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, bpf_freelist;
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;
525 if (len < hlen || len - hlen > ifp->if_mtu)
528 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
531 m->m_pkthdr.len = m->m_len = len;
534 error = uiomove(mtod(m, u_char *), len, uio);
538 slen = bpf_filter(wfilter, mtod(m, u_char *), len, len);
544 /* Check for multicast destination */
547 eh = mtod(m, struct ether_header *);
548 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
549 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
550 ETHER_ADDR_LEN) == 0)
551 m->m_flags |= M_BCAST;
553 m->m_flags |= M_MCAST;
559 * Make room for link header, and copy it to sockaddr
562 if (sockp->sa_family == AF_IEEE80211) {
564 * Collect true length from the parameter header
565 * NB: sockp is known to be zero'd so if we do a
566 * short copy unspecified parameters will be
568 * NB: packet may not be aligned after stripping
572 p = mtod(m, const struct ieee80211_bpf_params *);
574 if (hlen > sizeof(sockp->sa_data)) {
579 bcopy(m->m_data, sockp->sa_data, hlen);
590 * Attach file to the bpf interface, i.e. make d listen on bp.
593 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
600 * Save sysctl value to protect from sysctl change
603 op_w = V_bpf_optimize_writers || d->bd_writer;
605 if (d->bd_bif != NULL)
606 bpf_detachd_locked(d);
608 * Point d at bp, and add d to the interface's list.
609 * Since there are many applications using BPF for
610 * sending raw packets only (dhcpd, cdpd are good examples)
611 * we can delay adding d to the list of active listeners until
612 * some filter is configured.
621 /* Add to writers-only list */
622 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
624 * We decrement bd_writer on every filter set operation.
625 * First BIOCSETF is done by pcap_open_live() to set up
626 * snap length. After that appliation usually sets its own filter
630 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
637 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
638 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
641 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
645 * Add d to the list of active bp filters.
646 * Requires bpf_attachd() to be called before.
649 bpf_upgraded(struct bpf_d *d)
658 * Filter can be set several times without specifying interface.
659 * Mark d as reader and exit.
671 /* Remove from writers-only list */
672 LIST_REMOVE(d, bd_next);
673 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
674 /* Mark d as reader */
680 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
682 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
686 * Detach a file from its interface.
689 bpf_detachd(struct bpf_d *d)
692 bpf_detachd_locked(d);
697 bpf_detachd_locked(struct bpf_d *d)
703 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
707 /* Check if descriptor is attached */
708 if ((bp = d->bd_bif) == NULL)
714 /* Save bd_writer value */
715 error = d->bd_writer;
718 * Remove d from the interface's descriptor list.
720 LIST_REMOVE(d, bd_next);
729 /* Call event handler iff d is attached */
731 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
734 * Check if this descriptor had requested promiscuous mode.
735 * If so, turn it off.
739 CURVNET_SET(ifp->if_vnet);
740 error = ifpromisc(ifp, 0);
742 if (error != 0 && error != ENXIO) {
744 * ENXIO can happen if a pccard is unplugged
745 * Something is really wrong if we were able to put
746 * the driver into promiscuous mode, but can't
749 if_printf(bp->bif_ifp,
750 "bpf_detach: ifpromisc failed (%d)\n", error);
756 * Close the descriptor by detaching it from its interface,
757 * deallocating its buffers, and marking it free.
762 struct bpf_d *d = data;
765 if (d->bd_state == BPF_WAITING)
766 callout_stop(&d->bd_callout);
767 d->bd_state = BPF_IDLE;
769 funsetown(&d->bd_sigio);
772 mac_bpfdesc_destroy(d);
774 seldrain(&d->bd_sel);
775 knlist_destroy(&d->bd_sel.si_note);
776 callout_drain(&d->bd_callout);
782 * Open ethernet device. Returns ENXIO for illegal minor device number,
783 * EBUSY if file is open by another process.
787 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
792 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
793 error = devfs_set_cdevpriv(d, bpf_dtor);
800 * For historical reasons, perform a one-time initialization call to
801 * the buffer routines, even though we're not yet committed to a
802 * particular buffer method.
805 if ((flags & FREAD) == 0)
807 d->bd_hbuf_in_use = 0;
808 d->bd_bufmode = BPF_BUFMODE_BUFFER;
810 d->bd_direction = BPF_D_INOUT;
811 BPF_PID_REFRESH(d, td);
814 mac_bpfdesc_create(td->td_ucred, d);
816 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
817 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
818 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
824 * bpfread - read next chunk of packets from buffers
827 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
834 error = devfs_get_cdevpriv((void **)&d);
839 * Restrict application to use a buffer the same size as
842 if (uio->uio_resid != d->bd_bufsize)
845 non_block = ((ioflag & O_NONBLOCK) != 0);
848 BPF_PID_REFRESH_CUR(d);
849 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
853 if (d->bd_state == BPF_WAITING)
854 callout_stop(&d->bd_callout);
855 timed_out = (d->bd_state == BPF_TIMED_OUT);
856 d->bd_state = BPF_IDLE;
857 while (d->bd_hbuf_in_use) {
858 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
859 PRINET|PCATCH, "bd_hbuf", 0);
866 * If the hold buffer is empty, then do a timed sleep, which
867 * ends when the timeout expires or when enough packets
868 * have arrived to fill the store buffer.
