsulation type.
2023 		 *
2024 		 * This means that
2025 		 *
2026 		 *	programs that dissect Linux Classical IP frames
2027 		 *	would have to check for an LLC header and,
2028 		 *	depending on whether they see one or not, dissect
2029 		 *	the frame as LLC-encapsulated or as raw IP (I
2030 		 *	don't know whether there's any traffic other than
2031 		 *	IP that would show up on the socket, or whether
2032 		 *	there's any support for IPv6 in the Linux
2033 		 *	Classical IP code);
2034 		 *
2035 		 *	filter expressions would have to compile into
2036 		 *	code that checks for an LLC header and does
2037 		 *	the right thing.
2038 		 *
2039 		 * Both of those are a nuisance - and, at least on systems
2040 		 * that support PF_PACKET sockets, we don't have to put
2041 		 * up with those nuisances; instead, we can just capture
2042 		 * in cooked mode.  That's what we'll do, if we can.
2043 		 * Otherwise, we'll just fail.
2044 		 */
2045 		if (cooked_ok)
2046 			handle->linktype = DLT_LINUX_SLL;
2047 		else
2048 			handle->linktype = -1;
2049 		break;
2050 
2051 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2052 #define ARPHRD_IEEE80211 801
2053 #endif
2054 	case ARPHRD_IEEE80211:
2055 		handle->linktype = DLT_IEEE802_11;
2056 		break;
2057 
2058 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2059 #define ARPHRD_IEEE80211_PRISM 802
2060 #endif
2061 	case ARPHRD_IEEE80211_PRISM:
2062 		handle->linktype = DLT_PRISM_HEADER;
2063 		break;
2064 
2065 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2066 #define ARPHRD_IEEE80211_RADIOTAP 803
2067 #endif
2068 	case ARPHRD_IEEE80211_RADIOTAP:
2069 		handle->linktype = DLT_IEEE802_11_RADIO;
2070 		break;
2071 
2072 	case ARPHRD_PPP:
2073 		/*
2074 		 * Some PPP code in the kernel supplies no link-layer
2075 		 * header whatsoever to PF_PACKET sockets; other PPP
2076 		 * code supplies PPP link-layer headers ("syncppp.c");
2077 		 * some PPP code might supply random link-layer
2078 		 * headers (PPP over ISDN - there's code in Ethereal,
2079 		 * for example, to cope with PPP-over-ISDN captures
2080 		 * with which the Ethereal developers have had to cope,
2081 		 * heuristically trying to determine which of the
2082 		 * oddball link-layer headers particular packets have).
2083 		 *
2084 		 * As such, we just punt, and run all PPP interfaces
2085 		 * in cooked mode, if we can; otherwise, we just treat
2086 		 * it as DLT_RAW, for now - if somebody needs to capture,
2087 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2088 		 * link-layer header, they'll have to add code here to
2089 		 * map to the appropriate DLT_ type (possibly adding a
2090 		 * new DLT_ type, if necessary).
2091 		 */
2092 		if (cooked_ok)
2093 			handle->linktype = DLT_LINUX_SLL;
2094 		else {
2095 			/*
2096 			 * XXX - handle ISDN types here?  We can't fall
2097 			 * back on cooked sockets, so we'd have to
2098 			 * figure out from the device name what type of
2099 			 * link-layer encapsulation it's using, and map
2100 			 * that to an appropriate DLT_ value, meaning
2101 			 * we'd map "isdnN" devices to DLT_RAW (they
2102 			 * supply raw IP packets with no link-layer
2103 			 * header) and "isdY" devices to a new DLT_I4L_IP
2104 			 * type that has only an Ethernet packet type as
2105 			 * a link-layer header.
2106 			 *
2107 			 * But sometimes we seem to get random crap
2108 			 * in the link-layer header when capturing on
2109 			 * ISDN devices....
2110 			 */
2111 			handle->linktype = DLT_RAW;
2112 		}
2113 		break;
2114 
2115 #ifndef ARPHRD_CISCO
2116 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2117 #endif
2118 	case ARPHRD_CISCO:
2119 		handle->linktype = DLT_C_HDLC;
2120 		break;
2121 
2122 	/* Not sure if this is correct for all tunnels, but it
2123 	 * works for CIPE */
2124 	case ARPHRD_TUNNEL:
2125 #ifndef ARPHRD_SIT
2126 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2127 #endif
2128 	case ARPHRD_SIT:
2129 	case ARPHRD_CSLIP:
2130 	case ARPHRD_SLIP6:
2131 	case ARPHRD_CSLIP6:
2132 	case ARPHRD_ADAPT:
2133 	case ARPHRD_SLIP:
2134 #ifndef ARPHRD_RAWHDLC
2135 #define ARPHRD_RAWHDLC 518
2136 #endif
2137 	case ARPHRD_RAWHDLC:
2138 #ifndef ARPHRD_DLCI
2139 #define ARPHRD_DLCI 15
2140 #endif
2141 	case ARPHRD_DLCI:
2142 		/*
2143 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2144 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2145 		 */
2146 		handle->linktype = DLT_RAW;
2147 		break;
2148 
2149 #ifndef ARPHRD_FRAD
2150 #define ARPHRD_FRAD 770
2151 #endif
2152 	case ARPHRD_FRAD:
2153 		handle->linktype = DLT_FRELAY;
2154 		break;
2155 
2156 	case ARPHRD_LOCALTLK:
2157 		handle->linktype = DLT_LTALK;
2158 		break;
2159 
2160 	case 18:
2161 		/*
2162 		 * RFC 4338 defines an encapsulation for IP and ARP
2163 		 * packets that's compatible with the RFC 2625
2164 		 * encapsulation, but that uses a different ARP
2165 		 * hardware type and hardware addresses.  That
2166 		 * ARP hardware type is 18; Linux doesn't define
2167 		 * any ARPHRD_ value as 18, but if it ever officially
2168 		 * supports RFC 4338-style IP-over-FC, it should define
2169 		 * one.
2170 		 *
2171 		 * For now, we map it to DLT_IP_OVER_FC, in the hopes
2172 		 * that this will encourage its use in the future,
2173 		 * should Linux ever officially support RFC 4338-style
2174 		 * IP-over-FC.
2175 		 */
2176 		handle->linktype = DLT_IP_OVER_FC;
2177 		break;
2178 
2179 #ifndef ARPHRD_FCPP
2180 #define ARPHRD_FCPP	784
2181 #endif
2182 	case ARPHRD_FCPP:
2183 #ifndef ARPHRD_FCAL
2184 #define ARPHRD_FCAL	785
2185 #endif
2186 	case ARPHRD_FCAL:
2187 #ifndef ARPHRD_FCPL
2188 #define ARPHRD_FCPL	786
2189 #endif
2190 	case ARPHRD_FCPL:
2191 #ifndef ARPHRD_FCFABRIC
2192 #define ARPHRD_FCFABRIC	787
2193 #endif
2194 	case ARPHRD_FCFABRIC:
2195 		/*
2196 		 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
2197 		 * IP-over-FC:
2198 		 *
2199 		 *	https://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
2200 		 *
2201 		 * and one was assigned.
2202 		 *
2203 		 * In a later private discussion (spun off from a message
2204 		 * on the ethereal-users list) on how to get that DLT_
2205 		 * value in libpcap on Linux, I ended up deciding that
2206 		 * the best thing to do would be to have him tweak the
2207 		 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
2208 		 * type, and map all those types to DLT_IP_OVER_FC:
2209 		 *
2210 		 *	I've checked into the libpcap and tcpdump CVS tree
2211 		 *	support for DLT_IP_OVER_FC.  In order to use that,
2212 		 *	you'd have to modify your modified driver to return
2213 		 *	one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
2214 		 *	change it to set "dev->type" to ARPHRD_FCFABRIC, for
2215 		 *	example (the exact value doesn't matter, it can be
2216 		 *	any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
2217 		 *	ARPHRD_FCFABRIC).
2218 		 *
2219 		 * 11 years later, Christian Svensson wanted to map
2220 		 * various ARPHRD_ values to DLT_FC_2 and
2221 		 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
2222 		 * frames:
2223 		 *
2224 		 *	https://github.com/mcr/libpcap/pull/29
2225 		 *
2226 		 * There doesn't seem to be any network drivers that uses
2227 		 * any of the ARPHRD_FC* values for IP-over-FC, and
2228 		 * it's not exactly clear what the "Dummy types for non
2229 		 * ARP hardware" are supposed to mean (link-layer
2230 		 * header type?  Physical network type?), so it's
2231 		 * not exactly clear why the ARPHRD_FC* types exist
2232 		 * in the first place.
2233 		 *
2234 		 * For now, we map them to DLT_FC_2, and provide an
2235 		 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
2236 		 * DLT_IP_OVER_FC just in case there's some old
2237 		 * driver out there that uses one of those types for
2238 		 * IP-over-FC on which somebody wants to capture
2239 		 * packets.
2240 		 */
2241 		handle->linktype = DLT_FC_2;
2242 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 3);
2243 		if (handle->dlt_list == NULL) {
2244 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
2245 			    PCAP_ERRBUF_SIZE, errno, "malloc");
2246 			return (PCAP_ERROR);
2247 		}
2248 		handle->dlt_list[0] = DLT_FC_2;
2249 		handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
2250 		handle->dlt_list[2] = DLT_IP_OVER_FC;
2251 		handle->dlt_count = 3;
2252 		break;
2253 
2254 #ifndef ARPHRD_IRDA
2255 #define ARPHRD_IRDA	783
2256 #endif
2257 	case ARPHRD_IRDA:
2258 		/* Don't expect IP packet out of this interfaces... */
2259 		handle->linktype = DLT_LINUX_IRDA;
2260 		/* We need to save packet direction for IrDA decoding,
2261 		 * so let's use "Linux-cooked" mode. Jean II
2262 		 *
2263 		 * XXX - this is handled in setup_socket(). */
2264 		/* handlep->cooked = 1; */
2265 		break;
2266 
2267 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2268 	 * is needed, please report it to <daniele@orlandi.com> */
2269 #ifndef ARPHRD_LAPD
2270 #define ARPHRD_LAPD	8445
2271 #endif
2272 	case ARPHRD_LAPD:
2273 		/* Don't expect IP packet out of this interfaces... */
2274 		handle->linktype = DLT_LINUX_LAPD;
2275 		break;
2276 
2277 #ifndef ARPHRD_NONE
2278 #define ARPHRD_NONE	0xFFFE
2279 #endif
2280 	case ARPHRD_NONE:
2281 		/*
2282 		 * No link-layer header; packets are just IP
2283 		 * packets, so use DLT_RAW.
2284 		 */
2285 		handle->linktype = DLT_RAW;
2286 		break;
2287 
2288 #ifndef ARPHRD_IEEE802154
2289 #define ARPHRD_IEEE802154      804
2290 #endif
2291        case ARPHRD_IEEE802154:
2292                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2293                break;
2294 
2295 #ifndef ARPHRD_NETLINK
2296 #define ARPHRD_NETLINK	824
2297 #endif
2298 	case ARPHRD_NETLINK:
2299 		handle->linktype = DLT_NETLINK;
2300 		/*
2301 		 * We need to use cooked mode, so that in sll_protocol we
2302 		 * pick up the netlink protocol type such as NETLINK_ROUTE,
2303 		 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc.
2304 		 *
2305 		 * XXX - this is handled in setup_socket().
2306 		 */
2307 		/* handlep->cooked = 1; */
2308 		break;
2309 
2310 #ifndef ARPHRD_VSOCKMON
2311 #define ARPHRD_VSOCKMON	826
2312 #endif
2313 	case ARPHRD_VSOCKMON:
2314 		handle->linktype = DLT_VSOCK;
2315 		break;
2316 
2317 	default:
2318 		handle->linktype = -1;
2319 		break;
2320 	}
2321 	return (0);
2322 }
2323 
2324 /*
2325  * Try to set up a PF_PACKET socket.
2326  * Returns 0 or a PCAP_WARNING_ value on success and a PCAP_ERROR_ value
2327  * on failure.
2328  */
2329 static int
2330 setup_socket(pcap_t *handle, int is_any_device)
2331 {
2332 	struct pcap_linux *handlep = handle->priv;
2333 	const char		*device = handle->opt.device;
2334 	int			status = 0;
2335 	int			sock_fd, arptype;
2336 	int			val;
2337 	int			err = 0;
2338 	struct packet_mreq	mr;
2339 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
2340 	int			bpf_extensions;
2341 	socklen_t		len = sizeof(bpf_extensions);
2342 #endif
2343 
2344 	/*
2345 	 * Open a socket with protocol family packet. If cooked is true,
2346 	 * we open a SOCK_DGRAM socket for the cooked interface, otherwise
2347 	 * we open a SOCK_RAW socket for the raw interface.
2348 	 *
2349 	 * The protocol is set to 0.  This means we will receive no
2350 	 * packets until we "bind" the socket with a non-zero
2351 	 * protocol.  This allows us to setup the ring buffers without
2352 	 * dropping any packets.
2353 	 */
2354 	sock_fd = is_any_device ?
2355 		socket(PF_PACKET, SOCK_DGRAM, 0) :
2356 		socket(PF_PACKET, SOCK_RAW, 0);
2357 
2358 	if (sock_fd == -1) {
2359 		if (errno == EPERM || errno == EACCES) {
2360 			/*
2361 			 * You don't have permission to open the
2362 			 * socket.
2363 			 */
2364 			status = PCAP_ERROR_PERM_DENIED;
2365 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2366 			    "Attempt to create packet socket failed - CAP_NET_RAW may be required");
2367 		} else if (errno == EAFNOSUPPORT) {
2368 			/*
2369 			 * PF_PACKET sockets not supported.
2370 			 * Perhaps we're running on the WSL1 module
2371 			 * in the Windows NT kernel rather than on
2372 			 * a real Linux kernel.
2373 			 */
2374 			status = PCAP_ERROR_CAPTURE_NOTSUP;
2375 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2376 			    "PF_PACKET sockets not supported - is this WSL1?");
2377 		} else {
2378 			/*
2379 			 * Other error.
2380 			 */
2381 			status = PCAP_ERROR;
2382 		}
2383 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2384 		    errno, "socket");
2385 		return status;
2386 	}
2387 
2388 	/*
2389 	 * Get the interface index of the loopback device.
2390 	 * If the attempt fails, don't fail, just set the
2391 	 * "handlep->lo_ifindex" to -1.
2392 	 *
2393 	 * XXX - can there be more than one device that loops
2394 	 * packets back, i.e. devices other than "lo"?  If so,
2395 	 * we'd need to find them all, and have an array of
2396 	 * indices for them, and check all of them in
2397 	 * "pcap_read_packet()".
2398 	 */
2399 	handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
2400 
2401 	/*
2402 	 * Default value for offset to align link-layer payload
2403 	 * on a 4-byte boundary.
2404 	 */
2405 	handle->offset	 = 0;
2406 
2407 	/*
2408 	 * What kind of frames do we have to deal with? Fall back
2409 	 * to cooked mode if we have an unknown interface type
2410 	 * or a type we know doesn't work well in raw mode.
2411 	 */
2412 	if (!is_any_device) {
2413 		/* Assume for now we don't need cooked mode. */
2414 		handlep->cooked = 0;
2415 
2416 		if (handle->opt.rfmon) {
2417 			/*
2418 			 * We were asked to turn on monitor mode.
2419 			 * Do so before we get the link-layer type,
2420 			 * because entering monitor mode could change
2421 			 * the link-layer type.
2422 			 */
2423 			err = enter_rfmon_mode(handle, sock_fd, device);
2424 			if (err < 0) {
2425 				/* Hard failure */
2426 				close(sock_fd);
2427 				return err;
2428 			}
2429 			if (err == 0) {
2430 				/*
2431 				 * Nothing worked for turning monitor mode
2432 				 * on.
2433 				 */
2434 				close(sock_fd);
2435 
2436 				return PCAP_ERROR_RFMON_NOTSUP;
2437 			}
2438 
2439 			/*
2440 			 * Either monitor mode has been turned on for
2441 			 * the device, or we've been given a different
2442 			 * device to open for monitor mode.  If we've
2443 			 * been given a different device, use it.
2444 			 */
2445 			if (handlep->mondevice != NULL)
2446 				device = handlep->mondevice;
2447 		}
2448 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
2449 		if (arptype < 0) {
2450 			close(sock_fd);
2451 			return arptype;
2452 		}
2453 		status = map_arphrd_to_dlt(handle, arptype, device, 1);
2454 		if (status < 0) {
2455 			close(sock_fd);
2456 			return status;
2457 		}
2458 		if (handle->linktype == -1 ||
2459 		    handle->linktype == DLT_LINUX_SLL ||
2460 		    handle->linktype == DLT_LINUX_IRDA ||
2461 		    handle->linktype == DLT_LINUX_LAPD ||
2462 		    handle->linktype == DLT_NETLINK ||
2463 		    (handle->linktype == DLT_EN10MB &&
2464 		     (strncmp("isdn", device, 4) == 0 ||
2465 		      strncmp("isdY", device, 4) == 0))) {
2466 			/*
2467 			 * Unknown interface type (-1), or a
2468 			 * device we explicitly chose to run
2469 			 * in cooked mode (e.g., PPP devices),
2470 			 * or an ISDN device (whose link-layer
2471 			 * type we can only determine by using
2472 			 * APIs that may be different on different
2473 			 * kernels) - reopen in cooked mode.
