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 void bpf_drvinit(void *);
217 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
219 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
221 int bpf_maxinsns = BPF_MAXINSNS;
222 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
223 &bpf_maxinsns, 0, "Maximum bpf program instructions");
224 static int bpf_zerocopy_enable = 0;
225 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
226 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
227 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
228 bpf_stats_sysctl, "bpf statistics portal");
230 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
231 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
232 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
233 &VNET_NAME(bpf_optimize_writers), 0,
234 "Do not send packets until BPF program is set");
236 static d_open_t bpfopen;
237 static d_read_t bpfread;
238 static d_write_t bpfwrite;
239 static d_ioctl_t bpfioctl;
240 static d_poll_t bpfpoll;
241 static d_kqfilter_t bpfkqfilter;
243 static struct cdevsw bpf_cdevsw = {
244 .d_version = D_VERSION,
251 .d_kqfilter = bpfkqfilter,
254 static struct filterops bpfread_filtops = {
256 .f_detach = filt_bpfdetach,
257 .f_event = filt_bpfread,
261 * LOCKING MODEL USED BY BPF
264 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
265 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
266 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
267 * structure fields used by bpf_*tap* code.
269 * Lock order: global lock, then descriptor lock.
271 * There are several possible consumers:
273 * 1. The kernel registers interface pointer with bpfattach().
274 * Each call allocates new bpf_if structure, references ifnet pointer
275 * and links bpf_if into bpf_iflist chain. This is protected with global
278 * 2. An userland application uses ioctl() call to bpf_d descriptor.
279 * All such call are serialized with global lock. BPF filters can be
280 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
281 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
282 * filter pointers, even if change will happen during bpf_tap execution.
283 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
285 * 3. An userland application can write packets into bpf_d descriptor.
286 * There we need to be sure, that ifnet won't disappear during bpfwrite().
288 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
289 * bif_dlist is protected with net_epoch_preempt section. So, it should
290 * be safe to make access to bpf_d descriptor inside the section.
292 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
293 * are modified with global lock held and actual free() is done using
298 bpfif_free(epoch_context_t ctx)
302 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
303 if_rele(bp->bif_ifp);
308 bpfif_ref(struct bpf_if *bp)
311 refcount_acquire(&bp->bif_refcnt);
315 bpfif_rele(struct bpf_if *bp)
318 if (!refcount_release(&bp->bif_refcnt))
320 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
324 bpfd_ref(struct bpf_d *d)
327 refcount_acquire(&d->bd_refcnt);
331 bpfd_rele(struct bpf_d *d)
334 if (!refcount_release(&d->bd_refcnt))
336 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
339 static struct bpf_program_buffer*
340 bpf_program_buffer_alloc(size_t size, int flags)
343 return (malloc(sizeof(struct bpf_program_buffer) + size,
348 bpf_program_buffer_free(epoch_context_t ctx)
350 struct bpf_program_buffer *ptr;
352 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
354 if (ptr->func != NULL)
355 bpf_destroy_jit_filter(ptr->func);
361 * Wrapper functions for various buffering methods. If the set of buffer
362 * modes expands, we will probably want to introduce a switch data structure
363 * similar to protosw, et.
366 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
372 switch (d->bd_bufmode) {
373 case BPF_BUFMODE_BUFFER:
374 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
376 case BPF_BUFMODE_ZBUF:
377 counter_u64_add(d->bd_zcopy, 1);
378 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
381 panic("bpf_buf_append_bytes");
386 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
392 switch (d->bd_bufmode) {
393 case BPF_BUFMODE_BUFFER:
394 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
396 case BPF_BUFMODE_ZBUF:
397 counter_u64_add(d->bd_zcopy, 1);
398 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
401 panic("bpf_buf_append_mbuf");
406 * This function gets called when the free buffer is re-assigned.
409 bpf_buf_reclaimed(struct bpf_d *d)
414 switch (d->bd_bufmode) {
415 case BPF_BUFMODE_BUFFER:
418 case BPF_BUFMODE_ZBUF:
419 bpf_zerocopy_buf_reclaimed(d);
423 panic("bpf_buf_reclaimed");
428 * If the buffer mechanism has a way to decide that a held buffer can be made
429 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
430 * returned if the buffer can be discarded, (0) is returned if it cannot.
433 bpf_canfreebuf(struct bpf_d *d)
438 switch (d->bd_bufmode) {
439 case BPF_BUFMODE_ZBUF:
440 return (bpf_zerocopy_canfreebuf(d));
446 * Allow the buffer model to indicate that the current store buffer is
447 * immutable, regardless of the appearance of space. Return (1) if the
448 * buffer is writable, and (0) if not.
451 bpf_canwritebuf(struct bpf_d *d)
455 switch (d->bd_bufmode) {
456 case BPF_BUFMODE_ZBUF:
457 return (bpf_zerocopy_canwritebuf(d));
463 * Notify buffer model that an attempt to write to the store buffer has
464 * resulted in a dropped packet, in which case the buffer may be considered
468 bpf_buffull(struct bpf_d *d)
473 switch (d->bd_bufmode) {
474 case BPF_BUFMODE_ZBUF:
475 bpf_zerocopy_buffull(d);
481 * Notify the buffer model that a buffer has moved into the hold position.
484 bpf_bufheld(struct bpf_d *d)
489 switch (d->bd_bufmode) {
490 case BPF_BUFMODE_ZBUF:
491 bpf_zerocopy_bufheld(d);
497 bpf_free(struct bpf_d *d)
500 switch (d->bd_bufmode) {
501 case BPF_BUFMODE_BUFFER:
502 return (bpf_buffer_free(d));
504 case BPF_BUFMODE_ZBUF:
505 return (bpf_zerocopy_free(d));
508 panic("bpf_buf_free");
513 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
516 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
518 return (bpf_buffer_uiomove(d, buf, len, uio));
522 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
525 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
527 return (bpf_buffer_ioctl_sblen(d, i));
531 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
534 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
536 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
540 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
543 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
545 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
549 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
552 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
554 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
558 * General BPF functions.
561 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
562 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
564 const struct ieee80211_bpf_params *p;
565 struct ether_header *eh;
573 * Build a sockaddr based on the data link layer type.
574 * We do this at this level because the ethernet header
575 * is copied directly into the data field of the sockaddr.
576 * In the case of SLIP, there is no header and the packet
577 * is forwarded as is.
578 * Also, we are careful to leave room at the front of the mbuf
579 * for the link level header.
