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>
43 #include "opt_netgraph.h"
45 #include <sys/param.h>
47 #include <sys/eventhandler.h>
48 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
54 #include <sys/mutex.h>
58 #include <sys/signalvar.h>
59 #include <sys/filio.h>
60 #include <sys/sockio.h>
61 #include <sys/ttycom.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
66 #include <sys/event.h>
71 #include <sys/socket.h>
78 #include <net/if_var.h>
79 #include <net/if_vlan_var.h>
80 #include <net/if_dl.h>
82 #include <net/bpf_buffer.h>
84 #include <net/bpf_jitter.h>
86 #include <net/bpf_zerocopy.h>
87 #include <net/bpfdesc.h>
88 #include <net/route.h>
91 #include <netinet/in.h>
92 #include <netinet/if_ether.h>
93 #include <sys/kernel.h>
94 #include <sys/sysctl.h>
96 #include <net80211/ieee80211_freebsd.h>
98 #include <security/mac/mac_framework.h>
100 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
102 static struct bpf_if_ext dead_bpf_if = {
103 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
107 #define bif_next bif_ext.bif_next
108 #define bif_dlist bif_ext.bif_dlist
109 struct bpf_if_ext bif_ext; /* public members */
110 u_int bif_dlt; /* link layer type */
111 u_int bif_hdrlen; /* length of link header */
112 struct bpfd_list bif_wlist; /* writer-only list */
113 struct ifnet *bif_ifp; /* corresponding interface */
114 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
115 volatile u_int bif_refcnt;
116 struct epoch_context epoch_ctx;
119 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
121 struct bpf_program_buffer {
122 struct epoch_context epoch_ctx;
124 bpf_jit_filter *func;
129 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
131 #define PRINET 26 /* interruptible */
132 #define BPF_PRIO_MAX 7
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);
972 /* Disable VLAN pcp tagging. */
979 * bpfread - read next chunk of packets from buffers
982 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
989 error = devfs_get_cdevpriv((void **)&d);
994 * Restrict application to use a buffer the same size as
997 if (uio->uio_resid != d->bd_bufsize)
1000 non_block = ((ioflag & O_NONBLOCK) != 0);
1003 BPF_PID_REFRESH_CUR(d);
1004 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1006 return (EOPNOTSUPP);
1008 if (d->bd_state == BPF_WAITING)
1009 callout_stop(&d->bd_callout);
1010 timed_out = (d->bd_state == BPF_TIMED_OUT);
1011 d->bd_state = BPF_IDLE;
1012 while (d->bd_hbuf_in_use) {
1013 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1014 PRINET|PCATCH, "bd_hbuf", 0);
1021 * If the hold buffer is empty, then do a timed sleep, which
1022 * ends when the timeout expires or when enough packets
1023 * have arrived to fill the store buffer.
1025 while (d->bd_hbuf == NULL) {
1026 if (d->bd_slen != 0) {
1028 * A packet(s) either arrived since the previous
1029 * read or arrived while we were asleep.
1031 if (d->bd_immediate || non_block || timed_out) {
1033 * Rotate the buffers and return what's here
1034 * if we are in immediate mode, non-blocking
1035 * flag is set, or this descriptor timed out.
1043 * No data is available, check to see if the bpf device
1044 * is still pointed at a real interface. If not, return
1045 * ENXIO so that the userland process knows to rebind
1046 * it before using it again.
1048 if (d->bd_bif == NULL) {
1055 return (EWOULDBLOCK);
1057 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1058 "bpf", d->bd_rtout);
1059 if (error == EINTR || error == ERESTART) {
1063 if (error == EWOULDBLOCK) {
1065 * On a timeout, return what's in the buffer,
1066 * which may be nothing. If there is something
1067 * in the store buffer, we can rotate the buffers.
1071 * We filled up the buffer in between
1072 * getting the timeout and arriving
1073 * here, so we don't need to rotate.
1077 if (d->bd_slen == 0) {
1086 * At this point, we know we have something in the hold slot.
1088 d->bd_hbuf_in_use = 1;
1092 * Move data from hold buffer into user space.
1093 * We know the entire buffer is transferred since
1094 * we checked above that the read buffer is bpf_bufsize bytes.
1096 * We do not have to worry about simultaneous reads because
1097 * we waited for sole access to the hold buffer above.
1099 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1102 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1103 d->bd_fbuf = d->bd_hbuf;
1106 bpf_buf_reclaimed(d);
1107 d->bd_hbuf_in_use = 0;
1108 wakeup(&d->bd_hbuf_in_use);
1115 * If there are processes sleeping on this descriptor, wake them up.
