2 * Copyright (c) 1990, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from the Stanford/CMU enet packet filter,
6 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
7 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
41 #include "opt_compat.h"
42 #include "opt_netgraph.h"
44 #include <sys/types.h>
45 #include <sys/param.h>
47 #include <sys/rwlock.h>
48 #include <sys/systm.h>
50 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
57 #include <sys/signalvar.h>
58 #include <sys/filio.h>
59 #include <sys/sockio.h>
60 #include <sys/ttycom.h>
63 #include <sys/event.h>
68 #include <sys/socket.h>
73 #include <net/bpf_buffer.h>
75 #include <net/bpf_jitter.h>
77 #include <net/bpf_zerocopy.h>
78 #include <net/bpfdesc.h>
81 #include <netinet/in.h>
82 #include <netinet/if_ether.h>
83 #include <sys/kernel.h>
84 #include <sys/sysctl.h>
86 #include <net80211/ieee80211_freebsd.h>
88 #include <security/mac/mac_framework.h>
90 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
92 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
94 #define PRINET 26 /* interruptible */
96 #define SIZEOF_BPF_HDR(type) \
97 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
99 #ifdef COMPAT_FREEBSD32
100 #include <sys/mount.h>
101 #include <compat/freebsd32/freebsd32.h>
102 #define BPF_ALIGNMENT32 sizeof(int32_t)
103 #define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1))
107 * 32-bit version of structure prepended to each packet. We use this header
108 * instead of the standard one for 32-bit streams. We mark the a stream as
109 * 32-bit the first time we see a 32-bit compat ioctl request.
112 struct timeval32 bh_tstamp; /* time stamp */
113 uint32_t bh_caplen; /* length of captured portion */
114 uint32_t bh_datalen; /* original length of packet */
115 uint16_t bh_hdrlen; /* length of bpf header (this struct
116 plus alignment padding) */
120 struct bpf_program32 {
125 struct bpf_dltlist32 {
130 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
131 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
132 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
133 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
134 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
135 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
139 * bpf_iflist is a list of BPF interface structures, each corresponding to a
140 * specific DLT. The same network interface might have several BPF interface
141 * structures registered by different layers in the stack (i.e., 802.11
142 * frames, ethernet frames, etc).
144 static LIST_HEAD(, bpf_if) bpf_iflist;
145 static struct mtx bpf_mtx; /* bpf global lock */
146 static int bpf_bpfd_cnt;
148 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
149 static void bpf_detachd(struct bpf_d *);
150 static void bpf_detachd_locked(struct bpf_d *);
151 static void bpf_freed(struct bpf_d *);
152 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
153 struct sockaddr *, int *, struct bpf_insn *);
154 static int bpf_setif(struct bpf_d *, struct ifreq *);
155 static void bpf_timed_out(void *);
157 bpf_wakeup(struct bpf_d *);
158 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
159 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
161 static void reset_d(struct bpf_d *);
162 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
163 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
164 static int bpf_setdlt(struct bpf_d *, u_int);
165 static void filt_bpfdetach(struct knote *);
166 static int filt_bpfread(struct knote *, long);
167 static void bpf_drvinit(void *);
168 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
170 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
171 int bpf_maxinsns = BPF_MAXINSNS;
172 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
173 &bpf_maxinsns, 0, "Maximum bpf program instructions");
174 static int bpf_zerocopy_enable = 0;
175 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
176 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
177 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
178 bpf_stats_sysctl, "bpf statistics portal");
180 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
181 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
182 SYSCTL_VNET_INT(_net_bpf, OID_AUTO, optimize_writers,
183 CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0,
184 "Do not send packets until BPF program is set");
186 static d_open_t bpfopen;
187 static d_read_t bpfread;
188 static d_write_t bpfwrite;
189 static d_ioctl_t bpfioctl;
190 static d_poll_t bpfpoll;
191 static d_kqfilter_t bpfkqfilter;
193 static struct cdevsw bpf_cdevsw = {
194 .d_version = D_VERSION,
201 .d_kqfilter = bpfkqfilter,
204 static struct filterops bpfread_filtops = {
206 .f_detach = filt_bpfdetach,
207 .f_event = filt_bpfread,
210 eventhandler_tag bpf_ifdetach_cookie = NULL;
213 * LOCKING MODEL USED BY BPF:
215 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
216 * some global counters and every bpf_if reference.
217 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
218 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
219 * used by bpf_mtap code.
223 * Global lock, interface lock, descriptor lock
225 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
226 * working model. In many places (like bpf_detachd) we start with BPF descriptor
227 * (and we need to at least rlock it to get reliable interface pointer). This
228 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
229 * change in every such place.
231 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
232 * 3) descriptor main wlock.
233 * Reading bd_bif can be protected by any of these locks, typically global lock.
235 * Changing read/write BPF filter is protected by the same three locks,
236 * the same applies for reading.
238 * Sleeping in global lock is not allowed due to bpfdetach() using it.
242 * Wrapper functions for various buffering methods. If the set of buffer
243 * modes expands, we will probably want to introduce a switch data structure
244 * similar to protosw, et.
247 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
253 switch (d->bd_bufmode) {
254 case BPF_BUFMODE_BUFFER:
255 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
257 case BPF_BUFMODE_ZBUF:
259 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
262 panic("bpf_buf_append_bytes");
267 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
273 switch (d->bd_bufmode) {
274 case BPF_BUFMODE_BUFFER:
275 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
277 case BPF_BUFMODE_ZBUF:
279 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
282 panic("bpf_buf_append_mbuf");
287 * This function gets called when the free buffer is re-assigned.
290 bpf_buf_reclaimed(struct bpf_d *d)
295 switch (d->bd_bufmode) {
296 case BPF_BUFMODE_BUFFER:
299 case BPF_BUFMODE_ZBUF:
300 bpf_zerocopy_buf_reclaimed(d);
304 panic("bpf_buf_reclaimed");
309 * If the buffer mechanism has a way to decide that a held buffer can be made
310 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
311 * returned if the buffer can be discarded, (0) is returned if it cannot.
314 bpf_canfreebuf(struct bpf_d *d)
319 switch (d->bd_bufmode) {
320 case BPF_BUFMODE_ZBUF:
321 return (bpf_zerocopy_canfreebuf(d));
327 * Allow the buffer model to indicate that the current store buffer is
328 * immutable, regardless of the appearance of space. Return (1) if the
329 * buffer is writable, and (0) if not.
