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 * Check if we need to upgrade our descriptor @d from write-only mode.
665 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
667 int is_snap, need_upgrade;
670 * Check if we've already upgraded or new filter is empty.
672 if (d->bd_writer == 0 || fcode == NULL)
678 * Check if cmd looks like snaplen setting from
679 * pcap_bpf.c:pcap_open_live().
680 * Note we're not checking .k value here:
681 * while pcap_open_live() definitely sets to to non-zero value,
682 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
683 * do not consider upgrading immediately
685 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
692 * We're setting first filter and it doesn't look like
693 * setting snaplen. We're probably using bpf directly.
694 * Upgrade immediately.
699 * Do not require upgrade by first BIOCSETF
700 * (used to set snaplen) by pcap_open_live().
703 if (--d->bd_writer == 0) {
705 * First snaplen filter has already
706 * been set. This is probably catch-all
714 "%s: filter function set by pid %d, "
715 "bd_writer counter %d, snap %d upgrade %d",
716 __func__, d->bd_pid, d->bd_writer,
717 is_snap, need_upgrade);
719 return (need_upgrade);
723 * Add d to the list of active bp filters.
724 * Reuqires bpf_attachd() to be called before
727 bpf_upgraded(struct bpf_d *d)
736 * Filter can be set several times without specifying interface.
737 * Mark d as reader and exit.
749 /* Remove from writers-only list */
750 LIST_REMOVE(d, bd_next);
751 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
752 /* Mark d as reader */
758 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
760 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
764 * Detach a file from its interface.
767 bpf_detachd(struct bpf_d *d)
770 bpf_detachd_locked(d);
775 bpf_detachd_locked(struct bpf_d *d)
781 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
785 /* Check if descriptor is attached */
786 if ((bp = d->bd_bif) == NULL)
792 /* Save bd_writer value */
793 error = d->bd_writer;
796 * Remove d from the interface's descriptor list.
798 LIST_REMOVE(d, bd_next);
807 /* Call event handler iff d is attached */
809 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
812 * Check if this descriptor had requested promiscuous mode.
813 * If so, turn it off.
817 CURVNET_SET(ifp->if_vnet);
818 error = ifpromisc(ifp, 0);
820 if (error != 0 && error != ENXIO) {
822 * ENXIO can happen if a pccard is unplugged
823 * Something is really wrong if we were able to put
824 * the driver into promiscuous mode, but can't
827 if_printf(bp->bif_ifp,
828 "bpf_detach: ifpromisc failed (%d)\n", error);
834 * Close the descriptor by detaching it from its interface,
835 * deallocating its buffers, and marking it free.
840 struct bpf_d *d = data;
843 if (d->bd_state == BPF_WAITING)
844 callout_stop(&d->bd_callout);
845 d->bd_state = BPF_IDLE;
847 funsetown(&d->bd_sigio);
850 mac_bpfdesc_destroy(d);
852 seldrain(&d->bd_sel);
853 knlist_destroy(&d->bd_sel.si_note);
854 callout_drain(&d->bd_callout);
860 * Open ethernet device. Returns ENXIO for illegal minor device number,
861 * EBUSY if file is open by another process.
865 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
870 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
871 error = devfs_set_cdevpriv(d, bpf_dtor);
878 * For historical reasons, perform a one-time initialization call to
879 * the buffer routines, even though we're not yet committed to a
880 * particular buffer method.
883 d->bd_hbuf_in_use = 0;
884 d->bd_bufmode = BPF_BUFMODE_BUFFER;
886 d->bd_direction = BPF_D_INOUT;
887 BPF_PID_REFRESH(d, td);
890 mac_bpfdesc_create(td->td_ucred, d);
892 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
893 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
894 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
896 /* Allocate default buffers */
897 size = d->bd_bufsize;
898 bpf_buffer_ioctl_sblen(d, &size);
904 * bpfread - read next chunk of packets from buffers
907 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
914 error = devfs_get_cdevpriv((void **)&d);
919 * Restrict application to use a buffer the same size as
922 if (uio->uio_resid != d->bd_bufsize)
925 non_block = ((ioflag & O_NONBLOCK) != 0);
928 BPF_PID_REFRESH_CUR(d);
929 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
933 if (d->bd_state == BPF_WAITING)
934 callout_stop(&d->bd_callout);
935 timed_out = (d->bd_state == BPF_TIMED_OUT);
936 d->bd_state = BPF_IDLE;
937 while (d->bd_hbuf_in_use) {
938 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
939 PRINET|PCATCH, "bd_hbuf", 0);
946 * If the hold buffer is empty, then do a timed sleep, which
947 * ends when the timeout expires or when enough packets
948 * have arrived to fill the store buffer.
950 while (d->bd_hbuf == NULL) {
951 if (d->bd_slen != 0) {
953 * A packet(s) either arrived since the previous
954 * read or arrived while we were asleep.
956 if (d->bd_immediate || non_block || timed_out) {
958 * Rotate the buffers and return what's here
959 * if we are in immediate mode, non-blocking
960 * flag is set, or this descriptor timed out.