870 while (d->bd_hbuf == NULL) {
871 if (d->bd_slen != 0) {
873 * A packet(s) either arrived since the previous
874 * read or arrived while we were asleep.
876 if (d->bd_immediate || non_block || timed_out) {
878 * Rotate the buffers and return what's here
879 * if we are in immediate mode, non-blocking
880 * flag is set, or this descriptor timed out.
888 * No data is available, check to see if the bpf device
889 * is still pointed at a real interface. If not, return
890 * ENXIO so that the userland process knows to rebind
891 * it before using it again.
893 if (d->bd_bif == NULL) {
900 return (EWOULDBLOCK);
902 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
904 if (error == EINTR || error == ERESTART) {
908 if (error == EWOULDBLOCK) {
910 * On a timeout, return what's in the buffer,
911 * which may be nothing. If there is something
912 * in the store buffer, we can rotate the buffers.
916 * We filled up the buffer in between
917 * getting the timeout and arriving
918 * here, so we don't need to rotate.
922 if (d->bd_slen == 0) {
931 * At this point, we know we have something in the hold slot.
933 d->bd_hbuf_in_use = 1;
937 * Move data from hold buffer into user space.
938 * We know the entire buffer is transferred since
939 * we checked above that the read buffer is bpf_bufsize bytes.
941 * We do not have to worry about simultaneous reads because
942 * we waited for sole access to the hold buffer above.
944 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
947 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
948 d->bd_fbuf = d->bd_hbuf;
951 bpf_buf_reclaimed(d);
952 d->bd_hbuf_in_use = 0;
953 wakeup(&d->bd_hbuf_in_use);
960 * If there are processes sleeping on this descriptor, wake them up.
963 bpf_wakeup(struct bpf_d *d)
967 if (d->bd_state == BPF_WAITING) {
968 callout_stop(&d->bd_callout);
969 d->bd_state = BPF_IDLE;
972 if (d->bd_async && d->bd_sig && d->bd_sigio)
973 pgsigio(&d->bd_sigio, d->bd_sig, 0);
975 selwakeuppri(&d->bd_sel, PRINET);
976 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
980 bpf_timed_out(void *arg)
982 struct bpf_d *d = (struct bpf_d *)arg;
986 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
988 if (d->bd_state == BPF_WAITING) {
989 d->bd_state = BPF_TIMED_OUT;
996 bpf_ready(struct bpf_d *d)
1001 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1003 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1010 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1014 struct mbuf *m, *mc;
1015 struct sockaddr dst;
1018 error = devfs_get_cdevpriv((void **)&d);
1022 BPF_PID_REFRESH_CUR(d);
1024 /* XXX: locking required */
1025 if (d->bd_bif == NULL) {
1030 ifp = d->bd_bif->bif_ifp;
1032 if ((ifp->if_flags & IFF_UP) == 0) {
1037 if (uio->uio_resid == 0) {
1042 bzero(&dst, sizeof(dst));
1045 /* XXX: bpf_movein() can sleep */
1046 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1047 &m, &dst, &hlen, d->bd_wfilter);
1054 dst.sa_family = pseudo_AF_HDRCMPLT;
1056 if (d->bd_feedback) {
1057 mc = m_dup(m, M_NOWAIT);
1059 mc->m_pkthdr.rcvif = ifp;
1060 /* Set M_PROMISC for outgoing packets to be discarded. */
1061 if (d->bd_direction == BPF_D_INOUT)
1062 m->m_flags |= M_PROMISC;
1066 m->m_pkthdr.len -= hlen;
1068 m->m_data += hlen; /* XXX */
1070 CURVNET_SET(ifp->if_vnet);
1073 mac_bpfdesc_create_mbuf(d, m);
1075 mac_bpfdesc_create_mbuf(d, mc);
1079 error = (*ifp->if_output)(ifp, m, &dst, NULL);
1085 (*ifp->if_input)(ifp, mc);
1095 * Reset a descriptor by flushing its packet buffer and clearing the receive
1096 * and drop counts. This is doable for kernel-only buffers, but with
1097 * zero-copy buffers, we can't write to (or rotate) buffers that are
1098 * currently owned by userspace. It would be nice if we could encapsulate
1099 * this logic in the buffer code rather than here.
1102 reset_d(struct bpf_d *d)
1105 BPFD_LOCK_ASSERT(d);
1107 while (d->bd_hbuf_in_use)
1108 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1110 if ((d->bd_hbuf != NULL) &&
1111 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1112 /* Free the hold buffer. */
1113 d->bd_fbuf = d->bd_hbuf;
1116 bpf_buf_reclaimed(d);
1118 if (bpf_canwritebuf(d))
1130 * FIONREAD Check for read packet available.
1131 * SIOCGIFADDR Get interface address - convenient hook to driver.
1132 * BIOCGBLEN Get buffer len [for read()].
1133 * BIOCSETF Set read filter.
1134 * BIOCSETFNR Set read filter without resetting descriptor.
1135 * BIOCSETWF Set write filter.
1136 * BIOCFLUSH Flush read packet buffer.
