2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1990, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from the Stanford/CMU enet packet filter,
8 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
9 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10 * Berkeley Laboratory.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
43 #include "opt_compat.h"
45 #include "opt_netgraph.h"
47 #include <sys/types.h>
48 #include <sys/param.h>
50 #include <sys/rwlock.h>
51 #include <sys/systm.h>
53 #include <sys/fcntl.h>
55 #include <sys/malloc.h>
60 #include <sys/signalvar.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
63 #include <sys/ttycom.h>
65 #include <sys/sysent.h>
67 #include <sys/event.h>
72 #include <sys/socket.h>
79 #include <net/if_var.h>
80 #include <net/if_dl.h>
82 #include <net/bpf_buffer.h>
84 #include <net/bpf_jitter.h>
86 #include <net/bpf_zerocopy.h>
87 #include <net/bpfdesc.h>
88 #include <net/route.h>
91 #include <netinet/in.h>
92 #include <netinet/if_ether.h>
93 #include <sys/kernel.h>
94 #include <sys/sysctl.h>
96 #include <net80211/ieee80211_freebsd.h>
98 #include <security/mac/mac_framework.h>
100 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
103 #define bif_next bif_ext.bif_next
104 #define bif_dlist bif_ext.bif_dlist
105 struct bpf_if_ext bif_ext; /* public members */
106 u_int bif_dlt; /* link layer type */
107 u_int bif_hdrlen; /* length of link header */
108 struct ifnet *bif_ifp; /* corresponding interface */
109 struct rwlock bif_lock; /* interface lock */
110 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */
111 int bif_flags; /* Interface flags */
112 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
115 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
117 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
119 #define PRINET 26 /* interruptible */
121 #define SIZEOF_BPF_HDR(type) \
122 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
124 #ifdef COMPAT_FREEBSD32
125 #include <sys/mount.h>
126 #include <compat/freebsd32/freebsd32.h>
127 #define BPF_ALIGNMENT32 sizeof(int32_t)
128 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
132 * 32-bit version of structure prepended to each packet. We use this header
133 * instead of the standard one for 32-bit streams. We mark the a stream as
134 * 32-bit the first time we see a 32-bit compat ioctl request.
137 struct timeval32 bh_tstamp; /* time stamp */
138 uint32_t bh_caplen; /* length of captured portion */
139 uint32_t bh_datalen; /* original length of packet */
140 uint16_t bh_hdrlen; /* length of bpf header (this struct
141 plus alignment padding) */
145 struct bpf_program32 {
150 struct bpf_dltlist32 {
155 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
156 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
157 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
158 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
159 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
160 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
164 * bpf_iflist is a list of BPF interface structures, each corresponding to a
165 * specific DLT. The same network interface might have several BPF interface
166 * structures registered by different layers in the stack (i.e., 802.11
167 * frames, ethernet frames, etc).
169 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist;
170 static struct mtx bpf_mtx; /* bpf global lock */
171 static int bpf_bpfd_cnt;
173 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
174 static void bpf_detachd(struct bpf_d *);
175 static void bpf_detachd_locked(struct bpf_d *);
176 static void bpf_freed(struct bpf_d *);
177 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
178 struct sockaddr *, int *, struct bpf_d *);
179 static int bpf_setif(struct bpf_d *, struct ifreq *);
180 static void bpf_timed_out(void *);
182 bpf_wakeup(struct bpf_d *);
183 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
184 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
186 static void reset_d(struct bpf_d *);
187 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
188 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
189 static int bpf_setdlt(struct bpf_d *, u_int);
190 static void filt_bpfdetach(struct knote *);
191 static int filt_bpfread(struct knote *, long);
192 static void bpf_drvinit(void *);
193 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
195 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
196 int bpf_maxinsns = BPF_MAXINSNS;
197 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
198 &bpf_maxinsns, 0, "Maximum bpf program instructions");
199 static int bpf_zerocopy_enable = 0;
200 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
201 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
202 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
203 bpf_stats_sysctl, "bpf statistics portal");
205 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
206 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
207 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(bpf_optimize_writers), 0,
209 "Do not send packets until BPF program is set");
211 static d_open_t bpfopen;
212 static d_read_t bpfread;
213 static d_write_t bpfwrite;
214 static d_ioctl_t bpfioctl;
215 static d_poll_t bpfpoll;
216 static d_kqfilter_t bpfkqfilter;
218 static struct cdevsw bpf_cdevsw = {
219 .d_version = D_VERSION,
226 .d_kqfilter = bpfkqfilter,
229 static struct filterops bpfread_filtops = {
231 .f_detach = filt_bpfdetach,
232 .f_event = filt_bpfread,
235 eventhandler_tag bpf_ifdetach_cookie = NULL;
238 * LOCKING MODEL USED BY BPF:
240 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
241 * some global counters and every bpf_if reference.
242 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
243 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
244 * used by bpf_mtap code.
248 * Global lock, interface lock, descriptor lock
250 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
251 * working model. In many places (like bpf_detachd) we start with BPF descriptor
252 * (and we need to at least rlock it to get reliable interface pointer). This
253 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
254 * change in every such place.
256 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
257 * 3) descriptor main wlock.
258 * Reading bd_bif can be protected by any of these locks, typically global lock.
260 * Changing read/write BPF filter is protected by the same three locks,
261 * the same applies for reading.
263 * Sleeping in global lock is not allowed due to bpfdetach() using it.
267 * Wrapper functions for various buffering methods. If the set of buffer
268 * modes expands, we will probably want to introduce a switch data structure
269 * similar to protosw, et.
272 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
278 switch (d->bd_bufmode) {
279 case BPF_BUFMODE_BUFFER:
280 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
282 case BPF_BUFMODE_ZBUF:
284 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
287 panic("bpf_buf_append_bytes");
292 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
298 switch (d->bd_bufmode) {
299 case BPF_BUFMODE_BUFFER:
300 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
302 case BPF_BUFMODE_ZBUF:
304 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
307 panic("bpf_buf_append_mbuf");
312 * This function gets called when the free buffer is re-assigned.
315 bpf_buf_reclaimed(struct bpf_d *d)
320 switch (d->bd_bufmode) {
321 case BPF_BUFMODE_BUFFER:
324 case BPF_BUFMODE_ZBUF:
325 bpf_zerocopy_buf_reclaimed(d);
329 panic("bpf_buf_reclaimed");
334 * If the buffer mechanism has a way to decide that a held buffer can be made
335 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
336 * returned if the buffer can be discarded, (0) is returned if it cannot.