332 bpf_canwritebuf(struct bpf_d *d)
336 switch (d->bd_bufmode) {
337 case BPF_BUFMODE_ZBUF:
338 return (bpf_zerocopy_canwritebuf(d));
344 * Notify buffer model that an attempt to write to the store buffer has
345 * resulted in a dropped packet, in which case the buffer may be considered
349 bpf_buffull(struct bpf_d *d)
354 switch (d->bd_bufmode) {
355 case BPF_BUFMODE_ZBUF:
356 bpf_zerocopy_buffull(d);
362 * Notify the buffer model that a buffer has moved into the hold position.
365 bpf_bufheld(struct bpf_d *d)
370 switch (d->bd_bufmode) {
371 case BPF_BUFMODE_ZBUF:
372 bpf_zerocopy_bufheld(d);
378 bpf_free(struct bpf_d *d)
381 switch (d->bd_bufmode) {
382 case BPF_BUFMODE_BUFFER:
383 return (bpf_buffer_free(d));
385 case BPF_BUFMODE_ZBUF:
386 return (bpf_zerocopy_free(d));
389 panic("bpf_buf_free");
394 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
397 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
399 return (bpf_buffer_uiomove(d, buf, len, uio));
403 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
406 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
408 return (bpf_buffer_ioctl_sblen(d, i));
412 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
415 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
417 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
421 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
424 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
426 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
430 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
433 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
435 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
439 * General BPF functions.
442 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
443 struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter)
445 const struct ieee80211_bpf_params *p;
446 struct ether_header *eh;
454 * Build a sockaddr based on the data link layer type.
455 * We do this at this level because the ethernet header
456 * is copied directly into the data field of the sockaddr.
457 * In the case of SLIP, there is no header and the packet
458 * is forwarded as is.
459 * Also, we are careful to leave room at the front of the mbuf
460 * for the link level header.
465 sockp->sa_family = AF_INET;
470 sockp->sa_family = AF_UNSPEC;
471 /* XXX Would MAXLINKHDR be better? */
472 hlen = ETHER_HDR_LEN;
476 sockp->sa_family = AF_IMPLINK;
481 sockp->sa_family = AF_UNSPEC;
487 * null interface types require a 4 byte pseudo header which
488 * corresponds to the address family of the packet.
490 sockp->sa_family = AF_UNSPEC;
494 case DLT_ATM_RFC1483:
496 * en atm driver requires 4-byte atm pseudo header.
497 * though it isn't standard, vpi:vci needs to be
500 sockp->sa_family = AF_UNSPEC;
501 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
505 sockp->sa_family = AF_UNSPEC;
506 hlen = 4; /* This should match PPP_HDRLEN */
509 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
510 sockp->sa_family = AF_IEEE80211;
514 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
515 sockp->sa_family = AF_IEEE80211;
516 sockp->sa_len = 12; /* XXX != 0 */
517 hlen = sizeof(struct ieee80211_bpf_params);
524 len = uio->uio_resid;
526 if (len - hlen > ifp->if_mtu)
529 if ((unsigned)len > MJUM16BYTES)
533 MGETHDR(m, M_WAIT, MT_DATA);
534 else if (len <= MCLBYTES)
535 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
537 m = m_getjcl(M_WAIT, MT_DATA, M_PKTHDR,
538 #if (MJUMPAGESIZE > MCLBYTES)
539 len <= MJUMPAGESIZE ? MJUMPAGESIZE :
541 (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES));
542 m->m_pkthdr.len = m->m_len = len;
543 m->m_pkthdr.rcvif = NULL;
546 if (m->m_len < hlen) {
551 error = uiomove(mtod(m, u_char *), len, uio);
555 slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
561 /* Check for multicast destination */
564 eh = mtod(m, struct ether_header *);
565 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
566 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
567 ETHER_ADDR_LEN) == 0)
568 m->m_flags |= M_BCAST;
570 m->m_flags |= M_MCAST;
576 * Make room for link header, and copy it to sockaddr
579 if (sockp->sa_family == AF_IEEE80211) {
581 * Collect true length from the parameter header
582 * NB: sockp is known to be zero'd so if we do a
583 * short copy unspecified parameters will be
585 * NB: packet may not be aligned after stripping
589 p = mtod(m, const struct ieee80211_bpf_params *);
591 if (hlen > sizeof(sockp->sa_data)) {
596 bcopy(m->m_data, sockp->sa_data, hlen);
607 * Attach file to the bpf interface, i.e. make d listen on bp.
610 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
617 * Save sysctl value to protect from sysctl change
620 op_w = V_bpf_optimize_writers;
622 if (d->bd_bif != NULL)
623 bpf_detachd_locked(d);
625 * Point d at bp, and add d to the interface's list.
626 * Since there are many applicaiotns using BPF for
627 * sending raw packets only (dhcpd, cdpd are good examples)
628 * we can delay adding d to the list of active listeners until
629 * some filter is configured.
638 /* Add to writers-only list */
639 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
641 * We decrement bd_writer on every filter set operation.
642 * First BIOCSETF is done by pcap_open_live() to set up
643 * snap length. After that appliation usually sets its own filter
647 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
654 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
655 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
658 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
662 * Add d to the list of active bp filters.
663 * Reuqires bpf_attachd() to be called before
666 bpf_upgraded(struct bpf_d *d)
675 * Filter can be set several times without specifying interface.
676 * Mark d as reader and exit.
688 /* Remove from writers-only list */
689 LIST_REMOVE(d, bd_next);
690 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
691 /* Mark d as reader */
697 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
699 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
703 * Detach a file from its interface.
706 bpf_detachd(struct bpf_d *d)
709 bpf_detachd_locked(d);
714 bpf_detachd_locked(struct bpf_d *d)
720 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
724 /* Check if descriptor is attached */
725 if ((bp = d->bd_bif) == NULL)
731 /* Save bd_writer value */
732 error = d->bd_writer;
735 * Remove d from the interface's descriptor list.
737 LIST_REMOVE(d, bd_next);
746 /* Call event handler iff d is attached */
748 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
751 * Check if this descriptor had requested promiscuous mode.
752 * If so, turn it off.