2474 			 *
2475 			 * If the type is unknown, return a warning;
2476 			 * map_arphrd_to_dlt() has already set the
2477 			 * warning message.
2478 			 */
2479 			if (close(sock_fd) == -1) {
2480 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
2481 				    PCAP_ERRBUF_SIZE, errno, "close");
2482 				return PCAP_ERROR;
2483 			}
2484 			sock_fd = socket(PF_PACKET, SOCK_DGRAM, 0);
2485 			if (sock_fd < 0) {
2486 				/*
2487 				 * Fatal error.  We treat this as
2488 				 * a generic error; we already know
2489 				 * that we were able to open a
2490 				 * PF_PACKET/SOCK_RAW socket, so
2491 				 * any failure is a "this shouldn't
2492 				 * happen" case.
2493 				 */
2494 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
2495 				    PCAP_ERRBUF_SIZE, errno, "socket");
2496 				return PCAP_ERROR;
2497 			}
2498 			handlep->cooked = 1;
2499 
2500 			/*
2501 			 * Get rid of any link-layer type list
2502 			 * we allocated - this only supports cooked
2503 			 * capture.
2504 			 */
2505 			if (handle->dlt_list != NULL) {
2506 				free(handle->dlt_list);
2507 				handle->dlt_list = NULL;
2508 				handle->dlt_count = 0;
2509 			}
2510 
2511 			if (handle->linktype == -1) {
2512 				/*
2513 				 * Warn that we're falling back on
2514 				 * cooked mode; we may want to
2515 				 * update "map_arphrd_to_dlt()"
2516 				 * to handle the new type.
2517 				 */
2518 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2519 					"arptype %d not "
2520 					"supported by libpcap - "
2521 					"falling back to cooked "
2522 					"socket",
2523 					arptype);
2524 				status = PCAP_WARNING;
2525 			}
2526 
2527 			/*
2528 			 * IrDA capture is not a real "cooked" capture,
2529 			 * it's IrLAP frames, not IP packets.  The
2530 			 * same applies to LAPD capture.
2531 			 */
2532 			if (handle->linktype != DLT_LINUX_IRDA &&
2533 			    handle->linktype != DLT_LINUX_LAPD &&
2534 			    handle->linktype != DLT_NETLINK)
2535 				handle->linktype = DLT_LINUX_SLL;
2536 		}
2537 
2538 		handlep->ifindex = iface_get_id(sock_fd, device,
2539 		    handle->errbuf);
2540 		if (handlep->ifindex == -1) {
2541 			close(sock_fd);
2542 			return PCAP_ERROR;
2543 		}
2544 
2545 		if ((err = iface_bind(sock_fd, handlep->ifindex,
2546 		    handle->errbuf, 0)) != 0) {
2547 			close(sock_fd);
2548 			return err;
2549 		}
2550 	} else {
2551 		/*
2552 		 * The "any" device.
2553 		 */
2554 		if (handle->opt.rfmon) {
2555 			/*
2556 			 * It doesn't support monitor mode.
2557 			 */
2558 			close(sock_fd);
2559 			return PCAP_ERROR_RFMON_NOTSUP;
2560 		}
2561 
2562 		/*
2563 		 * It uses cooked mode.
2564 		 * Support both DLT_LINUX_SLL and DLT_LINUX_SLL2.
2565 		 */
2566 		handlep->cooked = 1;
2567 		handle->linktype = DLT_LINUX_SLL;
2568 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2569 		if (handle->dlt_list == NULL) {
2570 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
2571 			    PCAP_ERRBUF_SIZE, errno, "malloc");
2572 			return (PCAP_ERROR);
2573 		}
2574 		handle->dlt_list[0] = DLT_LINUX_SLL;
2575 		handle->dlt_list[1] = DLT_LINUX_SLL2;
2576 		handle->dlt_count = 2;
2577 
2578 		/*
2579 		 * We're not bound to a device.
2580 		 * For now, we're using this as an indication
2581 		 * that we can't transmit; stop doing that only
2582 		 * if we figure out how to transmit in cooked
2583 		 * mode.
2584 		 */
2585 		handlep->ifindex = -1;
2586 	}
2587 
2588 	/*
2589 	 * Select promiscuous mode on if "promisc" is set.
2590 	 *
2591 	 * Do not turn allmulti mode on if we don't select
2592 	 * promiscuous mode - on some devices (e.g., Orinoco
2593 	 * wireless interfaces), allmulti mode isn't supported
2594 	 * and the driver implements it by turning promiscuous
2595 	 * mode on, and that screws up the operation of the
2596 	 * card as a normal networking interface, and on no
2597 	 * other platform I know of does starting a non-
2598 	 * promiscuous capture affect which multicast packets
2599 	 * are received by the interface.
2600 	 */
2601 
2602 	/*
2603 	 * Hmm, how can we set promiscuous mode on all interfaces?
2604 	 * I am not sure if that is possible at all.  For now, we
2605 	 * silently ignore attempts to turn promiscuous mode on
2606 	 * for the "any" device (so you don't have to explicitly
2607 	 * disable it in programs such as tcpdump).
2608 	 */
2609 
2610 	if (!is_any_device && handle->opt.promisc) {
2611 		memset(&mr, 0, sizeof(mr));
2612 		mr.mr_ifindex = handlep->ifindex;
2613 		mr.mr_type    = PACKET_MR_PROMISC;
2614 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
2615 		    &mr, sizeof(mr)) == -1) {
2616 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
2617 			    PCAP_ERRBUF_SIZE, errno, "setsockopt (PACKET_ADD_MEMBERSHIP)");
2618 			close(sock_fd);
2619 			return PCAP_ERROR;
2620 		}
2621 	}
2622 
2623 	/*
2624 	 * Enable auxiliary data and reserve room for reconstructing
2625 	 * VLAN headers.
2626 	 *
2627 	 * XXX - is enabling auxiliary data necessary, now that we
2628 	 * only support memory-mapped capture?  The kernel's memory-mapped
2629 	 * capture code doesn't seem to check whether auxiliary data
2630 	 * is enabled, it seems to provide it whether it is or not.
2631 	 */
2632 	val = 1;
2633 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
2634 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
2635 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2636 		    errno, "setsockopt (PACKET_AUXDATA)");
2637 		close(sock_fd);
2638 		return PCAP_ERROR;
2639 	}
2640 	handle->offset += VLAN_TAG_LEN;
2641 
2642 	/*
2643 	 * If we're in cooked mode, make the snapshot length
2644 	 * large enough to hold a "cooked mode" header plus
2645 	 * 1 byte of packet data (so we don't pass a byte
2646 	 * count of 0 to "recvfrom()").
2647 	 * XXX - we don't know whether this will be DLT_LINUX_SLL
2648 	 * or DLT_LINUX_SLL2, so make sure it's big enough for
2649 	 * a DLT_LINUX_SLL2 "cooked mode" header; a snapshot length
2650 	 * that small is silly anyway.
2651 	 */
2652 	if (handlep->cooked) {
2653 		if (handle->snapshot < SLL2_HDR_LEN + 1)
2654 			handle->snapshot = SLL2_HDR_LEN + 1;
2655 	}
2656 	handle->bufsize = handle->snapshot;
2657 
2658 	/*
2659 	 * Set the offset at which to insert VLAN tags.
2660 	 */
2661 	set_vlan_offset(handle);
2662 
2663 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
2664 		int nsec_tstamps = 1;
2665 
2666 		if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
2667 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
2668 			close(sock_fd);
2669 			return PCAP_ERROR;
2670 		}
2671 	}
2672 
2673 	/*
2674 	 * We've succeeded. Save the socket FD in the pcap structure.
2675 	 */
2676 	handle->fd = sock_fd;
2677 
2678 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
2679 	/*
2680 	 * Can we generate special code for VLAN checks?
2681 	 * (XXX - what if we need the special code but it's not supported
2682 	 * by the OS?  Is that possible?)
2683 	 */
2684 	if (getsockopt(sock_fd, SOL_SOCKET, SO_BPF_EXTENSIONS,
2685 	    &bpf_extensions, &len) == 0) {
2686 		if (bpf_extensions >= SKF_AD_VLAN_TAG_PRESENT) {
2687 			/*
2688 			 * Yes, we can.  Request that we do so.
2689 			 */
2690 			handle->bpf_codegen_flags |= BPF_SPECIAL_VLAN_HANDLING;
2691 		}
2692 	}
2693 #endif /* defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT) */
2694 
2695 	return status;
2696 }
2697 
2698 /*
2699  * Attempt to setup memory-mapped access.
2700  *
2701  * On success, returns 0 if there are no warnings or a PCAP_WARNING_ code
2702  * if there is a warning.
2703  *
2704  * On error, returns the appropriate error code; if that is PCAP_ERROR,
2705  * sets handle->errbuf to the appropriate message.
2706  */
2707 static int
2708 setup_mmapped(pcap_t *handle)
2709 {
2710 	struct pcap_linux *handlep = handle->priv;
2711 	int status;
2712 
2713 	/*
2714 	 * Attempt to allocate a buffer to hold the contents of one
2715 	 * packet, for use by the oneshot callback.
2716 	 */
2717 	handlep->oneshot_buffer = malloc(handle->snapshot);
2718 	if (handlep->oneshot_buffer == NULL) {
2719 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2720 		    errno, "can't allocate oneshot buffer");
2721 		return PCAP_ERROR;
2722 	}
2723 
2724 	if (handle->opt.buffer_size == 0) {
2725 		/* by default request 2M for the ring buffer */
2726 		handle->opt.buffer_size = 2*1024*1024;
2727 	}
2728 	status = prepare_tpacket_socket(handle);
2729 	if (status == -1) {
2730 		free(handlep->oneshot_buffer);
2731 		handlep->oneshot_buffer = NULL;
2732 		return PCAP_ERROR;
2733 	}
2734 	status = create_ring(handle);
2735 	if (status < 0) {
2736 		/*
2737 		 * Error attempting to enable memory-mapped capture;
2738 		 * fail.  The return value is the status to return.
2739 		 */
2740 		free(handlep->oneshot_buffer);
2741 		handlep->oneshot_buffer = NULL;
2742 		return status;
2743 	}
2744 
2745 	/*
2746 	 * Success.  status has been set either to 0 if there are no
2747 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
2748 	 *
2749 	 * handle->offset is used to get the current position into the rx ring.
2750 	 * handle->cc is used to store the ring size.
2751 	 */
2752 
2753 	/*
2754 	 * Set the timeout to use in poll() before returning.
2755 	 */
2756 	set_poll_timeout(handlep);
2757 
2758 	return status;
2759 }
2760 
2761 /*
2762  * Attempt to set the socket to the specified version of the memory-mapped
2763  * header.
2764  *
2765  * Return 0 if we succeed; return 1 if we fail because that version isn't
2766  * supported; return -1 on any other error, and set handle->errbuf.
2767  */
2768 static int
2769 init_tpacket(pcap_t *handle, int version, const char *version_str)
2770 {
2771 	struct pcap_linux *handlep = handle->priv;
2772 	int val = version;
2773 	socklen_t len = sizeof(val);
2774 
2775 	/*
2776 	 * Probe whether kernel supports the specified TPACKET version;
2777 	 * this also gets the length of the header for that version.
2778 	 *
2779 	 * This socket option was introduced in 2.6.27, which was
2780 	 * also the first release with TPACKET_V2 support.
2781 	 */
2782 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
2783 		if (errno == EINVAL) {
2784 			/*
2785 			 * EINVAL means this specific version of TPACKET
2786 			 * is not supported. Tell the caller they can try
2787 			 * with a different one; if they've run out of
2788 			 * others to try, let them set the error message
2789 			 * appropriately.
2790 			 */
2791 			return 1;
2792 		}
2793 
2794 		/*
2795 		 * All other errors are fatal.
2796 		 */
2797 		if (errno == ENOPROTOOPT) {
2798 			/*
2799 			 * PACKET_HDRLEN isn't supported, which means
2800 			 * that memory-mapped capture isn't supported.
2801 			 * Indicate that in the message.
2802 			 */
2803 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2804 			    "Kernel doesn't support memory-mapped capture; a 2.6.27 or later 2.x kernel is required, with CONFIG_PACKET_MMAP specified for 2.x kernels");
2805 		} else {
2806 			/*
2807 			 * Some unexpected error.
2808 			 */
2809 			pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2810 			    errno, "can't get %s header len on packet socket",
2811 			    version_str);
2812 		}
2813 		return -1;
2814 	}
2815 	handlep->tp_hdrlen = val;
2816 
2817 	val = version;
2818 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
2819 			   sizeof(val)) < 0) {
2820 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2821 		    errno, "can't activate %s on packet socket", version_str);
2822 		return -1;
2823 	}
2824 	handlep->tp_version = version;
2825 
2826 	return 0;
2827 }
2828 
2829 /*
2830  * Attempt to set the socket to version 3 of the memory-mapped header and,
2831  * if that fails because version 3 isn't supported, attempt to fall
2832  * back to version 2.  If version 2 isn't supported, just fail.
2833  *
2834  * Return 0 if we succeed and -1 on any other error, and set handle->errbuf.
2835  */
2836 static int
2837 prepare_tpacket_socket(pcap_t *handle)
2838 {
2839 	int ret;
2840 
2841 #ifdef HAVE_TPACKET3
2842 	/*
2843 	 * Try setting the version to TPACKET_V3.
2844 	 *
2845 	 * The only mode in which buffering is done on PF_PACKET
2846 	 * sockets, so that packets might not be delivered
2847 	 * immediately, is TPACKET_V3 mode.
2848 	 *
2849 	 * The buffering cannot be disabled in that mode, so
2850 	 * if the user has requested immediate mode, we don't
2851 	 * use TPACKET_V3.
2852 	 */
2853 	if (!handle->opt.immediate) {
2854 		ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
2855 		if (ret == 0) {
2856 			/*
2857 			 * Success.
2858 			 */
2859 			return 0;
2860 		}
2861 		if (ret == -1) {
2862 			/*
2863 			 * We failed for some reason other than "the
2864 			 * kernel doesn't support TPACKET_V3".
2865 			 */
2866 			return -1;
2867 		}
2868 
2869 		/*
2870 		 * This means it returned 1, which means "the kernel
2871 		 * doesn't support TPACKET_V3"; try TPACKET_V2.
2872 		 */
2873 	}
2874 #endif /* HAVE_TPACKET3 */
2875 
2876 	/*
2877 	 * Try setting the version to TPACKET_V2.
2878 	 */
2879 	ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
2880 	if (ret == 0) {
2881 		/*
2882 		 * Success.
2883 		 */
2884 		return 0;
2885 	}
2886 
2887 	if (ret == 1) {
2888 		/*
2889 		 * OK, the kernel supports memory-mapped capture, but
2890 		 * not TPACKET_V2.  Set the error message appropriately.
2891 		 */
2892 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2893 		    "Kernel doesn't support TPACKET_V2; a 2.6.27 or later kernel is required");
2894 	}
2895 
2896 	/*
2897 	 * We failed.
2898 	 */
2899 	return -1;
2900 }
2901 
2902 #define MAX(a,b) ((a)>(b)?(a):(b))
2903 
2904 /*
2905  * Attempt to set up memory-mapped access.
2906  *
2907  * On success, returns 0 if there are no warnings or to a PCAP_WARNING_ code
2908  * if there is a warning.
2909  *
2910  * On error, returns the appropriate error code; if that is PCAP_ERROR,
2911  * sets handle->errbuf to the appropriate message.
2912  */
2913 static int
2914 create_ring(pcap_t *handle)
2915 {
2916 	struct pcap_linux *handlep = handle->priv;
2917 	unsigned i, j, frames_per_block;
2918 #ifdef HAVE_TPACKET3
2919 	/*
2920 	 * For sockets using TPACKET_V2, the extra stuff at the end of a
2921 	 * struct tpacket_req3 will be ignored, so this is OK even for
2922 	 * those sockets.
2923 	 */
2924 	struct tpacket_req3 req;
2925 #else
2926 	struct tpacket_req req;
2927 #endif
2928 	socklen_t len;
2929 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
2930 	unsigned int frame_size;
2931 	int status;
2932 
2933 	/*
2934 	 * Start out assuming no warnings.
2935 	 */
2936 	status = 0;
2937 
2938 	/*
2939 	 * Reserve space for VLAN tag reconstruction.
