2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1990, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org>
8 * This code is derived from the Stanford/CMU enet packet filter,
9 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
10 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
11 * Berkeley Laboratory.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
45 #include "opt_netgraph.h"
47 #include <sys/param.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
54 #include <sys/malloc.h>
56 #include <sys/mutex.h>
60 #include <sys/signalvar.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
63 #include <sys/ttycom.h>
65 #include <sys/sysent.h>
66 #include <sys/systm.h>
68 #include <sys/event.h>
73 #include <sys/socket.h>
80 #include <net/if_var.h>
81 #include <net/if_vlan_var.h>
82 #include <net/if_dl.h>
84 #include <net/bpf_buffer.h>
86 #include <net/bpf_jitter.h>
88 #include <net/bpf_zerocopy.h>
89 #include <net/bpfdesc.h>
90 #include <net/route.h>
93 #include <netinet/in.h>
94 #include <netinet/if_ether.h>
95 #include <sys/kernel.h>
96 #include <sys/sysctl.h>
98 #include <net80211/ieee80211_freebsd.h>
100 #include <security/mac/mac_framework.h>
102 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
104 static struct bpf_if_ext dead_bpf_if = {
105 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
109 #define bif_next bif_ext.bif_next
110 #define bif_dlist bif_ext.bif_dlist
111 struct bpf_if_ext bif_ext; /* public members */
112 u_int bif_dlt; /* link layer type */
113 u_int bif_hdrlen; /* length of link header */
114 struct bpfd_list bif_wlist; /* writer-only list */
115 struct ifnet *bif_ifp; /* corresponding interface */
116 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
117 volatile u_int bif_refcnt;
118 struct epoch_context epoch_ctx;
121 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
123 struct bpf_program_buffer {
124 struct epoch_context epoch_ctx;
126 bpf_jit_filter *func;
131 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
133 #define PRINET 26 /* interruptible */
134 #define BPF_PRIO_MAX 7
136 #define SIZEOF_BPF_HDR(type) \
137 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
139 #ifdef COMPAT_FREEBSD32
140 #include <sys/mount.h>
141 #include <compat/freebsd32/freebsd32.h>
142 #define BPF_ALIGNMENT32 sizeof(int32_t)
143 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
147 * 32-bit version of structure prepended to each packet. We use this header
148 * instead of the standard one for 32-bit streams. We mark the a stream as
149 * 32-bit the first time we see a 32-bit compat ioctl request.
152 struct timeval32 bh_tstamp; /* time stamp */
153 uint32_t bh_caplen; /* length of captured portion */
154 uint32_t bh_datalen; /* original length of packet */
155 uint16_t bh_hdrlen; /* length of bpf header (this struct
156 plus alignment padding) */
160 struct bpf_program32 {
165 struct bpf_dltlist32 {
170 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
171 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
172 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
173 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
174 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
175 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
178 #define BPF_LOCK() sx_xlock(&bpf_sx)
179 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
180 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
182 * bpf_iflist is a list of BPF interface structures, each corresponding to a
183 * specific DLT. The same network interface might have several BPF interface
184 * structures registered by different layers in the stack (i.e., 802.11
185 * frames, ethernet frames, etc).
187 CK_LIST_HEAD(bpf_iflist, bpf_if);
188 static struct bpf_iflist bpf_iflist;
189 static struct sx bpf_sx; /* bpf global lock */
190 static int bpf_bpfd_cnt;
192 static void bpfif_ref(struct bpf_if *);
193 static void bpfif_rele(struct bpf_if *);
195 static void bpfd_ref(struct bpf_d *);
196 static void bpfd_rele(struct bpf_d *);
197 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
198 static void bpf_detachd(struct bpf_d *);
199 static void bpf_detachd_locked(struct bpf_d *, bool);
200 static void bpfd_free(epoch_context_t);
201 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
202 struct sockaddr *, int *, struct bpf_d *);
203 static int bpf_setif(struct bpf_d *, struct ifreq *);
204 static void bpf_timed_out(void *);
206 bpf_wakeup(struct bpf_d *);
207 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
208 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
210 static void reset_d(struct bpf_d *);
211 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
212 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
213 static int bpf_setdlt(struct bpf_d *, u_int);
214 static void filt_bpfdetach(struct knote *);
215 static int filt_bpfread(struct knote *, long);
216 static int filt_bpfwrite(struct knote *, long);
217 static void bpf_drvinit(void *);
218 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
220 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
222 int bpf_maxinsns = BPF_MAXINSNS;
223 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
224 &bpf_maxinsns, 0, "Maximum bpf program instructions");
225 static int bpf_zerocopy_enable = 0;
226 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
227 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
228 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
229 bpf_stats_sysctl, "bpf statistics portal");
231 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
232 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
233 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
234 &VNET_NAME(bpf_optimize_writers), 0,
235 "Do not send packets until BPF program is set");
237 static d_open_t bpfopen;
238 static d_read_t bpfread;
239 static d_write_t bpfwrite;
240 static d_ioctl_t bpfioctl;
241 static d_poll_t bpfpoll;
242 static d_kqfilter_t bpfkqfilter;
244 static struct cdevsw bpf_cdevsw = {
245 .d_version = D_VERSION,
252 .d_kqfilter = bpfkqfilter,
255 static struct filterops bpfread_filtops = {
257 .f_detach = filt_bpfdetach,
258 .f_event = filt_bpfread,
261 static struct filterops bpfwrite_filtops = {
263 .f_detach = filt_bpfdetach,
264 .f_event = filt_bpfwrite,
268 * LOCKING MODEL USED BY BPF
271 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
272 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
273 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
274 * structure fields used by bpf_*tap* code.
276 * Lock order: global lock, then descriptor lock.
278 * There are several possible consumers:
280 * 1. The kernel registers interface pointer with bpfattach().
281 * Each call allocates new bpf_if structure, references ifnet pointer
282 * and links bpf_if into bpf_iflist chain. This is protected with global
285 * 2. An userland application uses ioctl() call to bpf_d descriptor.
286 * All such call are serialized with global lock. BPF filters can be
287 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
288 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
289 * filter pointers, even if change will happen during bpf_tap execution.
290 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
292 * 3. An userland application can write packets into bpf_d descriptor.
293 * There we need to be sure, that ifnet won't disappear during bpfwrite().
295 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
296 * bif_dlist is protected with net_epoch_preempt section. So, it should
297 * be safe to make access to bpf_d descriptor inside the section.
299 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
300 * are modified with global lock held and actual free() is done using
305 bpfif_free(epoch_context_t ctx)
309 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
310 if_rele(bp->bif_ifp);
315 bpfif_ref(struct bpf_if *bp)
318 refcount_acquire(&bp->bif_refcnt);
322 bpfif_rele(struct bpf_if *bp)
325 if (!refcount_release(&bp->bif_refcnt))
327 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
331 bpfd_ref(struct bpf_d *d)
334 refcount_acquire(&d->bd_refcnt);
338 bpfd_rele(struct bpf_d *d)
341 if (!refcount_release(&d->bd_refcnt))
343 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
346 static struct bpf_program_buffer*
347 bpf_program_buffer_alloc(size_t size, int flags)
350 return (malloc(sizeof(struct bpf_program_buffer) + size,
355 bpf_program_buffer_free(epoch_context_t ctx)
357 struct bpf_program_buffer *ptr;
359 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
361 if (ptr->func != NULL)
362 bpf_destroy_jit_filter(ptr->func);
368 * Wrapper functions for various buffering methods. If the set of buffer
369 * modes expands, we will probably want to introduce a switch data structure
370 * similar to protosw, et.