583 sockp->sa_family = AF_INET;
588 sockp->sa_family = AF_UNSPEC;
589 /* XXX Would MAXLINKHDR be better? */
590 hlen = ETHER_HDR_LEN;
594 sockp->sa_family = AF_IMPLINK;
599 sockp->sa_family = AF_UNSPEC;
605 * null interface types require a 4 byte pseudo header which
606 * corresponds to the address family of the packet.
608 sockp->sa_family = AF_UNSPEC;
612 case DLT_ATM_RFC1483:
614 * en atm driver requires 4-byte atm pseudo header.
615 * though it isn't standard, vpi:vci needs to be
618 sockp->sa_family = AF_UNSPEC;
619 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
623 sockp->sa_family = AF_UNSPEC;
624 hlen = 4; /* This should match PPP_HDRLEN */
627 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
628 sockp->sa_family = AF_IEEE80211;
632 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
633 sockp->sa_family = AF_IEEE80211;
634 sockp->sa_len = 12; /* XXX != 0 */
635 hlen = sizeof(struct ieee80211_bpf_params);
642 len = uio->uio_resid;
643 if (len < hlen || len - hlen > ifp->if_mtu)
646 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
649 m->m_pkthdr.len = m->m_len = len;
652 error = uiomove(mtod(m, u_char *), len, uio);
656 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
662 /* Check for multicast destination */
665 eh = mtod(m, struct ether_header *);
666 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
667 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
668 ETHER_ADDR_LEN) == 0)
669 m->m_flags |= M_BCAST;
671 m->m_flags |= M_MCAST;
673 if (d->bd_hdrcmplt == 0) {
674 memcpy(eh->ether_shost, IF_LLADDR(ifp),
675 sizeof(eh->ether_shost));
681 * Make room for link header, and copy it to sockaddr
684 if (sockp->sa_family == AF_IEEE80211) {
686 * Collect true length from the parameter header
687 * NB: sockp is known to be zero'd so if we do a
688 * short copy unspecified parameters will be
690 * NB: packet may not be aligned after stripping
694 p = mtod(m, const struct ieee80211_bpf_params *);
696 if (hlen > sizeof(sockp->sa_data)) {
701 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
712 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
713 * then reset its buffers and counters with reset_d().
716 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
723 * Save sysctl value to protect from sysctl change
726 op_w = V_bpf_optimize_writers || d->bd_writer;
728 if (d->bd_bif != NULL)
729 bpf_detachd_locked(d, false);
731 * Point d at bp, and add d to the interface's list.
732 * Since there are many applications using BPF for
733 * sending raw packets only (dhcpd, cdpd are good examples)
734 * we can delay adding d to the list of active listeners until
735 * some filter is configured.
740 * Hold reference to bpif while descriptor uses this interface.
745 /* Add to writers-only list */
746 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
748 * We decrement bd_writer on every filter set operation.
749 * First BIOCSETF is done by pcap_open_live() to set up
750 * snap length. After that appliation usually sets its own
755 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
761 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
762 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
765 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
769 * Check if we need to upgrade our descriptor @d from write-only mode.
772 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
775 int is_snap, need_upgrade;
778 * Check if we've already upgraded or new filter is empty.
780 if (d->bd_writer == 0 || fcode == NULL)
786 * Check if cmd looks like snaplen setting from
787 * pcap_bpf.c:pcap_open_live().
788 * Note we're not checking .k value here:
789 * while pcap_open_live() definitely sets to non-zero value,
790 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
791 * do not consider upgrading immediately
793 if (cmd == BIOCSETF && flen == 1 &&
794 fcode[0].code == (BPF_RET | BPF_K))
801 * We're setting first filter and it doesn't look like
802 * setting snaplen. We're probably using bpf directly.
803 * Upgrade immediately.
808 * Do not require upgrade by first BIOCSETF
809 * (used to set snaplen) by pcap_open_live().
812 if (--d->bd_writer == 0) {
814 * First snaplen filter has already
815 * been set. This is probably catch-all
823 "%s: filter function set by pid %d, "
824 "bd_writer counter %d, snap %d upgrade %d",
825 __func__, d->bd_pid, d->bd_writer,
826 is_snap, need_upgrade);
828 return (need_upgrade);
832 * Detach a file from its interface.
835 bpf_detachd(struct bpf_d *d)
838 bpf_detachd_locked(d, false);
843 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
850 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
852 /* Check if descriptor is attached */
853 if ((bp = d->bd_bif) == NULL)
857 /* Remove d from the interface's descriptor list. */
858 CK_LIST_REMOVE(d, bd_next);
859 /* Save bd_writer value */
860 error = d->bd_writer;
865 * Notify descriptor as it's detached, so that any
866 * sleepers wake up and get ENXIO.
873 /* Call event handler iff d is attached */
875 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
878 * Check if this descriptor had requested promiscuous mode.
879 * If so and ifnet is not detached, turn it off.
881 if (d->bd_promisc && !detached_ifp) {
883 CURVNET_SET(ifp->if_vnet);
884 error = ifpromisc(ifp, 0);
886 if (error != 0 && error != ENXIO) {
888 * ENXIO can happen if a pccard is unplugged
889 * Something is really wrong if we were able to put
890 * the driver into promiscuous mode, but can't
893 if_printf(bp->bif_ifp,
894 "bpf_detach: ifpromisc failed (%d)\n", error);
901 * Close the descriptor by detaching it from its interface,
902 * deallocating its buffers, and marking it free.
907 struct bpf_d *d = data;
910 if (d->bd_state == BPF_WAITING)
911 callout_stop(&d->bd_callout);
912 d->bd_state = BPF_IDLE;
914 funsetown(&d->bd_sigio);
917 mac_bpfdesc_destroy(d);
919 seldrain(&d->bd_sel);
920 knlist_destroy(&d->bd_sel.si_note);
921 callout_drain(&d->bd_callout);
926 * Open ethernet device. Returns ENXIO for illegal minor device number,
927 * EBUSY if file is open by another process.
931 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
936 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
937 error = devfs_set_cdevpriv(d, bpf_dtor);
944 d->bd_rcount = counter_u64_alloc(M_WAITOK);
945 d->bd_dcount = counter_u64_alloc(M_WAITOK);
946 d->bd_fcount = counter_u64_alloc(M_WAITOK);
947 d->bd_wcount = counter_u64_alloc(M_WAITOK);
948 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
949 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
950 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
953 * For historical reasons, perform a one-time initialization call to
954 * the buffer routines, even though we're not yet committed to a
955 * particular buffer method.