339 bpf_canfreebuf(struct bpf_d *d)
344 switch (d->bd_bufmode) {
345 case BPF_BUFMODE_ZBUF:
346 return (bpf_zerocopy_canfreebuf(d));
352 * Allow the buffer model to indicate that the current store buffer is
353 * immutable, regardless of the appearance of space. Return (1) if the
354 * buffer is writable, and (0) if not.
357 bpf_canwritebuf(struct bpf_d *d)
361 switch (d->bd_bufmode) {
362 case BPF_BUFMODE_ZBUF:
363 return (bpf_zerocopy_canwritebuf(d));
369 * Notify buffer model that an attempt to write to the store buffer has
370 * resulted in a dropped packet, in which case the buffer may be considered
374 bpf_buffull(struct bpf_d *d)
379 switch (d->bd_bufmode) {
380 case BPF_BUFMODE_ZBUF:
381 bpf_zerocopy_buffull(d);
387 * Notify the buffer model that a buffer has moved into the hold position.
390 bpf_bufheld(struct bpf_d *d)
395 switch (d->bd_bufmode) {
396 case BPF_BUFMODE_ZBUF:
397 bpf_zerocopy_bufheld(d);
403 bpf_free(struct bpf_d *d)
406 switch (d->bd_bufmode) {
407 case BPF_BUFMODE_BUFFER:
408 return (bpf_buffer_free(d));
410 case BPF_BUFMODE_ZBUF:
411 return (bpf_zerocopy_free(d));
414 panic("bpf_buf_free");
419 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
422 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
424 return (bpf_buffer_uiomove(d, buf, len, uio));
428 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
431 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
433 return (bpf_buffer_ioctl_sblen(d, i));
437 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
440 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
442 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
446 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
449 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
451 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
455 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
458 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
460 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
464 * General BPF functions.
467 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
468 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
470 const struct ieee80211_bpf_params *p;
471 struct ether_header *eh;
479 * Build a sockaddr based on the data link layer type.
480 * We do this at this level because the ethernet header
481 * is copied directly into the data field of the sockaddr.
482 * In the case of SLIP, there is no header and the packet
483 * is forwarded as is.
484 * Also, we are careful to leave room at the front of the mbuf
485 * for the link level header.
490 sockp->sa_family = AF_INET;
495 sockp->sa_family = AF_UNSPEC;
496 /* XXX Would MAXLINKHDR be better? */
497 hlen = ETHER_HDR_LEN;
501 sockp->sa_family = AF_IMPLINK;
506 sockp->sa_family = AF_UNSPEC;
512 * null interface types require a 4 byte pseudo header which
513 * corresponds to the address family of the packet.
515 sockp->sa_family = AF_UNSPEC;
519 case DLT_ATM_RFC1483:
521 * en atm driver requires 4-byte atm pseudo header.
522 * though it isn't standard, vpi:vci needs to be
525 sockp->sa_family = AF_UNSPEC;
526 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
530 sockp->sa_family = AF_UNSPEC;
531 hlen = 4; /* This should match PPP_HDRLEN */
534 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
535 sockp->sa_family = AF_IEEE80211;
539 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
540 sockp->sa_family = AF_IEEE80211;
541 sockp->sa_len = 12; /* XXX != 0 */
542 hlen = sizeof(struct ieee80211_bpf_params);
549 len = uio->uio_resid;
550 if (len < hlen || len - hlen > ifp->if_mtu)
553 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
556 m->m_pkthdr.len = m->m_len = len;
559 error = uiomove(mtod(m, u_char *), len, uio);
563 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
569 /* Check for multicast destination */
572 eh = mtod(m, struct ether_header *);
573 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
574 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
575 ETHER_ADDR_LEN) == 0)
576 m->m_flags |= M_BCAST;
578 m->m_flags |= M_MCAST;
580 if (d->bd_hdrcmplt == 0) {
581 memcpy(eh->ether_shost, IF_LLADDR(ifp),
582 sizeof(eh->ether_shost));
588 * Make room for link header, and copy it to sockaddr
591 if (sockp->sa_family == AF_IEEE80211) {
593 * Collect true length from the parameter header
594 * NB: sockp is known to be zero'd so if we do a
595 * short copy unspecified parameters will be
597 * NB: packet may not be aligned after stripping
601 p = mtod(m, const struct ieee80211_bpf_params *);
603 if (hlen > sizeof(sockp->sa_data)) {
608 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
619 * Attach file to the bpf interface, i.e. make d listen on bp.
622 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
629 * Save sysctl value to protect from sysctl change
632 op_w = V_bpf_optimize_writers || d->bd_writer;
634 if (d->bd_bif != NULL)
635 bpf_detachd_locked(d);
637 * Point d at bp, and add d to the interface's list.
638 * Since there are many applications using BPF for
639 * sending raw packets only (dhcpd, cdpd are good examples)
640 * we can delay adding d to the list of active listeners until
641 * some filter is configured.
650 /* Add to writers-only list */
651 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
653 * We decrement bd_writer on every filter set operation.
654 * First BIOCSETF is done by pcap_open_live() to set up
655 * snap length. After that appliation usually sets its own filter
659 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
666 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
667 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
670 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
674 * Check if we need to upgrade our descriptor @d from write-only mode.
677 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
679 int is_snap, need_upgrade;
682 * Check if we've already upgraded or new filter is empty.
684 if (d->bd_writer == 0 || fcode == NULL)
690 * Check if cmd looks like snaplen setting from
691 * pcap_bpf.c:pcap_open_live().
692 * Note we're not checking .k value here:
693 * while pcap_open_live() definitely sets to non-zero value,
694 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
695 * do not consider upgrading immediately
697 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
704 * We're setting first filter and it doesn't look like
705 * setting snaplen. We're probably using bpf directly.
706 * Upgrade immediately.
711 * Do not require upgrade by first BIOCSETF
712 * (used to set snaplen) by pcap_open_live().
715 if (--d->bd_writer == 0) {
717 * First snaplen filter has already
718 * been set. This is probably catch-all
726 "%s: filter function set by pid %d, "
727 "bd_writer counter %d, snap %d upgrade %d",
728 __func__, d->bd_pid, d->bd_writer,
729 is_snap, need_upgrade);
731 return (need_upgrade);
735 * Add d to the list of active bp filters.
736 * Requires bpf_attachd() to be called before.
739 bpf_upgraded(struct bpf_d *d)
748 * Filter can be set several times without specifying interface.
749 * Mark d as reader and exit.
761 /* Remove from writers-only list */
762 LIST_REMOVE(d, bd_next);
763 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
764 /* Mark d as reader */
770 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
772 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
776 * Detach a file from its interface.