373 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
379 switch (d->bd_bufmode) {
380 case BPF_BUFMODE_BUFFER:
381 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
383 case BPF_BUFMODE_ZBUF:
384 counter_u64_add(d->bd_zcopy, 1);
385 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
388 panic("bpf_buf_append_bytes");
393 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
399 switch (d->bd_bufmode) {
400 case BPF_BUFMODE_BUFFER:
401 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
403 case BPF_BUFMODE_ZBUF:
404 counter_u64_add(d->bd_zcopy, 1);
405 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
408 panic("bpf_buf_append_mbuf");
413 * This function gets called when the free buffer is re-assigned.
416 bpf_buf_reclaimed(struct bpf_d *d)
421 switch (d->bd_bufmode) {
422 case BPF_BUFMODE_BUFFER:
425 case BPF_BUFMODE_ZBUF:
426 bpf_zerocopy_buf_reclaimed(d);
430 panic("bpf_buf_reclaimed");
435 * If the buffer mechanism has a way to decide that a held buffer can be made
436 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
437 * returned if the buffer can be discarded, (0) is returned if it cannot.
440 bpf_canfreebuf(struct bpf_d *d)
445 switch (d->bd_bufmode) {
446 case BPF_BUFMODE_ZBUF:
447 return (bpf_zerocopy_canfreebuf(d));
453 * Allow the buffer model to indicate that the current store buffer is
454 * immutable, regardless of the appearance of space. Return (1) if the
455 * buffer is writable, and (0) if not.
458 bpf_canwritebuf(struct bpf_d *d)
462 switch (d->bd_bufmode) {
463 case BPF_BUFMODE_ZBUF:
464 return (bpf_zerocopy_canwritebuf(d));
470 * Notify buffer model that an attempt to write to the store buffer has
471 * resulted in a dropped packet, in which case the buffer may be considered
475 bpf_buffull(struct bpf_d *d)
480 switch (d->bd_bufmode) {
481 case BPF_BUFMODE_ZBUF:
482 bpf_zerocopy_buffull(d);
488 * Notify the buffer model that a buffer has moved into the hold position.
491 bpf_bufheld(struct bpf_d *d)
496 switch (d->bd_bufmode) {
497 case BPF_BUFMODE_ZBUF:
498 bpf_zerocopy_bufheld(d);
504 bpf_free(struct bpf_d *d)
507 switch (d->bd_bufmode) {
508 case BPF_BUFMODE_BUFFER:
509 return (bpf_buffer_free(d));
511 case BPF_BUFMODE_ZBUF:
512 return (bpf_zerocopy_free(d));
515 panic("bpf_buf_free");
520 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
523 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
525 return (bpf_buffer_uiomove(d, buf, len, uio));
529 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
532 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
534 return (bpf_buffer_ioctl_sblen(d, i));
538 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
541 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
543 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
547 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
550 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
552 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
556 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
559 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
561 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
565 * General BPF functions.
568 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
569 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
571 const struct ieee80211_bpf_params *p;
572 struct ether_header *eh;
580 * Build a sockaddr based on the data link layer type.
581 * We do this at this level because the ethernet header
582 * is copied directly into the data field of the sockaddr.
583 * In the case of SLIP, there is no header and the packet
584 * is forwarded as is.
585 * Also, we are careful to leave room at the front of the mbuf
586 * for the link level header.
590 sockp->sa_family = AF_INET;
595 sockp->sa_family = AF_UNSPEC;
596 /* XXX Would MAXLINKHDR be better? */
597 hlen = ETHER_HDR_LEN;
601 sockp->sa_family = AF_IMPLINK;
606 sockp->sa_family = AF_UNSPEC;
612 * null interface types require a 4 byte pseudo header which
613 * corresponds to the address family of the packet.
615 sockp->sa_family = AF_UNSPEC;
619 case DLT_ATM_RFC1483:
621 * en atm driver requires 4-byte atm pseudo header.
622 * though it isn't standard, vpi:vci needs to be
625 sockp->sa_family = AF_UNSPEC;
626 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
630 sockp->sa_family = AF_UNSPEC;
631 hlen = 4; /* This should match PPP_HDRLEN */
634 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
635 sockp->sa_family = AF_IEEE80211;
639 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
640 sockp->sa_family = AF_IEEE80211;
641 sockp->sa_len = 12; /* XXX != 0 */
642 hlen = sizeof(struct ieee80211_bpf_params);
649 len = uio->uio_resid;
650 if (len < hlen || len - hlen > ifp->if_mtu)
653 /* Allocate a mbuf for our write, since m_get2 fails if len >= to MJUMPAGESIZE, use m_getjcl for bigger buffers */
654 m = m_get3(len, M_WAITOK, MT_DATA, M_PKTHDR);
657 m->m_pkthdr.len = m->m_len = len;
660 error = uiomove(mtod(m, u_char *), len, uio);
664 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
670 /* Check for multicast destination */
673 eh = mtod(m, struct ether_header *);
674 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
675 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
676 ETHER_ADDR_LEN) == 0)
677 m->m_flags |= M_BCAST;
679 m->m_flags |= M_MCAST;
681 if (d->bd_hdrcmplt == 0) {
682 memcpy(eh->ether_shost, IF_LLADDR(ifp),
683 sizeof(eh->ether_shost));
689 * Make room for link header, and copy it to sockaddr
692 if (sockp->sa_family == AF_IEEE80211) {
694 * Collect true length from the parameter header
695 * NB: sockp is known to be zero'd so if we do a
696 * short copy unspecified parameters will be
698 * NB: packet may not be aligned after stripping
702 p = mtod(m, const struct ieee80211_bpf_params *);
704 if (hlen > sizeof(sockp->sa_data)) {
709 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
720 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
721 * then reset its buffers and counters with reset_d().
724 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
731 * Save sysctl value to protect from sysctl change
734 op_w = V_bpf_optimize_writers || d->bd_writer;
736 if (d->bd_bif != NULL)
737 bpf_detachd_locked(d, false);
739 * Point d at bp, and add d to the interface's list.
740 * Since there are many applications using BPF for
741 * sending raw packets only (dhcpd, cdpd are good examples)
742 * we can delay adding d to the list of active listeners until
743 * some filter is configured.
748 * Hold reference to bpif while descriptor uses this interface.
753 /* Add to writers-only list */
754 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
756 * We decrement bd_writer on every filter set operation.
757 * First BIOCSETF is done by pcap_open_live() to set up
758 * snap length. After that appliation usually sets its own
763 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
767 /* Trigger EVFILT_WRITE events. */
773 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
774 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
777 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
781 * Check if we need to upgrade our descriptor @d from write-only mode.
784 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
787 int is_snap, need_upgrade;
790 * Check if we've already upgraded or new filter is empty.
792 if (d->bd_writer == 0 || fcode == NULL)
798 * Check if cmd looks like snaplen setting from
799 * pcap_bpf.c:pcap_open_live().
