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 #ifdef COMPAT_FREEBSD32
1198 case BIOCGRTIMEOUT32:
1209 #ifdef COMPAT_FREEBSD32
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:
1238 CURVNET_SET(TD_TO_VNET(td));
1246 * Check for read packet available.
1266 if (d->bd_bif == NULL)
1269 ifp = d->bd_bif->bif_ifp;
1270 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1276 * Get buffer len [for read()].
1280 *(u_int *)addr = d->bd_bufsize;
1285 * Set buffer length.
1288 error = bpf_ioctl_sblen(d, (u_int *)addr);
1292 * Set link layer read filter.
1297 #ifdef COMPAT_FREEBSD32
1302 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1306 * Flush read packet buffer.
1315 * Put interface into promiscuous mode.
1318 if (d->bd_bif == NULL) {
1320 * No interface attached yet.
1325 if (d->bd_promisc == 0) {
1326 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1333 * Get current data link type.
1337 if (d->bd_bif == NULL)
1340 *(u_int *)addr = d->bd_bif->bif_dlt;
1345 * Get a list of supported data link types.
1347 #ifdef COMPAT_FREEBSD32
1348 case BIOCGDLTLIST32:
1350 struct bpf_dltlist32 *list32;
1351 struct bpf_dltlist dltlist;
1353 list32 = (struct bpf_dltlist32 *)addr;
1354 dltlist.bfl_len = list32->bfl_len;
1355 dltlist.bfl_list = PTRIN(list32->bfl_list);
1357 if (d->bd_bif == NULL)
1360 error = bpf_getdltlist(d, &dltlist);
1362 list32->bfl_len = dltlist.bfl_len;
1371 if (d->bd_bif == NULL)
1374 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1379 * Set data link type.
1383 if (d->bd_bif == NULL)
1386 error = bpf_setdlt(d, *(u_int *)addr);
1391 * Get interface name.
1395 if (d->bd_bif == NULL)
1398 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1399 struct ifreq *const ifr = (struct ifreq *)addr;
1401 strlcpy(ifr->ifr_name, ifp->if_xname,
1402 sizeof(ifr->ifr_name));
1412 error = bpf_setif(d, (struct ifreq *)addr);
1420 #ifdef COMPAT_FREEBSD32
1421 case BIOCSRTIMEOUT32:
1424 struct timeval *tv = (struct timeval *)addr;
1425 #ifdef COMPAT_FREEBSD32
1426 struct timeval32 *tv32;
1427 struct timeval tv64;
1429 if (cmd == BIOCSRTIMEOUT32) {
1430 tv32 = (struct timeval32 *)addr;
1432 tv->tv_sec = tv32->tv_sec;
1433 tv->tv_usec = tv32->tv_usec;
1436 tv = (struct timeval *)addr;
1439 * Subtract 1 tick from tvtohz() since this isn't
1442 if ((error = itimerfix(tv)) == 0)
1443 d->bd_rtout = tvtohz(tv) - 1;
1451 #ifdef COMPAT_FREEBSD32
1452 case BIOCGRTIMEOUT32:
1456 #ifdef COMPAT_FREEBSD32
1457 struct timeval32 *tv32;
1458 struct timeval tv64;
1460 if (cmd == BIOCGRTIMEOUT32)
1464 tv = (struct timeval *)addr;
1466 tv->tv_sec = d->bd_rtout / hz;
1467 tv->tv_usec = (d->bd_rtout % hz) * tick;
1468 #ifdef COMPAT_FREEBSD32
1469 if (cmd == BIOCGRTIMEOUT32) {
1470 tv32 = (struct timeval32 *)addr;
1471 tv32->tv_sec = tv->tv_sec;
1472 tv32->tv_usec = tv->tv_usec;
1484 struct bpf_stat *bs = (struct bpf_stat *)addr;
1486 /* XXXCSJP overflow */
1487 bs->bs_recv = d->bd_rcount;
1488 bs->bs_drop = d->bd_dcount;
1493 * Set immediate mode.