968 * No data is available, check to see if the bpf device
969 * is still pointed at a real interface. If not, return
970 * ENXIO so that the userland process knows to rebind
971 * it before using it again.
973 if (d->bd_bif == NULL) {
980 return (EWOULDBLOCK);
982 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
984 if (error == EINTR || error == ERESTART) {
988 if (error == EWOULDBLOCK) {
990 * On a timeout, return what's in the buffer,
991 * which may be nothing. If there is something
992 * in the store buffer, we can rotate the buffers.
996 * We filled up the buffer in between
997 * getting the timeout and arriving
998 * here, so we don't need to rotate.
1002 if (d->bd_slen == 0) {
1011 * At this point, we know we have something in the hold slot.
1013 d->bd_hbuf_in_use = 1;
1017 * Move data from hold buffer into user space.
1018 * We know the entire buffer is transferred since
1019 * we checked above that the read buffer is bpf_bufsize bytes.
1021 * We do not have to worry about simultaneous reads because
1022 * we waited for sole access to the hold buffer above.
1024 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1027 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1028 d->bd_fbuf = d->bd_hbuf;
1031 bpf_buf_reclaimed(d);
1032 d->bd_hbuf_in_use = 0;
1033 wakeup(&d->bd_hbuf_in_use);
1040 * If there are processes sleeping on this descriptor, wake them up.
1042 static __inline void
1043 bpf_wakeup(struct bpf_d *d)
1046 BPFD_LOCK_ASSERT(d);
1047 if (d->bd_state == BPF_WAITING) {
1048 callout_stop(&d->bd_callout);
1049 d->bd_state = BPF_IDLE;
1052 if (d->bd_async && d->bd_sig && d->bd_sigio)
1053 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1055 selwakeuppri(&d->bd_sel, PRINET);
1056 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1060 bpf_timed_out(void *arg)
1062 struct bpf_d *d = (struct bpf_d *)arg;
1064 BPFD_LOCK_ASSERT(d);
1066 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1068 if (d->bd_state == BPF_WAITING) {
1069 d->bd_state = BPF_TIMED_OUT;
1070 if (d->bd_slen != 0)
1076 bpf_ready(struct bpf_d *d)
1079 BPFD_LOCK_ASSERT(d);
1081 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1083 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1090 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1094 struct mbuf *m, *mc;
1095 struct sockaddr dst;
1098 error = devfs_get_cdevpriv((void **)&d);
1102 BPF_PID_REFRESH_CUR(d);
1104 /* XXX: locking required */
1105 if (d->bd_bif == NULL) {
1110 ifp = d->bd_bif->bif_ifp;
1112 if ((ifp->if_flags & IFF_UP) == 0) {
1117 if (uio->uio_resid == 0) {
1122 bzero(&dst, sizeof(dst));
1125 /* XXX: bpf_movein() can sleep */
1126 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1127 &m, &dst, &hlen, d->bd_wfilter);
1134 dst.sa_family = pseudo_AF_HDRCMPLT;
1136 if (d->bd_feedback) {
1137 mc = m_dup(m, M_DONTWAIT);
1139 mc->m_pkthdr.rcvif = ifp;
1140 /* Set M_PROMISC for outgoing packets to be discarded. */
1141 if (d->bd_direction == BPF_D_INOUT)
1142 m->m_flags |= M_PROMISC;
1146 m->m_pkthdr.len -= hlen;
1148 m->m_data += hlen; /* XXX */
1150 CURVNET_SET(ifp->if_vnet);
1153 mac_bpfdesc_create_mbuf(d, m);
1155 mac_bpfdesc_create_mbuf(d, mc);
1159 error = (*ifp->if_output)(ifp, m, &dst, NULL);
1165 (*ifp->if_input)(ifp, mc);
1175 * Reset a descriptor by flushing its packet buffer and clearing the receive
1176 * and drop counts. This is doable for kernel-only buffers, but with
1177 * zero-copy buffers, we can't write to (or rotate) buffers that are
1178 * currently owned by userspace. It would be nice if we could encapsulate
1179 * this logic in the buffer code rather than here.
1182 reset_d(struct bpf_d *d)
1185 BPFD_LOCK_ASSERT(d);
1187 while (d->bd_hbuf_in_use)
1188 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1190 if ((d->bd_hbuf != NULL) &&
1191 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1192 /* Free the hold buffer. */
1193 d->bd_fbuf = d->bd_hbuf;
1196 bpf_buf_reclaimed(d);
1198 if (bpf_canwritebuf(d))
1210 * FIONREAD Check for read packet available.
1211 * SIOCGIFADDR Get interface address - convenient hook to driver.
1212 * BIOCGBLEN Get buffer len [for read()].
1213 * BIOCSETF Set read filter.
1214 * BIOCSETFNR Set read filter without resetting descriptor.
1215 * BIOCSETWF Set write filter.
1216 * BIOCFLUSH Flush read packet buffer.
1217 * BIOCPROMISC Put interface into promiscuous mode.
1218 * BIOCGDLT Get link layer type.
