xref: /freebsd/contrib/libpcap/pcap-linux.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
1 /*
2  *  pcap-linux.c: Packet capture interface to the Linux kernel
3  *
4  *  Copyright (c) 2000 Torsten Landschoff <torsten@debian.org>
5  *  		       Sebastian Krahmer  <krahmer@cs.uni-potsdam.de>
6  *
7  *  License: BSD
8  *
9  *  Redistribution and use in source and binary forms, with or without
10  *  modification, are permitted provided that the following conditions
11  *  are met:
12  *
13  *  1. Redistributions of source code must retain the above copyright
14  *     notice, this list of conditions and the following disclaimer.
15  *  2. Redistributions in binary form must reproduce the above copyright
16  *     notice, this list of conditions and the following disclaimer in
17  *     the documentation and/or other materials provided with the
18  *     distribution.
19  *  3. The names of the authors may not be used to endorse or promote
20  *     products derived from this software without specific prior
21  *     written permission.
22  *
23  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR
24  *  IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED
25  *  WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
26  *
27  *  Modifications:     Added PACKET_MMAP support
28  *                     Paolo Abeni <paolo.abeni@email.it>
29  *                     Added TPACKET_V3 support
30  *                     Gabor Tatarka <gabor.tatarka@ericsson.com>
31  *
32  *                     based on previous works of:
33  *                     Simon Patarin <patarin@cs.unibo.it>
34  *                     Phil Wood <cpw@lanl.gov>
35  *
36  * Monitor-mode support for mac80211 includes code taken from the iw
37  * command; the copyright notice for that code is
38  *
39  * Copyright (c) 2007, 2008	Johannes Berg
40  * Copyright (c) 2007		Andy Lutomirski
41  * Copyright (c) 2007		Mike Kershaw
42  * Copyright (c) 2008		Gábor Stefanik
43  *
44  * All rights reserved.
45  *
46  * Redistribution and use in source and binary forms, with or without
47  * modification, are permitted provided that the following conditions
48  * are met:
49  * 1. Redistributions of source code must retain the above copyright
50  *    notice, this list of conditions and the following disclaimer.
51  * 2. Redistributions in binary form must reproduce the above copyright
52  *    notice, this list of conditions and the following disclaimer in the
53  *    documentation and/or other materials provided with the distribution.
54  * 3. The name of the author may not be used to endorse or promote products
55  *    derived from this software without specific prior written permission.
56  *
57  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
58  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
59  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
60  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
61  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
62  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
63  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
64  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
65  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
67  * SUCH DAMAGE.
68  */
69 
70 
71 #define _GNU_SOURCE
72 
73 #ifdef HAVE_CONFIG_H
74 #include <config.h>
75 #endif
76 
77 #include <errno.h>
78 #include <stdio.h>
79 #include <stdlib.h>
80 #include <unistd.h>
81 #include <fcntl.h>
82 #include <string.h>
83 #include <limits.h>
84 #include <sys/stat.h>
85 #include <sys/socket.h>
86 #include <sys/ioctl.h>
87 #include <sys/utsname.h>
88 #include <sys/mman.h>
89 #include <linux/if.h>
90 #include <linux/if_packet.h>
91 #include <linux/sockios.h>
92 #include <linux/ethtool.h>
93 #include <netinet/in.h>
94 #include <linux/if_ether.h>
95 #include <linux/if_arp.h>
96 #include <poll.h>
97 #include <dirent.h>
98 #include <sys/eventfd.h>
99 
100 #include "pcap-int.h"
101 #include "pcap/sll.h"
102 #include "pcap/vlan.h"
103 #include "pcap/can_socketcan.h"
104 
105 #include "diag-control.h"
106 
107 /*
108  * We require TPACKET_V2 support.
109  */
110 #ifndef TPACKET2_HDRLEN
111 #error "Libpcap will only work if TPACKET_V2 is supported; you must build for a 2.6.27 or later kernel"
112 #endif
113 
114 /* check for memory mapped access avaibility. We assume every needed
115  * struct is defined if the macro TPACKET_HDRLEN is defined, because it
116  * uses many ring related structs and macros */
117 #ifdef TPACKET3_HDRLEN
118 # define HAVE_TPACKET3
119 #endif /* TPACKET3_HDRLEN */
120 
121 /*
122  * Not all compilers that are used to compile code to run on Linux have
123  * these builtins.  For example, older versions of GCC don't, and at
124  * least some people are doing cross-builds for MIPS with older versions
125  * of GCC.
126  */
127 #ifndef HAVE___ATOMIC_LOAD_N
128 #define __atomic_load_n(ptr, memory_model)		(*(ptr))
129 #endif
130 #ifndef HAVE___ATOMIC_STORE_N
131 #define __atomic_store_n(ptr, val, memory_model)	*(ptr) = (val)
132 #endif
133 
134 #define packet_mmap_acquire(pkt) \
135 	(__atomic_load_n(&pkt->tp_status, __ATOMIC_ACQUIRE) != TP_STATUS_KERNEL)
136 #define packet_mmap_release(pkt) \
137 	(__atomic_store_n(&pkt->tp_status, TP_STATUS_KERNEL, __ATOMIC_RELEASE))
138 #define packet_mmap_v3_acquire(pkt) \
139 	(__atomic_load_n(&pkt->hdr.bh1.block_status, __ATOMIC_ACQUIRE) != TP_STATUS_KERNEL)
140 #define packet_mmap_v3_release(pkt) \
141 	(__atomic_store_n(&pkt->hdr.bh1.block_status, TP_STATUS_KERNEL, __ATOMIC_RELEASE))
142 
143 #include <linux/types.h>
144 #include <linux/filter.h>
145 
146 #ifdef HAVE_LINUX_NET_TSTAMP_H
147 #include <linux/net_tstamp.h>
148 #endif
149 
150 /*
151  * For checking whether a device is a bonding device.
152  */
153 #include <linux/if_bonding.h>
154 
155 /*
156  * Got libnl?
157  */
158 #ifdef HAVE_LIBNL
159 #include <linux/nl80211.h>
160 
161 #include <netlink/genl/genl.h>
162 #include <netlink/genl/family.h>
163 #include <netlink/genl/ctrl.h>
164 #include <netlink/msg.h>
165 #include <netlink/attr.h>
166 #endif /* HAVE_LIBNL */
167 
168 #ifndef HAVE_SOCKLEN_T
169 typedef int		socklen_t;
170 #endif
171 
172 #define MAX_LINKHEADER_SIZE	256
173 
174 /*
175  * When capturing on all interfaces we use this as the buffer size.
176  * Should be bigger then all MTUs that occur in real life.
177  * 64kB should be enough for now.
178  */
179 #define BIGGER_THAN_ALL_MTUS	(64*1024)
180 
181 /*
182  * Private data for capturing on Linux PF_PACKET sockets.
183  */
184 struct pcap_linux {
185 	long long sysfs_dropped; /* packets reported dropped by /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors */
186 	struct pcap_stat stat;
187 
188 	char	*device;	/* device name */
189 	int	filter_in_userland; /* must filter in userland */
190 	int	blocks_to_filter_in_userland;
191 	int	must_do_on_close; /* stuff we must do when we close */
192 	int	timeout;	/* timeout for buffering */
193 	int	cooked;		/* using SOCK_DGRAM rather than SOCK_RAW */
194 	int	ifindex;	/* interface index of device we're bound to */
195 	int	lo_ifindex;	/* interface index of the loopback device */
196 	int	netdown;	/* we got an ENETDOWN and haven't resolved it */
197 	bpf_u_int32 oldmode;	/* mode to restore when turning monitor mode off */
198 	char	*mondevice;	/* mac80211 monitor device we created */
199 	u_char	*mmapbuf;	/* memory-mapped region pointer */
200 	size_t	mmapbuflen;	/* size of region */
201 	int	vlan_offset;	/* offset at which to insert vlan tags; if -1, don't insert */
202 	u_int	tp_version;	/* version of tpacket_hdr for mmaped ring */
203 	u_int	tp_hdrlen;	/* hdrlen of tpacket_hdr for mmaped ring */
204 	u_char	*oneshot_buffer; /* buffer for copy of packet */
205 	int	poll_timeout;	/* timeout to use in poll() */
206 #ifdef HAVE_TPACKET3
207 	unsigned char *current_packet; /* Current packet within the TPACKET_V3 block. Move to next block if NULL. */
208 	int packets_left; /* Unhandled packets left within the block from previous call to pcap_read_linux_mmap_v3 in case of TPACKET_V3. */
209 #endif
210 	int poll_breakloop_fd; /* fd to an eventfd to break from blocking operations */
211 };
212 
213 /*
214  * Stuff to do when we close.
215  */
216 #define MUST_CLEAR_RFMON	0x00000001	/* clear rfmon (monitor) mode */
217 #define MUST_DELETE_MONIF	0x00000002	/* delete monitor-mode interface */
218 
219 /*
220  * Prototypes for internal functions and methods.
221  */
222 static int get_if_flags(const char *, bpf_u_int32 *, char *);
223 static int is_wifi(const char *);
224 static void map_arphrd_to_dlt(pcap_t *, int, const char *, int);
225 static int pcap_activate_linux(pcap_t *);
226 static int setup_socket(pcap_t *, int);
227 static int setup_mmapped(pcap_t *, int *);
228 static int pcap_can_set_rfmon_linux(pcap_t *);
229 static int pcap_inject_linux(pcap_t *, const void *, int);
230 static int pcap_stats_linux(pcap_t *, struct pcap_stat *);
231 static int pcap_setfilter_linux(pcap_t *, struct bpf_program *);
232 static int pcap_setdirection_linux(pcap_t *, pcap_direction_t);
233 static int pcap_set_datalink_linux(pcap_t *, int);
234 static void pcap_cleanup_linux(pcap_t *);
235 
236 union thdr {
237 	struct tpacket2_hdr		*h2;
238 #ifdef HAVE_TPACKET3
239 	struct tpacket_block_desc	*h3;
240 #endif
241 	u_char				*raw;
242 };
243 
244 #define RING_GET_FRAME_AT(h, offset) (((u_char **)h->buffer)[(offset)])
245 #define RING_GET_CURRENT_FRAME(h) RING_GET_FRAME_AT(h, h->offset)
246 
247 static void destroy_ring(pcap_t *handle);
248 static int create_ring(pcap_t *handle, int *status);
249 static int prepare_tpacket_socket(pcap_t *handle);
250 static int pcap_read_linux_mmap_v2(pcap_t *, int, pcap_handler , u_char *);
251 #ifdef HAVE_TPACKET3
252 static int pcap_read_linux_mmap_v3(pcap_t *, int, pcap_handler , u_char *);
253 #endif
254 static int pcap_setnonblock_linux(pcap_t *p, int nonblock);
255 static int pcap_getnonblock_linux(pcap_t *p);
256 static void pcap_oneshot_linux(u_char *user, const struct pcap_pkthdr *h,
257     const u_char *bytes);
258 
259 /*
260  * In pre-3.0 kernels, the tp_vlan_tci field is set to whatever the
261  * vlan_tci field in the skbuff is.  0 can either mean "not on a VLAN"
262  * or "on VLAN 0".  There is no flag set in the tp_status field to
263  * distinguish between them.
264  *
265  * In 3.0 and later kernels, if there's a VLAN tag present, the tp_vlan_tci
266  * field is set to the VLAN tag, and the TP_STATUS_VLAN_VALID flag is set
267  * in the tp_status field, otherwise the tp_vlan_tci field is set to 0 and
268  * the TP_STATUS_VLAN_VALID flag isn't set in the tp_status field.
269  *
270  * With a pre-3.0 kernel, we cannot distinguish between packets with no
271  * VLAN tag and packets on VLAN 0, so we will mishandle some packets, and
272  * there's nothing we can do about that.
273  *
274  * So, on those systems, which never set the TP_STATUS_VLAN_VALID flag, we
275  * continue the behavior of earlier libpcaps, wherein we treated packets
276  * with a VLAN tag of 0 as being packets without a VLAN tag rather than packets
277  * on VLAN 0.  We do this by treating packets with a tp_vlan_tci of 0 and
278  * with the TP_STATUS_VLAN_VALID flag not set in tp_status as not having
279  * VLAN tags.  This does the right thing on 3.0 and later kernels, and
280  * continues the old unfixably-imperfect behavior on pre-3.0 kernels.
281  *
282  * If TP_STATUS_VLAN_VALID isn't defined, we test it as the 0x10 bit; it
283  * has that value in 3.0 and later kernels.
284  */
285 #ifdef TP_STATUS_VLAN_VALID
286   #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & TP_STATUS_VLAN_VALID))
287 #else
288   /*
289    * This is being compiled on a system that lacks TP_STATUS_VLAN_VALID,
290    * so we testwith the value it has in the 3.0 and later kernels, so
291    * we can test it if we're running on a system that has it.  (If we're
292    * running on a system that doesn't have it, it won't be set in the
293    * tp_status field, so the tests of it will always fail; that means
294    * we behave the way we did before we introduced this macro.)
295    */
296   #define VLAN_VALID(hdr, hv)	((hv)->tp_vlan_tci != 0 || ((hdr)->tp_status & 0x10))
297 #endif
298 
299 #ifdef TP_STATUS_VLAN_TPID_VALID
300 # define VLAN_TPID(hdr, hv)	(((hv)->tp_vlan_tpid || ((hdr)->tp_status & TP_STATUS_VLAN_TPID_VALID)) ? (hv)->tp_vlan_tpid : ETH_P_8021Q)
301 #else
302 # define VLAN_TPID(hdr, hv)	ETH_P_8021Q
303 #endif
304 
305 /*
306  * Required select timeout if we're polling for an "interface disappeared"
307  * indication - 1 millisecond.
308  */
309 static const struct timeval netdown_timeout = {
310 	0, 1000		/* 1000 microseconds = 1 millisecond */
311 };
312 
313 /*
314  * Wrap some ioctl calls
315  */
316 static int	iface_get_id(int fd, const char *device, char *ebuf);
317 static int	iface_get_mtu(int fd, const char *device, char *ebuf);
318 static int 	iface_get_arptype(int fd, const char *device, char *ebuf);
319 static int 	iface_bind(int fd, int ifindex, char *ebuf, int protocol);
320 static int	enter_rfmon_mode(pcap_t *handle, int sock_fd,
321     const char *device);
322 static int	iface_get_ts_types(const char *device, pcap_t *handle,
323     char *ebuf);
324 static int	iface_get_offload(pcap_t *handle);
325 
326 static int	fix_program(pcap_t *handle, struct sock_fprog *fcode);
327 static int	fix_offset(pcap_t *handle, struct bpf_insn *p);
328 static int	set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode);
329 static int	reset_kernel_filter(pcap_t *handle);
330 
331 static struct sock_filter	total_insn
332 	= BPF_STMT(BPF_RET | BPF_K, 0);
333 static struct sock_fprog	total_fcode
334 	= { 1, &total_insn };
335 
336 static int	iface_dsa_get_proto_info(const char *device, pcap_t *handle);
337 
338 pcap_t *
339 pcap_create_interface(const char *device, char *ebuf)
340 {
341 	pcap_t *handle;
342 
343 	handle = PCAP_CREATE_COMMON(ebuf, struct pcap_linux);
344 	if (handle == NULL)
345 		return NULL;
346 
347 	handle->activate_op = pcap_activate_linux;
348 	handle->can_set_rfmon_op = pcap_can_set_rfmon_linux;
349 
350 	/*
351 	 * See what time stamp types we support.
352 	 */
353 	if (iface_get_ts_types(device, handle, ebuf) == -1) {
354 		pcap_close(handle);
355 		return NULL;
356 	}
357 
358 	/*
359 	 * We claim that we support microsecond and nanosecond time
360 	 * stamps.
361 	 *
362 	 * XXX - with adapter-supplied time stamps, can we choose
363 	 * microsecond or nanosecond time stamps on arbitrary
364 	 * adapters?
365 	 */
366 	handle->tstamp_precision_list = malloc(2 * sizeof(u_int));
367 	if (handle->tstamp_precision_list == NULL) {
368 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
369 		    errno, "malloc");
370 		pcap_close(handle);
371 		return NULL;
372 	}
373 	handle->tstamp_precision_list[0] = PCAP_TSTAMP_PRECISION_MICRO;
374 	handle->tstamp_precision_list[1] = PCAP_TSTAMP_PRECISION_NANO;
375 	handle->tstamp_precision_count = 2;
376 
377 	struct pcap_linux *handlep = handle->priv;
378 	handlep->poll_breakloop_fd = eventfd(0, EFD_NONBLOCK);
379 
380 	return handle;
381 }
382 
383 #ifdef HAVE_LIBNL
384 /*
385  * If interface {if_name} is a mac80211 driver, the file
386  * /sys/class/net/{if_name}/phy80211 is a symlink to
387  * /sys/class/ieee80211/{phydev_name}, for some {phydev_name}.
388  *
389  * On Fedora 9, with a 2.6.26.3-29 kernel, my Zydas stick, at
390  * least, has a "wmaster0" device and a "wlan0" device; the
391  * latter is the one with the IP address.  Both show up in
392  * "tcpdump -D" output.  Capturing on the wmaster0 device
393  * captures with 802.11 headers.
394  *
395  * airmon-ng searches through /sys/class/net for devices named
396  * monN, starting with mon0; as soon as one *doesn't* exist,
397  * it chooses that as the monitor device name.  If the "iw"
398  * command exists, it does
399  *
400  *    iw dev {if_name} interface add {monif_name} type monitor
401  *
402  * where {monif_name} is the monitor device.  It then (sigh) sleeps
403  * .1 second, and then configures the device up.  Otherwise, if
404  * /sys/class/ieee80211/{phydev_name}/add_iface is a file, it writes
405  * {mondev_name}, without a newline, to that file, and again (sigh)
406  * sleeps .1 second, and then iwconfig's that device into monitor
407  * mode and configures it up.  Otherwise, you can't do monitor mode.
408  *
409  * All these devices are "glued" together by having the
410  * /sys/class/net/{if_name}/phy80211 links pointing to the same
411  * place, so, given a wmaster, wlan, or mon device, you can
412  * find the other devices by looking for devices with
413  * the same phy80211 link.
414  *
415  * To turn monitor mode off, delete the monitor interface,
416  * either with
417  *
418  *    iw dev {monif_name} interface del
419  *
420  * or by sending {monif_name}, with no NL, down
421  * /sys/class/ieee80211/{phydev_name}/remove_iface
422  *
423  * Note: if you try to create a monitor device named "monN", and
424  * there's already a "monN" device, it fails, as least with
425  * the netlink interface (which is what iw uses), with a return
426  * value of -ENFILE.  (Return values are negative errnos.)  We
427  * could probably use that to find an unused device.
428  *
429  * Yes, you can have multiple monitor devices for a given
430  * physical device.
431  */
432 
433 /*
434  * Is this a mac80211 device?  If so, fill in the physical device path and
435  * return 1; if not, return 0.  On an error, fill in handle->errbuf and
436  * return PCAP_ERROR.
437  */
438 static int
439 get_mac80211_phydev(pcap_t *handle, const char *device, char *phydev_path,
440     size_t phydev_max_pathlen)
441 {
442 	char *pathstr;
443 	ssize_t bytes_read;
444 
445 	/*
446 	 * Generate the path string for the symlink to the physical device.
447 	 */
448 	if (asprintf(&pathstr, "/sys/class/net/%s/phy80211", device) == -1) {
449 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
450 		    "%s: Can't generate path name string for /sys/class/net device",
451 		    device);
452 		return PCAP_ERROR;
453 	}
454 	bytes_read = readlink(pathstr, phydev_path, phydev_max_pathlen);
455 	if (bytes_read == -1) {
456 		if (errno == ENOENT || errno == EINVAL) {
457 			/*
458 			 * Doesn't exist, or not a symlink; assume that
459 			 * means it's not a mac80211 device.
460 			 */
461 			free(pathstr);
462 			return 0;
463 		}
464 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
465 		    errno, "%s: Can't readlink %s", device, pathstr);
466 		free(pathstr);
467 		return PCAP_ERROR;
468 	}
469 	free(pathstr);
470 	phydev_path[bytes_read] = '\0';
471 	return 1;
472 }
473 
474 struct nl80211_state {
475 	struct nl_sock *nl_sock;
476 	struct nl_cache *nl_cache;
477 	struct genl_family *nl80211;
478 };
479 
480 static int
481 nl80211_init(pcap_t *handle, struct nl80211_state *state, const char *device)
482 {
483 	int err;
484 
485 	state->nl_sock = nl_socket_alloc();
486 	if (!state->nl_sock) {
487 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
488 		    "%s: failed to allocate netlink handle", device);
489 		return PCAP_ERROR;
490 	}
491 
492 	if (genl_connect(state->nl_sock)) {
493 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
494 		    "%s: failed to connect to generic netlink", device);
495 		goto out_handle_destroy;
496 	}
497 
498 	err = genl_ctrl_alloc_cache(state->nl_sock, &state->nl_cache);
499 	if (err < 0) {
500 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
501 		    "%s: failed to allocate generic netlink cache: %s",
502 		    device, nl_geterror(-err));
503 		goto out_handle_destroy;
504 	}
505 
506 	state->nl80211 = genl_ctrl_search_by_name(state->nl_cache, "nl80211");
507 	if (!state->nl80211) {
508 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
509 		    "%s: nl80211 not found", device);
510 		goto out_cache_free;
511 	}
512 
513 	return 0;
514 
515 out_cache_free:
516 	nl_cache_free(state->nl_cache);
517 out_handle_destroy:
518 	nl_socket_free(state->nl_sock);
519 	return PCAP_ERROR;
520 }
521 
522 static void
523 nl80211_cleanup(struct nl80211_state *state)
524 {
525 	genl_family_put(state->nl80211);
526 	nl_cache_free(state->nl_cache);
527 	nl_socket_free(state->nl_sock);
528 }
529 
530 static int
531 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
532     const char *device, const char *mondevice);
533 
534 static int
535 add_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
536     const char *device, const char *mondevice)
537 {
538 	struct pcap_linux *handlep = handle->priv;
539 	int ifindex;
540 	struct nl_msg *msg;
541 	int err;
542 
543 	ifindex = iface_get_id(sock_fd, device, handle->errbuf);
544 	if (ifindex == -1)
545 		return PCAP_ERROR;
546 
547 	msg = nlmsg_alloc();
548 	if (!msg) {
549 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
550 		    "%s: failed to allocate netlink msg", device);
551 		return PCAP_ERROR;
552 	}
553 
554 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
555 		    0, NL80211_CMD_NEW_INTERFACE, 0);
556 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
557 DIAG_OFF_NARROWING
558 	NLA_PUT_STRING(msg, NL80211_ATTR_IFNAME, mondevice);
559 DIAG_ON_NARROWING
560 	NLA_PUT_U32(msg, NL80211_ATTR_IFTYPE, NL80211_IFTYPE_MONITOR);
561 
562 	err = nl_send_auto_complete(state->nl_sock, msg);
563 	if (err < 0) {
564 		if (err == -NLE_FAILURE) {
565 			/*
566 			 * Device not available; our caller should just
567 			 * keep trying.  (libnl 2.x maps ENFILE to
568 			 * NLE_FAILURE; it can also map other errors
569 			 * to that, but there's not much we can do
570 			 * about that.)
571 			 */
572 			nlmsg_free(msg);
573 			return 0;
574 		} else {
575 			/*
576 			 * Real failure, not just "that device is not
577 			 * available.
578 			 */
579 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
580 			    "%s: nl_send_auto_complete failed adding %s interface: %s",
581 			    device, mondevice, nl_geterror(-err));
582 			nlmsg_free(msg);
583 			return PCAP_ERROR;
584 		}
585 	}
586 	err = nl_wait_for_ack(state->nl_sock);
587 	if (err < 0) {
588 		if (err == -NLE_FAILURE) {
589 			/*
590 			 * Device not available; our caller should just
591 			 * keep trying.  (libnl 2.x maps ENFILE to
592 			 * NLE_FAILURE; it can also map other errors
593 			 * to that, but there's not much we can do
594 			 * about that.)
595 			 */
596 			nlmsg_free(msg);
597 			return 0;
598 		} else {
599 			/*
600 			 * Real failure, not just "that device is not
601 			 * available.
602 			 */
603 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
604 			    "%s: nl_wait_for_ack failed adding %s interface: %s",
605 			    device, mondevice, nl_geterror(-err));
606 			nlmsg_free(msg);
607 			return PCAP_ERROR;
608 		}
609 	}
610 
611 	/*
612 	 * Success.
613 	 */
614 	nlmsg_free(msg);
615 
616 	/*
617 	 * Try to remember the monitor device.
618 	 */
619 	handlep->mondevice = strdup(mondevice);
620 	if (handlep->mondevice == NULL) {
621 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
622 		    errno, "strdup");
623 		/*
624 		 * Get rid of the monitor device.
625 		 */
626 		del_mon_if(handle, sock_fd, state, device, mondevice);
627 		return PCAP_ERROR;
628 	}
629 	return 1;
630 
631 nla_put_failure:
632 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
633 	    "%s: nl_put failed adding %s interface",
634 	    device, mondevice);
635 	nlmsg_free(msg);
636 	return PCAP_ERROR;
637 }
638 
639 static int
640 del_mon_if(pcap_t *handle, int sock_fd, struct nl80211_state *state,
641     const char *device, const char *mondevice)
642 {
643 	int ifindex;
644 	struct nl_msg *msg;
645 	int err;
646 
647 	ifindex = iface_get_id(sock_fd, mondevice, handle->errbuf);
648 	if (ifindex == -1)
649 		return PCAP_ERROR;
650 
651 	msg = nlmsg_alloc();
652 	if (!msg) {
653 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
654 		    "%s: failed to allocate netlink msg", device);
655 		return PCAP_ERROR;
656 	}
657 
658 	genlmsg_put(msg, 0, 0, genl_family_get_id(state->nl80211), 0,
659 		    0, NL80211_CMD_DEL_INTERFACE, 0);
660 	NLA_PUT_U32(msg, NL80211_ATTR_IFINDEX, ifindex);
661 
662 	err = nl_send_auto_complete(state->nl_sock, msg);
663 	if (err < 0) {
664 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
665 		    "%s: nl_send_auto_complete failed deleting %s interface: %s",
666 		    device, mondevice, nl_geterror(-err));
667 		nlmsg_free(msg);
668 		return PCAP_ERROR;
669 	}
670 	err = nl_wait_for_ack(state->nl_sock);
671 	if (err < 0) {
672 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
673 		    "%s: nl_wait_for_ack failed adding %s interface: %s",
674 		    device, mondevice, nl_geterror(-err));
675 		nlmsg_free(msg);
676 		return PCAP_ERROR;
677 	}
678 
679 	/*
680 	 * Success.