756 CURVNET_SET(ifp->if_vnet);
757 error = ifpromisc(ifp, 0);
759 if (error != 0 && error != ENXIO) {
761 * ENXIO can happen if a pccard is unplugged
762 * Something is really wrong if we were able to put
763 * the driver into promiscuous mode, but can't
766 if_printf(bp->bif_ifp,
767 "bpf_detach: ifpromisc failed (%d)\n", error);
773 * Close the descriptor by detaching it from its interface,
774 * deallocating its buffers, and marking it free.
779 struct bpf_d *d = data;
782 if (d->bd_state == BPF_WAITING)
783 callout_stop(&d->bd_callout);
784 d->bd_state = BPF_IDLE;
786 funsetown(&d->bd_sigio);
789 mac_bpfdesc_destroy(d);
791 seldrain(&d->bd_sel);
792 knlist_destroy(&d->bd_sel.si_note);
793 callout_drain(&d->bd_callout);
799 * Open ethernet device. Returns ENXIO for illegal minor device number,
800 * EBUSY if file is open by another process.
804 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
809 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
810 error = devfs_set_cdevpriv(d, bpf_dtor);
817 * For historical reasons, perform a one-time initialization call to
818 * the buffer routines, even though we're not yet committed to a
819 * particular buffer method.
822 d->bd_bufmode = BPF_BUFMODE_BUFFER;
824 d->bd_direction = BPF_D_INOUT;
825 BPF_PID_REFRESH(d, td);
828 mac_bpfdesc_create(td->td_ucred, d);
830 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
831 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
832 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
834 /* Allocate default buffers */
835 size = d->bd_bufsize;
836 bpf_buffer_ioctl_sblen(d, &size);
842 * bpfread - read next chunk of packets from buffers
845 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
852 error = devfs_get_cdevpriv((void **)&d);
857 * Restrict application to use a buffer the same size as
860 if (uio->uio_resid != d->bd_bufsize)
863 non_block = ((ioflag & O_NONBLOCK) != 0);
866 BPF_PID_REFRESH_CUR(d);
867 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
871 if (d->bd_state == BPF_WAITING)
872 callout_stop(&d->bd_callout);
873 timed_out = (d->bd_state == BPF_TIMED_OUT);
874 d->bd_state = BPF_IDLE;
876 * If the hold buffer is empty, then do a timed sleep, which
877 * ends when the timeout expires or when enough packets
878 * have arrived to fill the store buffer.
880 while (d->bd_hbuf == NULL) {
881 if (d->bd_slen != 0) {
883 * A packet(s) either arrived since the previous
884 * read or arrived while we were asleep.
886 if (d->bd_immediate || non_block || timed_out) {
888 * Rotate the buffers and return what's here
889 * if we are in immediate mode, non-blocking
890 * flag is set, or this descriptor timed out.
898 * No data is available, check to see if the bpf device
899 * is still pointed at a real interface. If not, return
900 * ENXIO so that the userland process knows to rebind
901 * it before using it again.
903 if (d->bd_bif == NULL) {
910 return (EWOULDBLOCK);
912 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
914 if (error == EINTR || error == ERESTART) {
918 if (error == EWOULDBLOCK) {
920 * On a timeout, return what's in the buffer,
921 * which may be nothing. If there is something
922 * in the store buffer, we can rotate the buffers.
926 * We filled up the buffer in between
927 * getting the timeout and arriving
928 * here, so we don't need to rotate.
932 if (d->bd_slen == 0) {
941 * At this point, we know we have something in the hold slot.
946 * Move data from hold buffer into user space.
947 * We know the entire buffer is transferred since
948 * we checked above that the read buffer is bpf_bufsize bytes.
950 * XXXRW: More synchronization needed here: what if a second thread
951 * issues a read on the same fd at the same time? Don't want this
952 * getting invalidated.
954 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
957 d->bd_fbuf = d->bd_hbuf;
960 bpf_buf_reclaimed(d);
967 * If there are processes sleeping on this descriptor, wake them up.
970 bpf_wakeup(struct bpf_d *d)
974 if (d->bd_state == BPF_WAITING) {
975 callout_stop(&d->bd_callout);
976 d->bd_state = BPF_IDLE;
979 if (d->bd_async && d->bd_sig && d->bd_sigio)
980 pgsigio(&d->bd_sigio, d->bd_sig, 0);
982 selwakeuppri(&d->bd_sel, PRINET);
983 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
987 bpf_timed_out(void *arg)
989 struct bpf_d *d = (struct bpf_d *)arg;
993 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
995 if (d->bd_state == BPF_WAITING) {
996 d->bd_state = BPF_TIMED_OUT;
1003 bpf_ready(struct bpf_d *d)
1006 BPFD_LOCK_ASSERT(d);
1008 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1010 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1017 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1021 struct mbuf *m, *mc;
1022 struct sockaddr dst;
1025 error = devfs_get_cdevpriv((void **)&d);
1029 BPF_PID_REFRESH_CUR(d);
1031 /* XXX: locking required */
1032 if (d->bd_bif == NULL) {
1037 ifp = d->bd_bif->bif_ifp;
1039 if ((ifp->if_flags & IFF_UP) == 0) {
1044 if (uio->uio_resid == 0) {
1049 bzero(&dst, sizeof(dst));
1052 /* XXX: bpf_movein() can sleep */
1053 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1054 &m, &dst, &hlen, d->bd_wfilter);
1061 dst.sa_family = pseudo_AF_HDRCMPLT;
1063 if (d->bd_feedback) {
1064 mc = m_dup(m, M_DONTWAIT);
1066 mc->m_pkthdr.rcvif = ifp;
1067 /* Set M_PROMISC for outgoing packets to be discarded. */
1068 if (d->bd_direction == BPF_D_INOUT)
1069 m->m_flags |= M_PROMISC;
1073 m->m_pkthdr.len -= hlen;
1075 m->m_data += hlen; /* XXX */
1077 CURVNET_SET(ifp->if_vnet);
1080 mac_bpfdesc_create_mbuf(d, m);
1082 mac_bpfdesc_create_mbuf(d, mc);
1086 error = (*ifp->if_output)(ifp, m, &dst, NULL);
1092 (*ifp->if_input)(ifp, mc);
1102 * Reset a descriptor by flushing its packet buffer and clearing the receive
1103 * and drop counts. This is doable for kernel-only buffers, but with
1104 * zero-copy buffers, we can't write to (or rotate) buffers that are
1105 * currently owned by userspace. It would be nice if we could encapsulate
1106 * this logic in the buffer code rather than here.