1219 * BIOCGETIF Get interface name.
1220 * BIOCSETIF Set interface.
1221 * BIOCSRTIMEOUT Set read timeout.
1222 * BIOCGRTIMEOUT Get read timeout.
1223 * BIOCGSTATS Get packet stats.
1224 * BIOCIMMEDIATE Set immediate mode.
1225 * BIOCVERSION Get filter language version.
1226 * BIOCGHDRCMPLT Get "header already complete" flag
1227 * BIOCSHDRCMPLT Set "header already complete" flag
1228 * BIOCGDIRECTION Get packet direction flag
1229 * BIOCSDIRECTION Set packet direction flag
1230 * BIOCGTSTAMP Get time stamp format and resolution.
1231 * BIOCSTSTAMP Set time stamp format and resolution.
1232 * BIOCLOCK Set "locked" flag
1233 * BIOCFEEDBACK Set packet feedback mode.
1234 * BIOCSETZBUF Set current zero-copy buffer locations.
1235 * BIOCGETZMAX Get maximum zero-copy buffer size.
1236 * BIOCROTZBUF Force rotation of zero-copy buffer
1237 * BIOCSETBUFMODE Set buffer mode.
1238 * BIOCGETBUFMODE Get current buffer mode.
1242 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1248 error = devfs_get_cdevpriv((void **)&d);
1253 * Refresh PID associated with this descriptor.
1256 BPF_PID_REFRESH(d, td);
1257 if (d->bd_state == BPF_WAITING)
1258 callout_stop(&d->bd_callout);
1259 d->bd_state = BPF_IDLE;
1262 if (d->bd_locked == 1) {
1268 #ifdef COMPAT_FREEBSD32
1269 case BIOCGDLTLIST32:
1273 #ifdef COMPAT_FREEBSD32
1274 case BIOCGRTIMEOUT32:
1285 #ifdef COMPAT_FREEBSD32
1286 case BIOCSRTIMEOUT32:
1296 #ifdef COMPAT_FREEBSD32
1298 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1299 * that it will get 32-bit packet headers.
1305 case BIOCGDLTLIST32:
1306 case BIOCGRTIMEOUT32:
1307 case BIOCSRTIMEOUT32:
1314 CURVNET_SET(TD_TO_VNET(td));
1322 * Check for read packet available.
1330 while (d->bd_hbuf_in_use)
1331 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1332 PRINET, "bd_hbuf", 0);
1345 if (d->bd_bif == NULL)
1348 ifp = d->bd_bif->bif_ifp;
1349 error = (*ifp->if_ioctl)(ifp, cmd, addr);
1355 * Get buffer len [for read()].
1359 *(u_int *)addr = d->bd_bufsize;
1364 * Set buffer length.
1367 error = bpf_ioctl_sblen(d, (u_int *)addr);
1371 * Set link layer read filter.
1376 #ifdef COMPAT_FREEBSD32
1381 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1385 * Flush read packet buffer.
1394 * Put interface into promiscuous mode.
1397 if (d->bd_bif == NULL) {
1399 * No interface attached yet.
1404 if (d->bd_promisc == 0) {
1405 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1412 * Get current data link type.
1416 if (d->bd_bif == NULL)
1419 *(u_int *)addr = d->bd_bif->bif_dlt;
1424 * Get a list of supported data link types.
1426 #ifdef COMPAT_FREEBSD32
1427 case BIOCGDLTLIST32:
1429 struct bpf_dltlist32 *list32;
1430 struct bpf_dltlist dltlist;
1432 list32 = (struct bpf_dltlist32 *)addr;
1433 dltlist.bfl_len = list32->bfl_len;
1434 dltlist.bfl_list = PTRIN(list32->bfl_list);
1436 if (d->bd_bif == NULL)
1439 error = bpf_getdltlist(d, &dltlist);
1441 list32->bfl_len = dltlist.bfl_len;
1450 if (d->bd_bif == NULL)
1453 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1458 * Set data link type.
1462 if (d->bd_bif == NULL)
1465 error = bpf_setdlt(d, *(u_int *)addr);
1470 * Get interface name.