681 	 */
682 	nlmsg_free(msg);
683 	return 1;
684 
685 nla_put_failure:
686 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
687 	    "%s: nl_put failed deleting %s interface",
688 	    device, mondevice);
689 	nlmsg_free(msg);
690 	return PCAP_ERROR;
691 }
692 #endif /* HAVE_LIBNL */
693 
694 static int pcap_protocol(pcap_t *handle)
695 {
696 	int protocol;
697 
698 	protocol = handle->opt.protocol;
699 	if (protocol == 0)
700 		protocol = ETH_P_ALL;
701 
702 	return htons(protocol);
703 }
704 
705 static int
706 pcap_can_set_rfmon_linux(pcap_t *handle)
707 {
708 #ifdef HAVE_LIBNL
709 	char phydev_path[PATH_MAX+1];
710 	int ret;
711 #endif
712 
713 	if (strcmp(handle->opt.device, "any") == 0) {
714 		/*
715 		 * Monitor mode makes no sense on the "any" device.
716 		 */
717 		return 0;
718 	}
719 
720 #ifdef HAVE_LIBNL
721 	/*
722 	 * Bleah.  There doesn't seem to be a way to ask a mac80211
723 	 * device, through libnl, whether it supports monitor mode;
724 	 * we'll just check whether the device appears to be a
725 	 * mac80211 device and, if so, assume the device supports
726 	 * monitor mode.
727 	 */
728 	ret = get_mac80211_phydev(handle, handle->opt.device, phydev_path,
729 	    PATH_MAX);
730 	if (ret < 0)
731 		return ret;	/* error */
732 	if (ret == 1)
733 		return 1;	/* mac80211 device */
734 #endif
735 
736 	return 0;
737 }
738 
739 /*
740  * Grabs the number of missed packets by the interface from
741  * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors.
742  *
743  * Compared to /proc/net/dev this avoids counting software drops,
744  * but may be unimplemented and just return 0.
745  * The author has found no straigthforward way to check for support.
746  */
747 static long long int
748 linux_get_stat(const char * if_name, const char * stat) {
749 	ssize_t bytes_read;
750 	int fd;
751 	char buffer[PATH_MAX];
752 
753 	snprintf(buffer, sizeof(buffer), "/sys/class/net/%s/statistics/%s", if_name, stat);
754 	fd = open(buffer, O_RDONLY);
755 	if (fd == -1)
756 		return 0;
757 
758 	bytes_read = read(fd, buffer, sizeof(buffer) - 1);
759 	close(fd);
760 	if (bytes_read == -1)
761 		return 0;
762 	buffer[bytes_read] = '\0';
763 
764 	return strtoll(buffer, NULL, 10);
765 }
766 
767 static long long int
768 linux_if_drops(const char * if_name)
769 {
770 	long long int missed = linux_get_stat(if_name, "rx_missed_errors");
771 	long long int fifo = linux_get_stat(if_name, "rx_fifo_errors");
772 	return missed + fifo;
773 }
774 
775 
776 /*
777  * Monitor mode is kind of interesting because we have to reset the
778  * interface before exiting. The problem can't really be solved without
779  * some daemon taking care of managing usage counts.  If we put the
780  * interface into monitor mode, we set a flag indicating that we must
781  * take it out of that mode when the interface is closed, and, when
782  * closing the interface, if that flag is set we take it out of monitor
783  * mode.
784  */
785 
786 static void	pcap_cleanup_linux( pcap_t *handle )
787 {
788 	struct pcap_linux *handlep = handle->priv;
789 #ifdef HAVE_LIBNL
790 	struct nl80211_state nlstate;
791 	int ret;
792 #endif /* HAVE_LIBNL */
793 
794 	if (handlep->must_do_on_close != 0) {
795 		/*
796 		 * There's something we have to do when closing this
797 		 * pcap_t.
798 		 */
799 #ifdef HAVE_LIBNL
800 		if (handlep->must_do_on_close & MUST_DELETE_MONIF) {
801 			ret = nl80211_init(handle, &nlstate, handlep->device);
802 			if (ret >= 0) {
803 				ret = del_mon_if(handle, handle->fd, &nlstate,
804 				    handlep->device, handlep->mondevice);
805 				nl80211_cleanup(&nlstate);
806 			}
807 			if (ret < 0) {
808 				fprintf(stderr,
809 				    "Can't delete monitor interface %s (%s).\n"
810 				    "Please delete manually.\n",
811 				    handlep->mondevice, handle->errbuf);
812 			}
813 		}
814 #endif /* HAVE_LIBNL */
815 
816 		/*
817 		 * Take this pcap out of the list of pcaps for which we
818 		 * have to take the interface out of some mode.
819 		 */
820 		pcap_remove_from_pcaps_to_close(handle);
821 	}
822 
823 	if (handle->fd != -1) {
824 		/*
825 		 * Destroy the ring buffer (assuming we've set it up),
826 		 * and unmap it if it's mapped.
827 		 */
828 		destroy_ring(handle);
829 	}
830 
831 	if (handlep->oneshot_buffer != NULL) {
832 		free(handlep->oneshot_buffer);
833 		handlep->oneshot_buffer = NULL;
834 	}
835 
836 	if (handlep->mondevice != NULL) {
837 		free(handlep->mondevice);
838 		handlep->mondevice = NULL;
839 	}
840 	if (handlep->device != NULL) {
841 		free(handlep->device);
842 		handlep->device = NULL;
843 	}
844 
845 	if (handlep->poll_breakloop_fd != -1) {
846 		close(handlep->poll_breakloop_fd);
847 		handlep->poll_breakloop_fd = -1;
848 	}
849 	pcap_cleanup_live_common(handle);
850 }
851 
852 #ifdef HAVE_TPACKET3
853 /*
854  * Some versions of TPACKET_V3 have annoying bugs/misfeatures
855  * around which we have to work.  Determine if we have those
856  * problems or not.
857  * 3.19 is the first release with a fixed version of
858  * TPACKET_V3.  We treat anything before that as
859  * not having a fixed version; that may really mean
860  * it has *no* version.
861  */
862 static int has_broken_tpacket_v3(void)
863 {
864 	struct utsname utsname;
865 	const char *release;
866 	long major, minor;
867 	int matches, verlen;
868 
869 	/* No version information, assume broken. */
870 	if (uname(&utsname) == -1)
871 		return 1;
872 	release = utsname.release;
873 
874 	/* A malformed version, ditto. */
875 	matches = sscanf(release, "%ld.%ld%n", &major, &minor, &verlen);
876 	if (matches != 2)
877 		return 1;
878 	if (release[verlen] != '.' && release[verlen] != '\0')
879 		return 1;
880 
881 	/* OK, a fixed version. */
882 	if (major > 3 || (major == 3 && minor >= 19))
883 		return 0;
884 
885 	/* Too old :( */
886 	return 1;
887 }
888 #endif
889 
890 /*
891  * Set the timeout to be used in poll() with memory-mapped packet capture.
892  */
893 static void
894 set_poll_timeout(struct pcap_linux *handlep)
895 {
896 #ifdef HAVE_TPACKET3
897 	int broken_tpacket_v3 = has_broken_tpacket_v3();
898 #endif
899 	if (handlep->timeout == 0) {
900 #ifdef HAVE_TPACKET3
901 		/*
902 		 * XXX - due to a set of (mis)features in the TPACKET_V3
903 		 * kernel code prior to the 3.19 kernel, blocking forever
904 		 * with a TPACKET_V3 socket can, if few packets are
905 		 * arriving and passing the socket filter, cause most
906 		 * packets to be dropped.  See libpcap issue #335 for the
907 		 * full painful story.
908 		 *
909 		 * The workaround is to have poll() time out very quickly,
910 		 * so we grab the frames handed to us, and return them to
911 		 * the kernel, ASAP.
912 		 */
913 		if (handlep->tp_version == TPACKET_V3 && broken_tpacket_v3)
914 			handlep->poll_timeout = 1;	/* don't block for very long */
915 		else
916 #endif
917 			handlep->poll_timeout = -1;	/* block forever */
918 	} else if (handlep->timeout > 0) {
919 #ifdef HAVE_TPACKET3
920 		/*
921 		 * For TPACKET_V3, the timeout is handled by the kernel,
922 		 * so block forever; that way, we don't get extra timeouts.
923 		 * Don't do that if we have a broken TPACKET_V3, though.
924 		 */
925 		if (handlep->tp_version == TPACKET_V3 && !broken_tpacket_v3)
926 			handlep->poll_timeout = -1;	/* block forever, let TPACKET_V3 wake us up */
927 		else
928 #endif
929 			handlep->poll_timeout = handlep->timeout;	/* block for that amount of time */
930 	} else {
931 		/*
932 		 * Non-blocking mode; we call poll() to pick up error
933 		 * indications, but we don't want it to wait for
934 		 * anything.
935 		 */
936 		handlep->poll_timeout = 0;
937 	}
938 }
939 
940 static void pcap_breakloop_linux(pcap_t *handle)
941 {
942 	pcap_breakloop_common(handle);
943 	struct pcap_linux *handlep = handle->priv;
944 
945 	uint64_t value = 1;
946 	/* XXX - what if this fails? */
947 	if (handlep->poll_breakloop_fd != -1)
948 		(void)write(handlep->poll_breakloop_fd, &value, sizeof(value));
949 }
950 
951 /*
952  * Set the offset at which to insert VLAN tags.
953  * That should be the offset of the type field.
954  */
955 static void
956 set_vlan_offset(pcap_t *handle)
957 {
958 	struct pcap_linux *handlep = handle->priv;
959 
960 	switch (handle->linktype) {
961 
962 	case DLT_EN10MB:
963 		/*
964 		 * The type field is after the destination and source
965 		 * MAC address.
966 		 */
967 		handlep->vlan_offset = 2 * ETH_ALEN;
968 		break;
969 
970 	case DLT_LINUX_SLL:
971 		/*
972 		 * The type field is in the last 2 bytes of the
973 		 * DLT_LINUX_SLL header.
974 		 */
975 		handlep->vlan_offset = SLL_HDR_LEN - 2;
976 		break;
977 
978 	default:
979 		handlep->vlan_offset = -1; /* unknown */
980 		break;
981 	}
982 }
983 
984 /*
985  *  Get a handle for a live capture from the given device. You can
986  *  pass NULL as device to get all packages (without link level
987  *  information of course). If you pass 1 as promisc the interface
988  *  will be set to promiscuous mode (XXX: I think this usage should
989  *  be deprecated and functions be added to select that later allow
990  *  modification of that values -- Torsten).
991  */
992 static int
993 pcap_activate_linux(pcap_t *handle)
994 {
995 	struct pcap_linux *handlep = handle->priv;
996 	const char	*device;
997 	int		is_any_device;
998 	struct ifreq	ifr;
999 	int		status = 0;
1000 	int		status2 = 0;
1001 	int		ret;
1002 
1003 	device = handle->opt.device;
1004 
1005 	/*
1006 	 * Make sure the name we were handed will fit into the ioctls we
1007 	 * might perform on the device; if not, return a "No such device"
1008 	 * indication, as the Linux kernel shouldn't support creating
1009 	 * a device whose name won't fit into those ioctls.
1010 	 *
1011 	 * "Will fit" means "will fit, complete with a null terminator",
1012 	 * so if the length, which does *not* include the null terminator,
1013 	 * is greater than *or equal to* the size of the field into which
1014 	 * we'll be copying it, that won't fit.
1015 	 */
1016 	if (strlen(device) >= sizeof(ifr.ifr_name)) {
1017 		/*
1018 		 * There's nothing more to say, so clear the error
1019 		 * message.
1020 		 */
1021 		handle->errbuf[0] = '\0';
1022 		status = PCAP_ERROR_NO_SUCH_DEVICE;
1023 		goto fail;
1024 	}
1025 
1026 	/*
1027 	 * Turn a negative snapshot value (invalid), a snapshot value of
1028 	 * 0 (unspecified), or a value bigger than the normal maximum
1029 	 * value, into the maximum allowed value.
1030 	 *
1031 	 * If some application really *needs* a bigger snapshot
1032 	 * length, we should just increase MAXIMUM_SNAPLEN.
1033 	 */
1034 	if (handle->snapshot <= 0 || handle->snapshot > MAXIMUM_SNAPLEN)
1035 		handle->snapshot = MAXIMUM_SNAPLEN;
1036 
1037 	handlep->device	= strdup(device);
1038 	if (handlep->device == NULL) {
1039 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1040 		    errno, "strdup");
1041 		status = PCAP_ERROR;
1042 		goto fail;
1043 	}
1044 
1045 	/*
1046 	 * The "any" device is a special device which causes us not
1047 	 * to bind to a particular device and thus to look at all
1048 	 * devices.
1049 	 */
1050 	is_any_device = (strcmp(device, "any") == 0);
1051 	if (is_any_device) {
1052 		if (handle->opt.promisc) {
1053 			handle->opt.promisc = 0;
1054 			/* Just a warning. */
1055 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
1056 			    "Promiscuous mode not supported on the \"any\" device");
1057 			status = PCAP_WARNING_PROMISC_NOTSUP;
1058 		}
1059 	}
1060 
1061 	/* copy timeout value */
1062 	handlep->timeout = handle->opt.timeout;
1063 
1064 	/*
1065 	 * If we're in promiscuous mode, then we probably want
1066 	 * to see when the interface drops packets too, so get an
1067 	 * initial count from
1068 	 * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors
1069 	 */
1070 	if (handle->opt.promisc)
1071 		handlep->sysfs_dropped = linux_if_drops(handlep->device);
1072 
1073 	/*
1074 	 * If the "any" device is specified, try to open a SOCK_DGRAM.
1075 	 * Otherwise, open a SOCK_RAW.
1076 	 */
1077 	ret = setup_socket(handle, is_any_device);
1078 	if (ret < 0) {
1079 		/*
1080 		 * Fatal error; the return value is the error code,
1081 		 * and handle->errbuf has been set to an appropriate
1082 		 * error message.
1083 		 */
1084 		status = ret;
1085 		goto fail;
1086 	}
1087 	/*
1088 	 * Success.
1089 	 * Try to set up memory-mapped access.
1090 	 */
1091 	ret = setup_mmapped(handle, &status);
1092 	if (ret == -1) {
1093 		/*
1094 		 * We failed to set up to use it, or the
1095 		 * kernel supports it, but we failed to
1096 		 * enable it.  status has been set to the
1097 		 * error status to return and, if it's
1098 		 * PCAP_ERROR, handle->errbuf contains
1099 		 * the error message.
1100 		 */
1101 		goto fail;
1102 	}
1103 
1104 	/*
1105 	 * We succeeded.  status has been set to the status to return,
1106 	 * which might be 0, or might be a PCAP_WARNING_ value.
1107 	 */
1108 	/*
1109 	 * Now that we have activated the mmap ring, we can
1110 	 * set the correct protocol.
1111 	 */
1112 	if ((status2 = iface_bind(handle->fd, handlep->ifindex,
1113 	    handle->errbuf, pcap_protocol(handle))) != 0) {
1114 		status = status2;
1115 		goto fail;
1116 	}
1117 
1118 	handle->inject_op = pcap_inject_linux;
1119 	handle->setfilter_op = pcap_setfilter_linux;
1120 	handle->setdirection_op = pcap_setdirection_linux;
1121 	handle->set_datalink_op = pcap_set_datalink_linux;
1122 	handle->setnonblock_op = pcap_setnonblock_linux;
1123 	handle->getnonblock_op = pcap_getnonblock_linux;
1124 	handle->cleanup_op = pcap_cleanup_linux;
1125 	handle->stats_op = pcap_stats_linux;
1126 	handle->breakloop_op = pcap_breakloop_linux;
1127 
1128 	switch (handlep->tp_version) {
1129 
1130 	case TPACKET_V2:
1131 		handle->read_op = pcap_read_linux_mmap_v2;
1132 		break;
1133 #ifdef HAVE_TPACKET3
1134 	case TPACKET_V3:
1135 		handle->read_op = pcap_read_linux_mmap_v3;
1136 		break;
1137 #endif
1138 	}
1139 	handle->oneshot_callback = pcap_oneshot_linux;
1140 	handle->selectable_fd = handle->fd;
1141 
1142 	return status;
1143 
1144 fail:
1145 	pcap_cleanup_linux(handle);
1146 	return status;
1147 }
1148 
1149 static int
1150 pcap_set_datalink_linux(pcap_t *handle, int dlt)
1151 {
1152 	handle->linktype = dlt;
1153 
1154 	/*
1155 	 * Update the offset at which to insert VLAN tags for the
1156 	 * new link-layer type.
1157 	 */
1158 	set_vlan_offset(handle);
1159 
1160 	return 0;
1161 }
1162 
1163 /*
1164  * linux_check_direction()
1165  *
1166  * Do checks based on packet direction.
1167  */
1168 static inline int
1169 linux_check_direction(const pcap_t *handle, const struct sockaddr_ll *sll)
1170 {
1171 	struct pcap_linux	*handlep = handle->priv;
1172 
1173 	if (sll->sll_pkttype == PACKET_OUTGOING) {
1174 		/*
1175 		 * Outgoing packet.
1176 		 * If this is from the loopback device, reject it;
1177 		 * we'll see the packet as an incoming packet as well,
1178 		 * and we don't want to see it twice.
1179 		 */
1180 		if (sll->sll_ifindex == handlep->lo_ifindex)
1181 			return 0;
1182 
1183 		/*
1184 		 * If this is an outgoing CAN or CAN FD frame, and
1185 		 * the user doesn't only want outgoing packets,
1186 		 * reject it; CAN devices and drivers, and the CAN
1187 		 * stack, always arrange to loop back transmitted
1188 		 * packets, so they also appear as incoming packets.
1189 		 * We don't want duplicate packets, and we can't
1190 		 * easily distinguish packets looped back by the CAN
1191 		 * layer than those received by the CAN layer, so we
1192 		 * eliminate this packet instead.
1193 		 *
1194 		 * We check whether this is a CAN or CAN FD frame
1195 		 * by checking whether the device's hardware type
1196 		 * is ARPHRD_CAN.
1197 		 */
1198 		if (sll->sll_hatype == ARPHRD_CAN &&
1199 		     handle->direction != PCAP_D_OUT)
1200 			return 0;
1201 
1202 		/*
1203 		 * If the user only wants incoming packets, reject it.
1204 		 */
1205 		if (handle->direction == PCAP_D_IN)
1206 			return 0;
1207 	} else {
1208 		/*
1209 		 * Incoming packet.
1210 		 * If the user only wants outgoing packets, reject it.
1211 		 */
1212 		if (handle->direction == PCAP_D_OUT)
1213 			return 0;
1214 	}
1215 	return 1;
1216 }
1217 
1218 /*
1219  * Check whether the device to which the pcap_t is bound still exists.
1220  * We do so by asking what address the socket is bound to, and checking
1221  * whether the ifindex in the address is -1, meaning "that device is gone",
1222  * or some other value, meaning "that device still exists".
1223  */
1224 static int
1225 device_still_exists(pcap_t *handle)
1226 {
1227 	struct pcap_linux *handlep = handle->priv;
1228 	struct sockaddr_ll addr;
1229 	socklen_t addr_len;
1230 
1231 	/*
1232 	 * If handlep->ifindex is -1, the socket isn't bound, meaning
1233 	 * we're capturing on the "any" device; that device never
1234 	 * disappears.  (It should also never be configured down, so
1235 	 * we shouldn't even get here, but let's make sure.)
1236 	 */
1237 	if (handlep->ifindex == -1)
1238 		return (1);	/* it's still here */
1239 
1240 	/*
1241 	 * OK, now try to get the address for the socket.
1242 	 */
1243 	addr_len = sizeof (addr);
1244 	if (getsockname(handle->fd, (struct sockaddr *) &addr, &addr_len) == -1) {
1245 		/*
1246 		 * Error - report an error and return -1.
1247 		 */
1248 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1249 		    errno, "getsockname failed");
1250 		return (-1);
1251 	}
1252 	if (addr.sll_ifindex == -1) {
1253 		/*
1254 		 * This means the device went away.
1255 		 */
1256 		return (0);
1257 	}
1258 
1259 	/*
1260 	 * The device presumably just went down.
1261 	 */
1262 	return (1);
1263 }
1264 
1265 static int
1266 pcap_inject_linux(pcap_t *handle, const void *buf, int size)
1267 {
1268 	struct pcap_linux *handlep = handle->priv;
1269 	int ret;
1270 
1271 	if (handlep->ifindex == -1) {
1272 		/*
1273 		 * We don't support sending on the "any" device.
1274 		 */
1275 		pcap_strlcpy(handle->errbuf,
1276 		    "Sending packets isn't supported on the \"any\" device",
1277 		    PCAP_ERRBUF_SIZE);
1278 		return (-1);
1279 	}
1280 
1281 	if (handlep->cooked) {
1282 		/*
1283 		 * We don't support sending on cooked-mode sockets.
1284 		 *
1285 		 * XXX - how do you send on a bound cooked-mode
1286 		 * socket?
1287 		 * Is a "sendto()" required there?
1288 		 */
1289 		pcap_strlcpy(handle->errbuf,
1290 		    "Sending packets isn't supported in cooked mode",
1291 		    PCAP_ERRBUF_SIZE);
1292 		return (-1);
1293 	}
1294 
1295 	ret = (int)send(handle->fd, buf, size, 0);
1296 	if (ret == -1) {
1297 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
1298 		    errno, "send");
1299 		return (-1);
1300 	}
1301 	return (ret);
1302 }
1303 
1304 /*
1305  *  Get the statistics for the given packet capture handle.
1306  */
1307 static int
1308 pcap_stats_linux(pcap_t *handle, struct pcap_stat *stats)
1309 {
1310 	struct pcap_linux *handlep = handle->priv;
1311 #ifdef HAVE_TPACKET3
1312 	/*
1313 	 * For sockets using TPACKET_V2, the extra stuff at the end
1314 	 * of a struct tpacket_stats_v3 will not be filled in, and
1315 	 * we don't look at it so this is OK even for those sockets.
1316 	 * In addition, the PF_PACKET socket code in the kernel only
1317 	 * uses the length parameter to compute how much data to
1318 	 * copy out and to indicate how much data was copied out, so
1319 	 * it's OK to base it on the size of a struct tpacket_stats.
1320 	 *
1321 	 * XXX - it's probably OK, in fact, to just use a
1322 	 * struct tpacket_stats for V3 sockets, as we don't
1323 	 * care about the tp_freeze_q_cnt stat.
1324 	 */
1325 	struct tpacket_stats_v3 kstats;
1326 #else /* HAVE_TPACKET3 */
1327 	struct tpacket_stats kstats;
1328 #endif /* HAVE_TPACKET3 */
1329 	socklen_t len = sizeof (struct tpacket_stats);
1330 
1331 	long long if_dropped = 0;
1332 
1333 	/*
1334 	 * To fill in ps_ifdrop, we parse
1335 	 * /sys/class/net/{if_name}/statistics/rx_{missed,fifo}_errors
1336 	 * for the numbers
1337 	 */
1338 	if (handle->opt.promisc)
1339 	{
1340 		/*
1341 		 * XXX - is there any reason to do this by remembering
1342 		 * the last counts value, subtracting it from the
1343 		 * current counts value, and adding that to stat.ps_ifdrop,
1344 		 * maintaining stat.ps_ifdrop as a count, rather than just
1345 		 * saving the *initial* counts value and setting
1346 		 * stat.ps_ifdrop to the difference between the current
1347 		 * value and the initial value?
1348 		 *
1349 		 * One reason might be to handle the count wrapping
1350 		 * around, on platforms where the count is 32 bits
1351 		 * and where you might get more than 2^32 dropped
1352 		 * packets; is there any other reason?
1353 		 *
1354 		 * (We maintain the count as a long long int so that,
1355 		 * if the kernel maintains the counts as 64-bit even
1356 		 * on 32-bit platforms, we can handle the real count.
1357 		 *
1358 		 * Unfortunately, we can't report 64-bit counts; we
1359 		 * need a better API for reporting statistics, such as
1360 		 * one that reports them in a style similar to the
1361 		 * pcapng Interface Statistics Block, so that 1) the
1362 		 * counts are 64-bit, 2) it's easier to add new statistics
1363 		 * without breaking the ABI, and 3) it's easier to
1364 		 * indicate to a caller that wants one particular
1365 		 * statistic that it's not available by just not supplying
1366 		 * it.)
1367 		 */
1368 		if_dropped = handlep->sysfs_dropped;
1369 		handlep->sysfs_dropped = linux_if_drops(handlep->device);
1370 		handlep->stat.ps_ifdrop += (u_int)(handlep->sysfs_dropped - if_dropped);
1371 	}
1372 
1373 	/*
1374 	 * Try to get the packet counts from the kernel.
1375 	 */
1376 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS,
1377 			&kstats, &len) > -1) {
1378 		/*
1379 		 * "ps_recv" counts only packets that *passed* the
1380 		 * filter, not packets that didn't pass the filter.
1381 		 * This includes packets later dropped because we
1382 		 * ran out of buffer space.
1383 		 *
1384 		 * "ps_drop" counts packets dropped because we ran
1385 		 * out of buffer space.  It doesn't count packets
1386 		 * dropped by the interface driver.  It counts only
1387 		 * packets that passed the filter.
1388 		 *
1389 		 * See above for ps_ifdrop.
1390 		 *
1391 		 * Both statistics include packets not yet read from
1392 		 * the kernel by libpcap, and thus not yet seen by
1393 		 * the application.
1394 		 *
1395 		 * In "linux/net/packet/af_packet.c", at least in 2.6.27
1396 		 * through 5.6 kernels, "tp_packets" is incremented for
1397 		 * every packet that passes the packet filter *and* is
1398 		 * successfully copied to the ring buffer; "tp_drops" is
1399 		 * incremented for every packet dropped because there's
1400 		 * not enough free space in the ring buffer.
1401 		 *
1402 		 * When the statistics are returned for a PACKET_STATISTICS
1403 		 * "getsockopt()" call, "tp_drops" is added to "tp_packets",
1404 		 * so that "tp_packets" counts all packets handed to
1405 		 * the PF_PACKET socket, including packets dropped because
1406 		 * there wasn't room on the socket buffer - but not
1407 		 * including packets that didn't pass the filter.
1408 		 *
1409 		 * In the BSD BPF, the count of received packets is
1410 		 * incremented for every packet handed to BPF, regardless
1411 		 * of whether it passed the filter.
1412 		 *
1413 		 * We can't make "pcap_stats()" work the same on both
1414 		 * platforms, but the best approximation is to return
1415 		 * "tp_packets" as the count of packets and "tp_drops"
1416 		 * as the count of drops.
1417 		 *
1418 		 * Keep a running total because each call to
1419 		 *    getsockopt(handle->fd, SOL_PACKET, PACKET_STATISTICS, ....
1420 		 * resets the counters to zero.
1421 		 */
1422 		handlep->stat.ps_recv += kstats.tp_packets;
1423 		handlep->stat.ps_drop += kstats.tp_drops;
1424 		*stats = handlep->stat;
1425 		return 0;
1426 	}
1427 
1428 	pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE, errno,
1429 	    "failed to get statistics from socket");
1430 	return -1;
1431 }
1432 
1433 /*
1434  * Description string for the "any" device.