958 if ((flags & FREAD) == 0)
960 d->bd_hbuf_in_use = 0;
961 d->bd_bufmode = BPF_BUFMODE_BUFFER;
963 d->bd_direction = BPF_D_INOUT;
965 BPF_PID_REFRESH(d, td);
968 mac_bpfdesc_create(td->td_ucred, d);
970 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
971 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
972 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
974 /* Disable VLAN pcp tagging. */
981 * bpfread - read next chunk of packets from buffers
984 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
991 error = devfs_get_cdevpriv((void **)&d);
996 * Restrict application to use a buffer the same size as
999 if (uio->uio_resid != d->bd_bufsize)
1002 non_block = ((ioflag & O_NONBLOCK) != 0);
1005 BPF_PID_REFRESH_CUR(d);
1006 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1008 return (EOPNOTSUPP);
1010 if (d->bd_state == BPF_WAITING)
1011 callout_stop(&d->bd_callout);
1012 timed_out = (d->bd_state == BPF_TIMED_OUT);
1013 d->bd_state = BPF_IDLE;
1014 while (d->bd_hbuf_in_use) {
1015 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1016 PRINET|PCATCH, "bd_hbuf", 0);
1023 * If the hold buffer is empty, then do a timed sleep, which
1024 * ends when the timeout expires or when enough packets
1025 * have arrived to fill the store buffer.
1027 while (d->bd_hbuf == NULL) {
1028 if (d->bd_slen != 0) {
1030 * A packet(s) either arrived since the previous
1031 * read or arrived while we were asleep.
1033 if (d->bd_immediate || non_block || timed_out) {
1035 * Rotate the buffers and return what's here
1036 * if we are in immediate mode, non-blocking
1037 * flag is set, or this descriptor timed out.
1045 * No data is available, check to see if the bpf device
1046 * is still pointed at a real interface. If not, return
1047 * ENXIO so that the userland process knows to rebind
1048 * it before using it again.
1050 if (d->bd_bif == NULL) {
1057 return (EWOULDBLOCK);
1059 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1060 "bpf", d->bd_rtout);
1061 if (error == EINTR || error == ERESTART) {
1065 if (error == EWOULDBLOCK) {
1067 * On a timeout, return what's in the buffer,
1068 * which may be nothing. If there is something
1069 * in the store buffer, we can rotate the buffers.
1073 * We filled up the buffer in between
1074 * getting the timeout and arriving
1075 * here, so we don't need to rotate.
1079 if (d->bd_slen == 0) {
1088 * At this point, we know we have something in the hold slot.
1090 d->bd_hbuf_in_use = 1;
1094 * Move data from hold buffer into user space.
1095 * We know the entire buffer is transferred since
1096 * we checked above that the read buffer is bpf_bufsize bytes.
1098 * We do not have to worry about simultaneous reads because
1099 * we waited for sole access to the hold buffer above.
1101 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1104 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1105 d->bd_fbuf = d->bd_hbuf;
1108 bpf_buf_reclaimed(d);
1109 d->bd_hbuf_in_use = 0;
1110 wakeup(&d->bd_hbuf_in_use);
1117 * If there are processes sleeping on this descriptor, wake them up.
1119 static __inline void
1120 bpf_wakeup(struct bpf_d *d)
1123 BPFD_LOCK_ASSERT(d);
1124 if (d->bd_state == BPF_WAITING) {
1125 callout_stop(&d->bd_callout);
1126 d->bd_state = BPF_IDLE;
1129 if (d->bd_async && d->bd_sig && d->bd_sigio)
1130 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1132 selwakeuppri(&d->bd_sel, PRINET);
1133 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1137 bpf_timed_out(void *arg)
1139 struct bpf_d *d = (struct bpf_d *)arg;
1141 BPFD_LOCK_ASSERT(d);
1143 if (callout_pending(&d->bd_callout) ||
1144 !callout_active(&d->bd_callout))
1146 if (d->bd_state == BPF_WAITING) {
1147 d->bd_state = BPF_TIMED_OUT;
1148 if (d->bd_slen != 0)
1154 bpf_ready(struct bpf_d *d)
1157 BPFD_LOCK_ASSERT(d);
1159 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1161 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1168 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1171 struct sockaddr dst;
1172 struct epoch_tracker et;
1176 struct mbuf *m, *mc;
1179 error = devfs_get_cdevpriv((void **)&d);
1183 NET_EPOCH_ENTER(et);
1185 BPF_PID_REFRESH_CUR(d);
1186 counter_u64_add(d->bd_wcount, 1);
1187 if ((bp = d->bd_bif) == NULL) {
1193 if ((ifp->if_flags & IFF_UP) == 0) {
1198 if (uio->uio_resid == 0)
1201 bzero(&dst, sizeof(dst));
1206 * Take extra reference, unlock d and exit from epoch section,
1207 * since bpf_movein() can sleep.
1213 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1214 &m, &dst, &hlen, d);
1217 counter_u64_add(d->bd_wdcount, 1);
1224 * Check that descriptor is still attached to the interface.
1225 * This can happen on bpfdetach(). To avoid access to detached
1226 * ifnet, free mbuf and return ENXIO.
1228 if (d->bd_bif == NULL) {
1229 counter_u64_add(d->bd_wdcount, 1);
1235 counter_u64_add(d->bd_wfcount, 1);
1237 dst.sa_family = pseudo_AF_HDRCMPLT;
1239 if (d->bd_feedback) {
1240 mc = m_dup(m, M_NOWAIT);
1242 mc->m_pkthdr.rcvif = ifp;
1243 /* Set M_PROMISC for outgoing packets to be discarded. */
1244 if (d->bd_direction == BPF_D_INOUT)
1245 m->m_flags |= M_PROMISC;
1249 m->m_pkthdr.len -= hlen;
1251 m->m_data += hlen; /* XXX */
1253 CURVNET_SET(ifp->if_vnet);
1255 mac_bpfdesc_create_mbuf(d, m);
1257 mac_bpfdesc_create_mbuf(d, mc);
1260 bzero(&ro, sizeof(ro));
1262 ro.ro_prepend = (u_char *)&dst.sa_data;
1264 ro.ro_flags = RT_HAS_HEADER;
1268 vlan_set_pcp(m, d->bd_pcp);
1270 /* Avoid possible recursion on BPFD_LOCK(). */
1271 NET_EPOCH_ENTER(et);
1273 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1275 counter_u64_add(d->bd_wdcount, 1);
1279 (*ifp->if_input)(ifp, mc);
1289 counter_u64_add(d->bd_wdcount, 1);
1296 * Reset a descriptor by flushing its packet buffer and clearing the receive
1297 * and drop counts. This is doable for kernel-only buffers, but with
1298 * zero-copy buffers, we can't write to (or rotate) buffers that are
1299 * currently owned by userspace. It would be nice if we could encapsulate
1300 * this logic in the buffer code rather than here.