1497 d->bd_immediate = *(u_int *)addr;
1503 struct bpf_version *bv = (struct bpf_version *)addr;
1505 bv->bv_major = BPF_MAJOR_VERSION;
1506 bv->bv_minor = BPF_MINOR_VERSION;
1511 * Get "header already complete" flag
1515 *(u_int *)addr = d->bd_hdrcmplt;
1520 * Set "header already complete" flag
1524 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1529 * Get packet direction flag
1531 case BIOCGDIRECTION:
1533 *(u_int *)addr = d->bd_direction;
1538 * Set packet direction flag
1540 case BIOCSDIRECTION:
1544 direction = *(u_int *)addr;
1545 switch (direction) {
1550 d->bd_direction = direction;
1560 * Get packet timestamp format and resolution.
1564 *(u_int *)addr = d->bd_tstamp;
1569 * Set packet timestamp format and resolution.
1575 func = *(u_int *)addr;
1576 if (BPF_T_VALID(func))
1577 d->bd_tstamp = func;
1585 d->bd_feedback = *(u_int *)addr;
1595 case FIONBIO: /* Non-blocking I/O */
1598 case FIOASYNC: /* Send signal on receive packets */
1600 d->bd_async = *(int *)addr;
1606 * XXX: Add some sort of locking here?
1607 * fsetown() can sleep.
1609 error = fsetown(*(int *)addr, &d->bd_sigio);
1614 *(int *)addr = fgetown(&d->bd_sigio);
1618 /* This is deprecated, FIOSETOWN should be used instead. */
1620 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1623 /* This is deprecated, FIOGETOWN should be used instead. */
1625 *(int *)addr = -fgetown(&d->bd_sigio);
1628 case BIOCSRSIG: /* Set receive signal */
1632 sig = *(u_int *)addr;
1645 *(u_int *)addr = d->bd_sig;
1649 case BIOCGETBUFMODE:
1651 *(u_int *)addr = d->bd_bufmode;
1655 case BIOCSETBUFMODE:
1657 * Allow the buffering mode to be changed as long as we
1658 * haven't yet committed to a particular mode. Our
1659 * definition of commitment, for now, is whether or not a
1660 * buffer has been allocated or an interface attached, since
1661 * that's the point where things get tricky.
1663 switch (*(u_int *)addr) {
1664 case BPF_BUFMODE_BUFFER:
1667 case BPF_BUFMODE_ZBUF:
1668 if (bpf_zerocopy_enable)
1678 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1679 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1684 d->bd_bufmode = *(u_int *)addr;
1689 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1693 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1697 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1705 * Set d's packet filter program to fp. If this file already has a filter,
1706 * free it and replace it. Returns EINVAL for bogus requests.
1708 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1709 * since reading d->bd_bif can't be protected by d or interface lock due to
1712 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1713 * interface read lock to read all filers.
1717 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1719 #ifdef COMPAT_FREEBSD32
1720 struct bpf_program fp_swab;
1721 struct bpf_program32 *fp32;
1723 struct bpf_insn *fcode, *old;
1725 bpf_jit_filter *jfunc, *ofunc;
1731 #ifdef COMPAT_FREEBSD32
1736 fp32 = (struct bpf_program32 *)fp;
1737 fp_swab.bf_len = fp32->bf_len;
1738 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1754 jfunc = ofunc = NULL;
1759 * Check new filter validness before acquiring any locks.
1760 * Allocate memory for new filter, if needed.
1763 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1765 size = flen * sizeof(*fp->bf_insns);
1767 /* We're setting up new filter. Copy and check actual data. */
1768 fcode = malloc(size, M_BPF, M_WAITOK);
1769 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1770 !bpf_validate(fcode, flen)) {
1775 /* Filter is copied inside fcode and is perfectly valid. */
1776 jfunc = bpf_jitter(fcode, flen);
1783 * Set up new filter.
1784 * Protect filter change by interface lock.
1785 * Additionally, we are protected by global lock here.
