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"
43 #include "opt_netgraph.h"
45 #include <sys/types.h>
46 #include <sys/param.h>
48 #include <sys/rwlock.h>
49 #include <sys/systm.h>
51 #include <sys/fcntl.h>
53 #include <sys/malloc.h>
58 #include <sys/signalvar.h>
59 #include <sys/filio.h>
60 #include <sys/sockio.h>
61 #include <sys/ttycom.h>
64 #include <sys/event.h>
69 #include <sys/socket.h>
76 #include <net/if_var.h>
77 #include <net/if_dl.h>
79 #include <net/bpf_buffer.h>
81 #include <net/bpf_jitter.h>
83 #include <net/bpf_zerocopy.h>
84 #include <net/bpfdesc.h>
85 #include <net/route.h>
88 #include <netinet/in.h>
89 #include <netinet/if_ether.h>
90 #include <sys/kernel.h>
91 #include <sys/sysctl.h>
93 #include <net80211/ieee80211_freebsd.h>
95 #include <security/mac/mac_framework.h>
97 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
100 #define bif_next bif_ext.bif_next
101 #define bif_dlist bif_ext.bif_dlist
102 struct bpf_if_ext bif_ext; /* public members */
103 u_int bif_dlt; /* link layer type */
104 u_int bif_hdrlen; /* length of link header */
105 struct ifnet *bif_ifp; /* corresponding interface */
106 struct rwlock bif_lock; /* interface lock */
107 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */
108 int bif_flags; /* Interface flags */
111 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
113 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
115 #define PRINET 26 /* interruptible */
117 #define SIZEOF_BPF_HDR(type) \
118 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
120 #ifdef COMPAT_FREEBSD32
121 #include <sys/mount.h>
122 #include <compat/freebsd32/freebsd32.h>
123 #define BPF_ALIGNMENT32 sizeof(int32_t)
124 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
128 * 32-bit version of structure prepended to each packet. We use this header
129 * instead of the standard one for 32-bit streams. We mark the a stream as
130 * 32-bit the first time we see a 32-bit compat ioctl request.
133 struct timeval32 bh_tstamp; /* time stamp */
134 uint32_t bh_caplen; /* length of captured portion */
135 uint32_t bh_datalen; /* original length of packet */
136 uint16_t bh_hdrlen; /* length of bpf header (this struct
137 plus alignment padding) */
141 struct bpf_program32 {
146 struct bpf_dltlist32 {
151 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
152 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
153 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
154 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
155 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
156 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
160 * bpf_iflist is a list of BPF interface structures, each corresponding to a
161 * specific DLT. The same network interface might have several BPF interface
162 * structures registered by different layers in the stack (i.e., 802.11
163 * frames, ethernet frames, etc).
165 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist;
166 static struct mtx bpf_mtx; /* bpf global lock */
167 static int bpf_bpfd_cnt;
169 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
170 static void bpf_detachd(struct bpf_d *);
171 static void bpf_detachd_locked(struct bpf_d *);
172 static void bpf_freed(struct bpf_d *);
173 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
174 struct sockaddr *, int *, struct bpf_d *);
175 static int bpf_setif(struct bpf_d *, struct ifreq *);
176 static void bpf_timed_out(void *);
178 bpf_wakeup(struct bpf_d *);
179 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
180 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
182 static void reset_d(struct bpf_d *);
183 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
184 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
185 static int bpf_setdlt(struct bpf_d *, u_int);
186 static void filt_bpfdetach(struct knote *);
187 static int filt_bpfread(struct knote *, long);
188 static void bpf_drvinit(void *);
189 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
191 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
192 int bpf_maxinsns = BPF_MAXINSNS;
193 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
194 &bpf_maxinsns, 0, "Maximum bpf program instructions");
195 static int bpf_zerocopy_enable = 0;
196 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
197 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
198 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
199 bpf_stats_sysctl, "bpf statistics portal");
201 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
202 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
203 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(bpf_optimize_writers), 0,
205 "Do not send packets until BPF program is set");
207 static d_open_t bpfopen;
208 static d_read_t bpfread;
209 static d_write_t bpfwrite;
210 static d_ioctl_t bpfioctl;
211 static d_poll_t bpfpoll;
212 static d_kqfilter_t bpfkqfilter;
214 static struct cdevsw bpf_cdevsw = {
215 .d_version = D_VERSION,
222 .d_kqfilter = bpfkqfilter,
225 static struct filterops bpfread_filtops = {
227 .f_detach = filt_bpfdetach,
228 .f_event = filt_bpfread,
231 eventhandler_tag bpf_ifdetach_cookie = NULL;
234 * LOCKING MODEL USED BY BPF:
236 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
237 * some global counters and every bpf_if reference.
238 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
239 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
240 * used by bpf_mtap code.
244 * Global lock, interface lock, descriptor lock
246 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
247 * working model. In many places (like bpf_detachd) we start with BPF descriptor
248 * (and we need to at least rlock it to get reliable interface pointer). This
249 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
250 * change in every such place.
252 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
253 * 3) descriptor main wlock.
254 * Reading bd_bif can be protected by any of these locks, typically global lock.
256 * Changing read/write BPF filter is protected by the same three locks,
257 * the same applies for reading.
259 * Sleeping in global lock is not allowed due to bpfdetach() using it.
263 * Wrapper functions for various buffering methods. If the set of buffer
264 * modes expands, we will probably want to introduce a switch data structure
265 * similar to protosw, et.
268 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
274 switch (d->bd_bufmode) {
275 case BPF_BUFMODE_BUFFER:
276 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
278 case BPF_BUFMODE_ZBUF:
280 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
283 panic("bpf_buf_append_bytes");
288 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
294 switch (d->bd_bufmode) {
295 case BPF_BUFMODE_BUFFER:
296 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
298 case BPF_BUFMODE_ZBUF:
300 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
303 panic("bpf_buf_append_mbuf");
308 * This function gets called when the free buffer is re-assigned.
311 bpf_buf_reclaimed(struct bpf_d *d)
316 switch (d->bd_bufmode) {
317 case BPF_BUFMODE_BUFFER:
320 case BPF_BUFMODE_ZBUF:
321 bpf_zerocopy_buf_reclaimed(d);
325 panic("bpf_buf_reclaimed");
330 * If the buffer mechanism has a way to decide that a held buffer can be made
331 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
332 * returned if the buffer can be discarded, (0) is returned if it cannot.