1435  */
1436 static const char any_descr[] = "Pseudo-device that captures on all interfaces";
1437 
1438 /*
1439  * A PF_PACKET socket can be bound to any network interface.
1440  */
1441 static int
1442 can_be_bound(const char *name _U_)
1443 {
1444 	return (1);
1445 }
1446 
1447 /*
1448  * Get a socket to use with various interface ioctls.
1449  */
1450 static int
1451 get_if_ioctl_socket(void)
1452 {
1453 	int fd;
1454 
1455 	/*
1456 	 * This is a bit ugly.
1457 	 *
1458 	 * There isn't a socket type that's guaranteed to work.
1459 	 *
1460 	 * AF_NETLINK will work *if* you have Netlink configured into the
1461 	 * kernel (can it be configured out if you have any networking
1462 	 * support at all?) *and* if you're running a sufficiently recent
1463 	 * kernel, but not all the kernels we support are sufficiently
1464 	 * recent - that feature was introduced in Linux 4.6.
1465 	 *
1466 	 * AF_UNIX will work *if* you have UNIX-domain sockets configured
1467 	 * into the kernel and *if* you're not on a system that doesn't
1468 	 * allow them - some SELinux systems don't allow you create them.
1469 	 * Most systems probably have them configured in, but not all systems
1470 	 * have them configured in and allow them to be created.
1471 	 *
1472 	 * AF_INET will work *if* you have IPv4 configured into the kernel,
1473 	 * but, apparently, some systems have network adapters but have
1474 	 * kernels without IPv4 support.
1475 	 *
1476 	 * AF_INET6 will work *if* you have IPv6 configured into the
1477 	 * kernel, but if you don't have AF_INET, you might not have
1478 	 * AF_INET6, either (that is, independently on its own grounds).
1479 	 *
1480 	 * AF_PACKET would work, except that some of these calls should
1481 	 * work even if you *don't* have capture permission (you should be
1482 	 * able to enumerate interfaces and get information about them
1483 	 * without capture permission; you shouldn't get a failure until
1484 	 * you try pcap_activate()).  (If you don't allow programs to
1485 	 * get as much information as possible about interfaces if you
1486 	 * don't have permission to capture, you run the risk of users
1487 	 * asking "why isn't it showing XXX" - or, worse, if you don't
1488 	 * show interfaces *at all* if you don't have permission to
1489 	 * capture on them, "why do no interfaces show up?" - when the
1490 	 * real problem is a permissions problem.  Error reports of that
1491 	 * type require a lot more back-and-forth to debug, as evidenced
1492 	 * by many Wireshark bugs/mailing list questions/Q&A questions.)
1493 	 *
1494 	 * So:
1495 	 *
1496 	 * we first try an AF_NETLINK socket, where "try" includes
1497 	 * "try to do a device ioctl on it", as, in the future, once
1498 	 * pre-4.6 kernels are sufficiently rare, that will probably
1499 	 * be the mechanism most likely to work;
1500 	 *
1501 	 * if that fails, we try an AF_UNIX socket, as that's less
1502 	 * likely to be configured out on a networking-capable system
1503 	 * than is IP;
1504 	 *
1505 	 * if that fails, we try an AF_INET6 socket;
1506 	 *
1507 	 * if that fails, we try an AF_INET socket.
1508 	 */
1509 	fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
1510 	if (fd != -1) {
1511 		/*
1512 		 * OK, let's make sure we can do an SIOCGIFNAME
1513 		 * ioctl.
1514 		 */
1515 		struct ifreq ifr;
1516 
1517 		memset(&ifr, 0, sizeof(ifr));
1518 		if (ioctl(fd, SIOCGIFNAME, &ifr) == 0 ||
1519 		    errno != EOPNOTSUPP) {
1520 			/*
1521 			 * It succeeded, or failed for some reason
1522 			 * other than "netlink sockets don't support
1523 			 * device ioctls".  Go with the AF_NETLINK
1524 			 * socket.
1525 			 */
1526 			return (fd);
1527 		}
1528 
1529 		/*
1530 		 * OK, that didn't work, so it's as bad as "netlink
1531 		 * sockets aren't available".  Close the socket and
1532 		 * drive on.
1533 		 */
1534 		close(fd);
1535 	}
1536 
1537 	/*
1538 	 * Now try an AF_UNIX socket.
1539 	 */
1540 	fd = socket(AF_UNIX, SOCK_RAW, 0);
1541 	if (fd != -1) {
1542 		/*
1543 		 * OK, we got it!
1544 		 */
1545 		return (fd);
1546 	}
1547 
1548 	/*
1549 	 * Now try an AF_INET6 socket.
1550 	 */
1551 	fd = socket(AF_INET6, SOCK_DGRAM, 0);
1552 	if (fd != -1) {
1553 		return (fd);
1554 	}
1555 
1556 	/*
1557 	 * Now try an AF_INET socket.
1558 	 *
1559 	 * XXX - if that fails, is there anything else we should try?
1560 	 * AF_CAN, for embedded systems in vehicles, in case they're
1561 	 * built without Internet protocol support?  Any other socket
1562 	 * types popular in non-Internet embedded systems?
1563 	 */
1564 	return (socket(AF_INET, SOCK_DGRAM, 0));
1565 }
1566 
1567 /*
1568  * Get additional flags for a device, using SIOCGIFMEDIA.
1569  */
1570 static int
1571 get_if_flags(const char *name, bpf_u_int32 *flags, char *errbuf)
1572 {
1573 	int sock;
1574 	FILE *fh;
1575 	unsigned int arptype;
1576 	struct ifreq ifr;
1577 	struct ethtool_value info;
1578 
1579 	if (*flags & PCAP_IF_LOOPBACK) {
1580 		/*
1581 		 * Loopback devices aren't wireless, and "connected"/
1582 		 * "disconnected" doesn't apply to them.
1583 		 */
1584 		*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
1585 		return 0;
1586 	}
1587 
1588 	sock = get_if_ioctl_socket();
1589 	if (sock == -1) {
1590 		pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE, errno,
1591 		    "Can't create socket to get ethtool information for %s",
1592 		    name);
1593 		return -1;
1594 	}
1595 
1596 	/*
1597 	 * OK, what type of network is this?
1598 	 * In particular, is it wired or wireless?
1599 	 */
1600 	if (is_wifi(name)) {
1601 		/*
1602 		 * Wi-Fi, hence wireless.
1603 		 */
1604 		*flags |= PCAP_IF_WIRELESS;
1605 	} else {
1606 		/*
1607 		 * OK, what does /sys/class/net/{if_name}/type contain?
1608 		 * (We don't use that for Wi-Fi, as it'll report
1609 		 * "Ethernet", i.e. ARPHRD_ETHER, for non-monitor-
1610 		 * mode devices.)
1611 		 */
1612 		char *pathstr;
1613 
1614 		if (asprintf(&pathstr, "/sys/class/net/%s/type", name) == -1) {
1615 			snprintf(errbuf, PCAP_ERRBUF_SIZE,
1616 			    "%s: Can't generate path name string for /sys/class/net device",
1617 			    name);
1618 			close(sock);
1619 			return -1;
1620 		}
1621 		fh = fopen(pathstr, "r");
1622 		if (fh != NULL) {
1623 			if (fscanf(fh, "%u", &arptype) == 1) {
1624 				/*
1625 				 * OK, we got an ARPHRD_ type; what is it?
1626 				 */
1627 				switch (arptype) {
1628 
1629 				case ARPHRD_LOOPBACK:
1630 					/*
1631 					 * These are types to which
1632 					 * "connected" and "disconnected"
1633 					 * don't apply, so don't bother
1634 					 * asking about it.
1635 					 *
1636 					 * XXX - add other types?
1637 					 */
1638 					close(sock);
1639 					fclose(fh);
1640 					free(pathstr);
1641 					return 0;
1642 
1643 				case ARPHRD_IRDA:
1644 				case ARPHRD_IEEE80211:
1645 				case ARPHRD_IEEE80211_PRISM:
1646 				case ARPHRD_IEEE80211_RADIOTAP:
1647 #ifdef ARPHRD_IEEE802154
1648 				case ARPHRD_IEEE802154:
1649 #endif
1650 #ifdef ARPHRD_IEEE802154_MONITOR
1651 				case ARPHRD_IEEE802154_MONITOR:
1652 #endif
1653 #ifdef ARPHRD_6LOWPAN
1654 				case ARPHRD_6LOWPAN:
1655 #endif
1656 					/*
1657 					 * Various wireless types.
1658 					 */
1659 					*flags |= PCAP_IF_WIRELESS;
1660 					break;
1661 				}
1662 			}
1663 			fclose(fh);
1664 		}
1665 		free(pathstr);
1666 	}
1667 
1668 #ifdef ETHTOOL_GLINK
1669 	memset(&ifr, 0, sizeof(ifr));
1670 	pcap_strlcpy(ifr.ifr_name, name, sizeof(ifr.ifr_name));
1671 	info.cmd = ETHTOOL_GLINK;
1672 	/*
1673 	 * XXX - while Valgrind handles SIOCETHTOOL and knows that
1674 	 * the ETHTOOL_GLINK command sets the .data member of the
1675 	 * structure, Memory Sanitizer doesn't yet do so:
1676 	 *
1677 	 *    https://bugs.llvm.org/show_bug.cgi?id=45814
1678 	 *
1679 	 * For now, we zero it out to squelch warnings; if the bug
1680 	 * in question is fixed, we can remove this.
1681 	 */
1682 	info.data = 0;
1683 	ifr.ifr_data = (caddr_t)&info;
1684 	if (ioctl(sock, SIOCETHTOOL, &ifr) == -1) {
1685 		int save_errno = errno;
1686 
1687 		switch (save_errno) {
1688 
1689 		case EOPNOTSUPP:
1690 		case EINVAL:
1691 			/*
1692 			 * OK, this OS version or driver doesn't support
1693 			 * asking for this information.
1694 			 * XXX - distinguish between "this doesn't
1695 			 * support ethtool at all because it's not
1696 			 * that type of device" vs. "this doesn't
1697 			 * support ethtool even though it's that
1698 			 * type of device", and return "unknown".
1699 			 */
1700 			*flags |= PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE;
1701 			close(sock);
1702 			return 0;
1703 
1704 		case ENODEV:
1705 			/*
1706 			 * OK, no such device.
1707 			 * The user will find that out when they try to
1708 			 * activate the device; just say "OK" and
1709 			 * don't set anything.
1710 			 */
1711 			close(sock);
1712 			return 0;
1713 
1714 		default:
1715 			/*
1716 			 * Other error.
1717 			 */
1718 			pcap_fmt_errmsg_for_errno(errbuf, PCAP_ERRBUF_SIZE,
1719 			    save_errno,
1720 			    "%s: SIOCETHTOOL(ETHTOOL_GLINK) ioctl failed",
1721 			    name);
1722 			close(sock);
1723 			return -1;
1724 		}
1725 	}
1726 
1727 	/*
1728 	 * Is it connected?
1729 	 */
1730 	if (info.data) {
1731 		/*
1732 		 * It's connected.
1733 		 */
1734 		*flags |= PCAP_IF_CONNECTION_STATUS_CONNECTED;
1735 	} else {
1736 		/*
1737 		 * It's disconnected.
1738 		 */
1739 		*flags |= PCAP_IF_CONNECTION_STATUS_DISCONNECTED;
1740 	}
1741 #endif
1742 
1743 	close(sock);
1744 	return 0;
1745 }
1746 
1747 int
1748 pcap_platform_finddevs(pcap_if_list_t *devlistp, char *errbuf)
1749 {
1750 	/*
1751 	 * Get the list of regular interfaces first.
1752 	 */
1753 	if (pcap_findalldevs_interfaces(devlistp, errbuf, can_be_bound,
1754 	    get_if_flags) == -1)
1755 		return (-1);	/* failure */
1756 
1757 	/*
1758 	 * Add the "any" device.
1759 	 * As it refers to all network devices, not to any particular
1760 	 * network device, the notion of "connected" vs. "disconnected"
1761 	 * doesn't apply.
1762 	 */
1763 	if (add_dev(devlistp, "any",
1764 	    PCAP_IF_UP|PCAP_IF_RUNNING|PCAP_IF_CONNECTION_STATUS_NOT_APPLICABLE,
1765 	    any_descr, errbuf) == NULL)
1766 		return (-1);
1767 
1768 	return (0);
1769 }
1770 
1771 /*
1772  * Set direction flag: Which packets do we accept on a forwarding
1773  * single device? IN, OUT or both?
1774  */
1775 static int
1776 pcap_setdirection_linux(pcap_t *handle, pcap_direction_t d)
1777 {
1778 	/*
1779 	 * It's guaranteed, at this point, that d is a valid
1780 	 * direction value.
1781 	 */
1782 	handle->direction = d;
1783 	return 0;
1784 }
1785 
1786 static int
1787 is_wifi(const char *device)
1788 {
1789 	char *pathstr;
1790 	struct stat statb;
1791 
1792 	/*
1793 	 * See if there's a sysfs wireless directory for it.
1794 	 * If so, it's a wireless interface.
1795 	 */
1796 	if (asprintf(&pathstr, "/sys/class/net/%s/wireless", device) == -1) {
1797 		/*
1798 		 * Just give up here.
1799 		 */
1800 		return 0;
1801 	}
1802 	if (stat(pathstr, &statb) == 0) {
1803 		free(pathstr);
1804 		return 1;
1805 	}
1806 	free(pathstr);
1807 
1808 	return 0;
1809 }
1810 
1811 /*
1812  *  Linux uses the ARP hardware type to identify the type of an
1813  *  interface. pcap uses the DLT_xxx constants for this. This
1814  *  function takes a pointer to a "pcap_t", and an ARPHRD_xxx
1815  *  constant, as arguments, and sets "handle->linktype" to the
1816  *  appropriate DLT_XXX constant and sets "handle->offset" to
1817  *  the appropriate value (to make "handle->offset" plus link-layer
1818  *  header length be a multiple of 4, so that the link-layer payload
1819  *  will be aligned on a 4-byte boundary when capturing packets).
1820  *  (If the offset isn't set here, it'll be 0; add code as appropriate
1821  *  for cases where it shouldn't be 0.)
1822  *
1823  *  If "cooked_ok" is non-zero, we can use DLT_LINUX_SLL and capture
1824  *  in cooked mode; otherwise, we can't use cooked mode, so we have
1825  *  to pick some type that works in raw mode, or fail.
1826  *
1827  *  Sets the link type to -1 if unable to map the type.
1828  */
1829 static void map_arphrd_to_dlt(pcap_t *handle, int arptype,
1830 			      const char *device, int cooked_ok)
1831 {
1832 	static const char cdma_rmnet[] = "cdma_rmnet";
1833 
1834 	switch (arptype) {
1835 
1836 	case ARPHRD_ETHER:
1837 		/*
1838 		 * For various annoying reasons having to do with DHCP
1839 		 * software, some versions of Android give the mobile-
1840 		 * phone-network interface an ARPHRD_ value of
1841 		 * ARPHRD_ETHER, even though the packets supplied by
1842 		 * that interface have no link-layer header, and begin
1843 		 * with an IP header, so that the ARPHRD_ value should
1844 		 * be ARPHRD_NONE.
1845 		 *
1846 		 * Detect those devices by checking the device name, and
1847 		 * use DLT_RAW for them.
1848 		 */
1849 		if (strncmp(device, cdma_rmnet, sizeof cdma_rmnet - 1) == 0) {
1850 			handle->linktype = DLT_RAW;
1851 			return;
1852 		}
1853 
1854 		/*
1855 		 * Is this a real Ethernet device?  If so, give it a
1856 		 * link-layer-type list with DLT_EN10MB and DLT_DOCSIS, so
1857 		 * that an application can let you choose it, in case you're
1858 		 * capturing DOCSIS traffic that a Cisco Cable Modem
1859 		 * Termination System is putting out onto an Ethernet (it
1860 		 * doesn't put an Ethernet header onto the wire, it puts raw
1861 		 * DOCSIS frames out on the wire inside the low-level
1862 		 * Ethernet framing).
1863 		 *
1864 		 * XXX - are there any other sorts of "fake Ethernet" that
1865 		 * have ARPHRD_ETHER but that shouldn't offer DLT_DOCSIS as
1866 		 * a Cisco CMTS won't put traffic onto it or get traffic
1867 		 * bridged onto it?  ISDN is handled in "setup_socket()",
1868 		 * as we fall back on cooked mode there, and we use
1869 		 * is_wifi() to check for 802.11 devices; are there any
1870 		 * others?
1871 		 */
1872 		if (!is_wifi(device)) {
1873 			int ret;
1874 
1875 			/*
1876 			 * This is not a Wi-Fi device but it could be
1877 			 * a DSA master/management network device.
1878 			 */
1879 			ret = iface_dsa_get_proto_info(device, handle);
1880 			if (ret < 0)
1881 				return;
1882 
1883 			if (ret == 1) {
1884 				/*
1885 				 * This is a DSA master/management network
1886 				 * device linktype is already set by
1887 				 * iface_dsa_get_proto_info() set an
1888 				 * appropriate offset here.
1889 				 */
1890 				handle->offset = 2;
1891 				break;
1892 			}
1893 
1894 			/*
1895 			 * It's not a Wi-Fi device; offer DOCSIS.
1896 			 */
1897 			handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
1898 			/*
1899 			 * If that fails, just leave the list empty.
1900 			 */
1901 			if (handle->dlt_list != NULL) {
1902 				handle->dlt_list[0] = DLT_EN10MB;
1903 				handle->dlt_list[1] = DLT_DOCSIS;
1904 				handle->dlt_count = 2;
1905 			}
1906 		}
1907 		/* FALLTHROUGH */
1908 
1909 	case ARPHRD_METRICOM:
1910 	case ARPHRD_LOOPBACK:
1911 		handle->linktype = DLT_EN10MB;
1912 		handle->offset = 2;
1913 		break;
1914 
1915 	case ARPHRD_EETHER:
1916 		handle->linktype = DLT_EN3MB;
1917 		break;
1918 
1919 	case ARPHRD_AX25:
1920 		handle->linktype = DLT_AX25_KISS;
1921 		break;
1922 
1923 	case ARPHRD_PRONET:
1924 		handle->linktype = DLT_PRONET;
1925 		break;
1926 
1927 	case ARPHRD_CHAOS:
1928 		handle->linktype = DLT_CHAOS;
1929 		break;
1930 #ifndef ARPHRD_CAN
1931 #define ARPHRD_CAN 280
1932 #endif
1933 	case ARPHRD_CAN:
1934 		handle->linktype = DLT_CAN_SOCKETCAN;
1935 		break;
1936 
1937 #ifndef ARPHRD_IEEE802_TR
1938 #define ARPHRD_IEEE802_TR 800	/* From Linux 2.4 */
1939 #endif
1940 	case ARPHRD_IEEE802_TR:
1941 	case ARPHRD_IEEE802:
1942 		handle->linktype = DLT_IEEE802;
1943 		handle->offset = 2;
1944 		break;
1945 
1946 	case ARPHRD_ARCNET:
1947 		handle->linktype = DLT_ARCNET_LINUX;
1948 		break;
1949 
1950 #ifndef ARPHRD_FDDI	/* From Linux 2.2.13 */
1951 #define ARPHRD_FDDI	774
1952 #endif
1953 	case ARPHRD_FDDI:
1954 		handle->linktype = DLT_FDDI;
1955 		handle->offset = 3;
1956 		break;
1957 
1958 #ifndef ARPHRD_ATM  /* FIXME: How to #include this? */
1959 #define ARPHRD_ATM 19
1960 #endif
1961 	case ARPHRD_ATM:
1962 		/*
1963 		 * The Classical IP implementation in ATM for Linux
1964 		 * supports both what RFC 1483 calls "LLC Encapsulation",
1965 		 * in which each packet has an LLC header, possibly
1966 		 * with a SNAP header as well, prepended to it, and
1967 		 * what RFC 1483 calls "VC Based Multiplexing", in which
1968 		 * different virtual circuits carry different network
1969 		 * layer protocols, and no header is prepended to packets.
1970 		 *
1971 		 * They both have an ARPHRD_ type of ARPHRD_ATM, so
1972 		 * you can't use the ARPHRD_ type to find out whether
1973 		 * captured packets will have an LLC header, and,
1974 		 * while there's a socket ioctl to *set* the encapsulation
1975 		 * type, there's no ioctl to *get* the encapsulation type.
1976 		 *
1977 		 * This means that
1978 		 *
1979 		 *	programs that dissect Linux Classical IP frames
1980 		 *	would have to check for an LLC header and,
1981 		 *	depending on whether they see one or not, dissect
1982 		 *	the frame as LLC-encapsulated or as raw IP (I
1983 		 *	don't know whether there's any traffic other than
1984 		 *	IP that would show up on the socket, or whether
1985 		 *	there's any support for IPv6 in the Linux
1986 		 *	Classical IP code);
1987 		 *
1988 		 *	filter expressions would have to compile into
1989 		 *	code that checks for an LLC header and does
1990 		 *	the right thing.
1991 		 *
1992 		 * Both of those are a nuisance - and, at least on systems
1993 		 * that support PF_PACKET sockets, we don't have to put
1994 		 * up with those nuisances; instead, we can just capture
1995 		 * in cooked mode.  That's what we'll do, if we can.
1996 		 * Otherwise, we'll just fail.
1997 		 */
1998 		if (cooked_ok)
1999 			handle->linktype = DLT_LINUX_SLL;
2000 		else
2001 			handle->linktype = -1;
2002 		break;
2003 
2004 #ifndef ARPHRD_IEEE80211  /* From Linux 2.4.6 */
2005 #define ARPHRD_IEEE80211 801
2006 #endif
2007 	case ARPHRD_IEEE80211:
2008 		handle->linktype = DLT_IEEE802_11;
2009 		break;
2010 
2011 #ifndef ARPHRD_IEEE80211_PRISM  /* From Linux 2.4.18 */
2012 #define ARPHRD_IEEE80211_PRISM 802
2013 #endif
2014 	case ARPHRD_IEEE80211_PRISM:
2015 		handle->linktype = DLT_PRISM_HEADER;
2016 		break;
2017 
2018 #ifndef ARPHRD_IEEE80211_RADIOTAP /* new */
2019 #define ARPHRD_IEEE80211_RADIOTAP 803
2020 #endif
2021 	case ARPHRD_IEEE80211_RADIOTAP:
2022 		handle->linktype = DLT_IEEE802_11_RADIO;
2023 		break;
2024 
2025 	case ARPHRD_PPP:
2026 		/*
2027 		 * Some PPP code in the kernel supplies no link-layer
2028 		 * header whatsoever to PF_PACKET sockets; other PPP
2029 		 * code supplies PPP link-layer headers ("syncppp.c");
2030 		 * some PPP code might supply random link-layer
2031 		 * headers (PPP over ISDN - there's code in Ethereal,
2032 		 * for example, to cope with PPP-over-ISDN captures
2033 		 * with which the Ethereal developers have had to cope,
2034 		 * heuristically trying to determine which of the
2035 		 * oddball link-layer headers particular packets have).
2036 		 *
2037 		 * As such, we just punt, and run all PPP interfaces
2038 		 * in cooked mode, if we can; otherwise, we just treat
2039 		 * it as DLT_RAW, for now - if somebody needs to capture,
2040 		 * on a 2.0[.x] kernel, on PPP devices that supply a
2041 		 * link-layer header, they'll have to add code here to
2042 		 * map to the appropriate DLT_ type (possibly adding a
2043 		 * new DLT_ type, if necessary).
2044 		 */
2045 		if (cooked_ok)
2046 			handle->linktype = DLT_LINUX_SLL;
2047 		else {
2048 			/*
2049 			 * XXX - handle ISDN types here?  We can't fall
2050 			 * back on cooked sockets, so we'd have to
2051 			 * figure out from the device name what type of
2052 			 * link-layer encapsulation it's using, and map
2053 			 * that to an appropriate DLT_ value, meaning
2054 			 * we'd map "isdnN" devices to DLT_RAW (they
2055 			 * supply raw IP packets with no link-layer
2056 			 * header) and "isdY" devices to a new DLT_I4L_IP
2057 			 * type that has only an Ethernet packet type as
2058 			 * a link-layer header.
2059 			 *
2060 			 * But sometimes we seem to get random crap
2061 			 * in the link-layer header when capturing on
2062 			 * ISDN devices....
2063 			 */
2064 			handle->linktype = DLT_RAW;
2065 		}
2066 		break;
2067 
2068 #ifndef ARPHRD_CISCO
2069 #define ARPHRD_CISCO 513 /* previously ARPHRD_HDLC */
2070 #endif
2071 	case ARPHRD_CISCO:
2072 		handle->linktype = DLT_C_HDLC;
2073 		break;
2074 
2075 	/* Not sure if this is correct for all tunnels, but it
2076 	 * works for CIPE */
2077 	case ARPHRD_TUNNEL:
2078 #ifndef ARPHRD_SIT
2079 #define ARPHRD_SIT 776	/* From Linux 2.2.13 */
2080 #endif
2081 	case ARPHRD_SIT:
2082 	case ARPHRD_CSLIP:
2083 	case ARPHRD_SLIP6:
2084 	case ARPHRD_CSLIP6:
2085 	case ARPHRD_ADAPT:
2086 	case ARPHRD_SLIP:
2087 #ifndef ARPHRD_RAWHDLC
2088 #define ARPHRD_RAWHDLC 518
2089 #endif
2090 	case ARPHRD_RAWHDLC:
2091 #ifndef ARPHRD_DLCI
2092 #define ARPHRD_DLCI 15
2093 #endif
2094 	case ARPHRD_DLCI:
2095 		/*
2096 		 * XXX - should some of those be mapped to DLT_LINUX_SLL
2097 		 * instead?  Should we just map all of them to DLT_LINUX_SLL?
2098 		 */
2099 		handle->linktype = DLT_RAW;
2100 		break;
2101 
2102 #ifndef ARPHRD_FRAD
2103 #define ARPHRD_FRAD 770
2104 #endif
2105 	case ARPHRD_FRAD:
2106 		handle->linktype = DLT_FRELAY;
2107 		break;
2108 
2109 	case ARPHRD_LOCALTLK:
2110 		handle->linktype = DLT_LTALK;
2111 		break;
2112 
2113 	case 18:
2114 		/*
2115 		 * RFC 4338 defines an encapsulation for IP and ARP
2116 		 * packets that's compatible with the RFC 2625
2117 		 * encapsulation, but that uses a different ARP
2118 		 * hardware type and hardware addresses.  That
2119 		 * ARP hardware type is 18; Linux doesn't define
2120 		 * any ARPHRD_ value as 18, but if it ever officially
2121 		 * supports RFC 4338-style IP-over-FC, it should define
2122 		 * one.
2123 		 *
2124 		 * For now, we map it to DLT_IP_OVER_FC, in the hopes
2125 		 * that this will encourage its use in the future,
2126 		 * should Linux ever officially support RFC 4338-style
2127 		 * IP-over-FC.