1117 static __inline void
1118 bpf_wakeup(struct bpf_d *d)
1121 BPFD_LOCK_ASSERT(d);
1122 if (d->bd_state == BPF_WAITING) {
1123 callout_stop(&d->bd_callout);
1124 d->bd_state = BPF_IDLE;
1127 if (d->bd_async && d->bd_sig && d->bd_sigio)
1128 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1130 selwakeuppri(&d->bd_sel, PRINET);
1131 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1135 bpf_timed_out(void *arg)
1137 struct bpf_d *d = (struct bpf_d *)arg;
1139 BPFD_LOCK_ASSERT(d);
1141 if (callout_pending(&d->bd_callout) ||
1142 !callout_active(&d->bd_callout))
1144 if (d->bd_state == BPF_WAITING) {
1145 d->bd_state = BPF_TIMED_OUT;
1146 if (d->bd_slen != 0)
1152 bpf_ready(struct bpf_d *d)
1155 BPFD_LOCK_ASSERT(d);
1157 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1159 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1166 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1169 struct sockaddr dst;
1170 struct epoch_tracker et;
1174 struct mbuf *m, *mc;
1177 error = devfs_get_cdevpriv((void **)&d);
1181 NET_EPOCH_ENTER(et);
1183 BPF_PID_REFRESH_CUR(d);
1184 counter_u64_add(d->bd_wcount, 1);
1185 if ((bp = d->bd_bif) == NULL) {
1191 if ((ifp->if_flags & IFF_UP) == 0) {
1196 if (uio->uio_resid == 0)
1199 bzero(&dst, sizeof(dst));
1204 * Take extra reference, unlock d and exit from epoch section,
1205 * since bpf_movein() can sleep.
1211 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1212 &m, &dst, &hlen, d);
1215 counter_u64_add(d->bd_wdcount, 1);
1222 * Check that descriptor is still attached to the interface.
1223 * This can happen on bpfdetach(). To avoid access to detached
1224 * ifnet, free mbuf and return ENXIO.
1226 if (d->bd_bif == NULL) {
1227 counter_u64_add(d->bd_wdcount, 1);
1233 counter_u64_add(d->bd_wfcount, 1);
1235 dst.sa_family = pseudo_AF_HDRCMPLT;
1237 if (d->bd_feedback) {
1238 mc = m_dup(m, M_NOWAIT);
1240 mc->m_pkthdr.rcvif = ifp;
1241 /* Set M_PROMISC for outgoing packets to be discarded. */
1242 if (d->bd_direction == BPF_D_INOUT)
1243 m->m_flags |= M_PROMISC;
1247 m->m_pkthdr.len -= hlen;
1249 m->m_data += hlen; /* XXX */
1251 CURVNET_SET(ifp->if_vnet);
1253 mac_bpfdesc_create_mbuf(d, m);
1255 mac_bpfdesc_create_mbuf(d, mc);
1258 bzero(&ro, sizeof(ro));
1260 ro.ro_prepend = (u_char *)&dst.sa_data;
1262 ro.ro_flags = RT_HAS_HEADER;
1266 vlan_set_pcp(m, d->bd_pcp);
1268 /* Avoid possible recursion on BPFD_LOCK(). */
1269 NET_EPOCH_ENTER(et);
1271 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1273 counter_u64_add(d->bd_wdcount, 1);
1277 (*ifp->if_input)(ifp, mc);
1287 counter_u64_add(d->bd_wdcount, 1);
1294 * Reset a descriptor by flushing its packet buffer and clearing the receive
1295 * and drop counts. This is doable for kernel-only buffers, but with
1296 * zero-copy buffers, we can't write to (or rotate) buffers that are
1297 * currently owned by userspace. It would be nice if we could encapsulate
1298 * this logic in the buffer code rather than here.
1301 reset_d(struct bpf_d *d)
1304 BPFD_LOCK_ASSERT(d);
1306 while (d->bd_hbuf_in_use)
1307 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1309 if ((d->bd_hbuf != NULL) &&
1310 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1311 /* Free the hold buffer. */
1312 d->bd_fbuf = d->bd_hbuf;
1315 bpf_buf_reclaimed(d);
1317 if (bpf_canwritebuf(d))
1319 counter_u64_zero(d->bd_rcount);
1320 counter_u64_zero(d->bd_dcount);
1321 counter_u64_zero(d->bd_fcount);
1322 counter_u64_zero(d->bd_wcount);
1323 counter_u64_zero(d->bd_wfcount);
1324 counter_u64_zero(d->bd_wdcount);
1325 counter_u64_zero(d->bd_zcopy);
1329 * FIONREAD Check for read packet available.
1330 * BIOCGBLEN Get buffer len [for read()].
1331 * BIOCSETF Set read filter.
1332 * BIOCSETFNR Set read filter without resetting descriptor.
1333 * BIOCSETWF Set write filter.
1334 * BIOCFLUSH Flush read packet buffer.
1335 * BIOCPROMISC Put interface into promiscuous mode.
1336 * BIOCGDLT Get link layer type.
1337 * BIOCGETIF Get interface name.
1338 * BIOCSETIF Set interface.
1339 * BIOCSRTIMEOUT Set read timeout.
1340 * BIOCGRTIMEOUT Get read timeout.
1341 * BIOCGSTATS Get packet stats.
1342 * BIOCIMMEDIATE Set immediate mode.
1343 * BIOCVERSION Get filter language version.
1344 * BIOCGHDRCMPLT Get "header already complete" flag
1345 * BIOCSHDRCMPLT Set "header already complete" flag
1346 * BIOCGDIRECTION Get packet direction flag
1347 * BIOCSDIRECTION Set packet direction flag
1348 * BIOCGTSTAMP Get time stamp format and resolution.
1349 * BIOCSTSTAMP Set time stamp format and resolution.
1350 * BIOCLOCK Set "locked" flag
1351 * BIOCFEEDBACK Set packet feedback mode.