335 bpf_canfreebuf(struct bpf_d *d)
340 switch (d->bd_bufmode) {
341 case BPF_BUFMODE_ZBUF:
342 return (bpf_zerocopy_canfreebuf(d));
348 * Allow the buffer model to indicate that the current store buffer is
349 * immutable, regardless of the appearance of space. Return (1) if the
350 * buffer is writable, and (0) if not.
353 bpf_canwritebuf(struct bpf_d *d)
357 switch (d->bd_bufmode) {
358 case BPF_BUFMODE_ZBUF:
359 return (bpf_zerocopy_canwritebuf(d));
365 * Notify buffer model that an attempt to write to the store buffer has
366 * resulted in a dropped packet, in which case the buffer may be considered
370 bpf_buffull(struct bpf_d *d)
375 switch (d->bd_bufmode) {
376 case BPF_BUFMODE_ZBUF:
377 bpf_zerocopy_buffull(d);
383 * Notify the buffer model that a buffer has moved into the hold position.
386 bpf_bufheld(struct bpf_d *d)
391 switch (d->bd_bufmode) {
392 case BPF_BUFMODE_ZBUF:
393 bpf_zerocopy_bufheld(d);
399 bpf_free(struct bpf_d *d)
402 switch (d->bd_bufmode) {
403 case BPF_BUFMODE_BUFFER:
404 return (bpf_buffer_free(d));
406 case BPF_BUFMODE_ZBUF:
407 return (bpf_zerocopy_free(d));
410 panic("bpf_buf_free");
415 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
418 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
420 return (bpf_buffer_uiomove(d, buf, len, uio));
424 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
427 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
429 return (bpf_buffer_ioctl_sblen(d, i));
433 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
436 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
438 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
442 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
445 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
447 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
451 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
454 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
456 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
460 * General BPF functions.
463 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
464 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
466 const struct ieee80211_bpf_params *p;
467 struct ether_header *eh;
475 * Build a sockaddr based on the data link layer type.
476 * We do this at this level because the ethernet header
477 * is copied directly into the data field of the sockaddr.
478 * In the case of SLIP, there is no header and the packet
479 * is forwarded as is.
480 * Also, we are careful to leave room at the front of the mbuf
481 * for the link level header.
486 sockp->sa_family = AF_INET;
491 sockp->sa_family = AF_UNSPEC;
492 /* XXX Would MAXLINKHDR be better? */
493 hlen = ETHER_HDR_LEN;
497 sockp->sa_family = AF_IMPLINK;
502 sockp->sa_family = AF_UNSPEC;
508 * null interface types require a 4 byte pseudo header which
509 * corresponds to the address family of the packet.
511 sockp->sa_family = AF_UNSPEC;
515 case DLT_ATM_RFC1483:
517 * en atm driver requires 4-byte atm pseudo header.
518 * though it isn't standard, vpi:vci needs to be
521 sockp->sa_family = AF_UNSPEC;
522 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
526 sockp->sa_family = AF_UNSPEC;
527 hlen = 4; /* This should match PPP_HDRLEN */
530 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
531 sockp->sa_family = AF_IEEE80211;
535 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
536 sockp->sa_family = AF_IEEE80211;
537 sockp->sa_len = 12; /* XXX != 0 */
538 hlen = sizeof(struct ieee80211_bpf_params);
545 len = uio->uio_resid;
546 if (len < hlen || len - hlen > ifp->if_mtu)
549 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
552 m->m_pkthdr.len = m->m_len = len;
555 error = uiomove(mtod(m, u_char *), len, uio);
559 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
565 /* Check for multicast destination */
568 eh = mtod(m, struct ether_header *);
569 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
570 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
571 ETHER_ADDR_LEN) == 0)
572 m->m_flags |= M_BCAST;
574 m->m_flags |= M_MCAST;
576 if (d->bd_hdrcmplt == 0) {
577 memcpy(eh->ether_shost, IF_LLADDR(ifp),
578 sizeof(eh->ether_shost));
584 * Make room for link header, and copy it to sockaddr
587 if (sockp->sa_family == AF_IEEE80211) {
589 * Collect true length from the parameter header
590 * NB: sockp is known to be zero'd so if we do a
591 * short copy unspecified parameters will be
593 * NB: packet may not be aligned after stripping
597 p = mtod(m, const struct ieee80211_bpf_params *);
599 if (hlen > sizeof(sockp->sa_data)) {
604 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
615 * Attach file to the bpf interface, i.e. make d listen on bp.
618 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
625 * Save sysctl value to protect from sysctl change
628 op_w = V_bpf_optimize_writers || d->bd_writer;
630 if (d->bd_bif != NULL)
631 bpf_detachd_locked(d);
633 * Point d at bp, and add d to the interface's list.
634 * Since there are many applications using BPF for
635 * sending raw packets only (dhcpd, cdpd are good examples)
636 * we can delay adding d to the list of active listeners until
637 * some filter is configured.
646 /* Add to writers-only list */
647 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
649 * We decrement bd_writer on every filter set operation.
650 * First BIOCSETF is done by pcap_open_live() to set up
651 * snap length. After that appliation usually sets its own filter
655 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
662 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
663 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
666 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
670 * Check if we need to upgrade our descriptor @d from write-only mode.
673 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
675 int is_snap, need_upgrade;
678 * Check if we've already upgraded or new filter is empty.
680 if (d->bd_writer == 0 || fcode == NULL)
686 * Check if cmd looks like snaplen setting from
687 * pcap_bpf.c:pcap_open_live().
688 * Note we're not checking .k value here:
689 * while pcap_open_live() definitely sets to to non-zero value,
690 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
691 * do not consider upgrading immediately
693 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
700 * We're setting first filter and it doesn't look like
701 * setting snaplen. We're probably using bpf directly.
702 * Upgrade immediately.
707 * Do not require upgrade by first BIOCSETF
708 * (used to set snaplen) by pcap_open_live().
711 if (--d->bd_writer == 0) {
713 * First snaplen filter has already
714 * been set. This is probably catch-all
722 "%s: filter function set by pid %d, "
723 "bd_writer counter %d, snap %d upgrade %d",
724 __func__, d->bd_pid, d->bd_writer,
725 is_snap, need_upgrade);
727 return (need_upgrade);
731 * Add d to the list of active bp filters.
732 * Requires bpf_attachd() to be called before.
735 bpf_upgraded(struct bpf_d *d)
744 * Filter can be set several times without specifying interface.