1137 * BIOCPROMISC Put interface into promiscuous mode.
1138 * BIOCGDLT Get link layer type.
1139 * BIOCGETIF Get interface name.
1140 * BIOCSETIF Set interface.
1141 * BIOCSRTIMEOUT Set read timeout.
1142 * BIOCGRTIMEOUT Get read timeout.
1143 * BIOCGSTATS Get packet stats.
1144 * BIOCIMMEDIATE Set immediate mode.
1145 * BIOCVERSION Get filter language version.
1146 * BIOCGHDRCMPLT Get "header already complete" flag
1147 * BIOCSHDRCMPLT Set "header already complete" flag
1148 * BIOCGDIRECTION Get packet direction flag
1149 * BIOCSDIRECTION Set packet direction flag
1150 * BIOCGTSTAMP Get time stamp format and resolution.
1151 * BIOCSTSTAMP Set time stamp format and resolution.
1152 * BIOCLOCK Set "locked" flag
1153 * BIOCFEEDBACK Set packet feedback mode.
1154 * BIOCSETZBUF Set current zero-copy buffer locations.
1155 * BIOCGETZMAX Get maximum zero-copy buffer size.
1156 * BIOCROTZBUF Force rotation of zero-copy buffer
1157 * BIOCSETBUFMODE Set buffer mode.
1158 * BIOCGETBUFMODE Get current buffer mode.
1162 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1168 error = devfs_get_cdevpriv((void **)&d);
1173 * Refresh PID associated with this descriptor.
1176 BPF_PID_REFRESH(d, td);
1177 if (d->bd_state == BPF_WAITING)
1178 callout_stop(&d->bd_callout);
1179 d->bd_state = BPF_IDLE;
1182 if (d->bd_locked == 1) {
1188 #ifdef COMPAT_FREEBSD32
1189 case BIOCGDLTLIST32:
1193 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1194 case BIOCGRTIMEOUT32:
1205 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1206 case BIOCSRTIMEOUT32:
1216 #ifdef COMPAT_FREEBSD32
1218 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1219 * that it will get 32-bit packet headers.
1225 case BIOCGDLTLIST32:
1226 case BIOCGRTIMEOUT32:
1227 case BIOCSRTIMEOUT32:
1234 CURVNET_SET(TD_TO_VNET(td));
1242 * Check for read packet available.
1250 while (d->bd_hbuf_in_use)
1251 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1252 PRINET, "bd_hbuf", 0);
1265 if (d->bd_bif == NULL)
1268 ifp = d->bd_bif->bif_ifp;
1269 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1275 * Get buffer len [for read()].
1279 *(u_int *)addr = d->bd_bufsize;
1284 * Set buffer length.
1287 error = bpf_ioctl_sblen(d, (u_int *)addr);
1291 * Set link layer read filter.
1296 #ifdef COMPAT_FREEBSD32
1301 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1305 * Flush read packet buffer.
1314 * Put interface into promiscuous mode.
1317 if (d->bd_bif == NULL) {
1319 * No interface attached yet.
1324 if (d->bd_promisc == 0) {
1325 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1332 * Get current data link type.
1336 if (d->bd_bif == NULL)
1339 *(u_int *)addr = d->bd_bif->bif_dlt;
1344 * Get a list of supported data link types.
1346 #ifdef COMPAT_FREEBSD32
1347 case BIOCGDLTLIST32:
1349 struct bpf_dltlist32 *list32;
1350 struct bpf_dltlist dltlist;
1352 list32 = (struct bpf_dltlist32 *)addr;
1353 dltlist.bfl_len = list32->bfl_len;
1354 dltlist.bfl_list = PTRIN(list32->bfl_list);
1356 if (d->bd_bif == NULL)
1359 error = bpf_getdltlist(d, &dltlist);
1361 list32->bfl_len = dltlist.bfl_len;
1370 if (d->bd_bif == NULL)
1373 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1378 * Set data link type.
1382 if (d->bd_bif == NULL)
1385 error = bpf_setdlt(d, *(u_int *)addr);
1390 * Get interface name.
1394 if (d->bd_bif == NULL)
1397 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1398 struct ifreq *const ifr = (struct ifreq *)addr;
1400 strlcpy(ifr->ifr_name, ifp->if_xname,
1401 sizeof(ifr->ifr_name));
1411 int alloc_buf, size;
1414 * Behavior here depends on the buffering model. If
1415 * we're using kernel memory buffers, then we can
1416 * allocate them here. If we're using zero-copy,
1417 * then the user process must have registered buffers
1418 * by the time we get here.