1474 if (d->bd_bif == NULL)
1477 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1478 struct ifreq *const ifr = (struct ifreq *)addr;
1480 strlcpy(ifr->ifr_name, ifp->if_xname,
1481 sizeof(ifr->ifr_name));
1491 error = bpf_setif(d, (struct ifreq *)addr);
1499 #ifdef COMPAT_FREEBSD32
1500 case BIOCSRTIMEOUT32:
1503 struct timeval *tv = (struct timeval *)addr;
1504 #ifdef COMPAT_FREEBSD32
1505 struct timeval32 *tv32;
1506 struct timeval tv64;
1508 if (cmd == BIOCSRTIMEOUT32) {
1509 tv32 = (struct timeval32 *)addr;
1511 tv->tv_sec = tv32->tv_sec;
1512 tv->tv_usec = tv32->tv_usec;
1515 tv = (struct timeval *)addr;
1518 * Subtract 1 tick from tvtohz() since this isn't
1521 if ((error = itimerfix(tv)) == 0)
1522 d->bd_rtout = tvtohz(tv) - 1;
1530 #ifdef COMPAT_FREEBSD32
1531 case BIOCGRTIMEOUT32:
1535 #ifdef COMPAT_FREEBSD32
1536 struct timeval32 *tv32;
1537 struct timeval tv64;
1539 if (cmd == BIOCGRTIMEOUT32)
1543 tv = (struct timeval *)addr;
1545 tv->tv_sec = d->bd_rtout / hz;
1546 tv->tv_usec = (d->bd_rtout % hz) * tick;
1547 #ifdef COMPAT_FREEBSD32
1548 if (cmd == BIOCGRTIMEOUT32) {
1549 tv32 = (struct timeval32 *)addr;
1550 tv32->tv_sec = tv->tv_sec;
1551 tv32->tv_usec = tv->tv_usec;
1563 struct bpf_stat *bs = (struct bpf_stat *)addr;
1565 /* XXXCSJP overflow */
1566 bs->bs_recv = d->bd_rcount;
1567 bs->bs_drop = d->bd_dcount;
1572 * Set immediate mode.
1576 d->bd_immediate = *(u_int *)addr;
1582 struct bpf_version *bv = (struct bpf_version *)addr;
1584 bv->bv_major = BPF_MAJOR_VERSION;
1585 bv->bv_minor = BPF_MINOR_VERSION;
1590 * Get "header already complete" flag
1594 *(u_int *)addr = d->bd_hdrcmplt;
1599 * Set "header already complete" flag
1603 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1608 * Get packet direction flag
1610 case BIOCGDIRECTION:
1612 *(u_int *)addr = d->bd_direction;
1617 * Set packet direction flag
1619 case BIOCSDIRECTION:
1623 direction = *(u_int *)addr;
1624 switch (direction) {
1629 d->bd_direction = direction;
1639 * Get packet timestamp format and resolution.
1643 *(u_int *)addr = d->bd_tstamp;
1648 * Set packet timestamp format and resolution.
1654 func = *(u_int *)addr;
1655 if (BPF_T_VALID(func))
1656 d->bd_tstamp = func;
1664 d->bd_feedback = *(u_int *)addr;
1674 case FIONBIO: /* Non-blocking I/O */
1677 case FIOASYNC: /* Send signal on receive packets */
1679 d->bd_async = *(int *)addr;
1685 * XXX: Add some sort of locking here?
1686 * fsetown() can sleep.
1688 error = fsetown(*(int *)addr, &d->bd_sigio);
1693 *(int *)addr = fgetown(&d->bd_sigio);
1697 /* This is deprecated, FIOSETOWN should be used instead. */
1699 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1702 /* This is deprecated, FIOGETOWN should be used instead. */
1704 *(int *)addr = -fgetown(&d->bd_sigio);
1707 case BIOCSRSIG: /* Set receive signal */
1711 sig = *(u_int *)addr;
1724 *(u_int *)addr = d->bd_sig;
1728 case BIOCGETBUFMODE:
1730 *(u_int *)addr = d->bd_bufmode;
1734 case BIOCSETBUFMODE:
1736 * Allow the buffering mode to be changed as long as we
1737 * haven't yet committed to a particular mode. Our
1738 * definition of commitment, for now, is whether or not a
1739 * buffer has been allocated or an interface attached, since
1740 * that's the point where things get tricky.
1742 switch (*(u_int *)addr) {
1743 case BPF_BUFMODE_BUFFER:
1746 case BPF_BUFMODE_ZBUF:
1747 if (bpf_zerocopy_enable)
1757 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1758 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1763 d->bd_bufmode = *(u_int *)addr;
1768 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1772 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1776 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1784 * Set d's packet filter program to fp. If this file already has a filter,
1785 * free it and replace it. Returns EINVAL for bogus requests.
1787 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1788 * since reading d->bd_bif can't be protected by d or interface lock due to
1791 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1792 * interface read lock to read all filers.
1796 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1798 #ifdef COMPAT_FREEBSD32
1799 struct bpf_program fp_swab;
1800 struct bpf_program32 *fp32;
1802 struct bpf_insn *fcode, *old;
1804 bpf_jit_filter *jfunc, *ofunc;
1810 #ifdef COMPAT_FREEBSD32
1815 fp32 = (struct bpf_program32 *)fp;
1816 fp_swab.bf_len = fp32->bf_len;
1817 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1833 jfunc = ofunc = NULL;
1838 * Check new filter validness before acquiring any locks.
1839 * Allocate memory for new filter, if needed.
1842 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1844 size = flen * sizeof(*fp->bf_insns);
1846 /* We're setting up new filter. Copy and check actual data. */
1847 fcode = malloc(size, M_BPF, M_WAITOK);
1848 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1849 !bpf_validate(fcode, flen)) {
1854 /* Filter is copied inside fcode and is perfectly valid. */
1855 jfunc = bpf_jitter(fcode, flen);
1862 * Set up new filter.
1863 * Protect filter change by interface lock.
1864 * Additionally, we are protected by global lock here.