745 * Mark d as reader and exit.
757 /* Remove from writers-only list */
758 LIST_REMOVE(d, bd_next);
759 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
760 /* Mark d as reader */
766 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
768 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
772 * Detach a file from its interface.
775 bpf_detachd(struct bpf_d *d)
778 bpf_detachd_locked(d);
783 bpf_detachd_locked(struct bpf_d *d)
789 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
793 /* Check if descriptor is attached */
794 if ((bp = d->bd_bif) == NULL)
800 /* Save bd_writer value */
801 error = d->bd_writer;
804 * Remove d from the interface's descriptor list.
806 LIST_REMOVE(d, bd_next);
815 /* Call event handler iff d is attached */
817 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
820 * Check if this descriptor had requested promiscuous mode.
821 * If so, turn it off.
825 CURVNET_SET(ifp->if_vnet);
826 error = ifpromisc(ifp, 0);
828 if (error != 0 && error != ENXIO) {
830 * ENXIO can happen if a pccard is unplugged
831 * Something is really wrong if we were able to put
832 * the driver into promiscuous mode, but can't
835 if_printf(bp->bif_ifp,
836 "bpf_detach: ifpromisc failed (%d)\n", error);
842 * Close the descriptor by detaching it from its interface,
843 * deallocating its buffers, and marking it free.
848 struct bpf_d *d = data;
851 if (d->bd_state == BPF_WAITING)
852 callout_stop(&d->bd_callout);
853 d->bd_state = BPF_IDLE;
855 funsetown(&d->bd_sigio);
858 mac_bpfdesc_destroy(d);
860 seldrain(&d->bd_sel);
861 knlist_destroy(&d->bd_sel.si_note);
862 callout_drain(&d->bd_callout);
868 * Open ethernet device. Returns ENXIO for illegal minor device number,
869 * EBUSY if file is open by another process.
873 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
878 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
879 error = devfs_set_cdevpriv(d, bpf_dtor);
886 * For historical reasons, perform a one-time initialization call to
887 * the buffer routines, even though we're not yet committed to a
888 * particular buffer method.
891 if ((flags & FREAD) == 0)
893 d->bd_hbuf_in_use = 0;
894 d->bd_bufmode = BPF_BUFMODE_BUFFER;
896 d->bd_direction = BPF_D_INOUT;
897 BPF_PID_REFRESH(d, td);
900 mac_bpfdesc_create(td->td_ucred, d);
902 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
903 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
904 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
910 * bpfread - read next chunk of packets from buffers
913 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
920 error = devfs_get_cdevpriv((void **)&d);
925 * Restrict application to use a buffer the same size as
928 if (uio->uio_resid != d->bd_bufsize)
931 non_block = ((ioflag & O_NONBLOCK) != 0);
934 BPF_PID_REFRESH_CUR(d);
935 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
939 if (d->bd_state == BPF_WAITING)
940 callout_stop(&d->bd_callout);
941 timed_out = (d->bd_state == BPF_TIMED_OUT);
942 d->bd_state = BPF_IDLE;
943 while (d->bd_hbuf_in_use) {
944 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
945 PRINET|PCATCH, "bd_hbuf", 0);
952 * If the hold buffer is empty, then do a timed sleep, which
953 * ends when the timeout expires or when enough packets
954 * have arrived to fill the store buffer.
956 while (d->bd_hbuf == NULL) {
957 if (d->bd_slen != 0) {
959 * A packet(s) either arrived since the previous
960 * read or arrived while we were asleep.
962 if (d->bd_immediate || non_block || timed_out) {
964 * Rotate the buffers and return what's here
965 * if we are in immediate mode, non-blocking
966 * flag is set, or this descriptor timed out.
974 * No data is available, check to see if the bpf device
975 * is still pointed at a real interface. If not, return
976 * ENXIO so that the userland process knows to rebind
977 * it before using it again.
979 if (d->bd_bif == NULL) {
986 return (EWOULDBLOCK);
988 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
990 if (error == EINTR || error == ERESTART) {
994 if (error == EWOULDBLOCK) {
996 * On a timeout, return what's in the buffer,
997 * which may be nothing. If there is something
998 * in the store buffer, we can rotate the buffers.
1002 * We filled up the buffer in between
1003 * getting the timeout and arriving
1004 * here, so we don't need to rotate.
1008 if (d->bd_slen == 0) {
1017 * At this point, we know we have something in the hold slot.
1019 d->bd_hbuf_in_use = 1;
1023 * Move data from hold buffer into user space.
1024 * We know the entire buffer is transferred since
1025 * we checked above that the read buffer is bpf_bufsize bytes.
1027 * We do not have to worry about simultaneous reads because
1028 * we waited for sole access to the hold buffer above.
1030 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1033 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1034 d->bd_fbuf = d->bd_hbuf;
1037 bpf_buf_reclaimed(d);
1038 d->bd_hbuf_in_use = 0;
1039 wakeup(&d->bd_hbuf_in_use);
1046 * If there are processes sleeping on this descriptor, wake them up.