2128 		 */
2129 		handle->linktype = DLT_IP_OVER_FC;
2130 		break;
2131 
2132 #ifndef ARPHRD_FCPP
2133 #define ARPHRD_FCPP	784
2134 #endif
2135 	case ARPHRD_FCPP:
2136 #ifndef ARPHRD_FCAL
2137 #define ARPHRD_FCAL	785
2138 #endif
2139 	case ARPHRD_FCAL:
2140 #ifndef ARPHRD_FCPL
2141 #define ARPHRD_FCPL	786
2142 #endif
2143 	case ARPHRD_FCPL:
2144 #ifndef ARPHRD_FCFABRIC
2145 #define ARPHRD_FCFABRIC	787
2146 #endif
2147 	case ARPHRD_FCFABRIC:
2148 		/*
2149 		 * Back in 2002, Donald Lee at Cray wanted a DLT_ for
2150 		 * IP-over-FC:
2151 		 *
2152 		 *	https://www.mail-archive.com/tcpdump-workers@sandelman.ottawa.on.ca/msg01043.html
2153 		 *
2154 		 * and one was assigned.
2155 		 *
2156 		 * In a later private discussion (spun off from a message
2157 		 * on the ethereal-users list) on how to get that DLT_
2158 		 * value in libpcap on Linux, I ended up deciding that
2159 		 * the best thing to do would be to have him tweak the
2160 		 * driver to set the ARPHRD_ value to some ARPHRD_FCxx
2161 		 * type, and map all those types to DLT_IP_OVER_FC:
2162 		 *
2163 		 *	I've checked into the libpcap and tcpdump CVS tree
2164 		 *	support for DLT_IP_OVER_FC.  In order to use that,
2165 		 *	you'd have to modify your modified driver to return
2166 		 *	one of the ARPHRD_FCxxx types, in "fcLINUXfcp.c" -
2167 		 *	change it to set "dev->type" to ARPHRD_FCFABRIC, for
2168 		 *	example (the exact value doesn't matter, it can be
2169 		 *	any of ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, or
2170 		 *	ARPHRD_FCFABRIC).
2171 		 *
2172 		 * 11 years later, Christian Svensson wanted to map
2173 		 * various ARPHRD_ values to DLT_FC_2 and
2174 		 * DLT_FC_2_WITH_FRAME_DELIMS for raw Fibre Channel
2175 		 * frames:
2176 		 *
2177 		 *	https://github.com/mcr/libpcap/pull/29
2178 		 *
2179 		 * There doesn't seem to be any network drivers that uses
2180 		 * any of the ARPHRD_FC* values for IP-over-FC, and
2181 		 * it's not exactly clear what the "Dummy types for non
2182 		 * ARP hardware" are supposed to mean (link-layer
2183 		 * header type?  Physical network type?), so it's
2184 		 * not exactly clear why the ARPHRD_FC* types exist
2185 		 * in the first place.
2186 		 *
2187 		 * For now, we map them to DLT_FC_2, and provide an
2188 		 * option of DLT_FC_2_WITH_FRAME_DELIMS, as well as
2189 		 * DLT_IP_OVER_FC just in case there's some old
2190 		 * driver out there that uses one of those types for
2191 		 * IP-over-FC on which somebody wants to capture
2192 		 * packets.
2193 		 */
2194 		handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 3);
2195 		/*
2196 		 * If that fails, just leave the list empty.
2197 		 */
2198 		if (handle->dlt_list != NULL) {
2199 			handle->dlt_list[0] = DLT_FC_2;
2200 			handle->dlt_list[1] = DLT_FC_2_WITH_FRAME_DELIMS;
2201 			handle->dlt_list[2] = DLT_IP_OVER_FC;
2202 			handle->dlt_count = 3;
2203 		}
2204 		handle->linktype = DLT_FC_2;
2205 		break;
2206 
2207 #ifndef ARPHRD_IRDA
2208 #define ARPHRD_IRDA	783
2209 #endif
2210 	case ARPHRD_IRDA:
2211 		/* Don't expect IP packet out of this interfaces... */
2212 		handle->linktype = DLT_LINUX_IRDA;
2213 		/* We need to save packet direction for IrDA decoding,
2214 		 * so let's use "Linux-cooked" mode. Jean II
2215 		 *
2216 		 * XXX - this is handled in setup_socket(). */
2217 		/* handlep->cooked = 1; */
2218 		break;
2219 
2220 	/* ARPHRD_LAPD is unofficial and randomly allocated, if reallocation
2221 	 * is needed, please report it to <daniele@orlandi.com> */
2222 #ifndef ARPHRD_LAPD
2223 #define ARPHRD_LAPD	8445
2224 #endif
2225 	case ARPHRD_LAPD:
2226 		/* Don't expect IP packet out of this interfaces... */
2227 		handle->linktype = DLT_LINUX_LAPD;
2228 		break;
2229 
2230 #ifndef ARPHRD_NONE
2231 #define ARPHRD_NONE	0xFFFE
2232 #endif
2233 	case ARPHRD_NONE:
2234 		/*
2235 		 * No link-layer header; packets are just IP
2236 		 * packets, so use DLT_RAW.
2237 		 */
2238 		handle->linktype = DLT_RAW;
2239 		break;
2240 
2241 #ifndef ARPHRD_IEEE802154
2242 #define ARPHRD_IEEE802154      804
2243 #endif
2244        case ARPHRD_IEEE802154:
2245                handle->linktype =  DLT_IEEE802_15_4_NOFCS;
2246                break;
2247 
2248 #ifndef ARPHRD_NETLINK
2249 #define ARPHRD_NETLINK	824
2250 #endif
2251 	case ARPHRD_NETLINK:
2252 		handle->linktype = DLT_NETLINK;
2253 		/*
2254 		 * We need to use cooked mode, so that in sll_protocol we
2255 		 * pick up the netlink protocol type such as NETLINK_ROUTE,
2256 		 * NETLINK_GENERIC, NETLINK_FIB_LOOKUP, etc.
2257 		 *
2258 		 * XXX - this is handled in setup_socket().
2259 		 */
2260 		/* handlep->cooked = 1; */
2261 		break;
2262 
2263 #ifndef ARPHRD_VSOCKMON
2264 #define ARPHRD_VSOCKMON	826
2265 #endif
2266 	case ARPHRD_VSOCKMON:
2267 		handle->linktype = DLT_VSOCK;
2268 		break;
2269 
2270 	default:
2271 		handle->linktype = -1;
2272 		break;
2273 	}
2274 }
2275 
2276 static void
2277 set_dlt_list_cooked(pcap_t *handle)
2278 {
2279 	/*
2280 	 * Support both DLT_LINUX_SLL and DLT_LINUX_SLL2.
2281 	 */
2282 	handle->dlt_list = (u_int *) malloc(sizeof(u_int) * 2);
2283 
2284 	/*
2285 	 * If that failed, just leave the list empty.
2286 	 */
2287 	if (handle->dlt_list != NULL) {
2288 		handle->dlt_list[0] = DLT_LINUX_SLL;
2289 		handle->dlt_list[1] = DLT_LINUX_SLL2;
2290 		handle->dlt_count = 2;
2291 	}
2292 }
2293 
2294 /*
2295  * Try to set up a PF_PACKET socket.
2296  * Returns 0 on success and a PCAP_ERROR_ value on failure.
2297  */
2298 static int
2299 setup_socket(pcap_t *handle, int is_any_device)
2300 {
2301 	struct pcap_linux *handlep = handle->priv;
2302 	const char		*device = handle->opt.device;
2303 	int			status = 0;
2304 	int			sock_fd, arptype;
2305 	int			val;
2306 	int			err = 0;
2307 	struct packet_mreq	mr;
2308 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
2309 	int			bpf_extensions;
2310 	socklen_t		len = sizeof(bpf_extensions);
2311 #endif
2312 
2313 	/*
2314 	 * Open a socket with protocol family packet. If cooked is true,
2315 	 * we open a SOCK_DGRAM socket for the cooked interface, otherwise
2316 	 * we open a SOCK_RAW socket for the raw interface.
2317 	 *
2318 	 * The protocol is set to 0.  This means we will receive no
2319 	 * packets until we "bind" the socket with a non-zero
2320 	 * protocol.  This allows us to setup the ring buffers without
2321 	 * dropping any packets.
2322 	 */
2323 	sock_fd = is_any_device ?
2324 		socket(PF_PACKET, SOCK_DGRAM, 0) :
2325 		socket(PF_PACKET, SOCK_RAW, 0);
2326 
2327 	if (sock_fd == -1) {
2328 		if (errno == EPERM || errno == EACCES) {
2329 			/*
2330 			 * You don't have permission to open the
2331 			 * socket.
2332 			 */
2333 			status = PCAP_ERROR_PERM_DENIED;
2334 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2335 			    "Attempt to create packet socket failed - CAP_NET_RAW may be required");
2336 		} else {
2337 			/*
2338 			 * Other error.
2339 			 */
2340 			status = PCAP_ERROR;
2341 		}
2342 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2343 		    errno, "socket");
2344 		return status;
2345 	}
2346 
2347 	/*
2348 	 * Get the interface index of the loopback device.
2349 	 * If the attempt fails, don't fail, just set the
2350 	 * "handlep->lo_ifindex" to -1.
2351 	 *
2352 	 * XXX - can there be more than one device that loops
2353 	 * packets back, i.e. devices other than "lo"?  If so,
2354 	 * we'd need to find them all, and have an array of
2355 	 * indices for them, and check all of them in
2356 	 * "pcap_read_packet()".
2357 	 */
2358 	handlep->lo_ifindex = iface_get_id(sock_fd, "lo", handle->errbuf);
2359 
2360 	/*
2361 	 * Default value for offset to align link-layer payload
2362 	 * on a 4-byte boundary.
2363 	 */
2364 	handle->offset	 = 0;
2365 
2366 	/*
2367 	 * What kind of frames do we have to deal with? Fall back
2368 	 * to cooked mode if we have an unknown interface type
2369 	 * or a type we know doesn't work well in raw mode.
2370 	 */
2371 	if (!is_any_device) {
2372 		/* Assume for now we don't need cooked mode. */
2373 		handlep->cooked = 0;
2374 
2375 		if (handle->opt.rfmon) {
2376 			/*
2377 			 * We were asked to turn on monitor mode.
2378 			 * Do so before we get the link-layer type,
2379 			 * because entering monitor mode could change
2380 			 * the link-layer type.
2381 			 */
2382 			err = enter_rfmon_mode(handle, sock_fd, device);
2383 			if (err < 0) {
2384 				/* Hard failure */
2385 				close(sock_fd);
2386 				return err;
2387 			}
2388 			if (err == 0) {
2389 				/*
2390 				 * Nothing worked for turning monitor mode
2391 				 * on.
2392 				 */
2393 				close(sock_fd);
2394 				return PCAP_ERROR_RFMON_NOTSUP;
2395 			}
2396 
2397 			/*
2398 			 * Either monitor mode has been turned on for
2399 			 * the device, or we've been given a different
2400 			 * device to open for monitor mode.  If we've
2401 			 * been given a different device, use it.
2402 			 */
2403 			if (handlep->mondevice != NULL)
2404 				device = handlep->mondevice;
2405 		}
2406 		arptype	= iface_get_arptype(sock_fd, device, handle->errbuf);
2407 		if (arptype < 0) {
2408 			close(sock_fd);
2409 			return arptype;
2410 		}
2411 		map_arphrd_to_dlt(handle, arptype, device, 1);
2412 		if (handle->linktype == -1 ||
2413 		    handle->linktype == DLT_LINUX_SLL ||
2414 		    handle->linktype == DLT_LINUX_IRDA ||
2415 		    handle->linktype == DLT_LINUX_LAPD ||
2416 		    handle->linktype == DLT_NETLINK ||
2417 		    (handle->linktype == DLT_EN10MB &&
2418 		     (strncmp("isdn", device, 4) == 0 ||
2419 		      strncmp("isdY", device, 4) == 0))) {
2420 			/*
2421 			 * Unknown interface type (-1), or a
2422 			 * device we explicitly chose to run
2423 			 * in cooked mode (e.g., PPP devices),
2424 			 * or an ISDN device (whose link-layer
2425 			 * type we can only determine by using
2426 			 * APIs that may be different on different
2427 			 * kernels) - reopen in cooked mode.
2428 			 *
2429 			 * If the type is unknown, return a warning;
2430 			 * map_arphrd_to_dlt() has already set the
2431 			 * warning message.
2432 			 */
2433 			if (close(sock_fd) == -1) {
2434 				pcap_fmt_errmsg_for_errno(handle->errbuf,
2435 				    PCAP_ERRBUF_SIZE, errno, "close");
2436 				return PCAP_ERROR;
2437 			}
2438 			sock_fd = socket(PF_PACKET, SOCK_DGRAM, 0);
2439 			if (sock_fd < 0) {
2440 				/*
2441 				 * Fatal error.  We treat this as
2442 				 * a generic error; we already know
2443 				 * that we were able to open a
2444 				 * PF_PACKET/SOCK_RAW socket, so
2445 				 * any failure is a "this shouldn't
2446 				 * happen" case.
2447 				 */
2448 				pcap_fmt_errmsg_for_errno(handle->errbuf,
2449 				    PCAP_ERRBUF_SIZE, errno, "socket");
2450 				return PCAP_ERROR;
2451 			}
2452 			handlep->cooked = 1;
2453 
2454 			/*
2455 			 * Get rid of any link-layer type list
2456 			 * we allocated - this only supports cooked
2457 			 * capture.
2458 			 */
2459 			if (handle->dlt_list != NULL) {
2460 				free(handle->dlt_list);
2461 				handle->dlt_list = NULL;
2462 				handle->dlt_count = 0;
2463 				set_dlt_list_cooked(handle);
2464 			}
2465 
2466 			if (handle->linktype == -1) {
2467 				/*
2468 				 * Warn that we're falling back on
2469 				 * cooked mode; we may want to
2470 				 * update "map_arphrd_to_dlt()"
2471 				 * to handle the new type.
2472 				 */
2473 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2474 					"arptype %d not "
2475 					"supported by libpcap - "
2476 					"falling back to cooked "
2477 					"socket",
2478 					arptype);
2479 			}
2480 
2481 			/*
2482 			 * IrDA capture is not a real "cooked" capture,
2483 			 * it's IrLAP frames, not IP packets.  The
2484 			 * same applies to LAPD capture.
2485 			 */
2486 			if (handle->linktype != DLT_LINUX_IRDA &&
2487 			    handle->linktype != DLT_LINUX_LAPD &&
2488 			    handle->linktype != DLT_NETLINK)
2489 				handle->linktype = DLT_LINUX_SLL;
2490 			if (handle->linktype == -1) {
2491 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2492 				    "unknown arptype %d, defaulting to cooked mode",
2493 				    arptype);
2494 				status = PCAP_WARNING;
2495 			}
2496 		}
2497 
2498 		handlep->ifindex = iface_get_id(sock_fd, device,
2499 		    handle->errbuf);
2500 		if (handlep->ifindex == -1) {
2501 			close(sock_fd);
2502 			return PCAP_ERROR;
2503 		}
2504 
2505 		if ((err = iface_bind(sock_fd, handlep->ifindex,
2506 		    handle->errbuf, 0)) != 0) {
2507 			close(sock_fd);
2508 			return err;
2509 		}
2510 	} else {
2511 		/*
2512 		 * The "any" device.
2513 		 */
2514 		if (handle->opt.rfmon) {
2515 			/*
2516 			 * It doesn't support monitor mode.
2517 			 */
2518 			close(sock_fd);
2519 			return PCAP_ERROR_RFMON_NOTSUP;
2520 		}
2521 
2522 		/*
2523 		 * It uses cooked mode.
2524 		 */
2525 		handlep->cooked = 1;
2526 		handle->linktype = DLT_LINUX_SLL;
2527 		handle->dlt_list = NULL;
2528 		handle->dlt_count = 0;
2529 		set_dlt_list_cooked(handle);
2530 
2531 		/*
2532 		 * We're not bound to a device.
2533 		 * For now, we're using this as an indication
2534 		 * that we can't transmit; stop doing that only
2535 		 * if we figure out how to transmit in cooked
2536 		 * mode.
2537 		 */
2538 		handlep->ifindex = -1;
2539 	}
2540 
2541 	/*
2542 	 * Select promiscuous mode on if "promisc" is set.
2543 	 *
2544 	 * Do not turn allmulti mode on if we don't select
2545 	 * promiscuous mode - on some devices (e.g., Orinoco
2546 	 * wireless interfaces), allmulti mode isn't supported
2547 	 * and the driver implements it by turning promiscuous
2548 	 * mode on, and that screws up the operation of the
2549 	 * card as a normal networking interface, and on no
2550 	 * other platform I know of does starting a non-
2551 	 * promiscuous capture affect which multicast packets
2552 	 * are received by the interface.
2553 	 */
2554 
2555 	/*
2556 	 * Hmm, how can we set promiscuous mode on all interfaces?
2557 	 * I am not sure if that is possible at all.  For now, we
2558 	 * silently ignore attempts to turn promiscuous mode on
2559 	 * for the "any" device (so you don't have to explicitly
2560 	 * disable it in programs such as tcpdump).
2561 	 */
2562 
2563 	if (!is_any_device && handle->opt.promisc) {
2564 		memset(&mr, 0, sizeof(mr));
2565 		mr.mr_ifindex = handlep->ifindex;
2566 		mr.mr_type    = PACKET_MR_PROMISC;
2567 		if (setsockopt(sock_fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP,
2568 		    &mr, sizeof(mr)) == -1) {
2569 			pcap_fmt_errmsg_for_errno(handle->errbuf,
2570 			    PCAP_ERRBUF_SIZE, errno, "setsockopt (PACKET_ADD_MEMBERSHIP)");
2571 			close(sock_fd);
2572 			return PCAP_ERROR;
2573 		}
2574 	}
2575 
2576 	/*
2577 	 * Enable auxiliary data and reserve room for reconstructing
2578 	 * VLAN headers.
2579 	 *
2580 	 * XXX - is enabling auxiliary data necessary, now that we
2581 	 * only support memory-mapped capture?  The kernel's memory-mapped
2582 	 * capture code doesn't seem to check whether auxiliary data
2583 	 * is enabled, it seems to provide it whether it is or not.
2584 	 */
2585 	val = 1;
2586 	if (setsockopt(sock_fd, SOL_PACKET, PACKET_AUXDATA, &val,
2587 		       sizeof(val)) == -1 && errno != ENOPROTOOPT) {
2588 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2589 		    errno, "setsockopt (PACKET_AUXDATA)");
2590 		close(sock_fd);
2591 		return PCAP_ERROR;
2592 	}
2593 	handle->offset += VLAN_TAG_LEN;
2594 
2595 	/*
2596 	 * If we're in cooked mode, make the snapshot length
2597 	 * large enough to hold a "cooked mode" header plus
2598 	 * 1 byte of packet data (so we don't pass a byte
2599 	 * count of 0 to "recvfrom()").
2600 	 * XXX - we don't know whether this will be DLT_LINUX_SLL
2601 	 * or DLT_LINUX_SLL2, so make sure it's big enough for
2602 	 * a DLT_LINUX_SLL2 "cooked mode" header; a snapshot length
2603 	 * that small is silly anyway.
2604 	 */
2605 	if (handlep->cooked) {
2606 		if (handle->snapshot < SLL2_HDR_LEN + 1)
2607 			handle->snapshot = SLL2_HDR_LEN + 1;
2608 	}
2609 	handle->bufsize = handle->snapshot;
2610 
2611 	/*
2612 	 * Set the offset at which to insert VLAN tags.
2613 	 */
2614 	set_vlan_offset(handle);
2615 
2616 	if (handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO) {
2617 		int nsec_tstamps = 1;
2618 
2619 		if (setsockopt(sock_fd, SOL_SOCKET, SO_TIMESTAMPNS, &nsec_tstamps, sizeof(nsec_tstamps)) < 0) {
2620 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE, "setsockopt: unable to set SO_TIMESTAMPNS");
2621 			close(sock_fd);
2622 			return PCAP_ERROR;
2623 		}
2624 	}
2625 
2626 	/*
2627 	 * We've succeeded. Save the socket FD in the pcap structure.
2628 	 */
2629 	handle->fd = sock_fd;
2630 
2631 #if defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT)
2632 	/*
2633 	 * Can we generate special code for VLAN checks?
2634 	 * (XXX - what if we need the special code but it's not supported
2635 	 * by the OS?  Is that possible?)
2636 	 */
2637 	if (getsockopt(sock_fd, SOL_SOCKET, SO_BPF_EXTENSIONS,
2638 	    &bpf_extensions, &len) == 0) {
2639 		if (bpf_extensions >= SKF_AD_VLAN_TAG_PRESENT) {
2640 			/*
2641 			 * Yes, we can.  Request that we do so.
2642 			 */
2643 			handle->bpf_codegen_flags |= BPF_SPECIAL_VLAN_HANDLING;
2644 		}
2645 	}
2646 #endif /* defined(SO_BPF_EXTENSIONS) && defined(SKF_AD_VLAN_TAG_PRESENT) */
2647 
2648 	return status;
2649 }
2650 
2651 /*
2652  * Attempt to setup memory-mapped access.
2653  *
2654  * On success, returns 1, and sets *status to 0 if there are no warnings
2655  * or to a PCAP_WARNING_ code if there is a warning.
2656  *
2657  * On error, returns -1, and sets *status to the appropriate error code;
2658  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
2659  */
2660 static int
2661 setup_mmapped(pcap_t *handle, int *status)
2662 {
2663 	struct pcap_linux *handlep = handle->priv;
2664 	int ret;
2665 
2666 	/*
2667 	 * Attempt to allocate a buffer to hold the contents of one
2668 	 * packet, for use by the oneshot callback.
2669 	 */
2670 	handlep->oneshot_buffer = malloc(handle->snapshot);
2671 	if (handlep->oneshot_buffer == NULL) {
2672 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2673 		    errno, "can't allocate oneshot buffer");
2674 		*status = PCAP_ERROR;
2675 		return -1;
2676 	}
2677 
2678 	if (handle->opt.buffer_size == 0) {
2679 		/* by default request 2M for the ring buffer */
2680 		handle->opt.buffer_size = 2*1024*1024;
2681 	}
2682 	ret = prepare_tpacket_socket(handle);
2683 	if (ret == -1) {
2684 		free(handlep->oneshot_buffer);
2685 		handlep->oneshot_buffer = NULL;
2686 		*status = PCAP_ERROR;
2687 		return ret;
2688 	}
2689 	ret = create_ring(handle, status);
2690 	if (ret == -1) {
2691 		/*
2692 		 * Error attempting to enable memory-mapped capture;
2693 		 * fail.  create_ring() has set *status.
2694 		 */
2695 		free(handlep->oneshot_buffer);
2696 		handlep->oneshot_buffer = NULL;
2697 		return -1;
2698 	}
2699 
2700 	/*
2701 	 * Success.  *status has been set either to 0 if there are no
2702 	 * warnings or to a PCAP_WARNING_ value if there is a warning.
2703 	 *
2704 	 * handle->offset is used to get the current position into the rx ring.
2705 	 * handle->cc is used to store the ring size.
2706 	 */
2707 
2708 	/*
2709 	 * Set the timeout to use in poll() before returning.
2710 	 */
2711 	set_poll_timeout(handlep);
2712 
2713 	return 1;
2714 }
2715 
2716 /*
2717  * Attempt to set the socket to the specified version of the memory-mapped
2718  * header.
2719  *
2720  * Return 0 if we succeed; return 1 if we fail because that version isn't
2721  * supported; return -1 on any other error, and set handle->errbuf.
2722  */
2723 static int
2724 init_tpacket(pcap_t *handle, int version, const char *version_str)
2725 {
2726 	struct pcap_linux *handlep = handle->priv;
2727 	int val = version;
2728 	socklen_t len = sizeof(val);
2729 
2730 	/*
2731 	 * Probe whether kernel supports the specified TPACKET version;
2732 	 * this also gets the length of the header for that version.
2733 	 *
2734 	 * This socket option was introduced in 2.6.27, which was
2735 	 * also the first release with TPACKET_V2 support.
2736 	 */
2737 	if (getsockopt(handle->fd, SOL_PACKET, PACKET_HDRLEN, &val, &len) < 0) {
2738 		if (errno == EINVAL) {
2739 			/*
2740 			 * EINVAL means this specific version of TPACKET
2741 			 * is not supported. Tell the caller they can try
2742 			 * with a different one; if they've run out of
2743 			 * others to try, let them set the error message
2744 			 * appropriately.
2745 			 */
2746 			return 1;
2747 		}
2748 
2749 		/*
2750 		 * All other errors are fatal.
2751 		 */
2752 		if (errno == ENOPROTOOPT) {
2753 			/*
2754 			 * PACKET_HDRLEN isn't supported, which means
2755 			 * that memory-mapped capture isn't supported.
2756 			 * Indicate that in the message.
2757 			 */
2758 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2759 			    "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");
2760 		} else {
2761 			/*
2762 			 * Some unexpected error.
2763 			 */
2764 			pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2765 			    errno, "can't get %s header len on packet socket",
2766 			    version_str);
2767 		}
2768 		return -1;
2769 	}
2770 	handlep->tp_hdrlen = val;
2771 
2772 	val = version;
2773 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_VERSION, &val,
2774 			   sizeof(val)) < 0) {
2775 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
2776 		    errno, "can't activate %s on packet socket", version_str);
2777 		return -1;
2778 	}
2779 	handlep->tp_version = version;
2780 
2781 	return 0;
2782 }
2783 
2784 /*
2785  * Attempt to set the socket to version 3 of the memory-mapped header and,
2786  * if that fails because version 3 isn't supported, attempt to fall
2787  * back to version 2.  If version 2 isn't supported, just fail.
2788  *
2789  * Return 0 if we succeed and -1 on any other error, and set handle->errbuf.
2790  */
2791 static int
2792 prepare_tpacket_socket(pcap_t *handle)
2793 {
2794 	int ret;
2795 
2796 #ifdef HAVE_TPACKET3
2797 	/*
2798 	 * Try setting the version to TPACKET_V3.
2799 	 *
2800 	 * The only mode in which buffering is done on PF_PACKET
2801 	 * sockets, so that packets might not be delivered
2802 	 * immediately, is TPACKET_V3 mode.
2803 	 *
2804 	 * The buffering cannot be disabled in that mode, so
2805 	 * if the user has requested immediate mode, we don't
2806 	 * use TPACKET_V3.
2807 	 */
2808 	if (!handle->opt.immediate) {
2809 		ret = init_tpacket(handle, TPACKET_V3, "TPACKET_V3");
2810 		if (ret == 0) {
2811 			/*
2812 			 * Success.
2813 			 */
2814 			return 0;
2815 		}
2816 		if (ret == -1) {
2817 			/*
2818 			 * We failed for some reason other than "the
2819 			 * kernel doesn't support TPACKET_V3".