1352 * BIOCSETZBUF Set current zero-copy buffer locations.
1353 * BIOCGETZMAX Get maximum zero-copy buffer size.
1354 * BIOCROTZBUF Force rotation of zero-copy buffer
1355 * BIOCSETBUFMODE Set buffer mode.
1356 * BIOCGETBUFMODE Get current buffer mode.
1357 * BIOCSETVLANPCP Set VLAN PCP tag.
1361 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1367 error = devfs_get_cdevpriv((void **)&d);
1372 * Refresh PID associated with this descriptor.
1375 BPF_PID_REFRESH(d, td);
1376 if (d->bd_state == BPF_WAITING)
1377 callout_stop(&d->bd_callout);
1378 d->bd_state = BPF_IDLE;
1381 if (d->bd_locked == 1) {
1387 #ifdef COMPAT_FREEBSD32
1388 case BIOCGDLTLIST32:
1392 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1393 case BIOCGRTIMEOUT32:
1404 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1405 case BIOCSRTIMEOUT32:
1415 #ifdef COMPAT_FREEBSD32
1417 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1418 * that it will get 32-bit packet headers.
1424 case BIOCGDLTLIST32:
1425 case BIOCGRTIMEOUT32:
1426 case BIOCSRTIMEOUT32:
1427 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1435 CURVNET_SET(TD_TO_VNET(td));
1442 * Check for read packet available.
1450 while (d->bd_hbuf_in_use)
1451 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1452 PRINET, "bd_hbuf", 0);
1462 * Get buffer len [for read()].
1466 *(u_int *)addr = d->bd_bufsize;
1471 * Set buffer length.
1474 error = bpf_ioctl_sblen(d, (u_int *)addr);
1478 * Set link layer read filter.
1483 #ifdef COMPAT_FREEBSD32
1488 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1492 * Flush read packet buffer.
1501 * Put interface into promiscuous mode.
1505 if (d->bd_bif == NULL) {
1507 * No interface attached yet.
1510 } else if (d->bd_promisc == 0) {
1511 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1519 * Get current data link type.
1523 if (d->bd_bif == NULL)
1526 *(u_int *)addr = d->bd_bif->bif_dlt;
1531 * Get a list of supported data link types.
1533 #ifdef COMPAT_FREEBSD32
1534 case BIOCGDLTLIST32:
1536 struct bpf_dltlist32 *list32;
1537 struct bpf_dltlist dltlist;
1539 list32 = (struct bpf_dltlist32 *)addr;
1540 dltlist.bfl_len = list32->bfl_len;
1541 dltlist.bfl_list = PTRIN(list32->bfl_list);
1543 if (d->bd_bif == NULL)
1546 error = bpf_getdltlist(d, &dltlist);
1548 list32->bfl_len = dltlist.bfl_len;
1557 if (d->bd_bif == NULL)
1560 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1565 * Set data link type.
1569 if (d->bd_bif == NULL)
1572 error = bpf_setdlt(d, *(u_int *)addr);
1577 * Get interface name.
1581 if (d->bd_bif == NULL)
1584 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1585 struct ifreq *const ifr = (struct ifreq *)addr;
1587 strlcpy(ifr->ifr_name, ifp->if_xname,
1588 sizeof(ifr->ifr_name));
1598 int alloc_buf, size;
1601 * Behavior here depends on the buffering model. If
1602 * we're using kernel memory buffers, then we can
1603 * allocate them here. If we're using zero-copy,
1604 * then the user process must have registered buffers
1605 * by the time we get here.
1609 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1614 size = d->bd_bufsize;
1615 error = bpf_buffer_ioctl_sblen(d, &size);
1620 error = bpf_setif(d, (struct ifreq *)addr);
1629 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1630 case BIOCSRTIMEOUT32:
1633 struct timeval *tv = (struct timeval *)addr;
1634 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1635 struct timeval32 *tv32;
1636 struct timeval tv64;
1638 if (cmd == BIOCSRTIMEOUT32) {
1639 tv32 = (struct timeval32 *)addr;
1641 tv->tv_sec = tv32->tv_sec;
1642 tv->tv_usec = tv32->tv_usec;
1645 tv = (struct timeval *)addr;
1648 * Subtract 1 tick from tvtohz() since this isn't
1651 if ((error = itimerfix(tv)) == 0)
1652 d->bd_rtout = tvtohz(tv) - 1;
1660 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1661 case BIOCGRTIMEOUT32:
1665 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1666 struct timeval32 *tv32;
1667 struct timeval tv64;
1669 if (cmd == BIOCGRTIMEOUT32)
1673 tv = (struct timeval *)addr;
1675 tv->tv_sec = d->bd_rtout / hz;
1676 tv->tv_usec = (d->bd_rtout % hz) * tick;
1677 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1678 if (cmd == BIOCGRTIMEOUT32) {
1679 tv32 = (struct timeval32 *)addr;
1680 tv32->tv_sec = tv->tv_sec;
1681 tv32->tv_usec = tv->tv_usec;
1693 struct bpf_stat *bs = (struct bpf_stat *)addr;
1695 /* XXXCSJP overflow */
1696 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1697 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1702 * Set immediate mode.