1048 static __inline void
1049 bpf_wakeup(struct bpf_d *d)
1052 BPFD_LOCK_ASSERT(d);
1053 if (d->bd_state == BPF_WAITING) {
1054 callout_stop(&d->bd_callout);
1055 d->bd_state = BPF_IDLE;
1058 if (d->bd_async && d->bd_sig && d->bd_sigio)
1059 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1061 selwakeuppri(&d->bd_sel, PRINET);
1062 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1066 bpf_timed_out(void *arg)
1068 struct bpf_d *d = (struct bpf_d *)arg;
1070 BPFD_LOCK_ASSERT(d);
1072 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1074 if (d->bd_state == BPF_WAITING) {
1075 d->bd_state = BPF_TIMED_OUT;
1076 if (d->bd_slen != 0)
1082 bpf_ready(struct bpf_d *d)
1085 BPFD_LOCK_ASSERT(d);
1087 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1089 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1096 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1100 struct mbuf *m, *mc;
1101 struct sockaddr dst;
1105 error = devfs_get_cdevpriv((void **)&d);
1109 BPF_PID_REFRESH_CUR(d);
1111 /* XXX: locking required */
1112 if (d->bd_bif == NULL) {
1117 ifp = d->bd_bif->bif_ifp;
1119 if ((ifp->if_flags & IFF_UP) == 0) {
1124 if (uio->uio_resid == 0) {
1129 bzero(&dst, sizeof(dst));
1132 /* XXX: bpf_movein() can sleep */
1133 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1134 &m, &dst, &hlen, d);
1141 dst.sa_family = pseudo_AF_HDRCMPLT;
1143 if (d->bd_feedback) {
1144 mc = m_dup(m, M_NOWAIT);
1146 mc->m_pkthdr.rcvif = ifp;
1147 /* Set M_PROMISC for outgoing packets to be discarded. */
1148 if (d->bd_direction == BPF_D_INOUT)
1149 m->m_flags |= M_PROMISC;
1153 m->m_pkthdr.len -= hlen;
1155 m->m_data += hlen; /* XXX */
1157 CURVNET_SET(ifp->if_vnet);
1160 mac_bpfdesc_create_mbuf(d, m);
1162 mac_bpfdesc_create_mbuf(d, mc);
1166 bzero(&ro, sizeof(ro));
1168 ro.ro_prepend = (u_char *)&dst.sa_data;
1170 ro.ro_flags = RT_HAS_HEADER;
1173 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1179 (*ifp->if_input)(ifp, mc);
1189 * Reset a descriptor by flushing its packet buffer and clearing the receive
1190 * and drop counts. This is doable for kernel-only buffers, but with
1191 * zero-copy buffers, we can't write to (or rotate) buffers that are
1192 * currently owned by userspace. It would be nice if we could encapsulate
1193 * this logic in the buffer code rather than here.
1196 reset_d(struct bpf_d *d)
1199 BPFD_LOCK_ASSERT(d);
1201 while (d->bd_hbuf_in_use)
1202 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1204 if ((d->bd_hbuf != NULL) &&
1205 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1206 /* Free the hold buffer. */
1207 d->bd_fbuf = d->bd_hbuf;
1210 bpf_buf_reclaimed(d);
1212 if (bpf_canwritebuf(d))
1224 * FIONREAD Check for read packet available.
1225 * BIOCGBLEN Get buffer len [for read()].
1226 * BIOCSETF Set read filter.
1227 * BIOCSETFNR Set read filter without resetting descriptor.
1228 * BIOCSETWF Set write filter.
1229 * BIOCFLUSH Flush read packet buffer.
1230 * BIOCPROMISC Put interface into promiscuous mode.
1231 * BIOCGDLT Get link layer type.
1232 * BIOCGETIF Get interface name.
1233 * BIOCSETIF Set interface.
1234 * BIOCSRTIMEOUT Set read timeout.
1235 * BIOCGRTIMEOUT Get read timeout.
1236 * BIOCGSTATS Get packet stats.
1237 * BIOCIMMEDIATE Set immediate mode.
1238 * BIOCVERSION Get filter language version.
1239 * BIOCGHDRCMPLT Get "header already complete" flag
1240 * BIOCSHDRCMPLT Set "header already complete" flag
1241 * BIOCGDIRECTION Get packet direction flag
1242 * BIOCSDIRECTION Set packet direction flag
1243 * BIOCGTSTAMP Get time stamp format and resolution.
1244 * BIOCSTSTAMP Set time stamp format and resolution.
1245 * BIOCLOCK Set "locked" flag
1246 * BIOCFEEDBACK Set packet feedback mode.
1247 * BIOCSETZBUF Set current zero-copy buffer locations.
1248 * BIOCGETZMAX Get maximum zero-copy buffer size.
1249 * BIOCROTZBUF Force rotation of zero-copy buffer
1250 * BIOCSETBUFMODE Set buffer mode.
1251 * BIOCGETBUFMODE Get current buffer mode.
1255 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1261 error = devfs_get_cdevpriv((void **)&d);
1266 * Refresh PID associated with this descriptor.
1269 BPF_PID_REFRESH(d, td);
1270 if (d->bd_state == BPF_WAITING)
1271 callout_stop(&d->bd_callout);
1272 d->bd_state = BPF_IDLE;
1275 if (d->bd_locked == 1) {
1281 #ifdef COMPAT_FREEBSD32
1282 case BIOCGDLTLIST32:
1286 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1287 case BIOCGRTIMEOUT32:
1298 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1299 case BIOCSRTIMEOUT32:
1309 #ifdef COMPAT_FREEBSD32
1311 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1312 * that it will get 32-bit packet headers.
1318 case BIOCGDLTLIST32:
1319 case BIOCGRTIMEOUT32:
1320 case BIOCSRTIMEOUT32:
1327 CURVNET_SET(TD_TO_VNET(td));
1335 * Check for read packet available.
1343 while (d->bd_hbuf_in_use)
1344 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1345 PRINET, "bd_hbuf", 0);
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 int alloc_buf, size;
1494 * Behavior here depends on the buffering model. If
1495 * we're using kernel memory buffers, then we can
1496 * allocate them here. If we're using zero-copy,
1497 * then the user process must have registered buffers
1498 * by the time we get here.
1502 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1507 size = d->bd_bufsize;
1508 error = bpf_buffer_ioctl_sblen(d, &size);
1513 error = bpf_setif(d, (struct ifreq *)addr);
1522 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1523 case BIOCSRTIMEOUT32:
1526 struct timeval *tv = (struct timeval *)addr;
1527 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1528 struct timeval32 *tv32;
1529 struct timeval tv64;
1531 if (cmd == BIOCSRTIMEOUT32) {
1532 tv32 = (struct timeval32 *)addr;
1534 tv->tv_sec = tv32->tv_sec;
1535 tv->tv_usec = tv32->tv_usec;
1538 tv = (struct timeval *)addr;
1541 * Subtract 1 tick from tvtohz() since this isn't
1544 if ((error = itimerfix(tv)) == 0)
1545 d->bd_rtout = tvtohz(tv) - 1;
1553 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1554 case BIOCGRTIMEOUT32:
1558 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1559 struct timeval32 *tv32;
1560 struct timeval tv64;
1562 if (cmd == BIOCGRTIMEOUT32)
1566 tv = (struct timeval *)addr;
1568 tv->tv_sec = d->bd_rtout / hz;
1569 tv->tv_usec = (d->bd_rtout % hz) * tick;
1570 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1571 if (cmd == BIOCGRTIMEOUT32) {
1572 tv32 = (struct timeval32 *)addr;
1573 tv32->tv_sec = tv->tv_sec;
1574 tv32->tv_usec = tv->tv_usec;
1586 struct bpf_stat *bs = (struct bpf_stat *)addr;
1588 /* XXXCSJP overflow */
1589 bs->bs_recv = d->bd_rcount;
1590 bs->bs_drop = d->bd_dcount;
1595 * Set immediate mode.