1866 if (d->bd_bif != NULL)
1867 BPFIF_WLOCK(d->bd_bif);
1869 if (cmd == BIOCSETWF) {
1870 old = d->bd_wfilter;
1871 d->bd_wfilter = fcode;
1873 old = d->bd_rfilter;
1874 d->bd_rfilter = fcode;
1876 ofunc = d->bd_bfilter;
1877 d->bd_bfilter = jfunc;
1879 if (cmd == BIOCSETF)
1882 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1885 if (d->bd_bif != NULL)
1886 BPFIF_WUNLOCK(d->bd_bif);
1891 bpf_destroy_jit_filter(ofunc);
1894 /* Move d to active readers list. */
1895 if (need_upgrade != 0)
1903 * Detach a file from its current interface (if attached at all) and attach
1904 * to the interface indicated by the name stored in ifr.
1905 * Return an errno or 0.
1908 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1911 struct ifnet *theywant;
1915 theywant = ifunit(ifr->ifr_name);
1916 if (theywant == NULL || theywant->if_bpf == NULL)
1919 bp = theywant->if_bpf;
1921 /* Check if interface is not being detached from BPF */
1923 if (bp->flags & BPFIF_FLAG_DYING) {
1930 * Behavior here depends on the buffering model. If we're using
1931 * kernel memory buffers, then we can allocate them here. If we're
1932 * using zero-copy, then the user process must have registered
1933 * buffers by the time we get here. If not, return an error.
1935 switch (d->bd_bufmode) {
1936 case BPF_BUFMODE_BUFFER:
1937 case BPF_BUFMODE_ZBUF:
1938 if (d->bd_sbuf == NULL)
1943 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1945 if (bp != d->bd_bif)
1954 * Support for select() and poll() system calls
1956 * Return true iff the specific operation will not block indefinitely.
1957 * Otherwise, return false but make a note that a selwakeup() must be done.
1960 bpfpoll(struct cdev *dev, int events, struct thread *td)
1965 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1967 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1970 * Refresh PID associated with this descriptor.
1972 revents = events & (POLLOUT | POLLWRNORM);
1974 BPF_PID_REFRESH(d, td);
1975 if (events & (POLLIN | POLLRDNORM)) {
1977 revents |= events & (POLLIN | POLLRDNORM);
1979 selrecord(td, &d->bd_sel);
1980 /* Start the read timeout if necessary. */
1981 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
1982 callout_reset(&d->bd_callout, d->bd_rtout,
1984 d->bd_state = BPF_WAITING;
1993 * Support for kevent() system call. Register EVFILT_READ filters and
1994 * reject all others.
1997 bpfkqfilter(struct cdev *dev, struct knote *kn)
2001 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2002 kn->kn_filter != EVFILT_READ)
2006 * Refresh PID associated with this descriptor.
2009 BPF_PID_REFRESH_CUR(d);
2010 kn->kn_fop = &bpfread_filtops;
2012 knlist_add(&d->bd_sel.si_note, kn, 1);
2019 filt_bpfdetach(struct knote *kn)
2021 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2023 knlist_remove(&d->bd_sel.si_note, kn, 0);
2027 filt_bpfread(struct knote *kn, long hint)
2029 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2032 BPFD_LOCK_ASSERT(d);
2033 ready = bpf_ready(d);
2035 kn->kn_data = d->bd_slen;
2036 while (d->bd_hbuf_in_use)
2037 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2038 PRINET, "bd_hbuf", 0);
2040 kn->kn_data += d->bd_hlen;
2041 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2042 callout_reset(&d->bd_callout, d->bd_rtout,
2044 d->bd_state = BPF_WAITING;
2050 #define BPF_TSTAMP_NONE 0
2051 #define BPF_TSTAMP_FAST 1
2052 #define BPF_TSTAMP_NORMAL 2
2053 #define BPF_TSTAMP_EXTERN 3
2056 bpf_ts_quality(int tstype)
2059 if (tstype == BPF_T_NONE)
2060 return (BPF_TSTAMP_NONE);
2061 if ((tstype & BPF_T_FAST) != 0)
2062 return (BPF_TSTAMP_FAST);
2064 return (BPF_TSTAMP_NORMAL);
2068 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2073 quality = bpf_ts_quality(tstype);
2074 if (quality == BPF_TSTAMP_NONE)
2078 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2080 *bt = *(struct bintime *)(tag + 1);
2081 return (BPF_TSTAMP_EXTERN);
2084 if (quality == BPF_TSTAMP_NORMAL)
2093 * Incoming linkage from device drivers. Process the packet pkt, of length
2094 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2095 * by each process' filter, and if accepted, stashed into the corresponding
2099 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2109 gottime = BPF_TSTAMP_NONE;
2113 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2115 * We are not using any locks for d here because:
2116 * 1) any filter change is protected by interface
2118 * 2) destroying/detaching d is protected by interface
2122 /* XXX: Do not protect counter for the sake of performance. */
2125 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2126 * way for the caller to indiciate to us whether this packet
2127 * is inbound or outbound. In the bpf_mtap() routines, we use
2128 * the interface pointers on the mbuf to figure it out.
