2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1990, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org>
8 * This code is derived from the Stanford/CMU enet packet filter,
9 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
10 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
11 * Berkeley Laboratory.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
45 #include "opt_netgraph.h"
47 #include <sys/param.h>
49 #include <sys/eventhandler.h>
50 #include <sys/fcntl.h>
54 #include <sys/malloc.h>
56 #include <sys/mutex.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>
66 #include <sys/systm.h>
68 #include <sys/event.h>
73 #include <sys/socket.h>
80 #include <net/if_var.h>
81 #include <net/if_dl.h>
83 #include <net/bpf_buffer.h>
85 #include <net/bpf_jitter.h>
87 #include <net/bpf_zerocopy.h>
88 #include <net/bpfdesc.h>
89 #include <net/route.h>
92 #include <netinet/in.h>
93 #include <netinet/if_ether.h>
94 #include <sys/kernel.h>
95 #include <sys/sysctl.h>
97 #include <net80211/ieee80211_freebsd.h>
99 #include <security/mac/mac_framework.h>
101 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
103 static struct bpf_if_ext dead_bpf_if = {
104 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
108 #define bif_next bif_ext.bif_next
109 #define bif_dlist bif_ext.bif_dlist
110 struct bpf_if_ext bif_ext; /* public members */
111 u_int bif_dlt; /* link layer type */
112 u_int bif_hdrlen; /* length of link header */
113 struct bpfd_list bif_wlist; /* writer-only list */
114 struct ifnet *bif_ifp; /* corresponding interface */
115 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
116 volatile u_int bif_refcnt;
117 struct epoch_context epoch_ctx;
120 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
122 struct bpf_program_buffer {
123 struct epoch_context epoch_ctx;
125 bpf_jit_filter *func;
130 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
132 #define PRINET 26 /* interruptible */
134 #define SIZEOF_BPF_HDR(type) \
135 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
137 #ifdef COMPAT_FREEBSD32
138 #include <sys/mount.h>
139 #include <compat/freebsd32/freebsd32.h>
140 #define BPF_ALIGNMENT32 sizeof(int32_t)
141 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
145 * 32-bit version of structure prepended to each packet. We use this header
146 * instead of the standard one for 32-bit streams. We mark the a stream as
147 * 32-bit the first time we see a 32-bit compat ioctl request.
150 struct timeval32 bh_tstamp; /* time stamp */
151 uint32_t bh_caplen; /* length of captured portion */
152 uint32_t bh_datalen; /* original length of packet */
153 uint16_t bh_hdrlen; /* length of bpf header (this struct
154 plus alignment padding) */
158 struct bpf_program32 {
163 struct bpf_dltlist32 {
168 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
169 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
170 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
171 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
172 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
173 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
176 #define BPF_LOCK() sx_xlock(&bpf_sx)
177 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
178 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
180 * bpf_iflist is a list of BPF interface structures, each corresponding to a
181 * specific DLT. The same network interface might have several BPF interface
182 * structures registered by different layers in the stack (i.e., 802.11
183 * frames, ethernet frames, etc).
185 CK_LIST_HEAD(bpf_iflist, bpf_if);
186 static struct bpf_iflist bpf_iflist;
187 static struct sx bpf_sx; /* bpf global lock */
188 static int bpf_bpfd_cnt;
190 static void bpfif_ref(struct bpf_if *);
191 static void bpfif_rele(struct bpf_if *);
193 static void bpfd_ref(struct bpf_d *);
194 static void bpfd_rele(struct bpf_d *);
195 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
196 static void bpf_detachd(struct bpf_d *);
197 static void bpf_detachd_locked(struct bpf_d *, bool);
198 static void bpfd_free(epoch_context_t);
199 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
200 struct sockaddr *, int *, struct bpf_d *);
201 static int bpf_setif(struct bpf_d *, struct ifreq *);
202 static void bpf_timed_out(void *);
204 bpf_wakeup(struct bpf_d *);
205 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
206 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
208 static void reset_d(struct bpf_d *);
209 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
210 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
211 static int bpf_setdlt(struct bpf_d *, u_int);
212 static void filt_bpfdetach(struct knote *);
213 static int filt_bpfread(struct knote *, long);
214 static void bpf_drvinit(void *);
215 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
217 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
219 int bpf_maxinsns = BPF_MAXINSNS;
220 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
221 &bpf_maxinsns, 0, "Maximum bpf program instructions");
222 static int bpf_zerocopy_enable = 0;
223 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
224 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
225 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
226 bpf_stats_sysctl, "bpf statistics portal");
228 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0;
229 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
230 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RWTUN,
231 &VNET_NAME(bpf_optimize_writers), 0,
232 "Do not send packets until BPF program is set");
234 static d_open_t bpfopen;
235 static d_read_t bpfread;
236 static d_write_t bpfwrite;
237 static d_ioctl_t bpfioctl;
238 static d_poll_t bpfpoll;
239 static d_kqfilter_t bpfkqfilter;
241 static struct cdevsw bpf_cdevsw = {
242 .d_version = D_VERSION,
249 .d_kqfilter = bpfkqfilter,
252 static struct filterops bpfread_filtops = {
254 .f_detach = filt_bpfdetach,
255 .f_event = filt_bpfread,
259 * LOCKING MODEL USED BY BPF
262 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
263 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
264 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
265 * structure fields used by bpf_*tap* code.
267 * Lock order: global lock, then descriptor lock.
269 * There are several possible consumers:
271 * 1. The kernel registers interface pointer with bpfattach().
272 * Each call allocates new bpf_if structure, references ifnet pointer
273 * and links bpf_if into bpf_iflist chain. This is protected with global
276 * 2. An userland application uses ioctl() call to bpf_d descriptor.
277 * All such call are serialized with global lock. BPF filters can be
278 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
279 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
280 * filter pointers, even if change will happen during bpf_tap execution.
281 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
283 * 3. An userland application can write packets into bpf_d descriptor.
284 * There we need to be sure, that ifnet won't disappear during bpfwrite().
286 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
287 * bif_dlist is protected with net_epoch_preempt section. So, it should
288 * be safe to make access to bpf_d descriptor inside the section.
290 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
291 * are modified with global lock held and actual free() is done using
296 bpfif_free(epoch_context_t ctx)
300 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
301 if_rele(bp->bif_ifp);
306 bpfif_ref(struct bpf_if *bp)
309 refcount_acquire(&bp->bif_refcnt);
313 bpfif_rele(struct bpf_if *bp)
316 if (!refcount_release(&bp->bif_refcnt))
318 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
322 bpfd_ref(struct bpf_d *d)
325 refcount_acquire(&d->bd_refcnt);
329 bpfd_rele(struct bpf_d *d)
332 if (!refcount_release(&d->bd_refcnt))
334 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
337 static struct bpf_program_buffer*
338 bpf_program_buffer_alloc(size_t size, int flags)
341 return (malloc(sizeof(struct bpf_program_buffer) + size,
346 bpf_program_buffer_free(epoch_context_t ctx)
348 struct bpf_program_buffer *ptr;
350 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
352 if (ptr->func != NULL)
353 bpf_destroy_jit_filter(ptr->func);
359 * Wrapper functions for various buffering methods. If the set of buffer
360 * modes expands, we will probably want to introduce a switch data structure
361 * similar to protosw, et.