2940 	 */
2941 	tp_reserve = VLAN_TAG_LEN;
2942 
2943 	/*
2944 	 * If we're capturing in cooked mode, reserve space for
2945 	 * a DLT_LINUX_SLL2 header; we don't know yet whether
2946 	 * we'll be using DLT_LINUX_SLL or DLT_LINUX_SLL2, as
2947 	 * that can be changed on an open device, so we reserve
2948 	 * space for the larger of the two.
2949 	 *
2950 	 * XXX - we assume that the kernel is still adding
2951 	 * 16 bytes of extra space, so we subtract 16 from
2952 	 * SLL2_HDR_LEN to get the additional space needed.
2953 	 * (Are they doing that for DLT_LINUX_SLL, the link-
2954 	 * layer header for which is 16 bytes?)
2955 	 *
2956 	 * XXX - should we use TPACKET_ALIGN(SLL2_HDR_LEN - 16)?
2957 	 */
2958 	if (handlep->cooked)
2959 		tp_reserve += SLL2_HDR_LEN - 16;
2960 
2961 	/*
2962 	 * Try to request that amount of reserve space.
2963 	 * This must be done before creating the ring buffer.
2964 	 */
2965 	len = sizeof(tp_reserve);
2966 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
2967 	    &tp_reserve, len) < 0) {
2968 		pcapint_fmt_errmsg_for_errno(handle->errbuf,
2969 		    PCAP_ERRBUF_SIZE, errno,
2970 		    "setsockopt (PACKET_RESERVE)");
2971 		return PCAP_ERROR;
2972 	}
2973 
2974 	switch (handlep->tp_version) {
2975 
2976 	case TPACKET_V2:
2977 		/* Note that with large snapshot length (say 256K, which is
2978 		 * the default for recent versions of tcpdump, Wireshark,
2979 		 * TShark, dumpcap or 64K, the value that "-s 0" has given for
2980 		 * a long time with tcpdump), if we use the snapshot
2981 		 * length to calculate the frame length, only a few frames
2982 		 * will be available in the ring even with pretty
2983 		 * large ring size (and a lot of memory will be unused).
2984 		 *
2985 		 * Ideally, we should choose a frame length based on the
2986 		 * minimum of the specified snapshot length and the maximum
2987 		 * packet size.  That's not as easy as it sounds; consider,
2988 		 * for example, an 802.11 interface in monitor mode, where
2989 		 * the frame would include a radiotap header, where the
2990 		 * maximum radiotap header length is device-dependent.
2991 		 *
2992 		 * So, for now, we just do this for Ethernet devices, where
2993 		 * there's no metadata header, and the link-layer header is
2994 		 * fixed length.  We can get the maximum packet size by
2995 		 * adding 18, the Ethernet header length plus the CRC length
2996 		 * (just in case we happen to get the CRC in the packet), to
2997 		 * the MTU of the interface; we fetch the MTU in the hopes
2998 		 * that it reflects support for jumbo frames.  (Even if the
2999 		 * interface is just being used for passive snooping, the
3000 		 * driver might set the size of buffers in the receive ring
3001 		 * based on the MTU, so that the MTU limits the maximum size
3002 		 * of packets that we can receive.)
3003 		 *
3004 		 * If segmentation/fragmentation or receive offload are
3005 		 * enabled, we can get reassembled/aggregated packets larger
3006 		 * than MTU, but bounded to 65535 plus the Ethernet overhead,
3007 		 * due to kernel and protocol constraints */
3008 		frame_size = handle->snapshot;
3009 		if (handle->linktype == DLT_EN10MB) {
3010 			unsigned int max_frame_len;
3011 			int mtu;
3012 			int offload;
3013 
3014 			mtu = iface_get_mtu(handle->fd, handle->opt.device,
3015 			    handle->errbuf);
3016 			if (mtu == -1)
3017 				return PCAP_ERROR;
3018 			offload = iface_get_offload(handle);
3019 			if (offload == -1)
3020 				return PCAP_ERROR;
3021 			if (offload)
3022 				max_frame_len = MAX(mtu, 65535);
3023 			else
3024 				max_frame_len = mtu;
3025 			max_frame_len += 18;
3026 
3027 			if (frame_size > max_frame_len)
3028 				frame_size = max_frame_len;
3029 		}
3030 
3031 		/* NOTE: calculus matching those in tpacket_rcv()
3032 		 * in linux-2.6/net/packet/af_packet.c
3033 		 */
3034 		len = sizeof(sk_type);
3035 		if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
3036 		    &len) < 0) {
3037 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
3038 			    PCAP_ERRBUF_SIZE, errno, "getsockopt (SO_TYPE)");
3039 			return PCAP_ERROR;
3040 		}
3041 		maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3042 			/* XXX: in the kernel maclen is calculated from
3043 			 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3044 			 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3045 			 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3046 			 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3047 			 * but I see no way to get those sizes in userspace,
3048 			 * like for instance with an ifreq ioctl();
3049 			 * the best thing I've found so far is MAX_HEADER in
3050 			 * the kernel part of linux-2.6/include/linux/netdevice.h
3051 			 * which goes up to 128+48=176; since pcap-linux.c
3052 			 * defines a MAX_LINKHEADER_SIZE of 256 which is
3053 			 * greater than that, let's use it.. maybe is it even
3054 			 * large enough to directly replace macoff..
3055 			 */
3056 		tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3057 		netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3058 			/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
3059 			 * of netoff, which contradicts
3060 			 * linux-2.6/Documentation/networking/packet_mmap.txt
3061 			 * documenting that:
3062 			 * "- Gap, chosen so that packet data (Start+tp_net)
3063 			 * aligns to TPACKET_ALIGNMENT=16"
3064 			 */
3065 			/* NOTE: in linux-2.6/include/linux/skbuff.h:
3066 			 * "CPUs often take a performance hit
3067 			 *  when accessing unaligned memory locations"
3068 			 */
3069 		macoff = netoff - maclen;
3070 		req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3071 		/*
3072 		 * Round the buffer size up to a multiple of the
3073 		 * frame size (rather than rounding down, which
3074 		 * would give a buffer smaller than our caller asked
3075 		 * for, and possibly give zero frames if the requested
3076 		 * buffer size is too small for one frame).
3077 		 */
3078 		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
3079 		break;
3080 
3081 #ifdef HAVE_TPACKET3
3082 	case TPACKET_V3:
3083 		/* The "frames" for this are actually buffers that
3084 		 * contain multiple variable-sized frames.
3085 		 *
3086 		 * We pick a "frame" size of MAXIMUM_SNAPLEN to leave
3087 		 * enough room for at least one reasonably-sized packet
3088 		 * in the "frame". */
3089 		req.tp_frame_size = MAXIMUM_SNAPLEN;
3090 		/*
3091 		 * Round the buffer size up to a multiple of the
3092 		 * "frame" size (rather than rounding down, which
3093 		 * would give a buffer smaller than our caller asked
3094 		 * for, and possibly give zero "frames" if the requested
3095 		 * buffer size is too small for one "frame").
3096 		 */
3097 		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
3098 		break;
3099 #endif
3100 	default:
3101 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3102 		    "Internal error: unknown TPACKET_ value %u",
3103 		    handlep->tp_version);
3104 		return PCAP_ERROR;
3105 	}
3106 
3107 	/* compute the minimum block size that will handle this frame.
3108 	 * The block has to be page size aligned.
3109 	 * The max block size allowed by the kernel is arch-dependent and
3110 	 * it's not explicitly checked here. */
3111 	req.tp_block_size = getpagesize();
3112 	while (req.tp_block_size < req.tp_frame_size)
3113 		req.tp_block_size <<= 1;
3114 
3115 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3116 
3117 	/*
3118 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3119 	 * so we check for PACKET_TIMESTAMP.  We check for
3120 	 * linux/net_tstamp.h just in case a system somehow has
3121 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3122 	 * be unnecessary.
3123 	 *
3124 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3125 	 * linux/net_tstamp.h, so we don't bother checking whether
3126 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3127 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3128 	 * Linux system is badly broken).
3129 	 */
3130 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3131 	/*
3132 	 * If we were told to do so, ask the kernel and the driver
3133 	 * to use hardware timestamps.
3134 	 *
3135 	 * Hardware timestamps are only supported with mmapped
3136 	 * captures.
3137 	 */
3138 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3139 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3140 		struct hwtstamp_config hwconfig;
3141 		struct ifreq ifr;
3142 		int timesource;
3143 
3144 		/*
3145 		 * Ask for hardware time stamps on all packets,
3146 		 * including transmitted packets.
3147 		 */
3148 		memset(&hwconfig, 0, sizeof(hwconfig));
3149 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3150 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3151 
3152 		memset(&ifr, 0, sizeof(ifr));
3153 		pcapint_strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
3154 		ifr.ifr_data = (void *)&hwconfig;
3155 
3156 		/*
3157 		 * This may require CAP_NET_ADMIN.
3158 		 */
3159 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3160 			switch (errno) {
3161 
3162 			case EPERM:
3163 				/*
3164 				 * Treat this as an error, as the
3165 				 * user should try to run this
3166 				 * with the appropriate privileges -
3167 				 * and, if they can't, shouldn't
3168 				 * try requesting hardware time stamps.
3169 				 */
3170 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3171 				    "Attempt to set hardware timestamp failed - CAP_NET_ADMIN may be required");
3172 				return PCAP_ERROR_PERM_DENIED;
3173 
3174 			case EOPNOTSUPP:
3175 			case ERANGE:
3176 				/*
3177 				 * Treat this as a warning, as the
3178 				 * only way to fix the warning is to
3179 				 * get an adapter that supports hardware
3180 				 * time stamps for *all* packets.
3181 				 * (ERANGE means "we support hardware
3182 				 * time stamps, but for packets matching
3183 				 * that particular filter", so it means
3184 				 * "we don't support hardware time stamps
3185 				 * for all incoming packets" here.)
3186 				 *
3187 				 * We'll just fall back on the standard
3188 				 * host time stamps.
3189 				 */
3190 				status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3191 				break;
3192 
3193 			default:
3194 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
3195 				    PCAP_ERRBUF_SIZE, errno,
3196 				    "SIOCSHWTSTAMP failed");
3197 				return PCAP_ERROR;
3198 			}
3199 		} else {
3200 			/*
3201 			 * Well, that worked.  Now specify the type of
3202 			 * hardware time stamp we want for this
3203 			 * socket.
3204 			 */
3205 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3206 				/*
3207 				 * Hardware timestamp, synchronized
3208 				 * with the system clock.
3209 				 */
3210 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3211 			} else {
3212 				/*
3213 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3214 				 * timestamp, not synchronized with the
3215 				 * system clock.
3216 				 */
3217 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3218 			}
3219 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3220 				(void *)&timesource, sizeof(timesource))) {
3221 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
3222 				    PCAP_ERRBUF_SIZE, errno,
3223 				    "can't set PACKET_TIMESTAMP");
3224 				return PCAP_ERROR;
3225 			}
3226 		}
3227 	}
3228 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3229 
3230 	/* ask the kernel to create the ring */
3231 retry:
3232 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3233 
3234 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3235 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3236 
3237 #ifdef HAVE_TPACKET3
3238 	/* timeout value to retire block - use the configured buffering timeout, or default if <0. */
3239 	if (handlep->timeout > 0) {
3240 		/* Use the user specified timeout as the block timeout */
3241 		req.tp_retire_blk_tov = handlep->timeout;
3242 	} else if (handlep->timeout == 0) {
3243 		/*
3244 		 * In pcap, this means "infinite timeout"; TPACKET_V3
3245 		 * doesn't support that, so just set it to UINT_MAX
3246 		 * milliseconds.  In the TPACKET_V3 loop, if the
3247 		 * timeout is 0, and we haven't yet seen any packets,
3248 		 * and we block and still don't have any packets, we
3249 		 * keep blocking until we do.
3250 		 */
3251 		req.tp_retire_blk_tov = UINT_MAX;
3252 	} else {
3253 		/*
3254 		 * XXX - this is not valid; use 0, meaning "have the
3255 		 * kernel pick a default", for now.
3256 		 */
3257 		req.tp_retire_blk_tov = 0;
3258 	}
3259 	/* private data not used */
3260 	req.tp_sizeof_priv = 0;
3261 	/* Rx ring - feature request bits - none (rxhash will not be filled) */
3262 	req.tp_feature_req_word = 0;
3263 #endif
3264 
3265 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3266 					(void *) &req, sizeof(req))) {
3267 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3268 			/*
3269 			 * Memory failure; try to reduce the requested ring
3270 			 * size.
3271 			 *
3272 			 * We used to reduce this by half -- do 5% instead.
3273 			 * That may result in more iterations and a longer
3274 			 * startup, but the user will be much happier with
3275 			 * the resulting buffer size.
3276 			 */
3277 			if (req.tp_frame_nr < 20)
3278 				req.tp_frame_nr -= 1;
3279 			else
3280 				req.tp_frame_nr -= req.tp_frame_nr/20;
3281 			goto retry;
3282 		}
3283 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3284 		    errno, "can't create rx ring on packet socket");
3285 		return PCAP_ERROR;
3286 	}
3287 
3288 	/* memory map the rx ring */
3289 	handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
3290 	handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
3291 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3292 	if (handlep->mmapbuf == MAP_FAILED) {
3293 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3294 		    errno, "can't mmap rx ring");
3295 
3296 		/* clear the allocated ring on error*/
3297 		destroy_ring(handle);
3298 		return PCAP_ERROR;
3299 	}
3300 
3301 	/* allocate a ring for each frame header pointer*/
3302 	handle->cc = req.tp_frame_nr;
3303 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3304 	if (!handle->buffer) {
3305 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3306 		    errno, "can't allocate ring of frame headers");
3307 
3308 		destroy_ring(handle);
3309 		return PCAP_ERROR;
3310 	}
3311 
3312 	/* fill the header ring with proper frame ptr*/
3313 	handle->offset = 0;
3314 	for (i=0; i<req.tp_block_nr; ++i) {
3315 		u_char *base = &handlep->mmapbuf[i*req.tp_block_size];
3316 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3317 			RING_GET_CURRENT_FRAME(handle) = base;
3318 			base += req.tp_frame_size;
3319 		}
3320 	}
3321 
3322 	handle->bufsize = req.tp_frame_size;
3323 	handle->offset = 0;
3324 	return status;
3325 }
3326 
3327 /* free all ring related resources*/
3328 static void
3329 destroy_ring(pcap_t *handle)
3330 {
3331 	struct pcap_linux *handlep = handle->priv;
3332 
3333 	/*
3334 	 * Tell the kernel to destroy the ring.
3335 	 * We don't check for setsockopt failure, as 1) we can't recover
3336 	 * from an error and 2) we might not yet have set it up in the
3337 	 * first place.
3338 	 */
3339 	struct tpacket_req req;
3340 	memset(&req, 0, sizeof(req));
3341 	(void)setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3342 				(void *) &req, sizeof(req));
3343 
3344 	/* if ring is mapped, unmap it*/
3345 	if (handlep->mmapbuf) {
3346 		/* do not test for mmap failure, as we can't recover from any error */
3347 		(void)munmap(handlep->mmapbuf, handlep->mmapbuflen);
3348 		handlep->mmapbuf = NULL;
3349 	}
3350 }
3351 
3352 /*
3353  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3354  * for Linux mmapped capture.
3355  *
3356  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3357  * data handed to the callback to be valid after the callback returns,
3358  * but pcap_read_linux_mmap() has to release that packet as soon as
3359  * the callback returns (otherwise, the kernel thinks there's still
3360  * at least one unprocessed packet available in the ring, so a select()
3361  * will immediately return indicating that there's data to process), so,
3362  * in the callback, we have to make a copy of the packet.
3363  *
3364  * Yes, this means that, if the capture is using the ring buffer, using
3365  * pcap_next() or pcap_next_ex() requires more copies than using
3366  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3367  * pcap_next() or pcap_next_ex().
3368  */
3369 static void
3370 pcapint_oneshot_linux(u_char *user, const struct pcap_pkthdr *h,
3371     const u_char *bytes)
3372 {
3373 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3374 	pcap_t *handle = sp->pd;
3375 	struct pcap_linux *handlep = handle->priv;
3376 
3377 	*sp->hdr = *h;
3378 	memcpy(handlep->oneshot_buffer, bytes, h->caplen);
3379 	*sp->pkt = handlep->oneshot_buffer;
3380 }
3381 
3382 static int
3383 pcap_getnonblock_linux(pcap_t *handle)
3384 {
3385 	struct pcap_linux *handlep = handle->priv;
3386 
3387 	/* use negative value of timeout to indicate non blocking ops */
3388 	return (handlep->timeout<0);
3389 }
3390 
3391 static int
3392 pcap_setnonblock_linux(pcap_t *handle, int nonblock)
3393 {
3394 	struct pcap_linux *handlep = handle->priv;
3395 
3396 	/*
3397 	 * Set the file descriptor to the requested mode, as we use
3398 	 * it for sending packets.