800 * Note we're not checking .k value here:
801 * while pcap_open_live() definitely sets to non-zero value,
802 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
803 * do not consider upgrading immediately
805 if (cmd == BIOCSETF && flen == 1 &&
806 fcode[0].code == (BPF_RET | BPF_K))
813 * We're setting first filter and it doesn't look like
814 * setting snaplen. We're probably using bpf directly.
815 * Upgrade immediately.
820 * Do not require upgrade by first BIOCSETF
821 * (used to set snaplen) by pcap_open_live().
824 if (--d->bd_writer == 0) {
826 * First snaplen filter has already
827 * been set. This is probably catch-all
835 "%s: filter function set by pid %d, "
836 "bd_writer counter %d, snap %d upgrade %d",
837 __func__, d->bd_pid, d->bd_writer,
838 is_snap, need_upgrade);
840 return (need_upgrade);
844 * Detach a file from its interface.
847 bpf_detachd(struct bpf_d *d)
850 bpf_detachd_locked(d, false);
855 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
862 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
864 /* Check if descriptor is attached */
865 if ((bp = d->bd_bif) == NULL)
869 /* Remove d from the interface's descriptor list. */
870 CK_LIST_REMOVE(d, bd_next);
871 /* Save bd_writer value */
872 error = d->bd_writer;
877 * Notify descriptor as it's detached, so that any
878 * sleepers wake up and get ENXIO.
885 /* Call event handler iff d is attached */
887 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
890 * Check if this descriptor had requested promiscuous mode.
891 * If so and ifnet is not detached, turn it off.
893 if (d->bd_promisc && !detached_ifp) {
895 CURVNET_SET(ifp->if_vnet);
896 error = ifpromisc(ifp, 0);
898 if (error != 0 && error != ENXIO) {
900 * ENXIO can happen if a pccard is unplugged
901 * Something is really wrong if we were able to put
902 * the driver into promiscuous mode, but can't
905 if_printf(bp->bif_ifp,
906 "bpf_detach: ifpromisc failed (%d)\n", error);
913 * Close the descriptor by detaching it from its interface,
914 * deallocating its buffers, and marking it free.
919 struct bpf_d *d = data;
922 if (d->bd_state == BPF_WAITING)
923 callout_stop(&d->bd_callout);
924 d->bd_state = BPF_IDLE;
926 funsetown(&d->bd_sigio);
929 mac_bpfdesc_destroy(d);
931 seldrain(&d->bd_sel);
932 knlist_destroy(&d->bd_sel.si_note);
933 callout_drain(&d->bd_callout);
938 * Open ethernet device. Returns ENXIO for illegal minor device number,
939 * EBUSY if file is open by another process.
943 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
948 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
949 error = devfs_set_cdevpriv(d, bpf_dtor);
956 d->bd_rcount = counter_u64_alloc(M_WAITOK);
957 d->bd_dcount = counter_u64_alloc(M_WAITOK);
958 d->bd_fcount = counter_u64_alloc(M_WAITOK);
959 d->bd_wcount = counter_u64_alloc(M_WAITOK);
960 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
961 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
962 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
965 * For historical reasons, perform a one-time initialization call to
966 * the buffer routines, even though we're not yet committed to a
967 * particular buffer method.
970 if ((flags & FREAD) == 0)
972 d->bd_hbuf_in_use = 0;
973 d->bd_bufmode = BPF_BUFMODE_BUFFER;
975 d->bd_direction = BPF_D_INOUT;
977 BPF_PID_REFRESH(d, td);
980 mac_bpfdesc_create(td->td_ucred, d);
982 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
983 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
984 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
986 /* Disable VLAN pcp tagging. */
993 * bpfread - read next chunk of packets from buffers
996 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
1003 error = devfs_get_cdevpriv((void **)&d);
1008 * Restrict application to use a buffer the same size as
1009 * as kernel buffers.
1011 if (uio->uio_resid != d->bd_bufsize)
1014 non_block = ((ioflag & O_NONBLOCK) != 0);
1017 BPF_PID_REFRESH_CUR(d);
1018 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1020 return (EOPNOTSUPP);
1022 if (d->bd_state == BPF_WAITING)
1023 callout_stop(&d->bd_callout);
1024 timed_out = (d->bd_state == BPF_TIMED_OUT);
1025 d->bd_state = BPF_IDLE;
1026 while (d->bd_hbuf_in_use) {
1027 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1028 PRINET|PCATCH, "bd_hbuf", 0);
1035 * If the hold buffer is empty, then do a timed sleep, which
1036 * ends when the timeout expires or when enough packets
1037 * have arrived to fill the store buffer.
1039 while (d->bd_hbuf == NULL) {
1040 if (d->bd_slen != 0) {
1042 * A packet(s) either arrived since the previous
1043 * read or arrived while we were asleep.
1045 if (d->bd_immediate || non_block || timed_out) {
1047 * Rotate the buffers and return what's here
1048 * if we are in immediate mode, non-blocking
1049 * flag is set, or this descriptor timed out.
1057 * No data is available, check to see if the bpf device
1058 * is still pointed at a real interface. If not, return
1059 * ENXIO so that the userland process knows to rebind
1060 * it before using it again.
1062 if (d->bd_bif == NULL) {
1069 return (EWOULDBLOCK);
1071 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1072 "bpf", d->bd_rtout);
1073 if (error == EINTR || error == ERESTART) {
1077 if (error == EWOULDBLOCK) {
1079 * On a timeout, return what's in the buffer,
1080 * which may be nothing. If there is something
1081 * in the store buffer, we can rotate the buffers.
1085 * We filled up the buffer in between
1086 * getting the timeout and arriving
1087 * here, so we don't need to rotate.
1091 if (d->bd_slen == 0) {
1100 * At this point, we know we have something in the hold slot.
1102 d->bd_hbuf_in_use = 1;
1106 * Move data from hold buffer into user space.
1107 * We know the entire buffer is transferred since
1108 * we checked above that the read buffer is bpf_bufsize bytes.
1110 * We do not have to worry about simultaneous reads because
1111 * we waited for sole access to the hold buffer above.
1113 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1116 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1117 d->bd_fbuf = d->bd_hbuf;
1120 bpf_buf_reclaimed(d);
1121 d->bd_hbuf_in_use = 0;
1122 wakeup(&d->bd_hbuf_in_use);
1129 * If there are processes sleeping on this descriptor, wake them up.
1131 static __inline void
1132 bpf_wakeup(struct bpf_d *d)
1135 BPFD_LOCK_ASSERT(d);
1136 if (d->bd_state == BPF_WAITING) {
1137 callout_stop(&d->bd_callout);
1138 d->bd_state = BPF_IDLE;
1141 if (d->bd_async && d->bd_sig && d->bd_sigio)
1142 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1144 selwakeuppri(&d->bd_sel, PRINET);
1145 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1149 bpf_timed_out(void *arg)
1151 struct bpf_d *d = (struct bpf_d *)arg;
1153 BPFD_LOCK_ASSERT(d);
1155 if (callout_pending(&d->bd_callout) ||
1156 !callout_active(&d->bd_callout))
1158 if (d->bd_state == BPF_WAITING) {
1159 d->bd_state = BPF_TIMED_OUT;
1160 if (d->bd_slen != 0)
1166 bpf_ready(struct bpf_d *d)
1169 BPFD_LOCK_ASSERT(d);
1171 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1173 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1180 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1183 struct sockaddr dst;
1184 struct epoch_tracker et;
1188 struct mbuf *m, *mc;
1191 error = devfs_get_cdevpriv((void **)&d);
1195 NET_EPOCH_ENTER(et);
1197 BPF_PID_REFRESH_CUR(d);
1198 counter_u64_add(d->bd_wcount, 1);
1199 if ((bp = d->bd_bif) == NULL) {
1205 if ((ifp->if_flags & IFF_UP) == 0) {
1210 if (uio->uio_resid == 0)
1213 bzero(&dst, sizeof(dst));
1218 * Take extra reference, unlock d and exit from epoch section,
1219 * since bpf_movein() can sleep.