1109 reset_d(struct bpf_d *d)
1112 BPFD_LOCK_ASSERT(d);
1114 if ((d->bd_hbuf != NULL) &&
1115 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1116 /* Free the hold buffer. */
1117 d->bd_fbuf = d->bd_hbuf;
1120 bpf_buf_reclaimed(d);
1122 if (bpf_canwritebuf(d))
1134 * FIONREAD Check for read packet available.
1135 * SIOCGIFADDR Get interface address - convenient hook to driver.
1136 * BIOCGBLEN Get buffer len [for read()].
1137 * BIOCSETF Set read filter.
1138 * BIOCSETFNR Set read filter without resetting descriptor.
1139 * BIOCSETWF Set write filter.
1140 * BIOCFLUSH Flush read packet buffer.
1141 * BIOCPROMISC Put interface into promiscuous mode.
1142 * BIOCGDLT Get link layer type.
1143 * BIOCGETIF Get interface name.
1144 * BIOCSETIF Set interface.
1145 * BIOCSRTIMEOUT Set read timeout.
1146 * BIOCGRTIMEOUT Get read timeout.
1147 * BIOCGSTATS Get packet stats.
1148 * BIOCIMMEDIATE Set immediate mode.
1149 * BIOCVERSION Get filter language version.
1150 * BIOCGHDRCMPLT Get "header already complete" flag
1151 * BIOCSHDRCMPLT Set "header already complete" flag
1152 * BIOCGDIRECTION Get packet direction flag
1153 * BIOCSDIRECTION Set packet direction flag
1154 * BIOCGTSTAMP Get time stamp format and resolution.
1155 * BIOCSTSTAMP Set time stamp format and resolution.
1156 * BIOCLOCK Set "locked" flag
1157 * BIOCFEEDBACK Set packet feedback mode.
1158 * BIOCSETZBUF Set current zero-copy buffer locations.
1159 * BIOCGETZMAX Get maximum zero-copy buffer size.
1160 * BIOCROTZBUF Force rotation of zero-copy buffer
1161 * BIOCSETBUFMODE Set buffer mode.
1162 * BIOCGETBUFMODE Get current buffer mode.
1166 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1172 error = devfs_get_cdevpriv((void **)&d);
1177 * Refresh PID associated with this descriptor.
1180 BPF_PID_REFRESH(d, td);
1181 if (d->bd_state == BPF_WAITING)
1182 callout_stop(&d->bd_callout);
1183 d->bd_state = BPF_IDLE;
1186 if (d->bd_locked == 1) {
1192 #ifdef COMPAT_FREEBSD32
1193 case BIOCGDLTLIST32:
1197 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1198 case BIOCGRTIMEOUT32:
1209 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1210 case BIOCSRTIMEOUT32:
1220 #ifdef COMPAT_FREEBSD32
1222 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1223 * that it will get 32-bit packet headers.
1229 case BIOCGDLTLIST32:
1230 case BIOCGRTIMEOUT32:
1231 case BIOCSRTIMEOUT32:
1236 CURVNET_SET(TD_TO_VNET(td));
1244 * Check for read packet available.
1264 if (d->bd_bif == NULL)
1267 ifp = d->bd_bif->bif_ifp;
1268 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1274 * Get buffer len [for read()].
1277 *(u_int *)addr = d->bd_bufsize;
1281 * Set buffer length.
1284 error = bpf_ioctl_sblen(d, (u_int *)addr);
1288 * Set link layer read filter.
1293 #ifdef COMPAT_FREEBSD32
1298 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1302 * Flush read packet buffer.
1311 * Put interface into promiscuous mode.
1314 if (d->bd_bif == NULL) {
1316 * No interface attached yet.
1321 if (d->bd_promisc == 0) {
1322 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1329 * Get current data link type.
1332 if (d->bd_bif == NULL)
1335 *(u_int *)addr = d->bd_bif->bif_dlt;
1339 * Get a list of supported data link types.
1341 #ifdef COMPAT_FREEBSD32
1342 case BIOCGDLTLIST32:
1344 struct bpf_dltlist32 *list32;
1345 struct bpf_dltlist dltlist;
1347 list32 = (struct bpf_dltlist32 *)addr;
1348 dltlist.bfl_len = list32->bfl_len;
1349 dltlist.bfl_list = PTRIN(list32->bfl_list);
1350 if (d->bd_bif == NULL)
1353 error = bpf_getdltlist(d, &dltlist);
1355 list32->bfl_len = dltlist.bfl_len;
1362 if (d->bd_bif == NULL)
1365 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1369 * Set data link type.
1373 if (d->bd_bif == NULL)
1376 error = bpf_setdlt(d, *(u_int *)addr);
1381 * Get interface name.
1384 if (d->bd_bif == NULL)
1387 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1388 struct ifreq *const ifr = (struct ifreq *)addr;
1390 strlcpy(ifr->ifr_name, ifp->if_xname,
1391 sizeof(ifr->ifr_name));
1400 error = bpf_setif(d, (struct ifreq *)addr);
1408 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1409 case BIOCSRTIMEOUT32:
1412 struct timeval *tv = (struct timeval *)addr;
1413 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1414 struct timeval32 *tv32;
1415 struct timeval tv64;
1417 if (cmd == BIOCSRTIMEOUT32) {
1418 tv32 = (struct timeval32 *)addr;
1420 tv->tv_sec = tv32->tv_sec;
1421 tv->tv_usec = tv32->tv_usec;
1424 tv = (struct timeval *)addr;
1427 * Subtract 1 tick from tvtohz() since this isn't
1430 if ((error = itimerfix(tv)) == 0)
1431 d->bd_rtout = tvtohz(tv) - 1;
1439 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1440 case BIOCGRTIMEOUT32:
1444 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1445 struct timeval32 *tv32;
1446 struct timeval tv64;
1448 if (cmd == BIOCGRTIMEOUT32)
1452 tv = (struct timeval *)addr;
1454 tv->tv_sec = d->bd_rtout / hz;
1455 tv->tv_usec = (d->bd_rtout % hz) * tick;
1456 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1457 if (cmd == BIOCGRTIMEOUT32) {
1458 tv32 = (struct timeval32 *)addr;
1459 tv32->tv_sec = tv->tv_sec;
1460 tv32->tv_usec = tv->tv_usec;
1472 struct bpf_stat *bs = (struct bpf_stat *)addr;
1474 /* XXXCSJP overflow */
1475 bs->bs_recv = d->bd_rcount;
1476 bs->bs_drop = d->bd_dcount;
1481 * Set immediate mode.