3399 	 */
3400 	if (pcapint_setnonblock_fd(handle, nonblock) == -1)
3401 		return -1;
3402 
3403 	/*
3404 	 * Map each value to their corresponding negation to
3405 	 * preserve the timeout value provided with pcap_set_timeout.
3406 	 */
3407 	if (nonblock) {
3408 		/*
3409 		 * We're setting the mode to non-blocking mode.
3410 		 */
3411 		if (handlep->timeout >= 0) {
3412 			/*
3413 			 * Indicate that we're switching to
3414 			 * non-blocking mode.
3415 			 */
3416 			handlep->timeout = ~handlep->timeout;
3417 		}
3418 		if (handlep->poll_breakloop_fd != -1) {
3419 			/* Close the eventfd; we do not need it in nonblock mode. */
3420 			close(handlep->poll_breakloop_fd);
3421 			handlep->poll_breakloop_fd = -1;
3422 		}
3423 	} else {
3424 		/*
3425 		 * We're setting the mode to blocking mode.
3426 		 */
3427 		if (handlep->poll_breakloop_fd == -1) {
3428 			/* If we did not have an eventfd, open one now that we are blocking. */
3429 			if ( ( handlep->poll_breakloop_fd = eventfd(0, EFD_NONBLOCK) ) == -1 ) {
3430 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
3431 				    PCAP_ERRBUF_SIZE, errno,
3432 				    "could not open eventfd");
3433 				return -1;
3434 			}
3435 		}
3436 		if (handlep->timeout < 0) {
3437 			handlep->timeout = ~handlep->timeout;
3438 		}
3439 	}
3440 	/* Update the timeout to use in poll(). */
3441 	set_poll_timeout(handlep);
3442 	return 0;
3443 }
3444 
3445 /*
3446  * Get the status field of the ring buffer frame at a specified offset.
3447  */
3448 static inline u_int
3449 pcap_get_ring_frame_status(pcap_t *handle, int offset)
3450 {
3451 	struct pcap_linux *handlep = handle->priv;
3452 	union thdr h;
3453 
3454 	h.raw = RING_GET_FRAME_AT(handle, offset);
3455 	switch (handlep->tp_version) {
3456 	case TPACKET_V2:
3457 		return __atomic_load_n(&h.h2->tp_status, __ATOMIC_ACQUIRE);
3458 		break;
3459 #ifdef HAVE_TPACKET3
3460 	case TPACKET_V3:
3461 		return __atomic_load_n(&h.h3->hdr.bh1.block_status, __ATOMIC_ACQUIRE);
3462 		break;
3463 #endif
3464 	}
3465 	/* This should not happen. */
3466 	return 0;
3467 }
3468 
3469 /*
3470  * Block waiting for frames to be available.
3471  */
3472 static int pcap_wait_for_frames_mmap(pcap_t *handle)
3473 {
3474 	struct pcap_linux *handlep = handle->priv;
3475 	int timeout;
3476 	struct ifreq ifr;
3477 	int ret;
3478 	struct pollfd pollinfo[2];
3479 	int numpollinfo;
3480 	pollinfo[0].fd = handle->fd;
3481 	pollinfo[0].events = POLLIN;
3482 	if ( handlep->poll_breakloop_fd == -1 ) {
3483 		numpollinfo = 1;
3484 		pollinfo[1].revents = 0;
3485 		/*
3486 		 * We set pollinfo[1].revents to zero, even though
3487 		 * numpollinfo = 1 meaning that poll() doesn't see
3488 		 * pollinfo[1], so that we do not have to add a
3489 		 * conditional of numpollinfo > 1 below when we
3490 		 * test pollinfo[1].revents.
3491 		 */
3492 	} else {
3493 		pollinfo[1].fd = handlep->poll_breakloop_fd;
3494 		pollinfo[1].events = POLLIN;
3495 		numpollinfo = 2;
3496 	}
3497 
3498 	/*
3499 	 * Keep polling until we either get some packets to read, see
3500 	 * that we got told to break out of the loop, get a fatal error,
3501 	 * or discover that the device went away.
3502 	 *
3503 	 * In non-blocking mode, we must still do one poll() to catch
3504 	 * any pending error indications, but the poll() has a timeout
3505 	 * of 0, so that it doesn't block, and we quit after that one
3506 	 * poll().
3507 	 *
3508 	 * If we've seen an ENETDOWN, it might be the first indication
3509 	 * that the device went away, or it might just be that it was
3510 	 * configured down.  Unfortunately, there's no guarantee that
3511 	 * the device has actually been removed as an interface, because:
3512 	 *
3513 	 * 1) if, as appears to be the case at least some of the time,
3514 	 * the PF_PACKET socket code first gets a NETDEV_DOWN indication
3515 	 * for the device and then gets a NETDEV_UNREGISTER indication
3516 	 * for it, the first indication will cause a wakeup with ENETDOWN
3517 	 * but won't set the packet socket's field for the interface index
3518 	 * to -1, and the second indication won't cause a wakeup (because
3519 	 * the first indication also caused the protocol hook to be
3520 	 * unregistered) but will set the packet socket's field for the
3521 	 * interface index to -1;
3522 	 *
3523 	 * 2) even if just a NETDEV_UNREGISTER indication is registered,
3524 	 * the packet socket's field for the interface index only gets
3525 	 * set to -1 after the wakeup, so there's a small but non-zero
3526 	 * risk that a thread blocked waiting for the wakeup will get
3527 	 * to the "fetch the socket name" code before the interface index
3528 	 * gets set to -1, so it'll get the old interface index.
3529 	 *
3530 	 * Therefore, if we got an ENETDOWN and haven't seen a packet
3531 	 * since then, we assume that we might be waiting for the interface
3532 	 * to disappear, and poll with a timeout to try again in a short
3533 	 * period of time.  If we *do* see a packet, the interface has
3534 	 * come back up again, and is *definitely* still there, so we
3535 	 * don't need to poll.
3536 	 */
3537 	for (;;) {
3538 		/*
3539 		 * Yes, we do this even in non-blocking mode, as it's
3540 		 * the only way to get error indications from a
3541 		 * tpacket socket.
3542 		 *
3543 		 * The timeout is 0 in non-blocking mode, so poll()
3544 		 * returns immediately.
3545 		 */
3546 		timeout = handlep->poll_timeout;
3547 
3548 		/*
3549 		 * If we got an ENETDOWN and haven't gotten an indication
3550 		 * that the device has gone away or that the device is up,
3551 		 * we don't yet know for certain whether the device has
3552 		 * gone away or not, do a poll() with a 1-millisecond timeout,
3553 		 * as we have to poll indefinitely for "device went away"
3554 		 * indications until we either get one or see that the
3555 		 * device is up.
3556 		 */
3557 		if (handlep->netdown) {
3558 			if (timeout != 0)
3559 				timeout = 1;
3560 		}
3561 		ret = poll(pollinfo, numpollinfo, timeout);
3562 		if (ret < 0) {
3563 			/*
3564 			 * Error.  If it's not EINTR, report it.
3565 			 */
3566 			if (errno != EINTR) {
3567 				pcapint_fmt_errmsg_for_errno(handle->errbuf,
3568 				    PCAP_ERRBUF_SIZE, errno,
3569 				    "can't poll on packet socket");
3570 				return PCAP_ERROR;
3571 			}
3572 
3573 			/*
3574 			 * It's EINTR; if we were told to break out of
3575 			 * the loop, do so.
3576 			 */
3577 			if (handle->break_loop) {
3578 				handle->break_loop = 0;
3579 				return PCAP_ERROR_BREAK;
3580 			}
3581 		} else if (ret > 0) {
3582 			/*
3583 			 * OK, some descriptor is ready.
3584 			 * Check the socket descriptor first.
3585 			 *
3586 			 * As I read the Linux man page, pollinfo[0].revents
3587 			 * will either be POLLIN, POLLERR, POLLHUP, or POLLNVAL.
3588 			 */
3589 			if (pollinfo[0].revents == POLLIN) {
3590 				/*
3591 				 * OK, we may have packets to
3592 				 * read.
3593 				 */
3594 				break;
3595 			}
3596 			if (pollinfo[0].revents != 0) {
3597 				/*
3598 				 * There's some indication other than
3599 				 * "you can read on this descriptor" on
3600 				 * the descriptor.
3601 				 */
3602 				if (pollinfo[0].revents & POLLNVAL) {
3603 					snprintf(handle->errbuf,
3604 					    PCAP_ERRBUF_SIZE,
3605 					    "Invalid polling request on packet socket");
3606 					return PCAP_ERROR;
3607 				}
3608 				if (pollinfo[0].revents & (POLLHUP | POLLRDHUP)) {
3609 					snprintf(handle->errbuf,
3610 					    PCAP_ERRBUF_SIZE,
3611 					    "Hangup on packet socket");
3612 					return PCAP_ERROR;
3613 				}
3614 				if (pollinfo[0].revents & POLLERR) {
3615 					/*
3616 					 * Get the error.
3617 					 */
3618 					int err;
3619 					socklen_t errlen;
3620 
3621 					errlen = sizeof(err);
3622 					if (getsockopt(handle->fd, SOL_SOCKET,
3623 					    SO_ERROR, &err, &errlen) == -1) {
3624 						/*
3625 						 * The call *itself* returned
3626 						 * an error; make *that*
3627 						 * the error.
3628 						 */
3629 						err = errno;
3630 					}
3631 
3632 					/*
3633 					 * OK, we have the error.
3634 					 */
3635 					if (err == ENETDOWN) {
3636 						/*
3637 						 * The device on which we're
3638 						 * capturing went away or the
3639 						 * interface was taken down.
3640 						 *
3641 						 * We don't know for certain
3642 						 * which happened, and the
3643 						 * next poll() may indicate
3644 						 * that there are packets
3645 						 * to be read, so just set
3646 						 * a flag to get us to do
3647 						 * checks later, and set
3648 						 * the required select
3649 						 * timeout to 1 millisecond
3650 						 * so that event loops that
3651 						 * check our socket descriptor
3652 						 * also time out so that
3653 						 * they can call us and we
3654 						 * can do the checks.
3655 						 */
3656 						handlep->netdown = 1;
3657 						handle->required_select_timeout = &netdown_timeout;
3658 					} else if (err == 0) {
3659 						/*
3660 						 * This shouldn't happen, so
3661 						 * report a special indication
3662 						 * that it did.
3663 						 */
3664 						snprintf(handle->errbuf,
3665 						    PCAP_ERRBUF_SIZE,
3666 						    "Error condition on packet socket: Reported error was 0");
3667 						return PCAP_ERROR;
3668 					} else {
3669 						pcapint_fmt_errmsg_for_errno(handle->errbuf,
3670 						    PCAP_ERRBUF_SIZE,
3671 						    err,
3672 						    "Error condition on packet socket");
3673 						return PCAP_ERROR;
3674 					}
3675 				}
3676 			}
3677 			/*
3678 			 * Now check the event device.
3679 			 */
3680 			if (pollinfo[1].revents & POLLIN) {
3681 				ssize_t nread;
3682 				uint64_t value;
3683 
3684 				/*
3685 				 * This should never fail, but, just
3686 				 * in case....
3687 				 */
3688 				nread = read(handlep->poll_breakloop_fd, &value,
3689 				    sizeof(value));
3690 				if (nread == -1) {
3691 					pcapint_fmt_errmsg_for_errno(handle->errbuf,
3692 					    PCAP_ERRBUF_SIZE,
3693 					    errno,
3694 					    "Error reading from event FD");
3695 					return PCAP_ERROR;
3696 				}
3697 
3698 				/*
3699 				 * According to the Linux read(2) man
3700 				 * page, read() will transfer at most
3701 				 * 2^31-1 bytes, so the return value is
3702 				 * either -1 or a value between 0
3703 				 * and 2^31-1, so it's non-negative.
3704 				 *
3705 				 * Cast it to size_t to squelch
3706 				 * warnings from the compiler; add this
3707 				 * comment to squelch warnings from
3708 				 * humans reading the code. :-)
3709 				 *
3710 				 * Don't treat an EOF as an error, but
3711 				 * *do* treat a short read as an error;
3712 				 * that "shouldn't happen", but....
3713 				 */
3714 				if (nread != 0 &&
3715 				    (size_t)nread < sizeof(value)) {
3716 					snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3717 					    "Short read from event FD: expected %zu, got %zd",
3718 					    sizeof(value), nread);
3719 					return PCAP_ERROR;
3720 				}
3721 
3722 				/*
3723 				 * This event gets signaled by a
3724 				 * pcap_breakloop() call; if we were told
3725 				 * to break out of the loop, do so.
3726 				 */
3727 				if (handle->break_loop) {
3728 					handle->break_loop = 0;
3729 					return PCAP_ERROR_BREAK;
3730 				}
3731 			}
3732 		}
3733 
3734 		/*
3735 		 * Either:
3736 		 *
3737 		 *   1) we got neither an error from poll() nor any
3738 		 *      readable descriptors, in which case there
3739 		 *      are no packets waiting to read
3740 		 *
3741 		 * or
3742 		 *
3743 		 *   2) We got readable descriptors but the PF_PACKET
3744 		 *      socket wasn't one of them, in which case there
3745 		 *      are no packets waiting to read
3746 		 *
3747 		 * so, if we got an ENETDOWN, we've drained whatever
3748 		 * packets were available to read at the point of the
3749 		 * ENETDOWN.
3750 		 *
3751 		 * So, if we got an ENETDOWN and haven't gotten an indication
3752 		 * that the device has gone away or that the device is up,
3753 		 * we don't yet know for certain whether the device has
3754 		 * gone away or not, check whether the device exists and is
3755 		 * up.
3756 		 */
3757 		if (handlep->netdown) {
3758 			if (!device_still_exists(handle)) {
3759 				/*
3760 				 * The device doesn't exist any more;
3761 				 * report that.
3762 				 *
3763 				 * XXX - we should really return an
3764 				 * appropriate error for that, but
3765 				 * pcap_dispatch() etc. aren't documented
3766 				 * as having error returns other than
3767 				 * PCAP_ERROR or PCAP_ERROR_BREAK.
3768 				 */
3769 				snprintf(handle->errbuf,  PCAP_ERRBUF_SIZE,
3770 				    "The interface disappeared");
3771 				return PCAP_ERROR;
3772 			}
3773 
3774 			/*
3775 			 * The device still exists; try to see if it's up.
3776 			 */
3777 			memset(&ifr, 0, sizeof(ifr));
3778 			pcapint_strlcpy(ifr.ifr_name, handlep->device,
3779 			    sizeof(ifr.ifr_name));
3780 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
3781 				if (errno == ENXIO || errno == ENODEV) {
3782 					/*
3783 					 * OK, *now* it's gone.
3784 					 *
3785 					 * XXX - see above comment.
3786 					 */
3787 					snprintf(handle->errbuf,
3788 					    PCAP_ERRBUF_SIZE,
3789 					    "The interface disappeared");
3790 					return PCAP_ERROR;
3791 				} else {
3792 					pcapint_fmt_errmsg_for_errno(handle->errbuf,
3793 					    PCAP_ERRBUF_SIZE, errno,
3794 					    "%s: Can't get flags",
3795 					    handlep->device);
3796 					return PCAP_ERROR;
3797 				}
3798 			}
3799 			if (ifr.ifr_flags & IFF_UP) {
3800 				/*
3801 				 * It's up, so it definitely still exists.
3802 				 * Cancel the ENETDOWN indication - we
3803 				 * presumably got it due to the interface
3804 				 * going down rather than the device going
3805 				 * away - and revert to "no required select
3806 				 * timeout.
3807 				 */
3808 				handlep->netdown = 0;
3809 				handle->required_select_timeout = NULL;
3810 			}
3811 		}
3812 
3813 		/*
3814 		 * If we're in non-blocking mode, just quit now, rather
3815 		 * than spinning in a loop doing poll()s that immediately
3816 		 * time out if there's no indication on any descriptor.
3817 		 */
3818 		if (handlep->poll_timeout == 0)
3819 			break;
3820 	}
3821 	return 0;
3822 }
3823 
3824 /* handle a single memory mapped packet */
3825 static int pcap_handle_packet_mmap(
3826 		pcap_t *handle,
3827 		pcap_handler callback,
3828 		u_char *user,
3829 		unsigned char *frame,
3830 		unsigned int tp_len,
3831 		unsigned int tp_mac,
3832 		unsigned int tp_snaplen,
3833 		unsigned int tp_sec,
3834 		unsigned int tp_usec,
3835 		int tp_vlan_tci_valid,
3836 		__u16 tp_vlan_tci,
3837 		__u16 tp_vlan_tpid)
3838 {
3839 	struct pcap_linux *handlep = handle->priv;
3840 	unsigned char *bp;
3841 	struct sockaddr_ll *sll;
3842 	struct pcap_pkthdr pcaphdr;
3843 	unsigned int snaplen = tp_snaplen;
3844 	struct utsname utsname;
3845 
3846 	/* perform sanity check on internal offset. */
3847 	if (tp_mac + tp_snaplen > handle->bufsize) {
3848 		/*
3849 		 * Report some system information as a debugging aid.