1787 if (d->bd_bif != NULL)
1788 BPFIF_WLOCK(d->bd_bif);
1790 if (cmd == BIOCSETWF) {
1791 old = d->bd_wfilter;
1792 d->bd_wfilter = fcode;
1794 old = d->bd_rfilter;
1795 d->bd_rfilter = fcode;
1797 ofunc = d->bd_bfilter;
1798 d->bd_bfilter = jfunc;
1800 if (cmd == BIOCSETF)
1803 if (fcode != NULL) {
1805 * Do not require upgrade by first BIOCSETF
1806 * (used to set snaplen) by pcap_open_live().
1808 if (d->bd_writer != 0 && --d->bd_writer == 0)
1810 CTR4(KTR_NET, "%s: filter function set by pid %d, "
1811 "bd_writer counter %d, need_upgrade %d",
1812 __func__, d->bd_pid, d->bd_writer, need_upgrade);
1816 if (d->bd_bif != NULL)
1817 BPFIF_WUNLOCK(d->bd_bif);
1822 bpf_destroy_jit_filter(ofunc);
1825 /* Move d to active readers list. */
1834 * Detach a file from its current interface (if attached at all) and attach
1835 * to the interface indicated by the name stored in ifr.
1836 * Return an errno or 0.
1839 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1842 struct ifnet *theywant;
1846 theywant = ifunit(ifr->ifr_name);
1847 if (theywant == NULL || theywant->if_bpf == NULL)
1850 bp = theywant->if_bpf;
1852 /* Check if interface is not being detached from BPF */
1854 if (bp->flags & BPFIF_FLAG_DYING) {
1861 * Behavior here depends on the buffering model. If we're using
1862 * kernel memory buffers, then we can allocate them here. If we're
1863 * using zero-copy, then the user process must have registered
1864 * buffers by the time we get here. If not, return an error.
1866 switch (d->bd_bufmode) {
1867 case BPF_BUFMODE_BUFFER:
1868 case BPF_BUFMODE_ZBUF:
1869 if (d->bd_sbuf == NULL)
1874 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1876 if (bp != d->bd_bif)
1885 * Support for select() and poll() system calls
1887 * Return true iff the specific operation will not block indefinitely.
1888 * Otherwise, return false but make a note that a selwakeup() must be done.
1891 bpfpoll(struct cdev *dev, int events, struct thread *td)
1896 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1898 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1901 * Refresh PID associated with this descriptor.
1903 revents = events & (POLLOUT | POLLWRNORM);
1905 BPF_PID_REFRESH(d, td);
1906 if (events & (POLLIN | POLLRDNORM)) {
1908 revents |= events & (POLLIN | POLLRDNORM);
1910 selrecord(td, &d->bd_sel);
1911 /* Start the read timeout if necessary. */
1912 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1913 callout_reset(&d->bd_callout, d->bd_rtout,
1915 d->bd_state = BPF_WAITING;
1924 * Support for kevent() system call. Register EVFILT_READ filters and
1925 * reject all others.
1928 bpfkqfilter(struct cdev *dev, struct knote *kn)
1932 if (devfs_get_cdevpriv((void **)&d) != 0 ||
1933 kn->kn_filter != EVFILT_READ)
1937 * Refresh PID associated with this descriptor.