1484 d->bd_immediate = *(u_int *)addr;
1489 struct bpf_version *bv = (struct bpf_version *)addr;
1491 bv->bv_major = BPF_MAJOR_VERSION;
1492 bv->bv_minor = BPF_MINOR_VERSION;
1497 * Get "header already complete" flag
1500 *(u_int *)addr = d->bd_hdrcmplt;
1504 * Set "header already complete" flag
1507 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1511 * Get packet direction flag
1513 case BIOCGDIRECTION:
1514 *(u_int *)addr = d->bd_direction;
1518 * Set packet direction flag
1520 case BIOCSDIRECTION:
1524 direction = *(u_int *)addr;
1525 switch (direction) {
1529 d->bd_direction = direction;
1538 * Get packet timestamp format and resolution.
1541 *(u_int *)addr = d->bd_tstamp;
1545 * Set packet timestamp format and resolution.
1551 func = *(u_int *)addr;
1552 if (BPF_T_VALID(func))
1553 d->bd_tstamp = func;
1560 d->bd_feedback = *(u_int *)addr;
1567 case FIONBIO: /* Non-blocking I/O */
1570 case FIOASYNC: /* Send signal on receive packets */
1571 d->bd_async = *(int *)addr;
1575 error = fsetown(*(int *)addr, &d->bd_sigio);
1579 *(int *)addr = fgetown(&d->bd_sigio);
1582 /* This is deprecated, FIOSETOWN should be used instead. */
1584 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1587 /* This is deprecated, FIOGETOWN should be used instead. */
1589 *(int *)addr = -fgetown(&d->bd_sigio);
1592 case BIOCSRSIG: /* Set receive signal */
1596 sig = *(u_int *)addr;
1605 *(u_int *)addr = d->bd_sig;
1608 case BIOCGETBUFMODE:
1609 *(u_int *)addr = d->bd_bufmode;
1612 case BIOCSETBUFMODE:
1614 * Allow the buffering mode to be changed as long as we
1615 * haven't yet committed to a particular mode. Our
1616 * definition of commitment, for now, is whether or not a
1617 * buffer has been allocated or an interface attached, since
1618 * that's the point where things get tricky.
1620 switch (*(u_int *)addr) {
1621 case BPF_BUFMODE_BUFFER:
1624 case BPF_BUFMODE_ZBUF:
1625 if (bpf_zerocopy_enable)
1635 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1636 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1641 d->bd_bufmode = *(u_int *)addr;
1646 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1650 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1654 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1662 * Set d's packet filter program to fp. If this file already has a filter,
1663 * free it and replace it. Returns EINVAL for bogus requests.
1666 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1668 struct bpf_insn *fcode, *old;
1669 u_int wfilter, flen, size;
1671 bpf_jit_filter *ofunc, *jfunc;
1674 #ifdef COMPAT_FREEBSD32
1675 struct bpf_program32 *fp32;
1676 struct bpf_program fp_swab;
1678 if (cmd == BIOCSETWF32 || cmd == BIOCSETF32 || cmd == BIOCSETFNR32) {
1679 fp32 = (struct bpf_program32 *)fp;
1680 fp_swab.bf_len = fp32->bf_len;
1681 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1683 if (cmd == BIOCSETWF32)
1688 * Check new filter validness before acquiring any locks.
1689 * Allocate memory for new filter, if needed.
1692 if ((flen > bpf_maxinsns) || ((fp->bf_insns == NULL) && (flen != 0)))
1696 size = flen * sizeof(*fp->bf_insns);
1698 fcode = (struct bpf_insn *)malloc(size, M_BPF, M_WAITOK);
1700 fcode = NULL; /* Make compiler happy */
1703 if (fp->bf_insns != NULL)
1704 jfunc = bpf_jitter(fcode, flen);
1706 jfunc = NULL; /* Make compiler happy */
1711 if (cmd == BIOCSETWF) {
1712 old = d->bd_wfilter;
1719 old = d->bd_rfilter;
1721 ofunc = d->bd_bfilter;
1724 if (fp->bf_insns == NULL) {
1726 * Protect filter removal by interface lock.
1727 * Additionally, we are protected by global lock here.
1729 if (d->bd_bif != NULL)
1730 BPFIF_WLOCK(d->bd_bif);
1733 d->bd_wfilter = NULL;
1735 d->bd_rfilter = NULL;
1737 d->bd_bfilter = NULL;
1739 if (cmd == BIOCSETF)
1743 if (d->bd_bif != NULL)
1744 BPFIF_WUNLOCK(d->bd_bif);
1746 free((caddr_t)old, M_BPF);
1749 bpf_destroy_jit_filter(ofunc);
1755 if (copyin((caddr_t)fp->bf_insns, (caddr_t)fcode, size) == 0 &&
1756 bpf_validate(fcode, (int)flen)) {
1758 * Protect filter change by interface lock
1759 * Additionally, we are protected by global lock here.
1761 if (d->bd_bif != NULL)
1762 BPFIF_WLOCK(d->bd_bif);
1765 d->bd_wfilter = fcode;
1767 d->bd_rfilter = fcode;
1769 d->bd_bfilter = jfunc;
1771 if (cmd == BIOCSETF)
1775 * Do not require upgrade by first BIOCSETF
1776 * (used to set snaplen) by pcap_open_live()
1778 if ((d->bd_writer != 0) && (--d->bd_writer == 0))
1780 CTR4(KTR_NET, "%s: filter function set by pid %d, "
1781 "bd_writer counter %d, need_upgrade %d",
1782 __func__, d->bd_pid, d->bd_writer, need_upgrade);
1785 if (d->bd_bif != NULL)
1786 BPFIF_WUNLOCK(d->bd_bif);
1788 free((caddr_t)old, M_BPF);
1791 bpf_destroy_jit_filter(ofunc);
1794 /* Move d to active readers list */
1795 if (need_upgrade != 0)
1801 free((caddr_t)fcode, M_BPF);
1807 * Detach a file from its current interface (if attached at all) and attach
1808 * to the interface indicated by the name stored in ifr.