1599 d->bd_immediate = *(u_int *)addr;
1605 struct bpf_version *bv = (struct bpf_version *)addr;
1607 bv->bv_major = BPF_MAJOR_VERSION;
1608 bv->bv_minor = BPF_MINOR_VERSION;
1613 * Get "header already complete" flag
1617 *(u_int *)addr = d->bd_hdrcmplt;
1622 * Set "header already complete" flag
1626 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1631 * Get packet direction flag
1633 case BIOCGDIRECTION:
1635 *(u_int *)addr = d->bd_direction;
1640 * Set packet direction flag
1642 case BIOCSDIRECTION:
1646 direction = *(u_int *)addr;
1647 switch (direction) {
1652 d->bd_direction = direction;
1662 * Get packet timestamp format and resolution.
1666 *(u_int *)addr = d->bd_tstamp;
1671 * Set packet timestamp format and resolution.
1677 func = *(u_int *)addr;
1678 if (BPF_T_VALID(func))
1679 d->bd_tstamp = func;
1687 d->bd_feedback = *(u_int *)addr;
1697 case FIONBIO: /* Non-blocking I/O */
1700 case FIOASYNC: /* Send signal on receive packets */
1702 d->bd_async = *(int *)addr;
1708 * XXX: Add some sort of locking here?
1709 * fsetown() can sleep.
1711 error = fsetown(*(int *)addr, &d->bd_sigio);
1716 *(int *)addr = fgetown(&d->bd_sigio);
1720 /* This is deprecated, FIOSETOWN should be used instead. */
1722 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1725 /* This is deprecated, FIOGETOWN should be used instead. */
1727 *(int *)addr = -fgetown(&d->bd_sigio);
1730 case BIOCSRSIG: /* Set receive signal */
1734 sig = *(u_int *)addr;
1747 *(u_int *)addr = d->bd_sig;
1751 case BIOCGETBUFMODE:
1753 *(u_int *)addr = d->bd_bufmode;
1757 case BIOCSETBUFMODE:
1759 * Allow the buffering mode to be changed as long as we
1760 * haven't yet committed to a particular mode. Our
1761 * definition of commitment, for now, is whether or not a
1762 * buffer has been allocated or an interface attached, since
1763 * that's the point where things get tricky.
1765 switch (*(u_int *)addr) {
1766 case BPF_BUFMODE_BUFFER:
1769 case BPF_BUFMODE_ZBUF:
1770 if (bpf_zerocopy_enable)
1780 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1781 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1786 d->bd_bufmode = *(u_int *)addr;
1791 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1795 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1799 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1807 * Set d's packet filter program to fp. If this file already has a filter,
1808 * free it and replace it. Returns EINVAL for bogus requests.
1810 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1811 * since reading d->bd_bif can't be protected by d or interface lock due to
1814 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1815 * interface read lock to read all filers.
1819 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1821 #ifdef COMPAT_FREEBSD32
1822 struct bpf_program fp_swab;
1823 struct bpf_program32 *fp32;
1825 struct bpf_insn *fcode, *old;
1827 bpf_jit_filter *jfunc, *ofunc;
1833 #ifdef COMPAT_FREEBSD32
1838 fp32 = (struct bpf_program32 *)fp;
1839 fp_swab.bf_len = fp32->bf_len;
1840 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1856 jfunc = ofunc = NULL;
1861 * Check new filter validness before acquiring any locks.
1862 * Allocate memory for new filter, if needed.
1865 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1867 size = flen * sizeof(*fp->bf_insns);
1869 /* We're setting up new filter. Copy and check actual data. */
1870 fcode = malloc(size, M_BPF, M_WAITOK);
1871 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1872 !bpf_validate(fcode, flen)) {
1877 /* Filter is copied inside fcode and is perfectly valid. */
1878 jfunc = bpf_jitter(fcode, flen);
1885 * Set up new filter.
1886 * Protect filter change by interface lock.
1887 * Additionally, we are protected by global lock here.
1889 if (d->bd_bif != NULL)
1890 BPFIF_WLOCK(d->bd_bif);
1892 if (cmd == BIOCSETWF) {
1893 old = d->bd_wfilter;
1894 d->bd_wfilter = fcode;
1896 old = d->bd_rfilter;
1897 d->bd_rfilter = fcode;
1899 ofunc = d->bd_bfilter;
1900 d->bd_bfilter = jfunc;
1902 if (cmd == BIOCSETF)
1905 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1908 if (d->bd_bif != NULL)
1909 BPFIF_WUNLOCK(d->bd_bif);
1914 bpf_destroy_jit_filter(ofunc);
1917 /* Move d to active readers list. */
1918 if (need_upgrade != 0)
1926 * Detach a file from its current interface (if attached at all) and attach
1927 * to the interface indicated by the name stored in ifr.
1928 * Return an errno or 0.
1931 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1934 struct ifnet *theywant;
1938 theywant = ifunit(ifr->ifr_name);
1939 if (theywant == NULL || theywant->if_bpf == NULL)
1942 bp = theywant->if_bpf;
1944 /* Check if interface is not being detached from BPF */
1946 if (bp->bif_flags & BPFIF_FLAG_DYING) {
1953 * At this point, we expect the buffer is already allocated. If not,
1956 switch (d->bd_bufmode) {
1957 case BPF_BUFMODE_BUFFER:
1958 case BPF_BUFMODE_ZBUF:
1959 if (d->bd_sbuf == NULL)
1964 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1966 if (bp != d->bd_bif)
1975 * Support for select() and poll() system calls
1977 * Return true iff the specific operation will not block indefinitely.
1978 * Otherwise, return false but make a note that a selwakeup() must be done.
1981 bpfpoll(struct cdev *dev, int events, struct thread *td)
1986 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1988 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1991 * Refresh PID associated with this descriptor.
1993 revents = events & (POLLOUT | POLLWRNORM);
1995 BPF_PID_REFRESH(d, td);
1996 if (events & (POLLIN | POLLRDNORM)) {
1998 revents |= events & (POLLIN | POLLRDNORM);
2000 selrecord(td, &d->bd_sel);
2001 /* Start the read timeout if necessary. */
2002 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2003 callout_reset(&d->bd_callout, d->bd_rtout,
2005 d->bd_state = BPF_WAITING;
2014 * Support for kevent() system call. Register EVFILT_READ filters and
2015 * reject all others.