3850 		 */
3851 		if (uname(&utsname) != -1) {
3852 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3853 				"corrupted frame on kernel ring mac "
3854 				"offset %u + caplen %u > frame len %d "
3855 				"(kernel %.32s version %s, machine %.16s)",
3856 				tp_mac, tp_snaplen, handle->bufsize,
3857 				utsname.release, utsname.version,
3858 				utsname.machine);
3859 		} else {
3860 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3861 				"corrupted frame on kernel ring mac "
3862 				"offset %u + caplen %u > frame len %d",
3863 				tp_mac, tp_snaplen, handle->bufsize);
3864 		}
3865 		return -1;
3866 	}
3867 
3868 	/* run filter on received packet
3869 	 * If the kernel filtering is enabled we need to run the
3870 	 * filter until all the frames present into the ring
3871 	 * at filter creation time are processed.
3872 	 * In this case, blocks_to_filter_in_userland is used
3873 	 * as a counter for the packet we need to filter.
3874 	 * Note: alternatively it could be possible to stop applying
3875 	 * the filter when the ring became empty, but it can possibly
3876 	 * happen a lot later... */
3877 	bp = frame + tp_mac;
3878 
3879 	/* if required build in place the sll header*/
3880 	sll = (void *)(frame + TPACKET_ALIGN(handlep->tp_hdrlen));
3881 	if (handlep->cooked) {
3882 		if (handle->linktype == DLT_LINUX_SLL2) {
3883 			struct sll2_header *hdrp;
3884 
3885 			/*
3886 			 * The kernel should have left us with enough
3887 			 * space for an sll header; back up the packet
3888 			 * data pointer into that space, as that'll be
3889 			 * the beginning of the packet we pass to the
3890 			 * callback.
3891 			 */
3892 			bp -= SLL2_HDR_LEN;
3893 
3894 			/*
3895 			 * Let's make sure that's past the end of
3896 			 * the tpacket header, i.e. >=
3897 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3898 			 * don't step on the header when we construct
3899 			 * the sll header.
3900 			 */
3901 			if (bp < (u_char *)frame +
3902 					   TPACKET_ALIGN(handlep->tp_hdrlen) +
3903 					   sizeof(struct sockaddr_ll)) {
3904 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3905 					"cooked-mode frame doesn't have room for sll header");
3906 				return -1;
3907 			}
3908 
3909 			/*
3910 			 * OK, that worked; construct the sll header.
3911 			 */
3912 			hdrp = (struct sll2_header *)bp;
3913 			hdrp->sll2_protocol = sll->sll_protocol;
3914 			hdrp->sll2_reserved_mbz = 0;
3915 			hdrp->sll2_if_index = htonl(sll->sll_ifindex);
3916 			hdrp->sll2_hatype = htons(sll->sll_hatype);
3917 			hdrp->sll2_pkttype = sll->sll_pkttype;
3918 			hdrp->sll2_halen = sll->sll_halen;
3919 			memcpy(hdrp->sll2_addr, sll->sll_addr, SLL_ADDRLEN);
3920 
3921 			snaplen += sizeof(struct sll2_header);
3922 		} else {
3923 			struct sll_header *hdrp;
3924 
3925 			/*
3926 			 * The kernel should have left us with enough
3927 			 * space for an sll header; back up the packet
3928 			 * data pointer into that space, as that'll be
3929 			 * the beginning of the packet we pass to the
3930 			 * callback.
3931 			 */
3932 			bp -= SLL_HDR_LEN;
3933 
3934 			/*
3935 			 * Let's make sure that's past the end of
3936 			 * the tpacket header, i.e. >=
3937 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3938 			 * don't step on the header when we construct
3939 			 * the sll header.
3940 			 */
3941 			if (bp < (u_char *)frame +
3942 					   TPACKET_ALIGN(handlep->tp_hdrlen) +
3943 					   sizeof(struct sockaddr_ll)) {
3944 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3945 					"cooked-mode frame doesn't have room for sll header");
3946 				return -1;
3947 			}
3948 
3949 			/*
3950 			 * OK, that worked; construct the sll header.
3951 			 */
3952 			hdrp = (struct sll_header *)bp;
3953 			hdrp->sll_pkttype = htons(sll->sll_pkttype);
3954 			hdrp->sll_hatype = htons(sll->sll_hatype);
3955 			hdrp->sll_halen = htons(sll->sll_halen);
3956 			memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
3957 			hdrp->sll_protocol = sll->sll_protocol;
3958 
3959 			snaplen += sizeof(struct sll_header);
3960 		}
3961 	} else {
3962 		/*
3963 		 * If this is a packet from a CAN device, so that
3964 		 * sll->sll_hatype is ARPHRD_CAN, then, as we're
3965 		 * not capturing in cooked mode, its link-layer
3966 		 * type is DLT_CAN_SOCKETCAN.  Fix up the header
3967 		 * provided by the code below us to match what
3968 		 * DLT_CAN_SOCKETCAN is expected to provide.
3969 		 */
3970 		if (sll->sll_hatype == ARPHRD_CAN) {
3971 			pcap_can_socketcan_hdr *canhdr = (pcap_can_socketcan_hdr *)bp;
3972 			uint16_t protocol = ntohs(sll->sll_protocol);
3973 
3974 			/*
3975 			 * Check the protocol field from the sll header.
3976 			 * If it's one of the known CAN protocol types,
3977 			 * make sure the appropriate flags are set, so
3978 			 * that a program can tell what type of frame
3979 			 * it is.
3980 			 *
3981 			 * The two flags are:
3982 			 *
3983 			 *   CANFD_FDF, which is in the fd_flags field
3984 			 *   of the CAN classic/CAN FD header;
3985 			 *
3986 			 *   CANXL_XLF, which is in the flags field
3987 			 *   of the CAN XL header, which overlaps
3988 			 *   the payload_length field of the CAN
3989 			 *   classic/CAN FD header.
3990 			 */
3991 			switch (protocol) {
3992 
3993 			case LINUX_SLL_P_CAN:
3994 				/*
3995 				 * CAN classic.
3996 				 *
3997 				 * Zero out the fd_flags and reserved
3998 				 * fields, in case they're uninitialized
3999 				 * crap, and clear the CANXL_XLF bit in
4000 				 * the payload_length field.
4001 				 *
4002 				 * This means that the CANFD_FDF flag isn't
4003 				 * set in the fd_flags field, and that
4004 				 * the CANXL_XLF bit isn't set in the
4005 				 * payload_length field, so this frame
4006 				 * will appear to be a CAN classic frame.
4007 				 */
4008 				canhdr->payload_length &= ~CANXL_XLF;
4009 				canhdr->fd_flags = 0;
4010 				canhdr->reserved1 = 0;
4011 				canhdr->reserved2 = 0;
4012 				break;
4013 
4014 			case LINUX_SLL_P_CANFD:
4015 				/*
4016 				 * Set CANFD_FDF in the fd_flags field,
4017 				 * and clear the CANXL_XLF bit in the
4018 				 * payload_length field, so this frame
4019 				 * will appear to be a CAN FD frame.
4020 				 */
4021 				canhdr->payload_length &= ~CANXL_XLF;
4022 				canhdr->fd_flags |= CANFD_FDF;
4023 
4024 				/*
4025 				 * Zero out all the unknown bits in fd_flags
4026 				 * and clear the reserved fields, so that
4027 				 * a program reading this can assume that
4028 				 * CANFD_FDF is set because we set it, not
4029 				 * because some uninitialized crap was
4030 				 * provided in the fd_flags field.
4031 				 *
4032 				 * (At least some LINKTYPE_CAN_SOCKETCAN
4033 				 * files attached to Wireshark bugs had
4034 				 * uninitialized junk there, so it does
4035 				 * happen.)
4036 				 *
4037 				 * Update this if Linux adds more flag bits
4038 				 * to the fd_flags field or uses either of
4039 				 * the reserved fields for FD frames.
4040 				 */
4041 				canhdr->fd_flags &= (CANFD_FDF|CANFD_ESI|CANFD_BRS);
4042 				canhdr->reserved1 = 0;
4043 				canhdr->reserved2 = 0;
4044 				break;
4045 
4046 			case LINUX_SLL_P_CANXL:
4047 				/*
4048 				 * CAN XL frame.
4049 				 *
4050 				 * Make sure the CANXL_XLF bit is set in
4051 				 * the payload_length field, so that
4052 				 * this frame will appear to be a
4053 				 * CAN XL frame.
4054 				 */
4055 				canhdr->payload_length |= CANXL_XLF;
4056 				break;
4057 			}
4058 
4059 			/*
4060 			 * Put multi-byte header fields in a byte-order
4061 			 *-independent format.
4062 			 */
4063 			if (canhdr->payload_length & CANXL_XLF) {
4064 				/*
4065 				 * This is a CAN XL frame.
4066 				 *
4067 				 * DLT_CAN_SOCKETCAN is specified as having
4068 				 * the Priority ID/VCID field in big--
4069 				 * endian byte order, and the payload length
4070 				 * and Acceptance Field in little-endian byte
4071 				 * order. but capturing on a CAN device
4072 				 * provides them in host byte order.
4073 				 * Convert them to the appropriate byte
4074 				 * orders.
4075 				 *
4076 				 * The reason we put the first field
4077 				 * into big-endian byte order is that
4078 				 * older libpcap code, ignorant of
4079 				 * CAN XL, treated it as the CAN ID
4080 				 * field and put it into big-endian
4081 				 * byte order, and we don't want to
4082 				 * break code that understands CAN XL
4083 				 * headers, and treats that field as
4084 				 * being big-endian.
4085 				 *
4086 				 * The other fields are put in little-
4087 				 * endian byte order is that older
4088 				 * libpcap code, ignorant of CAN XL,
4089 				 * left those fields alone, and the
4090 				 * processors on which the CAN XL
4091 				 * frames were captured are likely
4092 				 * to be little-endian processors.
4093 				 */
4094 				pcap_can_socketcan_xl_hdr *canxl_hdr = (pcap_can_socketcan_xl_hdr *)bp;
4095 
4096 #if __BYTE_ORDER == __LITTLE_ENDIAN
4097 				/*
4098 				 * We're capturing on a little-endian
4099 				 * machine, so we put the priority/VCID
4100 				 * field into big-endian byte order, and
4101 				 * leave the payload length and acceptance
4102 				 * field in little-endian byte order.
4103 				 */
4104 				/* Byte-swap priority/VCID. */
4105 				canxl_hdr->priority_vcid = SWAPLONG(canxl_hdr->priority_vcid);
4106 #elif __BYTE_ORDER == __BIG_ENDIAN
4107 				/*
4108 				 * We're capturing on a big-endian
4109 				 * machine, so we want to leave the
4110 				 * priority/VCID field alone, and byte-swap
4111 				 * the payload length and acceptance
4112 				 * fields to little-endian.
4113 				 */
4114 				/* Byte-swap the payload length */
4115 				canxl_hdr->payload_length = SWAPSHORT(canxl_hdr->payload_length);
4116 
4117 				/*
4118 				 * Byte-swap the acceptance field.
4119 				 *
4120 				 * XXX - is it just a 4-octet string,
4121 				 * not in any byte order?
4122 				 */
4123 				canxl_hdr->acceptance_field = SWAPLONG(canxl_hdr->acceptance_field);
4124 #else
4125 #error "Unknown byte order"
4126 #endif
4127 			} else {
4128 				/*
4129 				 * CAN or CAN FD frame.
4130 				 *
4131 				 * DLT_CAN_SOCKETCAN is specified as having
4132 				 * the CAN ID and flags in network byte
4133 				 * order, but capturing on a CAN device
4134 				 * provides it in host byte order.  Convert
4135 				 * it to network byte order.
4136 				 */
4137 				canhdr->can_id = htonl(canhdr->can_id);
4138 			}
4139 		}
4140 	}
4141 
4142 	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
4143 		struct pcap_bpf_aux_data aux_data;
4144 
4145 		aux_data.vlan_tag_present = tp_vlan_tci_valid;
4146 		aux_data.vlan_tag = tp_vlan_tci & 0x0fff;
4147 
4148 		if (pcapint_filter_with_aux_data(handle->fcode.bf_insns,
4149 					      bp,
4150 					      tp_len,
4151 					      snaplen,
4152 					      &aux_data) == 0)
4153 			return 0;
4154 	}
4155 
4156 	if (!linux_check_direction(handle, sll))
4157 		return 0;
4158 
4159 	/* get required packet info from ring header */
4160 	pcaphdr.ts.tv_sec = tp_sec;
4161 	pcaphdr.ts.tv_usec = tp_usec;
4162 	pcaphdr.caplen = tp_snaplen;
4163 	pcaphdr.len = tp_len;
4164 
4165 	/* if required build in place the sll header*/
4166 	if (handlep->cooked) {
4167 		/* update packet len */
4168 		if (handle->linktype == DLT_LINUX_SLL2) {
4169 			pcaphdr.caplen += SLL2_HDR_LEN;
4170 			pcaphdr.len += SLL2_HDR_LEN;
4171 		} else {
4172 			pcaphdr.caplen += SLL_HDR_LEN;
4173 			pcaphdr.len += SLL_HDR_LEN;
4174 		}
4175 	}
4176 
4177 	if (tp_vlan_tci_valid &&
4178 		handlep->vlan_offset != -1 &&
4179 		tp_snaplen >= (unsigned int) handlep->vlan_offset)
4180 	{
4181 		struct vlan_tag *tag;
4182 
4183 		/*
4184 		 * Move everything in the header, except the type field,
4185 		 * down VLAN_TAG_LEN bytes, to allow us to insert the
4186 		 * VLAN tag between that stuff and the type field.
4187 		 */
4188 		bp -= VLAN_TAG_LEN;
4189 		memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
4190 
4191 		/*
4192 		 * Now insert the tag.
4193 		 */
4194 		tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
4195 		tag->vlan_tpid = htons(tp_vlan_tpid);
4196 		tag->vlan_tci = htons(tp_vlan_tci);
4197 
4198 		/*
4199 		 * Add the tag to the packet lengths.
4200 		 */
4201 		pcaphdr.caplen += VLAN_TAG_LEN;
4202 		pcaphdr.len += VLAN_TAG_LEN;
4203 	}
4204 
4205 	/*
4206 	 * The only way to tell the kernel to cut off the
4207 	 * packet at a snapshot length is with a filter program;
4208 	 * if there's no filter program, the kernel won't cut
4209 	 * the packet off.
4210 	 *
4211 	 * Trim the snapshot length to be no longer than the
4212 	 * specified snapshot length.
4213 	 *
4214 	 * XXX - an alternative is to put a filter, consisting
4215 	 * of a "ret <snaplen>" instruction, on the socket
4216 	 * in the activate routine, so that the truncation is
4217 	 * done in the kernel even if nobody specified a filter;
4218 	 * that means that less buffer space is consumed in
4219 	 * the memory-mapped buffer.
4220 	 */
4221 	if (pcaphdr.caplen > (bpf_u_int32)handle->snapshot)
4222 		pcaphdr.caplen = handle->snapshot;
4223 
4224 	/* pass the packet to the user */
4225 	callback(user, &pcaphdr, bp);
4226 
4227 	return 1;
4228 }
4229 
4230 static int
4231 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
4232 		u_char *user)
4233 {
4234 	struct pcap_linux *handlep = handle->priv;
4235 	union thdr h;
4236 	int pkts = 0;
4237 	int ret;
4238 
4239 	/* wait for frames availability.*/
4240 	h.raw = RING_GET_CURRENT_FRAME(handle);
4241 	if (!packet_mmap_acquire(h.h2)) {
4242 		/*
4243 		 * The current frame is owned by the kernel; wait for
4244 		 * a frame to be handed to us.
4245 		 */
4246 		ret = pcap_wait_for_frames_mmap(handle);
4247 		if (ret) {
4248 			return ret;
4249 		}
4250 	}
4251 
4252 	/*
4253 	 * This can conceivably process more than INT_MAX packets,
4254 	 * which would overflow the packet count, causing it either
4255 	 * to look like a negative number, and thus cause us to
4256 	 * return a value that looks like an error, or overflow
4257 	 * back into positive territory, and thus cause us to
4258 	 * return a too-low count.
4259 	 *
4260 	 * Therefore, if the packet count is unlimited, we clip
4261 	 * it at INT_MAX; this routine is not expected to
4262 	 * process packets indefinitely, so that's not an issue.
4263 	 */
4264 	if (PACKET_COUNT_IS_UNLIMITED(max_packets))
4265 		max_packets = INT_MAX;
4266 
4267 	while (pkts < max_packets) {
4268 		/*
4269 		 * Get the current ring buffer frame, and break if
4270 		 * it's still owned by the kernel.