1422 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1427 size = d->bd_bufsize;
1428 error = bpf_buffer_ioctl_sblen(d, &size);
1433 error = bpf_setif(d, (struct ifreq *)addr);
1442 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1443 case BIOCSRTIMEOUT32:
1446 struct timeval *tv = (struct timeval *)addr;
1447 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1448 struct timeval32 *tv32;
1449 struct timeval tv64;
1451 if (cmd == BIOCSRTIMEOUT32) {
1452 tv32 = (struct timeval32 *)addr;
1454 tv->tv_sec = tv32->tv_sec;
1455 tv->tv_usec = tv32->tv_usec;
1458 tv = (struct timeval *)addr;
1461 * Subtract 1 tick from tvtohz() since this isn't
1464 if ((error = itimerfix(tv)) == 0)
1465 d->bd_rtout = tvtohz(tv) - 1;
1473 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1474 case BIOCGRTIMEOUT32:
1478 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1479 struct timeval32 *tv32;
1480 struct timeval tv64;
1482 if (cmd == BIOCGRTIMEOUT32)
1486 tv = (struct timeval *)addr;
1488 tv->tv_sec = d->bd_rtout / hz;
1489 tv->tv_usec = (d->bd_rtout % hz) * tick;
1490 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1491 if (cmd == BIOCGRTIMEOUT32) {
1492 tv32 = (struct timeval32 *)addr;
1493 tv32->tv_sec = tv->tv_sec;
1494 tv32->tv_usec = tv->tv_usec;
1506 struct bpf_stat *bs = (struct bpf_stat *)addr;
1508 /* XXXCSJP overflow */
1509 bs->bs_recv = d->bd_rcount;
1510 bs->bs_drop = d->bd_dcount;
1515 * Set immediate mode.
1519 d->bd_immediate = *(u_int *)addr;
1525 struct bpf_version *bv = (struct bpf_version *)addr;
1527 bv->bv_major = BPF_MAJOR_VERSION;
1528 bv->bv_minor = BPF_MINOR_VERSION;
1533 * Get "header already complete" flag
1537 *(u_int *)addr = d->bd_hdrcmplt;
1542 * Set "header already complete" flag
1546 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1551 * Get packet direction flag
1553 case BIOCGDIRECTION:
1555 *(u_int *)addr = d->bd_direction;
1560 * Set packet direction flag
1562 case BIOCSDIRECTION:
1566 direction = *(u_int *)addr;
1567 switch (direction) {
1572 d->bd_direction = direction;
1582 * Get packet timestamp format and resolution.
1586 *(u_int *)addr = d->bd_tstamp;
1591 * Set packet timestamp format and resolution.
1597 func = *(u_int *)addr;
1598 if (BPF_T_VALID(func))
1599 d->bd_tstamp = func;
1607 d->bd_feedback = *(u_int *)addr;
1617 case FIONBIO: /* Non-blocking I/O */
1620 case FIOASYNC: /* Send signal on receive packets */
1622 d->bd_async = *(int *)addr;
1628 * XXX: Add some sort of locking here?
1629 * fsetown() can sleep.
1631 error = fsetown(*(int *)addr, &d->bd_sigio);
1636 *(int *)addr = fgetown(&d->bd_sigio);
1640 /* This is deprecated, FIOSETOWN should be used instead. */
1642 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1645 /* This is deprecated, FIOGETOWN should be used instead. */
1647 *(int *)addr = -fgetown(&d->bd_sigio);
1650 case BIOCSRSIG: /* Set receive signal */
1654 sig = *(u_int *)addr;
1667 *(u_int *)addr = d->bd_sig;
1671 case BIOCGETBUFMODE:
1673 *(u_int *)addr = d->bd_bufmode;
1677 case BIOCSETBUFMODE:
1679 * Allow the buffering mode to be changed as long as we
1680 * haven't yet committed to a particular mode. Our
1681 * definition of commitment, for now, is whether or not a
1682 * buffer has been allocated or an interface attached, since
1683 * that's the point where things get tricky.
1685 switch (*(u_int *)addr) {
1686 case BPF_BUFMODE_BUFFER:
1689 case BPF_BUFMODE_ZBUF:
1690 if (bpf_zerocopy_enable)
1700 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1701 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1706 d->bd_bufmode = *(u_int *)addr;
1711 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1715 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1719 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1727 * Set d's packet filter program to fp. If this file already has a filter,
1728 * free it and replace it. Returns EINVAL for bogus requests.
1730 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1731 * since reading d->bd_bif can't be protected by d or interface lock due to
1734 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1735 * interface read lock to read all filers.
1739 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1741 #ifdef COMPAT_FREEBSD32
1742 struct bpf_program fp_swab;
1743 struct bpf_program32 *fp32;
1745 struct bpf_insn *fcode, *old;
1747 bpf_jit_filter *jfunc, *ofunc;
1753 #ifdef COMPAT_FREEBSD32
1758 fp32 = (struct bpf_program32 *)fp;
1759 fp_swab.bf_len = fp32->bf_len;
1760 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1776 jfunc = ofunc = NULL;
1781 * Check new filter validness before acquiring any locks.
1782 * Allocate memory for new filter, if needed.
1785 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1787 size = flen * sizeof(*fp->bf_insns);
1789 /* We're setting up new filter. Copy and check actual data. */
1790 fcode = malloc(size, M_BPF, M_WAITOK);
1791 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1792 !bpf_validate(fcode, flen)) {
1797 /* Filter is copied inside fcode and is perfectly valid. */
1798 jfunc = bpf_jitter(fcode, flen);
1805 * Set up new filter.
1806 * Protect filter change by interface lock.
1807 * Additionally, we are protected by global lock here.