1303 reset_d(struct bpf_d *d)
1306 BPFD_LOCK_ASSERT(d);
1308 while (d->bd_hbuf_in_use)
1309 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1311 if ((d->bd_hbuf != NULL) &&
1312 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1313 /* Free the hold buffer. */
1314 d->bd_fbuf = d->bd_hbuf;
1317 bpf_buf_reclaimed(d);
1319 if (bpf_canwritebuf(d))
1321 counter_u64_zero(d->bd_rcount);
1322 counter_u64_zero(d->bd_dcount);
1323 counter_u64_zero(d->bd_fcount);
1324 counter_u64_zero(d->bd_wcount);
1325 counter_u64_zero(d->bd_wfcount);
1326 counter_u64_zero(d->bd_wdcount);
1327 counter_u64_zero(d->bd_zcopy);
1331 * FIONREAD Check for read packet available.
1332 * BIOCGBLEN Get buffer len [for read()].
1333 * BIOCSETF Set read filter.
1334 * BIOCSETFNR Set read filter without resetting descriptor.
1335 * BIOCSETWF Set write filter.
1336 * BIOCFLUSH Flush read packet buffer.
1337 * BIOCPROMISC Put interface into promiscuous mode.
1338 * BIOCGDLT Get link layer type.
1339 * BIOCGETIF Get interface name.
1340 * BIOCSETIF Set interface.
1341 * BIOCSRTIMEOUT Set read timeout.
1342 * BIOCGRTIMEOUT Get read timeout.
1343 * BIOCGSTATS Get packet stats.
1344 * BIOCIMMEDIATE Set immediate mode.
1345 * BIOCVERSION Get filter language version.
1346 * BIOCGHDRCMPLT Get "header already complete" flag
1347 * BIOCSHDRCMPLT Set "header already complete" flag
1348 * BIOCGDIRECTION Get packet direction flag
1349 * BIOCSDIRECTION Set packet direction flag
1350 * BIOCGTSTAMP Get time stamp format and resolution.
1351 * BIOCSTSTAMP Set time stamp format and resolution.
1352 * BIOCLOCK Set "locked" flag
1353 * BIOCFEEDBACK Set packet feedback mode.
1354 * BIOCSETZBUF Set current zero-copy buffer locations.
1355 * BIOCGETZMAX Get maximum zero-copy buffer size.
1356 * BIOCROTZBUF Force rotation of zero-copy buffer
1357 * BIOCSETBUFMODE Set buffer mode.
1358 * BIOCGETBUFMODE Get current buffer mode.
1359 * BIOCSETVLANPCP Set VLAN PCP tag.
1363 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1369 error = devfs_get_cdevpriv((void **)&d);
1374 * Refresh PID associated with this descriptor.
1377 BPF_PID_REFRESH(d, td);
1378 if (d->bd_state == BPF_WAITING)
1379 callout_stop(&d->bd_callout);
1380 d->bd_state = BPF_IDLE;
1383 if (d->bd_locked == 1) {
1389 #ifdef COMPAT_FREEBSD32
1390 case BIOCGDLTLIST32:
1394 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1395 case BIOCGRTIMEOUT32:
1406 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1407 case BIOCSRTIMEOUT32:
1417 #ifdef COMPAT_FREEBSD32
1419 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1420 * that it will get 32-bit packet headers.
1426 case BIOCGDLTLIST32:
1427 case BIOCGRTIMEOUT32:
1428 case BIOCSRTIMEOUT32:
1429 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1437 CURVNET_SET(TD_TO_VNET(td));
1444 * Check for read packet available.
1452 while (d->bd_hbuf_in_use)
1453 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1454 PRINET, "bd_hbuf", 0);
1464 * Get buffer len [for read()].
1468 *(u_int *)addr = d->bd_bufsize;
1473 * Set buffer length.
1476 error = bpf_ioctl_sblen(d, (u_int *)addr);
1480 * Set link layer read filter.
1485 #ifdef COMPAT_FREEBSD32
1490 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1494 * Flush read packet buffer.
1503 * Put interface into promiscuous mode.
1506 if (d->bd_bif == NULL) {
1508 * No interface attached yet.
1513 if (d->bd_promisc == 0) {
1514 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1521 * Get current data link type.
1525 if (d->bd_bif == NULL)
1528 *(u_int *)addr = d->bd_bif->bif_dlt;
1533 * Get a list of supported data link types.
1535 #ifdef COMPAT_FREEBSD32
1536 case BIOCGDLTLIST32:
1538 struct bpf_dltlist32 *list32;
1539 struct bpf_dltlist dltlist;
1541 list32 = (struct bpf_dltlist32 *)addr;
1542 dltlist.bfl_len = list32->bfl_len;
1543 dltlist.bfl_list = PTRIN(list32->bfl_list);
1545 if (d->bd_bif == NULL)
1548 error = bpf_getdltlist(d, &dltlist);
1550 list32->bfl_len = dltlist.bfl_len;
1559 if (d->bd_bif == NULL)
1562 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1567 * Set data link type.
1571 if (d->bd_bif == NULL)
1574 error = bpf_setdlt(d, *(u_int *)addr);
1579 * Get interface name.
1583 if (d->bd_bif == NULL)
1586 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1587 struct ifreq *const ifr = (struct ifreq *)addr;
1589 strlcpy(ifr->ifr_name, ifp->if_xname,
1590 sizeof(ifr->ifr_name));
1600 int alloc_buf, size;
1603 * Behavior here depends on the buffering model. If
1604 * we're using kernel memory buffers, then we can
1605 * allocate them here. If we're using zero-copy,
1606 * then the user process must have registered buffers
1607 * by the time we get here.
1611 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1616 size = d->bd_bufsize;
1617 error = bpf_buffer_ioctl_sblen(d, &size);
1622 error = bpf_setif(d, (struct ifreq *)addr);
1631 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1632 case BIOCSRTIMEOUT32:
1635 struct timeval *tv = (struct timeval *)addr;
1636 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1637 struct timeval32 *tv32;
1638 struct timeval tv64;
1640 if (cmd == BIOCSRTIMEOUT32) {
1641 tv32 = (struct timeval32 *)addr;
1643 tv->tv_sec = tv32->tv_sec;
1644 tv->tv_usec = tv32->tv_usec;
1647 tv = (struct timeval *)addr;
1650 * Subtract 1 tick from tvtohz() since this isn't
1653 if ((error = itimerfix(tv)) == 0)
1654 d->bd_rtout = tvtohz(tv) - 1;
1662 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1663 case BIOCGRTIMEOUT32:
1667 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1668 struct timeval32 *tv32;
1669 struct timeval tv64;
1671 if (cmd == BIOCGRTIMEOUT32)
1675 tv = (struct timeval *)addr;
1677 tv->tv_sec = d->bd_rtout / hz;
1678 tv->tv_usec = (d->bd_rtout % hz) * tick;
1679 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1680 if (cmd == BIOCGRTIMEOUT32) {
1681 tv32 = (struct timeval32 *)addr;
1682 tv32->tv_sec = tv->tv_sec;
1683 tv32->tv_usec = tv->tv_usec;
1695 struct bpf_stat *bs = (struct bpf_stat *)addr;
1697 /* XXXCSJP overflow */
1698 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1699 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1704 * Set immediate mode.