4271 		 */
4272 		h.raw = RING_GET_CURRENT_FRAME(handle);
4273 		if (!packet_mmap_acquire(h.h2))
4274 			break;
4275 
4276 		ret = pcap_handle_packet_mmap(
4277 				handle,
4278 				callback,
4279 				user,
4280 				h.raw,
4281 				h.h2->tp_len,
4282 				h.h2->tp_mac,
4283 				h.h2->tp_snaplen,
4284 				h.h2->tp_sec,
4285 				handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
4286 				VLAN_VALID(h.h2, h.h2),
4287 				h.h2->tp_vlan_tci,
4288 				VLAN_TPID(h.h2, h.h2));
4289 		if (ret == 1) {
4290 			pkts++;
4291 		} else if (ret < 0) {
4292 			return ret;
4293 		}
4294 
4295 		/*
4296 		 * Hand this block back to the kernel, and, if we're
4297 		 * counting blocks that need to be filtered in userland
4298 		 * after having been filtered by the kernel, count
4299 		 * the one we've just processed.
4300 		 */
4301 		packet_mmap_release(h.h2);
4302 		if (handlep->blocks_to_filter_in_userland > 0) {
4303 			handlep->blocks_to_filter_in_userland--;
4304 			if (handlep->blocks_to_filter_in_userland == 0) {
4305 				/*
4306 				 * No more blocks need to be filtered
4307 				 * in userland.
4308 				 */
4309 				handlep->filter_in_userland = 0;
4310 			}
4311 		}
4312 
4313 		/* next block */
4314 		if (++handle->offset >= handle->cc)
4315 			handle->offset = 0;
4316 
4317 		/* check for break loop condition*/
4318 		if (handle->break_loop) {
4319 			handle->break_loop = 0;
4320 			return PCAP_ERROR_BREAK;
4321 		}
4322 	}
4323 	return pkts;
4324 }
4325 
4326 #ifdef HAVE_TPACKET3
4327 static int
4328 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
4329 		u_char *user)
4330 {
4331 	struct pcap_linux *handlep = handle->priv;
4332 	union thdr h;
4333 	int pkts = 0;
4334 	int ret;
4335 
4336 again:
4337 	if (handlep->current_packet == NULL) {
4338 		/* wait for frames availability.*/
4339 		h.raw = RING_GET_CURRENT_FRAME(handle);
4340 		if (!packet_mmap_v3_acquire(h.h3)) {
4341 			/*
4342 			 * The current frame is owned by the kernel; wait
4343 			 * for a frame to be handed to us.
4344 			 */
4345 			ret = pcap_wait_for_frames_mmap(handle);
4346 			if (ret) {
4347 				return ret;
4348 			}
4349 		}
4350 	}
4351 	h.raw = RING_GET_CURRENT_FRAME(handle);
4352 	if (!packet_mmap_v3_acquire(h.h3)) {
4353 		if (pkts == 0 && handlep->timeout == 0) {
4354 			/* Block until we see a packet. */
4355 			goto again;
4356 		}
4357 		return pkts;
4358 	}
4359 
4360 	/*
4361 	 * This can conceivably process more than INT_MAX packets,
4362 	 * which would overflow the packet count, causing it either
4363 	 * to look like a negative number, and thus cause us to
4364 	 * return a value that looks like an error, or overflow
4365 	 * back into positive territory, and thus cause us to
4366 	 * return a too-low count.
4367 	 *
4368 	 * Therefore, if the packet count is unlimited, we clip
4369 	 * it at INT_MAX; this routine is not expected to
4370 	 * process packets indefinitely, so that's not an issue.
4371 	 */
4372 	if (PACKET_COUNT_IS_UNLIMITED(max_packets))
4373 		max_packets = INT_MAX;
4374 
4375 	while (pkts < max_packets) {
4376 		int packets_to_read;
4377 
4378 		if (handlep->current_packet == NULL) {
4379 			h.raw = RING_GET_CURRENT_FRAME(handle);
4380 			if (!packet_mmap_v3_acquire(h.h3))
4381 				break;
4382 
4383 			handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
4384 			handlep->packets_left = h.h3->hdr.bh1.num_pkts;
4385 		}
4386 		packets_to_read = handlep->packets_left;
4387 
4388 		if (packets_to_read > (max_packets - pkts)) {
4389 			/*
4390 			 * There are more packets in the buffer than
4391 			 * the number of packets we have left to
4392 			 * process to get up to the maximum number
4393 			 * of packets to process.  Only process enough
4394 			 * of them to get us up to that maximum.
4395 			 */
4396 			packets_to_read = max_packets - pkts;
4397 		}
4398 
4399 		while (packets_to_read-- && !handle->break_loop) {
4400 			struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
4401 			ret = pcap_handle_packet_mmap(
4402 					handle,
4403 					callback,
4404 					user,
4405 					handlep->current_packet,
4406 					tp3_hdr->tp_len,
4407 					tp3_hdr->tp_mac,
4408 					tp3_hdr->tp_snaplen,
4409 					tp3_hdr->tp_sec,
4410 					handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
4411 					VLAN_VALID(tp3_hdr, &tp3_hdr->hv1),
4412 					tp3_hdr->hv1.tp_vlan_tci,
4413 					VLAN_TPID(tp3_hdr, &tp3_hdr->hv1));
4414 			if (ret == 1) {
4415 				pkts++;
4416 			} else if (ret < 0) {
4417 				handlep->current_packet = NULL;
4418 				return ret;
4419 			}
4420 			handlep->current_packet += tp3_hdr->tp_next_offset;
4421 			handlep->packets_left--;
4422 		}
4423 
4424 		if (handlep->packets_left <= 0) {
4425 			/*
4426 			 * Hand this block back to the kernel, and, if
4427 			 * we're counting blocks that need to be
4428 			 * filtered in userland after having been
4429 			 * filtered by the kernel, count the one we've
4430 			 * just processed.
4431 			 */
4432 			packet_mmap_v3_release(h.h3);
4433 			if (handlep->blocks_to_filter_in_userland > 0) {
4434 				handlep->blocks_to_filter_in_userland--;
4435 				if (handlep->blocks_to_filter_in_userland == 0) {
4436 					/*
4437 					 * No more blocks need to be filtered
4438 					 * in userland.
4439 					 */
4440 					handlep->filter_in_userland = 0;
4441 				}
4442 			}
4443 
4444 			/* next block */
4445 			if (++handle->offset >= handle->cc)
4446 				handle->offset = 0;
4447 
4448 			handlep->current_packet = NULL;
4449 		}
4450 
4451 		/* check for break loop condition*/
4452 		if (handle->break_loop) {
4453 			handle->break_loop = 0;
4454 			return PCAP_ERROR_BREAK;
4455 		}
4456 	}
4457 	if (pkts == 0 && handlep->timeout == 0) {
4458 		/* Block until we see a packet. */
4459 		goto again;
4460 	}
4461 	return pkts;
4462 }
4463 #endif /* HAVE_TPACKET3 */
4464 
4465 /*
4466  *  Attach the given BPF code to the packet capture device.
4467  */
4468 static int
4469 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
4470 {
4471 	struct pcap_linux *handlep;
4472 	struct sock_fprog	fcode;
4473 	int			can_filter_in_kernel;
4474 	int			err = 0;
4475 	int			n, offset;
4476 
4477 	if (!handle)
4478 		return -1;
4479 	if (!filter) {
4480 	        pcapint_strlcpy(handle->errbuf, "setfilter: No filter specified",
4481 			PCAP_ERRBUF_SIZE);
4482 		return -1;
4483 	}
4484 
4485 	handlep = handle->priv;
4486 
4487 	/* Make our private copy of the filter */
4488 
4489 	if (pcapint_install_bpf_program(handle, filter) < 0)
4490 		/* pcapint_install_bpf_program() filled in errbuf */
4491 		return -1;
4492 
4493 	/*
4494 	 * Run user level packet filter by default. Will be overridden if
4495 	 * installing a kernel filter succeeds.
4496 	 */
4497 	handlep->filter_in_userland = 1;
4498 
4499 	/* Install kernel level filter if possible */
4500 
4501 #ifdef USHRT_MAX
4502 	if (handle->fcode.bf_len > USHRT_MAX) {
4503 		/*
4504 		 * fcode.len is an unsigned short for current kernel.
4505 		 * I have yet to see BPF-Code with that much
4506 		 * instructions but still it is possible. So for the
4507 		 * sake of correctness I added this check.
4508 		 */
4509 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
4510 		fcode.len = 0;
4511 		fcode.filter = NULL;
4512 		can_filter_in_kernel = 0;
4513 	} else
4514 #endif /* USHRT_MAX */
4515 	{
4516 		/*
4517 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
4518 		 * of struct bpf_program and of course the length field is
4519 		 * of different size. Pointed out by Sebastian
4520 		 *
4521 		 * Oh, and we also need to fix it up so that all "ret"
4522 		 * instructions with non-zero operands have MAXIMUM_SNAPLEN
4523 		 * as the operand if we're not capturing in memory-mapped
4524 		 * mode, and so that, if we're in cooked mode, all memory-
4525 		 * reference instructions use special magic offsets in
4526 		 * references to the link-layer header and assume that the
4527 		 * link-layer payload begins at 0; "fix_program()" will do
4528 		 * that.
4529 		 */
4530 		switch (fix_program(handle, &fcode)) {
4531 
4532 		case -1:
4533 		default:
4534 			/*
4535 			 * Fatal error; just quit.
4536 			 * (The "default" case shouldn't happen; we
4537 			 * return -1 for that reason.)
4538 			 */
4539 			return -1;
4540 
4541 		case 0:
4542 			/*
4543 			 * The program performed checks that we can't make
4544 			 * work in the kernel.
4545 			 */
4546 			can_filter_in_kernel = 0;
4547 			break;
4548 
4549 		case 1:
4550 			/*
4551 			 * We have a filter that'll work in the kernel.
4552 			 */
4553 			can_filter_in_kernel = 1;
4554 			break;
4555 		}
4556 	}
4557 
4558 	/*
4559 	 * NOTE: at this point, we've set both the "len" and "filter"
4560 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
4561 	 * those are the only members of the "sock_fprog" structure,
4562 	 * so we initialize every member of that structure.
4563 	 *
4564 	 * If there is anything in "fcode" that is not initialized,
4565 	 * it is either a field added in a later kernel, or it's
4566 	 * padding.
4567 	 *
4568 	 * If a new field is added, this code needs to be updated
4569 	 * to set it correctly.
4570 	 *
4571 	 * If there are no other fields, then:
4572 	 *
4573 	 *	if the Linux kernel looks at the padding, it's
4574 	 *	buggy;
4575 	 *
4576 	 *	if the Linux kernel doesn't look at the padding,
4577 	 *	then if some tool complains that we're passing
4578 	 *	uninitialized data to the kernel, then the tool
4579 	 *	is buggy and needs to understand that it's just
4580 	 *	padding.
4581 	 */
4582 	if (can_filter_in_kernel) {
4583 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
4584 		{
4585 			/*
4586 			 * Installation succeeded - using kernel filter,
4587 			 * so userland filtering not needed.
4588 			 */
4589 			handlep->filter_in_userland = 0;
4590 		}
4591 		else if (err == -1)	/* Non-fatal error */
4592 		{
4593 			/*
4594 			 * Print a warning if we weren't able to install
4595 			 * the filter for a reason other than "this kernel
4596 			 * isn't configured to support socket filters.
4597 			 */
4598 			if (errno == ENOMEM) {
4599 				/*
4600 				 * Either a kernel memory allocation
4601 				 * failure occurred, or there's too
4602 				 * much "other/option memory" allocated
4603 				 * for this socket.  Suggest that they
4604 				 * increase the "other/option memory"
4605 				 * limit.
4606 				 */
4607 				fprintf(stderr,
4608 				    "Warning: Couldn't allocate kernel memory for filter: try increasing net.core.optmem_max with sysctl\n");
4609 			} else if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
4610 				fprintf(stderr,
4611 				    "Warning: Kernel filter failed: %s\n",
4612 					pcap_strerror(errno));
4613 			}
4614 		}
4615 	}
4616 
4617 	/*
4618 	 * If we're not using the kernel filter, get rid of any kernel
4619 	 * filter that might've been there before, e.g. because the
4620 	 * previous filter could work in the kernel, or because some other
4621 	 * code attached a filter to the socket by some means other than
4622 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
4623 	 * filter out packets that would pass the new userland filter.
4624 	 */
4625 	if (handlep->filter_in_userland) {
4626 		if (reset_kernel_filter(handle) == -1) {
4627 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
4628 			    PCAP_ERRBUF_SIZE, errno,
4629 			    "can't remove kernel filter");
4630 			err = -2;	/* fatal error */
4631 		}
4632 	}
4633 
4634 	/*
4635 	 * Free up the copy of the filter that was made by "fix_program()".
4636 	 */
4637 	if (fcode.filter != NULL)
4638 		free(fcode.filter);
4639 
4640 	if (err == -2)
4641 		/* Fatal error */
4642 		return -1;
4643 
4644 	/*
4645 	 * If we're filtering in userland, there's nothing to do;
4646 	 * the new filter will be used for the next packet.
4647 	 */
4648 	if (handlep->filter_in_userland)
4649 		return 0;
4650 
4651 	/*
4652 	 * We're filtering in the kernel; the packets present in
4653 	 * all blocks currently in the ring were already filtered
4654 	 * by the old filter, and so will need to be filtered in
4655 	 * userland by the new filter.
4656 	 *
4657 	 * Get an upper bound for the number of such blocks; first,
4658 	 * walk the ring backward and count the free blocks.
4659 	 */
4660 	offset = handle->offset;
4661 	if (--offset < 0)
4662 		offset = handle->cc - 1;
4663 	for (n=0; n < handle->cc; ++n) {
4664 		if (--offset < 0)
4665 			offset = handle->cc - 1;
4666 		if (pcap_get_ring_frame_status(handle, offset) != TP_STATUS_KERNEL)
4667 			break;
4668 	}
4669 
4670 	/*
4671 	 * If we found free blocks, decrement the count of free
4672 	 * blocks by 1, just in case we lost a race with another
4673 	 * thread of control that was adding a packet while
4674 	 * we were counting and that had run the filter before
4675 	 * we changed it.
4676 	 *
4677 	 * XXX - could there be more than one block added in
4678 	 * this fashion?
4679 	 *
4680 	 * XXX - is there a way to avoid that race, e.g. somehow
4681 	 * wait for all packets that passed the old filter to
4682 	 * be added to the ring?
4683 	 */
4684 	if (n != 0)
4685 		n--;
4686 
4687 	/*
4688 	 * Set the count of blocks worth of packets to filter
4689 	 * in userland to the total number of blocks in the
4690 	 * ring minus the number of free blocks we found, and
4691 	 * turn on userland filtering.  (The count of blocks
4692 	 * worth of packets to filter in userland is guaranteed
4693 	 * not to be zero - n, above, couldn't be set to a
4694 	 * value > handle->cc, and if it were equal to
4695 	 * handle->cc, it wouldn't be zero, and thus would
4696 	 * be decremented to handle->cc - 1.)
4697 	 */
4698 	handlep->blocks_to_filter_in_userland = handle->cc - n;
4699 	handlep->filter_in_userland = 1;
4700 
4701 	return 0;
4702 }
4703 
4704 /*
4705  *  Return the index of the given device name. Fill ebuf and return
4706  *  -1 on failure.
4707  */
4708 static int
4709 iface_get_id(int fd, const char *device, char *ebuf)
4710 {
4711 	struct ifreq	ifr;
4712 
4713 	memset(&ifr, 0, sizeof(ifr));
4714 	pcapint_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4715 
4716 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4717 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4718 		    errno, "SIOCGIFINDEX");
4719 		return -1;
4720 	}
4721 
4722 	return ifr.ifr_ifindex;
4723 }
4724 
4725 /*
4726  *  Bind the socket associated with FD to the given device.
4727  *  Return 0 on success or a PCAP_ERROR_ value on a hard error.
4728  */
4729 static int
4730 iface_bind(int fd, int ifindex, char *ebuf, int protocol)
4731 {
4732 	struct sockaddr_ll	sll;
4733 	int			ret, err;
4734 	socklen_t		errlen = sizeof(err);
4735 
4736 	memset(&sll, 0, sizeof(sll));
4737 	sll.sll_family		= AF_PACKET;
4738 	sll.sll_ifindex		= ifindex < 0 ? 0 : ifindex;
4739 	sll.sll_protocol	= protocol;
4740 
4741 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4742 		if (errno == ENETDOWN) {
4743 			/*
4744 			 * Return a "network down" indication, so that
4745 			 * the application can report that rather than
4746 			 * saying we had a mysterious failure and
4747 			 * suggest that they report a problem to the
4748 			 * libpcap developers.
4749 			 */
4750 			return PCAP_ERROR_IFACE_NOT_UP;
4751 		}
4752 		if (errno == ENODEV) {
4753 			/*
4754 			 * There's nothing more to say, so clear the
4755 			 * error message.