1809 if (d->bd_bif != NULL)
1810 BPFIF_WLOCK(d->bd_bif);
1812 if (cmd == BIOCSETWF) {
1813 old = d->bd_wfilter;
1814 d->bd_wfilter = fcode;
1816 old = d->bd_rfilter;
1817 d->bd_rfilter = fcode;
1819 ofunc = d->bd_bfilter;
1820 d->bd_bfilter = jfunc;
1822 if (cmd == BIOCSETF)
1825 if (fcode != NULL) {
1827 * Do not require upgrade by first BIOCSETF
1828 * (used to set snaplen) by pcap_open_live().
1830 if (d->bd_writer != 0 && --d->bd_writer == 0)
1832 CTR4(KTR_NET, "%s: filter function set by pid %d, "
1833 "bd_writer counter %d, need_upgrade %d",
1834 __func__, d->bd_pid, d->bd_writer, need_upgrade);
1838 if (d->bd_bif != NULL)
1839 BPFIF_WUNLOCK(d->bd_bif);
1844 bpf_destroy_jit_filter(ofunc);
1847 /* Move d to active readers list. */
1856 * Detach a file from its current interface (if attached at all) and attach
1857 * to the interface indicated by the name stored in ifr.
1858 * Return an errno or 0.
1861 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1864 struct ifnet *theywant;
1868 theywant = ifunit(ifr->ifr_name);
1869 if (theywant == NULL || theywant->if_bpf == NULL)
1872 bp = theywant->if_bpf;
1874 /* Check if interface is not being detached from BPF */
1876 if (bp->flags & BPFIF_FLAG_DYING) {
1883 * At this point, we expect the buffer is already allocated. If not,
1886 switch (d->bd_bufmode) {
1887 case BPF_BUFMODE_BUFFER:
1888 case BPF_BUFMODE_ZBUF:
1889 if (d->bd_sbuf == NULL)
1894 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1896 if (bp != d->bd_bif)
1905 * Support for select() and poll() system calls
1907 * Return true iff the specific operation will not block indefinitely.
1908 * Otherwise, return false but make a note that a selwakeup() must be done.
1911 bpfpoll(struct cdev *dev, int events, struct thread *td)
1916 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1918 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1921 * Refresh PID associated with this descriptor.
1923 revents = events & (POLLOUT | POLLWRNORM);
1925 BPF_PID_REFRESH(d, td);
1926 if (events & (POLLIN | POLLRDNORM)) {
1928 revents |= events & (POLLIN | POLLRDNORM);
1930 selrecord(td, &d->bd_sel);
1931 /* Start the read timeout if necessary. */
1932 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1933 callout_reset(&d->bd_callout, d->bd_rtout,
1935 d->bd_state = BPF_WAITING;
1944 * Support for kevent() system call. Register EVFILT_READ filters and
1945 * reject all others.
1948 bpfkqfilter(struct cdev *dev, struct knote *kn)
1952 if (devfs_get_cdevpriv((void **)&d) != 0 ||
1953 kn->kn_filter != EVFILT_READ)
1957 * Refresh PID associated with this descriptor.
1960 BPF_PID_REFRESH_CUR(d);
1961 kn->kn_fop = &bpfread_filtops;
1963 knlist_add(&d->bd_sel.si_note, kn, 1);
1970 filt_bpfdetach(struct knote *kn)
1972 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1974 knlist_remove(&d->bd_sel.si_note, kn, 0);
1978 filt_bpfread(struct knote *kn, long hint)
1980 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1983 BPFD_LOCK_ASSERT(d);
1984 ready = bpf_ready(d);
1986 kn->kn_data = d->bd_slen;
1988 * Ignore the hold buffer if it is being copied to user space.
1990 if (!d->bd_hbuf_in_use && d->bd_hbuf)
1991 kn->kn_data += d->bd_hlen;
1992 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1993 callout_reset(&d->bd_callout, d->bd_rtout,
1995 d->bd_state = BPF_WAITING;
2001 #define BPF_TSTAMP_NONE 0
2002 #define BPF_TSTAMP_FAST 1
2003 #define BPF_TSTAMP_NORMAL 2
2004 #define BPF_TSTAMP_EXTERN 3
2007 bpf_ts_quality(int tstype)
2010 if (tstype == BPF_T_NONE)
2011 return (BPF_TSTAMP_NONE);
2012 if ((tstype & BPF_T_FAST) != 0)
2013 return (BPF_TSTAMP_FAST);
2015 return (BPF_TSTAMP_NORMAL);
2019 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2024 quality = bpf_ts_quality(tstype);
2025 if (quality == BPF_TSTAMP_NONE)
2029 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2031 *bt = *(struct bintime *)(tag + 1);
2032 return (BPF_TSTAMP_EXTERN);
2035 if (quality == BPF_TSTAMP_NORMAL)
2044 * Incoming linkage from device drivers. Process the packet pkt, of length
2045 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2046 * by each process' filter, and if accepted, stashed into the corresponding
2050 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2060 gottime = BPF_TSTAMP_NONE;
2064 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2066 * We are not using any locks for d here because:
2067 * 1) any filter change is protected by interface
2069 * 2) destroying/detaching d is protected by interface
2073 /* XXX: Do not protect counter for the sake of performance. */
2076 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2077 * way for the caller to indiciate to us whether this packet
2078 * is inbound or outbound. In the bpf_mtap() routines, we use
2079 * the interface pointers on the mbuf to figure it out.