1940 BPF_PID_REFRESH_CUR(d);
1941 kn->kn_fop = &bpfread_filtops;
1943 knlist_add(&d->bd_sel.si_note, kn, 1);
1950 filt_bpfdetach(struct knote *kn)
1952 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1954 knlist_remove(&d->bd_sel.si_note, kn, 0);
1958 filt_bpfread(struct knote *kn, long hint)
1960 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1963 BPFD_LOCK_ASSERT(d);
1964 ready = bpf_ready(d);
1966 kn->kn_data = d->bd_slen;
1968 kn->kn_data += d->bd_hlen;
1969 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1970 callout_reset(&d->bd_callout, d->bd_rtout,
1972 d->bd_state = BPF_WAITING;
1978 #define BPF_TSTAMP_NONE 0
1979 #define BPF_TSTAMP_FAST 1
1980 #define BPF_TSTAMP_NORMAL 2
1981 #define BPF_TSTAMP_EXTERN 3
1984 bpf_ts_quality(int tstype)
1987 if (tstype == BPF_T_NONE)
1988 return (BPF_TSTAMP_NONE);
1989 if ((tstype & BPF_T_FAST) != 0)
1990 return (BPF_TSTAMP_FAST);
1992 return (BPF_TSTAMP_NORMAL);
1996 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2001 quality = bpf_ts_quality(tstype);
2002 if (quality == BPF_TSTAMP_NONE)
2006 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2008 *bt = *(struct bintime *)(tag + 1);
2009 return (BPF_TSTAMP_EXTERN);
2012 if (quality == BPF_TSTAMP_NORMAL)
2021 * Incoming linkage from device drivers. Process the packet pkt, of length
2022 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2023 * by each process' filter, and if accepted, stashed into the corresponding
2027 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2037 gottime = BPF_TSTAMP_NONE;
2041 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2043 * We are not using any locks for d here because:
2044 * 1) any filter change is protected by interface
2046 * 2) destroying/detaching d is protected by interface
2050 /* XXX: Do not protect counter for the sake of performance. */
2053 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2054 * way for the caller to indiciate to us whether this packet
2055 * is inbound or outbound. In the bpf_mtap() routines, we use
2056 * the interface pointers on the mbuf to figure it out.
2059 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2061 slen = (*(bf->func))(pkt, pktlen, pktlen);
2064 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2067 * Filter matches. Let's to acquire write lock.
2072 if (gottime < bpf_ts_quality(d->bd_tstamp))
2073 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2075 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2077 catchpacket(d, pkt, pktlen, slen,
2078 bpf_append_bytes, &bt);
2085 #define BPF_CHECK_DIRECTION(d, r, i) \
2086 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2087 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2090 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2091 * Locking model is explained in bpf_tap().
2094 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2104 /* Skip outgoing duplicate packets. */
2105 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2106 m->m_flags &= ~M_PROMISC;
2110 pktlen = m_length(m, NULL);
2111 gottime = BPF_TSTAMP_NONE;
2115 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2116 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2120 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2121 /* XXX We cannot handle multiple mbufs. */
2122 if (bf != NULL && m->m_next == NULL)
2123 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2126 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2131 if (gottime < bpf_ts_quality(d->bd_tstamp))
2132 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2134 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2136 catchpacket(d, (u_char *)m, pktlen, slen,
2137 bpf_append_mbuf, &bt);
2145 * Incoming linkage from device drivers, when packet is in
2146 * an mbuf chain and to be prepended by a contiguous header.
2149 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2157 /* Skip outgoing duplicate packets. */
2158 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2159 m->m_flags &= ~M_PROMISC;
2163 pktlen = m_length(m, NULL);
2165 * Craft on-stack mbuf suitable for passing to bpf_filter.
2166 * Note that we cut corners here; we only setup what's
2167 * absolutely needed--this mbuf should never go anywhere else.
2174 gottime = BPF_TSTAMP_NONE;
2178 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2179 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2182 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2187 if (gottime < bpf_ts_quality(d->bd_tstamp))
2188 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2190 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2192 catchpacket(d, (u_char *)&mb, pktlen, slen,
2193 bpf_append_mbuf, &bt);
2200 #undef BPF_CHECK_DIRECTION
2202 #undef BPF_TSTAMP_NONE
2203 #undef BPF_TSTAMP_FAST
2204 #undef BPF_TSTAMP_NORMAL
2205 #undef BPF_TSTAMP_EXTERN
2208 bpf_hdrlen(struct bpf_d *d)
2212 hdrlen = d->bd_bif->bif_hdrlen;
2213 #ifndef BURN_BRIDGES
2214 if (d->bd_tstamp == BPF_T_NONE ||
2215 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2216 #ifdef COMPAT_FREEBSD32
2218 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2221 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2224 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2225 #ifdef COMPAT_FREEBSD32
2227 hdrlen = BPF_WORDALIGN32(hdrlen);
2230 hdrlen = BPF_WORDALIGN(hdrlen);
2232 return (hdrlen - d->bd_bif->bif_hdrlen);
2236 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2240 struct timespec tsn;
2242 if ((tstype & BPF_T_MONOTONIC) == 0) {
2244 bintime_add(&bt2, &boottimebin);
2247 switch (BPF_T_FORMAT(tstype)) {
2248 case BPF_T_MICROTIME:
2249 bintime2timeval(bt, &tsm);
2250 ts->bt_sec = tsm.tv_sec;
2251 ts->bt_frac = tsm.tv_usec;
2253 case BPF_T_NANOTIME:
2254 bintime2timespec(bt, &tsn);
2255 ts->bt_sec = tsn.tv_sec;
2256 ts->bt_frac = tsn.tv_nsec;
2259 ts->bt_sec = bt->sec;
2260 ts->bt_frac = bt->frac;
2266 * Move the packet data from interface memory (pkt) into the
2267 * store buffer. "cpfn" is the routine called to do the actual data
2268 * transfer. bcopy is passed in to copy contiguous chunks, while
2269 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2270 * pkt is really an mbuf.