4756 			 */
4757 			ebuf[0] = '\0';
4758 			ret = PCAP_ERROR_NO_SUCH_DEVICE;
4759 		} else {
4760 			ret = PCAP_ERROR;
4761 			pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4762 			    errno, "bind");
4763 		}
4764 		return ret;
4765 	}
4766 
4767 	/* Any pending errors, e.g., network is down? */
4768 
4769 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4770 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4771 		    errno, "getsockopt (SO_ERROR)");
4772 		return PCAP_ERROR;
4773 	}
4774 
4775 	if (err == ENETDOWN) {
4776 		/*
4777 		 * Return a "network down" indication, so that
4778 		 * the application can report that rather than
4779 		 * saying we had a mysterious failure and
4780 		 * suggest that they report a problem to the
4781 		 * libpcap developers.
4782 		 */
4783 		return PCAP_ERROR_IFACE_NOT_UP;
4784 	} else if (err > 0) {
4785 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4786 		    err, "bind");
4787 		return PCAP_ERROR;
4788 	}
4789 
4790 	return 0;
4791 }
4792 
4793 /*
4794  * Try to enter monitor mode.
4795  * If we have libnl, try to create a new monitor-mode device and
4796  * capture on that; otherwise, just say "not supported".
4797  */
4798 #ifdef HAVE_LIBNL
4799 static int
4800 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
4801 {
4802 	struct pcap_linux *handlep = handle->priv;
4803 	int ret;
4804 	char phydev_path[PATH_MAX+1];
4805 	struct nl80211_state nlstate;
4806 	struct ifreq ifr;
4807 	u_int n;
4808 
4809 	/*
4810 	 * Is this a mac80211 device?
4811 	 */
4812 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
4813 	if (ret < 0)
4814 		return ret;	/* error */
4815 	if (ret == 0)
4816 		return 0;	/* no error, but not mac80211 device */
4817 
4818 	/*
4819 	 * XXX - is this already a monN device?
4820 	 * If so, we're done.
4821 	 */
4822 
4823 	/*
4824 	 * OK, it's apparently a mac80211 device.
4825 	 * Try to find an unused monN device for it.
4826 	 */
4827 	ret = nl80211_init(handle, &nlstate, device);
4828 	if (ret != 0)
4829 		return ret;
4830 	for (n = 0; n < UINT_MAX; n++) {
4831 		/*
4832 		 * Try mon{n}.
4833 		 */
4834 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
4835 
4836 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
4837 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
4838 		if (ret == 1) {
4839 			/*
4840 			 * Success.  We don't clean up the libnl state
4841 			 * yet, as we'll be using it later.
4842 			 */
4843 			goto added;
4844 		}
4845 		if (ret < 0) {
4846 			/*
4847 			 * Hard failure.  Just return ret; handle->errbuf
4848 			 * has already been set.
4849 			 */
4850 			nl80211_cleanup(&nlstate);
4851 			return ret;
4852 		}
4853 	}
4854 
4855 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4856 	    "%s: No free monN interfaces", device);
4857 	nl80211_cleanup(&nlstate);
4858 	return PCAP_ERROR;
4859 
4860 added:
4861 
4862 #if 0
4863 	/*
4864 	 * Sleep for .1 seconds.
4865 	 */
4866 	delay.tv_sec = 0;
4867 	delay.tv_nsec = 500000000;
4868 	nanosleep(&delay, NULL);
4869 #endif
4870 
4871 	/*
4872 	 * If we haven't already done so, arrange to have
4873 	 * "pcap_close_all()" called when we exit.
4874 	 */
4875 	if (!pcapint_do_addexit(handle)) {
4876 		/*
4877 		 * "atexit()" failed; don't put the interface
4878 		 * in rfmon mode, just give up.
4879 		 */
4880 		del_mon_if(handle, sock_fd, &nlstate, device,
4881 		    handlep->mondevice);
4882 		nl80211_cleanup(&nlstate);
4883 		return PCAP_ERROR;
4884 	}
4885 
4886 	/*
4887 	 * Now configure the monitor interface up.
4888 	 */
4889 	memset(&ifr, 0, sizeof(ifr));
4890 	pcapint_strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
4891 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
4892 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4893 		    errno, "%s: Can't get flags for %s", device,
4894 		    handlep->mondevice);
4895 		del_mon_if(handle, sock_fd, &nlstate, device,
4896 		    handlep->mondevice);
4897 		nl80211_cleanup(&nlstate);
4898 		return PCAP_ERROR;
4899 	}
4900 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
4901 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4902 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4903 		    errno, "%s: Can't set flags for %s", device,
4904 		    handlep->mondevice);
4905 		del_mon_if(handle, sock_fd, &nlstate, device,
4906 		    handlep->mondevice);
4907 		nl80211_cleanup(&nlstate);
4908 		return PCAP_ERROR;
4909 	}
4910 
4911 	/*
4912 	 * Success.  Clean up the libnl state.
4913 	 */
4914 	nl80211_cleanup(&nlstate);
4915 
4916 	/*
4917 	 * Note that we have to delete the monitor device when we close
4918 	 * the handle.
4919 	 */
4920 	handlep->must_do_on_close |= MUST_DELETE_MONIF;
4921 
4922 	/*
4923 	 * Add this to the list of pcaps to close when we exit.
4924 	 */
4925 	pcapint_add_to_pcaps_to_close(handle);
4926 
4927 	return 1;
4928 }
4929 #else /* HAVE_LIBNL */
4930 static int
4931 enter_rfmon_mode(pcap_t *handle _U_, int sock_fd _U_, const char *device _U_)
4932 {
4933 	/*
4934 	 * We don't have libnl, so we can't do monitor mode.
4935 	 */
4936 	return 0;
4937 }
4938 #endif /* HAVE_LIBNL */
4939 
4940 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
4941 /*
4942  * Map SOF_TIMESTAMPING_ values to PCAP_TSTAMP_ values.
4943  */
4944 static const struct {
4945 	int soft_timestamping_val;
4946 	int pcap_tstamp_val;
4947 } sof_ts_type_map[3] = {
4948 	{ SOF_TIMESTAMPING_SOFTWARE, PCAP_TSTAMP_HOST },
4949 	{ SOF_TIMESTAMPING_SYS_HARDWARE, PCAP_TSTAMP_ADAPTER },
4950 	{ SOF_TIMESTAMPING_RAW_HARDWARE, PCAP_TSTAMP_ADAPTER_UNSYNCED }
4951 };
4952 #define NUM_SOF_TIMESTAMPING_TYPES	(sizeof sof_ts_type_map / sizeof sof_ts_type_map[0])
4953 
4954 /*
4955  * Set the list of time stamping types to include all types.
4956  */
4957 static int
4958 iface_set_all_ts_types(pcap_t *handle, char *ebuf)
4959 {
4960 	u_int i;
4961 
4962 	handle->tstamp_type_list = malloc(NUM_SOF_TIMESTAMPING_TYPES * sizeof(u_int));
4963 	if (handle->tstamp_type_list == NULL) {
4964 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4965 		    errno, "malloc");
4966 		return -1;
4967 	}
4968 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++)
4969 		handle->tstamp_type_list[i] = sof_ts_type_map[i].pcap_tstamp_val;
4970 	handle->tstamp_type_count = NUM_SOF_TIMESTAMPING_TYPES;
4971 	return 0;
4972 }
4973 
4974 /*
4975  * Get a list of time stamp types.
4976  */
4977 #ifdef ETHTOOL_GET_TS_INFO
4978 static int
4979 iface_get_ts_types(const char *device, pcap_t *handle, char *ebuf)
4980 {
4981 	int fd;
4982 	struct ifreq ifr;
4983 	struct ethtool_ts_info info;
4984 	int num_ts_types;
4985 	u_int i, j;
4986 
4987 	/*
4988 	 * This doesn't apply to the "any" device; you can't say "turn on
4989 	 * hardware time stamping for all devices that exist now and arrange
4990 	 * that it be turned on for any device that appears in the future",
4991 	 * and not all devices even necessarily *support* hardware time
4992 	 * stamping, so don't report any time stamp types.
4993 	 */
4994 	if (strcmp(device, "any") == 0) {
4995 		handle->tstamp_type_list = NULL;
4996 		return 0;
4997 	}
4998 
4999 	/*
5000 	 * Create a socket from which to fetch time stamping capabilities.
5001 	 */
5002 	fd = get_if_ioctl_socket();
5003 	if (fd < 0) {
5004 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5005 		    errno, "socket for SIOCETHTOOL(ETHTOOL_GET_TS_INFO)");
5006 		return -1;
5007 	}
5008 
5009 	memset(&ifr, 0, sizeof(ifr));
5010 	pcapint_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5011 	memset(&info, 0, sizeof(info));
5012 	info.cmd = ETHTOOL_GET_TS_INFO;
5013 	ifr.ifr_data = (caddr_t)&info;
5014 	if (ioctl(fd, SIOCETHTOOL, &ifr) == -1) {
5015 		int save_errno = errno;
5016 
5017 		close(fd);
5018 		switch (save_errno) {
5019 
5020 		case EOPNOTSUPP:
5021 		case EINVAL:
5022 			/*
5023 			 * OK, this OS version or driver doesn't support
5024 			 * asking for the time stamping types, so let's
5025 			 * just return all the possible types.
5026 			 */
5027 			if (iface_set_all_ts_types(handle, ebuf) == -1)
5028 				return -1;
5029 			return 0;
5030 
5031 		case ENODEV:
5032 			/*
5033 			 * OK, no such device.
5034 			 * The user will find that out when they try to
5035 			 * activate the device; just return an empty
5036 			 * list of time stamp types.
5037 			 */
5038 			handle->tstamp_type_list = NULL;
5039 			return 0;
5040 
5041 		default:
5042 			/*
5043 			 * Other error.
5044 			 */
5045 			pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5046 			    save_errno,
5047 			    "%s: SIOCETHTOOL(ETHTOOL_GET_TS_INFO) ioctl failed",
5048 			    device);
5049 			return -1;
5050 		}
5051 	}
5052 	close(fd);
5053 
5054 	/*
5055 	 * Do we support hardware time stamping of *all* packets?
5056 	 */
5057 	if (!(info.rx_filters & (1 << HWTSTAMP_FILTER_ALL))) {
5058 		/*
5059 		 * No, so don't report any time stamp types.
5060 		 *
5061 		 * XXX - some devices either don't report
5062 		 * HWTSTAMP_FILTER_ALL when they do support it, or
5063 		 * report HWTSTAMP_FILTER_ALL but map it to only
5064 		 * time stamping a few PTP packets.  See
5065 		 * http://marc.info/?l=linux-netdev&m=146318183529571&w=2
5066 		 *
5067 		 * Maybe that got fixed later.
5068 		 */
5069 		handle->tstamp_type_list = NULL;
5070 		return 0;
5071 	}
5072 
5073 	num_ts_types = 0;
5074 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
5075 		if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val)
5076 			num_ts_types++;
5077 	}
5078 	if (num_ts_types != 0) {
5079 		handle->tstamp_type_list = malloc(num_ts_types * sizeof(u_int));
5080 		if (handle->tstamp_type_list == NULL) {
5081 			pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5082 			    errno, "malloc");
5083 			return -1;
5084 		}
5085 		for (i = 0, j = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
5086 			if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) {
5087 				handle->tstamp_type_list[j] = sof_ts_type_map[i].pcap_tstamp_val;
5088 				j++;
5089 			}
5090 		}
5091 		handle->tstamp_type_count = num_ts_types;
5092 	} else
5093 		handle->tstamp_type_list = NULL;
5094 
5095 	return 0;
5096 }
5097 #else /* ETHTOOL_GET_TS_INFO */
5098 static int
5099 iface_get_ts_types(const char *device, pcap_t *handle, char *ebuf)
5100 {
5101 	/*
5102 	 * This doesn't apply to the "any" device; you can't say "turn on
5103 	 * hardware time stamping for all devices that exist now and arrange
5104 	 * that it be turned on for any device that appears in the future",
5105 	 * and not all devices even necessarily *support* hardware time
5106 	 * stamping, so don't report any time stamp types.
5107 	 */
5108 	if (strcmp(device, "any") == 0) {
5109 		handle->tstamp_type_list = NULL;
5110 		return 0;
5111 	}
5112 
5113 	/*
5114 	 * We don't have an ioctl to use to ask what's supported,
5115 	 * so say we support everything.
5116 	 */
5117 	if (iface_set_all_ts_types(handle, ebuf) == -1)
5118 		return -1;
5119 	return 0;
5120 }
5121 #endif /* ETHTOOL_GET_TS_INFO */
5122 #else  /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
5123 static int
5124 iface_get_ts_types(const char *device _U_, pcap_t *p _U_, char *ebuf _U_)
5125 {
5126 	/*
5127 	 * Nothing to fetch, so it always "succeeds".
5128 	 */
5129 	return 0;
5130 }
5131 #endif /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
5132 
5133 /*
5134  * Find out if we have any form of fragmentation/reassembly offloading.
5135  *
5136  * We do so using SIOCETHTOOL checking for various types of offloading;
5137  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
5138  * of the types of offloading, there's nothing we can do to check, so
5139  * we just say "no, we don't".
5140  *
5141  * We treat EOPNOTSUPP, EINVAL and, if eperm_ok is true, EPERM as
5142  * indications that the operation isn't supported.  We do EPERM
5143  * weirdly because the SIOCETHTOOL code in later kernels 1) doesn't
5144  * support ETHTOOL_GUFO, 2) also doesn't include it in the list
5145  * of ethtool operations that don't require CAP_NET_ADMIN privileges,
5146  * and 3) does the "is this permitted" check before doing the "is
5147  * this even supported" check, so it fails with "this is not permitted"
5148  * rather than "this is not even supported".  To work around this
5149  * annoyance, we only treat EPERM as an error for the first feature,
5150  * and assume that they all do the same permission checks, so if the
5151  * first one is allowed all the others are allowed if supported.
5152  */
5153 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
5154 static int
5155 iface_ethtool_flag_ioctl(pcap_t *handle, int cmd, const char *cmdname,
5156     int eperm_ok)
5157 {
5158 	struct ifreq	ifr;
5159 	struct ethtool_value eval;
5160 
5161 	memset(&ifr, 0, sizeof(ifr));
5162 	pcapint_strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
5163 	eval.cmd = cmd;
5164 	eval.data = 0;
5165 	ifr.ifr_data = (caddr_t)&eval;
5166 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
5167 		if (errno == EOPNOTSUPP || errno == EINVAL ||
5168 		    (errno == EPERM && eperm_ok)) {
5169 			/*
5170 			 * OK, let's just return 0, which, in our
5171 			 * case, either means "no, what we're asking
5172 			 * about is not enabled" or "all the flags
5173 			 * are clear (i.e., nothing is enabled)".
5174 			 */
5175 			return 0;
5176 		}
5177 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5178 		    errno, "%s: SIOCETHTOOL(%s) ioctl failed",
5179 		    handle->opt.device, cmdname);
5180 		return -1;
5181 	}
5182 	return eval.data;
5183 }
5184 
5185 /*
5186  * XXX - it's annoying that we have to check for offloading at all, but,
5187  * given that we have to, it's still annoying that we have to check for
5188  * particular types of offloading, especially that shiny new types of
5189  * offloading may be added - and, worse, may not be checkable with
5190  * a particular ETHTOOL_ operation; ETHTOOL_GFEATURES would, in
5191  * theory, give those to you, but the actual flags being used are
5192  * opaque (defined in a non-uapi header), and there doesn't seem to
5193  * be any obvious way to ask the kernel what all the offloading flags
5194  * are - at best, you can ask for a set of strings(!) to get *names*
5195  * for various flags.  (That whole mechanism appears to have been
5196  * designed for the sole purpose of letting ethtool report flags
5197  * by name and set flags by name, with the names having no semantics
5198  * ethtool understands.)
5199  */
5200 static int
5201 iface_get_offload(pcap_t *handle)
5202 {
5203 	int ret;
5204 
5205 #ifdef ETHTOOL_GTSO
5206 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO", 0);
5207 	if (ret == -1)
5208 		return -1;
5209 	if (ret)
5210 		return 1;	/* TCP segmentation offloading on */
5211 #endif
5212 
5213 #ifdef ETHTOOL_GGSO
5214 	/*
5215 	 * XXX - will this cause large unsegmented packets to be
5216 	 * handed to PF_PACKET sockets on transmission?  If not,
5217 	 * this need not be checked.
5218 	 */
5219 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO", 0);
5220 	if (ret == -1)
5221 		return -1;
5222 	if (ret)
5223 		return 1;	/* generic segmentation offloading on */
5224 #endif
5225 
5226 #ifdef ETHTOOL_GFLAGS
5227 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS", 0);
5228 	if (ret == -1)
5229 		return -1;
5230 	if (ret & ETH_FLAG_LRO)
5231 		return 1;	/* large receive offloading on */
5232 #endif
5233 
5234 #ifdef ETHTOOL_GGRO
5235 	/*
5236 	 * XXX - will this cause large reassembled packets to be
5237 	 * handed to PF_PACKET sockets on receipt?  If not,
5238 	 * this need not be checked.