2018 bpfkqfilter(struct cdev *dev, struct knote *kn)
2022 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2023 kn->kn_filter != EVFILT_READ)
2027 * Refresh PID associated with this descriptor.
2030 BPF_PID_REFRESH_CUR(d);
2031 kn->kn_fop = &bpfread_filtops;
2033 knlist_add(&d->bd_sel.si_note, kn, 1);
2040 filt_bpfdetach(struct knote *kn)
2042 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2044 knlist_remove(&d->bd_sel.si_note, kn, 0);
2048 filt_bpfread(struct knote *kn, long hint)
2050 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2053 BPFD_LOCK_ASSERT(d);
2054 ready = bpf_ready(d);
2056 kn->kn_data = d->bd_slen;
2058 * Ignore the hold buffer if it is being copied to user space.
2060 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2061 kn->kn_data += d->bd_hlen;
2062 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2063 callout_reset(&d->bd_callout, d->bd_rtout,
2065 d->bd_state = BPF_WAITING;
2071 #define BPF_TSTAMP_NONE 0
2072 #define BPF_TSTAMP_FAST 1
2073 #define BPF_TSTAMP_NORMAL 2
2074 #define BPF_TSTAMP_EXTERN 3
2077 bpf_ts_quality(int tstype)
2080 if (tstype == BPF_T_NONE)
2081 return (BPF_TSTAMP_NONE);
2082 if ((tstype & BPF_T_FAST) != 0)
2083 return (BPF_TSTAMP_FAST);
2085 return (BPF_TSTAMP_NORMAL);
2089 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2094 quality = bpf_ts_quality(tstype);
2095 if (quality == BPF_TSTAMP_NONE)
2099 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2101 *bt = *(struct bintime *)(tag + 1);
2102 return (BPF_TSTAMP_EXTERN);
2105 if (quality == BPF_TSTAMP_NORMAL)
2114 * Incoming linkage from device drivers. Process the packet pkt, of length
2115 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2116 * by each process' filter, and if accepted, stashed into the corresponding
2120 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2130 gottime = BPF_TSTAMP_NONE;
2134 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2136 * We are not using any locks for d here because:
2137 * 1) any filter change is protected by interface
2139 * 2) destroying/detaching d is protected by interface
2143 /* XXX: Do not protect counter for the sake of performance. */
2146 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2147 * way for the caller to indiciate to us whether this packet
2148 * is inbound or outbound. In the bpf_mtap() routines, we use
2149 * the interface pointers on the mbuf to figure it out.
2152 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2154 slen = (*(bf->func))(pkt, pktlen, pktlen);
2157 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2160 * Filter matches. Let's to acquire write lock.
2165 if (gottime < bpf_ts_quality(d->bd_tstamp))
2166 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2168 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2170 catchpacket(d, pkt, pktlen, slen,
2171 bpf_append_bytes, &bt);
2178 #define BPF_CHECK_DIRECTION(d, r, i) \
2179 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2180 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2183 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2184 * Locking model is explained in bpf_tap().
2187 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2197 /* Skip outgoing duplicate packets. */
2198 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2199 m->m_flags &= ~M_PROMISC;
2203 pktlen = m_length(m, NULL);
2204 gottime = BPF_TSTAMP_NONE;
2208 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2209 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2213 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2214 /* XXX We cannot handle multiple mbufs. */
2215 if (bf != NULL && m->m_next == NULL)
2216 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2219 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2224 if (gottime < bpf_ts_quality(d->bd_tstamp))
2225 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2227 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2229 catchpacket(d, (u_char *)m, pktlen, slen,
2230 bpf_append_mbuf, &bt);
2238 * Incoming linkage from device drivers, when packet is in
2239 * an mbuf chain and to be prepended by a contiguous header.
2242 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2250 /* Skip outgoing duplicate packets. */
2251 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2252 m->m_flags &= ~M_PROMISC;
2256 pktlen = m_length(m, NULL);
2258 * Craft on-stack mbuf suitable for passing to bpf_filter.
2259 * Note that we cut corners here; we only setup what's
2260 * absolutely needed--this mbuf should never go anywhere else.
2267 gottime = BPF_TSTAMP_NONE;
2271 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2272 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2275 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2280 if (gottime < bpf_ts_quality(d->bd_tstamp))
2281 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2283 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2285 catchpacket(d, (u_char *)&mb, pktlen, slen,
2286 bpf_append_mbuf, &bt);
2293 #undef BPF_CHECK_DIRECTION
2295 #undef BPF_TSTAMP_NONE
2296 #undef BPF_TSTAMP_FAST
2297 #undef BPF_TSTAMP_NORMAL
2298 #undef BPF_TSTAMP_EXTERN
2301 bpf_hdrlen(struct bpf_d *d)
2305 hdrlen = d->bd_bif->bif_hdrlen;
2306 #ifndef BURN_BRIDGES
2307 if (d->bd_tstamp == BPF_T_NONE ||
2308 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2309 #ifdef COMPAT_FREEBSD32
2311 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2314 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2317 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2318 #ifdef COMPAT_FREEBSD32
2320 hdrlen = BPF_WORDALIGN32(hdrlen);
2323 hdrlen = BPF_WORDALIGN(hdrlen);
2325 return (hdrlen - d->bd_bif->bif_hdrlen);
2329 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2331 struct bintime bt2, boottimebin;
2333 struct timespec tsn;
2335 if ((tstype & BPF_T_MONOTONIC) == 0) {
2337 getboottimebin(&boottimebin);
2338 bintime_add(&bt2, &boottimebin);
2341 switch (BPF_T_FORMAT(tstype)) {
2342 case BPF_T_MICROTIME:
2343 bintime2timeval(bt, &tsm);
2344 ts->bt_sec = tsm.tv_sec;
2345 ts->bt_frac = tsm.tv_usec;
2347 case BPF_T_NANOTIME:
2348 bintime2timespec(bt, &tsn);
2349 ts->bt_sec = tsn.tv_sec;
2350 ts->bt_frac = tsn.tv_nsec;
2353 ts->bt_sec = bt->sec;
2354 ts->bt_frac = bt->frac;
2360 * Move the packet data from interface memory (pkt) into the
2361 * store buffer. "cpfn" is the routine called to do the actual data
2362 * transfer. bcopy is passed in to copy contiguous chunks, while
2363 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2364 * pkt is really an mbuf.