364 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
370 switch (d->bd_bufmode) {
371 case BPF_BUFMODE_BUFFER:
372 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
374 case BPF_BUFMODE_ZBUF:
375 counter_u64_add(d->bd_zcopy, 1);
376 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
379 panic("bpf_buf_append_bytes");
384 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
390 switch (d->bd_bufmode) {
391 case BPF_BUFMODE_BUFFER:
392 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
394 case BPF_BUFMODE_ZBUF:
395 counter_u64_add(d->bd_zcopy, 1);
396 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
399 panic("bpf_buf_append_mbuf");
404 * This function gets called when the free buffer is re-assigned.
407 bpf_buf_reclaimed(struct bpf_d *d)
412 switch (d->bd_bufmode) {
413 case BPF_BUFMODE_BUFFER:
416 case BPF_BUFMODE_ZBUF:
417 bpf_zerocopy_buf_reclaimed(d);
421 panic("bpf_buf_reclaimed");
426 * If the buffer mechanism has a way to decide that a held buffer can be made
427 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
428 * returned if the buffer can be discarded, (0) is returned if it cannot.
431 bpf_canfreebuf(struct bpf_d *d)
436 switch (d->bd_bufmode) {
437 case BPF_BUFMODE_ZBUF:
438 return (bpf_zerocopy_canfreebuf(d));
444 * Allow the buffer model to indicate that the current store buffer is
445 * immutable, regardless of the appearance of space. Return (1) if the
446 * buffer is writable, and (0) if not.
449 bpf_canwritebuf(struct bpf_d *d)
453 switch (d->bd_bufmode) {
454 case BPF_BUFMODE_ZBUF:
455 return (bpf_zerocopy_canwritebuf(d));
461 * Notify buffer model that an attempt to write to the store buffer has
462 * resulted in a dropped packet, in which case the buffer may be considered
466 bpf_buffull(struct bpf_d *d)
471 switch (d->bd_bufmode) {
472 case BPF_BUFMODE_ZBUF:
473 bpf_zerocopy_buffull(d);
479 * Notify the buffer model that a buffer has moved into the hold position.
482 bpf_bufheld(struct bpf_d *d)
487 switch (d->bd_bufmode) {
488 case BPF_BUFMODE_ZBUF:
489 bpf_zerocopy_bufheld(d);
495 bpf_free(struct bpf_d *d)
498 switch (d->bd_bufmode) {
499 case BPF_BUFMODE_BUFFER:
500 return (bpf_buffer_free(d));
502 case BPF_BUFMODE_ZBUF:
503 return (bpf_zerocopy_free(d));
506 panic("bpf_buf_free");
511 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
514 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
516 return (bpf_buffer_uiomove(d, buf, len, uio));
520 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
523 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
525 return (bpf_buffer_ioctl_sblen(d, i));
529 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
532 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
534 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
538 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
541 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
543 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
547 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
550 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
552 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
556 * General BPF functions.
559 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
560 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
562 const struct ieee80211_bpf_params *p;
563 struct ether_header *eh;
571 * Build a sockaddr based on the data link layer type.
572 * We do this at this level because the ethernet header
573 * is copied directly into the data field of the sockaddr.
574 * In the case of SLIP, there is no header and the packet
575 * is forwarded as is.
576 * Also, we are careful to leave room at the front of the mbuf
577 * for the link level header.
581 sockp->sa_family = AF_INET;
586 sockp->sa_family = AF_UNSPEC;
587 /* XXX Would MAXLINKHDR be better? */
588 hlen = ETHER_HDR_LEN;
592 sockp->sa_family = AF_IMPLINK;
597 sockp->sa_family = AF_UNSPEC;
603 * null interface types require a 4 byte pseudo header which
604 * corresponds to the address family of the packet.
606 sockp->sa_family = AF_UNSPEC;
610 case DLT_ATM_RFC1483:
612 * en atm driver requires 4-byte atm pseudo header.
613 * though it isn't standard, vpi:vci needs to be
616 sockp->sa_family = AF_UNSPEC;
617 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
621 sockp->sa_family = AF_UNSPEC;
622 hlen = 4; /* This should match PPP_HDRLEN */
625 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
626 sockp->sa_family = AF_IEEE80211;
630 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
631 sockp->sa_family = AF_IEEE80211;
632 sockp->sa_len = 12; /* XXX != 0 */
633 hlen = sizeof(struct ieee80211_bpf_params);
640 len = uio->uio_resid;
641 if (len < hlen || len - hlen > ifp->if_mtu)
644 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
647 m->m_pkthdr.len = m->m_len = len;
650 error = uiomove(mtod(m, u_char *), len, uio);
654 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
660 /* Check for multicast destination */
663 eh = mtod(m, struct ether_header *);
664 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
665 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
666 ETHER_ADDR_LEN) == 0)
667 m->m_flags |= M_BCAST;
669 m->m_flags |= M_MCAST;
671 if (d->bd_hdrcmplt == 0) {
672 memcpy(eh->ether_shost, IF_LLADDR(ifp),
673 sizeof(eh->ether_shost));
679 * Make room for link header, and copy it to sockaddr
682 if (sockp->sa_family == AF_IEEE80211) {
684 * Collect true length from the parameter header
685 * NB: sockp is known to be zero'd so if we do a
686 * short copy unspecified parameters will be
688 * NB: packet may not be aligned after stripping
692 p = mtod(m, const struct ieee80211_bpf_params *);
694 if (hlen > sizeof(sockp->sa_data)) {
699 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
710 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
711 * then reset its buffers and counters with reset_d().
714 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
721 * Save sysctl value to protect from sysctl change
724 op_w = V_bpf_optimize_writers || d->bd_writer;
726 if (d->bd_bif != NULL)
727 bpf_detachd_locked(d, false);
729 * Point d at bp, and add d to the interface's list.
730 * Since there are many applications using BPF for
731 * sending raw packets only (dhcpd, cdpd are good examples)
732 * we can delay adding d to the list of active listeners until
733 * some filter is configured.
738 * Hold reference to bpif while descriptor uses this interface.
743 /* Add to writers-only list */
744 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
746 * We decrement bd_writer on every filter set operation.
747 * First BIOCSETF is done by pcap_open_live() to set up
748 * snap length. After that appliation usually sets its own
753 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
759 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
760 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
763 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
767 * Check if we need to upgrade our descriptor @d from write-only mode.
770 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
773 int is_snap, need_upgrade;
776 * Check if we've already upgraded or new filter is empty.
778 if (d->bd_writer == 0 || fcode == NULL)
784 * Check if cmd looks like snaplen setting from
785 * pcap_bpf.c:pcap_open_live().
786 * Note we're not checking .k value here:
787 * while pcap_open_live() definitely sets to non-zero value,
788 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
789 * do not consider upgrading immediately
791 if (cmd == BIOCSETF && flen == 1 &&
792 fcode[0].code == (BPF_RET | BPF_K))
799 * We're setting first filter and it doesn't look like
800 * setting snaplen. We're probably using bpf directly.