2131 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2133 slen = (*(bf->func))(pkt, pktlen, pktlen);
2136 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2139 * Filter matches. Let's to acquire write lock.
2144 if (gottime < bpf_ts_quality(d->bd_tstamp))
2145 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2147 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2149 catchpacket(d, pkt, pktlen, slen,
2150 bpf_append_bytes, &bt);
2157 #define BPF_CHECK_DIRECTION(d, r, i) \
2158 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2159 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2162 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2163 * Locking model is explained in bpf_tap().
2166 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2176 /* Skip outgoing duplicate packets. */
2177 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2178 m->m_flags &= ~M_PROMISC;
2182 pktlen = m_length(m, NULL);
2183 gottime = BPF_TSTAMP_NONE;
2187 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2188 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2192 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2193 /* XXX We cannot handle multiple mbufs. */
2194 if (bf != NULL && m->m_next == NULL)
2195 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2198 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2203 if (gottime < bpf_ts_quality(d->bd_tstamp))
2204 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2206 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2208 catchpacket(d, (u_char *)m, pktlen, slen,
2209 bpf_append_mbuf, &bt);
2217 * Incoming linkage from device drivers, when packet is in
2218 * an mbuf chain and to be prepended by a contiguous header.
2221 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2229 /* Skip outgoing duplicate packets. */
2230 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2231 m->m_flags &= ~M_PROMISC;
2235 pktlen = m_length(m, NULL);
2237 * Craft on-stack mbuf suitable for passing to bpf_filter.
2238 * Note that we cut corners here; we only setup what's
2239 * absolutely needed--this mbuf should never go anywhere else.
2246 gottime = BPF_TSTAMP_NONE;
2250 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2251 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2254 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2259 if (gottime < bpf_ts_quality(d->bd_tstamp))
2260 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2262 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2264 catchpacket(d, (u_char *)&mb, pktlen, slen,
2265 bpf_append_mbuf, &bt);
2272 #undef BPF_CHECK_DIRECTION
2274 #undef BPF_TSTAMP_NONE
2275 #undef BPF_TSTAMP_FAST
2276 #undef BPF_TSTAMP_NORMAL
2277 #undef BPF_TSTAMP_EXTERN
2280 bpf_hdrlen(struct bpf_d *d)
2284 hdrlen = d->bd_bif->bif_hdrlen;
2285 #ifndef BURN_BRIDGES
2286 if (d->bd_tstamp == BPF_T_NONE ||
2287 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2288 #ifdef COMPAT_FREEBSD32
2290 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2293 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2296 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2297 #ifdef COMPAT_FREEBSD32
2299 hdrlen = BPF_WORDALIGN32(hdrlen);
2302 hdrlen = BPF_WORDALIGN(hdrlen);
2304 return (hdrlen - d->bd_bif->bif_hdrlen);
2308 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2312 struct timespec tsn;
2314 if ((tstype & BPF_T_MONOTONIC) == 0) {
2316 bintime_add(&bt2, &boottimebin);
2319 switch (BPF_T_FORMAT(tstype)) {
2320 case BPF_T_MICROTIME:
2321 bintime2timeval(bt, &tsm);
2322 ts->bt_sec = tsm.tv_sec;
2323 ts->bt_frac = tsm.tv_usec;
2325 case BPF_T_NANOTIME:
2326 bintime2timespec(bt, &tsn);
2327 ts->bt_sec = tsn.tv_sec;
2328 ts->bt_frac = tsn.tv_nsec;
2331 ts->bt_sec = bt->sec;
2332 ts->bt_frac = bt->frac;
2338 * Move the packet data from interface memory (pkt) into the
2339 * store buffer. "cpfn" is the routine called to do the actual data
2340 * transfer. bcopy is passed in to copy contiguous chunks, while
2341 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2342 * pkt is really an mbuf.
2345 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2346 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2349 struct bpf_xhdr hdr;
2350 #ifndef BURN_BRIDGES
2351 struct bpf_hdr hdr_old;
2352 #ifdef COMPAT_FREEBSD32
2353 struct bpf_hdr32 hdr32_old;
2356 int caplen, curlen, hdrlen, totlen;
2361 BPFD_LOCK_ASSERT(d);
2364 * Detect whether user space has released a buffer back to us, and if
2365 * so, move it from being a hold buffer to a free buffer. This may
2366 * not be the best place to do it (for example, we might only want to
2367 * run this check if we need the space), but for now it's a reliable
2370 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2371 while (d->bd_hbuf_in_use)
2372 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2373 PRINET, "bd_hbuf", 0);
2374 d->bd_fbuf = d->bd_hbuf;
2377 bpf_buf_reclaimed(d);
2381 * Figure out how many bytes to move. If the packet is
2382 * greater or equal to the snapshot length, transfer that
2383 * much. Otherwise, transfer the whole packet (unless
2384 * we hit the buffer size limit).