2820 			 */
2821 			return -1;
2822 		}
2823 
2824 		/*
2825 		 * This means it returned 1, which means "the kernel
2826 		 * doesn't support TPACKET_V3"; try TPACKET_V2.
2827 		 */
2828 	}
2829 #endif /* HAVE_TPACKET3 */
2830 
2831 	/*
2832 	 * Try setting the version to TPACKET_V2.
2833 	 */
2834 	ret = init_tpacket(handle, TPACKET_V2, "TPACKET_V2");
2835 	if (ret == 0) {
2836 		/*
2837 		 * Success.
2838 		 */
2839 		return 0;
2840 	}
2841 
2842 	if (ret == 1) {
2843 		/*
2844 		 * OK, the kernel supports memory-mapped capture, but
2845 		 * not TPACKET_V2.  Set the error message appropriately.
2846 		 */
2847 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
2848 		    "Kernel doesn't support TPACKET_V2; a 2.6.27 or later kernel is required");
2849 	}
2850 
2851 	/*
2852 	 * We failed.
2853 	 */
2854 	return -1;
2855 }
2856 
2857 #define MAX(a,b) ((a)>(b)?(a):(b))
2858 
2859 /*
2860  * Attempt to set up memory-mapped access.
2861  *
2862  * On success, returns 1, and sets *status to 0 if there are no warnings
2863  * or to a PCAP_WARNING_ code if there is a warning.
2864  *
2865  * On error, returns -1, and sets *status to the appropriate error code;
2866  * if that is PCAP_ERROR, sets handle->errbuf to the appropriate message.
2867  */
2868 static int
2869 create_ring(pcap_t *handle, int *status)
2870 {
2871 	struct pcap_linux *handlep = handle->priv;
2872 	unsigned i, j, frames_per_block;
2873 #ifdef HAVE_TPACKET3
2874 	/*
2875 	 * For sockets using TPACKET_V2, the extra stuff at the end of a
2876 	 * struct tpacket_req3 will be ignored, so this is OK even for
2877 	 * those sockets.
2878 	 */
2879 	struct tpacket_req3 req;
2880 #else
2881 	struct tpacket_req req;
2882 #endif
2883 	socklen_t len;
2884 	unsigned int sk_type, tp_reserve, maclen, tp_hdrlen, netoff, macoff;
2885 	unsigned int frame_size;
2886 
2887 	/*
2888 	 * Start out assuming no warnings or errors.
2889 	 */
2890 	*status = 0;
2891 
2892 	/*
2893 	 * Reserve space for VLAN tag reconstruction.
2894 	 */
2895 	tp_reserve = VLAN_TAG_LEN;
2896 
2897 	/*
2898 	 * If we're capturing in cooked mode, reserve space for
2899 	 * a DLT_LINUX_SLL2 header; we don't know yet whether
2900 	 * we'll be using DLT_LINUX_SLL or DLT_LINUX_SLL2, as
2901 	 * that can be changed on an open device, so we reserve
2902 	 * space for the larger of the two.
2903 	 *
2904 	 * XXX - we assume that the kernel is still adding
2905 	 * 16 bytes of extra space, so we subtract 16 from
2906 	 * SLL2_HDR_LEN to get the additional space needed.
2907 	 * (Are they doing that for DLT_LINUX_SLL, the link-
2908 	 * layer header for which is 16 bytes?)
2909 	 *
2910 	 * XXX - should we use TPACKET_ALIGN(SLL2_HDR_LEN - 16)?
2911 	 */
2912 	if (handlep->cooked)
2913 		tp_reserve += SLL2_HDR_LEN - 16;
2914 
2915 	/*
2916 	 * Try to request that amount of reserve space.
2917 	 * This must be done before creating the ring buffer.
2918 	 */
2919 	len = sizeof(tp_reserve);
2920 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RESERVE,
2921 	    &tp_reserve, len) < 0) {
2922 		pcap_fmt_errmsg_for_errno(handle->errbuf,
2923 		    PCAP_ERRBUF_SIZE, errno,
2924 		    "setsockopt (PACKET_RESERVE)");
2925 		*status = PCAP_ERROR;
2926 		return -1;
2927 	}
2928 
2929 	switch (handlep->tp_version) {
2930 
2931 	case TPACKET_V2:
2932 		/* Note that with large snapshot length (say 256K, which is
2933 		 * the default for recent versions of tcpdump, Wireshark,
2934 		 * TShark, dumpcap or 64K, the value that "-s 0" has given for
2935 		 * a long time with tcpdump), if we use the snapshot
2936 		 * length to calculate the frame length, only a few frames
2937 		 * will be available in the ring even with pretty
2938 		 * large ring size (and a lot of memory will be unused).
2939 		 *
2940 		 * Ideally, we should choose a frame length based on the
2941 		 * minimum of the specified snapshot length and the maximum
2942 		 * packet size.  That's not as easy as it sounds; consider,
2943 		 * for example, an 802.11 interface in monitor mode, where
2944 		 * the frame would include a radiotap header, where the
2945 		 * maximum radiotap header length is device-dependent.
2946 		 *
2947 		 * So, for now, we just do this for Ethernet devices, where
2948 		 * there's no metadata header, and the link-layer header is
2949 		 * fixed length.  We can get the maximum packet size by
2950 		 * adding 18, the Ethernet header length plus the CRC length
2951 		 * (just in case we happen to get the CRC in the packet), to
2952 		 * the MTU of the interface; we fetch the MTU in the hopes
2953 		 * that it reflects support for jumbo frames.  (Even if the
2954 		 * interface is just being used for passive snooping, the
2955 		 * driver might set the size of buffers in the receive ring
2956 		 * based on the MTU, so that the MTU limits the maximum size
2957 		 * of packets that we can receive.)
2958 		 *
2959 		 * If segmentation/fragmentation or receive offload are
2960 		 * enabled, we can get reassembled/aggregated packets larger
2961 		 * than MTU, but bounded to 65535 plus the Ethernet overhead,
2962 		 * due to kernel and protocol constraints */
2963 		frame_size = handle->snapshot;
2964 		if (handle->linktype == DLT_EN10MB) {
2965 			unsigned int max_frame_len;
2966 			int mtu;
2967 			int offload;
2968 
2969 			mtu = iface_get_mtu(handle->fd, handle->opt.device,
2970 			    handle->errbuf);
2971 			if (mtu == -1) {
2972 				*status = PCAP_ERROR;
2973 				return -1;
2974 			}
2975 			offload = iface_get_offload(handle);
2976 			if (offload == -1) {
2977 				*status = PCAP_ERROR;
2978 				return -1;
2979 			}
2980 			if (offload)
2981 				max_frame_len = MAX(mtu, 65535);
2982 			else
2983 				max_frame_len = mtu;
2984 			max_frame_len += 18;
2985 
2986 			if (frame_size > max_frame_len)
2987 				frame_size = max_frame_len;
2988 		}
2989 
2990 		/* NOTE: calculus matching those in tpacket_rcv()
2991 		 * in linux-2.6/net/packet/af_packet.c
2992 		 */
2993 		len = sizeof(sk_type);
2994 		if (getsockopt(handle->fd, SOL_SOCKET, SO_TYPE, &sk_type,
2995 		    &len) < 0) {
2996 			pcap_fmt_errmsg_for_errno(handle->errbuf,
2997 			    PCAP_ERRBUF_SIZE, errno, "getsockopt (SO_TYPE)");
2998 			*status = PCAP_ERROR;
2999 			return -1;
3000 		}
3001 		maclen = (sk_type == SOCK_DGRAM) ? 0 : MAX_LINKHEADER_SIZE;
3002 			/* XXX: in the kernel maclen is calculated from
3003 			 * LL_ALLOCATED_SPACE(dev) and vnet_hdr.hdr_len
3004 			 * in:  packet_snd()           in linux-2.6/net/packet/af_packet.c
3005 			 * then packet_alloc_skb()     in linux-2.6/net/packet/af_packet.c
3006 			 * then sock_alloc_send_pskb() in linux-2.6/net/core/sock.c
3007 			 * but I see no way to get those sizes in userspace,
3008 			 * like for instance with an ifreq ioctl();
3009 			 * the best thing I've found so far is MAX_HEADER in
3010 			 * the kernel part of linux-2.6/include/linux/netdevice.h
3011 			 * which goes up to 128+48=176; since pcap-linux.c
3012 			 * defines a MAX_LINKHEADER_SIZE of 256 which is
3013 			 * greater than that, let's use it.. maybe is it even
3014 			 * large enough to directly replace macoff..
3015 			 */
3016 		tp_hdrlen = TPACKET_ALIGN(handlep->tp_hdrlen) + sizeof(struct sockaddr_ll) ;
3017 		netoff = TPACKET_ALIGN(tp_hdrlen + (maclen < 16 ? 16 : maclen)) + tp_reserve;
3018 			/* NOTE: AFAICS tp_reserve may break the TPACKET_ALIGN
3019 			 * of netoff, which contradicts
3020 			 * linux-2.6/Documentation/networking/packet_mmap.txt
3021 			 * documenting that:
3022 			 * "- Gap, chosen so that packet data (Start+tp_net)
3023 			 * aligns to TPACKET_ALIGNMENT=16"
3024 			 */
3025 			/* NOTE: in linux-2.6/include/linux/skbuff.h:
3026 			 * "CPUs often take a performance hit
3027 			 *  when accessing unaligned memory locations"
3028 			 */
3029 		macoff = netoff - maclen;
3030 		req.tp_frame_size = TPACKET_ALIGN(macoff + frame_size);
3031 		/*
3032 		 * Round the buffer size up to a multiple of the
3033 		 * frame size (rather than rounding down, which
3034 		 * would give a buffer smaller than our caller asked
3035 		 * for, and possibly give zero frames if the requested
3036 		 * buffer size is too small for one frame).
3037 		 */
3038 		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
3039 		break;
3040 
3041 #ifdef HAVE_TPACKET3
3042 	case TPACKET_V3:
3043 		/* The "frames" for this are actually buffers that
3044 		 * contain multiple variable-sized frames.
3045 		 *
3046 		 * We pick a "frame" size of MAXIMUM_SNAPLEN to leave
3047 		 * enough room for at least one reasonably-sized packet
3048 		 * in the "frame". */
3049 		req.tp_frame_size = MAXIMUM_SNAPLEN;
3050 		/*
3051 		 * Round the buffer size up to a multiple of the
3052 		 * "frame" size (rather than rounding down, which
3053 		 * would give a buffer smaller than our caller asked
3054 		 * for, and possibly give zero "frames" if the requested
3055 		 * buffer size is too small for one "frame").
3056 		 */
3057 		req.tp_frame_nr = (handle->opt.buffer_size + req.tp_frame_size - 1)/req.tp_frame_size;
3058 		break;
3059 #endif
3060 	default:
3061 		snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3062 		    "Internal error: unknown TPACKET_ value %u",
3063 		    handlep->tp_version);
3064 		*status = PCAP_ERROR;
3065 		return -1;
3066 	}
3067 
3068 	/* compute the minimum block size that will handle this frame.
3069 	 * The block has to be page size aligned.
3070 	 * The max block size allowed by the kernel is arch-dependent and
3071 	 * it's not explicitly checked here. */
3072 	req.tp_block_size = getpagesize();
3073 	while (req.tp_block_size < req.tp_frame_size)
3074 		req.tp_block_size <<= 1;
3075 
3076 	frames_per_block = req.tp_block_size/req.tp_frame_size;
3077 
3078 	/*
3079 	 * PACKET_TIMESTAMP was added after linux/net_tstamp.h was,
3080 	 * so we check for PACKET_TIMESTAMP.  We check for
3081 	 * linux/net_tstamp.h just in case a system somehow has
3082 	 * PACKET_TIMESTAMP but not linux/net_tstamp.h; that might
3083 	 * be unnecessary.
3084 	 *
3085 	 * SIOCSHWTSTAMP was introduced in the patch that introduced
3086 	 * linux/net_tstamp.h, so we don't bother checking whether
3087 	 * SIOCSHWTSTAMP is defined (if your Linux system has
3088 	 * linux/net_tstamp.h but doesn't define SIOCSHWTSTAMP, your
3089 	 * Linux system is badly broken).
3090 	 */
3091 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
3092 	/*
3093 	 * If we were told to do so, ask the kernel and the driver
3094 	 * to use hardware timestamps.
3095 	 *
3096 	 * Hardware timestamps are only supported with mmapped
3097 	 * captures.
3098 	 */
3099 	if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER ||
3100 	    handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER_UNSYNCED) {
3101 		struct hwtstamp_config hwconfig;
3102 		struct ifreq ifr;
3103 		int timesource;
3104 
3105 		/*
3106 		 * Ask for hardware time stamps on all packets,
3107 		 * including transmitted packets.
3108 		 */
3109 		memset(&hwconfig, 0, sizeof(hwconfig));
3110 		hwconfig.tx_type = HWTSTAMP_TX_ON;
3111 		hwconfig.rx_filter = HWTSTAMP_FILTER_ALL;
3112 
3113 		memset(&ifr, 0, sizeof(ifr));
3114 		pcap_strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
3115 		ifr.ifr_data = (void *)&hwconfig;
3116 
3117 		/*
3118 		 * This may require CAP_NET_ADMIN.
3119 		 */
3120 		if (ioctl(handle->fd, SIOCSHWTSTAMP, &ifr) < 0) {
3121 			switch (errno) {
3122 
3123 			case EPERM:
3124 				/*
3125 				 * Treat this as an error, as the
3126 				 * user should try to run this
3127 				 * with the appropriate privileges -
3128 				 * and, if they can't, shouldn't
3129 				 * try requesting hardware time stamps.
3130 				 */
3131 				*status = PCAP_ERROR_PERM_DENIED;
3132 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3133 				    "Attempt to set hardware timestamp failed - CAP_NET_ADMIN may be required");
3134 				return -1;
3135 
3136 			case EOPNOTSUPP:
3137 			case ERANGE:
3138 				/*
3139 				 * Treat this as a warning, as the
3140 				 * only way to fix the warning is to
3141 				 * get an adapter that supports hardware
3142 				 * time stamps for *all* packets.
3143 				 * (ERANGE means "we support hardware
3144 				 * time stamps, but for packets matching
3145 				 * that particular filter", so it means
3146 				 * "we don't support hardware time stamps
3147 				 * for all incoming packets" here.)
3148 				 *
3149 				 * We'll just fall back on the standard
3150 				 * host time stamps.
3151 				 */
3152 				*status = PCAP_WARNING_TSTAMP_TYPE_NOTSUP;
3153 				break;
3154 
3155 			default:
3156 				pcap_fmt_errmsg_for_errno(handle->errbuf,
3157 				    PCAP_ERRBUF_SIZE, errno,
3158 				    "SIOCSHWTSTAMP failed");
3159 				*status = PCAP_ERROR;
3160 				return -1;
3161 			}
3162 		} else {
3163 			/*
3164 			 * Well, that worked.  Now specify the type of
3165 			 * hardware time stamp we want for this
3166 			 * socket.
3167 			 */
3168 			if (handle->opt.tstamp_type == PCAP_TSTAMP_ADAPTER) {
3169 				/*
3170 				 * Hardware timestamp, synchronized
3171 				 * with the system clock.
3172 				 */
3173 				timesource = SOF_TIMESTAMPING_SYS_HARDWARE;
3174 			} else {
3175 				/*
3176 				 * PCAP_TSTAMP_ADAPTER_UNSYNCED - hardware
3177 				 * timestamp, not synchronized with the
3178 				 * system clock.
3179 				 */
3180 				timesource = SOF_TIMESTAMPING_RAW_HARDWARE;
3181 			}
3182 			if (setsockopt(handle->fd, SOL_PACKET, PACKET_TIMESTAMP,
3183 				(void *)&timesource, sizeof(timesource))) {
3184 				pcap_fmt_errmsg_for_errno(handle->errbuf,
3185 				    PCAP_ERRBUF_SIZE, errno,
3186 				    "can't set PACKET_TIMESTAMP");
3187 				*status = PCAP_ERROR;
3188 				return -1;
3189 			}
3190 		}
3191 	}
3192 #endif /* HAVE_LINUX_NET_TSTAMP_H && PACKET_TIMESTAMP */
3193 
3194 	/* ask the kernel to create the ring */
3195 retry:
3196 	req.tp_block_nr = req.tp_frame_nr / frames_per_block;
3197 
3198 	/* req.tp_frame_nr is requested to match frames_per_block*req.tp_block_nr */
3199 	req.tp_frame_nr = req.tp_block_nr * frames_per_block;
3200 
3201 #ifdef HAVE_TPACKET3
3202 	/* timeout value to retire block - use the configured buffering timeout, or default if <0. */
3203 	if (handlep->timeout > 0) {
3204 		/* Use the user specified timeout as the block timeout */
3205 		req.tp_retire_blk_tov = handlep->timeout;
3206 	} else if (handlep->timeout == 0) {
3207 		/*
3208 		 * In pcap, this means "infinite timeout"; TPACKET_V3
3209 		 * doesn't support that, so just set it to UINT_MAX
3210 		 * milliseconds.  In the TPACKET_V3 loop, if the
3211 		 * timeout is 0, and we haven't yet seen any packets,
3212 		 * and we block and still don't have any packets, we
3213 		 * keep blocking until we do.
3214 		 */
3215 		req.tp_retire_blk_tov = UINT_MAX;
3216 	} else {
3217 		/*
3218 		 * XXX - this is not valid; use 0, meaning "have the
3219 		 * kernel pick a default", for now.
3220 		 */
3221 		req.tp_retire_blk_tov = 0;
3222 	}
3223 	/* private data not used */
3224 	req.tp_sizeof_priv = 0;
3225 	/* Rx ring - feature request bits - none (rxhash will not be filled) */
3226 	req.tp_feature_req_word = 0;
3227 #endif
3228 
3229 	if (setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3230 					(void *) &req, sizeof(req))) {
3231 		if ((errno == ENOMEM) && (req.tp_block_nr > 1)) {
3232 			/*
3233 			 * Memory failure; try to reduce the requested ring
3234 			 * size.
3235 			 *
3236 			 * We used to reduce this by half -- do 5% instead.
3237 			 * That may result in more iterations and a longer
3238 			 * startup, but the user will be much happier with
3239 			 * the resulting buffer size.
3240 			 */
3241 			if (req.tp_frame_nr < 20)
3242 				req.tp_frame_nr -= 1;
3243 			else
3244 				req.tp_frame_nr -= req.tp_frame_nr/20;
3245 			goto retry;
3246 		}
3247 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3248 		    errno, "can't create rx ring on packet socket");
3249 		*status = PCAP_ERROR;
3250 		return -1;
3251 	}
3252 
3253 	/* memory map the rx ring */
3254 	handlep->mmapbuflen = req.tp_block_nr * req.tp_block_size;
3255 	handlep->mmapbuf = mmap(0, handlep->mmapbuflen,
3256 	    PROT_READ|PROT_WRITE, MAP_SHARED, handle->fd, 0);
3257 	if (handlep->mmapbuf == MAP_FAILED) {
3258 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3259 		    errno, "can't mmap rx ring");
3260 
3261 		/* clear the allocated ring on error*/
3262 		destroy_ring(handle);
3263 		*status = PCAP_ERROR;
3264 		return -1;
3265 	}
3266 
3267 	/* allocate a ring for each frame header pointer*/
3268 	handle->cc = req.tp_frame_nr;
3269 	handle->buffer = malloc(handle->cc * sizeof(union thdr *));
3270 	if (!handle->buffer) {
3271 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
3272 		    errno, "can't allocate ring of frame headers");
3273 
3274 		destroy_ring(handle);
3275 		*status = PCAP_ERROR;
3276 		return -1;
3277 	}
3278 
3279 	/* fill the header ring with proper frame ptr*/
3280 	handle->offset = 0;
3281 	for (i=0; i<req.tp_block_nr; ++i) {
3282 		u_char *base = &handlep->mmapbuf[i*req.tp_block_size];
3283 		for (j=0; j<frames_per_block; ++j, ++handle->offset) {
3284 			RING_GET_CURRENT_FRAME(handle) = base;
3285 			base += req.tp_frame_size;
3286 		}
3287 	}
3288 
3289 	handle->bufsize = req.tp_frame_size;
3290 	handle->offset = 0;
3291 	return 1;
3292 }
3293 
3294 /* free all ring related resources*/
3295 static void
3296 destroy_ring(pcap_t *handle)
3297 {
3298 	struct pcap_linux *handlep = handle->priv;
3299 
3300 	/*
3301 	 * Tell the kernel to destroy the ring.
3302 	 * We don't check for setsockopt failure, as 1) we can't recover
3303 	 * from an error and 2) we might not yet have set it up in the
3304 	 * first place.
3305 	 */
3306 	struct tpacket_req req;
3307 	memset(&req, 0, sizeof(req));
3308 	(void)setsockopt(handle->fd, SOL_PACKET, PACKET_RX_RING,
3309 				(void *) &req, sizeof(req));
3310 
3311 	/* if ring is mapped, unmap it*/
3312 	if (handlep->mmapbuf) {
3313 		/* do not test for mmap failure, as we can't recover from any error */
3314 		(void)munmap(handlep->mmapbuf, handlep->mmapbuflen);
3315 		handlep->mmapbuf = NULL;
3316 	}
3317 }
3318 
3319 /*
3320  * Special one-shot callback, used for pcap_next() and pcap_next_ex(),
3321  * for Linux mmapped capture.
3322  *
3323  * The problem is that pcap_next() and pcap_next_ex() expect the packet
3324  * data handed to the callback to be valid after the callback returns,
3325  * but pcap_read_linux_mmap() has to release that packet as soon as
3326  * the callback returns (otherwise, the kernel thinks there's still
3327  * at least one unprocessed packet available in the ring, so a select()
3328  * will immediately return indicating that there's data to process), so,
3329  * in the callback, we have to make a copy of the packet.
3330  *
3331  * Yes, this means that, if the capture is using the ring buffer, using
3332  * pcap_next() or pcap_next_ex() requires more copies than using
3333  * pcap_loop() or pcap_dispatch().  If that bothers you, don't use
3334  * pcap_next() or pcap_next_ex().
3335  */
3336 static void
3337 pcap_oneshot_linux(u_char *user, const struct pcap_pkthdr *h,
3338     const u_char *bytes)
3339 {
3340 	struct oneshot_userdata *sp = (struct oneshot_userdata *)user;
3341 	pcap_t *handle = sp->pd;
3342 	struct pcap_linux *handlep = handle->priv;
3343 
3344 	*sp->hdr = *h;
3345 	memcpy(handlep->oneshot_buffer, bytes, h->caplen);
3346 	*sp->pkt = handlep->oneshot_buffer;
3347 }
3348 
3349 static int
3350 pcap_getnonblock_linux(pcap_t *handle)
3351 {
3352 	struct pcap_linux *handlep = handle->priv;
3353 
3354 	/* use negative value of timeout to indicate non blocking ops */
3355 	return (handlep->timeout<0);
3356 }
3357 
3358 static int
3359 pcap_setnonblock_linux(pcap_t *handle, int nonblock)
3360 {
3361 	struct pcap_linux *handlep = handle->priv;
3362 
3363 	/*
3364 	 * Set the file descriptor to non-blocking mode, as we use
3365 	 * it for sending packets.
3366 	 */
3367 	if (pcap_setnonblock_fd(handle, nonblock) == -1)
3368 		return -1;
3369 
3370 	/*
3371 	 * Map each value to their corresponding negation to
3372 	 * preserve the timeout value provided with pcap_set_timeout.
3373 	 */
3374 	if (nonblock) {
3375 		if (handlep->timeout >= 0) {
3376 			/*
3377 			 * Indicate that we're switching to
3378 			 * non-blocking mode.
3379 			 */
3380 			handlep->timeout = ~handlep->timeout;
3381 		}
3382 		if (handlep->poll_breakloop_fd != -1) {
3383 			/* Close the eventfd; we do not need it in nonblock mode. */
3384 			close(handlep->poll_breakloop_fd);
3385 			handlep->poll_breakloop_fd = -1;
3386 		}
3387 	} else {
3388 		if (handlep->poll_breakloop_fd == -1) {
3389 			/* If we did not have an eventfd, open one now that we are blocking. */
3390 			if ( ( handlep->poll_breakloop_fd = eventfd(0, EFD_NONBLOCK) ) == -1 ) {
3391 				int save_errno = errno;
3392 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3393 						"Could not open eventfd: %s",
3394 						strerror(errno));
3395 				errno = save_errno;
3396 				return -1;
3397 			}
3398 		}
3399 		if (handlep->timeout < 0) {
3400 			handlep->timeout = ~handlep->timeout;
3401 		}
3402 	}
3403 	/* Update the timeout to use in poll(). */
3404 	set_poll_timeout(handlep);
3405 	return 0;
3406 }
3407 
3408 /*
3409  * Get the status field of the ring buffer frame at a specified offset.
3410  */
3411 static inline u_int
3412 pcap_get_ring_frame_status(pcap_t *handle, int offset)
3413 {
3414 	struct pcap_linux *handlep = handle->priv;
3415 	union thdr h;
3416 
3417 	h.raw = RING_GET_FRAME_AT(handle, offset);
3418 	switch (handlep->tp_version) {
3419 	case TPACKET_V2:
3420 		return __atomic_load_n(&h.h2->tp_status, __ATOMIC_ACQUIRE);
3421 		break;
3422 #ifdef HAVE_TPACKET3
3423 	case TPACKET_V3:
3424 		return __atomic_load_n(&h.h3->hdr.bh1.block_status, __ATOMIC_ACQUIRE);
3425 		break;
3426 #endif
3427 	}
3428 	/* This should not happen. */
3429 	return 0;
3430 }
3431 
3432 /*
3433  * Block waiting for frames to be available.
3434  */
3435 static int pcap_wait_for_frames_mmap(pcap_t *handle)
3436 {
3437 	struct pcap_linux *handlep = handle->priv;
3438 	int timeout;
3439 	struct ifreq ifr;
3440 	int ret;
3441 	struct pollfd pollinfo[2];
3442 	int numpollinfo;
3443 	pollinfo[0].fd = handle->fd;
3444 	pollinfo[0].events = POLLIN;
3445 	if ( handlep->poll_breakloop_fd == -1 ) {
3446 		numpollinfo = 1;
3447 		pollinfo[1].revents = 0;
3448 		/*
3449 		 * We set pollinfo[1].revents to zero, even though
3450 		 * numpollinfo = 1 meaning that poll() doesn't see
3451 		 * pollinfo[1], so that we do not have to add a
3452 		 * conditional of numpollinfo > 1 below when we
3453 		 * test pollinfo[1].revents.
3454 		 */
3455 	} else {
3456 		pollinfo[1].fd = handlep->poll_breakloop_fd;
3457 		pollinfo[1].events = POLLIN;
3458 		numpollinfo = 2;
3459 	}
3460 
3461 	/*
3462 	 * Keep polling until we either get some packets to read, see
3463 	 * that we got told to break out of the loop, get a fatal error,
3464 	 * or discover that the device went away.