2273 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2274 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2277 struct bpf_xhdr hdr;
2278 #ifndef BURN_BRIDGES
2279 struct bpf_hdr hdr_old;
2280 #ifdef COMPAT_FREEBSD32
2281 struct bpf_hdr32 hdr32_old;
2284 int caplen, curlen, hdrlen, totlen;
2289 BPFD_LOCK_ASSERT(d);
2292 * Detect whether user space has released a buffer back to us, and if
2293 * so, move it from being a hold buffer to a free buffer. This may
2294 * not be the best place to do it (for example, we might only want to
2295 * run this check if we need the space), but for now it's a reliable
2298 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2299 d->bd_fbuf = d->bd_hbuf;
2302 bpf_buf_reclaimed(d);
2306 * Figure out how many bytes to move. If the packet is
2307 * greater or equal to the snapshot length, transfer that
2308 * much. Otherwise, transfer the whole packet (unless
2309 * we hit the buffer size limit).
2311 hdrlen = bpf_hdrlen(d);
2312 totlen = hdrlen + min(snaplen, pktlen);
2313 if (totlen > d->bd_bufsize)
2314 totlen = d->bd_bufsize;
2317 * Round up the end of the previous packet to the next longword.
2319 * Drop the packet if there's no room and no hope of room
2320 * If the packet would overflow the storage buffer or the storage
2321 * buffer is considered immutable by the buffer model, try to rotate
2322 * the buffer and wakeup pending processes.
2324 #ifdef COMPAT_FREEBSD32
2326 curlen = BPF_WORDALIGN32(d->bd_slen);
2329 curlen = BPF_WORDALIGN(d->bd_slen);
2330 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2331 if (d->bd_fbuf == NULL) {
2333 * There's no room in the store buffer, and no
2334 * prospect of room, so drop the packet. Notify the
2344 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2346 * Immediate mode is set, or the read timeout has already
2347 * expired during a select call. A packet arrived, so the
2348 * reader should be woken up.
2351 caplen = totlen - hdrlen;
2352 tstype = d->bd_tstamp;
2353 do_timestamp = tstype != BPF_T_NONE;
2354 #ifndef BURN_BRIDGES
2355 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2358 bpf_bintime2ts(bt, &ts, tstype);
2359 #ifdef COMPAT_FREEBSD32
2360 if (d->bd_compat32) {
2361 bzero(&hdr32_old, sizeof(hdr32_old));
2363 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2364 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2366 hdr32_old.bh_datalen = pktlen;
2367 hdr32_old.bh_hdrlen = hdrlen;
2368 hdr32_old.bh_caplen = caplen;
2369 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2374 bzero(&hdr_old, sizeof(hdr_old));
2376 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2377 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2379 hdr_old.bh_datalen = pktlen;
2380 hdr_old.bh_hdrlen = hdrlen;
2381 hdr_old.bh_caplen = caplen;
2382 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2389 * Append the bpf header. Note we append the actual header size, but
2390 * move forward the length of the header plus padding.