1809 * Return an errno or 0.
1812 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1815 struct ifnet *theywant;
1819 theywant = ifunit(ifr->ifr_name);
1820 if (theywant == NULL || theywant->if_bpf == NULL)
1823 bp = theywant->if_bpf;
1825 /* Check if interface is not being detached from BPF */
1827 if (bp->flags & BPFIF_FLAG_DYING) {
1834 * Behavior here depends on the buffering model. If we're using
1835 * kernel memory buffers, then we can allocate them here. If we're
1836 * using zero-copy, then the user process must have registered
1837 * buffers by the time we get here. If not, return an error.
1839 switch (d->bd_bufmode) {
1840 case BPF_BUFMODE_BUFFER:
1841 case BPF_BUFMODE_ZBUF:
1842 if (d->bd_sbuf == NULL)
1847 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1849 if (bp != d->bd_bif)
1858 * Support for select() and poll() system calls
1860 * Return true iff the specific operation will not block indefinitely.
1861 * Otherwise, return false but make a note that a selwakeup() must be done.
1864 bpfpoll(struct cdev *dev, int events, struct thread *td)
1869 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1871 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1874 * Refresh PID associated with this descriptor.
1876 revents = events & (POLLOUT | POLLWRNORM);
1878 BPF_PID_REFRESH(d, td);
1879 if (events & (POLLIN | POLLRDNORM)) {
1881 revents |= events & (POLLIN | POLLRDNORM);
1883 selrecord(td, &d->bd_sel);
1884 /* Start the read timeout if necessary. */
1885 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1886 callout_reset(&d->bd_callout, d->bd_rtout,
1888 d->bd_state = BPF_WAITING;
1897 * Support for kevent() system call. Register EVFILT_READ filters and
1898 * reject all others.
1901 bpfkqfilter(struct cdev *dev, struct knote *kn)
1905 if (devfs_get_cdevpriv((void **)&d) != 0 ||
1906 kn->kn_filter != EVFILT_READ)
1910 * Refresh PID associated with this descriptor.
1913 BPF_PID_REFRESH_CUR(d);
1914 kn->kn_fop = &bpfread_filtops;
1916 knlist_add(&d->bd_sel.si_note, kn, 1);
1923 filt_bpfdetach(struct knote *kn)
1925 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1927 knlist_remove(&d->bd_sel.si_note, kn, 0);
1931 filt_bpfread(struct knote *kn, long hint)
1933 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1936 BPFD_LOCK_ASSERT(d);
1937 ready = bpf_ready(d);
1939 kn->kn_data = d->bd_slen;
1941 kn->kn_data += d->bd_hlen;
1942 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1943 callout_reset(&d->bd_callout, d->bd_rtout,
1945 d->bd_state = BPF_WAITING;
1951 #define BPF_TSTAMP_NONE 0
1952 #define BPF_TSTAMP_FAST 1
1953 #define BPF_TSTAMP_NORMAL 2
1954 #define BPF_TSTAMP_EXTERN 3
1957 bpf_ts_quality(int tstype)
1960 if (tstype == BPF_T_NONE)
1961 return (BPF_TSTAMP_NONE);
1962 if ((tstype & BPF_T_FAST) != 0)
1963 return (BPF_TSTAMP_FAST);
1965 return (BPF_TSTAMP_NORMAL);
1969 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
1974 quality = bpf_ts_quality(tstype);
1975 if (quality == BPF_TSTAMP_NONE)
1979 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
1981 *bt = *(struct bintime *)(tag + 1);
1982 return (BPF_TSTAMP_EXTERN);
1985 if (quality == BPF_TSTAMP_NORMAL)
1994 * Incoming linkage from device drivers. Process the packet pkt, of length
1995 * pktlen, which is stored in a contiguous buffer. The packet is parsed
1996 * by each process' filter, and if accepted, stashed into the corresponding
2000 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2010 gottime = BPF_TSTAMP_NONE;
2014 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2016 * We are not using any locks for d here because:
2017 * 1) any filter change is protected by interface
2019 * 2) destroying/detaching d is protected by interface
2023 /* XXX: Do not protect counter for the sake of performance. */
2026 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2027 * way for the caller to indiciate to us whether this packet
2028 * is inbound or outbound. In the bpf_mtap() routines, we use
2029 * the interface pointers on the mbuf to figure it out.
2032 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2034 slen = (*(bf->func))(pkt, pktlen, pktlen);
2037 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2040 * Filter matches. Let's to acquire write lock.
2045 if (gottime < bpf_ts_quality(d->bd_tstamp))
2046 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2048 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2050 catchpacket(d, pkt, pktlen, slen,
2051 bpf_append_bytes, &bt);
2058 #define BPF_CHECK_DIRECTION(d, r, i) \
2059 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2060 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2063 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2064 * Locking model is explained in bpf_tap().
2067 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2077 /* Skip outgoing duplicate packets. */
2078 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2079 m->m_flags &= ~M_PROMISC;
2083 pktlen = m_length(m, NULL);
2084 gottime = BPF_TSTAMP_NONE;
2088 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2089 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2093 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2094 /* XXX We cannot handle multiple mbufs. */
2095 if (bf != NULL && m->m_next == NULL)
2096 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2099 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2104 if (gottime < bpf_ts_quality(d->bd_tstamp))
2105 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2107 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2109 catchpacket(d, (u_char *)m, pktlen, slen,
2110 bpf_append_mbuf, &bt);
2118 * Incoming linkage from device drivers, when packet is in
2119 * an mbuf chain and to be prepended by a contiguous header.
2122 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2130 /* Skip outgoing duplicate packets. */
2131 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2132 m->m_flags &= ~M_PROMISC;
2136 pktlen = m_length(m, NULL);
2138 * Craft on-stack mbuf suitable for passing to bpf_filter.
2139 * Note that we cut corners here; we only setup what's
2140 * absolutely needed--this mbuf should never go anywhere else.