801 * Upgrade immediately.
806 * Do not require upgrade by first BIOCSETF
807 * (used to set snaplen) by pcap_open_live().
810 if (--d->bd_writer == 0) {
812 * First snaplen filter has already
813 * been set. This is probably catch-all
821 "%s: filter function set by pid %d, "
822 "bd_writer counter %d, snap %d upgrade %d",
823 __func__, d->bd_pid, d->bd_writer,
824 is_snap, need_upgrade);
826 return (need_upgrade);
830 * Detach a file from its interface.
833 bpf_detachd(struct bpf_d *d)
836 bpf_detachd_locked(d, false);
841 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
848 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
850 /* Check if descriptor is attached */
851 if ((bp = d->bd_bif) == NULL)
855 /* Remove d from the interface's descriptor list. */
856 CK_LIST_REMOVE(d, bd_next);
857 /* Save bd_writer value */
858 error = d->bd_writer;
863 * Notify descriptor as it's detached, so that any
864 * sleepers wake up and get ENXIO.
871 /* Call event handler iff d is attached */
873 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
876 * Check if this descriptor had requested promiscuous mode.
877 * If so and ifnet is not detached, turn it off.
879 if (d->bd_promisc && !detached_ifp) {
881 CURVNET_SET(ifp->if_vnet);
882 error = ifpromisc(ifp, 0);
884 if (error != 0 && error != ENXIO) {
886 * ENXIO can happen if a pccard is unplugged
887 * Something is really wrong if we were able to put
888 * the driver into promiscuous mode, but can't
891 if_printf(bp->bif_ifp,
892 "bpf_detach: ifpromisc failed (%d)\n", error);
899 * Close the descriptor by detaching it from its interface,
900 * deallocating its buffers, and marking it free.
905 struct bpf_d *d = data;
908 if (d->bd_state == BPF_WAITING)
909 callout_stop(&d->bd_callout);
910 d->bd_state = BPF_IDLE;
912 funsetown(&d->bd_sigio);
915 mac_bpfdesc_destroy(d);
917 seldrain(&d->bd_sel);
918 knlist_destroy(&d->bd_sel.si_note);
919 callout_drain(&d->bd_callout);
924 * Open ethernet device. Returns ENXIO for illegal minor device number,
925 * EBUSY if file is open by another process.
929 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
934 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
935 error = devfs_set_cdevpriv(d, bpf_dtor);
942 d->bd_rcount = counter_u64_alloc(M_WAITOK);
943 d->bd_dcount = counter_u64_alloc(M_WAITOK);
944 d->bd_fcount = counter_u64_alloc(M_WAITOK);
945 d->bd_wcount = counter_u64_alloc(M_WAITOK);
946 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
947 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
948 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
951 * For historical reasons, perform a one-time initialization call to
952 * the buffer routines, even though we're not yet committed to a
953 * particular buffer method.
956 if ((flags & FREAD) == 0)
958 d->bd_hbuf_in_use = 0;
959 d->bd_bufmode = BPF_BUFMODE_BUFFER;
961 d->bd_direction = BPF_D_INOUT;
963 BPF_PID_REFRESH(d, td);
966 mac_bpfdesc_create(td->td_ucred, d);
968 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
969 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
970 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
976 * bpfread - read next chunk of packets from buffers
979 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
986 error = devfs_get_cdevpriv((void **)&d);
991 * Restrict application to use a buffer the same size as
994 if (uio->uio_resid != d->bd_bufsize)
997 non_block = ((ioflag & O_NONBLOCK) != 0);
1000 BPF_PID_REFRESH_CUR(d);
1001 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1003 return (EOPNOTSUPP);
1005 if (d->bd_state == BPF_WAITING)
1006 callout_stop(&d->bd_callout);
1007 timed_out = (d->bd_state == BPF_TIMED_OUT);
1008 d->bd_state = BPF_IDLE;
1009 while (d->bd_hbuf_in_use) {
1010 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1011 PRINET|PCATCH, "bd_hbuf", 0);
1018 * If the hold buffer is empty, then do a timed sleep, which
1019 * ends when the timeout expires or when enough packets
1020 * have arrived to fill the store buffer.
1022 while (d->bd_hbuf == NULL) {
1023 if (d->bd_slen != 0) {
1025 * A packet(s) either arrived since the previous
1026 * read or arrived while we were asleep.
1028 if (d->bd_immediate || non_block || timed_out) {
1030 * Rotate the buffers and return what's here
1031 * if we are in immediate mode, non-blocking
1032 * flag is set, or this descriptor timed out.
1040 * No data is available, check to see if the bpf device
1041 * is still pointed at a real interface. If not, return
1042 * ENXIO so that the userland process knows to rebind
1043 * it before using it again.
1045 if (d->bd_bif == NULL) {
1052 return (EWOULDBLOCK);
1054 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1055 "bpf", d->bd_rtout);
1056 if (error == EINTR || error == ERESTART) {
1060 if (error == EWOULDBLOCK) {
1062 * On a timeout, return what's in the buffer,
1063 * which may be nothing. If there is something
1064 * in the store buffer, we can rotate the buffers.
1068 * We filled up the buffer in between
1069 * getting the timeout and arriving
1070 * here, so we don't need to rotate.
1074 if (d->bd_slen == 0) {
1083 * At this point, we know we have something in the hold slot.
1085 d->bd_hbuf_in_use = 1;
1089 * Move data from hold buffer into user space.
1090 * We know the entire buffer is transferred since
1091 * we checked above that the read buffer is bpf_bufsize bytes.
1093 * We do not have to worry about simultaneous reads because
1094 * we waited for sole access to the hold buffer above.
1096 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1099 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1100 d->bd_fbuf = d->bd_hbuf;
1103 bpf_buf_reclaimed(d);
1104 d->bd_hbuf_in_use = 0;
1105 wakeup(&d->bd_hbuf_in_use);
1112 * If there are processes sleeping on this descriptor, wake them up.
1114 static __inline void
1115 bpf_wakeup(struct bpf_d *d)
1118 BPFD_LOCK_ASSERT(d);
1119 if (d->bd_state == BPF_WAITING) {
1120 callout_stop(&d->bd_callout);
1121 d->bd_state = BPF_IDLE;
1124 if (d->bd_async && d->bd_sig && d->bd_sigio)
1125 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1127 selwakeuppri(&d->bd_sel, PRINET);
1128 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1132 bpf_timed_out(void *arg)
1134 struct bpf_d *d = (struct bpf_d *)arg;
1136 BPFD_LOCK_ASSERT(d);
1138 if (callout_pending(&d->bd_callout) ||
1139 !callout_active(&d->bd_callout))
1141 if (d->bd_state == BPF_WAITING) {
1142 d->bd_state = BPF_TIMED_OUT;
1143 if (d->bd_slen != 0)
1149 bpf_ready(struct bpf_d *d)
1152 BPFD_LOCK_ASSERT(d);
1154 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1156 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1163 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1166 struct sockaddr dst;
1167 struct epoch_tracker et;
1171 struct mbuf *m, *mc;
1174 error = devfs_get_cdevpriv((void **)&d);
1178 NET_EPOCH_ENTER(et);
1180 BPF_PID_REFRESH_CUR(d);
1181 counter_u64_add(d->bd_wcount, 1);
1182 if ((bp = d->bd_bif) == NULL) {
1188 if ((ifp->if_flags & IFF_UP) == 0) {
1193 if (uio->uio_resid == 0)
1196 bzero(&dst, sizeof(dst));
1201 * Take extra reference, unlock d and exit from epoch section,
1202 * since bpf_movein() can sleep.