2082 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2084 slen = (*(bf->func))(pkt, pktlen, pktlen);
2087 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2090 * Filter matches. Let's to acquire write lock.
2095 if (gottime < bpf_ts_quality(d->bd_tstamp))
2096 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2098 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2100 catchpacket(d, pkt, pktlen, slen,
2101 bpf_append_bytes, &bt);
2108 #define BPF_CHECK_DIRECTION(d, r, i) \
2109 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2110 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2113 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2114 * Locking model is explained in bpf_tap().
2117 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2127 /* Skip outgoing duplicate packets. */
2128 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2129 m->m_flags &= ~M_PROMISC;
2133 pktlen = m_length(m, NULL);
2134 gottime = BPF_TSTAMP_NONE;
2138 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2139 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2143 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2144 /* XXX We cannot handle multiple mbufs. */
2145 if (bf != NULL && m->m_next == NULL)
2146 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2149 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2154 if (gottime < bpf_ts_quality(d->bd_tstamp))
2155 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2157 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2159 catchpacket(d, (u_char *)m, pktlen, slen,
2160 bpf_append_mbuf, &bt);
2168 * Incoming linkage from device drivers, when packet is in
2169 * an mbuf chain and to be prepended by a contiguous header.
2172 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2180 /* Skip outgoing duplicate packets. */
2181 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2182 m->m_flags &= ~M_PROMISC;
2186 pktlen = m_length(m, NULL);
2188 * Craft on-stack mbuf suitable for passing to bpf_filter.
2189 * Note that we cut corners here; we only setup what's
2190 * absolutely needed--this mbuf should never go anywhere else.
2197 gottime = BPF_TSTAMP_NONE;
2201 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2202 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2205 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2210 if (gottime < bpf_ts_quality(d->bd_tstamp))
2211 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2213 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2215 catchpacket(d, (u_char *)&mb, pktlen, slen,
2216 bpf_append_mbuf, &bt);
2223 #undef BPF_CHECK_DIRECTION
2225 #undef BPF_TSTAMP_NONE
2226 #undef BPF_TSTAMP_FAST
2227 #undef BPF_TSTAMP_NORMAL
2228 #undef BPF_TSTAMP_EXTERN
2231 bpf_hdrlen(struct bpf_d *d)
2235 hdrlen = d->bd_bif->bif_hdrlen;
2236 #ifndef BURN_BRIDGES
2237 if (d->bd_tstamp == BPF_T_NONE ||
2238 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2239 #ifdef COMPAT_FREEBSD32
2241 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2244 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2247 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2248 #ifdef COMPAT_FREEBSD32
2250 hdrlen = BPF_WORDALIGN32(hdrlen);
2253 hdrlen = BPF_WORDALIGN(hdrlen);
2255 return (hdrlen - d->bd_bif->bif_hdrlen);
2259 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2263 struct timespec tsn;
2265 if ((tstype & BPF_T_MONOTONIC) == 0) {
2267 bintime_add(&bt2, &boottimebin);
2270 switch (BPF_T_FORMAT(tstype)) {
2271 case BPF_T_MICROTIME:
2272 bintime2timeval(bt, &tsm);
2273 ts->bt_sec = tsm.tv_sec;
2274 ts->bt_frac = tsm.tv_usec;
2276 case BPF_T_NANOTIME:
2277 bintime2timespec(bt, &tsn);
2278 ts->bt_sec = tsn.tv_sec;
2279 ts->bt_frac = tsn.tv_nsec;
2282 ts->bt_sec = bt->sec;
2283 ts->bt_frac = bt->frac;
2289 * Move the packet data from interface memory (pkt) into the
2290 * store buffer. "cpfn" is the routine called to do the actual data
2291 * transfer. bcopy is passed in to copy contiguous chunks, while
2292 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2293 * pkt is really an mbuf.
2296 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2297 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2300 struct bpf_xhdr hdr;
2301 #ifndef BURN_BRIDGES
2302 struct bpf_hdr hdr_old;
2303 #ifdef COMPAT_FREEBSD32
2304 struct bpf_hdr32 hdr32_old;
2307 int caplen, curlen, hdrlen, totlen;
2312 BPFD_LOCK_ASSERT(d);
2315 * Detect whether user space has released a buffer back to us, and if
2316 * so, move it from being a hold buffer to a free buffer. This may
2317 * not be the best place to do it (for example, we might only want to
2318 * run this check if we need the space), but for now it's a reliable
2321 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2322 d->bd_fbuf = d->bd_hbuf;
2325 bpf_buf_reclaimed(d);
2329 * Figure out how many bytes to move. If the packet is
2330 * greater or equal to the snapshot length, transfer that
2331 * much. Otherwise, transfer the whole packet (unless
2332 * we hit the buffer size limit).
2334 hdrlen = bpf_hdrlen(d);
2335 totlen = hdrlen + min(snaplen, pktlen);
2336 if (totlen > d->bd_bufsize)
2337 totlen = d->bd_bufsize;
2340 * Round up the end of the previous packet to the next longword.
