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_hbuf_in_use = 0;
823 d->bd_bufmode = BPF_BUFMODE_BUFFER;
825 d->bd_direction = BPF_D_INOUT;
826 BPF_PID_REFRESH(d, td);
829 mac_bpfdesc_create(td->td_ucred, d);
831 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
832 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
833 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
835 /* Allocate default buffers */
836 size = d->bd_bufsize;
837 bpf_buffer_ioctl_sblen(d, &size);
843 * bpfread - read next chunk of packets from buffers
846 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
853 error = devfs_get_cdevpriv((void **)&d);
858 * Restrict application to use a buffer the same size as
861 if (uio->uio_resid != d->bd_bufsize)
864 non_block = ((ioflag & O_NONBLOCK) != 0);
867 BPF_PID_REFRESH_CUR(d);
868 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
872 if (d->bd_state == BPF_WAITING)
873 callout_stop(&d->bd_callout);
874 timed_out = (d->bd_state == BPF_TIMED_OUT);
875 d->bd_state = BPF_IDLE;
876 while (d->bd_hbuf_in_use) {
877 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
878 PRINET|PCATCH, "bd_hbuf", 0);
885 * If the hold buffer is empty, then do a timed sleep, which
886 * ends when the timeout expires or when enough packets
887 * have arrived to fill the store buffer.
889 while (d->bd_hbuf == NULL) {
890 if (d->bd_slen != 0) {
892 * A packet(s) either arrived since the previous
893 * read or arrived while we were asleep.
895 if (d->bd_immediate || non_block || timed_out) {
897 * Rotate the buffers and return what's here
898 * if we are in immediate mode, non-blocking
899 * flag is set, or this descriptor timed out.
907 * No data is available, check to see if the bpf device
908 * is still pointed at a real interface. If not, return
909 * ENXIO so that the userland process knows to rebind
910 * it before using it again.
912 if (d->bd_bif == NULL) {
919 return (EWOULDBLOCK);
921 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
923 if (error == EINTR || error == ERESTART) {
927 if (error == EWOULDBLOCK) {
929 * On a timeout, return what's in the buffer,
930 * which may be nothing. If there is something
931 * in the store buffer, we can rotate the buffers.
935 * We filled up the buffer in between
936 * getting the timeout and arriving
937 * here, so we don't need to rotate.
941 if (d->bd_slen == 0) {
950 * At this point, we know we have something in the hold slot.
952 d->bd_hbuf_in_use = 1;
956 * Move data from hold buffer into user space.
957 * We know the entire buffer is transferred since
958 * we checked above that the read buffer is bpf_bufsize bytes.
960 * We do not have to worry about simultaneous reads because
961 * we waited for sole access to the hold buffer above.
963 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
966 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
967 d->bd_fbuf = d->bd_hbuf;
970 bpf_buf_reclaimed(d);
971 d->bd_hbuf_in_use = 0;
972 wakeup(&d->bd_hbuf_in_use);
979 * If there are processes sleeping on this descriptor, wake them up.
982 bpf_wakeup(struct bpf_d *d)
986 if (d->bd_state == BPF_WAITING) {
987 callout_stop(&d->bd_callout);
988 d->bd_state = BPF_IDLE;
991 if (d->bd_async && d->bd_sig && d->bd_sigio)
992 pgsigio(&d->bd_sigio, d->bd_sig, 0);
994 selwakeuppri(&d->bd_sel, PRINET);
995 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
999 bpf_timed_out(void *arg)
1001 struct bpf_d *d = (struct bpf_d *)arg;
1003 BPFD_LOCK_ASSERT(d);
1005 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1007 if (d->bd_state == BPF_WAITING) {
1008 d->bd_state = BPF_TIMED_OUT;
1009 if (d->bd_slen != 0)
1015 bpf_ready(struct bpf_d *d)
1018 BPFD_LOCK_ASSERT(d);
1020 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1022 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1029 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1033 struct mbuf *m, *mc;
1034 struct sockaddr dst;
1037 error = devfs_get_cdevpriv((void **)&d);
1041 BPF_PID_REFRESH_CUR(d);
1043 /* XXX: locking required */
1044 if (d->bd_bif == NULL) {
1049 ifp = d->bd_bif->bif_ifp;
1051 if ((ifp->if_flags & IFF_UP) == 0) {
1056 if (uio->uio_resid == 0) {
1061 bzero(&dst, sizeof(dst));
1064 /* XXX: bpf_movein() can sleep */
1065 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1066 &m, &dst, &hlen, d->bd_wfilter);
1073 dst.sa_family = pseudo_AF_HDRCMPLT;
1075 if (d->bd_feedback) {
1076 mc = m_dup(m, M_DONTWAIT);
1078 mc->m_pkthdr.rcvif = ifp;
1079 /* Set M_PROMISC for outgoing packets to be discarded. */
1080 if (d->bd_direction == BPF_D_INOUT)
1081 m->m_flags |= M_PROMISC;
1085 m->m_pkthdr.len -= hlen;
1087 m->m_data += hlen; /* XXX */
1089 CURVNET_SET(ifp->if_vnet);
1092 mac_bpfdesc_create_mbuf(d, m);
1094 mac_bpfdesc_create_mbuf(d, mc);
1098 error = (*ifp->if_output)(ifp, m, &dst, NULL);
1104 (*ifp->if_input)(ifp, mc);
1114 * Reset a descriptor by flushing its packet buffer and clearing the receive
1115 * and drop counts. This is doable for kernel-only buffers, but with
1116 * zero-copy buffers, we can't write to (or rotate) buffers that are
1117 * currently owned by userspace. It would be nice if we could encapsulate
1118 * this logic in the buffer code rather than here.