1208 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1209 &m, &dst, &hlen, d);
1212 counter_u64_add(d->bd_wdcount, 1);
1219 * Check that descriptor is still attached to the interface.
1220 * This can happen on bpfdetach(). To avoid access to detached
1221 * ifnet, free mbuf and return ENXIO.
1223 if (d->bd_bif == NULL) {
1224 counter_u64_add(d->bd_wdcount, 1);
1230 counter_u64_add(d->bd_wfcount, 1);
1232 dst.sa_family = pseudo_AF_HDRCMPLT;
1234 if (d->bd_feedback) {
1235 mc = m_dup(m, M_NOWAIT);
1237 mc->m_pkthdr.rcvif = ifp;
1238 /* Set M_PROMISC for outgoing packets to be discarded. */
1239 if (d->bd_direction == BPF_D_INOUT)
1240 m->m_flags |= M_PROMISC;
1244 m->m_pkthdr.len -= hlen;
1246 m->m_data += hlen; /* XXX */
1248 CURVNET_SET(ifp->if_vnet);
1250 mac_bpfdesc_create_mbuf(d, m);
1252 mac_bpfdesc_create_mbuf(d, mc);
1255 bzero(&ro, sizeof(ro));
1257 ro.ro_prepend = (u_char *)&dst.sa_data;
1259 ro.ro_flags = RT_HAS_HEADER;
1262 /* Avoid possible recursion on BPFD_LOCK(). */
1263 NET_EPOCH_ENTER(et);
1265 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1267 counter_u64_add(d->bd_wdcount, 1);
1271 (*ifp->if_input)(ifp, mc);
1281 counter_u64_add(d->bd_wdcount, 1);
1288 * Reset a descriptor by flushing its packet buffer and clearing the receive
1289 * and drop counts. This is doable for kernel-only buffers, but with
1290 * zero-copy buffers, we can't write to (or rotate) buffers that are
1291 * currently owned by userspace. It would be nice if we could encapsulate
1292 * this logic in the buffer code rather than here.
1295 reset_d(struct bpf_d *d)
1298 BPFD_LOCK_ASSERT(d);
1300 while (d->bd_hbuf_in_use)
1301 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1303 if ((d->bd_hbuf != NULL) &&
1304 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1305 /* Free the hold buffer. */
1306 d->bd_fbuf = d->bd_hbuf;
1309 bpf_buf_reclaimed(d);
1311 if (bpf_canwritebuf(d))
1313 counter_u64_zero(d->bd_rcount);
1314 counter_u64_zero(d->bd_dcount);
1315 counter_u64_zero(d->bd_fcount);
1316 counter_u64_zero(d->bd_wcount);
1317 counter_u64_zero(d->bd_wfcount);
1318 counter_u64_zero(d->bd_wdcount);
1319 counter_u64_zero(d->bd_zcopy);
1323 * FIONREAD Check for read packet available.
1324 * BIOCGBLEN Get buffer len [for read()].
1325 * BIOCSETF Set read filter.
1326 * BIOCSETFNR Set read filter without resetting descriptor.
1327 * BIOCSETWF Set write filter.
1328 * BIOCFLUSH Flush read packet buffer.
1329 * BIOCPROMISC Put interface into promiscuous mode.
1330 * BIOCGDLT Get link layer type.
1331 * BIOCGETIF Get interface name.
1332 * BIOCSETIF Set interface.
1333 * BIOCSRTIMEOUT Set read timeout.
1334 * BIOCGRTIMEOUT Get read timeout.
1335 * BIOCGSTATS Get packet stats.
1336 * BIOCIMMEDIATE Set immediate mode.
1337 * BIOCVERSION Get filter language version.
1338 * BIOCGHDRCMPLT Get "header already complete" flag
1339 * BIOCSHDRCMPLT Set "header already complete" flag
1340 * BIOCGDIRECTION Get packet direction flag
1341 * BIOCSDIRECTION Set packet direction flag
1342 * BIOCGTSTAMP Get time stamp format and resolution.
1343 * BIOCSTSTAMP Set time stamp format and resolution.
1344 * BIOCLOCK Set "locked" flag
1345 * BIOCFEEDBACK Set packet feedback mode.
1346 * BIOCSETZBUF Set current zero-copy buffer locations.
1347 * BIOCGETZMAX Get maximum zero-copy buffer size.
1348 * BIOCROTZBUF Force rotation of zero-copy buffer
1349 * BIOCSETBUFMODE Set buffer mode.
1350 * BIOCGETBUFMODE Get current buffer mode.
1354 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1360 error = devfs_get_cdevpriv((void **)&d);
1365 * Refresh PID associated with this descriptor.
1368 BPF_PID_REFRESH(d, td);
1369 if (d->bd_state == BPF_WAITING)
1370 callout_stop(&d->bd_callout);
1371 d->bd_state = BPF_IDLE;
1374 if (d->bd_locked == 1) {
1380 #ifdef COMPAT_FREEBSD32
1381 case BIOCGDLTLIST32:
1385 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1386 case BIOCGRTIMEOUT32:
1397 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1398 case BIOCSRTIMEOUT32:
1408 #ifdef COMPAT_FREEBSD32
1410 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1411 * that it will get 32-bit packet headers.
1417 case BIOCGDLTLIST32:
1418 case BIOCGRTIMEOUT32:
1419 case BIOCSRTIMEOUT32:
1420 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1428 CURVNET_SET(TD_TO_VNET(td));
1435 * Check for read packet available.
1443 while (d->bd_hbuf_in_use)
1444 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1445 PRINET, "bd_hbuf", 0);
1455 * Get buffer len [for read()].
1459 *(u_int *)addr = d->bd_bufsize;
1464 * Set buffer length.
1467 error = bpf_ioctl_sblen(d, (u_int *)addr);
1471 * Set link layer read filter.
1476 #ifdef COMPAT_FREEBSD32
1481 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1485 * Flush read packet buffer.
1494 * Put interface into promiscuous mode.
1497 if (d->bd_bif == NULL) {
1499 * No interface attached yet.
1504 if (d->bd_promisc == 0) {
1505 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1512 * Get current data link type.
1516 if (d->bd_bif == NULL)
1519 *(u_int *)addr = d->bd_bif->bif_dlt;
1524 * Get a list of supported data link types.