2147 gottime = BPF_TSTAMP_NONE;
2151 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2152 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2155 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2160 if (gottime < bpf_ts_quality(d->bd_tstamp))
2161 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2163 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2165 catchpacket(d, (u_char *)&mb, pktlen, slen,
2166 bpf_append_mbuf, &bt);
2173 #undef BPF_CHECK_DIRECTION
2175 #undef BPF_TSTAMP_NONE
2176 #undef BPF_TSTAMP_FAST
2177 #undef BPF_TSTAMP_NORMAL
2178 #undef BPF_TSTAMP_EXTERN
2181 bpf_hdrlen(struct bpf_d *d)
2185 hdrlen = d->bd_bif->bif_hdrlen;
2186 #ifndef BURN_BRIDGES
2187 if (d->bd_tstamp == BPF_T_NONE ||
2188 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2189 #ifdef COMPAT_FREEBSD32
2191 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2194 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2197 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2198 #ifdef COMPAT_FREEBSD32
2200 hdrlen = BPF_WORDALIGN32(hdrlen);
2203 hdrlen = BPF_WORDALIGN(hdrlen);
2205 return (hdrlen - d->bd_bif->bif_hdrlen);
2209 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2213 struct timespec tsn;
2215 if ((tstype & BPF_T_MONOTONIC) == 0) {
2217 bintime_add(&bt2, &boottimebin);
2220 switch (BPF_T_FORMAT(tstype)) {
2221 case BPF_T_MICROTIME:
2222 bintime2timeval(bt, &tsm);
2223 ts->bt_sec = tsm.tv_sec;
2224 ts->bt_frac = tsm.tv_usec;
2226 case BPF_T_NANOTIME:
2227 bintime2timespec(bt, &tsn);
2228 ts->bt_sec = tsn.tv_sec;
2229 ts->bt_frac = tsn.tv_nsec;
2232 ts->bt_sec = bt->sec;
2233 ts->bt_frac = bt->frac;
2239 * Move the packet data from interface memory (pkt) into the
2240 * store buffer. "cpfn" is the routine called to do the actual data
2241 * transfer. bcopy is passed in to copy contiguous chunks, while
2242 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2243 * pkt is really an mbuf.
2246 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2247 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2250 struct bpf_xhdr hdr;
2251 #ifndef BURN_BRIDGES
2252 struct bpf_hdr hdr_old;
2253 #ifdef COMPAT_FREEBSD32
2254 struct bpf_hdr32 hdr32_old;
2257 int caplen, curlen, hdrlen, totlen;
2262 BPFD_LOCK_ASSERT(d);
2265 * Detect whether user space has released a buffer back to us, and if
2266 * so, move it from being a hold buffer to a free buffer. This may
2267 * not be the best place to do it (for example, we might only want to
2268 * run this check if we need the space), but for now it's a reliable
2271 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2272 d->bd_fbuf = d->bd_hbuf;
2275 bpf_buf_reclaimed(d);
2279 * Figure out how many bytes to move. If the packet is
2280 * greater or equal to the snapshot length, transfer that
2281 * much. Otherwise, transfer the whole packet (unless
2282 * we hit the buffer size limit).
2284 hdrlen = bpf_hdrlen(d);
2285 totlen = hdrlen + min(snaplen, pktlen);
2286 if (totlen > d->bd_bufsize)
2287 totlen = d->bd_bufsize;
2290 * Round up the end of the previous packet to the next longword.
2292 * Drop the packet if there's no room and no hope of room
2293 * If the packet would overflow the storage buffer or the storage
2294 * buffer is considered immutable by the buffer model, try to rotate
2295 * the buffer and wakeup pending processes.
2297 #ifdef COMPAT_FREEBSD32
2299 curlen = BPF_WORDALIGN32(d->bd_slen);
2302 curlen = BPF_WORDALIGN(d->bd_slen);
2303 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2304 if (d->bd_fbuf == NULL) {
2306 * There's no room in the store buffer, and no
2307 * prospect of room, so drop the packet. Notify the
2317 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2319 * Immediate mode is set, or the read timeout has already
2320 * expired during a select call. A packet arrived, so the
2321 * reader should be woken up.
2324 caplen = totlen - hdrlen;
2325 tstype = d->bd_tstamp;
2326 do_timestamp = tstype != BPF_T_NONE;
2327 #ifndef BURN_BRIDGES
2328 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2331 bpf_bintime2ts(bt, &ts, tstype);
2332 #ifdef COMPAT_FREEBSD32
2333 if (d->bd_compat32) {
2334 bzero(&hdr32_old, sizeof(hdr32_old));
2336 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2337 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2339 hdr32_old.bh_datalen = pktlen;
2340 hdr32_old.bh_hdrlen = hdrlen;
2341 hdr32_old.bh_caplen = caplen;
2342 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2347 bzero(&hdr_old, sizeof(hdr_old));
2349 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2350 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2352 hdr_old.bh_datalen = pktlen;
2353 hdr_old.bh_hdrlen = hdrlen;
2354 hdr_old.bh_caplen = caplen;
2355 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2362 * Append the bpf header. Note we append the actual header size, but
2363 * move forward the length of the header plus padding.
2365 bzero(&hdr, sizeof(hdr));
2367 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2368 hdr.bh_datalen = pktlen;
2369 hdr.bh_hdrlen = hdrlen;
2370 hdr.bh_caplen = caplen;
2371 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2374 * Copy the packet data into the store buffer and update its length.
2376 #ifndef BURN_BRIDGES
2379 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2380 d->bd_slen = curlen + totlen;
2387 * Free buffers currently in use by a descriptor.
2391 bpf_freed(struct bpf_d *d)
2395 * We don't need to lock out interrupts since this descriptor has
2396 * been detached from its interface and it yet hasn't been marked
2400 if (d->bd_rfilter != NULL) {
2401 free((caddr_t)d->bd_rfilter, M_BPF);
2403 if (d->bd_bfilter != NULL)
2404 bpf_destroy_jit_filter(d->bd_bfilter);
2407 if (d->bd_wfilter != NULL)
2408 free((caddr_t)d->bd_wfilter, M_BPF);
2409 mtx_destroy(&d->bd_lock);
2413 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2414 * fixed size of the link header (variable length headers not yet supported).
2417 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2420 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2424 * Attach an interface to bpf. ifp is a pointer to the structure
2425 * defining the interface to be attached, dlt is the link layer type,
2426 * and hdrlen is the fixed size of the link header (variable length
2427 * headers are not yet supporrted).