1121 reset_d(struct bpf_d *d)
1124 BPFD_LOCK_ASSERT(d);
1126 while (d->bd_hbuf_in_use)
1127 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1129 if ((d->bd_hbuf != NULL) &&
1130 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1131 /* Free the hold buffer. */
1132 d->bd_fbuf = d->bd_hbuf;
1135 bpf_buf_reclaimed(d);
1137 if (bpf_canwritebuf(d))
1149 * FIONREAD Check for read packet available.
1150 * SIOCGIFADDR Get interface address - convenient hook to driver.
1151 * BIOCGBLEN Get buffer len [for read()].
1152 * BIOCSETF Set read filter.
1153 * BIOCSETFNR Set read filter without resetting descriptor.
1154 * BIOCSETWF Set write filter.
1155 * BIOCFLUSH Flush read packet buffer.
1156 * BIOCPROMISC Put interface into promiscuous mode.
1157 * BIOCGDLT Get link layer type.
1158 * BIOCGETIF Get interface name.
1159 * BIOCSETIF Set interface.
1160 * BIOCSRTIMEOUT Set read timeout.
1161 * BIOCGRTIMEOUT Get read timeout.
1162 * BIOCGSTATS Get packet stats.
1163 * BIOCIMMEDIATE Set immediate mode.
1164 * BIOCVERSION Get filter language version.
1165 * BIOCGHDRCMPLT Get "header already complete" flag
1166 * BIOCSHDRCMPLT Set "header already complete" flag
1167 * BIOCGDIRECTION Get packet direction flag
1168 * BIOCSDIRECTION Set packet direction flag
1169 * BIOCGTSTAMP Get time stamp format and resolution.
1170 * BIOCSTSTAMP Set time stamp format and resolution.
1171 * BIOCLOCK Set "locked" flag
1172 * BIOCFEEDBACK Set packet feedback mode.
1173 * BIOCSETZBUF Set current zero-copy buffer locations.
1174 * BIOCGETZMAX Get maximum zero-copy buffer size.
1175 * BIOCROTZBUF Force rotation of zero-copy buffer
1176 * BIOCSETBUFMODE Set buffer mode.
1177 * BIOCGETBUFMODE Get current buffer mode.
1181 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1187 error = devfs_get_cdevpriv((void **)&d);
1192 * Refresh PID associated with this descriptor.
1195 BPF_PID_REFRESH(d, td);
1196 if (d->bd_state == BPF_WAITING)
1197 callout_stop(&d->bd_callout);
1198 d->bd_state = BPF_IDLE;
1201 if (d->bd_locked == 1) {
1207 #ifdef COMPAT_FREEBSD32
1208 case BIOCGDLTLIST32:
1212 #ifdef COMPAT_FREEBSD32
1213 case BIOCGRTIMEOUT32:
1224 #ifdef COMPAT_FREEBSD32
1225 case BIOCSRTIMEOUT32:
1235 #ifdef COMPAT_FREEBSD32
1237 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1238 * that it will get 32-bit packet headers.
1244 case BIOCGDLTLIST32:
1245 case BIOCGRTIMEOUT32:
1246 case BIOCSRTIMEOUT32:
1253 CURVNET_SET(TD_TO_VNET(td));
1261 * Check for read packet available.
1269 while (d->bd_hbuf_in_use)
1270 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1271 PRINET, "bd_hbuf", 0);
1284 if (d->bd_bif == NULL)
1287 ifp = d->bd_bif->bif_ifp;
1288 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1294 * Get buffer len [for read()].
1298 *(u_int *)addr = d->bd_bufsize;
1303 * Set buffer length.