1526 #ifdef COMPAT_FREEBSD32
1527 case BIOCGDLTLIST32:
1529 struct bpf_dltlist32 *list32;
1530 struct bpf_dltlist dltlist;
1532 list32 = (struct bpf_dltlist32 *)addr;
1533 dltlist.bfl_len = list32->bfl_len;
1534 dltlist.bfl_list = PTRIN(list32->bfl_list);
1536 if (d->bd_bif == NULL)
1539 error = bpf_getdltlist(d, &dltlist);
1541 list32->bfl_len = dltlist.bfl_len;
1550 if (d->bd_bif == NULL)
1553 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1558 * Set data link type.
1562 if (d->bd_bif == NULL)
1565 error = bpf_setdlt(d, *(u_int *)addr);
1570 * Get interface name.
1574 if (d->bd_bif == NULL)
1577 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1578 struct ifreq *const ifr = (struct ifreq *)addr;
1580 strlcpy(ifr->ifr_name, ifp->if_xname,
1581 sizeof(ifr->ifr_name));
1591 int alloc_buf, size;
1594 * Behavior here depends on the buffering model. If
1595 * we're using kernel memory buffers, then we can
1596 * allocate them here. If we're using zero-copy,
1597 * then the user process must have registered buffers
1598 * by the time we get here.
1602 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1607 size = d->bd_bufsize;
1608 error = bpf_buffer_ioctl_sblen(d, &size);
1613 error = bpf_setif(d, (struct ifreq *)addr);
1622 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1623 case BIOCSRTIMEOUT32:
1626 struct timeval *tv = (struct timeval *)addr;
1627 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1628 struct timeval32 *tv32;
1629 struct timeval tv64;
1631 if (cmd == BIOCSRTIMEOUT32) {
1632 tv32 = (struct timeval32 *)addr;
1634 tv->tv_sec = tv32->tv_sec;
1635 tv->tv_usec = tv32->tv_usec;
1638 tv = (struct timeval *)addr;
1641 * Subtract 1 tick from tvtohz() since this isn't
1644 if ((error = itimerfix(tv)) == 0)
1645 d->bd_rtout = tvtohz(tv) - 1;
1653 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1654 case BIOCGRTIMEOUT32:
1658 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1659 struct timeval32 *tv32;
1660 struct timeval tv64;
1662 if (cmd == BIOCGRTIMEOUT32)
1666 tv = (struct timeval *)addr;
1668 tv->tv_sec = d->bd_rtout / hz;
1669 tv->tv_usec = (d->bd_rtout % hz) * tick;
1670 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1671 if (cmd == BIOCGRTIMEOUT32) {
1672 tv32 = (struct timeval32 *)addr;
1673 tv32->tv_sec = tv->tv_sec;
1674 tv32->tv_usec = tv->tv_usec;
1686 struct bpf_stat *bs = (struct bpf_stat *)addr;
1688 /* XXXCSJP overflow */
1689 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1690 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1695 * Set immediate mode.
1699 d->bd_immediate = *(u_int *)addr;
1705 struct bpf_version *bv = (struct bpf_version *)addr;
1707 bv->bv_major = BPF_MAJOR_VERSION;
1708 bv->bv_minor = BPF_MINOR_VERSION;
1713 * Get "header already complete" flag
1717 *(u_int *)addr = d->bd_hdrcmplt;
1722 * Set "header already complete" flag
1726 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1731 * Get packet direction flag
1733 case BIOCGDIRECTION:
1735 *(u_int *)addr = d->bd_direction;
1740 * Set packet direction flag
1742 case BIOCSDIRECTION:
1746 direction = *(u_int *)addr;
1747 switch (direction) {
1752 d->bd_direction = direction;
1762 * Get packet timestamp format and resolution.
1766 *(u_int *)addr = d->bd_tstamp;
1771 * Set packet timestamp format and resolution.
1777 func = *(u_int *)addr;
1778 if (BPF_T_VALID(func))
1779 d->bd_tstamp = func;
1787 d->bd_feedback = *(u_int *)addr;
1797 case FIONBIO: /* Non-blocking I/O */
1800 case FIOASYNC: /* Send signal on receive packets */
1802 d->bd_async = *(int *)addr;
1808 * XXX: Add some sort of locking here?
1809 * fsetown() can sleep.
1811 error = fsetown(*(int *)addr, &d->bd_sigio);
1816 *(int *)addr = fgetown(&d->bd_sigio);
1820 /* This is deprecated, FIOSETOWN should be used instead. */
1822 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1825 /* This is deprecated, FIOGETOWN should be used instead. */
1827 *(int *)addr = -fgetown(&d->bd_sigio);
1830 case BIOCSRSIG: /* Set receive signal */
1834 sig = *(u_int *)addr;
1847 *(u_int *)addr = d->bd_sig;
1851 case BIOCGETBUFMODE:
1853 *(u_int *)addr = d->bd_bufmode;
1857 case BIOCSETBUFMODE:
1859 * Allow the buffering mode to be changed as long as we
1860 * haven't yet committed to a particular mode. Our
1861 * definition of commitment, for now, is whether or not a
1862 * buffer has been allocated or an interface attached, since
1863 * that's the point where things get tricky.
1865 switch (*(u_int *)addr) {
1866 case BPF_BUFMODE_BUFFER:
1869 case BPF_BUFMODE_ZBUF:
1870 if (bpf_zerocopy_enable)
1880 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1881 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1886 d->bd_bufmode = *(u_int *)addr;
1891 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1895 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1899 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1907 * Set d's packet filter program to fp. If this file already has a filter,
1908 * free it and replace it. Returns EINVAL for bogus requests.
1910 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1914 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1916 #ifdef COMPAT_FREEBSD32
1917 struct bpf_program fp_swab;
1918 struct bpf_program32 *fp32;
1920 struct bpf_program_buffer *fcode;
1921 struct bpf_insn *filter;
1923 bpf_jit_filter *jfunc;
1929 #ifdef COMPAT_FREEBSD32
1934 fp32 = (struct bpf_program32 *)fp;
1935 fp_swab.bf_len = fp32->bf_len;
1937 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1956 * Check new filter validness before acquiring any locks.
1957 * Allocate memory for new filter, if needed.
1960 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1962 size = flen * sizeof(*fp->bf_insns);
1964 /* We're setting up new filter. Copy and check actual data. */
1965 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1966 filter = (struct bpf_insn *)fcode->buffer;
1967 if (copyin(fp->bf_insns, filter, size) != 0 ||
1968 !bpf_validate(filter, flen)) {
1973 if (cmd != BIOCSETWF) {
1975 * Filter is copied inside fcode and is
1978 jfunc = bpf_jitter(filter, flen);
1983 track_event = false;
1988 /* Set up new filter. */
1989 if (cmd == BIOCSETWF) {
1990 if (d->bd_wfilter != NULL) {
1991 fcode = __containerof((void *)d->bd_wfilter,
1992 struct bpf_program_buffer, buffer);
1997 d->bd_wfilter = filter;
1999 if (d->bd_rfilter != NULL) {
2000 fcode = __containerof((void *)d->bd_rfilter,
2001 struct bpf_program_buffer, buffer);
2003 fcode->func = d->bd_bfilter;
2006 d->bd_rfilter = filter;
2008 d->bd_bfilter = jfunc;
2010 if (cmd == BIOCSETF)
2013 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2015 * Filter can be set several times without
2016 * specifying interface. In this case just mark d
2020 if (d->bd_bif != NULL) {
2022 * Remove descriptor from writers-only list
2023 * and add it to active readers list.