2392 bzero(&hdr, sizeof(hdr));
2394 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2395 hdr.bh_datalen = pktlen;
2396 hdr.bh_hdrlen = hdrlen;
2397 hdr.bh_caplen = caplen;
2398 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2401 * Copy the packet data into the store buffer and update its length.
2403 #ifndef BURN_BRIDGES
2406 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2407 d->bd_slen = curlen + totlen;
2414 * Free buffers currently in use by a descriptor.
2418 bpf_freed(struct bpf_d *d)
2422 * We don't need to lock out interrupts since this descriptor has
2423 * been detached from its interface and it yet hasn't been marked
2427 if (d->bd_rfilter != NULL) {
2428 free((caddr_t)d->bd_rfilter, M_BPF);
2430 if (d->bd_bfilter != NULL)
2431 bpf_destroy_jit_filter(d->bd_bfilter);
2434 if (d->bd_wfilter != NULL)
2435 free((caddr_t)d->bd_wfilter, M_BPF);
2436 mtx_destroy(&d->bd_lock);
2440 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2441 * fixed size of the link header (variable length headers not yet supported).
2444 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2447 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2451 * Attach an interface to bpf. ifp is a pointer to the structure
2452 * defining the interface to be attached, dlt is the link layer type,
2453 * and hdrlen is the fixed size of the link header (variable length
2454 * headers are not yet supporrted).
2457 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2461 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2465 LIST_INIT(&bp->bif_dlist);
2466 LIST_INIT(&bp->bif_wlist);
2469 rw_init(&bp->bif_lock, "bpf interface lock");
2470 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2474 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2477 bp->bif_hdrlen = hdrlen;
2480 if_printf(ifp, "bpf attached\n");
2484 * Detach bpf from an interface. This involves detaching each descriptor
2485 * associated with the interface. Notify each descriptor as it's detached
2486 * so that any sleepers wake up and get ENXIO.
2489 bpfdetach(struct ifnet *ifp)
2500 /* Find all bpf_if struct's which reference ifp and detach them. */
2502 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2503 if (ifp == bp->bif_ifp)
2507 LIST_REMOVE(bp, bif_next);
2513 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2514 bpf_detachd_locked(d);
2519 /* Free writer-only descriptors */
2520 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2521 bpf_detachd_locked(d);
2528 * Delay freing bp till interface is detached
2529 * and all routes through this interface are removed.
2530 * Mark bp as detached to restrict new consumers.
2533 bp->flags |= BPFIF_FLAG_DYING;
2536 } while (bp != NULL);
2541 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2546 * Interface departure handler.
2547 * Note departure event does not guarantee interface is going down.
2550 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2555 if ((bp = ifp->if_bpf) == NULL) {
2560 /* Check if bpfdetach() was called previously */
2561 if ((bp->flags & BPFIF_FLAG_DYING) == 0) {
2566 CTR3(KTR_NET, "%s: freing BPF instance %p for interface %p",
2572 rw_destroy(&bp->bif_lock);
2577 * Get a list of available data link type of the interface.
2580 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2588 ifp = d->bd_bif->bif_ifp;
2591 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2592 if (bp->bif_ifp != ifp)
2594 if (bfl->bfl_list != NULL) {
2595 if (n >= bfl->bfl_len)
2597 error = copyout(&bp->bif_dlt,
2598 bfl->bfl_list + n, sizeof(u_int));
2607 * Set the data link type of a BPF instance.
2610 bpf_setdlt(struct bpf_d *d, u_int dlt)
2612 int error, opromisc;
2618 if (d->bd_bif->bif_dlt == dlt)
2620 ifp = d->bd_bif->bif_ifp;
2622 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2623 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2628 opromisc = d->bd_promisc;
2634 error = ifpromisc(bp->bif_ifp, 1);
2636 if_printf(bp->bif_ifp,
2637 "bpf_setdlt: ifpromisc failed (%d)\n",
2643 return (bp == NULL ? EINVAL : 0);
2647 bpf_drvinit(void *unused)
2651 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2652 LIST_INIT(&bpf_iflist);
2654 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2655 /* For compatibility */
2656 make_dev_alias(dev, "bpf0");
2658 /* Register interface departure handler */
2659 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2660 ifnet_departure_event, bpf_ifdetach, NULL,
2661 EVENTHANDLER_PRI_ANY);
2665 * Zero out the various packet counters associated with all of the bpf
2666 * descriptors. At some point, we will probably want to get a bit more
2667 * granular and allow the user to specify descriptors to be zeroed.