1306 error = bpf_ioctl_sblen(d, (u_int *)addr);
1310 * Set link layer read filter.
1315 #ifdef COMPAT_FREEBSD32
1320 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1324 * Flush read packet buffer.
1333 * Put interface into promiscuous mode.
1336 if (d->bd_bif == NULL) {
1338 * No interface attached yet.
1343 if (d->bd_promisc == 0) {
1344 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1351 * Get current data link type.
1355 if (d->bd_bif == NULL)
1358 *(u_int *)addr = d->bd_bif->bif_dlt;
1363 * Get a list of supported data link types.
1365 #ifdef COMPAT_FREEBSD32
1366 case BIOCGDLTLIST32:
1368 struct bpf_dltlist32 *list32;
1369 struct bpf_dltlist dltlist;
1371 list32 = (struct bpf_dltlist32 *)addr;
1372 dltlist.bfl_len = list32->bfl_len;
1373 dltlist.bfl_list = PTRIN(list32->bfl_list);
1375 if (d->bd_bif == NULL)
1378 error = bpf_getdltlist(d, &dltlist);
1380 list32->bfl_len = dltlist.bfl_len;
1389 if (d->bd_bif == NULL)
1392 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1397 * Set data link type.
1401 if (d->bd_bif == NULL)
1404 error = bpf_setdlt(d, *(u_int *)addr);
1409 * Get interface name.
1413 if (d->bd_bif == NULL)
1416 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1417 struct ifreq *const ifr = (struct ifreq *)addr;
1419 strlcpy(ifr->ifr_name, ifp->if_xname,
1420 sizeof(ifr->ifr_name));
1430 error = bpf_setif(d, (struct ifreq *)addr);
1438 #ifdef COMPAT_FREEBSD32
1439 case BIOCSRTIMEOUT32:
1442 struct timeval *tv = (struct timeval *)addr;
1443 #ifdef COMPAT_FREEBSD32
1444 struct timeval32 *tv32;
1445 struct timeval tv64;
1447 if (cmd == BIOCSRTIMEOUT32) {
1448 tv32 = (struct timeval32 *)addr;
1450 tv->tv_sec = tv32->tv_sec;
1451 tv->tv_usec = tv32->tv_usec;
1454 tv = (struct timeval *)addr;
1457 * Subtract 1 tick from tvtohz() since this isn't
1460 if ((error = itimerfix(tv)) == 0)
1461 d->bd_rtout = tvtohz(tv) - 1;
1469 #ifdef COMPAT_FREEBSD32
1470 case BIOCGRTIMEOUT32:
1474 #ifdef COMPAT_FREEBSD32
1475 struct timeval32 *tv32;
1476 struct timeval tv64;
1478 if (cmd == BIOCGRTIMEOUT32)
1482 tv = (struct timeval *)addr;
1484 tv->tv_sec = d->bd_rtout / hz;
1485 tv->tv_usec = (d->bd_rtout % hz) * tick;
1486 #ifdef COMPAT_FREEBSD32
1487 if (cmd == BIOCGRTIMEOUT32) {
1488 tv32 = (struct timeval32 *)addr;
1489 tv32->tv_sec = tv->tv_sec;
1490 tv32->tv_usec = tv->tv_usec;
1502 struct bpf_stat *bs = (struct bpf_stat *)addr;
1504 /* XXXCSJP overflow */
1505 bs->bs_recv = d->bd_rcount;
1506 bs->bs_drop = d->bd_dcount;
1511 * Set immediate mode.
1515 d->bd_immediate = *(u_int *)addr;
1521 struct bpf_version *bv = (struct bpf_version *)addr;
1523 bv->bv_major = BPF_MAJOR_VERSION;
1524 bv->bv_minor = BPF_MINOR_VERSION;
1529 * Get "header already complete" flag
1533 *(u_int *)addr = d->bd_hdrcmplt;
1538 * Set "header already complete" flag
1542 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1547 * Get packet direction flag
1549 case BIOCGDIRECTION:
1551 *(u_int *)addr = d->bd_direction;
1556 * Set packet direction flag
1558 case BIOCSDIRECTION:
1562 direction = *(u_int *)addr;
1563 switch (direction) {
1568 d->bd_direction = direction;
1578 * Get packet timestamp format and resolution.
1582 *(u_int *)addr = d->bd_tstamp;
1587 * Set packet timestamp format and resolution.
1593 func = *(u_int *)addr;
1594 if (BPF_T_VALID(func))
1595 d->bd_tstamp = func;
1603 d->bd_feedback = *(u_int *)addr;
1613 case FIONBIO: /* Non-blocking I/O */
1616 case FIOASYNC: /* Send signal on receive packets */
1618 d->bd_async = *(int *)addr;
1624 * XXX: Add some sort of locking here?