2342 * Drop the packet if there's no room and no hope of room
2343 * If the packet would overflow the storage buffer or the storage
2344 * buffer is considered immutable by the buffer model, try to rotate
2345 * the buffer and wakeup pending processes.
2347 #ifdef COMPAT_FREEBSD32
2349 curlen = BPF_WORDALIGN32(d->bd_slen);
2352 curlen = BPF_WORDALIGN(d->bd_slen);
2353 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2354 if (d->bd_fbuf == NULL) {
2356 * There's no room in the store buffer, and no
2357 * prospect of room, so drop the packet. Notify the
2364 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2368 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2370 * Immediate mode is set, or the read timeout has already
2371 * expired during a select call. A packet arrived, so the
2372 * reader should be woken up.
2375 caplen = totlen - hdrlen;
2376 tstype = d->bd_tstamp;
2377 do_timestamp = tstype != BPF_T_NONE;
2378 #ifndef BURN_BRIDGES
2379 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2382 bpf_bintime2ts(bt, &ts, tstype);
2383 #ifdef COMPAT_FREEBSD32
2384 if (d->bd_compat32) {
2385 bzero(&hdr32_old, sizeof(hdr32_old));
2387 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2388 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2390 hdr32_old.bh_datalen = pktlen;
2391 hdr32_old.bh_hdrlen = hdrlen;
2392 hdr32_old.bh_caplen = caplen;
2393 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2398 bzero(&hdr_old, sizeof(hdr_old));
2400 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2401 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2403 hdr_old.bh_datalen = pktlen;
2404 hdr_old.bh_hdrlen = hdrlen;
2405 hdr_old.bh_caplen = caplen;
2406 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2413 * Append the bpf header. Note we append the actual header size, but
2414 * move forward the length of the header plus padding.
2416 bzero(&hdr, sizeof(hdr));
2418 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2419 hdr.bh_datalen = pktlen;
2420 hdr.bh_hdrlen = hdrlen;
2421 hdr.bh_caplen = caplen;
2422 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2425 * Copy the packet data into the store buffer and update its length.
2427 #ifndef BURN_BRIDGES
2430 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2431 d->bd_slen = curlen + totlen;
2438 * Free buffers currently in use by a descriptor.
2442 bpf_freed(struct bpf_d *d)
2446 * We don't need to lock out interrupts since this descriptor has
2447 * been detached from its interface and it yet hasn't been marked
2451 if (d->bd_rfilter != NULL) {
2452 free((caddr_t)d->bd_rfilter, M_BPF);
2454 if (d->bd_bfilter != NULL)
2455 bpf_destroy_jit_filter(d->bd_bfilter);
2458 if (d->bd_wfilter != NULL)
2459 free((caddr_t)d->bd_wfilter, M_BPF);
2460 mtx_destroy(&d->bd_lock);
2464 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2465 * fixed size of the link header (variable length headers not yet supported).
2468 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2471 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2475 * Attach an interface to bpf. ifp is a pointer to the structure
2476 * defining the interface to be attached, dlt is the link layer type,
2477 * and hdrlen is the fixed size of the link header (variable length
2478 * headers are not yet supporrted).
2481 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2485 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2489 LIST_INIT(&bp->bif_dlist);
2490 LIST_INIT(&bp->bif_wlist);
2493 rw_init(&bp->bif_lock, "bpf interface lock");
2494 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2498 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2501 bp->bif_hdrlen = hdrlen;
2504 if_printf(ifp, "bpf attached\n");
2508 * Detach bpf from an interface. This involves detaching each descriptor
2509 * associated with the interface. Notify each descriptor as it's detached
2510 * so that any sleepers wake up and get ENXIO.
2513 bpfdetach(struct ifnet *ifp)
2515 struct bpf_if *bp, *bp_temp;
2522 /* Find all bpf_if struct's which reference ifp and detach them. */
2523 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2524 if (ifp != bp->bif_ifp)
2527 LIST_REMOVE(bp, bif_next);
2528 /* Add to to-be-freed list */
2529 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2533 * Delay freeing bp till interface is detached
2534 * and all routes through this interface are removed.
2535 * Mark bp as detached to restrict new consumers.
2538 bp->flags |= BPFIF_FLAG_DYING;
2541 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2542 __func__, bp->bif_dlt, bp, ifp);
2544 /* Free common descriptors */
2545 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2546 bpf_detachd_locked(d);
2552 /* Free writer-only descriptors */
2553 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2554 bpf_detachd_locked(d);
2564 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2569 * Interface departure handler.
2570 * Note departure event does not guarantee interface is going down.
2571 * Interface renaming is currently done via departure/arrival event set.
2573 * Departure handled is called after all routes pointing to
2574 * given interface are removed and interface is in down state
2575 * restricting any packets to be sent/received. We assume it is now safe
2576 * to free data allocated by BPF.
2579 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2581 struct bpf_if *bp, *bp_temp;
2586 * Find matching entries in free list.
2587 * Nothing should be found if bpfdetach() was not called.
2589 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2590 if (ifp != bp->bif_ifp)
2593 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2596 LIST_REMOVE(bp, bif_next);
2598 rw_destroy(&bp->bif_lock);
2606 * Note that we cannot zero other pointers to
2607 * custom DLTs possibly used by given interface.