1706 d->bd_immediate = *(u_int *)addr;
1712 struct bpf_version *bv = (struct bpf_version *)addr;
1714 bv->bv_major = BPF_MAJOR_VERSION;
1715 bv->bv_minor = BPF_MINOR_VERSION;
1720 * Get "header already complete" flag
1724 *(u_int *)addr = d->bd_hdrcmplt;
1729 * Set "header already complete" flag
1733 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1738 * Get packet direction flag
1740 case BIOCGDIRECTION:
1742 *(u_int *)addr = d->bd_direction;
1747 * Set packet direction flag
1749 case BIOCSDIRECTION:
1753 direction = *(u_int *)addr;
1754 switch (direction) {
1759 d->bd_direction = direction;
1769 * Get packet timestamp format and resolution.
1773 *(u_int *)addr = d->bd_tstamp;
1778 * Set packet timestamp format and resolution.
1784 func = *(u_int *)addr;
1785 if (BPF_T_VALID(func))
1786 d->bd_tstamp = func;
1794 d->bd_feedback = *(u_int *)addr;
1804 case FIONBIO: /* Non-blocking I/O */
1807 case FIOASYNC: /* Send signal on receive packets */
1809 d->bd_async = *(int *)addr;
1815 * XXX: Add some sort of locking here?
1816 * fsetown() can sleep.
1818 error = fsetown(*(int *)addr, &d->bd_sigio);
1823 *(int *)addr = fgetown(&d->bd_sigio);
1827 /* This is deprecated, FIOSETOWN should be used instead. */
1829 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1832 /* This is deprecated, FIOGETOWN should be used instead. */
1834 *(int *)addr = -fgetown(&d->bd_sigio);
1837 case BIOCSRSIG: /* Set receive signal */
1841 sig = *(u_int *)addr;
1854 *(u_int *)addr = d->bd_sig;
1858 case BIOCGETBUFMODE:
1860 *(u_int *)addr = d->bd_bufmode;
1864 case BIOCSETBUFMODE:
1866 * Allow the buffering mode to be changed as long as we
1867 * haven't yet committed to a particular mode. Our
1868 * definition of commitment, for now, is whether or not a
1869 * buffer has been allocated or an interface attached, since
1870 * that's the point where things get tricky.
1872 switch (*(u_int *)addr) {
1873 case BPF_BUFMODE_BUFFER:
1876 case BPF_BUFMODE_ZBUF:
1877 if (bpf_zerocopy_enable)
1887 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1888 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1893 d->bd_bufmode = *(u_int *)addr;
1898 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1902 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1906 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1909 case BIOCSETVLANPCP:
1913 pcp = *(u_int *)addr;
1914 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1927 * Set d's packet filter program to fp. If this file already has a filter,
1928 * free it and replace it. Returns EINVAL for bogus requests.
1930 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1934 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1936 #ifdef COMPAT_FREEBSD32
1937 struct bpf_program fp_swab;
1938 struct bpf_program32 *fp32;
1940 struct bpf_program_buffer *fcode;
1941 struct bpf_insn *filter;
1943 bpf_jit_filter *jfunc;
1949 #ifdef COMPAT_FREEBSD32
1954 fp32 = (struct bpf_program32 *)fp;
1955 fp_swab.bf_len = fp32->bf_len;
1957 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1976 * Check new filter validness before acquiring any locks.
1977 * Allocate memory for new filter, if needed.
1980 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1982 size = flen * sizeof(*fp->bf_insns);
1984 /* We're setting up new filter. Copy and check actual data. */
1985 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1986 filter = (struct bpf_insn *)fcode->buffer;
1987 if (copyin(fp->bf_insns, filter, size) != 0 ||
1988 !bpf_validate(filter, flen)) {
1993 if (cmd != BIOCSETWF) {
1995 * Filter is copied inside fcode and is
1998 jfunc = bpf_jitter(filter, flen);
2003 track_event = false;
2008 /* Set up new filter. */
2009 if (cmd == BIOCSETWF) {
2010 if (d->bd_wfilter != NULL) {
2011 fcode = __containerof((void *)d->bd_wfilter,
2012 struct bpf_program_buffer, buffer);
2017 d->bd_wfilter = filter;
2019 if (d->bd_rfilter != NULL) {
2020 fcode = __containerof((void *)d->bd_rfilter,
2021 struct bpf_program_buffer, buffer);
2023 fcode->func = d->bd_bfilter;
2026 d->bd_rfilter = filter;
2028 d->bd_bfilter = jfunc;
2030 if (cmd == BIOCSETF)
2033 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2035 * Filter can be set several times without
2036 * specifying interface. In this case just mark d
2040 if (d->bd_bif != NULL) {
2042 * Remove descriptor from writers-only list
2043 * and add it to active readers list.
2045 CK_LIST_REMOVE(d, bd_next);
2046 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2049 "%s: upgrade required by pid %d",
2050 __func__, d->bd_pid);
2058 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2061 EVENTHANDLER_INVOKE(bpf_track,
2062 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2069 * Detach a file from its current interface (if attached at all) and attach
2070 * to the interface indicated by the name stored in ifr.
2071 * Return an errno or 0.