1225 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1226 &m, &dst, &hlen, d);
1229 counter_u64_add(d->bd_wdcount, 1);
1236 * Check that descriptor is still attached to the interface.
1237 * This can happen on bpfdetach(). To avoid access to detached
1238 * ifnet, free mbuf and return ENXIO.
1240 if (d->bd_bif == NULL) {
1241 counter_u64_add(d->bd_wdcount, 1);
1247 counter_u64_add(d->bd_wfcount, 1);
1249 dst.sa_family = pseudo_AF_HDRCMPLT;
1251 if (d->bd_feedback) {
1252 mc = m_dup(m, M_NOWAIT);
1254 mc->m_pkthdr.rcvif = ifp;
1255 /* Set M_PROMISC for outgoing packets to be discarded. */
1256 if (d->bd_direction == BPF_D_INOUT)
1257 m->m_flags |= M_PROMISC;
1261 m->m_pkthdr.len -= hlen;
1263 m->m_data += hlen; /* XXX */
1265 CURVNET_SET(ifp->if_vnet);
1267 mac_bpfdesc_create_mbuf(d, m);
1269 mac_bpfdesc_create_mbuf(d, mc);
1272 bzero(&ro, sizeof(ro));
1274 ro.ro_prepend = (u_char *)&dst.sa_data;
1276 ro.ro_flags = RT_HAS_HEADER;
1280 vlan_set_pcp(m, d->bd_pcp);
1282 /* Avoid possible recursion on BPFD_LOCK(). */
1283 NET_EPOCH_ENTER(et);
1285 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1287 counter_u64_add(d->bd_wdcount, 1);
1291 (*ifp->if_input)(ifp, mc);
1301 counter_u64_add(d->bd_wdcount, 1);
1308 * Reset a descriptor by flushing its packet buffer and clearing the receive
1309 * and drop counts. This is doable for kernel-only buffers, but with
1310 * zero-copy buffers, we can't write to (or rotate) buffers that are
1311 * currently owned by userspace. It would be nice if we could encapsulate
1312 * this logic in the buffer code rather than here.
1315 reset_d(struct bpf_d *d)
1318 BPFD_LOCK_ASSERT(d);
1320 while (d->bd_hbuf_in_use)
1321 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1323 if ((d->bd_hbuf != NULL) &&
1324 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1325 /* Free the hold buffer. */
1326 d->bd_fbuf = d->bd_hbuf;
1329 bpf_buf_reclaimed(d);
1331 if (bpf_canwritebuf(d))
1333 counter_u64_zero(d->bd_rcount);
1334 counter_u64_zero(d->bd_dcount);
1335 counter_u64_zero(d->bd_fcount);
1336 counter_u64_zero(d->bd_wcount);
1337 counter_u64_zero(d->bd_wfcount);
1338 counter_u64_zero(d->bd_wdcount);
1339 counter_u64_zero(d->bd_zcopy);
1343 * FIONREAD Check for read packet available.
1344 * BIOCGBLEN Get buffer len [for read()].
1345 * BIOCSETF Set read filter.
1346 * BIOCSETFNR Set read filter without resetting descriptor.
1347 * BIOCSETWF Set write filter.
1348 * BIOCFLUSH Flush read packet buffer.
1349 * BIOCPROMISC Put interface into promiscuous mode.
1350 * BIOCGDLT Get link layer type.
1351 * BIOCGETIF Get interface name.
1352 * BIOCSETIF Set interface.
1353 * BIOCSRTIMEOUT Set read timeout.
1354 * BIOCGRTIMEOUT Get read timeout.
1355 * BIOCGSTATS Get packet stats.
1356 * BIOCIMMEDIATE Set immediate mode.
1357 * BIOCVERSION Get filter language version.
1358 * BIOCGHDRCMPLT Get "header already complete" flag
1359 * BIOCSHDRCMPLT Set "header already complete" flag
1360 * BIOCGDIRECTION Get packet direction flag
1361 * BIOCSDIRECTION Set packet direction flag
1362 * BIOCGTSTAMP Get time stamp format and resolution.
1363 * BIOCSTSTAMP Set time stamp format and resolution.
1364 * BIOCLOCK Set "locked" flag
1365 * BIOCFEEDBACK Set packet feedback mode.
1366 * BIOCSETZBUF Set current zero-copy buffer locations.
1367 * BIOCGETZMAX Get maximum zero-copy buffer size.
1368 * BIOCROTZBUF Force rotation of zero-copy buffer
1369 * BIOCSETBUFMODE Set buffer mode.
1370 * BIOCGETBUFMODE Get current buffer mode.
1371 * BIOCSETVLANPCP Set VLAN PCP tag.
1375 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1381 error = devfs_get_cdevpriv((void **)&d);
1386 * Refresh PID associated with this descriptor.
1389 BPF_PID_REFRESH(d, td);
1390 if (d->bd_state == BPF_WAITING)
1391 callout_stop(&d->bd_callout);
1392 d->bd_state = BPF_IDLE;
1395 if (d->bd_locked == 1) {
1401 #ifdef COMPAT_FREEBSD32
1402 case BIOCGDLTLIST32:
1406 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1407 case BIOCGRTIMEOUT32:
1418 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1419 case BIOCSRTIMEOUT32:
1429 #ifdef COMPAT_FREEBSD32
1431 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1432 * that it will get 32-bit packet headers.
1438 case BIOCGDLTLIST32:
1439 case BIOCGRTIMEOUT32:
1440 case BIOCSRTIMEOUT32:
1441 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1449 CURVNET_SET(TD_TO_VNET(td));
1456 * Check for read packet available.
1464 while (d->bd_hbuf_in_use)
1465 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1466 PRINET, "bd_hbuf", 0);
1476 * Get buffer len [for read()].
1480 *(u_int *)addr = d->bd_bufsize;
1485 * Set buffer length.
1488 error = bpf_ioctl_sblen(d, (u_int *)addr);
1492 * Set link layer read filter.
1497 #ifdef COMPAT_FREEBSD32
1502 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1506 * Flush read packet buffer.
1515 * Put interface into promiscuous mode.
1518 if (d->bd_bif == NULL) {
1520 * No interface attached yet.
1525 if (d->bd_promisc == 0) {
1526 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1533 * Get current data link type.
1537 if (d->bd_bif == NULL)
1540 *(u_int *)addr = d->bd_bif->bif_dlt;
1545 * Get a list of supported data link types.