779 bpf_detachd(struct bpf_d *d)
782 bpf_detachd_locked(d);
787 bpf_detachd_locked(struct bpf_d *d)
793 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
797 /* Check if descriptor is attached */
798 if ((bp = d->bd_bif) == NULL)
804 /* Save bd_writer value */
805 error = d->bd_writer;
808 * Remove d from the interface's descriptor list.
810 LIST_REMOVE(d, bd_next);
819 /* Call event handler iff d is attached */
821 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
824 * Check if this descriptor had requested promiscuous mode.
825 * If so, turn it off.
829 CURVNET_SET(ifp->if_vnet);
830 error = ifpromisc(ifp, 0);
832 if (error != 0 && error != ENXIO) {
834 * ENXIO can happen if a pccard is unplugged
835 * Something is really wrong if we were able to put
836 * the driver into promiscuous mode, but can't
839 if_printf(bp->bif_ifp,
840 "bpf_detach: ifpromisc failed (%d)\n", error);
846 * Close the descriptor by detaching it from its interface,
847 * deallocating its buffers, and marking it free.
852 struct bpf_d *d = data;
855 if (d->bd_state == BPF_WAITING)
856 callout_stop(&d->bd_callout);
857 d->bd_state = BPF_IDLE;
859 funsetown(&d->bd_sigio);
862 mac_bpfdesc_destroy(d);
864 seldrain(&d->bd_sel);
865 knlist_destroy(&d->bd_sel.si_note);
866 callout_drain(&d->bd_callout);
872 * Open ethernet device. Returns ENXIO for illegal minor device number,
873 * EBUSY if file is open by another process.
877 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
882 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
883 error = devfs_set_cdevpriv(d, bpf_dtor);
890 * For historical reasons, perform a one-time initialization call to
891 * the buffer routines, even though we're not yet committed to a
892 * particular buffer method.
895 if ((flags & FREAD) == 0)
897 d->bd_hbuf_in_use = 0;
898 d->bd_bufmode = BPF_BUFMODE_BUFFER;
900 d->bd_direction = BPF_D_INOUT;
901 BPF_PID_REFRESH(d, td);
904 mac_bpfdesc_create(td->td_ucred, d);
906 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
907 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
908 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
914 * bpfread - read next chunk of packets from buffers
917 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
924 error = devfs_get_cdevpriv((void **)&d);
929 * Restrict application to use a buffer the same size as
932 if (uio->uio_resid != d->bd_bufsize)
935 non_block = ((ioflag & O_NONBLOCK) != 0);
938 BPF_PID_REFRESH_CUR(d);
939 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
943 if (d->bd_state == BPF_WAITING)
944 callout_stop(&d->bd_callout);
945 timed_out = (d->bd_state == BPF_TIMED_OUT);
946 d->bd_state = BPF_IDLE;
947 while (d->bd_hbuf_in_use) {
948 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
949 PRINET|PCATCH, "bd_hbuf", 0);
956 * If the hold buffer is empty, then do a timed sleep, which
957 * ends when the timeout expires or when enough packets
958 * have arrived to fill the store buffer.
960 while (d->bd_hbuf == NULL) {
961 if (d->bd_slen != 0) {
963 * A packet(s) either arrived since the previous
964 * read or arrived while we were asleep.
966 if (d->bd_immediate || non_block || timed_out) {
968 * Rotate the buffers and return what's here
969 * if we are in immediate mode, non-blocking
970 * flag is set, or this descriptor timed out.
978 * No data is available, check to see if the bpf device
979 * is still pointed at a real interface. If not, return
980 * ENXIO so that the userland process knows to rebind
981 * it before using it again.
983 if (d->bd_bif == NULL) {
990 return (EWOULDBLOCK);
992 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
994 if (error == EINTR || error == ERESTART) {
998 if (error == EWOULDBLOCK) {
1000 * On a timeout, return what's in the buffer,
1001 * which may be nothing. If there is something
1002 * in the store buffer, we can rotate the buffers.
1006 * We filled up the buffer in between
1007 * getting the timeout and arriving
1008 * here, so we don't need to rotate.
1012 if (d->bd_slen == 0) {
1021 * At this point, we know we have something in the hold slot.
1023 d->bd_hbuf_in_use = 1;
1027 * Move data from hold buffer into user space.
1028 * We know the entire buffer is transferred since
1029 * we checked above that the read buffer is bpf_bufsize bytes.
1031 * We do not have to worry about simultaneous reads because
1032 * we waited for sole access to the hold buffer above.
1034 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1037 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1038 d->bd_fbuf = d->bd_hbuf;
1041 bpf_buf_reclaimed(d);
1042 d->bd_hbuf_in_use = 0;
1043 wakeup(&d->bd_hbuf_in_use);
1050 * If there are processes sleeping on this descriptor, wake them up.
1052 static __inline void
1053 bpf_wakeup(struct bpf_d *d)
1056 BPFD_LOCK_ASSERT(d);
1057 if (d->bd_state == BPF_WAITING) {
1058 callout_stop(&d->bd_callout);
1059 d->bd_state = BPF_IDLE;
1062 if (d->bd_async && d->bd_sig && d->bd_sigio)
1063 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1065 selwakeuppri(&d->bd_sel, PRINET);
1066 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1070 bpf_timed_out(void *arg)
1072 struct bpf_d *d = (struct bpf_d *)arg;
1074 BPFD_LOCK_ASSERT(d);
1076 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1078 if (d->bd_state == BPF_WAITING) {
1079 d->bd_state = BPF_TIMED_OUT;
1080 if (d->bd_slen != 0)
1086 bpf_ready(struct bpf_d *d)
1089 BPFD_LOCK_ASSERT(d);
1091 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1093 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1100 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1104 struct mbuf *m, *mc;
1105 struct sockaddr dst;
1109 error = devfs_get_cdevpriv((void **)&d);
1113 BPF_PID_REFRESH_CUR(d);
1115 /* XXX: locking required */
1116 if (d->bd_bif == NULL) {
1121 ifp = d->bd_bif->bif_ifp;
1123 if ((ifp->if_flags & IFF_UP) == 0) {
1128 if (uio->uio_resid == 0) {
1133 bzero(&dst, sizeof(dst));
1136 /* XXX: bpf_movein() can sleep */
1137 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1138 &m, &dst, &hlen, d);
1145 dst.sa_family = pseudo_AF_HDRCMPLT;
1147 if (d->bd_feedback) {
1148 mc = m_dup(m, M_NOWAIT);
1150 mc->m_pkthdr.rcvif = ifp;
1151 /* Set M_PROMISC for outgoing packets to be discarded. */
1152 if (d->bd_direction == BPF_D_INOUT)
1153 m->m_flags |= M_PROMISC;
1157 m->m_pkthdr.len -= hlen;
1159 m->m_data += hlen; /* XXX */
1161 CURVNET_SET(ifp->if_vnet);
1164 mac_bpfdesc_create_mbuf(d, m);
1166 mac_bpfdesc_create_mbuf(d, mc);
1170 bzero(&ro, sizeof(ro));
1172 ro.ro_prepend = (u_char *)&dst.sa_data;
1174 ro.ro_flags = RT_HAS_HEADER;
1177 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1183 (*ifp->if_input)(ifp, mc);
1193 * Reset a descriptor by flushing its packet buffer and clearing the receive
1194 * and drop counts. This is doable for kernel-only buffers, but with
1195 * zero-copy buffers, we can't write to (or rotate) buffers that are
1196 * currently owned by userspace. It would be nice if we could encapsulate
1197 * this logic in the buffer code rather than here.