2074 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2077 struct ifnet *theywant;
2081 theywant = ifunit(ifr->ifr_name);
2082 if (theywant == NULL || theywant->if_bpf == NULL)
2085 bp = theywant->if_bpf;
2087 * At this point, we expect the buffer is already allocated. If not,
2090 switch (d->bd_bufmode) {
2091 case BPF_BUFMODE_BUFFER:
2092 case BPF_BUFMODE_ZBUF:
2093 if (d->bd_sbuf == NULL)
2098 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2100 if (bp != d->bd_bif)
2111 * Support for select() and poll() system calls
2113 * Return true iff the specific operation will not block indefinitely.
2114 * Otherwise, return false but make a note that a selwakeup() must be done.
2117 bpfpoll(struct cdev *dev, int events, struct thread *td)
2122 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2124 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2127 * Refresh PID associated with this descriptor.
2129 revents = events & (POLLOUT | POLLWRNORM);
2131 BPF_PID_REFRESH(d, td);
2132 if (events & (POLLIN | POLLRDNORM)) {
2134 revents |= events & (POLLIN | POLLRDNORM);
2136 selrecord(td, &d->bd_sel);
2137 /* Start the read timeout if necessary. */
2138 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2139 callout_reset(&d->bd_callout, d->bd_rtout,
2141 d->bd_state = BPF_WAITING;
2150 * Support for kevent() system call. Register EVFILT_READ filters and
2151 * reject all others.
2154 bpfkqfilter(struct cdev *dev, struct knote *kn)
2158 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2159 kn->kn_filter != EVFILT_READ)
2163 * Refresh PID associated with this descriptor.
2166 BPF_PID_REFRESH_CUR(d);
2167 kn->kn_fop = &bpfread_filtops;
2169 knlist_add(&d->bd_sel.si_note, kn, 1);
2176 filt_bpfdetach(struct knote *kn)
2178 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2180 knlist_remove(&d->bd_sel.si_note, kn, 0);
2184 filt_bpfread(struct knote *kn, long hint)
2186 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2189 BPFD_LOCK_ASSERT(d);
2190 ready = bpf_ready(d);
2192 kn->kn_data = d->bd_slen;
2194 * Ignore the hold buffer if it is being copied to user space.
2196 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2197 kn->kn_data += d->bd_hlen;
2198 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2199 callout_reset(&d->bd_callout, d->bd_rtout,
2201 d->bd_state = BPF_WAITING;
2207 #define BPF_TSTAMP_NONE 0
2208 #define BPF_TSTAMP_FAST 1
2209 #define BPF_TSTAMP_NORMAL 2
2210 #define BPF_TSTAMP_EXTERN 3
2213 bpf_ts_quality(int tstype)
2216 if (tstype == BPF_T_NONE)
2217 return (BPF_TSTAMP_NONE);
2218 if ((tstype & BPF_T_FAST) != 0)
2219 return (BPF_TSTAMP_FAST);
2221 return (BPF_TSTAMP_NORMAL);
2225 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2230 quality = bpf_ts_quality(tstype);
2231 if (quality == BPF_TSTAMP_NONE)
2235 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2237 *bt = *(struct bintime *)(tag + 1);
2238 return (BPF_TSTAMP_EXTERN);
2241 if (quality == BPF_TSTAMP_NORMAL)
2250 * Incoming linkage from device drivers. Process the packet pkt, of length
2251 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2252 * by each process' filter, and if accepted, stashed into the corresponding
2256 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2258 struct epoch_tracker et;
2267 gottime = BPF_TSTAMP_NONE;
2268 NET_EPOCH_ENTER(et);
2269 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2270 counter_u64_add(d->bd_rcount, 1);
2272 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2273 * is no way for the caller to indiciate to us whether this
2274 * packet is inbound or outbound. In the bpf_mtap() routines,
2275 * we use the interface pointers on the mbuf to figure it out.
2278 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2280 slen = (*(bf->func))(pkt, pktlen, pktlen);
2283 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2286 * Filter matches. Let's to acquire write lock.
2289 counter_u64_add(d->bd_fcount, 1);
2290 if (gottime < bpf_ts_quality(d->bd_tstamp))
2291 gottime = bpf_gettime(&bt, d->bd_tstamp,
2294 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2296 catchpacket(d, pkt, pktlen, slen,
2297 bpf_append_bytes, &bt);
2304 #define BPF_CHECK_DIRECTION(d, r, i) \
2305 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2306 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2309 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2310 * Locking model is explained in bpf_tap().
2313 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2315 struct epoch_tracker et;
2324 /* Skip outgoing duplicate packets. */
2325 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2326 m->m_flags &= ~M_PROMISC;
2330 pktlen = m_length(m, NULL);
2331 gottime = BPF_TSTAMP_NONE;
2333 NET_EPOCH_ENTER(et);
2334 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2335 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2337 counter_u64_add(d->bd_rcount, 1);
2339 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2340 /* XXX We cannot handle multiple mbufs. */
2341 if (bf != NULL && m->m_next == NULL)
2342 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2346 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2350 counter_u64_add(d->bd_fcount, 1);
2351 if (gottime < bpf_ts_quality(d->bd_tstamp))
2352 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2354 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2356 catchpacket(d, (u_char *)m, pktlen, slen,
2357 bpf_append_mbuf, &bt);
2365 * Incoming linkage from device drivers, when packet is in
2366 * an mbuf chain and to be prepended by a contiguous header.