2614 * Get a list of available data link type of the interface.
2617 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2625 ifp = d->bd_bif->bif_ifp;
2628 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2629 if (bp->bif_ifp != ifp)
2631 if (bfl->bfl_list != NULL) {
2632 if (n >= bfl->bfl_len)
2634 error = copyout(&bp->bif_dlt,
2635 bfl->bfl_list + n, sizeof(u_int));
2644 * Set the data link type of a BPF instance.
2647 bpf_setdlt(struct bpf_d *d, u_int dlt)
2649 int error, opromisc;
2655 if (d->bd_bif->bif_dlt == dlt)
2657 ifp = d->bd_bif->bif_ifp;
2659 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2660 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2665 opromisc = d->bd_promisc;
2671 error = ifpromisc(bp->bif_ifp, 1);
2673 if_printf(bp->bif_ifp,
2674 "bpf_setdlt: ifpromisc failed (%d)\n",
2680 return (bp == NULL ? EINVAL : 0);
2684 bpf_drvinit(void *unused)
2688 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2689 LIST_INIT(&bpf_iflist);
2690 LIST_INIT(&bpf_freelist);
2692 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2693 /* For compatibility */
2694 make_dev_alias(dev, "bpf0");
2696 /* Register interface departure handler */
2697 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2698 ifnet_departure_event, bpf_ifdetach, NULL,
2699 EVENTHANDLER_PRI_ANY);
2703 * Zero out the various packet counters associated with all of the bpf
2704 * descriptors. At some point, we will probably want to get a bit more
2705 * granular and allow the user to specify descriptors to be zeroed.
2708 bpf_zero_counters(void)
2714 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2716 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2732 * Fill filter statistics
2735 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2738 bzero(d, sizeof(*d));
2739 BPFD_LOCK_ASSERT(bd);
2740 d->bd_structsize = sizeof(*d);
2741 /* XXX: reading should be protected by global lock */
2742 d->bd_immediate = bd->bd_immediate;
2743 d->bd_promisc = bd->bd_promisc;
2744 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2745 d->bd_direction = bd->bd_direction;
2746 d->bd_feedback = bd->bd_feedback;
2747 d->bd_async = bd->bd_async;
2748 d->bd_rcount = bd->bd_rcount;
2749 d->bd_dcount = bd->bd_dcount;
2750 d->bd_fcount = bd->bd_fcount;
2751 d->bd_sig = bd->bd_sig;
2752 d->bd_slen = bd->bd_slen;
2753 d->bd_hlen = bd->bd_hlen;
2754 d->bd_bufsize = bd->bd_bufsize;
2755 d->bd_pid = bd->bd_pid;
2756 strlcpy(d->bd_ifname,
2757 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2758 d->bd_locked = bd->bd_locked;
2759 d->bd_wcount = bd->bd_wcount;
2760 d->bd_wdcount = bd->bd_wdcount;
2761 d->bd_wfcount = bd->bd_wfcount;
2762 d->bd_zcopy = bd->bd_zcopy;
2763 d->bd_bufmode = bd->bd_bufmode;
2767 * Handle `netstat -B' stats request
2770 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2772 static const struct xbpf_d zerostats;
2773 struct xbpf_d *xbdbuf, *xbd, tempstats;
2779 * XXX This is not technically correct. It is possible for non
2780 * privileged users to open bpf devices. It would make sense
2781 * if the users who opened the devices were able to retrieve
2782 * the statistics for them, too.
2784 error = priv_check(req->td, PRIV_NET_BPF);
2788 * Check to see if the user is requesting that the counters be
2789 * zeroed out. Explicitly check that the supplied data is zeroed,
2790 * as we aren't allowing the user to set the counters currently.
2792 if (req->newptr != NULL) {
2793 if (req->newlen != sizeof(tempstats))
2795 memset(&tempstats, 0, sizeof(tempstats));
2796 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2799 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2801 bpf_zero_counters();
2804 if (req->oldptr == NULL)
2805 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2806 if (bpf_bpfd_cnt == 0)
2807 return (SYSCTL_OUT(req, 0, 0));
2808 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2810 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2812 free(xbdbuf, M_BPF);
2816 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2818 /* Send writers-only first */
2819 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2820 xbd = &xbdbuf[index++];
2822 bpfstats_fill_xbpf(xbd, bd);
2825 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2826 xbd = &xbdbuf[index++];
2828 bpfstats_fill_xbpf(xbd, bd);
2834 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2835 free(xbdbuf, M_BPF);
2839 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2841 #else /* !DEV_BPF && !NETGRAPH_BPF */
2843 * NOP stubs to allow bpf-using drivers to load and function.
2845 * A 'better' implementation would allow the core bpf functionality
2846 * to be loaded at runtime.
2848 static struct bpf_if bp_null;
2851 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2856 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2861 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2866 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2869 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2873 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2876 *driverp = &bp_null;
2880 bpfdetach(struct ifnet *ifp)
2885 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2887 return -1; /* "no filter" behaviour */
2891 bpf_validate(const struct bpf_insn *f, int len)
2893 return 0; /* false */
2896 #endif /* !DEV_BPF && !NETGRAPH_BPF */