1200 reset_d(struct bpf_d *d)
1203 BPFD_LOCK_ASSERT(d);
1205 while (d->bd_hbuf_in_use)
1206 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1208 if ((d->bd_hbuf != NULL) &&
1209 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1210 /* Free the hold buffer. */
1211 d->bd_fbuf = d->bd_hbuf;
1214 bpf_buf_reclaimed(d);
1216 if (bpf_canwritebuf(d))
1228 * FIONREAD Check for read packet available.
1229 * BIOCGBLEN Get buffer len [for read()].
1230 * BIOCSETF Set read filter.
1231 * BIOCSETFNR Set read filter without resetting descriptor.
1232 * BIOCSETWF Set write filter.
1233 * BIOCFLUSH Flush read packet buffer.
1234 * BIOCPROMISC Put interface into promiscuous mode.
1235 * BIOCGDLT Get link layer type.
1236 * BIOCGETIF Get interface name.
1237 * BIOCSETIF Set interface.
1238 * BIOCSRTIMEOUT Set read timeout.
1239 * BIOCGRTIMEOUT Get read timeout.
1240 * BIOCGSTATS Get packet stats.
1241 * BIOCIMMEDIATE Set immediate mode.
1242 * BIOCVERSION Get filter language version.
1243 * BIOCGHDRCMPLT Get "header already complete" flag
1244 * BIOCSHDRCMPLT Set "header already complete" flag
1245 * BIOCGDIRECTION Get packet direction flag
1246 * BIOCSDIRECTION Set packet direction flag
1247 * BIOCGTSTAMP Get time stamp format and resolution.
1248 * BIOCSTSTAMP Set time stamp format and resolution.
1249 * BIOCLOCK Set "locked" flag
1250 * BIOCFEEDBACK Set packet feedback mode.
1251 * BIOCSETZBUF Set current zero-copy buffer locations.
1252 * BIOCGETZMAX Get maximum zero-copy buffer size.
1253 * BIOCROTZBUF Force rotation of zero-copy buffer
1254 * BIOCSETBUFMODE Set buffer mode.
1255 * BIOCGETBUFMODE Get current buffer mode.
1259 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1265 error = devfs_get_cdevpriv((void **)&d);
1270 * Refresh PID associated with this descriptor.
1273 BPF_PID_REFRESH(d, td);
1274 if (d->bd_state == BPF_WAITING)
1275 callout_stop(&d->bd_callout);
1276 d->bd_state = BPF_IDLE;
1279 if (d->bd_locked == 1) {
1285 #ifdef COMPAT_FREEBSD32
1286 case BIOCGDLTLIST32:
1290 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1291 case BIOCGRTIMEOUT32:
1302 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1303 case BIOCSRTIMEOUT32:
1313 #ifdef COMPAT_FREEBSD32
1315 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1316 * that it will get 32-bit packet headers.
1322 case BIOCGDLTLIST32:
1323 case BIOCGRTIMEOUT32:
1324 case BIOCSRTIMEOUT32:
1325 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1333 CURVNET_SET(TD_TO_VNET(td));
1341 * Check for read packet available.
1349 while (d->bd_hbuf_in_use)
1350 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1351 PRINET, "bd_hbuf", 0);
1361 * Get buffer len [for read()].
1365 *(u_int *)addr = d->bd_bufsize;
1370 * Set buffer length.
1373 error = bpf_ioctl_sblen(d, (u_int *)addr);
1377 * Set link layer read filter.
1382 #ifdef COMPAT_FREEBSD32
1387 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1391 * Flush read packet buffer.
1400 * Put interface into promiscuous mode.
1403 if (d->bd_bif == NULL) {
1405 * No interface attached yet.
1410 if (d->bd_promisc == 0) {
1411 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1418 * Get current data link type.
1422 if (d->bd_bif == NULL)
1425 *(u_int *)addr = d->bd_bif->bif_dlt;
1430 * Get a list of supported data link types.
1432 #ifdef COMPAT_FREEBSD32
1433 case BIOCGDLTLIST32:
1435 struct bpf_dltlist32 *list32;
1436 struct bpf_dltlist dltlist;
1438 list32 = (struct bpf_dltlist32 *)addr;
1439 dltlist.bfl_len = list32->bfl_len;
1440 dltlist.bfl_list = PTRIN(list32->bfl_list);
1442 if (d->bd_bif == NULL)
1445 error = bpf_getdltlist(d, &dltlist);
1447 list32->bfl_len = dltlist.bfl_len;
1456 if (d->bd_bif == NULL)
1459 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1464 * Set data link type.
1468 if (d->bd_bif == NULL)
1471 error = bpf_setdlt(d, *(u_int *)addr);
1476 * Get interface name.
1480 if (d->bd_bif == NULL)
1483 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1484 struct ifreq *const ifr = (struct ifreq *)addr;
1486 strlcpy(ifr->ifr_name, ifp->if_xname,
1487 sizeof(ifr->ifr_name));
1497 int alloc_buf, size;
1500 * Behavior here depends on the buffering model. If
1501 * we're using kernel memory buffers, then we can
1502 * allocate them here. If we're using zero-copy,
1503 * then the user process must have registered buffers
1504 * by the time we get here.