2369 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2371 struct epoch_tracker et;
2378 /* Skip outgoing duplicate packets. */
2379 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2380 m->m_flags &= ~M_PROMISC;
2384 pktlen = m_length(m, NULL);
2386 * Craft on-stack mbuf suitable for passing to bpf_filter.
2387 * Note that we cut corners here; we only setup what's
2388 * absolutely needed--this mbuf should never go anywhere else.
2396 gottime = BPF_TSTAMP_NONE;
2398 NET_EPOCH_ENTER(et);
2399 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2400 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2402 counter_u64_add(d->bd_rcount, 1);
2403 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2407 counter_u64_add(d->bd_fcount, 1);
2408 if (gottime < bpf_ts_quality(d->bd_tstamp))
2409 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2411 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2413 catchpacket(d, (u_char *)&mb, pktlen, slen,
2414 bpf_append_mbuf, &bt);
2421 #undef BPF_CHECK_DIRECTION
2422 #undef BPF_TSTAMP_NONE
2423 #undef BPF_TSTAMP_FAST
2424 #undef BPF_TSTAMP_NORMAL
2425 #undef BPF_TSTAMP_EXTERN
2428 bpf_hdrlen(struct bpf_d *d)
2432 hdrlen = d->bd_bif->bif_hdrlen;
2433 #ifndef BURN_BRIDGES
2434 if (d->bd_tstamp == BPF_T_NONE ||
2435 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2436 #ifdef COMPAT_FREEBSD32
2438 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2441 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2444 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2445 #ifdef COMPAT_FREEBSD32
2447 hdrlen = BPF_WORDALIGN32(hdrlen);
2450 hdrlen = BPF_WORDALIGN(hdrlen);
2452 return (hdrlen - d->bd_bif->bif_hdrlen);
2456 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2458 struct bintime bt2, boottimebin;
2460 struct timespec tsn;
2462 if ((tstype & BPF_T_MONOTONIC) == 0) {
2464 getboottimebin(&boottimebin);
2465 bintime_add(&bt2, &boottimebin);
2468 switch (BPF_T_FORMAT(tstype)) {
2469 case BPF_T_MICROTIME:
2470 bintime2timeval(bt, &tsm);
2471 ts->bt_sec = tsm.tv_sec;
2472 ts->bt_frac = tsm.tv_usec;
2474 case BPF_T_NANOTIME:
2475 bintime2timespec(bt, &tsn);
2476 ts->bt_sec = tsn.tv_sec;
2477 ts->bt_frac = tsn.tv_nsec;
2480 ts->bt_sec = bt->sec;
2481 ts->bt_frac = bt->frac;
2487 * Move the packet data from interface memory (pkt) into the
2488 * store buffer. "cpfn" is the routine called to do the actual data
2489 * transfer. bcopy is passed in to copy contiguous chunks, while
2490 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2491 * pkt is really an mbuf.
2494 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2495 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2498 static char zeroes[BPF_ALIGNMENT];
2499 struct bpf_xhdr hdr;
2500 #ifndef BURN_BRIDGES
2501 struct bpf_hdr hdr_old;
2502 #ifdef COMPAT_FREEBSD32
2503 struct bpf_hdr32 hdr32_old;
2506 int caplen, curlen, hdrlen, pad, totlen;
2511 BPFD_LOCK_ASSERT(d);
2512 if (d->bd_bif == NULL) {
2513 /* Descriptor was detached in concurrent thread */
2514 counter_u64_add(d->bd_dcount, 1);
2519 * Detect whether user space has released a buffer back to us, and if
2520 * so, move it from being a hold buffer to a free buffer. This may
2521 * not be the best place to do it (for example, we might only want to
2522 * run this check if we need the space), but for now it's a reliable
2525 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2526 d->bd_fbuf = d->bd_hbuf;
2529 bpf_buf_reclaimed(d);
2533 * Figure out how many bytes to move. If the packet is
2534 * greater or equal to the snapshot length, transfer that
2535 * much. Otherwise, transfer the whole packet (unless
2536 * we hit the buffer size limit).
2538 hdrlen = bpf_hdrlen(d);
2539 totlen = hdrlen + min(snaplen, pktlen);
2540 if (totlen > d->bd_bufsize)
2541 totlen = d->bd_bufsize;
2544 * Round up the end of the previous packet to the next longword.
2546 * Drop the packet if there's no room and no hope of room
2547 * If the packet would overflow the storage buffer or the storage
2548 * buffer is considered immutable by the buffer model, try to rotate
2549 * the buffer and wakeup pending processes.
2551 #ifdef COMPAT_FREEBSD32
2553 curlen = BPF_WORDALIGN32(d->bd_slen);
2556 curlen = BPF_WORDALIGN(d->bd_slen);
2557 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2558 if (d->bd_fbuf == NULL) {
2560 * There's no room in the store buffer, and no
2561 * prospect of room, so drop the packet. Notify the
2565 counter_u64_add(d->bd_dcount, 1);
2568 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2573 if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2575 * Immediate mode is set, or the read timeout has
2576 * already expired during a select call. A packet
2577 * arrived, so the reader should be woken up.