1625 * fsetown() can sleep.
1627 error = fsetown(*(int *)addr, &d->bd_sigio);
1632 *(int *)addr = fgetown(&d->bd_sigio);
1636 /* This is deprecated, FIOSETOWN should be used instead. */
1638 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1641 /* This is deprecated, FIOGETOWN should be used instead. */
1643 *(int *)addr = -fgetown(&d->bd_sigio);
1646 case BIOCSRSIG: /* Set receive signal */
1650 sig = *(u_int *)addr;
1663 *(u_int *)addr = d->bd_sig;
1667 case BIOCGETBUFMODE:
1669 *(u_int *)addr = d->bd_bufmode;
1673 case BIOCSETBUFMODE:
1675 * Allow the buffering mode to be changed as long as we
1676 * haven't yet committed to a particular mode. Our
1677 * definition of commitment, for now, is whether or not a
1678 * buffer has been allocated or an interface attached, since
1679 * that's the point where things get tricky.
1681 switch (*(u_int *)addr) {
1682 case BPF_BUFMODE_BUFFER:
1685 case BPF_BUFMODE_ZBUF:
1686 if (bpf_zerocopy_enable)
1696 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1697 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1702 d->bd_bufmode = *(u_int *)addr;
1707 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1711 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1715 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1723 * Set d's packet filter program to fp. If this file already has a filter,
1724 * free it and replace it. Returns EINVAL for bogus requests.
1726 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1727 * since reading d->bd_bif can't be protected by d or interface lock due to
1730 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1731 * interface read lock to read all filers.
1735 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1737 #ifdef COMPAT_FREEBSD32
1738 struct bpf_program fp_swab;
1739 struct bpf_program32 *fp32;
1741 struct bpf_insn *fcode, *old;
1743 bpf_jit_filter *jfunc, *ofunc;
1749 #ifdef COMPAT_FREEBSD32
1754 fp32 = (struct bpf_program32 *)fp;
1755 fp_swab.bf_len = fp32->bf_len;
1756 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1772 jfunc = ofunc = NULL;
1777 * Check new filter validness before acquiring any locks.
1778 * Allocate memory for new filter, if needed.
1781 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1783 size = flen * sizeof(*fp->bf_insns);
1785 /* We're setting up new filter. Copy and check actual data. */
1786 fcode = malloc(size, M_BPF, M_WAITOK);
1787 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1788 !bpf_validate(fcode, flen)) {
1793 /* Filter is copied inside fcode and is perfectly valid. */
1794 jfunc = bpf_jitter(fcode, flen);
1801 * Set up new filter.
1802 * Protect filter change by interface lock.
1803 * Additionally, we are protected by global lock here.
1805 if (d->bd_bif != NULL)
1806 BPFIF_WLOCK(d->bd_bif);
1808 if (cmd == BIOCSETWF) {
1809 old = d->bd_wfilter;
1810 d->bd_wfilter = fcode;
1812 old = d->bd_rfilter;
1813 d->bd_rfilter = fcode;
1815 ofunc = d->bd_bfilter;
1816 d->bd_bfilter = jfunc;
1818 if (cmd == BIOCSETF)
1821 if (fcode != NULL) {
1823 * Do not require upgrade by first BIOCSETF
1824 * (used to set snaplen) by pcap_open_live().
1826 if (d->bd_writer != 0 && --d->bd_writer == 0)
1828 CTR4(KTR_NET, "%s: filter function set by pid %d, "
1829 "bd_writer counter %d, need_upgrade %d",
1830 __func__, d->bd_pid, d->bd_writer, need_upgrade);
1834 if (d->bd_bif != NULL)
1835 BPFIF_WUNLOCK(d->bd_bif);
1840 bpf_destroy_jit_filter(ofunc);
1843 /* Move d to active readers list. */
1852 * Detach a file from its current interface (if attached at all) and attach
1853 * to the interface indicated by the name stored in ifr.
1854 * Return an errno or 0.
1857 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1860 struct ifnet *theywant;
1864 theywant = ifunit(ifr->ifr_name);
1865 if (theywant == NULL || theywant->if_bpf == NULL)
1868 bp = theywant->if_bpf;
1870 /* Check if interface is not being detached from BPF */
1872 if (bp->flags & BPFIF_FLAG_DYING) {
1879 * Behavior here depends on the buffering model. If we're using
1880 * kernel memory buffers, then we can allocate them here. If we're
1881 * using zero-copy, then the user process must have registered
1882 * buffers by the time we get here. If not, return an error.