1708 d->bd_immediate = *(u_int *)addr;
1714 struct bpf_version *bv = (struct bpf_version *)addr;
1716 bv->bv_major = BPF_MAJOR_VERSION;
1717 bv->bv_minor = BPF_MINOR_VERSION;
1722 * Get "header already complete" flag
1726 *(u_int *)addr = d->bd_hdrcmplt;
1731 * Set "header already complete" flag
1735 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1740 * Get packet direction flag
1742 case BIOCGDIRECTION:
1744 *(u_int *)addr = d->bd_direction;
1749 * Set packet direction flag
1751 case BIOCSDIRECTION:
1755 direction = *(u_int *)addr;
1756 switch (direction) {
1761 d->bd_direction = direction;
1771 * Get packet timestamp format and resolution.
1775 *(u_int *)addr = d->bd_tstamp;
1780 * Set packet timestamp format and resolution.
1786 func = *(u_int *)addr;
1787 if (BPF_T_VALID(func))
1788 d->bd_tstamp = func;
1796 d->bd_feedback = *(u_int *)addr;
1806 case FIONBIO: /* Non-blocking I/O */
1809 case FIOASYNC: /* Send signal on receive packets */
1811 d->bd_async = *(int *)addr;
1817 * XXX: Add some sort of locking here?
1818 * fsetown() can sleep.
1820 error = fsetown(*(int *)addr, &d->bd_sigio);
1825 *(int *)addr = fgetown(&d->bd_sigio);
1829 /* This is deprecated, FIOSETOWN should be used instead. */
1831 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1834 /* This is deprecated, FIOGETOWN should be used instead. */
1836 *(int *)addr = -fgetown(&d->bd_sigio);
1839 case BIOCSRSIG: /* Set receive signal */
1843 sig = *(u_int *)addr;
1856 *(u_int *)addr = d->bd_sig;
1860 case BIOCGETBUFMODE:
1862 *(u_int *)addr = d->bd_bufmode;
1866 case BIOCSETBUFMODE:
1868 * Allow the buffering mode to be changed as long as we
1869 * haven't yet committed to a particular mode. Our
1870 * definition of commitment, for now, is whether or not a
1871 * buffer has been allocated or an interface attached, since
1872 * that's the point where things get tricky.
1874 switch (*(u_int *)addr) {
1875 case BPF_BUFMODE_BUFFER:
1878 case BPF_BUFMODE_ZBUF:
1879 if (bpf_zerocopy_enable)
1889 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1890 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1895 d->bd_bufmode = *(u_int *)addr;
1900 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1904 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1908 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1911 case BIOCSETVLANPCP:
1915 pcp = *(u_int *)addr;
1916 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1929 * Set d's packet filter program to fp. If this file already has a filter,
1930 * free it and replace it. Returns EINVAL for bogus requests.
1932 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1936 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1938 #ifdef COMPAT_FREEBSD32
1939 struct bpf_program fp_swab;
1940 struct bpf_program32 *fp32;
1942 struct bpf_program_buffer *fcode;
1943 struct bpf_insn *filter;
1945 bpf_jit_filter *jfunc;
1951 #ifdef COMPAT_FREEBSD32
1956 fp32 = (struct bpf_program32 *)fp;
1957 fp_swab.bf_len = fp32->bf_len;
1959 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1978 * Check new filter validness before acquiring any locks.
1979 * Allocate memory for new filter, if needed.
1982 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1984 size = flen * sizeof(*fp->bf_insns);
1986 /* We're setting up new filter. Copy and check actual data. */
1987 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1988 filter = (struct bpf_insn *)fcode->buffer;
1989 if (copyin(fp->bf_insns, filter, size) != 0 ||
1990 !bpf_validate(filter, flen)) {
1995 if (cmd != BIOCSETWF) {
1997 * Filter is copied inside fcode and is
2000 jfunc = bpf_jitter(filter, flen);
2005 track_event = false;
2010 /* Set up new filter. */
2011 if (cmd == BIOCSETWF) {
2012 if (d->bd_wfilter != NULL) {
2013 fcode = __containerof((void *)d->bd_wfilter,
2014 struct bpf_program_buffer, buffer);
2019 d->bd_wfilter = filter;
2021 if (d->bd_rfilter != NULL) {
2022 fcode = __containerof((void *)d->bd_rfilter,
2023 struct bpf_program_buffer, buffer);
2025 fcode->func = d->bd_bfilter;
2028 d->bd_rfilter = filter;
2030 d->bd_bfilter = jfunc;
2032 if (cmd == BIOCSETF)
2035 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2037 * Filter can be set several times without
2038 * specifying interface. In this case just mark d
2042 if (d->bd_bif != NULL) {
2044 * Remove descriptor from writers-only list
2045 * and add it to active readers list.
2047 CK_LIST_REMOVE(d, bd_next);
2048 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2051 "%s: upgrade required by pid %d",
2052 __func__, d->bd_pid);
2060 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2063 EVENTHANDLER_INVOKE(bpf_track,
2064 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2071 * Detach a file from its current interface (if attached at all) and attach
2072 * to the interface indicated by the name stored in ifr.
2073 * Return an errno or 0.
2076 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2079 struct ifnet *theywant;
2083 theywant = ifunit(ifr->ifr_name);
2084 if (theywant == NULL || theywant->if_bpf == NULL)
2087 bp = theywant->if_bpf;
2089 * At this point, we expect the buffer is already allocated. If not,
2092 switch (d->bd_bufmode) {
2093 case BPF_BUFMODE_BUFFER:
2094 case BPF_BUFMODE_ZBUF:
2095 if (d->bd_sbuf == NULL)
2100 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2102 if (bp != d->bd_bif)
2113 * Support for select() and poll() system calls
2115 * Return true iff the specific operation will not block indefinitely.
2116 * Otherwise, return false but make a note that a selwakeup() must be done.
2119 bpfpoll(struct cdev *dev, int events, struct thread *td)
2124 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2126 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2129 * Refresh PID associated with this descriptor.