3465 	 *
3466 	 * In non-blocking mode, we must still do one poll() to catch
3467 	 * any pending error indications, but the poll() has a timeout
3468 	 * of 0, so that it doesn't block, and we quit after that one
3469 	 * poll().
3470 	 *
3471 	 * If we've seen an ENETDOWN, it might be the first indication
3472 	 * that the device went away, or it might just be that it was
3473 	 * configured down.  Unfortunately, there's no guarantee that
3474 	 * the device has actually been removed as an interface, because:
3475 	 *
3476 	 * 1) if, as appears to be the case at least some of the time,
3477 	 * the PF_PACKET socket code first gets a NETDEV_DOWN indication
3478 	 * for the device and then gets a NETDEV_UNREGISTER indication
3479 	 * for it, the first indication will cause a wakeup with ENETDOWN
3480 	 * but won't set the packet socket's field for the interface index
3481 	 * to -1, and the second indication won't cause a wakeup (because
3482 	 * the first indication also caused the protocol hook to be
3483 	 * unregistered) but will set the packet socket's field for the
3484 	 * interface index to -1;
3485 	 *
3486 	 * 2) even if just a NETDEV_UNREGISTER indication is registered,
3487 	 * the packet socket's field for the interface index only gets
3488 	 * set to -1 after the wakeup, so there's a small but non-zero
3489 	 * risk that a thread blocked waiting for the wakeup will get
3490 	 * to the "fetch the socket name" code before the interface index
3491 	 * gets set to -1, so it'll get the old interface index.
3492 	 *
3493 	 * Therefore, if we got an ENETDOWN and haven't seen a packet
3494 	 * since then, we assume that we might be waiting for the interface
3495 	 * to disappear, and poll with a timeout to try again in a short
3496 	 * period of time.  If we *do* see a packet, the interface has
3497 	 * come back up again, and is *definitely* still there, so we
3498 	 * don't need to poll.
3499 	 */
3500 	for (;;) {
3501 	 	/*
3502 		 * Yes, we do this even in non-blocking mode, as it's
3503 		 * the only way to get error indications from a
3504 		 * tpacket socket.
3505 		 *
3506 		 * The timeout is 0 in non-blocking mode, so poll()
3507 		 * returns immediately.
3508 		 */
3509 		timeout = handlep->poll_timeout;
3510 
3511 		/*
3512 		 * If we got an ENETDOWN and haven't gotten an indication
3513 		 * that the device has gone away or that the device is up,
3514 		 * we don't yet know for certain whether the device has
3515 		 * gone away or not, do a poll() with a 1-millisecond timeout,
3516 		 * as we have to poll indefinitely for "device went away"
3517 		 * indications until we either get one or see that the
3518 		 * device is up.
3519 		 */
3520 		if (handlep->netdown) {
3521 			if (timeout != 0)
3522 				timeout = 1;
3523 		}
3524 		ret = poll(pollinfo, numpollinfo, timeout);
3525 		if (ret < 0) {
3526 			/*
3527 			 * Error.  If it's not EINTR, report it.
3528 			 */
3529 			if (errno != EINTR) {
3530 				pcap_fmt_errmsg_for_errno(handle->errbuf,
3531 				    PCAP_ERRBUF_SIZE, errno,
3532 				    "can't poll on packet socket");
3533 				return PCAP_ERROR;
3534 			}
3535 
3536 			/*
3537 			 * It's EINTR; if we were told to break out of
3538 			 * the loop, do so.
3539 			 */
3540 			if (handle->break_loop) {
3541 				handle->break_loop = 0;
3542 				return PCAP_ERROR_BREAK;
3543 			}
3544 		} else if (ret > 0) {
3545 			/*
3546 			 * OK, some descriptor is ready.
3547 			 * Check the socket descriptor first.
3548 			 *
3549 			 * As I read the Linux man page, pollinfo[0].revents
3550 			 * will either be POLLIN, POLLERR, POLLHUP, or POLLNVAL.
3551 			 */
3552 			if (pollinfo[0].revents == POLLIN) {
3553 				/*
3554 				 * OK, we may have packets to
3555 				 * read.
3556 				 */
3557 				break;
3558 			}
3559 			if (pollinfo[0].revents != 0) {
3560 				/*
3561 				 * There's some indication other than
3562 				 * "you can read on this descriptor" on
3563 				 * the descriptor.
3564 				 */
3565 				if (pollinfo[0].revents & POLLNVAL) {
3566 					snprintf(handle->errbuf,
3567 					    PCAP_ERRBUF_SIZE,
3568 					    "Invalid polling request on packet socket");
3569 					return PCAP_ERROR;
3570 				}
3571 				if (pollinfo[0].revents & (POLLHUP | POLLRDHUP)) {
3572 					snprintf(handle->errbuf,
3573 					    PCAP_ERRBUF_SIZE,
3574 					    "Hangup on packet socket");
3575 					return PCAP_ERROR;
3576 				}
3577 				if (pollinfo[0].revents & POLLERR) {
3578 					/*
3579 					 * Get the error.
3580 					 */
3581 					int err;
3582 					socklen_t errlen;
3583 
3584 					errlen = sizeof(err);
3585 					if (getsockopt(handle->fd, SOL_SOCKET,
3586 					    SO_ERROR, &err, &errlen) == -1) {
3587 						/*
3588 						 * The call *itself* returned
3589 						 * an error; make *that*
3590 						 * the error.
3591 						 */
3592 						err = errno;
3593 					}
3594 
3595 					/*
3596 					 * OK, we have the error.
3597 					 */
3598 					if (err == ENETDOWN) {
3599 						/*
3600 						 * The device on which we're
3601 						 * capturing went away or the
3602 						 * interface was taken down.
3603 						 *
3604 						 * We don't know for certain
3605 						 * which happened, and the
3606 						 * next poll() may indicate
3607 						 * that there are packets
3608 						 * to be read, so just set
3609 						 * a flag to get us to do
3610 						 * checks later, and set
3611 						 * the required select
3612 						 * timeout to 1 millisecond
3613 						 * so that event loops that
3614 						 * check our socket descriptor
3615 						 * also time out so that
3616 						 * they can call us and we
3617 						 * can do the checks.
3618 						 */
3619 						handlep->netdown = 1;
3620 						handle->required_select_timeout = &netdown_timeout;
3621 					} else if (err == 0) {
3622 						/*
3623 						 * This shouldn't happen, so
3624 						 * report a special indication
3625 						 * that it did.
3626 						 */
3627 						snprintf(handle->errbuf,
3628 						    PCAP_ERRBUF_SIZE,
3629 						    "Error condition on packet socket: Reported error was 0");
3630 						return PCAP_ERROR;
3631 					} else {
3632 						pcap_fmt_errmsg_for_errno(handle->errbuf,
3633 						    PCAP_ERRBUF_SIZE,
3634 						    err,
3635 						    "Error condition on packet socket");
3636 						return PCAP_ERROR;
3637 					}
3638 				}
3639 			}
3640 			/*
3641 			 * Now check the event device.
3642 			 */
3643 			if (pollinfo[1].revents & POLLIN) {
3644 				ssize_t nread;
3645 				uint64_t value;
3646 
3647 				/*
3648 				 * This should never fail, but, just
3649 				 * in case....
3650 				 */
3651 				nread = read(handlep->poll_breakloop_fd, &value,
3652 				    sizeof(value));
3653 				if (nread == -1) {
3654 					pcap_fmt_errmsg_for_errno(handle->errbuf,
3655 					    PCAP_ERRBUF_SIZE,
3656 					    errno,
3657 					    "Error reading from event FD");
3658 					return PCAP_ERROR;
3659 				}
3660 
3661 				/*
3662 				 * According to the Linux read(2) man
3663 				 * page, read() will transfer at most
3664 				 * 2^31-1 bytes, so the return value is
3665 				 * either -1 or a value between 0
3666 				 * and 2^31-1, so it's non-negative.
3667 				 *
3668 				 * Cast it to size_t to squelch
3669 				 * warnings from the compiler; add this
3670 				 * comment to squelch warnings from
3671 				 * humans reading the code. :-)
3672 				 *
3673 				 * Don't treat an EOF as an error, but
3674 				 * *do* treat a short read as an error;
3675 				 * that "shouldn't happen", but....
3676 				 */
3677 				if (nread != 0 &&
3678 				    (size_t)nread < sizeof(value)) {
3679 					snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3680 					    "Short read from event FD: expected %zu, got %zd",
3681 					    sizeof(value), nread);
3682 					return PCAP_ERROR;
3683 				}
3684 
3685 				/*
3686 				 * This event gets signaled by a
3687 				 * pcap_breakloop() call; if we were told
3688 				 * to break out of the loop, do so.
3689 				 */
3690 				if (handle->break_loop) {
3691 					handle->break_loop = 0;
3692 					return PCAP_ERROR_BREAK;
3693 				}
3694 			}
3695 		}
3696 
3697 		/*
3698 		 * Either:
3699 		 *
3700 		 *   1) we got neither an error from poll() nor any
3701 		 *      readable descriptors, in which case there
3702 		 *      are no packets waiting to read
3703 		 *
3704 		 * or
3705 		 *
3706 		 *   2) We got readable descriptors but the PF_PACKET
3707 		 *      socket wasn't one of them, in which case there
3708 		 *      are no packets waiting to read
3709 		 *
3710 		 * so, if we got an ENETDOWN, we've drained whatever
3711 		 * packets were available to read at the point of the
3712 		 * ENETDOWN.
3713 		 *
3714 		 * So, if we got an ENETDOWN and haven't gotten an indication
3715 		 * that the device has gone away or that the device is up,
3716 		 * we don't yet know for certain whether the device has
3717 		 * gone away or not, check whether the device exists and is
3718 		 * up.
3719 		 */
3720 		if (handlep->netdown) {
3721 			if (!device_still_exists(handle)) {
3722 				/*
3723 				 * The device doesn't exist any more;
3724 				 * report that.
3725 				 *
3726 				 * XXX - we should really return an
3727 				 * appropriate error for that, but
3728 				 * pcap_dispatch() etc. aren't documented
3729 				 * as having error returns other than
3730 				 * PCAP_ERROR or PCAP_ERROR_BREAK.
3731 				 */
3732 				snprintf(handle->errbuf,  PCAP_ERRBUF_SIZE,
3733 				    "The interface disappeared");
3734 				return PCAP_ERROR;
3735 			}
3736 
3737 			/*
3738 			 * The device still exists; try to see if it's up.
3739 			 */
3740 			memset(&ifr, 0, sizeof(ifr));
3741 			pcap_strlcpy(ifr.ifr_name, handlep->device,
3742 			    sizeof(ifr.ifr_name));
3743 			if (ioctl(handle->fd, SIOCGIFFLAGS, &ifr) == -1) {
3744 				if (errno == ENXIO || errno == ENODEV) {
3745 					/*
3746 					 * OK, *now* it's gone.
3747 					 *
3748 					 * XXX - see above comment.
3749 					 */
3750 					snprintf(handle->errbuf,
3751 					    PCAP_ERRBUF_SIZE,
3752 					    "The interface disappeared");
3753 					return PCAP_ERROR;
3754 				} else {
3755 					pcap_fmt_errmsg_for_errno(handle->errbuf,
3756 					    PCAP_ERRBUF_SIZE, errno,
3757 					    "%s: Can't get flags",
3758 					    handlep->device);
3759 					return PCAP_ERROR;
3760 				}
3761 			}
3762 			if (ifr.ifr_flags & IFF_UP) {
3763 				/*
3764 				 * It's up, so it definitely still exists.
3765 				 * Cancel the ENETDOWN indication - we
3766 				 * presumably got it due to the interface
3767 				 * going down rather than the device going
3768 				 * away - and revert to "no required select
3769 				 * timeout.
3770 				 */
3771 				handlep->netdown = 0;
3772 				handle->required_select_timeout = NULL;
3773 			}
3774 		}
3775 
3776 		/*
3777 		 * If we're in non-blocking mode, just quit now, rather
3778 		 * than spinning in a loop doing poll()s that immediately
3779 		 * time out if there's no indication on any descriptor.
3780 		 */
3781 		if (handlep->poll_timeout == 0)
3782 			break;
3783 	}
3784 	return 0;
3785 }
3786 
3787 /* handle a single memory mapped packet */
3788 static int pcap_handle_packet_mmap(
3789 		pcap_t *handle,
3790 		pcap_handler callback,
3791 		u_char *user,
3792 		unsigned char *frame,
3793 		unsigned int tp_len,
3794 		unsigned int tp_mac,
3795 		unsigned int tp_snaplen,
3796 		unsigned int tp_sec,
3797 		unsigned int tp_usec,
3798 		int tp_vlan_tci_valid,
3799 		__u16 tp_vlan_tci,
3800 		__u16 tp_vlan_tpid)
3801 {
3802 	struct pcap_linux *handlep = handle->priv;
3803 	unsigned char *bp;
3804 	struct sockaddr_ll *sll;
3805 	struct pcap_pkthdr pcaphdr;
3806 	pcap_can_socketcan_hdr *canhdr;
3807 	unsigned int snaplen = tp_snaplen;
3808 	struct utsname utsname;
3809 
3810 	/* perform sanity check on internal offset. */
3811 	if (tp_mac + tp_snaplen > handle->bufsize) {
3812 		/*
3813 		 * Report some system information as a debugging aid.
3814 		 */
3815 		if (uname(&utsname) != -1) {
3816 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3817 				"corrupted frame on kernel ring mac "
3818 				"offset %u + caplen %u > frame len %d "
3819 				"(kernel %.32s version %s, machine %.16s)",
3820 				tp_mac, tp_snaplen, handle->bufsize,
3821 				utsname.release, utsname.version,
3822 				utsname.machine);
3823 		} else {
3824 			snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3825 				"corrupted frame on kernel ring mac "
3826 				"offset %u + caplen %u > frame len %d",
3827 				tp_mac, tp_snaplen, handle->bufsize);
3828 		}
3829 		return -1;
3830 	}
3831 
3832 	/* run filter on received packet
3833 	 * If the kernel filtering is enabled we need to run the
3834 	 * filter until all the frames present into the ring
3835 	 * at filter creation time are processed.
3836 	 * In this case, blocks_to_filter_in_userland is used
3837 	 * as a counter for the packet we need to filter.
3838 	 * Note: alternatively it could be possible to stop applying
3839 	 * the filter when the ring became empty, but it can possibly
3840 	 * happen a lot later... */
3841 	bp = frame + tp_mac;
3842 
3843 	/* if required build in place the sll header*/
3844 	sll = (void *)(frame + TPACKET_ALIGN(handlep->tp_hdrlen));
3845 	if (handlep->cooked) {
3846 		if (handle->linktype == DLT_LINUX_SLL2) {
3847 			struct sll2_header *hdrp;
3848 
3849 			/*
3850 			 * The kernel should have left us with enough
3851 			 * space for an sll header; back up the packet
3852 			 * data pointer into that space, as that'll be
3853 			 * the beginning of the packet we pass to the
3854 			 * callback.
3855 			 */
3856 			bp -= SLL2_HDR_LEN;
3857 
3858 			/*
3859 			 * Let's make sure that's past the end of
3860 			 * the tpacket header, i.e. >=
3861 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3862 			 * don't step on the header when we construct
3863 			 * the sll header.
3864 			 */
3865 			if (bp < (u_char *)frame +
3866 					   TPACKET_ALIGN(handlep->tp_hdrlen) +
3867 					   sizeof(struct sockaddr_ll)) {
3868 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3869 					"cooked-mode frame doesn't have room for sll header");
3870 				return -1;
3871 			}
3872 
3873 			/*
3874 			 * OK, that worked; construct the sll header.
3875 			 */
3876 			hdrp = (struct sll2_header *)bp;
3877 			hdrp->sll2_protocol = sll->sll_protocol;
3878 			hdrp->sll2_reserved_mbz = 0;
3879 			hdrp->sll2_if_index = htonl(sll->sll_ifindex);
3880 			hdrp->sll2_hatype = htons(sll->sll_hatype);
3881 			hdrp->sll2_pkttype = sll->sll_pkttype;
3882 			hdrp->sll2_halen = sll->sll_halen;
3883 			memcpy(hdrp->sll2_addr, sll->sll_addr, SLL_ADDRLEN);
3884 
3885 			snaplen += sizeof(struct sll2_header);
3886 		} else {
3887 			struct sll_header *hdrp;
3888 
3889 			/*
3890 			 * The kernel should have left us with enough
3891 			 * space for an sll header; back up the packet
3892 			 * data pointer into that space, as that'll be
3893 			 * the beginning of the packet we pass to the
3894 			 * callback.
3895 			 */
3896 			bp -= SLL_HDR_LEN;
3897 
3898 			/*
3899 			 * Let's make sure that's past the end of
3900 			 * the tpacket header, i.e. >=
3901 			 * ((u_char *)thdr + TPACKET_HDRLEN), so we
3902 			 * don't step on the header when we construct
3903 			 * the sll header.
3904 			 */
3905 			if (bp < (u_char *)frame +
3906 					   TPACKET_ALIGN(handlep->tp_hdrlen) +
3907 					   sizeof(struct sockaddr_ll)) {
3908 				snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
3909 					"cooked-mode frame doesn't have room for sll header");
3910 				return -1;
3911 			}
3912 
3913 			/*
3914 			 * OK, that worked; construct the sll header.
3915 			 */
3916 			hdrp = (struct sll_header *)bp;
3917 			hdrp->sll_pkttype = htons(sll->sll_pkttype);
3918 			hdrp->sll_hatype = htons(sll->sll_hatype);
3919 			hdrp->sll_halen = htons(sll->sll_halen);
3920 			memcpy(hdrp->sll_addr, sll->sll_addr, SLL_ADDRLEN);
3921 			hdrp->sll_protocol = sll->sll_protocol;
3922 
3923 			snaplen += sizeof(struct sll_header);
3924 		}
3925 	} else {
3926 		/*
3927 		 * If this is a packet from a CAN device, so that
3928 		 * sll->sll_hatype is ARPHRD_CAN, then, as we're
3929 		 * not capturing in cooked mode, its link-layer
3930 		 * type is DLT_CAN_SOCKETCAN.  Fix up the header
3931 		 * provided by the code below us to match what
3932 		 * DLT_CAN_SOCKETCAN is expected to provide.
3933 		 */
3934 		if (sll->sll_hatype == ARPHRD_CAN) {
3935 			/*
3936 			 * DLT_CAN_SOCKETCAN is specified as having the
3937 			 * CAN ID and flags in network byte order, but
3938 			 * capturing on a CAN device provides it in host
3939 			 * byte order.  Convert it to network byte order.
3940 			 */
3941 			canhdr = (pcap_can_socketcan_hdr *)bp;
3942 			canhdr->can_id = htonl(canhdr->can_id);
3943 
3944 			/*
3945 			 * In addition, set the CANFD_FDF flag if
3946 			 * the protocol is LINUX_SLL_P_CANFD, as
3947 			 * the protocol field itself isn't in
3948 			 * the packet to indicate that it's a
3949 			 * CAN FD packet.
3950 			 */
3951 			uint16_t protocol = ntohs(sll->sll_protocol);
3952 			if (protocol == LINUX_SLL_P_CANFD) {
3953 				canhdr->fd_flags |= CANFD_FDF;
3954 
3955 				/*
3956 				 * Zero out all the unknown bits in
3957 				 * fd_flags and clear the reserved
3958 				 * fields, so that a program reading
3959 				 * this can assume that CANFD_FDF
3960 				 * is set because we set it, not
3961 				 * because some uninitialized crap
3962 				 * was provided in the fd_flags
3963 				 * field.
3964 				 *
3965 				 * (At least some LINKTYPE_CAN_SOCKETCAN
3966 				 * files attached to Wireshark bugs
3967 				 * had uninitialized junk there, so it
3968 				 * does happen.)
3969 				 *
3970 				 * Update this if Linux adds more flag
3971 				 * bits to the fd_flags field or uses
3972 				 * either of the reserved fields for
3973 				 * FD frames.
3974 				 */
3975 				canhdr->fd_flags &= ~(CANFD_FDF|CANFD_ESI|CANFD_BRS);
3976 				canhdr->reserved1 = 0;
3977 				canhdr->reserved2 = 0;
3978 			} else {
3979 				/*
3980 				 * Clear CANFD_FDF if it's set (probably
3981 				 * again meaning that this field is
3982 				 * uninitialized junk).
3983 				 */
3984 				canhdr->fd_flags &= ~CANFD_FDF;
3985 			}
3986 		}
3987 	}
3988 
3989 	if (handlep->filter_in_userland && handle->fcode.bf_insns) {
3990 		struct pcap_bpf_aux_data aux_data;
3991 
3992 		aux_data.vlan_tag_present = tp_vlan_tci_valid;
3993 		aux_data.vlan_tag = tp_vlan_tci & 0x0fff;
3994 
3995 		if (pcap_filter_with_aux_data(handle->fcode.bf_insns,
3996 					      bp,
3997 					      tp_len,
3998 					      snaplen,
3999 					      &aux_data) == 0)
4000 			return 0;
4001 	}
4002 
4003 	if (!linux_check_direction(handle, sll))
4004 		return 0;
4005 
4006 	/* get required packet info from ring header */
4007 	pcaphdr.ts.tv_sec = tp_sec;
4008 	pcaphdr.ts.tv_usec = tp_usec;
4009 	pcaphdr.caplen = tp_snaplen;
4010 	pcaphdr.len = tp_len;
4011 
4012 	/* if required build in place the sll header*/
4013 	if (handlep->cooked) {
4014 		/* update packet len */
4015 		if (handle->linktype == DLT_LINUX_SLL2) {
4016 			pcaphdr.caplen += SLL2_HDR_LEN;
4017 			pcaphdr.len += SLL2_HDR_LEN;
4018 		} else {
4019 			pcaphdr.caplen += SLL_HDR_LEN;
4020 			pcaphdr.len += SLL_HDR_LEN;
4021 		}
4022 	}
4023 
4024 	if (tp_vlan_tci_valid &&
4025 		handlep->vlan_offset != -1 &&
4026 		tp_snaplen >= (unsigned int) handlep->vlan_offset)
4027 	{
4028 		struct vlan_tag *tag;
4029 
4030 		/*
4031 		 * Move everything in the header, except the type field,
4032 		 * down VLAN_TAG_LEN bytes, to allow us to insert the
4033 		 * VLAN tag between that stuff and the type field.
4034 		 */
4035 		bp -= VLAN_TAG_LEN;
4036 		memmove(bp, bp + VLAN_TAG_LEN, handlep->vlan_offset);
4037 
4038 		/*
4039 		 * Now insert the tag.
4040 		 */
4041 		tag = (struct vlan_tag *)(bp + handlep->vlan_offset);
4042 		tag->vlan_tpid = htons(tp_vlan_tpid);
4043 		tag->vlan_tci = htons(tp_vlan_tci);
4044 
4045 		/*
4046 		 * Add the tag to the packet lengths.
4047 		 */
4048 		pcaphdr.caplen += VLAN_TAG_LEN;
4049 		pcaphdr.len += VLAN_TAG_LEN;
4050 	}
4051 
4052 	/*
4053 	 * The only way to tell the kernel to cut off the
4054 	 * packet at a snapshot length is with a filter program;
4055 	 * if there's no filter program, the kernel won't cut
4056 	 * the packet off.
4057 	 *
4058 	 * Trim the snapshot length to be no longer than the
4059 	 * specified snapshot length.
4060 	 *
4061 	 * XXX - an alternative is to put a filter, consisting
4062 	 * of a "ret <snaplen>" instruction, on the socket
4063 	 * in the activate routine, so that the truncation is
4064 	 * done in the kernel even if nobody specified a filter;
4065 	 * that means that less buffer space is consumed in
4066 	 * the memory-mapped buffer.
4067 	 */
4068 	if (pcaphdr.caplen > (bpf_u_int32)handle->snapshot)
4069 		pcaphdr.caplen = handle->snapshot;
4070 
4071 	/* pass the packet to the user */
4072 	callback(user, &pcaphdr, bp);
4073 
4074 	return 1;
4075 }
4076 
4077 static int
4078 pcap_read_linux_mmap_v2(pcap_t *handle, int max_packets, pcap_handler callback,
4079 		u_char *user)
4080 {
4081 	struct pcap_linux *handlep = handle->priv;
4082 	union thdr h;
4083 	int pkts = 0;
4084 	int ret;
4085 
4086 	/* wait for frames availability.*/
4087 	h.raw = RING_GET_CURRENT_FRAME(handle);
4088 	if (!packet_mmap_acquire(h.h2)) {
4089 		/*
4090 		 * The current frame is owned by the kernel; wait for
4091 		 * a frame to be handed to us.
4092 		 */
4093 		ret = pcap_wait_for_frames_mmap(handle);
4094 		if (ret) {
4095 			return ret;
4096 		}
4097 	}
4098 
4099 	/*
4100 	 * This can conceivably process more than INT_MAX packets,
4101 	 * which would overflow the packet count, causing it either
4102 	 * to look like a negative number, and thus cause us to
4103 	 * return a value that looks like an error, or overflow
4104 	 * back into positive territory, and thus cause us to
4105 	 * return a too-low count.
4106 	 *
4107 	 * Therefore, if the packet count is unlimited, we clip
4108 	 * it at INT_MAX; this routine is not expected to
4109 	 * process packets indefinitely, so that's not an issue.
4110 	 */
4111 	if (PACKET_COUNT_IS_UNLIMITED(max_packets))
4112 		max_packets = INT_MAX;
4113 
4114 	while (pkts < max_packets) {
4115 		/*
4116 		 * Get the current ring buffer frame, and break if
4117 		 * it's still owned by the kernel.
4118 		 */
4119 		h.raw = RING_GET_CURRENT_FRAME(handle);
4120 		if (!packet_mmap_acquire(h.h2))
4121 			break;
4122 
4123 		ret = pcap_handle_packet_mmap(
4124 				handle,
4125 				callback,
4126 				user,
4127 				h.raw,
4128 				h.h2->tp_len,
4129 				h.h2->tp_mac,
4130 				h.h2->tp_snaplen,
4131 				h.h2->tp_sec,
4132 				handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? h.h2->tp_nsec : h.h2->tp_nsec / 1000,
4133 				VLAN_VALID(h.h2, h.h2),
4134 				h.h2->tp_vlan_tci,
4135 				VLAN_TPID(h.h2, h.h2));
4136 		if (ret == 1) {
4137 			pkts++;
4138 		} else if (ret < 0) {
4139 			return ret;
4140 		}
4141 
4142 		/*
4143 		 * Hand this block back to the kernel, and, if we're
4144 		 * counting blocks that need to be filtered in userland
4145 		 * after having been filtered by the kernel, count
4146 		 * the one we've just processed.