1884 switch (d->bd_bufmode) {
1885 case BPF_BUFMODE_BUFFER:
1886 case BPF_BUFMODE_ZBUF:
1887 if (d->bd_sbuf == NULL)
1892 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1894 if (bp != d->bd_bif)
1903 * Support for select() and poll() system calls
1905 * Return true iff the specific operation will not block indefinitely.
1906 * Otherwise, return false but make a note that a selwakeup() must be done.
1909 bpfpoll(struct cdev *dev, int events, struct thread *td)
1914 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1916 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1919 * Refresh PID associated with this descriptor.
1921 revents = events & (POLLOUT | POLLWRNORM);
1923 BPF_PID_REFRESH(d, td);
1924 if (events & (POLLIN | POLLRDNORM)) {
1926 revents |= events & (POLLIN | POLLRDNORM);
1928 selrecord(td, &d->bd_sel);
1929 /* Start the read timeout if necessary. */
1930 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1931 callout_reset(&d->bd_callout, d->bd_rtout,
1933 d->bd_state = BPF_WAITING;
1942 * Support for kevent() system call. Register EVFILT_READ filters and
1943 * reject all others.
1946 bpfkqfilter(struct cdev *dev, struct knote *kn)
1950 if (devfs_get_cdevpriv((void **)&d) != 0 ||
1951 kn->kn_filter != EVFILT_READ)
1955 * Refresh PID associated with this descriptor.
1958 BPF_PID_REFRESH_CUR(d);
1959 kn->kn_fop = &bpfread_filtops;
1961 knlist_add(&d->bd_sel.si_note, kn, 1);
1968 filt_bpfdetach(struct knote *kn)
1970 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1972 knlist_remove(&d->bd_sel.si_note, kn, 0);
1976 filt_bpfread(struct knote *kn, long hint)
1978 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
1981 BPFD_LOCK_ASSERT(d);
1982 ready = bpf_ready(d);
1984 kn->kn_data = d->bd_slen;
1985 while (d->bd_hbuf_in_use)
1986 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1987 PRINET, "bd_hbuf", 0);
1989 kn->kn_data += d->bd_hlen;
1990 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1991 callout_reset(&d->bd_callout, d->bd_rtout,
1993 d->bd_state = BPF_WAITING;
1999 #define BPF_TSTAMP_NONE 0
2000 #define BPF_TSTAMP_FAST 1
2001 #define BPF_TSTAMP_NORMAL 2
2002 #define BPF_TSTAMP_EXTERN 3
2005 bpf_ts_quality(int tstype)
2008 if (tstype == BPF_T_NONE)
2009 return (BPF_TSTAMP_NONE);
2010 if ((tstype & BPF_T_FAST) != 0)
2011 return (BPF_TSTAMP_FAST);
2013 return (BPF_TSTAMP_NORMAL);
2017 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2022 quality = bpf_ts_quality(tstype);
2023 if (quality == BPF_TSTAMP_NONE)
2027 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2029 *bt = *(struct bintime *)(tag + 1);
2030 return (BPF_TSTAMP_EXTERN);
2033 if (quality == BPF_TSTAMP_NORMAL)
2042 * Incoming linkage from device drivers. Process the packet pkt, of length
2043 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2044 * by each process' filter, and if accepted, stashed into the corresponding
2048 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2058 gottime = BPF_TSTAMP_NONE;
2062 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2064 * We are not using any locks for d here because:
2065 * 1) any filter change is protected by interface
2067 * 2) destroying/detaching d is protected by interface
2071 /* XXX: Do not protect counter for the sake of performance. */
2074 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2075 * way for the caller to indiciate to us whether this packet
2076 * is inbound or outbound. In the bpf_mtap() routines, we use
2077 * the interface pointers on the mbuf to figure it out.
2080 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2082 slen = (*(bf->func))(pkt, pktlen, pktlen);
2085 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2088 * Filter matches. Let's to acquire write lock.
2093 if (gottime < bpf_ts_quality(d->bd_tstamp))
2094 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2096 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2098 catchpacket(d, pkt, pktlen, slen,
2099 bpf_append_bytes, &bt);
2106 #define BPF_CHECK_DIRECTION(d, r, i) \
2107 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2108 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2111 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2112 * Locking model is explained in bpf_tap().
2115 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2125 /* Skip outgoing duplicate packets. */
2126 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2127 m->m_flags &= ~M_PROMISC;
2131 pktlen = m_length(m, NULL);
2132 gottime = BPF_TSTAMP_NONE;
2136 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2137 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2141 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2142 /* XXX We cannot handle multiple mbufs. */
2143 if (bf != NULL && m->m_next == NULL)
2144 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2147 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2152 if (gottime < bpf_ts_quality(d->bd_tstamp))
2153 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2155 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2157 catchpacket(d, (u_char *)m, pktlen, slen,
2158 bpf_append_mbuf, &bt);
2166 * Incoming linkage from device drivers, when packet is in
2167 * an mbuf chain and to be prepended by a contiguous header.