2670 bpf_zero_counters(void)
2676 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2678 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2694 * Fill filter statistics
2697 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2700 bzero(d, sizeof(*d));
2701 BPFD_LOCK_ASSERT(bd);
2702 d->bd_structsize = sizeof(*d);
2703 /* XXX: reading should be protected by global lock */
2704 d->bd_immediate = bd->bd_immediate;
2705 d->bd_promisc = bd->bd_promisc;
2706 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2707 d->bd_direction = bd->bd_direction;
2708 d->bd_feedback = bd->bd_feedback;
2709 d->bd_async = bd->bd_async;
2710 d->bd_rcount = bd->bd_rcount;
2711 d->bd_dcount = bd->bd_dcount;
2712 d->bd_fcount = bd->bd_fcount;
2713 d->bd_sig = bd->bd_sig;
2714 d->bd_slen = bd->bd_slen;
2715 d->bd_hlen = bd->bd_hlen;
2716 d->bd_bufsize = bd->bd_bufsize;
2717 d->bd_pid = bd->bd_pid;
2718 strlcpy(d->bd_ifname,
2719 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2720 d->bd_locked = bd->bd_locked;
2721 d->bd_wcount = bd->bd_wcount;
2722 d->bd_wdcount = bd->bd_wdcount;
2723 d->bd_wfcount = bd->bd_wfcount;
2724 d->bd_zcopy = bd->bd_zcopy;
2725 d->bd_bufmode = bd->bd_bufmode;
2729 * Handle `netstat -B' stats request
2732 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2734 struct xbpf_d *xbdbuf, *xbd, zerostats;
2740 * XXX This is not technically correct. It is possible for non
2741 * privileged users to open bpf devices. It would make sense
2742 * if the users who opened the devices were able to retrieve
2743 * the statistics for them, too.
2745 error = priv_check(req->td, PRIV_NET_BPF);
2749 * Check to see if the user is requesting that the counters be
2750 * zeroed out. Explicitly check that the supplied data is zeroed,
2751 * as we aren't allowing the user to set the counters currently.
2753 if (req->newptr != NULL) {
2754 if (req->newlen != sizeof(zerostats))
2756 bzero(&zerostats, sizeof(zerostats));
2758 if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2760 bpf_zero_counters();
2763 if (req->oldptr == NULL)
2764 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2765 if (bpf_bpfd_cnt == 0)
2766 return (SYSCTL_OUT(req, 0, 0));
2767 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2769 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2771 free(xbdbuf, M_BPF);
2775 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2777 /* Send writers-only first */
2778 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2779 xbd = &xbdbuf[index++];
2781 bpfstats_fill_xbpf(xbd, bd);
2784 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2785 xbd = &xbdbuf[index++];
2787 bpfstats_fill_xbpf(xbd, bd);
2793 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2794 free(xbdbuf, M_BPF);
2798 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2800 #else /* !DEV_BPF && !NETGRAPH_BPF */
2802 * NOP stubs to allow bpf-using drivers to load and function.
2804 * A 'better' implementation would allow the core bpf functionality
2805 * to be loaded at runtime.
2807 static struct bpf_if bp_null;
2810 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2815 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2820 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2825 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2828 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2832 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2835 *driverp = &bp_null;
2839 bpfdetach(struct ifnet *ifp)
2844 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2846 return -1; /* "no filter" behaviour */
2850 bpf_validate(const struct bpf_insn *f, int len)
2852 return 0; /* false */
2855 #endif /* !DEV_BPF && !NETGRAPH_BPF */