1547 #ifdef COMPAT_FREEBSD32
1548 case BIOCGDLTLIST32:
1550 struct bpf_dltlist32 *list32;
1551 struct bpf_dltlist dltlist;
1553 list32 = (struct bpf_dltlist32 *)addr;
1554 dltlist.bfl_len = list32->bfl_len;
1555 dltlist.bfl_list = PTRIN(list32->bfl_list);
1557 if (d->bd_bif == NULL)
1560 error = bpf_getdltlist(d, &dltlist);
1562 list32->bfl_len = dltlist.bfl_len;
1571 if (d->bd_bif == NULL)
1574 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1579 * Set data link type.
1583 if (d->bd_bif == NULL)
1586 error = bpf_setdlt(d, *(u_int *)addr);
1591 * Get interface name.
1595 if (d->bd_bif == NULL)
1598 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1599 struct ifreq *const ifr = (struct ifreq *)addr;
1601 strlcpy(ifr->ifr_name, ifp->if_xname,
1602 sizeof(ifr->ifr_name));
1612 int alloc_buf, size;
1615 * Behavior here depends on the buffering model. If
1616 * we're using kernel memory buffers, then we can
1617 * allocate them here. If we're using zero-copy,
1618 * then the user process must have registered buffers
1619 * by the time we get here.
1623 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1628 size = d->bd_bufsize;
1629 error = bpf_buffer_ioctl_sblen(d, &size);
1634 error = bpf_setif(d, (struct ifreq *)addr);
1643 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1644 case BIOCSRTIMEOUT32:
1647 struct timeval *tv = (struct timeval *)addr;
1648 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1649 struct timeval32 *tv32;
1650 struct timeval tv64;
1652 if (cmd == BIOCSRTIMEOUT32) {
1653 tv32 = (struct timeval32 *)addr;
1655 tv->tv_sec = tv32->tv_sec;
1656 tv->tv_usec = tv32->tv_usec;
1659 tv = (struct timeval *)addr;
1662 * Subtract 1 tick from tvtohz() since this isn't
1665 if ((error = itimerfix(tv)) == 0)
1666 d->bd_rtout = tvtohz(tv) - 1;
1674 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1675 case BIOCGRTIMEOUT32:
1679 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1680 struct timeval32 *tv32;
1681 struct timeval tv64;
1683 if (cmd == BIOCGRTIMEOUT32)
1687 tv = (struct timeval *)addr;
1689 tv->tv_sec = d->bd_rtout / hz;
1690 tv->tv_usec = (d->bd_rtout % hz) * tick;
1691 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1692 if (cmd == BIOCGRTIMEOUT32) {
1693 tv32 = (struct timeval32 *)addr;
1694 tv32->tv_sec = tv->tv_sec;
1695 tv32->tv_usec = tv->tv_usec;
1707 struct bpf_stat *bs = (struct bpf_stat *)addr;
1709 /* XXXCSJP overflow */
1710 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1711 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1716 * Set immediate mode.
1720 d->bd_immediate = *(u_int *)addr;
1726 struct bpf_version *bv = (struct bpf_version *)addr;
1728 bv->bv_major = BPF_MAJOR_VERSION;
1729 bv->bv_minor = BPF_MINOR_VERSION;
1734 * Get "header already complete" flag
1738 *(u_int *)addr = d->bd_hdrcmplt;
1743 * Set "header already complete" flag
1747 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1752 * Get packet direction flag
1754 case BIOCGDIRECTION:
1756 *(u_int *)addr = d->bd_direction;
1761 * Set packet direction flag
1763 case BIOCSDIRECTION:
1767 direction = *(u_int *)addr;
1768 switch (direction) {
1773 d->bd_direction = direction;
1783 * Get packet timestamp format and resolution.
1787 *(u_int *)addr = d->bd_tstamp;
1792 * Set packet timestamp format and resolution.
1798 func = *(u_int *)addr;
1799 if (BPF_T_VALID(func))
1800 d->bd_tstamp = func;
1808 d->bd_feedback = *(u_int *)addr;
1818 case FIONBIO: /* Non-blocking I/O */
1821 case FIOASYNC: /* Send signal on receive packets */
1823 d->bd_async = *(int *)addr;
1829 * XXX: Add some sort of locking here?
1830 * fsetown() can sleep.
1832 error = fsetown(*(int *)addr, &d->bd_sigio);
1837 *(int *)addr = fgetown(&d->bd_sigio);
1841 /* This is deprecated, FIOSETOWN should be used instead. */
1843 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1846 /* This is deprecated, FIOGETOWN should be used instead. */
1848 *(int *)addr = -fgetown(&d->bd_sigio);
1851 case BIOCSRSIG: /* Set receive signal */
1855 sig = *(u_int *)addr;
1868 *(u_int *)addr = d->bd_sig;
1872 case BIOCGETBUFMODE:
1874 *(u_int *)addr = d->bd_bufmode;
1878 case BIOCSETBUFMODE:
1880 * Allow the buffering mode to be changed as long as we
1881 * haven't yet committed to a particular mode. Our
1882 * definition of commitment, for now, is whether or not a
1883 * buffer has been allocated or an interface attached, since
1884 * that's the point where things get tricky.
1886 switch (*(u_int *)addr) {
1887 case BPF_BUFMODE_BUFFER:
1890 case BPF_BUFMODE_ZBUF:
1891 if (bpf_zerocopy_enable)
1901 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1902 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1907 d->bd_bufmode = *(u_int *)addr;
1912 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1916 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1920 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1923 case BIOCSETVLANPCP:
1927 pcp = *(u_int *)addr;
1928 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1941 * Set d's packet filter program to fp. If this file already has a filter,
1942 * free it and replace it. Returns EINVAL for bogus requests.
1944 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1948 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1950 #ifdef COMPAT_FREEBSD32
1951 struct bpf_program fp_swab;
1952 struct bpf_program32 *fp32;
1954 struct bpf_program_buffer *fcode;
1955 struct bpf_insn *filter;
1957 bpf_jit_filter *jfunc;
1963 #ifdef COMPAT_FREEBSD32
1968 fp32 = (struct bpf_program32 *)fp;
1969 fp_swab.bf_len = fp32->bf_len;
1971 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1990 * Check new filter validness before acquiring any locks.
1991 * Allocate memory for new filter, if needed.
1994 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1996 size = flen * sizeof(*fp->bf_insns);
1998 /* We're setting up new filter. Copy and check actual data. */
1999 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
2000 filter = (struct bpf_insn *)fcode->buffer;
2001 if (copyin(fp->bf_insns, filter, size) != 0 ||
2002 !bpf_validate(filter, flen)) {
2007 if (cmd != BIOCSETWF) {
2009 * Filter is copied inside fcode and is
2012 jfunc = bpf_jitter(filter, flen);
2017 track_event = false;
2022 /* Set up new filter. */
2023 if (cmd == BIOCSETWF) {
2024 if (d->bd_wfilter != NULL) {
2025 fcode = __containerof((void *)d->bd_wfilter,
2026 struct bpf_program_buffer, buffer);
2031 d->bd_wfilter = filter;
2033 if (d->bd_rfilter != NULL) {
2034 fcode = __containerof((void *)d->bd_rfilter,
2035 struct bpf_program_buffer, buffer);
2037 fcode->func = d->bd_bfilter;
2040 d->bd_rfilter = filter;
2042 d->bd_bfilter = jfunc;
2044 if (cmd == BIOCSETF)
2047 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2049 * Filter can be set several times without
2050 * specifying interface. In this case just mark d
2054 if (d->bd_bif != NULL) {
2056 * Remove descriptor from writers-only list
2057 * and add it to active readers list.