4147 		 */
4148 		packet_mmap_release(h.h2);
4149 		if (handlep->blocks_to_filter_in_userland > 0) {
4150 			handlep->blocks_to_filter_in_userland--;
4151 			if (handlep->blocks_to_filter_in_userland == 0) {
4152 				/*
4153 				 * No more blocks need to be filtered
4154 				 * in userland.
4155 				 */
4156 				handlep->filter_in_userland = 0;
4157 			}
4158 		}
4159 
4160 		/* next block */
4161 		if (++handle->offset >= handle->cc)
4162 			handle->offset = 0;
4163 
4164 		/* check for break loop condition*/
4165 		if (handle->break_loop) {
4166 			handle->break_loop = 0;
4167 			return PCAP_ERROR_BREAK;
4168 		}
4169 	}
4170 	return pkts;
4171 }
4172 
4173 #ifdef HAVE_TPACKET3
4174 static int
4175 pcap_read_linux_mmap_v3(pcap_t *handle, int max_packets, pcap_handler callback,
4176 		u_char *user)
4177 {
4178 	struct pcap_linux *handlep = handle->priv;
4179 	union thdr h;
4180 	int pkts = 0;
4181 	int ret;
4182 
4183 again:
4184 	if (handlep->current_packet == NULL) {
4185 		/* wait for frames availability.*/
4186 		h.raw = RING_GET_CURRENT_FRAME(handle);
4187 		if (!packet_mmap_v3_acquire(h.h3)) {
4188 			/*
4189 			 * The current frame is owned by the kernel; wait
4190 			 * for a frame to be handed to us.
4191 			 */
4192 			ret = pcap_wait_for_frames_mmap(handle);
4193 			if (ret) {
4194 				return ret;
4195 			}
4196 		}
4197 	}
4198 	h.raw = RING_GET_CURRENT_FRAME(handle);
4199 	if (!packet_mmap_v3_acquire(h.h3)) {
4200 		if (pkts == 0 && handlep->timeout == 0) {
4201 			/* Block until we see a packet. */
4202 			goto again;
4203 		}
4204 		return pkts;
4205 	}
4206 
4207 	/*
4208 	 * This can conceivably process more than INT_MAX packets,
4209 	 * which would overflow the packet count, causing it either
4210 	 * to look like a negative number, and thus cause us to
4211 	 * return a value that looks like an error, or overflow
4212 	 * back into positive territory, and thus cause us to
4213 	 * return a too-low count.
4214 	 *
4215 	 * Therefore, if the packet count is unlimited, we clip
4216 	 * it at INT_MAX; this routine is not expected to
4217 	 * process packets indefinitely, so that's not an issue.
4218 	 */
4219 	if (PACKET_COUNT_IS_UNLIMITED(max_packets))
4220 		max_packets = INT_MAX;
4221 
4222 	while (pkts < max_packets) {
4223 		int packets_to_read;
4224 
4225 		if (handlep->current_packet == NULL) {
4226 			h.raw = RING_GET_CURRENT_FRAME(handle);
4227 			if (!packet_mmap_v3_acquire(h.h3))
4228 				break;
4229 
4230 			handlep->current_packet = h.raw + h.h3->hdr.bh1.offset_to_first_pkt;
4231 			handlep->packets_left = h.h3->hdr.bh1.num_pkts;
4232 		}
4233 		packets_to_read = handlep->packets_left;
4234 
4235 		if (packets_to_read > (max_packets - pkts)) {
4236 			/*
4237 			 * There are more packets in the buffer than
4238 			 * the number of packets we have left to
4239 			 * process to get up to the maximum number
4240 			 * of packets to process.  Only process enough
4241 			 * of them to get us up to that maximum.
4242 			 */
4243 			packets_to_read = max_packets - pkts;
4244 		}
4245 
4246 		while (packets_to_read-- && !handle->break_loop) {
4247 			struct tpacket3_hdr* tp3_hdr = (struct tpacket3_hdr*) handlep->current_packet;
4248 			ret = pcap_handle_packet_mmap(
4249 					handle,
4250 					callback,
4251 					user,
4252 					handlep->current_packet,
4253 					tp3_hdr->tp_len,
4254 					tp3_hdr->tp_mac,
4255 					tp3_hdr->tp_snaplen,
4256 					tp3_hdr->tp_sec,
4257 					handle->opt.tstamp_precision == PCAP_TSTAMP_PRECISION_NANO ? tp3_hdr->tp_nsec : tp3_hdr->tp_nsec / 1000,
4258 					VLAN_VALID(tp3_hdr, &tp3_hdr->hv1),
4259 					tp3_hdr->hv1.tp_vlan_tci,
4260 					VLAN_TPID(tp3_hdr, &tp3_hdr->hv1));
4261 			if (ret == 1) {
4262 				pkts++;
4263 			} else if (ret < 0) {
4264 				handlep->current_packet = NULL;
4265 				return ret;
4266 			}
4267 			handlep->current_packet += tp3_hdr->tp_next_offset;
4268 			handlep->packets_left--;
4269 		}
4270 
4271 		if (handlep->packets_left <= 0) {
4272 			/*
4273 			 * Hand this block back to the kernel, and, if
4274 			 * we're counting blocks that need to be
4275 			 * filtered in userland after having been
4276 			 * filtered by the kernel, count the one we've
4277 			 * just processed.
4278 			 */
4279 			packet_mmap_v3_release(h.h3);
4280 			if (handlep->blocks_to_filter_in_userland > 0) {
4281 				handlep->blocks_to_filter_in_userland--;
4282 				if (handlep->blocks_to_filter_in_userland == 0) {
4283 					/*
4284 					 * No more blocks need to be filtered
4285 					 * in userland.
4286 					 */
4287 					handlep->filter_in_userland = 0;
4288 				}
4289 			}
4290 
4291 			/* next block */
4292 			if (++handle->offset >= handle->cc)
4293 				handle->offset = 0;
4294 
4295 			handlep->current_packet = NULL;
4296 		}
4297 
4298 		/* check for break loop condition*/
4299 		if (handle->break_loop) {
4300 			handle->break_loop = 0;
4301 			return PCAP_ERROR_BREAK;
4302 		}
4303 	}
4304 	if (pkts == 0 && handlep->timeout == 0) {
4305 		/* Block until we see a packet. */
4306 		goto again;
4307 	}
4308 	return pkts;
4309 }
4310 #endif /* HAVE_TPACKET3 */
4311 
4312 /*
4313  *  Attach the given BPF code to the packet capture device.
4314  */
4315 static int
4316 pcap_setfilter_linux(pcap_t *handle, struct bpf_program *filter)
4317 {
4318 	struct pcap_linux *handlep;
4319 	struct sock_fprog	fcode;
4320 	int			can_filter_in_kernel;
4321 	int			err = 0;
4322 	int			n, offset;
4323 
4324 	if (!handle)
4325 		return -1;
4326 	if (!filter) {
4327 	        pcap_strlcpy(handle->errbuf, "setfilter: No filter specified",
4328 			PCAP_ERRBUF_SIZE);
4329 		return -1;
4330 	}
4331 
4332 	handlep = handle->priv;
4333 
4334 	/* Make our private copy of the filter */
4335 
4336 	if (install_bpf_program(handle, filter) < 0)
4337 		/* install_bpf_program() filled in errbuf */
4338 		return -1;
4339 
4340 	/*
4341 	 * Run user level packet filter by default. Will be overridden if
4342 	 * installing a kernel filter succeeds.
4343 	 */
4344 	handlep->filter_in_userland = 1;
4345 
4346 	/* Install kernel level filter if possible */
4347 
4348 #ifdef USHRT_MAX
4349 	if (handle->fcode.bf_len > USHRT_MAX) {
4350 		/*
4351 		 * fcode.len is an unsigned short for current kernel.
4352 		 * I have yet to see BPF-Code with that much
4353 		 * instructions but still it is possible. So for the
4354 		 * sake of correctness I added this check.
4355 		 */
4356 		fprintf(stderr, "Warning: Filter too complex for kernel\n");
4357 		fcode.len = 0;
4358 		fcode.filter = NULL;
4359 		can_filter_in_kernel = 0;
4360 	} else
4361 #endif /* USHRT_MAX */
4362 	{
4363 		/*
4364 		 * Oh joy, the Linux kernel uses struct sock_fprog instead
4365 		 * of struct bpf_program and of course the length field is
4366 		 * of different size. Pointed out by Sebastian
4367 		 *
4368 		 * Oh, and we also need to fix it up so that all "ret"
4369 		 * instructions with non-zero operands have MAXIMUM_SNAPLEN
4370 		 * as the operand if we're not capturing in memory-mapped
4371 		 * mode, and so that, if we're in cooked mode, all memory-
4372 		 * reference instructions use special magic offsets in
4373 		 * references to the link-layer header and assume that the
4374 		 * link-layer payload begins at 0; "fix_program()" will do
4375 		 * that.
4376 		 */
4377 		switch (fix_program(handle, &fcode)) {
4378 
4379 		case -1:
4380 		default:
4381 			/*
4382 			 * Fatal error; just quit.
4383 			 * (The "default" case shouldn't happen; we
4384 			 * return -1 for that reason.)
4385 			 */
4386 			return -1;
4387 
4388 		case 0:
4389 			/*
4390 			 * The program performed checks that we can't make
4391 			 * work in the kernel.
4392 			 */
4393 			can_filter_in_kernel = 0;
4394 			break;
4395 
4396 		case 1:
4397 			/*
4398 			 * We have a filter that'll work in the kernel.
4399 			 */
4400 			can_filter_in_kernel = 1;
4401 			break;
4402 		}
4403 	}
4404 
4405 	/*
4406 	 * NOTE: at this point, we've set both the "len" and "filter"
4407 	 * fields of "fcode".  As of the 2.6.32.4 kernel, at least,
4408 	 * those are the only members of the "sock_fprog" structure,
4409 	 * so we initialize every member of that structure.
4410 	 *
4411 	 * If there is anything in "fcode" that is not initialized,
4412 	 * it is either a field added in a later kernel, or it's
4413 	 * padding.
4414 	 *
4415 	 * If a new field is added, this code needs to be updated
4416 	 * to set it correctly.
4417 	 *
4418 	 * If there are no other fields, then:
4419 	 *
4420 	 *	if the Linux kernel looks at the padding, it's
4421 	 *	buggy;
4422 	 *
4423 	 *	if the Linux kernel doesn't look at the padding,
4424 	 *	then if some tool complains that we're passing
4425 	 *	uninitialized data to the kernel, then the tool
4426 	 *	is buggy and needs to understand that it's just
4427 	 *	padding.
4428 	 */
4429 	if (can_filter_in_kernel) {
4430 		if ((err = set_kernel_filter(handle, &fcode)) == 0)
4431 		{
4432 			/*
4433 			 * Installation succeeded - using kernel filter,
4434 			 * so userland filtering not needed.
4435 			 */
4436 			handlep->filter_in_userland = 0;
4437 		}
4438 		else if (err == -1)	/* Non-fatal error */
4439 		{
4440 			/*
4441 			 * Print a warning if we weren't able to install
4442 			 * the filter for a reason other than "this kernel
4443 			 * isn't configured to support socket filters.
4444 			 */
4445 			if (errno == ENOMEM) {
4446 				/*
4447 				 * Either a kernel memory allocation
4448 				 * failure occurred, or there's too
4449 				 * much "other/option memory" allocated
4450 				 * for this socket.  Suggest that they
4451 				 * increase the "other/option memory"
4452 				 * limit.
4453 				 */
4454 				fprintf(stderr,
4455 				    "Warning: Couldn't allocate kernel memory for filter: try increasing net.core.optmem_max with sysctl\n");
4456 			} else if (errno != ENOPROTOOPT && errno != EOPNOTSUPP) {
4457 				fprintf(stderr,
4458 				    "Warning: Kernel filter failed: %s\n",
4459 					pcap_strerror(errno));
4460 			}
4461 		}
4462 	}
4463 
4464 	/*
4465 	 * If we're not using the kernel filter, get rid of any kernel
4466 	 * filter that might've been there before, e.g. because the
4467 	 * previous filter could work in the kernel, or because some other
4468 	 * code attached a filter to the socket by some means other than
4469 	 * calling "pcap_setfilter()".  Otherwise, the kernel filter may
4470 	 * filter out packets that would pass the new userland filter.
4471 	 */
4472 	if (handlep->filter_in_userland) {
4473 		if (reset_kernel_filter(handle) == -1) {
4474 			pcap_fmt_errmsg_for_errno(handle->errbuf,
4475 			    PCAP_ERRBUF_SIZE, errno,
4476 			    "can't remove kernel filter");
4477 			err = -2;	/* fatal error */
4478 		}
4479 	}
4480 
4481 	/*
4482 	 * Free up the copy of the filter that was made by "fix_program()".
4483 	 */
4484 	if (fcode.filter != NULL)
4485 		free(fcode.filter);
4486 
4487 	if (err == -2)
4488 		/* Fatal error */
4489 		return -1;
4490 
4491 	/*
4492 	 * If we're filtering in userland, there's nothing to do;
4493 	 * the new filter will be used for the next packet.
4494 	 */
4495 	if (handlep->filter_in_userland)
4496 		return 0;
4497 
4498 	/*
4499 	 * We're filtering in the kernel; the packets present in
4500 	 * all blocks currently in the ring were already filtered
4501 	 * by the old filter, and so will need to be filtered in
4502 	 * userland by the new filter.
4503 	 *
4504 	 * Get an upper bound for the number of such blocks; first,
4505 	 * walk the ring backward and count the free blocks.
4506 	 */
4507 	offset = handle->offset;
4508 	if (--offset < 0)
4509 		offset = handle->cc - 1;
4510 	for (n=0; n < handle->cc; ++n) {
4511 		if (--offset < 0)
4512 			offset = handle->cc - 1;
4513 		if (pcap_get_ring_frame_status(handle, offset) != TP_STATUS_KERNEL)
4514 			break;
4515 	}
4516 
4517 	/*
4518 	 * If we found free blocks, decrement the count of free
4519 	 * blocks by 1, just in case we lost a race with another
4520 	 * thread of control that was adding a packet while
4521 	 * we were counting and that had run the filter before
4522 	 * we changed it.
4523 	 *
4524 	 * XXX - could there be more than one block added in
4525 	 * this fashion?
4526 	 *
4527 	 * XXX - is there a way to avoid that race, e.g. somehow
4528 	 * wait for all packets that passed the old filter to
4529 	 * be added to the ring?
4530 	 */
4531 	if (n != 0)
4532 		n--;
4533 
4534 	/*
4535 	 * Set the count of blocks worth of packets to filter
4536 	 * in userland to the total number of blocks in the
4537 	 * ring minus the number of free blocks we found, and
4538 	 * turn on userland filtering.  (The count of blocks
4539 	 * worth of packets to filter in userland is guaranteed
4540 	 * not to be zero - n, above, couldn't be set to a
4541 	 * value > handle->cc, and if it were equal to
4542 	 * handle->cc, it wouldn't be zero, and thus would
4543 	 * be decremented to handle->cc - 1.)
4544 	 */
4545 	handlep->blocks_to_filter_in_userland = handle->cc - n;
4546 	handlep->filter_in_userland = 1;
4547 
4548 	return 0;
4549 }
4550 
4551 /*
4552  *  Return the index of the given device name. Fill ebuf and return
4553  *  -1 on failure.
4554  */
4555 static int
4556 iface_get_id(int fd, const char *device, char *ebuf)
4557 {
4558 	struct ifreq	ifr;
4559 
4560 	memset(&ifr, 0, sizeof(ifr));
4561 	pcap_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4562 
4563 	if (ioctl(fd, SIOCGIFINDEX, &ifr) == -1) {
4564 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4565 		    errno, "SIOCGIFINDEX");
4566 		return -1;
4567 	}
4568 
4569 	return ifr.ifr_ifindex;
4570 }
4571 
4572 /*
4573  *  Bind the socket associated with FD to the given device.
4574  *  Return 0 on success or a PCAP_ERROR_ value on a hard error.
4575  */
4576 static int
4577 iface_bind(int fd, int ifindex, char *ebuf, int protocol)
4578 {
4579 	struct sockaddr_ll	sll;
4580 	int			ret, err;
4581 	socklen_t		errlen = sizeof(err);
4582 
4583 	memset(&sll, 0, sizeof(sll));
4584 	sll.sll_family		= AF_PACKET;
4585 	sll.sll_ifindex		= ifindex < 0 ? 0 : ifindex;
4586 	sll.sll_protocol	= protocol;
4587 
4588 	if (bind(fd, (struct sockaddr *) &sll, sizeof(sll)) == -1) {
4589 		if (errno == ENETDOWN) {
4590 			/*
4591 			 * Return a "network down" indication, so that
4592 			 * the application can report that rather than
4593 			 * saying we had a mysterious failure and
4594 			 * suggest that they report a problem to the
4595 			 * libpcap developers.
4596 			 */
4597 			return PCAP_ERROR_IFACE_NOT_UP;
4598 		}
4599 		if (errno == ENODEV) {
4600 			/*
4601 			 * There's nothing more to say, so clear the
4602 			 * error message.
4603 			 */
4604 			ebuf[0] = '\0';
4605 			ret = PCAP_ERROR_NO_SUCH_DEVICE;
4606 		} else {
4607 			ret = PCAP_ERROR;
4608 			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4609 			    errno, "bind");
4610 		}
4611 		return ret;
4612 	}
4613 
4614 	/* Any pending errors, e.g., network is down? */
4615 
4616 	if (getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &errlen) == -1) {
4617 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4618 		    errno, "getsockopt (SO_ERROR)");
4619 		return PCAP_ERROR;
4620 	}
4621 
4622 	if (err == ENETDOWN) {
4623 		/*
4624 		 * Return a "network down" indication, so that
4625 		 * the application can report that rather than
4626 		 * saying we had a mysterious failure and
4627 		 * suggest that they report a problem to the
4628 		 * libpcap developers.
4629 		 */
4630 		return PCAP_ERROR_IFACE_NOT_UP;
4631 	} else if (err > 0) {
4632 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4633 		    err, "bind");
4634 		return PCAP_ERROR;
4635 	}
4636 
4637 	return 0;
4638 }
4639 
4640 /*
4641  * Try to enter monitor mode.
4642  * If we have libnl, try to create a new monitor-mode device and
4643  * capture on that; otherwise, just say "not supported".
4644  */
4645 #ifdef HAVE_LIBNL
4646 static int
4647 enter_rfmon_mode(pcap_t *handle, int sock_fd, const char *device)
4648 {
4649 	struct pcap_linux *handlep = handle->priv;
4650 	int ret;
4651 	char phydev_path[PATH_MAX+1];
4652 	struct nl80211_state nlstate;
4653 	struct ifreq ifr;
4654 	u_int n;
4655 
4656 	/*
4657 	 * Is this a mac80211 device?
4658 	 */
4659 	ret = get_mac80211_phydev(handle, device, phydev_path, PATH_MAX);
4660 	if (ret < 0)
4661 		return ret;	/* error */
4662 	if (ret == 0)
4663 		return 0;	/* no error, but not mac80211 device */
4664 
4665 	/*
4666 	 * XXX - is this already a monN device?
4667 	 * If so, we're done.
4668 	 */
4669 
4670 	/*
4671 	 * OK, it's apparently a mac80211 device.
4672 	 * Try to find an unused monN device for it.
4673 	 */
4674 	ret = nl80211_init(handle, &nlstate, device);
4675 	if (ret != 0)
4676 		return ret;
4677 	for (n = 0; n < UINT_MAX; n++) {
4678 		/*
4679 		 * Try mon{n}.
4680 		 */
4681 		char mondevice[3+10+1];	/* mon{UINT_MAX}\0 */
4682 
4683 		snprintf(mondevice, sizeof mondevice, "mon%u", n);
4684 		ret = add_mon_if(handle, sock_fd, &nlstate, device, mondevice);
4685 		if (ret == 1) {
4686 			/*
4687 			 * Success.  We don't clean up the libnl state
4688 			 * yet, as we'll be using it later.
4689 			 */
4690 			goto added;
4691 		}
4692 		if (ret < 0) {
4693 			/*
4694 			 * Hard failure.  Just return ret; handle->errbuf
4695 			 * has already been set.
4696 			 */
4697 			nl80211_cleanup(&nlstate);
4698 			return ret;
4699 		}
4700 	}
4701 
4702 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
4703 	    "%s: No free monN interfaces", device);
4704 	nl80211_cleanup(&nlstate);
4705 	return PCAP_ERROR;
4706 
4707 added:
4708 
4709 #if 0
4710 	/*
4711 	 * Sleep for .1 seconds.
4712 	 */
4713 	delay.tv_sec = 0;
4714 	delay.tv_nsec = 500000000;
4715 	nanosleep(&delay, NULL);
4716 #endif
4717 
4718 	/*
4719 	 * If we haven't already done so, arrange to have
4720 	 * "pcap_close_all()" called when we exit.
4721 	 */
4722 	if (!pcap_do_addexit(handle)) {
4723 		/*
4724 		 * "atexit()" failed; don't put the interface
4725 		 * in rfmon mode, just give up.
4726 		 */
4727 		del_mon_if(handle, sock_fd, &nlstate, device,
4728 		    handlep->mondevice);
4729 		nl80211_cleanup(&nlstate);
4730 		return PCAP_ERROR;
4731 	}
4732 
4733 	/*
4734 	 * Now configure the monitor interface up.
4735 	 */
4736 	memset(&ifr, 0, sizeof(ifr));
4737 	pcap_strlcpy(ifr.ifr_name, handlep->mondevice, sizeof(ifr.ifr_name));
4738 	if (ioctl(sock_fd, SIOCGIFFLAGS, &ifr) == -1) {
4739 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4740 		    errno, "%s: Can't get flags for %s", device,
4741 		    handlep->mondevice);
4742 		del_mon_if(handle, sock_fd, &nlstate, device,
4743 		    handlep->mondevice);
4744 		nl80211_cleanup(&nlstate);
4745 		return PCAP_ERROR;
4746 	}
4747 	ifr.ifr_flags |= IFF_UP|IFF_RUNNING;
4748 	if (ioctl(sock_fd, SIOCSIFFLAGS, &ifr) == -1) {
4749 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
4750 		    errno, "%s: Can't set flags for %s", device,
4751 		    handlep->mondevice);
4752 		del_mon_if(handle, sock_fd, &nlstate, device,
4753 		    handlep->mondevice);
4754 		nl80211_cleanup(&nlstate);
4755 		return PCAP_ERROR;
4756 	}
4757 
4758 	/*
4759 	 * Success.  Clean up the libnl state.
4760 	 */
4761 	nl80211_cleanup(&nlstate);
4762 
4763 	/*
4764 	 * Note that we have to delete the monitor device when we close
4765 	 * the handle.
4766 	 */
4767 	handlep->must_do_on_close |= MUST_DELETE_MONIF;
4768 
4769 	/*
4770 	 * Add this to the list of pcaps to close when we exit.
4771 	 */
4772 	pcap_add_to_pcaps_to_close(handle);
4773 
4774 	return 1;
4775 }
4776 #else /* HAVE_LIBNL */
4777 static int
4778 enter_rfmon_mode(pcap_t *handle _U_, int sock_fd _U_, const char *device _U_)
4779 {
4780 	/*
4781 	 * We don't have libnl, so we can't do monitor mode.
4782 	 */
4783 	return 0;
4784 }
4785 #endif /* HAVE_LIBNL */
4786 
4787 #if defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP)
4788 /*
4789  * Map SOF_TIMESTAMPING_ values to PCAP_TSTAMP_ values.
4790  */
4791 static const struct {
4792 	int soft_timestamping_val;
4793 	int pcap_tstamp_val;
4794 } sof_ts_type_map[3] = {
4795 	{ SOF_TIMESTAMPING_SOFTWARE, PCAP_TSTAMP_HOST },
4796 	{ SOF_TIMESTAMPING_SYS_HARDWARE, PCAP_TSTAMP_ADAPTER },
4797 	{ SOF_TIMESTAMPING_RAW_HARDWARE, PCAP_TSTAMP_ADAPTER_UNSYNCED }
4798 };
4799 #define NUM_SOF_TIMESTAMPING_TYPES	(sizeof sof_ts_type_map / sizeof sof_ts_type_map[0])
4800 
4801 /*
4802  * Set the list of time stamping types to include all types.
4803  */
4804 static int
4805 iface_set_all_ts_types(pcap_t *handle, char *ebuf)
4806 {
4807 	u_int i;
4808 
4809 	handle->tstamp_type_list = malloc(NUM_SOF_TIMESTAMPING_TYPES * sizeof(u_int));
4810 	if (handle->tstamp_type_list == NULL) {
4811 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4812 		    errno, "malloc");
4813 		return -1;
4814 	}
4815 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++)
4816 		handle->tstamp_type_list[i] = sof_ts_type_map[i].pcap_tstamp_val;
4817 	handle->tstamp_type_count = NUM_SOF_TIMESTAMPING_TYPES;
4818 	return 0;
4819 }
4820 
4821 /*
4822  * Get a list of time stamp types.
4823  */
4824 #ifdef ETHTOOL_GET_TS_INFO
4825 static int
4826 iface_get_ts_types(const char *device, pcap_t *handle, char *ebuf)
4827 {
4828 	int fd;
4829 	struct ifreq ifr;
4830 	struct ethtool_ts_info info;
4831 	int num_ts_types;
4832 	u_int i, j;
4833 
4834 	/*
4835 	 * This doesn't apply to the "any" device; you can't say "turn on
4836 	 * hardware time stamping for all devices that exist now and arrange
4837 	 * that it be turned on for any device that appears in the future",
4838 	 * and not all devices even necessarily *support* hardware time
4839 	 * stamping, so don't report any time stamp types.
4840 	 */
4841 	if (strcmp(device, "any") == 0) {
4842 		handle->tstamp_type_list = NULL;
4843 		return 0;
4844 	}
4845 
4846 	/*
4847 	 * Create a socket from which to fetch time stamping capabilities.
4848 	 */
4849 	fd = get_if_ioctl_socket();
4850 	if (fd < 0) {
4851 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4852 		    errno, "socket for SIOCETHTOOL(ETHTOOL_GET_TS_INFO)");
4853 		return -1;
4854 	}
4855 
4856 	memset(&ifr, 0, sizeof(ifr));
4857 	pcap_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
4858 	memset(&info, 0, sizeof(info));
4859 	info.cmd = ETHTOOL_GET_TS_INFO;
4860 	ifr.ifr_data = (caddr_t)&info;
4861 	if (ioctl(fd, SIOCETHTOOL, &ifr) == -1) {
4862 		int save_errno = errno;
4863 
4864 		close(fd);
4865 		switch (save_errno) {
4866 
4867 		case EOPNOTSUPP:
4868 		case EINVAL:
4869 			/*
4870 			 * OK, this OS version or driver doesn't support
4871 			 * asking for the time stamping types, so let's
4872 			 * just return all the possible types.