2581 pad = curlen - d->bd_slen;
2582 KASSERT(pad >= 0 && pad <= sizeof(zeroes),
2583 ("%s: invalid pad byte count %d", __func__, pad));
2585 /* Zero pad bytes. */
2586 bpf_append_bytes(d, d->bd_sbuf, d->bd_slen, zeroes,
2591 caplen = totlen - hdrlen;
2592 tstype = d->bd_tstamp;
2593 do_timestamp = tstype != BPF_T_NONE;
2594 #ifndef BURN_BRIDGES
2595 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2598 bpf_bintime2ts(bt, &ts, tstype);
2599 #ifdef COMPAT_FREEBSD32
2600 if (d->bd_compat32) {
2601 bzero(&hdr32_old, sizeof(hdr32_old));
2603 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2604 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2606 hdr32_old.bh_datalen = pktlen;
2607 hdr32_old.bh_hdrlen = hdrlen;
2608 hdr32_old.bh_caplen = caplen;
2609 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2614 bzero(&hdr_old, sizeof(hdr_old));
2616 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2617 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2619 hdr_old.bh_datalen = pktlen;
2620 hdr_old.bh_hdrlen = hdrlen;
2621 hdr_old.bh_caplen = caplen;
2622 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2629 * Append the bpf header. Note we append the actual header size, but
2630 * move forward the length of the header plus padding.
2632 bzero(&hdr, sizeof(hdr));
2634 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2635 hdr.bh_datalen = pktlen;
2636 hdr.bh_hdrlen = hdrlen;
2637 hdr.bh_caplen = caplen;
2638 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2641 * Copy the packet data into the store buffer and update its length.
2643 #ifndef BURN_BRIDGES
2646 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2647 d->bd_slen = curlen + totlen;
2654 * Free buffers currently in use by a descriptor.
2658 bpfd_free(epoch_context_t ctx)
2661 struct bpf_program_buffer *p;
2664 * We don't need to lock out interrupts since this descriptor has
2665 * been detached from its interface and it yet hasn't been marked
2668 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2670 if (d->bd_rfilter != NULL) {
2671 p = __containerof((void *)d->bd_rfilter,
2672 struct bpf_program_buffer, buffer);
2674 p->func = d->bd_bfilter;
2676 bpf_program_buffer_free(&p->epoch_ctx);
2678 if (d->bd_wfilter != NULL) {
2679 p = __containerof((void *)d->bd_wfilter,
2680 struct bpf_program_buffer, buffer);
2684 bpf_program_buffer_free(&p->epoch_ctx);
2687 mtx_destroy(&d->bd_lock);
2688 counter_u64_free(d->bd_rcount);
2689 counter_u64_free(d->bd_dcount);
2690 counter_u64_free(d->bd_fcount);
2691 counter_u64_free(d->bd_wcount);
2692 counter_u64_free(d->bd_wfcount);
2693 counter_u64_free(d->bd_wdcount);
2694 counter_u64_free(d->bd_zcopy);
2699 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2700 * fixed size of the link header (variable length headers not yet supported).
2703 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2706 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2710 * Attach an interface to bpf. ifp is a pointer to the structure
2711 * defining the interface to be attached, dlt is the link layer type,
2712 * and hdrlen is the fixed size of the link header (variable length
2713 * headers are not yet supporrted).
2716 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2717 struct bpf_if **driverp)
2721 KASSERT(*driverp == NULL,
2722 ("bpfattach2: driverp already initialized"));
2724 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2726 CK_LIST_INIT(&bp->bif_dlist);
2727 CK_LIST_INIT(&bp->bif_wlist);
2730 bp->bif_hdrlen = hdrlen;
2731 bp->bif_bpf = driverp;
2735 * Reference ifnet pointer, so it won't freed until
2740 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2743 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2744 if_printf(ifp, "bpf attached\n");
2749 * When moving interfaces between vnet instances we need a way to
2750 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2751 * after the vmove. We unfortunately have no device driver infrastructure
2752 * to query the interface for these values after creation/attach, thus
2753 * add this as a workaround.
2756 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2761 if (bif_dlt == NULL && bif_hdrlen == NULL)
2764 if (bif_dlt != NULL)
2765 *bif_dlt = bp->bif_dlt;
2766 if (bif_hdrlen != NULL)
2767 *bif_hdrlen = bp->bif_hdrlen;
2774 * Detach bpf from an interface. This involves detaching each descriptor
2775 * associated with the interface. Notify each descriptor as it's detached
2776 * so that any sleepers wake up and get ENXIO.
2779 bpfdetach(struct ifnet *ifp)
2781 struct bpf_if *bp, *bp_temp;
2785 /* Find all bpf_if struct's which reference ifp and detach them. */
2786 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2787 if (ifp != bp->bif_ifp)
2790 CK_LIST_REMOVE(bp, bif_next);
2791 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2794 "%s: sheduling free for encap %d (%p) for if %p",
2795 __func__, bp->bif_dlt, bp, ifp);
2797 /* Detach common descriptors */
2798 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2799 bpf_detachd_locked(d, true);
2802 /* Detach writer-only descriptors */
2803 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2804 bpf_detachd_locked(d, true);
2812 * Get a list of available data link type of the interface.
2815 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2824 ifp = d->bd_bif->bif_ifp;
2826 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2827 if (bp->bif_ifp == ifp)
2830 if (bfl->bfl_list == NULL) {
2834 if (n1 > bfl->bfl_len)
2837 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2839 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2840 if (bp->bif_ifp != ifp)
2842 lst[n++] = bp->bif_dlt;
2844 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2851 * Set the data link type of a BPF instance.