2059 CK_LIST_REMOVE(d, bd_next);
2060 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2063 "%s: upgrade required by pid %d",
2064 __func__, d->bd_pid);
2072 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2075 EVENTHANDLER_INVOKE(bpf_track,
2076 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2083 * Detach a file from its current interface (if attached at all) and attach
2084 * to the interface indicated by the name stored in ifr.
2085 * Return an errno or 0.
2088 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2091 struct ifnet *theywant;
2095 theywant = ifunit(ifr->ifr_name);
2096 if (theywant == NULL || theywant->if_bpf == NULL)
2099 bp = theywant->if_bpf;
2101 * At this point, we expect the buffer is already allocated. If not,
2104 switch (d->bd_bufmode) {
2105 case BPF_BUFMODE_BUFFER:
2106 case BPF_BUFMODE_ZBUF:
2107 if (d->bd_sbuf == NULL)
2112 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2114 if (bp != d->bd_bif)
2125 * Support for select() and poll() system calls
2127 * Return true iff the specific operation will not block indefinitely.
2128 * Otherwise, return false but make a note that a selwakeup() must be done.
2131 bpfpoll(struct cdev *dev, int events, struct thread *td)
2136 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2138 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2141 * Refresh PID associated with this descriptor.
2143 revents = events & (POLLOUT | POLLWRNORM);
2145 BPF_PID_REFRESH(d, td);
2146 if (events & (POLLIN | POLLRDNORM)) {
2148 revents |= events & (POLLIN | POLLRDNORM);
2150 selrecord(td, &d->bd_sel);
2151 /* Start the read timeout if necessary. */
2152 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2153 callout_reset(&d->bd_callout, d->bd_rtout,
2155 d->bd_state = BPF_WAITING;
2164 * Support for kevent() system call. Register EVFILT_READ filters and
2165 * reject all others.
2168 bpfkqfilter(struct cdev *dev, struct knote *kn)
2172 if (devfs_get_cdevpriv((void **)&d) != 0)
2175 switch (kn->kn_filter) {
2177 kn->kn_fop = &bpfread_filtops;
2181 kn->kn_fop = &bpfwrite_filtops;
2189 * Refresh PID associated with this descriptor.
2192 BPF_PID_REFRESH_CUR(d);
2194 knlist_add(&d->bd_sel.si_note, kn, 1);
2201 filt_bpfdetach(struct knote *kn)
2203 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2205 knlist_remove(&d->bd_sel.si_note, kn, 0);
2209 filt_bpfread(struct knote *kn, long hint)
2211 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2214 BPFD_LOCK_ASSERT(d);
2215 ready = bpf_ready(d);
2217 kn->kn_data = d->bd_slen;
2219 * Ignore the hold buffer if it is being copied to user space.
2221 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2222 kn->kn_data += d->bd_hlen;
2223 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2224 callout_reset(&d->bd_callout, d->bd_rtout,
2226 d->bd_state = BPF_WAITING;
2233 filt_bpfwrite(struct knote *kn, long hint)
2235 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2237 BPFD_LOCK_ASSERT(d);
2239 if (d->bd_bif == NULL) {
2243 kn->kn_data = d->bd_bif->bif_ifp->if_mtu;
2248 #define BPF_TSTAMP_NONE 0
2249 #define BPF_TSTAMP_FAST 1
2250 #define BPF_TSTAMP_NORMAL 2
2251 #define BPF_TSTAMP_EXTERN 3
2254 bpf_ts_quality(int tstype)
2257 if (tstype == BPF_T_NONE)
2258 return (BPF_TSTAMP_NONE);
2259 if ((tstype & BPF_T_FAST) != 0)
2260 return (BPF_TSTAMP_FAST);
2262 return (BPF_TSTAMP_NORMAL);
2266 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2271 quality = bpf_ts_quality(tstype);
2272 if (quality == BPF_TSTAMP_NONE)
2276 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2278 *bt = *(struct bintime *)(tag + 1);
2279 return (BPF_TSTAMP_EXTERN);
2282 if (quality == BPF_TSTAMP_NORMAL)
2291 * Incoming linkage from device drivers. Process the packet pkt, of length
2292 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2293 * by each process' filter, and if accepted, stashed into the corresponding
2297 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2299 struct epoch_tracker et;
2308 gottime = BPF_TSTAMP_NONE;
2309 NET_EPOCH_ENTER(et);
2310 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2311 counter_u64_add(d->bd_rcount, 1);
2313 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2314 * is no way for the caller to indiciate to us whether this
2315 * packet is inbound or outbound. In the bpf_mtap() routines,
2316 * we use the interface pointers on the mbuf to figure it out.
2319 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2321 slen = (*(bf->func))(pkt, pktlen, pktlen);
2324 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2327 * Filter matches. Let's to acquire write lock.
2330 counter_u64_add(d->bd_fcount, 1);
2331 if (gottime < bpf_ts_quality(d->bd_tstamp))
2332 gottime = bpf_gettime(&bt, d->bd_tstamp,
2335 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2337 catchpacket(d, pkt, pktlen, slen,
2338 bpf_append_bytes, &bt);
2345 #define BPF_CHECK_DIRECTION(d, r, i) \
2346 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2347 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2350 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2351 * Locking model is explained in bpf_tap().
2354 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2356 struct epoch_tracker et;
2365 /* Skip outgoing duplicate packets. */
2366 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2367 m->m_flags &= ~M_PROMISC;
2371 pktlen = m_length(m, NULL);
2372 gottime = BPF_TSTAMP_NONE;
2374 NET_EPOCH_ENTER(et);
2375 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2376 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2378 counter_u64_add(d->bd_rcount, 1);
2380 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2381 /* XXX We cannot handle multiple mbufs. */
2382 if (bf != NULL && m->m_next == NULL)
2383 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2387 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2391 counter_u64_add(d->bd_fcount, 1);
2392 if (gottime < bpf_ts_quality(d->bd_tstamp))
2393 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2395 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2397 catchpacket(d, (u_char *)m, pktlen, slen,
2398 bpf_append_mbuf, &bt);
2406 * Incoming linkage from device drivers, when packet is in
2407 * an mbuf chain and to be prepended by a contiguous header.
2410 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2412 struct epoch_tracker et;
2419 /* Skip outgoing duplicate packets. */
2420 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2421 m->m_flags &= ~M_PROMISC;
2425 pktlen = m_length(m, NULL);
2427 * Craft on-stack mbuf suitable for passing to bpf_filter.
2428 * Note that we cut corners here; we only setup what's
2429 * absolutely needed--this mbuf should never go anywhere else.