2386 hdrlen = bpf_hdrlen(d);
2387 totlen = hdrlen + min(snaplen, pktlen);
2388 if (totlen > d->bd_bufsize)
2389 totlen = d->bd_bufsize;
2392 * Round up the end of the previous packet to the next longword.
2394 * Drop the packet if there's no room and no hope of room
2395 * If the packet would overflow the storage buffer or the storage
2396 * buffer is considered immutable by the buffer model, try to rotate
2397 * the buffer and wakeup pending processes.
2399 #ifdef COMPAT_FREEBSD32
2401 curlen = BPF_WORDALIGN32(d->bd_slen);
2404 curlen = BPF_WORDALIGN(d->bd_slen);
2405 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2406 if (d->bd_fbuf == NULL) {
2408 * There's no room in the store buffer, and no
2409 * prospect of room, so drop the packet. Notify the
2416 while (d->bd_hbuf_in_use)
2417 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
2418 PRINET, "bd_hbuf", 0);
2422 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2424 * Immediate mode is set, or the read timeout has already
2425 * expired during a select call. A packet arrived, so the
2426 * reader should be woken up.
2429 caplen = totlen - hdrlen;
2430 tstype = d->bd_tstamp;
2431 do_timestamp = tstype != BPF_T_NONE;
2432 #ifndef BURN_BRIDGES
2433 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2436 bpf_bintime2ts(bt, &ts, tstype);
2437 #ifdef COMPAT_FREEBSD32
2438 if (d->bd_compat32) {
2439 bzero(&hdr32_old, sizeof(hdr32_old));
2441 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2442 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2444 hdr32_old.bh_datalen = pktlen;
2445 hdr32_old.bh_hdrlen = hdrlen;
2446 hdr32_old.bh_caplen = caplen;
2447 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2452 bzero(&hdr_old, sizeof(hdr_old));
2454 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2455 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2457 hdr_old.bh_datalen = pktlen;
2458 hdr_old.bh_hdrlen = hdrlen;
2459 hdr_old.bh_caplen = caplen;
2460 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2467 * Append the bpf header. Note we append the actual header size, but
2468 * move forward the length of the header plus padding.
2470 bzero(&hdr, sizeof(hdr));
2472 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2473 hdr.bh_datalen = pktlen;
2474 hdr.bh_hdrlen = hdrlen;
2475 hdr.bh_caplen = caplen;
2476 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2479 * Copy the packet data into the store buffer and update its length.
2481 #ifndef BURN_BRIDGES
2484 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2485 d->bd_slen = curlen + totlen;
2492 * Free buffers currently in use by a descriptor.
2496 bpf_freed(struct bpf_d *d)
2500 * We don't need to lock out interrupts since this descriptor has
2501 * been detached from its interface and it yet hasn't been marked
2505 if (d->bd_rfilter != NULL) {
2506 free((caddr_t)d->bd_rfilter, M_BPF);
2508 if (d->bd_bfilter != NULL)
2509 bpf_destroy_jit_filter(d->bd_bfilter);
2512 if (d->bd_wfilter != NULL)
2513 free((caddr_t)d->bd_wfilter, M_BPF);
2514 mtx_destroy(&d->bd_lock);
2518 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2519 * fixed size of the link header (variable length headers not yet supported).
2522 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2525 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2529 * Attach an interface to bpf. ifp is a pointer to the structure
2530 * defining the interface to be attached, dlt is the link layer type,
2531 * and hdrlen is the fixed size of the link header (variable length
2532 * headers are not yet supporrted).
2535 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2539 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2543 LIST_INIT(&bp->bif_dlist);
2544 LIST_INIT(&bp->bif_wlist);
2547 rw_init(&bp->bif_lock, "bpf interface lock");
2548 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2552 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2555 bp->bif_hdrlen = hdrlen;
2558 if_printf(ifp, "bpf attached\n");
2562 * Detach bpf from an interface. This involves detaching each descriptor
2563 * associated with the interface. Notify each descriptor as it's detached
2564 * so that any sleepers wake up and get ENXIO.
2567 bpfdetach(struct ifnet *ifp)
2578 /* Find all bpf_if struct's which reference ifp and detach them. */
2580 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2581 if (ifp == bp->bif_ifp)
2585 LIST_REMOVE(bp, bif_next);
2591 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2592 bpf_detachd_locked(d);
2597 /* Free writer-only descriptors */
2598 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2599 bpf_detachd_locked(d);
2606 * Delay freing bp till interface is detached
2607 * and all routes through this interface are removed.
2608 * Mark bp as detached to restrict new consumers.
2611 bp->flags |= BPFIF_FLAG_DYING;
2614 } while (bp != NULL);
2619 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2624 * Interface departure handler.
2625 * Note departure event does not guarantee interface is going down.
2628 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2633 if ((bp = ifp->if_bpf) == NULL) {
2638 /* Check if bpfdetach() was called previously */
2639 if ((bp->flags & BPFIF_FLAG_DYING) == 0) {
2644 CTR3(KTR_NET, "%s: freing BPF instance %p for interface %p",
2650 rw_destroy(&bp->bif_lock);
2655 * Get a list of available data link type of the interface.