2131 revents = events & (POLLOUT | POLLWRNORM);
2133 BPF_PID_REFRESH(d, td);
2134 if (events & (POLLIN | POLLRDNORM)) {
2136 revents |= events & (POLLIN | POLLRDNORM);
2138 selrecord(td, &d->bd_sel);
2139 /* Start the read timeout if necessary. */
2140 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2141 callout_reset(&d->bd_callout, d->bd_rtout,
2143 d->bd_state = BPF_WAITING;
2152 * Support for kevent() system call. Register EVFILT_READ filters and
2153 * reject all others.
2156 bpfkqfilter(struct cdev *dev, struct knote *kn)
2160 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2161 kn->kn_filter != EVFILT_READ)
2165 * Refresh PID associated with this descriptor.
2168 BPF_PID_REFRESH_CUR(d);
2169 kn->kn_fop = &bpfread_filtops;
2171 knlist_add(&d->bd_sel.si_note, kn, 1);
2178 filt_bpfdetach(struct knote *kn)
2180 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2182 knlist_remove(&d->bd_sel.si_note, kn, 0);
2186 filt_bpfread(struct knote *kn, long hint)
2188 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2191 BPFD_LOCK_ASSERT(d);
2192 ready = bpf_ready(d);
2194 kn->kn_data = d->bd_slen;
2196 * Ignore the hold buffer if it is being copied to user space.
2198 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2199 kn->kn_data += d->bd_hlen;
2200 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2201 callout_reset(&d->bd_callout, d->bd_rtout,
2203 d->bd_state = BPF_WAITING;
2209 #define BPF_TSTAMP_NONE 0
2210 #define BPF_TSTAMP_FAST 1
2211 #define BPF_TSTAMP_NORMAL 2
2212 #define BPF_TSTAMP_EXTERN 3
2215 bpf_ts_quality(int tstype)
2218 if (tstype == BPF_T_NONE)
2219 return (BPF_TSTAMP_NONE);
2220 if ((tstype & BPF_T_FAST) != 0)
2221 return (BPF_TSTAMP_FAST);
2223 return (BPF_TSTAMP_NORMAL);
2227 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2232 quality = bpf_ts_quality(tstype);
2233 if (quality == BPF_TSTAMP_NONE)
2237 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2239 *bt = *(struct bintime *)(tag + 1);
2240 return (BPF_TSTAMP_EXTERN);
2243 if (quality == BPF_TSTAMP_NORMAL)
2252 * Incoming linkage from device drivers. Process the packet pkt, of length
2253 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2254 * by each process' filter, and if accepted, stashed into the corresponding
2258 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2260 struct epoch_tracker et;
2269 gottime = BPF_TSTAMP_NONE;
2270 NET_EPOCH_ENTER(et);
2271 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2272 counter_u64_add(d->bd_rcount, 1);
2274 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2275 * is no way for the caller to indiciate to us whether this
2276 * packet is inbound or outbound. In the bpf_mtap() routines,
2277 * we use the interface pointers on the mbuf to figure it out.
2280 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2282 slen = (*(bf->func))(pkt, pktlen, pktlen);
2285 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2288 * Filter matches. Let's to acquire write lock.
2291 counter_u64_add(d->bd_fcount, 1);
2292 if (gottime < bpf_ts_quality(d->bd_tstamp))
2293 gottime = bpf_gettime(&bt, d->bd_tstamp,
2296 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2298 catchpacket(d, pkt, pktlen, slen,
2299 bpf_append_bytes, &bt);
2306 #define BPF_CHECK_DIRECTION(d, r, i) \
2307 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2308 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2311 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2312 * Locking model is explained in bpf_tap().
2315 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2317 struct epoch_tracker et;
2326 /* Skip outgoing duplicate packets. */
2327 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2328 m->m_flags &= ~M_PROMISC;
2332 pktlen = m_length(m, NULL);
2333 gottime = BPF_TSTAMP_NONE;
2335 NET_EPOCH_ENTER(et);
2336 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2337 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2339 counter_u64_add(d->bd_rcount, 1);
2341 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2342 /* XXX We cannot handle multiple mbufs. */
2343 if (bf != NULL && m->m_next == NULL)
2344 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2348 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2352 counter_u64_add(d->bd_fcount, 1);
2353 if (gottime < bpf_ts_quality(d->bd_tstamp))
2354 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2356 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2358 catchpacket(d, (u_char *)m, pktlen, slen,
2359 bpf_append_mbuf, &bt);
2367 * Incoming linkage from device drivers, when packet is in
2368 * an mbuf chain and to be prepended by a contiguous header.
2371 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2373 struct epoch_tracker et;
2380 /* Skip outgoing duplicate packets. */
2381 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2382 m->m_flags &= ~M_PROMISC;
2386 pktlen = m_length(m, NULL);
2388 * Craft on-stack mbuf suitable for passing to bpf_filter.
2389 * Note that we cut corners here; we only setup what's
2390 * absolutely needed--this mbuf should never go anywhere else.
2398 gottime = BPF_TSTAMP_NONE;
2400 NET_EPOCH_ENTER(et);
2401 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2402 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2404 counter_u64_add(d->bd_rcount, 1);
2405 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2409 counter_u64_add(d->bd_fcount, 1);
2410 if (gottime < bpf_ts_quality(d->bd_tstamp))
2411 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2413 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2415 catchpacket(d, (u_char *)&mb, pktlen, slen,
2416 bpf_append_mbuf, &bt);
2423 #undef BPF_CHECK_DIRECTION
2424 #undef BPF_TSTAMP_NONE
2425 #undef BPF_TSTAMP_FAST
2426 #undef BPF_TSTAMP_NORMAL
2427 #undef BPF_TSTAMP_EXTERN
2430 bpf_hdrlen(struct bpf_d *d)
2434 hdrlen = d->bd_bif->bif_hdrlen;
2435 #ifndef BURN_BRIDGES
2436 if (d->bd_tstamp == BPF_T_NONE ||
2437 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2438 #ifdef COMPAT_FREEBSD32
2440 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2443 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2446 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2447 #ifdef COMPAT_FREEBSD32
2449 hdrlen = BPF_WORDALIGN32(hdrlen);
2452 hdrlen = BPF_WORDALIGN(hdrlen);
2454 return (hdrlen - d->bd_bif->bif_hdrlen);
2458 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2460 struct bintime bt2, boottimebin;
2462 struct timespec tsn;
2464 if ((tstype & BPF_T_MONOTONIC) == 0) {
2466 getboottimebin(&boottimebin);
2467 bintime_add(&bt2, &boottimebin);
2470 switch (BPF_T_FORMAT(tstype)) {
2471 case BPF_T_MICROTIME:
2472 bintime2timeval(bt, &tsm);
2473 ts->bt_sec = tsm.tv_sec;
2474 ts->bt_frac = tsm.tv_usec;
2476 case BPF_T_NANOTIME:
2477 bintime2timespec(bt, &tsn);
2478 ts->bt_sec = tsn.tv_sec;
2479 ts->bt_frac = tsn.tv_nsec;
2482 ts->bt_sec = bt->sec;
2483 ts->bt_frac = bt->frac;
2489 * Move the packet data from interface memory (pkt) into the
2490 * store buffer. "cpfn" is the routine called to do the actual data
2491 * transfer. bcopy is passed in to copy contiguous chunks, while
2492 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2493 * pkt is really an mbuf.