1508 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1513 size = d->bd_bufsize;
1514 error = bpf_buffer_ioctl_sblen(d, &size);
1519 error = bpf_setif(d, (struct ifreq *)addr);
1528 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1529 case BIOCSRTIMEOUT32:
1532 struct timeval *tv = (struct timeval *)addr;
1533 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1534 struct timeval32 *tv32;
1535 struct timeval tv64;
1537 if (cmd == BIOCSRTIMEOUT32) {
1538 tv32 = (struct timeval32 *)addr;
1540 tv->tv_sec = tv32->tv_sec;
1541 tv->tv_usec = tv32->tv_usec;
1544 tv = (struct timeval *)addr;
1547 * Subtract 1 tick from tvtohz() since this isn't
1550 if ((error = itimerfix(tv)) == 0)
1551 d->bd_rtout = tvtohz(tv) - 1;
1559 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1560 case BIOCGRTIMEOUT32:
1564 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1565 struct timeval32 *tv32;
1566 struct timeval tv64;
1568 if (cmd == BIOCGRTIMEOUT32)
1572 tv = (struct timeval *)addr;
1574 tv->tv_sec = d->bd_rtout / hz;
1575 tv->tv_usec = (d->bd_rtout % hz) * tick;
1576 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1577 if (cmd == BIOCGRTIMEOUT32) {
1578 tv32 = (struct timeval32 *)addr;
1579 tv32->tv_sec = tv->tv_sec;
1580 tv32->tv_usec = tv->tv_usec;
1592 struct bpf_stat *bs = (struct bpf_stat *)addr;
1594 /* XXXCSJP overflow */
1595 bs->bs_recv = d->bd_rcount;
1596 bs->bs_drop = d->bd_dcount;
1601 * Set immediate mode.
1605 d->bd_immediate = *(u_int *)addr;
1611 struct bpf_version *bv = (struct bpf_version *)addr;
1613 bv->bv_major = BPF_MAJOR_VERSION;
1614 bv->bv_minor = BPF_MINOR_VERSION;
1619 * Get "header already complete" flag
1623 *(u_int *)addr = d->bd_hdrcmplt;
1628 * Set "header already complete" flag
1632 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1637 * Get packet direction flag
1639 case BIOCGDIRECTION:
1641 *(u_int *)addr = d->bd_direction;
1646 * Set packet direction flag
1648 case BIOCSDIRECTION:
1652 direction = *(u_int *)addr;
1653 switch (direction) {
1658 d->bd_direction = direction;
1668 * Get packet timestamp format and resolution.
1672 *(u_int *)addr = d->bd_tstamp;
1677 * Set packet timestamp format and resolution.
1683 func = *(u_int *)addr;
1684 if (BPF_T_VALID(func))
1685 d->bd_tstamp = func;
1693 d->bd_feedback = *(u_int *)addr;
1703 case FIONBIO: /* Non-blocking I/O */
1706 case FIOASYNC: /* Send signal on receive packets */
1708 d->bd_async = *(int *)addr;
1714 * XXX: Add some sort of locking here?
1715 * fsetown() can sleep.
1717 error = fsetown(*(int *)addr, &d->bd_sigio);
1722 *(int *)addr = fgetown(&d->bd_sigio);
1726 /* This is deprecated, FIOSETOWN should be used instead. */
1728 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1731 /* This is deprecated, FIOGETOWN should be used instead. */
1733 *(int *)addr = -fgetown(&d->bd_sigio);
1736 case BIOCSRSIG: /* Set receive signal */
1740 sig = *(u_int *)addr;
1753 *(u_int *)addr = d->bd_sig;
1757 case BIOCGETBUFMODE:
1759 *(u_int *)addr = d->bd_bufmode;
1763 case BIOCSETBUFMODE:
1765 * Allow the buffering mode to be changed as long as we
1766 * haven't yet committed to a particular mode. Our
1767 * definition of commitment, for now, is whether or not a
1768 * buffer has been allocated or an interface attached, since
1769 * that's the point where things get tricky.
1771 switch (*(u_int *)addr) {
1772 case BPF_BUFMODE_BUFFER:
1775 case BPF_BUFMODE_ZBUF:
1776 if (bpf_zerocopy_enable)
1786 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1787 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1792 d->bd_bufmode = *(u_int *)addr;
1797 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1801 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1805 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1813 * Set d's packet filter program to fp. If this file already has a filter,
1814 * free it and replace it. Returns EINVAL for bogus requests.
1816 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1817 * since reading d->bd_bif can't be protected by d or interface lock due to
1820 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1821 * interface read lock to read all filers.
1825 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1827 #ifdef COMPAT_FREEBSD32
1828 struct bpf_program fp_swab;
1829 struct bpf_program32 *fp32;
1831 struct bpf_insn *fcode, *old;
1833 bpf_jit_filter *jfunc, *ofunc;
1839 #ifdef COMPAT_FREEBSD32
1844 fp32 = (struct bpf_program32 *)fp;
1845 fp_swab.bf_len = fp32->bf_len;
1846 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1862 jfunc = ofunc = NULL;
1867 * Check new filter validness before acquiring any locks.
1868 * Allocate memory for new filter, if needed.
1871 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1873 size = flen * sizeof(*fp->bf_insns);
1875 /* We're setting up new filter. Copy and check actual data. */
1876 fcode = malloc(size, M_BPF, M_WAITOK);
1877 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1878 !bpf_validate(fcode, flen)) {
1883 /* Filter is copied inside fcode and is perfectly valid. */
1884 jfunc = bpf_jitter(fcode, flen);
1891 * Set up new filter.
1892 * Protect filter change by interface lock.
1893 * Additionally, we are protected by global lock here.
1895 if (d->bd_bif != NULL)
1896 BPFIF_WLOCK(d->bd_bif);
1898 if (cmd == BIOCSETWF) {
1899 old = d->bd_wfilter;
1900 d->bd_wfilter = fcode;
1902 old = d->bd_rfilter;
1903 d->bd_rfilter = fcode;
1905 ofunc = d->bd_bfilter;
1906 d->bd_bfilter = jfunc;
1908 if (cmd == BIOCSETF)
1911 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1914 if (d->bd_bif != NULL)
1915 BPFIF_WUNLOCK(d->bd_bif);
1920 bpf_destroy_jit_filter(ofunc);
1923 /* Move d to active readers list. */
1924 if (need_upgrade != 0)
1932 * Detach a file from its current interface (if attached at all) and attach
1933 * to the interface indicated by the name stored in ifr.
1934 * Return an errno or 0.
1937 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1940 struct ifnet *theywant;
1944 theywant = ifunit(ifr->ifr_name);
1945 if (theywant == NULL || theywant->if_bpf == NULL)
1948 bp = theywant->if_bpf;
1950 /* Check if interface is not being detached from BPF */
1952 if (bp->bif_flags & BPFIF_FLAG_DYING) {
1959 * At this point, we expect the buffer is already allocated. If not,
1962 switch (d->bd_bufmode) {
1963 case BPF_BUFMODE_BUFFER:
1964 case BPF_BUFMODE_ZBUF:
1965 if (d->bd_sbuf == NULL)
1970 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1972 if (bp != d->bd_bif)
1981 * Support for select() and poll() system calls
1983 * Return true iff the specific operation will not block indefinitely.