2658 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2666 ifp = d->bd_bif->bif_ifp;
2669 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2670 if (bp->bif_ifp != ifp)
2672 if (bfl->bfl_list != NULL) {
2673 if (n >= bfl->bfl_len)
2675 error = copyout(&bp->bif_dlt,
2676 bfl->bfl_list + n, sizeof(u_int));
2685 * Set the data link type of a BPF instance.
2688 bpf_setdlt(struct bpf_d *d, u_int dlt)
2690 int error, opromisc;
2696 if (d->bd_bif->bif_dlt == dlt)
2698 ifp = d->bd_bif->bif_ifp;
2700 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2701 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2706 opromisc = d->bd_promisc;
2712 error = ifpromisc(bp->bif_ifp, 1);
2714 if_printf(bp->bif_ifp,
2715 "bpf_setdlt: ifpromisc failed (%d)\n",
2721 return (bp == NULL ? EINVAL : 0);
2725 bpf_drvinit(void *unused)
2729 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2730 LIST_INIT(&bpf_iflist);
2732 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2733 /* For compatibility */
2734 make_dev_alias(dev, "bpf0");
2736 /* Register interface departure handler */
2737 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2738 ifnet_departure_event, bpf_ifdetach, NULL,
2739 EVENTHANDLER_PRI_ANY);
2743 * Zero out the various packet counters associated with all of the bpf
2744 * descriptors. At some point, we will probably want to get a bit more
2745 * granular and allow the user to specify descriptors to be zeroed.
2748 bpf_zero_counters(void)
2754 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2756 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2772 * Fill filter statistics
2775 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2778 bzero(d, sizeof(*d));
2779 BPFD_LOCK_ASSERT(bd);
2780 d->bd_structsize = sizeof(*d);
2781 /* XXX: reading should be protected by global lock */
2782 d->bd_immediate = bd->bd_immediate;
2783 d->bd_promisc = bd->bd_promisc;
2784 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2785 d->bd_direction = bd->bd_direction;
2786 d->bd_feedback = bd->bd_feedback;
2787 d->bd_async = bd->bd_async;
2788 d->bd_rcount = bd->bd_rcount;
2789 d->bd_dcount = bd->bd_dcount;
2790 d->bd_fcount = bd->bd_fcount;
2791 d->bd_sig = bd->bd_sig;
2792 d->bd_slen = bd->bd_slen;
2793 d->bd_hlen = bd->bd_hlen;
2794 d->bd_bufsize = bd->bd_bufsize;
2795 d->bd_pid = bd->bd_pid;
2796 strlcpy(d->bd_ifname,
2797 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2798 d->bd_locked = bd->bd_locked;
2799 d->bd_wcount = bd->bd_wcount;
2800 d->bd_wdcount = bd->bd_wdcount;
2801 d->bd_wfcount = bd->bd_wfcount;
2802 d->bd_zcopy = bd->bd_zcopy;
2803 d->bd_bufmode = bd->bd_bufmode;
2807 * Handle `netstat -B' stats request
2810 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2812 static const struct xbpf_d zerostats;
2813 struct xbpf_d *xbdbuf, *xbd, tempstats;
2819 * XXX This is not technically correct. It is possible for non
2820 * privileged users to open bpf devices. It would make sense
2821 * if the users who opened the devices were able to retrieve
2822 * the statistics for them, too.
2824 error = priv_check(req->td, PRIV_NET_BPF);
2828 * Check to see if the user is requesting that the counters be
2829 * zeroed out. Explicitly check that the supplied data is zeroed,
2830 * as we aren't allowing the user to set the counters currently.
2832 if (req->newptr != NULL) {
2833 if (req->newlen != sizeof(tempstats))
2835 memset(&tempstats, 0, sizeof(tempstats));
2836 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2839 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2841 bpf_zero_counters();
2844 if (req->oldptr == NULL)
2845 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2846 if (bpf_bpfd_cnt == 0)
2847 return (SYSCTL_OUT(req, 0, 0));
2848 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2850 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2852 free(xbdbuf, M_BPF);
2856 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2858 /* Send writers-only first */
2859 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2860 xbd = &xbdbuf[index++];
2862 bpfstats_fill_xbpf(xbd, bd);
2865 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2866 xbd = &xbdbuf[index++];
2868 bpfstats_fill_xbpf(xbd, bd);
2874 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2875 free(xbdbuf, M_BPF);
2879 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2881 #else /* !DEV_BPF && !NETGRAPH_BPF */
2883 * NOP stubs to allow bpf-using drivers to load and function.
2885 * A 'better' implementation would allow the core bpf functionality
2886 * to be loaded at runtime.
2888 static struct bpf_if bp_null;
2891 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2896 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2901 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2906 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2909 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2913 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2916 *driverp = &bp_null;
2920 bpfdetach(struct ifnet *ifp)
2925 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
2927 return -1; /* "no filter" behaviour */
2931 bpf_validate(const struct bpf_insn *f, int len)
2933 return 0; /* false */
2936 #endif /* !DEV_BPF && !NETGRAPH_BPF */