2437 gottime = BPF_TSTAMP_NONE;
2439 NET_EPOCH_ENTER(et);
2440 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2441 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2443 counter_u64_add(d->bd_rcount, 1);
2444 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2448 counter_u64_add(d->bd_fcount, 1);
2449 if (gottime < bpf_ts_quality(d->bd_tstamp))
2450 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2452 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2454 catchpacket(d, (u_char *)&mb, pktlen, slen,
2455 bpf_append_mbuf, &bt);
2462 #undef BPF_CHECK_DIRECTION
2463 #undef BPF_TSTAMP_NONE
2464 #undef BPF_TSTAMP_FAST
2465 #undef BPF_TSTAMP_NORMAL
2466 #undef BPF_TSTAMP_EXTERN
2469 bpf_hdrlen(struct bpf_d *d)
2473 hdrlen = d->bd_bif->bif_hdrlen;
2474 #ifndef BURN_BRIDGES
2475 if (d->bd_tstamp == BPF_T_NONE ||
2476 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2477 #ifdef COMPAT_FREEBSD32
2479 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2482 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2485 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2486 #ifdef COMPAT_FREEBSD32
2488 hdrlen = BPF_WORDALIGN32(hdrlen);
2491 hdrlen = BPF_WORDALIGN(hdrlen);
2493 return (hdrlen - d->bd_bif->bif_hdrlen);
2497 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2499 struct bintime bt2, boottimebin;
2501 struct timespec tsn;
2503 if ((tstype & BPF_T_MONOTONIC) == 0) {
2505 getboottimebin(&boottimebin);
2506 bintime_add(&bt2, &boottimebin);
2509 switch (BPF_T_FORMAT(tstype)) {
2510 case BPF_T_MICROTIME:
2511 bintime2timeval(bt, &tsm);
2512 ts->bt_sec = tsm.tv_sec;
2513 ts->bt_frac = tsm.tv_usec;
2515 case BPF_T_NANOTIME:
2516 bintime2timespec(bt, &tsn);
2517 ts->bt_sec = tsn.tv_sec;
2518 ts->bt_frac = tsn.tv_nsec;
2521 ts->bt_sec = bt->sec;
2522 ts->bt_frac = bt->frac;
2528 * Move the packet data from interface memory (pkt) into the
2529 * store buffer. "cpfn" is the routine called to do the actual data
2530 * transfer. bcopy is passed in to copy contiguous chunks, while
2531 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2532 * pkt is really an mbuf.
2535 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2536 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2539 struct bpf_xhdr hdr;
2540 #ifndef BURN_BRIDGES
2541 struct bpf_hdr hdr_old;
2542 #ifdef COMPAT_FREEBSD32
2543 struct bpf_hdr32 hdr32_old;
2546 int caplen, curlen, hdrlen, totlen;
2551 BPFD_LOCK_ASSERT(d);
2552 if (d->bd_bif == NULL) {
2553 /* Descriptor was detached in concurrent thread */
2554 counter_u64_add(d->bd_dcount, 1);
2559 * Detect whether user space has released a buffer back to us, and if
2560 * so, move it from being a hold buffer to a free buffer. This may
2561 * not be the best place to do it (for example, we might only want to
2562 * run this check if we need the space), but for now it's a reliable
2565 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2566 d->bd_fbuf = d->bd_hbuf;
2569 bpf_buf_reclaimed(d);
2573 * Figure out how many bytes to move. If the packet is
2574 * greater or equal to the snapshot length, transfer that
2575 * much. Otherwise, transfer the whole packet (unless
2576 * we hit the buffer size limit).
2578 hdrlen = bpf_hdrlen(d);
2579 totlen = hdrlen + min(snaplen, pktlen);
2580 if (totlen > d->bd_bufsize)
2581 totlen = d->bd_bufsize;
2584 * Round up the end of the previous packet to the next longword.
2586 * Drop the packet if there's no room and no hope of room
2587 * If the packet would overflow the storage buffer or the storage
2588 * buffer is considered immutable by the buffer model, try to rotate
2589 * the buffer and wakeup pending processes.
2591 #ifdef COMPAT_FREEBSD32
2593 curlen = BPF_WORDALIGN32(d->bd_slen);
2596 curlen = BPF_WORDALIGN(d->bd_slen);
2597 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2598 if (d->bd_fbuf == NULL) {
2600 * There's no room in the store buffer, and no
2601 * prospect of room, so drop the packet. Notify the
2605 counter_u64_add(d->bd_dcount, 1);
2608 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2612 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2614 * Immediate mode is set, or the read timeout has already
2615 * expired during a select call. A packet arrived, so the
2616 * reader should be woken up.
2619 caplen = totlen - hdrlen;
2620 tstype = d->bd_tstamp;
2621 do_timestamp = tstype != BPF_T_NONE;
2622 #ifndef BURN_BRIDGES
2623 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2626 bpf_bintime2ts(bt, &ts, tstype);
2627 #ifdef COMPAT_FREEBSD32
2628 if (d->bd_compat32) {
2629 bzero(&hdr32_old, sizeof(hdr32_old));
2631 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2632 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2634 hdr32_old.bh_datalen = pktlen;
2635 hdr32_old.bh_hdrlen = hdrlen;
2636 hdr32_old.bh_caplen = caplen;
2637 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2642 bzero(&hdr_old, sizeof(hdr_old));
2644 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2645 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2647 hdr_old.bh_datalen = pktlen;
2648 hdr_old.bh_hdrlen = hdrlen;
2649 hdr_old.bh_caplen = caplen;
2650 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2657 * Append the bpf header. Note we append the actual header size, but
2658 * move forward the length of the header plus padding.
2660 bzero(&hdr, sizeof(hdr));
2662 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2663 hdr.bh_datalen = pktlen;
2664 hdr.bh_hdrlen = hdrlen;
2665 hdr.bh_caplen = caplen;
2666 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2669 * Copy the packet data into the store buffer and update its length.
2671 #ifndef BURN_BRIDGES
2674 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2675 d->bd_slen = curlen + totlen;
2682 * Free buffers currently in use by a descriptor.
2686 bpfd_free(epoch_context_t ctx)
2689 struct bpf_program_buffer *p;
2692 * We don't need to lock out interrupts since this descriptor has
2693 * been detached from its interface and it yet hasn't been marked
2696 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2698 if (d->bd_rfilter != NULL) {
2699 p = __containerof((void *)d->bd_rfilter,
2700 struct bpf_program_buffer, buffer);
2702 p->func = d->bd_bfilter;
2704 bpf_program_buffer_free(&p->epoch_ctx);
2706 if (d->bd_wfilter != NULL) {
2707 p = __containerof((void *)d->bd_wfilter,
2708 struct bpf_program_buffer, buffer);
2712 bpf_program_buffer_free(&p->epoch_ctx);
2715 mtx_destroy(&d->bd_lock);
2716 counter_u64_free(d->bd_rcount);
2717 counter_u64_free(d->bd_dcount);
2718 counter_u64_free(d->bd_fcount);
2719 counter_u64_free(d->bd_wcount);
2720 counter_u64_free(d->bd_wfcount);
2721 counter_u64_free(d->bd_wdcount);
2722 counter_u64_free(d->bd_zcopy);
2727 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2728 * fixed size of the link header (variable length headers not yet supported).
2731 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2734 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2738 * Attach an interface to bpf. ifp is a pointer to the structure
2739 * defining the interface to be attached, dlt is the link layer type,
2740 * and hdrlen is the fixed size of the link header (variable length
2741 * headers are not yet supporrted).