2170 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2178 /* Skip outgoing duplicate packets. */
2179 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2180 m->m_flags &= ~M_PROMISC;
2184 pktlen = m_length(m, NULL);
2186 * Craft on-stack mbuf suitable for passing to bpf_filter.
2187 * Note that we cut corners here; we only setup what's
2188 * absolutely needed--this mbuf should never go anywhere else.
2195 gottime = BPF_TSTAMP_NONE;
2199 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2200 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2203 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2208 if (gottime < bpf_ts_quality(d->bd_tstamp))
2209 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2211 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2213 catchpacket(d, (u_char *)&mb, pktlen, slen,
2214 bpf_append_mbuf, &bt);
2221 #undef BPF_CHECK_DIRECTION
2223 #undef BPF_TSTAMP_NONE
2224 #undef BPF_TSTAMP_FAST
2225 #undef BPF_TSTAMP_NORMAL
2226 #undef BPF_TSTAMP_EXTERN
2229 bpf_hdrlen(struct bpf_d *d)
2233 hdrlen = d->bd_bif->bif_hdrlen;
2234 #ifndef BURN_BRIDGES
2235 if (d->bd_tstamp == BPF_T_NONE ||
2236 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2237 #ifdef COMPAT_FREEBSD32
2239 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2242 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2245 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2246 #ifdef COMPAT_FREEBSD32
2248 hdrlen = BPF_WORDALIGN32(hdrlen);
2251 hdrlen = BPF_WORDALIGN(hdrlen);
2253 return (hdrlen - d->bd_bif->bif_hdrlen);
2257 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2261 struct timespec tsn;
2263 if ((tstype & BPF_T_MONOTONIC) == 0) {
2265 bintime_add(&bt2, &boottimebin);
2268 switch (BPF_T_FORMAT(tstype)) {
2269 case BPF_T_MICROTIME:
2270 bintime2timeval(bt, &tsm);
2271 ts->bt_sec = tsm.tv_sec;
2272 ts->bt_frac = tsm.tv_usec;
2274 case BPF_T_NANOTIME:
2275 bintime2timespec(bt, &tsn);
2276 ts->bt_sec = tsn.tv_sec;
2277 ts->bt_frac = tsn.tv_nsec;
2280 ts->bt_sec = bt->sec;
2281 ts->bt_frac = bt->frac;
2287 * Move the packet data from interface memory (pkt) into the
2288 * store buffer. "cpfn" is the routine called to do the actual data
2289 * transfer. bcopy is passed in to copy contiguous chunks, while
2290 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2291 * pkt is really an mbuf.
2294 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2295 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2298 struct bpf_xhdr hdr;
2299 #ifndef BURN_BRIDGES
2300 struct bpf_hdr hdr_old;
2301 #ifdef COMPAT_FREEBSD32
2302 struct bpf_hdr32 hdr32_old;
2305 int caplen, curlen, hdrlen, totlen;
2310 BPFD_LOCK_ASSERT(d);
2313 * Detect whether user space has released a buffer back to us, and if
2314 * so, move it from being a hold buffer to a free buffer. This may
2315 * not be the best place to do it (for example, we might only want to
2316 * run this check if we need the space), but for now it's a reliable
2319 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2320 while (d->bd_hbuf_in_use)
2321 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2322 PRINET, "bd_hbuf", 0);
2323 d->bd_fbuf = d->bd_hbuf;
2326 bpf_buf_reclaimed(d);
2330 * Figure out how many bytes to move. If the packet is
2331 * greater or equal to the snapshot length, transfer that
2332 * much. Otherwise, transfer the whole packet (unless
2333 * we hit the buffer size limit).
2335 hdrlen = bpf_hdrlen(d);
2336 totlen = hdrlen + min(snaplen, pktlen);
2337 if (totlen > d->bd_bufsize)
2338 totlen = d->bd_bufsize;
2341 * Round up the end of the previous packet to the next longword.
2343 * Drop the packet if there's no room and no hope of room
2344 * If the packet would overflow the storage buffer or the storage
2345 * buffer is considered immutable by the buffer model, try to rotate
2346 * the buffer and wakeup pending processes.
2348 #ifdef COMPAT_FREEBSD32
2350 curlen = BPF_WORDALIGN32(d->bd_slen);
2353 curlen = BPF_WORDALIGN(d->bd_slen);
2354 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2355 if (d->bd_fbuf == NULL) {
2357 * There's no room in the store buffer, and no
2358 * prospect of room, so drop the packet. Notify the
2365 while (d->bd_hbuf_in_use)
2366 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2367 PRINET, "bd_hbuf", 0);
2371 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2373 * Immediate mode is set, or the read timeout has already
2374 * expired during a select call. A packet arrived, so the
2375 * reader should be woken up.