2496 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2497 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2500 struct bpf_xhdr hdr;
2501 #ifndef BURN_BRIDGES
2502 struct bpf_hdr hdr_old;
2503 #ifdef COMPAT_FREEBSD32
2504 struct bpf_hdr32 hdr32_old;
2507 int caplen, curlen, hdrlen, totlen;
2512 BPFD_LOCK_ASSERT(d);
2513 if (d->bd_bif == NULL) {
2514 /* Descriptor was detached in concurrent thread */
2515 counter_u64_add(d->bd_dcount, 1);
2520 * Detect whether user space has released a buffer back to us, and if
2521 * so, move it from being a hold buffer to a free buffer. This may
2522 * not be the best place to do it (for example, we might only want to
2523 * run this check if we need the space), but for now it's a reliable
2526 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2527 d->bd_fbuf = d->bd_hbuf;
2530 bpf_buf_reclaimed(d);
2534 * Figure out how many bytes to move. If the packet is
2535 * greater or equal to the snapshot length, transfer that
2536 * much. Otherwise, transfer the whole packet (unless
2537 * we hit the buffer size limit).
2539 hdrlen = bpf_hdrlen(d);
2540 totlen = hdrlen + min(snaplen, pktlen);
2541 if (totlen > d->bd_bufsize)
2542 totlen = d->bd_bufsize;
2545 * Round up the end of the previous packet to the next longword.
2547 * Drop the packet if there's no room and no hope of room
2548 * If the packet would overflow the storage buffer or the storage
2549 * buffer is considered immutable by the buffer model, try to rotate
2550 * the buffer and wakeup pending processes.
2552 #ifdef COMPAT_FREEBSD32
2554 curlen = BPF_WORDALIGN32(d->bd_slen);
2557 curlen = BPF_WORDALIGN(d->bd_slen);
2558 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2559 if (d->bd_fbuf == NULL) {
2561 * There's no room in the store buffer, and no
2562 * prospect of room, so drop the packet. Notify the
2566 counter_u64_add(d->bd_dcount, 1);
2569 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2573 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2575 * Immediate mode is set, or the read timeout has already
2576 * expired during a select call. A packet arrived, so the
2577 * reader should be woken up.
2580 caplen = totlen - hdrlen;
2581 tstype = d->bd_tstamp;
2582 do_timestamp = tstype != BPF_T_NONE;
2583 #ifndef BURN_BRIDGES
2584 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2587 bpf_bintime2ts(bt, &ts, tstype);
2588 #ifdef COMPAT_FREEBSD32
2589 if (d->bd_compat32) {
2590 bzero(&hdr32_old, sizeof(hdr32_old));
2592 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2593 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2595 hdr32_old.bh_datalen = pktlen;
2596 hdr32_old.bh_hdrlen = hdrlen;
2597 hdr32_old.bh_caplen = caplen;
2598 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2603 bzero(&hdr_old, sizeof(hdr_old));
2605 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2606 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2608 hdr_old.bh_datalen = pktlen;
2609 hdr_old.bh_hdrlen = hdrlen;
2610 hdr_old.bh_caplen = caplen;
2611 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2618 * Append the bpf header. Note we append the actual header size, but
2619 * move forward the length of the header plus padding.
2621 bzero(&hdr, sizeof(hdr));
2623 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2624 hdr.bh_datalen = pktlen;
2625 hdr.bh_hdrlen = hdrlen;
2626 hdr.bh_caplen = caplen;
2627 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2630 * Copy the packet data into the store buffer and update its length.
2632 #ifndef BURN_BRIDGES
2635 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2636 d->bd_slen = curlen + totlen;
2643 * Free buffers currently in use by a descriptor.
2647 bpfd_free(epoch_context_t ctx)
2650 struct bpf_program_buffer *p;
2653 * We don't need to lock out interrupts since this descriptor has
2654 * been detached from its interface and it yet hasn't been marked
2657 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2659 if (d->bd_rfilter != NULL) {
2660 p = __containerof((void *)d->bd_rfilter,
2661 struct bpf_program_buffer, buffer);
2663 p->func = d->bd_bfilter;
2665 bpf_program_buffer_free(&p->epoch_ctx);
2667 if (d->bd_wfilter != NULL) {
2668 p = __containerof((void *)d->bd_wfilter,
2669 struct bpf_program_buffer, buffer);
2673 bpf_program_buffer_free(&p->epoch_ctx);
2676 mtx_destroy(&d->bd_lock);
2677 counter_u64_free(d->bd_rcount);
2678 counter_u64_free(d->bd_dcount);
2679 counter_u64_free(d->bd_fcount);
2680 counter_u64_free(d->bd_wcount);
2681 counter_u64_free(d->bd_wfcount);
2682 counter_u64_free(d->bd_wdcount);
2683 counter_u64_free(d->bd_zcopy);
2688 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2689 * fixed size of the link header (variable length headers not yet supported).
2692 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2695 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2699 * Attach an interface to bpf. ifp is a pointer to the structure
2700 * defining the interface to be attached, dlt is the link layer type,
2701 * and hdrlen is the fixed size of the link header (variable length
2702 * headers are not yet supporrted).
2705 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2706 struct bpf_if **driverp)
2710 KASSERT(*driverp == NULL,
2711 ("bpfattach2: driverp already initialized"));
2713 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2715 CK_LIST_INIT(&bp->bif_dlist);
2716 CK_LIST_INIT(&bp->bif_wlist);
2719 bp->bif_hdrlen = hdrlen;
2720 bp->bif_bpf = driverp;
2724 * Reference ifnet pointer, so it won't freed until
2729 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2732 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2733 if_printf(ifp, "bpf attached\n");
2738 * When moving interfaces between vnet instances we need a way to
2739 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2740 * after the vmove. We unfortunately have no device driver infrastructure
2741 * to query the interface for these values after creation/attach, thus
2742 * add this as a workaround.