2430 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2434 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2438 LIST_INIT(&bp->bif_dlist);
2439 LIST_INIT(&bp->bif_wlist);
2442 rw_init(&bp->bif_lock, "bpf interface lock");
2443 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2447 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2450 bp->bif_hdrlen = hdrlen;
2453 if_printf(ifp, "bpf attached\n");
2457 * Detach bpf from an interface. This involves detaching each descriptor
2458 * associated with the interface. Notify each descriptor as it's detached
2459 * so that any sleepers wake up and get ENXIO.
2462 bpfdetach(struct ifnet *ifp)
2473 /* Find all bpf_if struct's which reference ifp and detach them. */
2475 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2476 if (ifp == bp->bif_ifp)
2480 LIST_REMOVE(bp, bif_next);
2486 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2487 bpf_detachd_locked(d);
2492 /* Free writer-only descriptors */
2493 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2494 bpf_detachd_locked(d);
2501 * Delay freing bp till interface is detached
2502 * and all routes through this interface are removed.
2503 * Mark bp as detached to restrict new consumers.
2506 bp->flags |= BPFIF_FLAG_DYING;
2509 } while (bp != NULL);
2514 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2519 * Interface departure handler
2522 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2526 if ((bp = ifp->if_bpf) == NULL)
2529 CTR3(KTR_NET, "%s: freing BPF instance %p for interface %p",
2533 rw_destroy(&bp->bif_lock);
2538 * Get a list of available data link type of the interface.
2541 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2547 ifp = d->bd_bif->bif_ifp;
2551 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2552 if (bp->bif_ifp != ifp)
2554 if (bfl->bfl_list != NULL) {
2555 if (n >= bfl->bfl_len) {
2559 error = copyout(&bp->bif_dlt,
2560 bfl->bfl_list + n, sizeof(u_int));
2570 * Set the data link type of a BPF instance.
2573 bpf_setdlt(struct bpf_d *d, u_int dlt)
2575 int error, opromisc;
2581 if (d->bd_bif->bif_dlt == dlt)
2583 ifp = d->bd_bif->bif_ifp;
2585 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2586 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2591 opromisc = d->bd_promisc;
2597 error = ifpromisc(bp->bif_ifp, 1);
2599 if_printf(bp->bif_ifp,
2600 "bpf_setdlt: ifpromisc failed (%d)\n",
2606 return (bp == NULL ? EINVAL : 0);
2610 bpf_drvinit(void *unused)
2614 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2615 LIST_INIT(&bpf_iflist);
2617 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2618 /* For compatibility */
2619 make_dev_alias(dev, "bpf0");
2621 /* Register interface departure handler */
2622 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2623 ifnet_departure_event, bpf_ifdetach, NULL,
2624 EVENTHANDLER_PRI_ANY);
2628 * Zero out the various packet counters associated with all of the bpf
2629 * descriptors. At some point, we will probably want to get a bit more
2630 * granular and allow the user to specify descriptors to be zeroed.
2633 bpf_zero_counters(void)
2639 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2641 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2657 * Fill filter statistics
2660 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2663 bzero(d, sizeof(*d));
2664 BPFD_LOCK_ASSERT(bd);
2665 d->bd_structsize = sizeof(*d);
2666 /* XXX: reading should be protected by global lock */
2667 d->bd_immediate = bd->bd_immediate;
2668 d->bd_promisc = bd->bd_promisc;
2669 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2670 d->bd_direction = bd->bd_direction;
2671 d->bd_feedback = bd->bd_feedback;
2672 d->bd_async = bd->bd_async;
2673 d->bd_rcount = bd->bd_rcount;
2674 d->bd_dcount = bd->bd_dcount;
2675 d->bd_fcount = bd->bd_fcount;
2676 d->bd_sig = bd->bd_sig;
2677 d->bd_slen = bd->bd_slen;
2678 d->bd_hlen = bd->bd_hlen;
2679 d->bd_bufsize = bd->bd_bufsize;
2680 d->bd_pid = bd->bd_pid;
2681 strlcpy(d->bd_ifname,
2682 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2683 d->bd_locked = bd->bd_locked;
2684 d->bd_wcount = bd->bd_wcount;
2685 d->bd_wdcount = bd->bd_wdcount;
2686 d->bd_wfcount = bd->bd_wfcount;
2687 d->bd_zcopy = bd->bd_zcopy;
2688 d->bd_bufmode = bd->bd_bufmode;
2692 * Handle `netstat -B' stats request
2695 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2697 struct xbpf_d *xbdbuf, *xbd, zerostats;
2703 * XXX This is not technically correct. It is possible for non
2704 * privileged users to open bpf devices. It would make sense
2705 * if the users who opened the devices were able to retrieve
2706 * the statistics for them, too.
2708 error = priv_check(req->td, PRIV_NET_BPF);
2712 * Check to see if the user is requesting that the counters be
2713 * zeroed out. Explicitly check that the supplied data is zeroed,
2714 * as we aren't allowing the user to set the counters currently.
2716 if (req->newptr != NULL) {
2717 if (req->newlen != sizeof(zerostats))
2719 bzero(&zerostats, sizeof(zerostats));
2721 if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2723 bpf_zero_counters();
2726 if (req->oldptr == NULL)
2727 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2728 if (bpf_bpfd_cnt == 0)
2729 return (SYSCTL_OUT(req, 0, 0));
2730 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2732 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2734 free(xbdbuf, M_BPF);
2738 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2740 /* Send writers-only first */
2741 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2742 xbd = &xbdbuf[index++];
2744 bpfstats_fill_xbpf(xbd, bd);
2747 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2748 xbd = &xbdbuf[index++];
2750 bpfstats_fill_xbpf(xbd, bd);
2756 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2757 free(xbdbuf, M_BPF);
2761 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2763 #else /* !DEV_BPF && !NETGRAPH_BPF */
2765 * NOP stubs to allow bpf-using drivers to load and function.
2767 * A 'better' implementation would allow the core bpf functionality
2768 * to be loaded at runtime.
2770 static struct bpf_if bp_null;
2773 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2778 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2783 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2788 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2791 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2795 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2798 *driverp = &bp_null;
2802 bpfdetach(struct ifnet *ifp)
2807 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2809 return -1; /* "no filter" behaviour */
2813 bpf_validate(const struct bpf_insn *f, int len)
2815 return 0; /* false */
2818 #endif /* !DEV_BPF && !NETGRAPH_BPF */