2367 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2368 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2371 struct bpf_xhdr hdr;
2372 #ifndef BURN_BRIDGES
2373 struct bpf_hdr hdr_old;
2374 #ifdef COMPAT_FREEBSD32
2375 struct bpf_hdr32 hdr32_old;
2378 int caplen, curlen, hdrlen, totlen;
2383 BPFD_LOCK_ASSERT(d);
2386 * Detect whether user space has released a buffer back to us, and if
2387 * so, move it from being a hold buffer to a free buffer. This may
2388 * not be the best place to do it (for example, we might only want to
2389 * run this check if we need the space), but for now it's a reliable
2392 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2393 d->bd_fbuf = d->bd_hbuf;
2396 bpf_buf_reclaimed(d);
2400 * Figure out how many bytes to move. If the packet is
2401 * greater or equal to the snapshot length, transfer that
2402 * much. Otherwise, transfer the whole packet (unless
2403 * we hit the buffer size limit).
2405 hdrlen = bpf_hdrlen(d);
2406 totlen = hdrlen + min(snaplen, pktlen);
2407 if (totlen > d->bd_bufsize)
2408 totlen = d->bd_bufsize;
2411 * Round up the end of the previous packet to the next longword.
2413 * Drop the packet if there's no room and no hope of room
2414 * If the packet would overflow the storage buffer or the storage
2415 * buffer is considered immutable by the buffer model, try to rotate
2416 * the buffer and wakeup pending processes.
2418 #ifdef COMPAT_FREEBSD32
2420 curlen = BPF_WORDALIGN32(d->bd_slen);
2423 curlen = BPF_WORDALIGN(d->bd_slen);
2424 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2425 if (d->bd_fbuf == NULL) {
2427 * There's no room in the store buffer, and no
2428 * prospect of room, so drop the packet. Notify the
2435 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2439 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2441 * Immediate mode is set, or the read timeout has already
2442 * expired during a select call. A packet arrived, so the
2443 * reader should be woken up.
2446 caplen = totlen - hdrlen;
2447 tstype = d->bd_tstamp;
2448 do_timestamp = tstype != BPF_T_NONE;
2449 #ifndef BURN_BRIDGES
2450 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2453 bpf_bintime2ts(bt, &ts, tstype);
2454 #ifdef COMPAT_FREEBSD32
2455 if (d->bd_compat32) {
2456 bzero(&hdr32_old, sizeof(hdr32_old));
2458 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2459 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2461 hdr32_old.bh_datalen = pktlen;
2462 hdr32_old.bh_hdrlen = hdrlen;
2463 hdr32_old.bh_caplen = caplen;
2464 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2469 bzero(&hdr_old, sizeof(hdr_old));
2471 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2472 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2474 hdr_old.bh_datalen = pktlen;
2475 hdr_old.bh_hdrlen = hdrlen;
2476 hdr_old.bh_caplen = caplen;
2477 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2484 * Append the bpf header. Note we append the actual header size, but
2485 * move forward the length of the header plus padding.
2487 bzero(&hdr, sizeof(hdr));
2489 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2490 hdr.bh_datalen = pktlen;
2491 hdr.bh_hdrlen = hdrlen;
2492 hdr.bh_caplen = caplen;
2493 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2496 * Copy the packet data into the store buffer and update its length.
2498 #ifndef BURN_BRIDGES
2501 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2502 d->bd_slen = curlen + totlen;
2509 * Free buffers currently in use by a descriptor.
2513 bpf_freed(struct bpf_d *d)
2517 * We don't need to lock out interrupts since this descriptor has
2518 * been detached from its interface and it yet hasn't been marked
2522 if (d->bd_rfilter != NULL) {
2523 free((caddr_t)d->bd_rfilter, M_BPF);
2525 if (d->bd_bfilter != NULL)
2526 bpf_destroy_jit_filter(d->bd_bfilter);
2529 if (d->bd_wfilter != NULL)
2530 free((caddr_t)d->bd_wfilter, M_BPF);
2531 mtx_destroy(&d->bd_lock);
2535 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2536 * fixed size of the link header (variable length headers not yet supported).
2539 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2542 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2546 * Attach an interface to bpf. ifp is a pointer to the structure
2547 * defining the interface to be attached, dlt is the link layer type,
2548 * and hdrlen is the fixed size of the link header (variable length
2549 * headers are not yet supporrted).
2552 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2556 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2560 LIST_INIT(&bp->bif_dlist);
2561 LIST_INIT(&bp->bif_wlist);
2564 rw_init(&bp->bif_lock, "bpf interface lock");
2565 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2569 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2572 bp->bif_hdrlen = hdrlen;
2574 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2575 if_printf(ifp, "bpf attached\n");
2580 * When moving interfaces between vnet instances we need a way to
2581 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2582 * after the vmove. We unfortunately have no device driver infrastructure
2583 * to query the interface for these values after creation/attach, thus
2584 * add this as a workaround.
2587 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2592 if (bif_dlt == NULL && bif_hdrlen == NULL)
2595 if (bif_dlt != NULL)
2596 *bif_dlt = bp->bif_dlt;
2597 if (bif_hdrlen != NULL)
2598 *bif_hdrlen = bp->bif_hdrlen;
2605 * Detach bpf from an interface. This involves detaching each descriptor
2606 * associated with the interface. Notify each descriptor as it's detached
2607 * so that any sleepers wake up and get ENXIO.
2610 bpfdetach(struct ifnet *ifp)
2612 struct bpf_if *bp, *bp_temp;
2619 /* Find all bpf_if struct's which reference ifp and detach them. */
2620 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2621 if (ifp != bp->bif_ifp)
2624 LIST_REMOVE(bp, bif_next);
2625 /* Add to to-be-freed list */
2626 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2630 * Delay freeing bp till interface is detached
2631 * and all routes through this interface are removed.
2632 * Mark bp as detached to restrict new consumers.
2635 bp->bif_flags |= BPFIF_FLAG_DYING;
2638 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2639 __func__, bp->bif_dlt, bp, ifp);
2641 /* Free common descriptors */
2642 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2643 bpf_detachd_locked(d);
2649 /* Free writer-only descriptors */
2650 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2651 bpf_detachd_locked(d);
2661 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2666 * Interface departure handler.
2667 * Note departure event does not guarantee interface is going down.