2378 caplen = totlen - hdrlen;
2379 tstype = d->bd_tstamp;
2380 do_timestamp = tstype != BPF_T_NONE;
2381 #ifndef BURN_BRIDGES
2382 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2385 bpf_bintime2ts(bt, &ts, tstype);
2386 #ifdef COMPAT_FREEBSD32
2387 if (d->bd_compat32) {
2388 bzero(&hdr32_old, sizeof(hdr32_old));
2390 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2391 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2393 hdr32_old.bh_datalen = pktlen;
2394 hdr32_old.bh_hdrlen = hdrlen;
2395 hdr32_old.bh_caplen = caplen;
2396 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2401 bzero(&hdr_old, sizeof(hdr_old));
2403 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2404 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2406 hdr_old.bh_datalen = pktlen;
2407 hdr_old.bh_hdrlen = hdrlen;
2408 hdr_old.bh_caplen = caplen;
2409 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2416 * Append the bpf header. Note we append the actual header size, but
2417 * move forward the length of the header plus padding.
2419 bzero(&hdr, sizeof(hdr));
2421 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2422 hdr.bh_datalen = pktlen;
2423 hdr.bh_hdrlen = hdrlen;
2424 hdr.bh_caplen = caplen;
2425 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2428 * Copy the packet data into the store buffer and update its length.
2430 #ifndef BURN_BRIDGES
2433 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2434 d->bd_slen = curlen + totlen;
2441 * Free buffers currently in use by a descriptor.
2445 bpf_freed(struct bpf_d *d)
2449 * We don't need to lock out interrupts since this descriptor has
2450 * been detached from its interface and it yet hasn't been marked
2454 if (d->bd_rfilter != NULL) {
2455 free((caddr_t)d->bd_rfilter, M_BPF);
2457 if (d->bd_bfilter != NULL)
2458 bpf_destroy_jit_filter(d->bd_bfilter);
2461 if (d->bd_wfilter != NULL)
2462 free((caddr_t)d->bd_wfilter, M_BPF);
2463 mtx_destroy(&d->bd_lock);
2467 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2468 * fixed size of the link header (variable length headers not yet supported).
2471 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2474 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2478 * Attach an interface to bpf. ifp is a pointer to the structure
2479 * defining the interface to be attached, dlt is the link layer type,
2480 * and hdrlen is the fixed size of the link header (variable length
2481 * headers are not yet supporrted).
2484 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2488 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2492 LIST_INIT(&bp->bif_dlist);
2493 LIST_INIT(&bp->bif_wlist);
2496 rw_init(&bp->bif_lock, "bpf interface lock");
2497 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2501 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2504 bp->bif_hdrlen = hdrlen;
2507 if_printf(ifp, "bpf attached\n");
2511 * Detach bpf from an interface. This involves detaching each descriptor
2512 * associated with the interface. Notify each descriptor as it's detached
2513 * so that any sleepers wake up and get ENXIO.
2516 bpfdetach(struct ifnet *ifp)
2527 /* Find all bpf_if struct's which reference ifp and detach them. */
2529 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2530 if (ifp == bp->bif_ifp)
2534 LIST_REMOVE(bp, bif_next);
2540 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2541 bpf_detachd_locked(d);
2546 /* Free writer-only descriptors */
2547 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2548 bpf_detachd_locked(d);
2555 * Delay freing bp till interface is detached
2556 * and all routes through this interface are removed.
2557 * Mark bp as detached to restrict new consumers.
2560 bp->flags |= BPFIF_FLAG_DYING;
2563 } while (bp != NULL);
2568 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2573 * Interface departure handler.
2574 * Note departure event does not guarantee interface is going down.
2577 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2582 if ((bp = ifp->if_bpf) == NULL) {
2587 /* Check if bpfdetach() was called previously */
2588 if ((bp->flags & BPFIF_FLAG_DYING) == 0) {
2593 CTR3(KTR_NET, "%s: freing BPF instance %p for interface %p",
2599 rw_destroy(&bp->bif_lock);
2604 * Get a list of available data link type of the interface.
2607 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2615 ifp = d->bd_bif->bif_ifp;
2618 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2619 if (bp->bif_ifp != ifp)
2621 if (bfl->bfl_list != NULL) {
2622 if (n >= bfl->bfl_len)
2624 error = copyout(&bp->bif_dlt,
2625 bfl->bfl_list + n, sizeof(u_int));
2634 * Set the data link type of a BPF instance.