2744 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2745 struct bpf_if **driverp)
2749 KASSERT(*driverp == NULL,
2750 ("bpfattach2: driverp already initialized"));
2752 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2754 CK_LIST_INIT(&bp->bif_dlist);
2755 CK_LIST_INIT(&bp->bif_wlist);
2758 bp->bif_hdrlen = hdrlen;
2759 bp->bif_bpf = driverp;
2763 * Reference ifnet pointer, so it won't freed until
2768 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2771 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2772 if_printf(ifp, "bpf attached\n");
2777 * When moving interfaces between vnet instances we need a way to
2778 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2779 * after the vmove. We unfortunately have no device driver infrastructure
2780 * to query the interface for these values after creation/attach, thus
2781 * add this as a workaround.
2784 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2789 if (bif_dlt == NULL && bif_hdrlen == NULL)
2792 if (bif_dlt != NULL)
2793 *bif_dlt = bp->bif_dlt;
2794 if (bif_hdrlen != NULL)
2795 *bif_hdrlen = bp->bif_hdrlen;
2802 * Detach bpf from an interface. This involves detaching each descriptor
2803 * associated with the interface. Notify each descriptor as it's detached
2804 * so that any sleepers wake up and get ENXIO.
2807 bpfdetach(struct ifnet *ifp)
2809 struct bpf_if *bp, *bp_temp;
2813 /* Find all bpf_if struct's which reference ifp and detach them. */
2814 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2815 if (ifp != bp->bif_ifp)
2818 CK_LIST_REMOVE(bp, bif_next);
2819 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2822 "%s: sheduling free for encap %d (%p) for if %p",
2823 __func__, bp->bif_dlt, bp, ifp);
2825 /* Detach common descriptors */
2826 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2827 bpf_detachd_locked(d, true);
2830 /* Detach writer-only descriptors */
2831 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2832 bpf_detachd_locked(d, true);
2840 * Get a list of available data link type of the interface.
2843 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2852 ifp = d->bd_bif->bif_ifp;
2854 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2855 if (bp->bif_ifp == ifp)
2858 if (bfl->bfl_list == NULL) {
2862 if (n1 > bfl->bfl_len)
2865 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2867 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2868 if (bp->bif_ifp != ifp)
2870 lst[n++] = bp->bif_dlt;
2872 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2879 * Set the data link type of a BPF instance.
2882 bpf_setdlt(struct bpf_d *d, u_int dlt)
2884 int error, opromisc;
2889 MPASS(d->bd_bif != NULL);
2892 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2893 * changed while we hold global lock.
2895 if (d->bd_bif->bif_dlt == dlt)
2898 ifp = d->bd_bif->bif_ifp;
2899 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2900 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2906 opromisc = d->bd_promisc;
2909 error = ifpromisc(bp->bif_ifp, 1);
2911 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2920 bpf_drvinit(void *unused)
2924 sx_init(&bpf_sx, "bpf global lock");
2925 CK_LIST_INIT(&bpf_iflist);
2927 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2928 /* For compatibility */
2929 make_dev_alias(dev, "bpf0");
2933 * Zero out the various packet counters associated with all of the bpf
2934 * descriptors. At some point, we will probably want to get a bit more
2935 * granular and allow the user to specify descriptors to be zeroed.
2938 bpf_zero_counters(void)
2945 * We are protected by global lock here, interfaces and
2946 * descriptors can not be deleted while we hold it.
2948 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2949 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2950 counter_u64_zero(bd->bd_rcount);
2951 counter_u64_zero(bd->bd_dcount);
2952 counter_u64_zero(bd->bd_fcount);
2953 counter_u64_zero(bd->bd_wcount);
2954 counter_u64_zero(bd->bd_wfcount);
2955 counter_u64_zero(bd->bd_zcopy);
2962 * Fill filter statistics
2965 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2969 bzero(d, sizeof(*d));
2970 d->bd_structsize = sizeof(*d);
2971 d->bd_immediate = bd->bd_immediate;
2972 d->bd_promisc = bd->bd_promisc;
2973 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2974 d->bd_direction = bd->bd_direction;
2975 d->bd_feedback = bd->bd_feedback;
2976 d->bd_async = bd->bd_async;
2977 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2978 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2979 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2980 d->bd_sig = bd->bd_sig;
2981 d->bd_slen = bd->bd_slen;
2982 d->bd_hlen = bd->bd_hlen;
2983 d->bd_bufsize = bd->bd_bufsize;
2984 d->bd_pid = bd->bd_pid;
2985 strlcpy(d->bd_ifname,
2986 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2987 d->bd_locked = bd->bd_locked;
2988 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2989 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2990 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2991 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2992 d->bd_bufmode = bd->bd_bufmode;
2996 * Handle `netstat -B' stats request
2999 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
3001 static const struct xbpf_d zerostats;
3002 struct xbpf_d *xbdbuf, *xbd, tempstats;
3008 * XXX This is not technically correct. It is possible for non
3009 * privileged users to open bpf devices. It would make sense
3010 * if the users who opened the devices were able to retrieve
3011 * the statistics for them, too.
3013 error = priv_check(req->td, PRIV_NET_BPF);
3017 * Check to see if the user is requesting that the counters be
3018 * zeroed out. Explicitly check that the supplied data is zeroed,
3019 * as we aren't allowing the user to set the counters currently.
3021 if (req->newptr != NULL) {
3022 if (req->newlen != sizeof(tempstats))
3024 memset(&tempstats, 0, sizeof(tempstats));
3025 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3028 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3030 bpf_zero_counters();
3033 if (req->oldptr == NULL)
3034 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3035 if (bpf_bpfd_cnt == 0)
3036 return (SYSCTL_OUT(req, 0, 0));
3037 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3039 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3041 free(xbdbuf, M_BPF);
3045 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3046 /* Send writers-only first */
3047 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3048 xbd = &xbdbuf[index++];
3049 bpfstats_fill_xbpf(xbd, bd);
3051 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3052 xbd = &xbdbuf[index++];
3053 bpfstats_fill_xbpf(xbd, bd);
3057 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3058 free(xbdbuf, M_BPF);
3062 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3064 #else /* !DEV_BPF && !NETGRAPH_BPF */
3067 * NOP stubs to allow bpf-using drivers to load and function.
3069 * A 'better' implementation would allow the core bpf functionality
3070 * to be loaded at runtime.
3074 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3079 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3084 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3089 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3092 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3096 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3099 *driverp = (struct bpf_if *)&dead_bpf_if;
3103 bpfdetach(struct ifnet *ifp)
3108 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3110 return -1; /* "no filter" behaviour */
3114 bpf_validate(const struct bpf_insn *f, int len)
3116 return 0; /* false */
3119 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3123 bpf_show_bpf_if(struct bpf_if *bpf_if)
3128 db_printf("%p:\n", bpf_if);
3129 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3130 /* bif_ext.bif_next */
3131 /* bif_ext.bif_dlist */
3132 BPF_DB_PRINTF("%#x", bif_dlt);
3133 BPF_DB_PRINTF("%u", bif_hdrlen);
3135 BPF_DB_PRINTF("%p", bif_ifp);
3136 BPF_DB_PRINTF("%p", bif_bpf);
3137 BPF_DB_PRINTF("%u", bif_refcnt);
3140 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3144 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3148 bpf_show_bpf_if((struct bpf_if *)addr);