2745 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2750 if (bif_dlt == NULL && bif_hdrlen == NULL)
2753 if (bif_dlt != NULL)
2754 *bif_dlt = bp->bif_dlt;
2755 if (bif_hdrlen != NULL)
2756 *bif_hdrlen = bp->bif_hdrlen;
2763 * Detach bpf from an interface. This involves detaching each descriptor
2764 * associated with the interface. Notify each descriptor as it's detached
2765 * so that any sleepers wake up and get ENXIO.
2768 bpfdetach(struct ifnet *ifp)
2770 struct bpf_if *bp, *bp_temp;
2774 /* Find all bpf_if struct's which reference ifp and detach them. */
2775 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2776 if (ifp != bp->bif_ifp)
2779 CK_LIST_REMOVE(bp, bif_next);
2780 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2783 "%s: sheduling free for encap %d (%p) for if %p",
2784 __func__, bp->bif_dlt, bp, ifp);
2786 /* Detach common descriptors */
2787 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2788 bpf_detachd_locked(d, true);
2791 /* Detach writer-only descriptors */
2792 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2793 bpf_detachd_locked(d, true);
2801 * Get a list of available data link type of the interface.
2804 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2813 ifp = d->bd_bif->bif_ifp;
2815 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2816 if (bp->bif_ifp == ifp)
2819 if (bfl->bfl_list == NULL) {
2823 if (n1 > bfl->bfl_len)
2826 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2828 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2829 if (bp->bif_ifp != ifp)
2831 lst[n++] = bp->bif_dlt;
2833 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2840 * Set the data link type of a BPF instance.
2843 bpf_setdlt(struct bpf_d *d, u_int dlt)
2845 int error, opromisc;
2850 MPASS(d->bd_bif != NULL);
2853 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2854 * changed while we hold global lock.
2856 if (d->bd_bif->bif_dlt == dlt)
2859 ifp = d->bd_bif->bif_ifp;
2860 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2861 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2867 opromisc = d->bd_promisc;
2870 error = ifpromisc(bp->bif_ifp, 1);
2872 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2881 bpf_drvinit(void *unused)
2885 sx_init(&bpf_sx, "bpf global lock");
2886 CK_LIST_INIT(&bpf_iflist);
2888 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2889 /* For compatibility */
2890 make_dev_alias(dev, "bpf0");
2894 * Zero out the various packet counters associated with all of the bpf
2895 * descriptors. At some point, we will probably want to get a bit more
2896 * granular and allow the user to specify descriptors to be zeroed.
2899 bpf_zero_counters(void)
2906 * We are protected by global lock here, interfaces and
2907 * descriptors can not be deleted while we hold it.
2909 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2910 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2911 counter_u64_zero(bd->bd_rcount);
2912 counter_u64_zero(bd->bd_dcount);
2913 counter_u64_zero(bd->bd_fcount);
2914 counter_u64_zero(bd->bd_wcount);
2915 counter_u64_zero(bd->bd_wfcount);
2916 counter_u64_zero(bd->bd_zcopy);
2923 * Fill filter statistics
2926 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2930 bzero(d, sizeof(*d));
2931 d->bd_structsize = sizeof(*d);
2932 d->bd_immediate = bd->bd_immediate;
2933 d->bd_promisc = bd->bd_promisc;
2934 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2935 d->bd_direction = bd->bd_direction;
2936 d->bd_feedback = bd->bd_feedback;
2937 d->bd_async = bd->bd_async;
2938 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2939 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2940 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2941 d->bd_sig = bd->bd_sig;
2942 d->bd_slen = bd->bd_slen;
2943 d->bd_hlen = bd->bd_hlen;
2944 d->bd_bufsize = bd->bd_bufsize;
2945 d->bd_pid = bd->bd_pid;
2946 strlcpy(d->bd_ifname,
2947 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2948 d->bd_locked = bd->bd_locked;
2949 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2950 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2951 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2952 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2953 d->bd_bufmode = bd->bd_bufmode;
2957 * Handle `netstat -B' stats request
2960 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2962 static const struct xbpf_d zerostats;
2963 struct xbpf_d *xbdbuf, *xbd, tempstats;
2969 * XXX This is not technically correct. It is possible for non
2970 * privileged users to open bpf devices. It would make sense
2971 * if the users who opened the devices were able to retrieve
2972 * the statistics for them, too.
2974 error = priv_check(req->td, PRIV_NET_BPF);
2978 * Check to see if the user is requesting that the counters be
2979 * zeroed out. Explicitly check that the supplied data is zeroed,
2980 * as we aren't allowing the user to set the counters currently.
2982 if (req->newptr != NULL) {
2983 if (req->newlen != sizeof(tempstats))
2985 memset(&tempstats, 0, sizeof(tempstats));
2986 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2989 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2991 bpf_zero_counters();
2994 if (req->oldptr == NULL)
2995 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2996 if (bpf_bpfd_cnt == 0)
2997 return (SYSCTL_OUT(req, 0, 0));
2998 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3000 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3002 free(xbdbuf, M_BPF);
3006 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3007 /* Send writers-only first */
3008 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3009 xbd = &xbdbuf[index++];
3010 bpfstats_fill_xbpf(xbd, bd);
3012 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3013 xbd = &xbdbuf[index++];
3014 bpfstats_fill_xbpf(xbd, bd);
3018 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3019 free(xbdbuf, M_BPF);
3023 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3025 #else /* !DEV_BPF && !NETGRAPH_BPF */
3028 * NOP stubs to allow bpf-using drivers to load and function.
3030 * A 'better' implementation would allow the core bpf functionality
3031 * to be loaded at runtime.
3035 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3040 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3045 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3050 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3053 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3057 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3060 *driverp = (struct bpf_if *)&dead_bpf_if;
3064 bpfdetach(struct ifnet *ifp)
3069 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3071 return -1; /* "no filter" behaviour */
3075 bpf_validate(const struct bpf_insn *f, int len)
3077 return 0; /* false */
3080 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3084 bpf_show_bpf_if(struct bpf_if *bpf_if)
3089 db_printf("%p:\n", bpf_if);
3090 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3091 /* bif_ext.bif_next */
3092 /* bif_ext.bif_dlist */
3093 BPF_DB_PRINTF("%#x", bif_dlt);
3094 BPF_DB_PRINTF("%u", bif_hdrlen);
3096 BPF_DB_PRINTF("%p", bif_ifp);
3097 BPF_DB_PRINTF("%p", bif_bpf);
3098 BPF_DB_PRINTF("%u", bif_refcnt);
3101 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3105 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3109 bpf_show_bpf_if((struct bpf_if *)addr);