2637 bpf_setdlt(struct bpf_d *d, u_int dlt)
2639 int error, opromisc;
2645 if (d->bd_bif->bif_dlt == dlt)
2647 ifp = d->bd_bif->bif_ifp;
2649 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2650 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2655 opromisc = d->bd_promisc;
2661 error = ifpromisc(bp->bif_ifp, 1);
2663 if_printf(bp->bif_ifp,
2664 "bpf_setdlt: ifpromisc failed (%d)\n",
2670 return (bp == NULL ? EINVAL : 0);
2674 bpf_drvinit(void *unused)
2678 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2679 LIST_INIT(&bpf_iflist);
2681 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2682 /* For compatibility */
2683 make_dev_alias(dev, "bpf0");
2685 /* Register interface departure handler */
2686 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2687 ifnet_departure_event, bpf_ifdetach, NULL,
2688 EVENTHANDLER_PRI_ANY);
2692 * Zero out the various packet counters associated with all of the bpf
2693 * descriptors. At some point, we will probably want to get a bit more
2694 * granular and allow the user to specify descriptors to be zeroed.
2697 bpf_zero_counters(void)
2703 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2705 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2721 * Fill filter statistics
2724 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2727 bzero(d, sizeof(*d));
2728 BPFD_LOCK_ASSERT(bd);
2729 d->bd_structsize = sizeof(*d);
2730 /* XXX: reading should be protected by global lock */
2731 d->bd_immediate = bd->bd_immediate;
2732 d->bd_promisc = bd->bd_promisc;
2733 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2734 d->bd_direction = bd->bd_direction;
2735 d->bd_feedback = bd->bd_feedback;
2736 d->bd_async = bd->bd_async;
2737 d->bd_rcount = bd->bd_rcount;
2738 d->bd_dcount = bd->bd_dcount;
2739 d->bd_fcount = bd->bd_fcount;
2740 d->bd_sig = bd->bd_sig;
2741 d->bd_slen = bd->bd_slen;
2742 d->bd_hlen = bd->bd_hlen;
2743 d->bd_bufsize = bd->bd_bufsize;
2744 d->bd_pid = bd->bd_pid;
2745 strlcpy(d->bd_ifname,
2746 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2747 d->bd_locked = bd->bd_locked;
2748 d->bd_wcount = bd->bd_wcount;
2749 d->bd_wdcount = bd->bd_wdcount;
2750 d->bd_wfcount = bd->bd_wfcount;
2751 d->bd_zcopy = bd->bd_zcopy;
2752 d->bd_bufmode = bd->bd_bufmode;
2756 * Handle `netstat -B' stats request
2759 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2761 struct xbpf_d *xbdbuf, *xbd, zerostats;
2767 * XXX This is not technically correct. It is possible for non
2768 * privileged users to open bpf devices. It would make sense
2769 * if the users who opened the devices were able to retrieve
2770 * the statistics for them, too.
2772 error = priv_check(req->td, PRIV_NET_BPF);
2776 * Check to see if the user is requesting that the counters be
2777 * zeroed out. Explicitly check that the supplied data is zeroed,
2778 * as we aren't allowing the user to set the counters currently.
2780 if (req->newptr != NULL) {
2781 if (req->newlen != sizeof(zerostats))
2783 bzero(&zerostats, sizeof(zerostats));
2785 if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0)
2787 bpf_zero_counters();
2790 if (req->oldptr == NULL)
2791 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2792 if (bpf_bpfd_cnt == 0)
2793 return (SYSCTL_OUT(req, 0, 0));
2794 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2796 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2798 free(xbdbuf, M_BPF);
2802 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2804 /* Send writers-only first */
2805 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2806 xbd = &xbdbuf[index++];
2808 bpfstats_fill_xbpf(xbd, bd);
2811 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2812 xbd = &xbdbuf[index++];
2814 bpfstats_fill_xbpf(xbd, bd);
2820 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2821 free(xbdbuf, M_BPF);
2825 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2827 #else /* !DEV_BPF && !NETGRAPH_BPF */
2829 * NOP stubs to allow bpf-using drivers to load and function.
2831 * A 'better' implementation would allow the core bpf functionality
2832 * to be loaded at runtime.
2834 static struct bpf_if bp_null;
2837 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2842 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2847 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2852 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2855 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2859 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2862 *driverp = &bp_null;
2866 bpfdetach(struct ifnet *ifp)
2871 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2873 return -1; /* "no filter" behaviour */
2877 bpf_validate(const struct bpf_insn *f, int len)
2879 return 0; /* false */
2882 #endif /* !DEV_BPF && !NETGRAPH_BPF */