2668 * Interface renaming is currently done via departure/arrival event set.
2670 * Departure handled is called after all routes pointing to
2671 * given interface are removed and interface is in down state
2672 * restricting any packets to be sent/received. We assume it is now safe
2673 * to free data allocated by BPF.
2676 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2678 struct bpf_if *bp, *bp_temp;
2683 * Find matching entries in free list.
2684 * Nothing should be found if bpfdetach() was not called.
2686 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2687 if (ifp != bp->bif_ifp)
2690 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2693 LIST_REMOVE(bp, bif_next);
2695 rw_destroy(&bp->bif_lock);
2703 * Note that we cannot zero other pointers to
2704 * custom DLTs possibly used by given interface.
2711 * Get a list of available data link type of the interface.
2714 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2723 ifp = d->bd_bif->bif_ifp;
2726 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2727 if (bp->bif_ifp == ifp)
2730 if (bfl->bfl_list == NULL) {
2734 if (n1 > bfl->bfl_len)
2737 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2740 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2741 if (bp->bif_ifp != ifp)
2747 lst[n] = bp->bif_dlt;
2751 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2759 * Set the data link type of a BPF instance.
2762 bpf_setdlt(struct bpf_d *d, u_int dlt)
2764 int error, opromisc;
2770 if (d->bd_bif->bif_dlt == dlt)
2772 ifp = d->bd_bif->bif_ifp;
2774 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2775 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2780 opromisc = d->bd_promisc;
2786 error = ifpromisc(bp->bif_ifp, 1);
2788 if_printf(bp->bif_ifp,
2789 "bpf_setdlt: ifpromisc failed (%d)\n",
2795 return (bp == NULL ? EINVAL : 0);
2799 bpf_drvinit(void *unused)
2803 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF);
2804 LIST_INIT(&bpf_iflist);
2805 LIST_INIT(&bpf_freelist);
2807 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2808 /* For compatibility */
2809 make_dev_alias(dev, "bpf0");
2811 /* Register interface departure handler */
2812 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2813 ifnet_departure_event, bpf_ifdetach, NULL,
2814 EVENTHANDLER_PRI_ANY);
2818 * Zero out the various packet counters associated with all of the bpf
2819 * descriptors. At some point, we will probably want to get a bit more
2820 * granular and allow the user to specify descriptors to be zeroed.
2823 bpf_zero_counters(void)
2829 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2831 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2847 * Fill filter statistics
2850 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2853 bzero(d, sizeof(*d));
2854 BPFD_LOCK_ASSERT(bd);
2855 d->bd_structsize = sizeof(*d);
2856 /* XXX: reading should be protected by global lock */
2857 d->bd_immediate = bd->bd_immediate;
2858 d->bd_promisc = bd->bd_promisc;
2859 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2860 d->bd_direction = bd->bd_direction;
2861 d->bd_feedback = bd->bd_feedback;
2862 d->bd_async = bd->bd_async;
2863 d->bd_rcount = bd->bd_rcount;
2864 d->bd_dcount = bd->bd_dcount;
2865 d->bd_fcount = bd->bd_fcount;
2866 d->bd_sig = bd->bd_sig;
2867 d->bd_slen = bd->bd_slen;
2868 d->bd_hlen = bd->bd_hlen;
2869 d->bd_bufsize = bd->bd_bufsize;
2870 d->bd_pid = bd->bd_pid;
2871 strlcpy(d->bd_ifname,
2872 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2873 d->bd_locked = bd->bd_locked;
2874 d->bd_wcount = bd->bd_wcount;
2875 d->bd_wdcount = bd->bd_wdcount;
2876 d->bd_wfcount = bd->bd_wfcount;
2877 d->bd_zcopy = bd->bd_zcopy;
2878 d->bd_bufmode = bd->bd_bufmode;
2882 * Handle `netstat -B' stats request
2885 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2887 static const struct xbpf_d zerostats;
2888 struct xbpf_d *xbdbuf, *xbd, tempstats;
2894 * XXX This is not technically correct. It is possible for non
2895 * privileged users to open bpf devices. It would make sense
2896 * if the users who opened the devices were able to retrieve
2897 * the statistics for them, too.
2899 error = priv_check(req->td, PRIV_NET_BPF);
2903 * Check to see if the user is requesting that the counters be
2904 * zeroed out. Explicitly check that the supplied data is zeroed,
2905 * as we aren't allowing the user to set the counters currently.
2907 if (req->newptr != NULL) {
2908 if (req->newlen != sizeof(tempstats))
2910 memset(&tempstats, 0, sizeof(tempstats));
2911 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2914 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2916 bpf_zero_counters();
2919 if (req->oldptr == NULL)
2920 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2921 if (bpf_bpfd_cnt == 0)
2922 return (SYSCTL_OUT(req, 0, 0));
2923 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2925 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2927 free(xbdbuf, M_BPF);
2931 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2933 /* Send writers-only first */
2934 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2935 xbd = &xbdbuf[index++];
2937 bpfstats_fill_xbpf(xbd, bd);
2940 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2941 xbd = &xbdbuf[index++];
2943 bpfstats_fill_xbpf(xbd, bd);
2949 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2950 free(xbdbuf, M_BPF);
2954 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2956 #else /* !DEV_BPF && !NETGRAPH_BPF */
2958 * NOP stubs to allow bpf-using drivers to load and function.
2960 * A 'better' implementation would allow the core bpf functionality
2961 * to be loaded at runtime.
2963 static struct bpf_if bp_null;
2966 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2971 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2976 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2981 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2984 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2988 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2991 *driverp = &bp_null;
2995 bpfdetach(struct ifnet *ifp)
3000 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3002 return -1; /* "no filter" behaviour */
3006 bpf_validate(const struct bpf_insn *f, int len)
3008 return 0; /* false */
3011 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3015 bpf_show_bpf_if(struct bpf_if *bpf_if)
3020 db_printf("%p:\n", bpf_if);
3021 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3022 /* bif_ext.bif_next */
3023 /* bif_ext.bif_dlist */
3024 BPF_DB_PRINTF("%#x", bif_dlt);
3025 BPF_DB_PRINTF("%u", bif_hdrlen);
3026 BPF_DB_PRINTF("%p", bif_ifp);
3029 BPF_DB_PRINTF("%#x", bif_flags);
3032 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3036 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3040 bpf_show_bpf_if((struct bpf_if *)addr);