1984 * Otherwise, return false but make a note that a selwakeup() must be done.
1987 bpfpoll(struct cdev *dev, int events, struct thread *td)
1992 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1994 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1997 * Refresh PID associated with this descriptor.
1999 revents = events & (POLLOUT | POLLWRNORM);
2001 BPF_PID_REFRESH(d, td);
2002 if (events & (POLLIN | POLLRDNORM)) {
2004 revents |= events & (POLLIN | POLLRDNORM);
2006 selrecord(td, &d->bd_sel);
2007 /* Start the read timeout if necessary. */
2008 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2009 callout_reset(&d->bd_callout, d->bd_rtout,
2011 d->bd_state = BPF_WAITING;
2020 * Support for kevent() system call. Register EVFILT_READ filters and
2021 * reject all others.
2024 bpfkqfilter(struct cdev *dev, struct knote *kn)
2028 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2029 kn->kn_filter != EVFILT_READ)
2033 * Refresh PID associated with this descriptor.
2036 BPF_PID_REFRESH_CUR(d);
2037 kn->kn_fop = &bpfread_filtops;
2039 knlist_add(&d->bd_sel.si_note, kn, 1);
2046 filt_bpfdetach(struct knote *kn)
2048 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2050 knlist_remove(&d->bd_sel.si_note, kn, 0);
2054 filt_bpfread(struct knote *kn, long hint)
2056 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2059 BPFD_LOCK_ASSERT(d);
2060 ready = bpf_ready(d);
2062 kn->kn_data = d->bd_slen;
2064 * Ignore the hold buffer if it is being copied to user space.
2066 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2067 kn->kn_data += d->bd_hlen;
2068 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2069 callout_reset(&d->bd_callout, d->bd_rtout,
2071 d->bd_state = BPF_WAITING;
2077 #define BPF_TSTAMP_NONE 0
2078 #define BPF_TSTAMP_FAST 1
2079 #define BPF_TSTAMP_NORMAL 2
2080 #define BPF_TSTAMP_EXTERN 3
2083 bpf_ts_quality(int tstype)
2086 if (tstype == BPF_T_NONE)
2087 return (BPF_TSTAMP_NONE);
2088 if ((tstype & BPF_T_FAST) != 0)
2089 return (BPF_TSTAMP_FAST);
2091 return (BPF_TSTAMP_NORMAL);
2095 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2100 quality = bpf_ts_quality(tstype);
2101 if (quality == BPF_TSTAMP_NONE)
2105 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2107 *bt = *(struct bintime *)(tag + 1);
2108 return (BPF_TSTAMP_EXTERN);
2111 if (quality == BPF_TSTAMP_NORMAL)
2120 * Incoming linkage from device drivers. Process the packet pkt, of length
2121 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2122 * by each process' filter, and if accepted, stashed into the corresponding
2126 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2136 gottime = BPF_TSTAMP_NONE;
2140 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2142 * We are not using any locks for d here because:
2143 * 1) any filter change is protected by interface
2145 * 2) destroying/detaching d is protected by interface
2149 /* XXX: Do not protect counter for the sake of performance. */
2152 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2153 * way for the caller to indiciate to us whether this packet
2154 * is inbound or outbound. In the bpf_mtap() routines, we use
2155 * the interface pointers on the mbuf to figure it out.
2158 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2160 slen = (*(bf->func))(pkt, pktlen, pktlen);
2163 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2166 * Filter matches. Let's to acquire write lock.
2171 if (gottime < bpf_ts_quality(d->bd_tstamp))
2172 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2174 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2176 catchpacket(d, pkt, pktlen, slen,
2177 bpf_append_bytes, &bt);
2184 #define BPF_CHECK_DIRECTION(d, r, i) \
2185 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2186 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2189 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2190 * Locking model is explained in bpf_tap().
2193 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2203 /* Skip outgoing duplicate packets. */
2204 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2205 m->m_flags &= ~M_PROMISC;
2209 pktlen = m_length(m, NULL);
2210 gottime = BPF_TSTAMP_NONE;
2214 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2215 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2219 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2220 /* XXX We cannot handle multiple mbufs. */
2221 if (bf != NULL && m->m_next == NULL)
2222 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2225 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2230 if (gottime < bpf_ts_quality(d->bd_tstamp))
2231 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2233 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2235 catchpacket(d, (u_char *)m, pktlen, slen,
2236 bpf_append_mbuf, &bt);
2244 * Incoming linkage from device drivers, when packet is in
2245 * an mbuf chain and to be prepended by a contiguous header.
2248 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2256 /* Skip outgoing duplicate packets. */
2257 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2258 m->m_flags &= ~M_PROMISC;
2262 pktlen = m_length(m, NULL);
2264 * Craft on-stack mbuf suitable for passing to bpf_filter.
2265 * Note that we cut corners here; we only setup what's
2266 * absolutely needed--this mbuf should never go anywhere else.
2273 gottime = BPF_TSTAMP_NONE;
2277 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2278 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2281 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2286 if (gottime < bpf_ts_quality(d->bd_tstamp))
2287 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2289 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2291 catchpacket(d, (u_char *)&mb, pktlen, slen,
2292 bpf_append_mbuf, &bt);
2299 #undef BPF_CHECK_DIRECTION
2301 #undef BPF_TSTAMP_NONE
2302 #undef BPF_TSTAMP_FAST
2303 #undef BPF_TSTAMP_NORMAL
2304 #undef BPF_TSTAMP_EXTERN
2307 bpf_hdrlen(struct bpf_d *d)
2311 hdrlen = d->bd_bif->bif_hdrlen;
2312 #ifndef BURN_BRIDGES
2313 if (d->bd_tstamp == BPF_T_NONE ||
2314 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2315 #ifdef COMPAT_FREEBSD32
2317 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2320 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2323 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2324 #ifdef COMPAT_FREEBSD32
2326 hdrlen = BPF_WORDALIGN32(hdrlen);
2329 hdrlen = BPF_WORDALIGN(hdrlen);
2331 return (hdrlen - d->bd_bif->bif_hdrlen);
2335 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2337 struct bintime bt2, boottimebin;
2339 struct timespec tsn;
2341 if ((tstype & BPF_T_MONOTONIC) == 0) {
2343 getboottimebin(&boottimebin);
2344 bintime_add(&bt2, &boottimebin);
2347 switch (BPF_T_FORMAT(tstype)) {
2348 case BPF_T_MICROTIME:
2349 bintime2timeval(bt, &tsm);
2350 ts->bt_sec = tsm.tv_sec;
2351 ts->bt_frac = tsm.tv_usec;
2353 case BPF_T_NANOTIME:
2354 bintime2timespec(bt, &tsn);
2355 ts->bt_sec = tsn.tv_sec;
2356 ts->bt_frac = tsn.tv_nsec;
2359 ts->bt_sec = bt->sec;
2360 ts->bt_frac = bt->frac;
2366 * Move the packet data from interface memory (pkt) into the
2367 * store buffer. "cpfn" is the routine called to do the actual data
2368 * transfer. bcopy is passed in to copy contiguous chunks, while
2369 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2370 * pkt is really an mbuf.