2854 bpf_setdlt(struct bpf_d *d, u_int dlt)
2856 int error, opromisc;
2861 MPASS(d->bd_bif != NULL);
2864 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2865 * changed while we hold global lock.
2867 if (d->bd_bif->bif_dlt == dlt)
2870 ifp = d->bd_bif->bif_ifp;
2871 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2872 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2878 opromisc = d->bd_promisc;
2881 error = ifpromisc(bp->bif_ifp, 1);
2883 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2892 bpf_drvinit(void *unused)
2896 sx_init(&bpf_sx, "bpf global lock");
2897 CK_LIST_INIT(&bpf_iflist);
2899 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2900 /* For compatibility */
2901 make_dev_alias(dev, "bpf0");
2905 * Zero out the various packet counters associated with all of the bpf
2906 * descriptors. At some point, we will probably want to get a bit more
2907 * granular and allow the user to specify descriptors to be zeroed.
2910 bpf_zero_counters(void)
2917 * We are protected by global lock here, interfaces and
2918 * descriptors can not be deleted while we hold it.
2920 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2921 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2922 counter_u64_zero(bd->bd_rcount);
2923 counter_u64_zero(bd->bd_dcount);
2924 counter_u64_zero(bd->bd_fcount);
2925 counter_u64_zero(bd->bd_wcount);
2926 counter_u64_zero(bd->bd_wfcount);
2927 counter_u64_zero(bd->bd_zcopy);
2934 * Fill filter statistics
2937 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2941 bzero(d, sizeof(*d));
2942 d->bd_structsize = sizeof(*d);
2943 d->bd_immediate = bd->bd_immediate;
2944 d->bd_promisc = bd->bd_promisc;
2945 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2946 d->bd_direction = bd->bd_direction;
2947 d->bd_feedback = bd->bd_feedback;
2948 d->bd_async = bd->bd_async;
2949 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
2950 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
2951 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
2952 d->bd_sig = bd->bd_sig;
2953 d->bd_slen = bd->bd_slen;
2954 d->bd_hlen = bd->bd_hlen;
2955 d->bd_bufsize = bd->bd_bufsize;
2956 d->bd_pid = bd->bd_pid;
2957 strlcpy(d->bd_ifname,
2958 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2959 d->bd_locked = bd->bd_locked;
2960 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
2961 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
2962 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
2963 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
2964 d->bd_bufmode = bd->bd_bufmode;
2968 * Handle `netstat -B' stats request
2971 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2973 static const struct xbpf_d zerostats;
2974 struct xbpf_d *xbdbuf, *xbd, tempstats;
2980 * XXX This is not technically correct. It is possible for non
2981 * privileged users to open bpf devices. It would make sense
2982 * if the users who opened the devices were able to retrieve
2983 * the statistics for them, too.
2985 error = priv_check(req->td, PRIV_NET_BPF);
2989 * Check to see if the user is requesting that the counters be
2990 * zeroed out. Explicitly check that the supplied data is zeroed,
2991 * as we aren't allowing the user to set the counters currently.
2993 if (req->newptr != NULL) {
2994 if (req->newlen != sizeof(tempstats))
2996 memset(&tempstats, 0, sizeof(tempstats));
2997 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3000 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3002 bpf_zero_counters();
3005 if (req->oldptr == NULL)
3006 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3007 if (bpf_bpfd_cnt == 0)
3008 return (SYSCTL_OUT(req, 0, 0));
3009 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3011 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3013 free(xbdbuf, M_BPF);
3017 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3018 /* Send writers-only first */
3019 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3020 xbd = &xbdbuf[index++];
3021 bpfstats_fill_xbpf(xbd, bd);
3023 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3024 xbd = &xbdbuf[index++];
3025 bpfstats_fill_xbpf(xbd, bd);
3029 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3030 free(xbdbuf, M_BPF);
3034 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3036 #else /* !DEV_BPF && !NETGRAPH_BPF */
3039 * NOP stubs to allow bpf-using drivers to load and function.
3041 * A 'better' implementation would allow the core bpf functionality
3042 * to be loaded at runtime.
3046 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3051 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3056 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3061 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3064 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3068 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3071 *driverp = (struct bpf_if *)&dead_bpf_if;
3075 bpfdetach(struct ifnet *ifp)
3080 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3082 return -1; /* "no filter" behaviour */
3086 bpf_validate(const struct bpf_insn *f, int len)
3088 return 0; /* false */
3091 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3095 bpf_show_bpf_if(struct bpf_if *bpf_if)
3100 db_printf("%p:\n", bpf_if);
3101 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3102 /* bif_ext.bif_next */
3103 /* bif_ext.bif_dlist */
3104 BPF_DB_PRINTF("%#x", bif_dlt);
3105 BPF_DB_PRINTF("%u", bif_hdrlen);
3107 BPF_DB_PRINTF("%p", bif_ifp);
3108 BPF_DB_PRINTF("%p", bif_bpf);
3109 BPF_DB_PRINTF("%u", bif_refcnt);
3112 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3116 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3120 bpf_show_bpf_if((struct bpf_if *)addr);