2025 CK_LIST_REMOVE(d, bd_next);
2026 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2029 "%s: upgrade required by pid %d",
2030 __func__, d->bd_pid);
2038 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2041 EVENTHANDLER_INVOKE(bpf_track,
2042 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2049 * Detach a file from its current interface (if attached at all) and attach
2050 * to the interface indicated by the name stored in ifr.
2051 * Return an errno or 0.
2054 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2057 struct ifnet *theywant;
2061 theywant = ifunit(ifr->ifr_name);
2062 if (theywant == NULL || theywant->if_bpf == NULL)
2065 bp = theywant->if_bpf;
2067 * At this point, we expect the buffer is already allocated. If not,
2070 switch (d->bd_bufmode) {
2071 case BPF_BUFMODE_BUFFER:
2072 case BPF_BUFMODE_ZBUF:
2073 if (d->bd_sbuf == NULL)
2078 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2080 if (bp != d->bd_bif)
2091 * Support for select() and poll() system calls
2093 * Return true iff the specific operation will not block indefinitely.
2094 * Otherwise, return false but make a note that a selwakeup() must be done.
2097 bpfpoll(struct cdev *dev, int events, struct thread *td)
2102 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2104 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2107 * Refresh PID associated with this descriptor.
2109 revents = events & (POLLOUT | POLLWRNORM);
2111 BPF_PID_REFRESH(d, td);
2112 if (events & (POLLIN | POLLRDNORM)) {
2114 revents |= events & (POLLIN | POLLRDNORM);
2116 selrecord(td, &d->bd_sel);
2117 /* Start the read timeout if necessary. */
2118 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2119 callout_reset(&d->bd_callout, d->bd_rtout,
2121 d->bd_state = BPF_WAITING;
2130 * Support for kevent() system call. Register EVFILT_READ filters and
2131 * reject all others.
2134 bpfkqfilter(struct cdev *dev, struct knote *kn)
2138 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2139 kn->kn_filter != EVFILT_READ)
2143 * Refresh PID associated with this descriptor.
2146 BPF_PID_REFRESH_CUR(d);
2147 kn->kn_fop = &bpfread_filtops;
2149 knlist_add(&d->bd_sel.si_note, kn, 1);
2156 filt_bpfdetach(struct knote *kn)
2158 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2160 knlist_remove(&d->bd_sel.si_note, kn, 0);
2164 filt_bpfread(struct knote *kn, long hint)
2166 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2169 BPFD_LOCK_ASSERT(d);
2170 ready = bpf_ready(d);
2172 kn->kn_data = d->bd_slen;
2174 * Ignore the hold buffer if it is being copied to user space.
2176 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2177 kn->kn_data += d->bd_hlen;
2178 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2179 callout_reset(&d->bd_callout, d->bd_rtout,
2181 d->bd_state = BPF_WAITING;
2187 #define BPF_TSTAMP_NONE 0
2188 #define BPF_TSTAMP_FAST 1
2189 #define BPF_TSTAMP_NORMAL 2
2190 #define BPF_TSTAMP_EXTERN 3
2193 bpf_ts_quality(int tstype)
2196 if (tstype == BPF_T_NONE)
2197 return (BPF_TSTAMP_NONE);
2198 if ((tstype & BPF_T_FAST) != 0)
2199 return (BPF_TSTAMP_FAST);
2201 return (BPF_TSTAMP_NORMAL);
2205 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2210 quality = bpf_ts_quality(tstype);
2211 if (quality == BPF_TSTAMP_NONE)
2215 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2217 *bt = *(struct bintime *)(tag + 1);
2218 return (BPF_TSTAMP_EXTERN);
2221 if (quality == BPF_TSTAMP_NORMAL)
2230 * Incoming linkage from device drivers. Process the packet pkt, of length
2231 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2232 * by each process' filter, and if accepted, stashed into the corresponding
2236 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2238 struct epoch_tracker et;
2247 gottime = BPF_TSTAMP_NONE;
2248 NET_EPOCH_ENTER(et);
2249 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2250 counter_u64_add(d->bd_rcount, 1);
2252 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2253 * is no way for the caller to indiciate to us whether this
2254 * packet is inbound or outbound. In the bpf_mtap() routines,
2255 * we use the interface pointers on the mbuf to figure it out.
2258 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2260 slen = (*(bf->func))(pkt, pktlen, pktlen);
2263 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2266 * Filter matches. Let's to acquire write lock.
2269 counter_u64_add(d->bd_fcount, 1);
2270 if (gottime < bpf_ts_quality(d->bd_tstamp))
2271 gottime = bpf_gettime(&bt, d->bd_tstamp,
2274 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2276 catchpacket(d, pkt, pktlen, slen,
2277 bpf_append_bytes, &bt);
2284 #define BPF_CHECK_DIRECTION(d, r, i) \
2285 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2286 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2289 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2290 * Locking model is explained in bpf_tap().
2293 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2295 struct epoch_tracker et;
2304 /* Skip outgoing duplicate packets. */
2305 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2306 m->m_flags &= ~M_PROMISC;
2310 pktlen = m_length(m, NULL);
2311 gottime = BPF_TSTAMP_NONE;
2313 NET_EPOCH_ENTER(et);
2314 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2315 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2317 counter_u64_add(d->bd_rcount, 1);
2319 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2320 /* XXX We cannot handle multiple mbufs. */
2321 if (bf != NULL && m->m_next == NULL)
2322 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2326 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2330 counter_u64_add(d->bd_fcount, 1);
2331 if (gottime < bpf_ts_quality(d->bd_tstamp))
2332 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2334 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2336 catchpacket(d, (u_char *)m, pktlen, slen,
2337 bpf_append_mbuf, &bt);
2345 * Incoming linkage from device drivers, when packet is in
2346 * an mbuf chain and to be prepended by a contiguous header.
2349 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2351 struct epoch_tracker et;
2358 /* Skip outgoing duplicate packets. */
2359 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2360 m->m_flags &= ~M_PROMISC;
2364 pktlen = m_length(m, NULL);
2366 * Craft on-stack mbuf suitable for passing to bpf_filter.
2367 * Note that we cut corners here; we only setup what's
2368 * absolutely needed--this mbuf should never go anywhere else.