2373 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2374 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2377 struct bpf_xhdr hdr;
2378 #ifndef BURN_BRIDGES
2379 struct bpf_hdr hdr_old;
2380 #ifdef COMPAT_FREEBSD32
2381 struct bpf_hdr32 hdr32_old;
2384 int caplen, curlen, hdrlen, totlen;
2389 BPFD_LOCK_ASSERT(d);
2392 * Detect whether user space has released a buffer back to us, and if
2393 * so, move it from being a hold buffer to a free buffer. This may
2394 * not be the best place to do it (for example, we might only want to
2395 * run this check if we need the space), but for now it's a reliable
2398 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2399 d->bd_fbuf = d->bd_hbuf;
2402 bpf_buf_reclaimed(d);
2406 * Figure out how many bytes to move. If the packet is
2407 * greater or equal to the snapshot length, transfer that
2408 * much. Otherwise, transfer the whole packet (unless
2409 * we hit the buffer size limit).
2411 hdrlen = bpf_hdrlen(d);
2412 totlen = hdrlen + min(snaplen, pktlen);
2413 if (totlen > d->bd_bufsize)
2414 totlen = d->bd_bufsize;
2417 * Round up the end of the previous packet to the next longword.
2419 * Drop the packet if there's no room and no hope of room
2420 * If the packet would overflow the storage buffer or the storage
2421 * buffer is considered immutable by the buffer model, try to rotate
2422 * the buffer and wakeup pending processes.
2424 #ifdef COMPAT_FREEBSD32
2426 curlen = BPF_WORDALIGN32(d->bd_slen);
2429 curlen = BPF_WORDALIGN(d->bd_slen);
2430 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2431 if (d->bd_fbuf == NULL) {
2433 * There's no room in the store buffer, and no
2434 * prospect of room, so drop the packet. Notify the
2441 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2445 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2447 * Immediate mode is set, or the read timeout has already
2448 * expired during a select call. A packet arrived, so the
2449 * reader should be woken up.
2452 caplen = totlen - hdrlen;
2453 tstype = d->bd_tstamp;
2454 do_timestamp = tstype != BPF_T_NONE;
2455 #ifndef BURN_BRIDGES
2456 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2459 bpf_bintime2ts(bt, &ts, tstype);
2460 #ifdef COMPAT_FREEBSD32
2461 if (d->bd_compat32) {
2462 bzero(&hdr32_old, sizeof(hdr32_old));
2464 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2465 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2467 hdr32_old.bh_datalen = pktlen;
2468 hdr32_old.bh_hdrlen = hdrlen;
2469 hdr32_old.bh_caplen = caplen;
2470 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2475 bzero(&hdr_old, sizeof(hdr_old));
2477 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2478 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2480 hdr_old.bh_datalen = pktlen;
2481 hdr_old.bh_hdrlen = hdrlen;
2482 hdr_old.bh_caplen = caplen;
2483 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2490 * Append the bpf header. Note we append the actual header size, but
2491 * move forward the length of the header plus padding.
2493 bzero(&hdr, sizeof(hdr));
2495 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2496 hdr.bh_datalen = pktlen;
2497 hdr.bh_hdrlen = hdrlen;
2498 hdr.bh_caplen = caplen;
2499 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2502 * Copy the packet data into the store buffer and update its length.
2504 #ifndef BURN_BRIDGES
2507 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2508 d->bd_slen = curlen + totlen;
2515 * Free buffers currently in use by a descriptor.
2519 bpf_freed(struct bpf_d *d)
2523 * We don't need to lock out interrupts since this descriptor has
2524 * been detached from its interface and it yet hasn't been marked
2528 if (d->bd_rfilter != NULL) {
2529 free((caddr_t)d->bd_rfilter, M_BPF);
2531 if (d->bd_bfilter != NULL)
2532 bpf_destroy_jit_filter(d->bd_bfilter);
2535 if (d->bd_wfilter != NULL)
2536 free((caddr_t)d->bd_wfilter, M_BPF);
2537 mtx_destroy(&d->bd_lock);
2541 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2542 * fixed size of the link header (variable length headers not yet supported).
2545 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2548 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2552 * Attach an interface to bpf. ifp is a pointer to the structure
2553 * defining the interface to be attached, dlt is the link layer type,
2554 * and hdrlen is the fixed size of the link header (variable length
2555 * headers are not yet supporrted).
2558 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2562 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2566 LIST_INIT(&bp->bif_dlist);
2567 LIST_INIT(&bp->bif_wlist);
2570 rw_init(&bp->bif_lock, "bpf interface lock");
2571 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2572 bp->bif_bpf = driverp;
2576 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2579 bp->bif_hdrlen = hdrlen;
2581 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2582 if_printf(ifp, "bpf attached\n");
2587 * When moving interfaces between vnet instances we need a way to
2588 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2589 * after the vmove. We unfortunately have no device driver infrastructure
2590 * to query the interface for these values after creation/attach, thus
2591 * add this as a workaround.
2594 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2599 if (bif_dlt == NULL && bif_hdrlen == NULL)
2602 if (bif_dlt != NULL)
2603 *bif_dlt = bp->bif_dlt;
2604 if (bif_hdrlen != NULL)
2605 *bif_hdrlen = bp->bif_hdrlen;
2612 * Detach bpf from an interface. This involves detaching each descriptor
2613 * associated with the interface. Notify each descriptor as it's detached
2614 * so that any sleepers wake up and get ENXIO.
2617 bpfdetach(struct ifnet *ifp)
2619 struct bpf_if *bp, *bp_temp;
2626 /* Find all bpf_if struct's which reference ifp and detach them. */
2627 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2628 if (ifp != bp->bif_ifp)
2631 LIST_REMOVE(bp, bif_next);
2632 /* Add to to-be-freed list */
2633 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2637 * Delay freeing bp till interface is detached
2638 * and all routes through this interface are removed.
2639 * Mark bp as detached to restrict new consumers.