4873 			 */
4874 			if (iface_set_all_ts_types(handle, ebuf) == -1)
4875 				return -1;
4876 			return 0;
4877 
4878 		case ENODEV:
4879 			/*
4880 			 * OK, no such device.
4881 			 * The user will find that out when they try to
4882 			 * activate the device; just return an empty
4883 			 * list of time stamp types.
4884 			 */
4885 			handle->tstamp_type_list = NULL;
4886 			return 0;
4887 
4888 		default:
4889 			/*
4890 			 * Other error.
4891 			 */
4892 			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4893 			    save_errno,
4894 			    "%s: SIOCETHTOOL(ETHTOOL_GET_TS_INFO) ioctl failed",
4895 			    device);
4896 			return -1;
4897 		}
4898 	}
4899 	close(fd);
4900 
4901 	/*
4902 	 * Do we support hardware time stamping of *all* packets?
4903 	 */
4904 	if (!(info.rx_filters & (1 << HWTSTAMP_FILTER_ALL))) {
4905 		/*
4906 		 * No, so don't report any time stamp types.
4907 		 *
4908 		 * XXX - some devices either don't report
4909 		 * HWTSTAMP_FILTER_ALL when they do support it, or
4910 		 * report HWTSTAMP_FILTER_ALL but map it to only
4911 		 * time stamping a few PTP packets.  See
4912 		 * http://marc.info/?l=linux-netdev&m=146318183529571&w=2
4913 		 *
4914 		 * Maybe that got fixed later.
4915 		 */
4916 		handle->tstamp_type_list = NULL;
4917 		return 0;
4918 	}
4919 
4920 	num_ts_types = 0;
4921 	for (i = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
4922 		if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val)
4923 			num_ts_types++;
4924 	}
4925 	if (num_ts_types != 0) {
4926 		handle->tstamp_type_list = malloc(num_ts_types * sizeof(u_int));
4927 		if (handle->tstamp_type_list == NULL) {
4928 			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
4929 			    errno, "malloc");
4930 			return -1;
4931 		}
4932 		for (i = 0, j = 0; i < NUM_SOF_TIMESTAMPING_TYPES; i++) {
4933 			if (info.so_timestamping & sof_ts_type_map[i].soft_timestamping_val) {
4934 				handle->tstamp_type_list[j] = sof_ts_type_map[i].pcap_tstamp_val;
4935 				j++;
4936 			}
4937 		}
4938 		handle->tstamp_type_count = num_ts_types;
4939 	} else
4940 		handle->tstamp_type_list = NULL;
4941 
4942 	return 0;
4943 }
4944 #else /* ETHTOOL_GET_TS_INFO */
4945 static int
4946 iface_get_ts_types(const char *device, pcap_t *handle, char *ebuf)
4947 {
4948 	/*
4949 	 * This doesn't apply to the "any" device; you can't say "turn on
4950 	 * hardware time stamping for all devices that exist now and arrange
4951 	 * that it be turned on for any device that appears in the future",
4952 	 * and not all devices even necessarily *support* hardware time
4953 	 * stamping, so don't report any time stamp types.
4954 	 */
4955 	if (strcmp(device, "any") == 0) {
4956 		handle->tstamp_type_list = NULL;
4957 		return 0;
4958 	}
4959 
4960 	/*
4961 	 * We don't have an ioctl to use to ask what's supported,
4962 	 * so say we support everything.
4963 	 */
4964 	if (iface_set_all_ts_types(handle, ebuf) == -1)
4965 		return -1;
4966 	return 0;
4967 }
4968 #endif /* ETHTOOL_GET_TS_INFO */
4969 #else  /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
4970 static int
4971 iface_get_ts_types(const char *device _U_, pcap_t *p _U_, char *ebuf _U_)
4972 {
4973 	/*
4974 	 * Nothing to fetch, so it always "succeeds".
4975 	 */
4976 	return 0;
4977 }
4978 #endif /* defined(HAVE_LINUX_NET_TSTAMP_H) && defined(PACKET_TIMESTAMP) */
4979 
4980 /*
4981  * Find out if we have any form of fragmentation/reassembly offloading.
4982  *
4983  * We do so using SIOCETHTOOL checking for various types of offloading;
4984  * if SIOCETHTOOL isn't defined, or we don't have any #defines for any
4985  * of the types of offloading, there's nothing we can do to check, so
4986  * we just say "no, we don't".
4987  *
4988  * We treat EOPNOTSUPP, EINVAL and, if eperm_ok is true, EPERM as
4989  * indications that the operation isn't supported.  We do EPERM
4990  * weirdly because the SIOCETHTOOL code in later kernels 1) doesn't
4991  * support ETHTOOL_GUFO, 2) also doesn't include it in the list
4992  * of ethtool operations that don't require CAP_NET_ADMIN privileges,
4993  * and 3) does the "is this permitted" check before doing the "is
4994  * this even supported" check, so it fails with "this is not permitted"
4995  * rather than "this is not even supported".  To work around this
4996  * annoyance, we only treat EPERM as an error for the first feature,
4997  * and assume that they all do the same permission checks, so if the
4998  * first one is allowed all the others are allowed if supported.
4999  */
5000 #if defined(SIOCETHTOOL) && (defined(ETHTOOL_GTSO) || defined(ETHTOOL_GUFO) || defined(ETHTOOL_GGSO) || defined(ETHTOOL_GFLAGS) || defined(ETHTOOL_GGRO))
5001 static int
5002 iface_ethtool_flag_ioctl(pcap_t *handle, int cmd, const char *cmdname,
5003     int eperm_ok)
5004 {
5005 	struct ifreq	ifr;
5006 	struct ethtool_value eval;
5007 
5008 	memset(&ifr, 0, sizeof(ifr));
5009 	pcap_strlcpy(ifr.ifr_name, handle->opt.device, sizeof(ifr.ifr_name));
5010 	eval.cmd = cmd;
5011 	eval.data = 0;
5012 	ifr.ifr_data = (caddr_t)&eval;
5013 	if (ioctl(handle->fd, SIOCETHTOOL, &ifr) == -1) {
5014 		if (errno == EOPNOTSUPP || errno == EINVAL ||
5015 		    (errno == EPERM && eperm_ok)) {
5016 			/*
5017 			 * OK, let's just return 0, which, in our
5018 			 * case, either means "no, what we're asking
5019 			 * about is not enabled" or "all the flags
5020 			 * are clear (i.e., nothing is enabled)".
5021 			 */
5022 			return 0;
5023 		}
5024 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5025 		    errno, "%s: SIOCETHTOOL(%s) ioctl failed",
5026 		    handle->opt.device, cmdname);
5027 		return -1;
5028 	}
5029 	return eval.data;
5030 }
5031 
5032 /*
5033  * XXX - it's annoying that we have to check for offloading at all, but,
5034  * given that we have to, it's still annoying that we have to check for
5035  * particular types of offloading, especially that shiny new types of
5036  * offloading may be added - and, worse, may not be checkable with
5037  * a particular ETHTOOL_ operation; ETHTOOL_GFEATURES would, in
5038  * theory, give those to you, but the actual flags being used are
5039  * opaque (defined in a non-uapi header), and there doesn't seem to
5040  * be any obvious way to ask the kernel what all the offloading flags
5041  * are - at best, you can ask for a set of strings(!) to get *names*
5042  * for various flags.  (That whole mechanism appears to have been
5043  * designed for the sole purpose of letting ethtool report flags
5044  * by name and set flags by name, with the names having no semantics
5045  * ethtool understands.)
5046  */
5047 static int
5048 iface_get_offload(pcap_t *handle)
5049 {
5050 	int ret;
5051 
5052 #ifdef ETHTOOL_GTSO
5053 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GTSO, "ETHTOOL_GTSO", 0);
5054 	if (ret == -1)
5055 		return -1;
5056 	if (ret)
5057 		return 1;	/* TCP segmentation offloading on */
5058 #endif
5059 
5060 #ifdef ETHTOOL_GGSO
5061 	/*
5062 	 * XXX - will this cause large unsegmented packets to be
5063 	 * handed to PF_PACKET sockets on transmission?  If not,
5064 	 * this need not be checked.
5065 	 */
5066 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGSO, "ETHTOOL_GGSO", 0);
5067 	if (ret == -1)
5068 		return -1;
5069 	if (ret)
5070 		return 1;	/* generic segmentation offloading on */
5071 #endif
5072 
5073 #ifdef ETHTOOL_GFLAGS
5074 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GFLAGS, "ETHTOOL_GFLAGS", 0);
5075 	if (ret == -1)
5076 		return -1;
5077 	if (ret & ETH_FLAG_LRO)
5078 		return 1;	/* large receive offloading on */
5079 #endif
5080 
5081 #ifdef ETHTOOL_GGRO
5082 	/*
5083 	 * XXX - will this cause large reassembled packets to be
5084 	 * handed to PF_PACKET sockets on receipt?  If not,
5085 	 * this need not be checked.
5086 	 */
5087 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GGRO, "ETHTOOL_GGRO", 0);
5088 	if (ret == -1)
5089 		return -1;
5090 	if (ret)
5091 		return 1;	/* generic (large) receive offloading on */
5092 #endif
5093 
5094 #ifdef ETHTOOL_GUFO
5095 	/*
5096 	 * Do this one last, as support for it was removed in later
5097 	 * kernels, and it fails with EPERM on those kernels rather
5098 	 * than with EOPNOTSUPP (see explanation in comment for
5099 	 * iface_ethtool_flag_ioctl()).
5100 	 */
5101 	ret = iface_ethtool_flag_ioctl(handle, ETHTOOL_GUFO, "ETHTOOL_GUFO", 1);
5102 	if (ret == -1)
5103 		return -1;
5104 	if (ret)
5105 		return 1;	/* UDP fragmentation offloading on */
5106 #endif
5107 
5108 	return 0;
5109 }
5110 #else /* SIOCETHTOOL */
5111 static int
5112 iface_get_offload(pcap_t *handle _U_)
5113 {
5114 	/*
5115 	 * XXX - do we need to get this information if we don't
5116 	 * have the ethtool ioctls?  If so, how do we do that?
5117 	 */
5118 	return 0;
5119 }
5120 #endif /* SIOCETHTOOL */
5121 
5122 static struct dsa_proto {
5123 	const char *name;
5124 	bpf_u_int32 linktype;
5125 } dsa_protos[] = {
5126 	/*
5127 	 * None is special and indicates that the interface does not have
5128 	 * any tagging protocol configured, and is therefore a standard
5129 	 * Ethernet interface.
5130 	 */
5131 	{ "none", DLT_EN10MB },
5132 	{ "brcm", DLT_DSA_TAG_BRCM },
5133 	{ "brcm-prepend", DLT_DSA_TAG_BRCM_PREPEND },
5134 	{ "dsa", DLT_DSA_TAG_DSA },
5135 	{ "edsa", DLT_DSA_TAG_EDSA },
5136 };
5137 
5138 static int
5139 iface_dsa_get_proto_info(const char *device, pcap_t *handle)
5140 {
5141 	char *pathstr;
5142 	unsigned int i;
5143 	/*
5144 	 * Make this significantly smaller than PCAP_ERRBUF_SIZE;
5145 	 * the tag *shouldn't* have some huge long name, and making
5146 	 * it smaller keeps newer versions of GCC from whining that
5147 	 * the error message if we don't support the tag could
5148 	 * overflow the error message buffer.
5149 	 */
5150 	char buf[128];
5151 	ssize_t r;
5152 	int fd;
5153 
5154 	fd = asprintf(&pathstr, "/sys/class/net/%s/dsa/tagging", device);
5155 	if (fd < 0) {
5156 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5157 					  fd, "asprintf");
5158 		return PCAP_ERROR;
5159 	}
5160 
5161 	fd = open(pathstr, O_RDONLY);
5162 	free(pathstr);
5163 	/*
5164 	 * This is not fatal, kernel >= 4.20 *might* expose this attribute
5165 	 */
5166 	if (fd < 0)
5167 		return 0;
5168 
5169 	r = read(fd, buf, sizeof(buf) - 1);
5170 	if (r <= 0) {
5171 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5172 					  errno, "read");
5173 		close(fd);
5174 		return PCAP_ERROR;
5175 	}
5176 	close(fd);
5177 
5178 	/*
5179 	 * Buffer should be LF terminated.
5180 	 */
5181 	if (buf[r - 1] == '\n')
5182 		r--;
5183 	buf[r] = '\0';
5184 
5185 	for (i = 0; i < sizeof(dsa_protos) / sizeof(dsa_protos[0]); i++) {
5186 		if (strlen(dsa_protos[i].name) == (size_t)r &&
5187 		    strcmp(buf, dsa_protos[i].name) == 0) {
5188 			handle->linktype = dsa_protos[i].linktype;
5189 			switch (dsa_protos[i].linktype) {
5190 			case DLT_EN10MB:
5191 				return 0;
5192 			default:
5193 				return 1;
5194 			}
5195 		}
5196 	}
5197 
5198 	snprintf(handle->errbuf, PCAP_ERRBUF_SIZE,
5199 		      "unsupported DSA tag: %s", buf);
5200 
5201 	return PCAP_ERROR;
5202 }
5203 
5204 /*
5205  *  Query the kernel for the MTU of the given interface.
5206  */
5207 static int
5208 iface_get_mtu(int fd, const char *device, char *ebuf)
5209 {
5210 	struct ifreq	ifr;
5211 
5212 	if (!device)
5213 		return BIGGER_THAN_ALL_MTUS;
5214 
5215 	memset(&ifr, 0, sizeof(ifr));
5216 	pcap_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5217 
5218 	if (ioctl(fd, SIOCGIFMTU, &ifr) == -1) {
5219 		pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5220 		    errno, "SIOCGIFMTU");
5221 		return -1;
5222 	}
5223 
5224 	return ifr.ifr_mtu;
5225 }
5226 
5227 /*
5228  *  Get the hardware type of the given interface as ARPHRD_xxx constant.
5229  */
5230 static int
5231 iface_get_arptype(int fd, const char *device, char *ebuf)
5232 {
5233 	struct ifreq	ifr;
5234 	int		ret;
5235 
5236 	memset(&ifr, 0, sizeof(ifr));
5237 	pcap_strlcpy(ifr.ifr_name, device, sizeof(ifr.ifr_name));
5238 
5239 	if (ioctl(fd, SIOCGIFHWADDR, &ifr) == -1) {
5240 		if (errno == ENODEV) {
5241 			/*
5242 			 * No such device.
5243 			 *
5244 			 * There's nothing more to say, so clear
5245 			 * the error message.
5246 			 */
5247 			ret = PCAP_ERROR_NO_SUCH_DEVICE;
5248 			ebuf[0] = '\0';
5249 		} else {
5250 			ret = PCAP_ERROR;
5251 			pcap_fmt_errmsg_for_errno(ebuf, PCAP_ERRBUF_SIZE,
5252 			    errno, "SIOCGIFHWADDR");
5253 		}
5254 		return ret;
5255 	}
5256 
5257 	return ifr.ifr_hwaddr.sa_family;
5258 }
5259 
5260 static int
5261 fix_program(pcap_t *handle, struct sock_fprog *fcode)
5262 {
5263 	struct pcap_linux *handlep = handle->priv;
5264 	size_t prog_size;
5265 	register int i;
5266 	register struct bpf_insn *p;
5267 	struct bpf_insn *f;
5268 	int len;
5269 
5270 	/*
5271 	 * Make a copy of the filter, and modify that copy if
5272 	 * necessary.
5273 	 */
5274 	prog_size = sizeof(*handle->fcode.bf_insns) * handle->fcode.bf_len;
5275 	len = handle->fcode.bf_len;
5276 	f = (struct bpf_insn *)malloc(prog_size);
5277 	if (f == NULL) {
5278 		pcap_fmt_errmsg_for_errno(handle->errbuf, PCAP_ERRBUF_SIZE,
5279 		    errno, "malloc");
5280 		return -1;
5281 	}
5282 	memcpy(f, handle->fcode.bf_insns, prog_size);
5283 	fcode->len = len;
5284 	fcode->filter = (struct sock_filter *) f;
5285 
5286 	for (i = 0; i < len; ++i) {
5287 		p = &f[i];
5288 		/*
5289 		 * What type of instruction is this?
5290 		 */
5291 		switch (BPF_CLASS(p->code)) {
5292 
5293 		case BPF_LD:
5294 		case BPF_LDX:
5295 			/*
5296 			 * It's a load instruction; is it loading
5297 			 * from the packet?
5298 			 */
5299 			switch (BPF_MODE(p->code)) {
5300 
5301 			case BPF_ABS:
5302 			case BPF_IND:
5303 			case BPF_MSH:
5304 				/*
5305 				 * Yes; are we in cooked mode?
5306 				 */
5307 				if (handlep->cooked) {
5308 					/*
5309 					 * Yes, so we need to fix this
5310 					 * instruction.
5311 					 */
5312 					if (fix_offset(handle, p) < 0) {
5313 						/*
5314 						 * We failed to do so.
5315 						 * Return 0, so our caller
5316 						 * knows to punt to userland.
5317 						 */
5318 						return 0;
5319 					}
5320 				}
5321 				break;
5322 			}
5323 			break;
5324 		}
5325 	}
5326 	return 1;	/* we succeeded */
5327 }
5328 
5329 static int
5330 fix_offset(pcap_t *handle, struct bpf_insn *p)
5331 {
5332 	/*
5333 	 * Existing references to auxiliary data shouldn't be adjusted.
5334 	 *
5335 	 * Note that SKF_AD_OFF is negative, but p->k is unsigned, so
5336 	 * we use >= and cast SKF_AD_OFF to unsigned.
5337 	 */
5338 	if (p->k >= (bpf_u_int32)SKF_AD_OFF)
5339 		return 0;
5340 	if (handle->linktype == DLT_LINUX_SLL2) {
5341 		/*
5342 		 * What's the offset?
5343 		 */
5344 		if (p->k >= SLL2_HDR_LEN) {
5345 			/*
5346 			 * It's within the link-layer payload; that starts
5347 			 * at an offset of 0, as far as the kernel packet
5348 			 * filter is concerned, so subtract the length of
5349 			 * the link-layer header.
5350 			 */
5351 			p->k -= SLL2_HDR_LEN;
5352 		} else if (p->k == 0) {
5353 			/*
5354 			 * It's the protocol field; map it to the
5355 			 * special magic kernel offset for that field.
5356 			 */
5357 			p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5358 		} else if (p->k == 4) {
5359 			/*
5360 			 * It's the ifindex field; map it to the
5361 			 * special magic kernel offset for that field.
5362 			 */
5363 			p->k = SKF_AD_OFF + SKF_AD_IFINDEX;
5364 		} else if (p->k == 10) {
5365 			/*
5366 			 * It's the packet type field; map it to the
5367 			 * special magic kernel offset for that field.
5368 			 */
5369 			p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5370 		} else if ((bpf_int32)(p->k) > 0) {
5371 			/*
5372 			 * It's within the header, but it's not one of
5373 			 * those fields; we can't do that in the kernel,
5374 			 * so punt to userland.
5375 			 */
5376 			return -1;
5377 		}
5378 	} else {
5379 		/*
5380 		 * What's the offset?
5381 		 */
5382 		if (p->k >= SLL_HDR_LEN) {
5383 			/*
5384 			 * It's within the link-layer payload; that starts
5385 			 * at an offset of 0, as far as the kernel packet
5386 			 * filter is concerned, so subtract the length of
5387 			 * the link-layer header.
5388 			 */
5389 			p->k -= SLL_HDR_LEN;
5390 		} else if (p->k == 0) {
5391 			/*
5392 			 * It's the packet type field; map it to the
5393 			 * special magic kernel offset for that field.
5394 			 */
5395 			p->k = SKF_AD_OFF + SKF_AD_PKTTYPE;
5396 		} else if (p->k == 14) {
5397 			/*
5398 			 * It's the protocol field; map it to the
5399 			 * special magic kernel offset for that field.
5400 			 */
5401 			p->k = SKF_AD_OFF + SKF_AD_PROTOCOL;
5402 		} else if ((bpf_int32)(p->k) > 0) {
5403 			/*
5404 			 * It's within the header, but it's not one of
5405 			 * those fields; we can't do that in the kernel,
5406 			 * so punt to userland.
5407 			 */
5408 			return -1;
5409 		}
5410 	}
5411 	return 0;
5412 }
5413 
5414 static int
5415 set_kernel_filter(pcap_t *handle, struct sock_fprog *fcode)
5416 {
5417 	int total_filter_on = 0;
5418 	int save_mode;
5419 	int ret;
5420 	int save_errno;
5421 
5422 	/*
5423 	 * The socket filter code doesn't discard all packets queued
5424 	 * up on the socket when the filter is changed; this means
5425 	 * that packets that don't match the new filter may show up
5426 	 * after the new filter is put onto the socket, if those
5427 	 * packets haven't yet been read.
5428 	 *
5429 	 * This means, for example, that if you do a tcpdump capture
5430 	 * with a filter, the first few packets in the capture might
5431 	 * be packets that wouldn't have passed the filter.
5432 	 *
5433 	 * We therefore discard all packets queued up on the socket
5434 	 * when setting a kernel filter.  (This isn't an issue for
5435 	 * userland filters, as the userland filtering is done after
5436 	 * packets are queued up.)
5437 	 *
5438 	 * To flush those packets, we put the socket in read-only mode,
5439 	 * and read packets from the socket until there are no more to
5440 	 * read.
5441 	 *
5442 	 * In order to keep that from being an infinite loop - i.e.,
5443 	 * to keep more packets from arriving while we're draining
5444 	 * the queue - we put the "total filter", which is a filter
5445 	 * that rejects all packets, onto the socket before draining
5446 	 * the queue.
5447 	 *
5448 	 * This code deliberately ignores any errors, so that you may
5449 	 * get bogus packets if an error occurs, rather than having
5450 	 * the filtering done in userland even if it could have been
5451 	 * done in the kernel.
5452 	 */
5453 	if (setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5454 		       &total_fcode, sizeof(total_fcode)) == 0) {
5455 		char drain[1];
5456 
5457 		/*
5458 		 * Note that we've put the total filter onto the socket.
5459 		 */
5460 		total_filter_on = 1;
5461 
5462 		/*
5463 		 * Save the socket's current mode, and put it in
5464 		 * non-blocking mode; we drain it by reading packets
5465 		 * until we get an error (which is normally a
5466 		 * "nothing more to be read" error).
5467 		 */
5468 		save_mode = fcntl(handle->fd, F_GETFL, 0);
5469 		if (save_mode == -1) {
5470 			pcap_fmt_errmsg_for_errno(handle->errbuf,
5471 			    PCAP_ERRBUF_SIZE, errno,
5472 			    "can't get FD flags when changing filter");
5473 			return -2;
5474 		}
5475 		if (fcntl(handle->fd, F_SETFL, save_mode | O_NONBLOCK) < 0) {
5476 			pcap_fmt_errmsg_for_errno(handle->errbuf,
5477 			    PCAP_ERRBUF_SIZE, errno,
5478 			    "can't set nonblocking mode when changing filter");
5479 			return -2;
5480 		}
5481 		while (recv(handle->fd, &drain, sizeof drain, MSG_TRUNC) >= 0)
5482 			;
5483 		save_errno = errno;
5484 		if (save_errno != EAGAIN) {
5485 			/*
5486 			 * Fatal error.
5487 			 *
5488 			 * If we can't restore the mode or reset the
5489 			 * kernel filter, there's nothing we can do.
5490 			 */
5491 			(void)fcntl(handle->fd, F_SETFL, save_mode);
5492 			(void)reset_kernel_filter(handle);
5493 			pcap_fmt_errmsg_for_errno(handle->errbuf,
5494 			    PCAP_ERRBUF_SIZE, save_errno,
5495 			    "recv failed when changing filter");
5496 			return -2;
5497 		}
5498 		if (fcntl(handle->fd, F_SETFL, save_mode) == -1) {
5499 			pcap_fmt_errmsg_for_errno(handle->errbuf,
5500 			    PCAP_ERRBUF_SIZE, errno,
5501 			    "can't restore FD flags when changing filter");
5502 			return -2;
5503 		}
5504 	}
5505 
5506 	/*
5507 	 * Now attach the new filter.
5508 	 */
5509 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_ATTACH_FILTER,
5510 			 fcode, sizeof(*fcode));
5511 	if (ret == -1 && total_filter_on) {
5512 		/*
5513 		 * Well, we couldn't set that filter on the socket,
5514 		 * but we could set the total filter on the socket.
5515 		 *
5516 		 * This could, for example, mean that the filter was
5517 		 * too big to put into the kernel, so we'll have to
5518 		 * filter in userland; in any case, we'll be doing
5519 		 * filtering in userland, so we need to remove the
5520 		 * total filter so we see packets.
5521 		 */
5522 		save_errno = errno;
5523 
5524 		/*
5525 		 * If this fails, we're really screwed; we have the
5526 		 * total filter on the socket, and it won't come off.
5527 		 * Report it as a fatal error.
5528 		 */
5529 		if (reset_kernel_filter(handle) == -1) {
5530 			pcap_fmt_errmsg_for_errno(handle->errbuf,
5531 			    PCAP_ERRBUF_SIZE, errno,
5532 			    "can't remove kernel total filter");
5533 			return -2;	/* fatal error */
5534 		}
5535 
5536 		errno = save_errno;
5537 	}
5538 	return ret;
5539 }
5540 
5541 static int
5542 reset_kernel_filter(pcap_t *handle)
5543 {
5544 	int ret;
5545 	/*
5546 	 * setsockopt() barfs unless it get a dummy parameter.
5547 	 * valgrind whines unless the value is initialized,
5548 	 * as it has no idea that setsockopt() ignores its
5549 	 * parameter.
5550 	 */
5551 	int dummy = 0;
5552 
5553 	ret = setsockopt(handle->fd, SOL_SOCKET, SO_DETACH_FILTER,
5554 				   &dummy, sizeof(dummy));
5555 	/*
5556 	 * Ignore ENOENT - it means "we don't have a filter", so there
5557 	 * was no filter to remove, and there's still no filter.
5558 	 *
5559 	 * Also ignore ENONET, as a lot of kernel versions had a
5560 	 * typo where ENONET, rather than ENOENT, was returned.
5561 	 */
5562 	if (ret == -1 && errno != ENOENT && errno != ENONET)
5563 		return -1;
5564 	return 0;
5565 }
5566 
5567 int
5568 pcap_set_protocol_linux(pcap_t *p, int protocol)
5569 {
5570 	if (pcap_check_activated(p))
5571 		return (PCAP_ERROR_ACTIVATED);
5572 	p->opt.protocol = protocol;
5573 	return (0);
5574 }
5575 
5576 /*
5577  * Libpcap version string.
5578  */
5579 const char *
5580 pcap_lib_version(void)
5581 {
5582 #if defined(HAVE_TPACKET3)
5583 	return (PCAP_VERSION_STRING " (with TPACKET_V3)");
5584 #else
5585 	return (PCAP_VERSION_STRING " (with TPACKET_V2)");
5586 #endif
5587 }
5588