2376 gottime = BPF_TSTAMP_NONE;
2378 NET_EPOCH_ENTER(et);
2379 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2380 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2382 counter_u64_add(d->bd_rcount, 1);
2383 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2387 counter_u64_add(d->bd_fcount, 1);
2388 if (gottime < bpf_ts_quality(d->bd_tstamp))
2389 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2391 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2393 catchpacket(d, (u_char *)&mb, pktlen, slen,
2394 bpf_append_mbuf, &bt);
2401 #undef BPF_CHECK_DIRECTION
2402 #undef BPF_TSTAMP_NONE
2403 #undef BPF_TSTAMP_FAST
2404 #undef BPF_TSTAMP_NORMAL
2405 #undef BPF_TSTAMP_EXTERN
2408 bpf_hdrlen(struct bpf_d *d)
2412 hdrlen = d->bd_bif->bif_hdrlen;
2413 #ifndef BURN_BRIDGES
2414 if (d->bd_tstamp == BPF_T_NONE ||
2415 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2416 #ifdef COMPAT_FREEBSD32
2418 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2421 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2424 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2425 #ifdef COMPAT_FREEBSD32
2427 hdrlen = BPF_WORDALIGN32(hdrlen);
2430 hdrlen = BPF_WORDALIGN(hdrlen);
2432 return (hdrlen - d->bd_bif->bif_hdrlen);
2436 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2438 struct bintime bt2, boottimebin;
2440 struct timespec tsn;
2442 if ((tstype & BPF_T_MONOTONIC) == 0) {
2444 getboottimebin(&boottimebin);
2445 bintime_add(&bt2, &boottimebin);
2448 switch (BPF_T_FORMAT(tstype)) {
2449 case BPF_T_MICROTIME:
2450 bintime2timeval(bt, &tsm);
2451 ts->bt_sec = tsm.tv_sec;
2452 ts->bt_frac = tsm.tv_usec;
2454 case BPF_T_NANOTIME:
2455 bintime2timespec(bt, &tsn);
2456 ts->bt_sec = tsn.tv_sec;
2457 ts->bt_frac = tsn.tv_nsec;
2460 ts->bt_sec = bt->sec;
2461 ts->bt_frac = bt->frac;
2467 * Move the packet data from interface memory (pkt) into the
2468 * store buffer. "cpfn" is the routine called to do the actual data
2469 * transfer. bcopy is passed in to copy contiguous chunks, while
2470 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2471 * pkt is really an mbuf.
2474 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2475 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2478 struct bpf_xhdr hdr;
2479 #ifndef BURN_BRIDGES
2480 struct bpf_hdr hdr_old;
2481 #ifdef COMPAT_FREEBSD32
2482 struct bpf_hdr32 hdr32_old;
2485 int caplen, curlen, hdrlen, totlen;
2490 BPFD_LOCK_ASSERT(d);
2491 if (d->bd_bif == NULL) {
2492 /* Descriptor was detached in concurrent thread */
2493 counter_u64_add(d->bd_dcount, 1);
2498 * Detect whether user space has released a buffer back to us, and if
2499 * so, move it from being a hold buffer to a free buffer. This may
2500 * not be the best place to do it (for example, we might only want to
2501 * run this check if we need the space), but for now it's a reliable
2504 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2505 d->bd_fbuf = d->bd_hbuf;
2508 bpf_buf_reclaimed(d);
2512 * Figure out how many bytes to move. If the packet is
2513 * greater or equal to the snapshot length, transfer that
2514 * much. Otherwise, transfer the whole packet (unless
2515 * we hit the buffer size limit).
2517 hdrlen = bpf_hdrlen(d);
2518 totlen = hdrlen + min(snaplen, pktlen);
2519 if (totlen > d->bd_bufsize)
2520 totlen = d->bd_bufsize;
2523 * Round up the end of the previous packet to the next longword.
2525 * Drop the packet if there's no room and no hope of room
2526 * If the packet would overflow the storage buffer or the storage
2527 * buffer is considered immutable by the buffer model, try to rotate
2528 * the buffer and wakeup pending processes.
2530 #ifdef COMPAT_FREEBSD32
2532 curlen = BPF_WORDALIGN32(d->bd_slen);
2535 curlen = BPF_WORDALIGN(d->bd_slen);
2536 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2537 if (d->bd_fbuf == NULL) {
2539 * There's no room in the store buffer, and no
2540 * prospect of room, so drop the packet. Notify the
2544 counter_u64_add(d->bd_dcount, 1);
2547 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2551 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2553 * Immediate mode is set, or the read timeout has already
2554 * expired during a select call. A packet arrived, so the
2555 * reader should be woken up.
2558 caplen = totlen - hdrlen;
2559 tstype = d->bd_tstamp;
2560 do_timestamp = tstype != BPF_T_NONE;
2561 #ifndef BURN_BRIDGES
2562 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2565 bpf_bintime2ts(bt, &ts, tstype);
2566 #ifdef COMPAT_FREEBSD32
2567 if (d->bd_compat32) {
2568 bzero(&hdr32_old, sizeof(hdr32_old));
2570 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2571 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2573 hdr32_old.bh_datalen = pktlen;
2574 hdr32_old.bh_hdrlen = hdrlen;
2575 hdr32_old.bh_caplen = caplen;
2576 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2581 bzero(&hdr_old, sizeof(hdr_old));
2583 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2584 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2586 hdr_old.bh_datalen = pktlen;
2587 hdr_old.bh_hdrlen = hdrlen;
2588 hdr_old.bh_caplen = caplen;
2589 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2596 * Append the bpf header. Note we append the actual header size, but
2597 * move forward the length of the header plus padding.
2599 bzero(&hdr, sizeof(hdr));
2601 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2602 hdr.bh_datalen = pktlen;
2603 hdr.bh_hdrlen = hdrlen;
2604 hdr.bh_caplen = caplen;
2605 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2608 * Copy the packet data into the store buffer and update its length.
2610 #ifndef BURN_BRIDGES
2613 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2614 d->bd_slen = curlen + totlen;
2621 * Free buffers currently in use by a descriptor.
2625 bpfd_free(epoch_context_t ctx)
2628 struct bpf_program_buffer *p;
2631 * We don't need to lock out interrupts since this descriptor has
2632 * been detached from its interface and it yet hasn't been marked
2635 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2637 if (d->bd_rfilter != NULL) {
2638 p = __containerof((void *)d->bd_rfilter,
2639 struct bpf_program_buffer, buffer);
2641 p->func = d->bd_bfilter;
2643 bpf_program_buffer_free(&p->epoch_ctx);
2645 if (d->bd_wfilter != NULL) {
2646 p = __containerof((void *)d->bd_wfilter,
2647 struct bpf_program_buffer, buffer);
2651 bpf_program_buffer_free(&p->epoch_ctx);
2654 mtx_destroy(&d->bd_lock);
2655 counter_u64_free(d->bd_rcount);
2656 counter_u64_free(d->bd_dcount);
2657 counter_u64_free(d->bd_fcount);
2658 counter_u64_free(d->bd_wcount);
2659 counter_u64_free(d->bd_wfcount);
2660 counter_u64_free(d->bd_wdcount);
2661 counter_u64_free(d->bd_zcopy);
2666 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2667 * fixed size of the link header (variable length headers not yet supported).
2670 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2673 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2677 * Attach an interface to bpf. ifp is a pointer to the structure
2678 * defining the interface to be attached, dlt is the link layer type,
2679 * and hdrlen is the fixed size of the link header (variable length
2680 * headers are not yet supporrted).