5239 	 */
5240 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO", 0);
5241 	if (ret == -1)
5242 		return -1;
5243 	if (ret)
5244 		return 1;	/* generic (large) receive offloading on */
5245 #endif
5246 
5247 #ifdef ETHTOOL_GUFO
5248 	/*
5249 	 * Do this one last, as support for it was removed in later
5250 	 * kernels, and it fails with EPERM on those kernels rather
5251 	 * than with EOPNOTSUPP (see explanation in comment for
5252 	 * iface_ethtool_flag_ioctl()).
5253 	 */
5254 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO", 1);
5255 	if (ret == -1)
5256 		return -1;
5257 	if (ret)
5258 		return 1;	/* UDP fragmentation offloading on */
5259 #endif
5260 
5261 	return 0;
5262 }
5263 #else /* SIOCETHTOOL */
5264 static int
5265 iface_get_offload(pcap_t *handle _U_)
5266 {
5267 	/*
5268 	 * XXX - do we need to get this information if we don't
5269 	 * have the ethtool ioctls?  If so, how do we do that?
5270 	 */
5271 	return 0;
5272 }
5273 #endif /* SIOCETHTOOL */
5274 
5275 static struct dsa_proto {
5276 	const char *name;
5277 	bpf_u_int32 linktype;
5278 } dsa_protos[] = {
5279 	/*
5280 	 * None is special and indicates that the interface does not have
5281 	 * any tagging protocol configured, and is therefore a standard
5282 	 * Ethernet interface.
5283 	 */
5284 	{ "none", DLT_EN10MB },
5285 	{ "brcm", DLT_DSA_TAG_BRCM },
5286 	{ "brcm-prepend", DLT_DSA_TAG_BRCM_PREPEND },
5287 	{ "dsa", DLT_DSA_TAG_DSA },
5288 	{ "edsa", DLT_DSA_TAG_EDSA },
5289 };
5290 
5291 static int
5292 iface_dsa_get_proto_info(const char *device, pcap_t *handle)
5293 {
5294 	char *pathstr;
5295 	unsigned int i;
5296 	/*
5297 	 * Make this significantly smaller than PCAP_ERRBUF_SIZE;
5298 	 * the tag *shouldn't* have some huge long name, and making
5299 	 * it smaller keeps newer versions of GCC from whining that
5300 	 * the error message if we don't support the tag could
5301 	 * overflow the error message buffer.
5302 	 */
5303 	char buf[128];
5304 	ssize_t r;
5305 	int fd;
5306 
5307 	fd = asprintf(&pathstr, "/sys/class/net/%s/dsa/tagging", device);
5308 	if (fd < 0) {
5309 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5310 					  fd, "asprintf");
5311 		return PCAP_ERROR;
5312 	}
5313 
5314 	fd = open(pathstr, O_RDONLY);
5315 	free(pathstr);
5316 	/*
5317 	 * This is not fatal, kernel >= 4.20 *might* expose this attribute
5318 	 */
5319 	if (fd < 0)
5320 		return 0;
5321 
5322 	r = read(fd, buf, sizeof(buf) - 1);
5323 	if (r <= 0) {
5324 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5325 					  errno, "read");
5326 		close(fd);
5327 		return PCAP_ERROR;
5328 	}
5329 	close(fd);
5330 
5331 	/*
5332 	 * Buffer should be LF terminated.
5333 	 */
5334 	if (buf[r - 1] == '\n')
5335 		r--;
5336 	buf[r] = '\0';
5337 
5338 	for (i = 0; i < sizeof(dsa_protos) / sizeof(dsa_protos[0]); i++) {
5339 		if (strlen(dsa_protos[i].name) == (size_t)r &&
5340 		    strcmp(buf, dsa_protos[i].name) == 0) {
5341 			handle->linktype = dsa_protos[i].linktype;
5342 			switch (dsa_protos[i].linktype) {
5343 			case DLT_EN10MB:
5344 				return 0;
5345 			default:
5346 				return 1;
5347 			}
5348 		}
5349 	}
5350 
5351 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5352 		      "unsupported DSA tag: %s", buf);
5353 
5354 	return PCAP_ERROR;
5355 }
5356 
5357 /*
5358  *  Query the kernel for the MTU of the given interface.
5359  */
5360 static int
5361 iface_get_mtu(int fd, const char *device, char *ebuf)
5362 {
5363 	struct ifreq	ifr;
5364 
5365 	if (!device)
5366 		return BIGGER_THAN_ALL_MTUS;
5367 
5368 	memset(&ifr, 0, sizeof(ifr));
5369 	pcapint_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5370 
5371 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5372 		pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5373 		    errno, "SIOCGIFMTU");
5374 		return -1;
5375 	}
5376 
5377 	return ifr.ifr_mtu;
5378 }
5379 
5380 /*
5381  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5382  */
5383 static int
5384 iface_get_arptype(int fd, const char *device, char *ebuf)
5385 {
5386 	struct ifreq	ifr;
5387 	int		ret;
5388 
5389 	memset(&ifr, 0, sizeof(ifr));
5390 	pcapint_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5391 
5392 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5393 		if (errno == ENODEV) {
5394 			/*
5395 			 * No such device.
5396 			 *
5397 			 * There's nothing more to say, so clear
5398 			 * the error message.
5399 			 */
5400 			ret = PCAP_ERROR_NO_SUCH_DEVICE;
5401 			ebuf[0] = '\0';
5402 		} else {
5403 			ret = PCAP_ERROR;
5404 			pcapint_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5405 			    errno, "SIOCGIFHWADDR");
5406 		}
5407 		return ret;
5408 	}
5409 
5410 	return ifr.ifr_hwaddr.sa_family;
5411 }
5412 
5413 static int
5414 fix_program(pcap_t *handle, struct sock_fprog *fcode)
5415 {
5416 	struct pcap_linux *handlep = handle->priv;
5417 	size_t prog_size;
5418 	register int i;
5419 	register struct bpf_insn *p;
5420 	struct bpf_insn *f;
5421 	int len;
5422 
5423 	/*
5424 	 * Make a copy of the filter, and modify that copy if
5425 	 * necessary.
5426 	 */
5427 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5428 	len = handle->fcode.bf_len;
5429 	f = (struct bpf_insn *)malloc(prog_size);
5430 	if (f == NULL) {
5431 		pcapint_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5432 		    errno, "malloc");
5433 		return -1;
5434 	}
5435 	memcpy(f, handle->fcode.bf_insns, prog_size);
5436 	fcode->len = len;
5437 	fcode->filter = (struct sock_filter *) f;
5438 
5439 	for (i = 0; i < len; ++i) {
5440 		p = &f[i];
5441 		/*
5442 		 * What type of instruction is this?
5443 		 */
5444 		switch (BPF_CLASS(p->code)) {
5445 
5446 		case BPF_LD:
5447 		case BPF_LDX:
5448 			/*
5449 			 * It's a load instruction; is it loading
5450 			 * from the packet?
5451 			 */
5452 			switch (BPF_MODE(p->code)) {
5453 
5454 			case BPF_ABS:
5455 			case BPF_IND:
5456 			case BPF_MSH:
5457 				/*
5458 				 * Yes; are we in cooked mode?
5459 				 */
5460 				if (handlep->cooked) {
5461 					/*
5462 					 * Yes, so we need to fix this
5463 					 * instruction.
5464 					 */
5465 					if (fix_offset(handle, p) < 0) {
5466 						/*
5467 						 * We failed to do so.
5468 						 * Return 0, so our caller
5469 						 * knows to punt to userland.
5470 						 */
5471 						return 0;
5472 					}
5473 				}
5474 				break;
5475 			}
5476 			break;
5477 		}
5478 	}
5479 	return 1;	/* we succeeded */
5480 }
5481 
5482 static int
5483 fix_offset(pcap_t *handle, struct bpf_insn *p)
5484 {
5485 	/*
5486 	 * Existing references to auxiliary data shouldn't be adjusted.
5487 	 *
5488 	 * Note that SKF_AD_OFF is negative, but p->k is unsigned, so
5489 	 * we use >= and cast SKF_AD_OFF to unsigned.
5490 	 */
5491 	if (p->k >= (bpf_u_int32)SKF_AD_OFF)
5492 		return 0;
5493 	if (handle->linktype == DLT_LINUX_SLL2) {
5494 		/*
5495 		 * What's the offset?
5496 		 */
5497 		if (p->k >= SLL2_HDR_LEN) {
5498 			/*
5499 			 * It's within the link-layer payload; that starts
5500 			 * at an offset of 0, as far as the kernel packet
5501 			 * filter is concerned, so subtract the length of
5502 			 * the link-layer header.
5503 			 */
5504 			p->k -= SLL2_HDR_LEN;
5505 		} else if (p->k == 0) {
5506 			/*
5507 			 * It's the protocol field; map it to the
5508 			 * special magic kernel offset for that field.
5509 			 */
5510 			p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5511 		} else if (p->k == 4) {
5512 			/*
5513 			 * It's the ifindex field; map it to the
5514 			 * special magic kernel offset for that field.
5515 			 */
5516 			p->k = SKF_AD_OFF + SKF_AD_IFINDEX;
5517 		} else if (p->k == 10) {
5518 			/*
5519 			 * It's the packet type field; map it to the
5520 			 * special magic kernel offset for that field.
5521 			 */
5522 			p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5523 		} else if ((bpf_int32)(p->k) > 0) {
5524 			/*
5525 			 * It's within the header, but it's not one of
5526 			 * those fields; we can't do that in the kernel,
5527 			 * so punt to userland.
5528 			 */
5529 			return -1;
5530 		}
5531 	} else {
5532 		/*
5533 		 * What's the offset?
5534 		 */
5535 		if (p->k >= SLL_HDR_LEN) {
5536 			/*
5537 			 * It's within the link-layer payload; that starts
5538 			 * at an offset of 0, as far as the kernel packet
5539 			 * filter is concerned, so subtract the length of
5540 			 * the link-layer header.
5541 			 */
5542 			p->k -= SLL_HDR_LEN;
5543 		} else if (p->k == 0) {
5544 			/*
5545 			 * It's the packet type field; map it to the
5546 			 * special magic kernel offset for that field.
5547 			 */
5548 			p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5549 		} else if (p->k == 14) {
5550 			/*
5551 			 * It's the protocol field; map it to the
5552 			 * special magic kernel offset for that field.
5553 			 */
5554 			p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5555 		} else if ((bpf_int32)(p->k) > 0) {
5556 			/*
5557 			 * It's within the header, but it's not one of
5558 			 * those fields; we can't do that in the kernel,
5559 			 * so punt to userland.
5560 			 */
5561 			return -1;
5562 		}
5563 	}
5564 	return 0;
5565 }
5566 
5567 static int
5568 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5569 {
5570 	int total_filter_on = 0;
5571 	int save_mode;
5572 	int ret;
5573 	int save_errno;
5574 
5575 	/*
5576 	 * The socket filter code doesn't discard all packets queued
5577 	 * up on the socket when the filter is changed; this means
5578 	 * that packets that don't match the new filter may show up
5579 	 * after the new filter is put onto the socket, if those
5580 	 * packets haven't yet been read.
5581 	 *
5582 	 * This means, for example, that if you do a tcpdump capture
5583 	 * with a filter, the first few packets in the capture might
5584 	 * be packets that wouldn't have passed the filter.
5585 	 *
5586 	 * We therefore discard all packets queued up on the socket
5587 	 * when setting a kernel filter.  (This isn't an issue for
5588 	 * userland filters, as the userland filtering is done after
5589 	 * packets are queued up.)
5590 	 *
5591 	 * To flush those packets, we put the socket in read-only mode,
5592 	 * and read packets from the socket until there are no more to
5593 	 * read.
5594 	 *
5595 	 * In order to keep that from being an infinite loop - i.e.,
5596 	 * to keep more packets from arriving while we're draining
5597 	 * the queue - we put the "total filter", which is a filter
5598 	 * that rejects all packets, onto the socket before draining
5599 	 * the queue.
5600 	 *
5601 	 * This code deliberately ignores any errors, so that you may
5602 	 * get bogus packets if an error occurs, rather than having
5603 	 * the filtering done in userland even if it could have been
5604 	 * done in the kernel.
5605 	 */
5606 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5607 		       &total_fcode, sizeof(total_fcode)) == 0) {
5608 		char drain[1];
5609 
5610 		/*
5611 		 * Note that we've put the total filter onto the socket.
5612 		 */
5613 		total_filter_on = 1;
5614 
5615 		/*
5616 		 * Save the socket's current mode, and put it in
5617 		 * non-blocking mode; we drain it by reading packets
5618 		 * until we get an error (which is normally a
5619 		 * "nothing more to be read" error).
5620 		 */
5621 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5622 		if (save_mode == -1) {
5623 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
5624 			    PCAP_ERRBUF_SIZE, errno,
5625 			    "can't get FD flags when changing filter");
5626 			return -2;
5627 		}
5628 		if (fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) < 0) {
5629 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
5630 			    PCAP_ERRBUF_SIZE, errno,
5631 			    "can't set nonblocking mode when changing filter");
5632 			return -2;
5633 		}
5634 		while (recv(handle->fd, &drain, sizeof drain, MSG_TRUNC) >= 0)
5635 			;
5636 		save_errno = errno;
5637 		if (save_errno != EAGAIN) {
5638 			/*
5639 			 * Fatal error.
5640 			 *
5641 			 * If we can't restore the mode or reset the
5642 			 * kernel filter, there's nothing we can do.
5643 			 */
5644 			(void)fcntl(handle->fd, F_SETFL, save_mode);
5645 			(void)reset_kernel_filter(handle);
5646 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
5647 			    PCAP_ERRBUF_SIZE, save_errno,
5648 			    "recv failed when changing filter");
5649 			return -2;
5650 		}
5651 		if (fcntl(handle->fd, F_SETFL, save_mode) == -1) {
5652 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
5653 			    PCAP_ERRBUF_SIZE, errno,
5654 			    "can't restore FD flags when changing filter");
5655 			return -2;
5656 		}
5657 	}
5658 
5659 	/*
5660 	 * Now attach the new filter.
5661 	 */
5662 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5663 			 fcode, sizeof(*fcode));
5664 	if (ret == -1 && total_filter_on) {
5665 		/*
5666 		 * Well, we couldn't set that filter on the socket,
5667 		 * but we could set the total filter on the socket.
5668 		 *
5669 		 * This could, for example, mean that the filter was
5670 		 * too big to put into the kernel, so we'll have to
5671 		 * filter in userland; in any case, we'll be doing
5672 		 * filtering in userland, so we need to remove the
5673 		 * total filter so we see packets.
5674 		 */
5675 		save_errno = errno;
5676 
5677 		/*
5678 		 * If this fails, we're really screwed; we have the
5679 		 * total filter on the socket, and it won't come off.
5680 		 * Report it as a fatal error.
5681 		 */
5682 		if (reset_kernel_filter(handle) == -1) {
5683 			pcapint_fmt_errmsg_for_errno(handle->errbuf,
5684 			    PCAP_ERRBUF_SIZE, errno,
5685 			    "can't remove kernel total filter");
5686 			return -2;	/* fatal error */
5687 		}
5688 
5689 		errno = save_errno;
5690 	}
5691 	return ret;
5692 }
5693 
5694 static int
5695 reset_kernel_filter(pcap_t *handle)
5696 {
5697 	int ret;
5698 	/*
5699 	 * setsockopt() barfs unless it get a dummy parameter.
5700 	 * valgrind whines unless the value is initialized,
5701 	 * as it has no idea that setsockopt() ignores its
5702 	 * parameter.
5703 	 */
5704 	int dummy = 0;
5705 
5706 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5707 				   &dummy, sizeof(dummy));
5708 	/*
5709 	 * Ignore ENOENT - it means "we don't have a filter", so there
5710 	 * was no filter to remove, and there's still no filter.
5711 	 *
5712 	 * Also ignore ENONET, as a lot of kernel versions had a
5713 	 * typo where ENONET, rather than ENOENT, was returned.
5714 	 */
5715 	if (ret == -1 && errno != ENOENT && errno != ENONET)
5716 		return -1;
5717 	return 0;
5718 }
5719 
5720 int
5721 pcap_set_protocol_linux(pcap_t *p, int protocol)
5722 {
5723 	if (pcapint_check_activated(p))
5724 		return (PCAP_ERROR_ACTIVATED);
5725 	p->opt.protocol = protocol;
5726 	return (0);
5727 }
5728 
5729 /*
5730  * Libpcap version string.
5731  */
5732 const char *
5733 pcap_lib_version(void)
5734 {
5735 #if defined(HAVE_TPACKET3)
5736 	return (PCAP_VERSION_STRING " (with TPACKET_V3)");
5737 #else
5738 	return (PCAP_VERSION_STRING " (with TPACKET_V2)");
5739 #endif
5740 }
5741