2642 bp->bif_flags |= BPFIF_FLAG_DYING;
2643 *bp->bif_bpf = NULL;
2646 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2647 __func__, bp->bif_dlt, bp, ifp);
2649 /* Free common descriptors */
2650 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2651 bpf_detachd_locked(d);
2657 /* Free writer-only descriptors */
2658 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2659 bpf_detachd_locked(d);
2669 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2674 * Interface departure handler.
2675 * Note departure event does not guarantee interface is going down.
2676 * Interface renaming is currently done via departure/arrival event set.
2678 * Departure handled is called after all routes pointing to
2679 * given interface are removed and interface is in down state
2680 * restricting any packets to be sent/received. We assume it is now safe
2681 * to free data allocated by BPF.
2684 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2686 struct bpf_if *bp, *bp_temp;
2689 /* Ignore ifnet renaming. */
2690 if (ifp->if_flags & IFF_RENAMING)
2695 * Find matching entries in free list.
2696 * Nothing should be found if bpfdetach() was not called.
2698 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2699 if (ifp != bp->bif_ifp)
2702 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2705 LIST_REMOVE(bp, bif_next);
2707 rw_destroy(&bp->bif_lock);
2716 * Get a list of available data link type of the interface.
2719 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2728 ifp = d->bd_bif->bif_ifp;
2731 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2732 if (bp->bif_ifp == ifp)
2735 if (bfl->bfl_list == NULL) {
2739 if (n1 > bfl->bfl_len)
2742 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2745 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2746 if (bp->bif_ifp != ifp)
2752 lst[n] = bp->bif_dlt;
2756 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2764 * Set the data link type of a BPF instance.
2767 bpf_setdlt(struct bpf_d *d, u_int dlt)
2769 int error, opromisc;
2775 if (d->bd_bif->bif_dlt == dlt)
2777 ifp = d->bd_bif->bif_ifp;
2779 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2780 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2785 opromisc = d->bd_promisc;
2791 error = ifpromisc(bp->bif_ifp, 1);
2793 if_printf(bp->bif_ifp,
2794 "bpf_setdlt: ifpromisc failed (%d)\n",
2800 return (bp == NULL ? EINVAL : 0);
2804 bpf_drvinit(void *unused)
2808 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2809 LIST_INIT(&bpf_iflist);
2810 LIST_INIT(&bpf_freelist);
2812 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2813 /* For compatibility */
2814 make_dev_alias(dev, "bpf0");
2816 /* Register interface departure handler */
2817 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2818 ifnet_departure_event, bpf_ifdetach, NULL,
2819 EVENTHANDLER_PRI_ANY);
2823 * Zero out the various packet counters associated with all of the bpf
2824 * descriptors. At some point, we will probably want to get a bit more
2825 * granular and allow the user to specify descriptors to be zeroed.
2828 bpf_zero_counters(void)
2834 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2836 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2852 * Fill filter statistics
2855 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2858 bzero(d, sizeof(*d));
2859 BPFD_LOCK_ASSERT(bd);
2860 d->bd_structsize = sizeof(*d);
2861 /* XXX: reading should be protected by global lock */
2862 d->bd_immediate = bd->bd_immediate;
2863 d->bd_promisc = bd->bd_promisc;
2864 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2865 d->bd_direction = bd->bd_direction;
2866 d->bd_feedback = bd->bd_feedback;
2867 d->bd_async = bd->bd_async;
2868 d->bd_rcount = bd->bd_rcount;
2869 d->bd_dcount = bd->bd_dcount;
2870 d->bd_fcount = bd->bd_fcount;
2871 d->bd_sig = bd->bd_sig;
2872 d->bd_slen = bd->bd_slen;
2873 d->bd_hlen = bd->bd_hlen;
2874 d->bd_bufsize = bd->bd_bufsize;
2875 d->bd_pid = bd->bd_pid;
2876 strlcpy(d->bd_ifname,
2877 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2878 d->bd_locked = bd->bd_locked;
2879 d->bd_wcount = bd->bd_wcount;
2880 d->bd_wdcount = bd->bd_wdcount;
2881 d->bd_wfcount = bd->bd_wfcount;
2882 d->bd_zcopy = bd->bd_zcopy;
2883 d->bd_bufmode = bd->bd_bufmode;
2887 * Handle `netstat -B' stats request
2890 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2892 static const struct xbpf_d zerostats;
2893 struct xbpf_d *xbdbuf, *xbd, tempstats;
2899 * XXX This is not technically correct. It is possible for non
2900 * privileged users to open bpf devices. It would make sense
2901 * if the users who opened the devices were able to retrieve
2902 * the statistics for them, too.
2904 error = priv_check(req->td, PRIV_NET_BPF);
2908 * Check to see if the user is requesting that the counters be
2909 * zeroed out. Explicitly check that the supplied data is zeroed,
2910 * as we aren't allowing the user to set the counters currently.
2912 if (req->newptr != NULL) {
2913 if (req->newlen != sizeof(tempstats))
2915 memset(&tempstats, 0, sizeof(tempstats));
2916 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2919 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2921 bpf_zero_counters();
2924 if (req->oldptr == NULL)
2925 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2926 if (bpf_bpfd_cnt == 0)
2927 return (SYSCTL_OUT(req, 0, 0));
2928 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2930 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2932 free(xbdbuf, M_BPF);
2936 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2938 /* Send writers-only first */
2939 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2940 xbd = &xbdbuf[index++];
2942 bpfstats_fill_xbpf(xbd, bd);
2945 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2946 xbd = &xbdbuf[index++];
2948 bpfstats_fill_xbpf(xbd, bd);
2954 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2955 free(xbdbuf, M_BPF);
2959 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2961 #else /* !DEV_BPF && !NETGRAPH_BPF */
2963 * NOP stubs to allow bpf-using drivers to load and function.
2965 * A 'better' implementation would allow the core bpf functionality
2966 * to be loaded at runtime.
2968 static struct bpf_if bp_null;
2971 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2976 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2981 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2986 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2989 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2993 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2996 *driverp = &bp_null;
3000 bpfdetach(struct ifnet *ifp)
3005 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3007 return -1; /* "no filter" behaviour */
3011 bpf_validate(const struct bpf_insn *f, int len)
3013 return 0; /* false */
3016 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3020 bpf_show_bpf_if(struct bpf_if *bpf_if)
3025 db_printf("%p:\n", bpf_if);
3026 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3027 /* bif_ext.bif_next */
3028 /* bif_ext.bif_dlist */
3029 BPF_DB_PRINTF("%#x", bif_dlt);
3030 BPF_DB_PRINTF("%u", bif_hdrlen);
3031 BPF_DB_PRINTF("%p", bif_ifp);
3034 BPF_DB_PRINTF("%#x", bif_flags);
3037 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3041 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3045 bpf_show_bpf_if((struct bpf_if *)addr);