2683 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2684 struct bpf_if **driverp)
2688 KASSERT(*driverp == NULL,
2689 ("bpfattach2: driverp already initialized"));
2691 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2693 CK_LIST_INIT(&bp->bif_dlist);
2694 CK_LIST_INIT(&bp->bif_wlist);
2697 bp->bif_hdrlen = hdrlen;
2698 bp->bif_bpf = driverp;
2702 * Reference ifnet pointer, so it won't freed until
2707 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2710 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2711 if_printf(ifp, "bpf attached\n");
2716 * When moving interfaces between vnet instances we need a way to
2717 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2718 * after the vmove. We unfortunately have no device driver infrastructure
2719 * to query the interface for these values after creation/attach, thus
2720 * add this as a workaround.
2723 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2728 if (bif_dlt == NULL && bif_hdrlen == NULL)
2731 if (bif_dlt != NULL)
2732 *bif_dlt = bp->bif_dlt;
2733 if (bif_hdrlen != NULL)
2734 *bif_hdrlen = bp->bif_hdrlen;
2741 * Detach bpf from an interface. This involves detaching each descriptor
2742 * associated with the interface. Notify each descriptor as it's detached
2743 * so that any sleepers wake up and get ENXIO.
2746 bpfdetach(struct ifnet *ifp)
2748 struct bpf_if *bp, *bp_temp;
2752 /* Find all bpf_if struct's which reference ifp and detach them. */
2753 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2754 if (ifp != bp->bif_ifp)
2757 CK_LIST_REMOVE(bp, bif_next);
2758 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2761 "%s: sheduling free for encap %d (%p) for if %p",
2762 __func__, bp->bif_dlt, bp, ifp);
2764 /* Detach common descriptors */
2765 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2766 bpf_detachd_locked(d, true);
2769 /* Detach writer-only descriptors */
2770 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2771 bpf_detachd_locked(d, true);
2779 * Get a list of available data link type of the interface.
2782 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2791 ifp = d->bd_bif->bif_ifp;
2793 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2794 if (bp->bif_ifp == ifp)
2797 if (bfl->bfl_list == NULL) {
2801 if (n1 > bfl->bfl_len)
2804 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2806 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2807 if (bp->bif_ifp != ifp)
2809 lst[n++] = bp->bif_dlt;
2811 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2818 * Set the data link type of a BPF instance.
2821 bpf_setdlt(struct bpf_d *d, u_int dlt)
2823 int error, opromisc;
2828 MPASS(d->bd_bif != NULL);
2831 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2832 * changed while we hold global lock.
2834 if (d->bd_bif->bif_dlt == dlt)
2837 ifp = d->bd_bif->bif_ifp;
2838 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2839 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2845 opromisc = d->bd_promisc;
2848 error = ifpromisc(bp->bif_ifp, 1);
2850 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2859 bpf_drvinit(void *unused)
2863 sx_init(&bpf_sx, "bpf global lock");
2864 CK_LIST_INIT(&bpf_iflist);
2866 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2867 /* For compatibility */
2868 make_dev_alias(dev, "bpf0");
2872 * Zero out the various packet counters associated with all of the bpf
2873 * descriptors. At some point, we will probably want to get a bit more
2874 * granular and allow the user to specify descriptors to be zeroed.
2877 bpf_zero_counters(void)
2884 * We are protected by global lock here, interfaces and
2885 * descriptors can not be deleted while we hold it.
2887 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2888 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2889 counter_u64_zero(bd->bd_rcount);
2890 counter_u64_zero(bd->bd_dcount);
2891 counter_u64_zero(bd->bd_fcount);
2892 counter_u64_zero(bd->bd_wcount);
2893 counter_u64_zero(bd->bd_wfcount);
2894 counter_u64_zero(bd->bd_zcopy);
2901 * Fill filter statistics
2904 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2908 bzero(d, sizeof(*d));
2909 d->bd_structsize = sizeof(*d);
2910 d->bd_immediate = bd->bd_immediate;
2911 d->bd_promisc = bd->bd_promisc;
2912 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2913 d->bd_direction = bd->bd_direction;
2914 d->bd_feedback = bd->bd_feedback;
2915 d->bd_async = bd->bd_async;
2916 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2917 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2918 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2919 d->bd_sig = bd->bd_sig;
2920 d->bd_slen = bd->bd_slen;
2921 d->bd_hlen = bd->bd_hlen;
2922 d->bd_bufsize = bd->bd_bufsize;
2923 d->bd_pid = bd->bd_pid;
2924 strlcpy(d->bd_ifname,
2925 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2926 d->bd_locked = bd->bd_locked;
2927 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2928 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2929 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2930 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2931 d->bd_bufmode = bd->bd_bufmode;
2935 * Handle `netstat -B' stats request
2938 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2940 static const struct xbpf_d zerostats;
2941 struct xbpf_d *xbdbuf, *xbd, tempstats;
2947 * XXX This is not technically correct. It is possible for non
2948 * privileged users to open bpf devices. It would make sense
2949 * if the users who opened the devices were able to retrieve
2950 * the statistics for them, too.
2952 error = priv_check(req->td, PRIV_NET_BPF);
2956 * Check to see if the user is requesting that the counters be
2957 * zeroed out. Explicitly check that the supplied data is zeroed,
2958 * as we aren't allowing the user to set the counters currently.
2960 if (req->newptr != NULL) {
2961 if (req->newlen != sizeof(tempstats))
2963 memset(&tempstats, 0, sizeof(tempstats));
2964 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2967 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2969 bpf_zero_counters();
2972 if (req->oldptr == NULL)
2973 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2974 if (bpf_bpfd_cnt == 0)
2975 return (SYSCTL_OUT(req, 0, 0));
2976 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2978 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2980 free(xbdbuf, M_BPF);
2984 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2985 /* Send writers-only first */
2986 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2987 xbd = &xbdbuf[index++];
2988 bpfstats_fill_xbpf(xbd, bd);
2990 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2991 xbd = &xbdbuf[index++];
2992 bpfstats_fill_xbpf(xbd, bd);
2996 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2997 free(xbdbuf, M_BPF);
3001 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3003 #else /* !DEV_BPF && !NETGRAPH_BPF */
3006 * NOP stubs to allow bpf-using drivers to load and function.
3008 * A 'better' implementation would allow the core bpf functionality
3009 * to be loaded at runtime.
3013 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3018 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3023 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3028 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3031 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3035 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3038 *driverp = (struct bpf_if *)&dead_bpf_if;
3042 bpfdetach(struct ifnet *ifp)
3047 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3049 return -1; /* "no filter" behaviour */
3053 bpf_validate(const struct bpf_insn *f, int len)
3055 return 0; /* false */
3058 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3062 bpf_show_bpf_if(struct bpf_if *bpf_if)
3067 db_printf("%p:\n", bpf_if);
3068 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3069 /* bif_ext.bif_next */
3070 /* bif_ext.bif_dlist */
3071 BPF_DB_PRINTF("%#x", bif_dlt);
3072 BPF_DB_PRINTF("%u", bif_hdrlen);
3074 BPF_DB_PRINTF("%p", bif_ifp);
3075 BPF_DB_PRINTF("%p", bif_bpf);
3076 BPF_DB_PRINTF("%u", bif_refcnt);
3079 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3083 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3087 bpf_show_bpf_if((struct bpf_if *)addr);