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
38 #include <sys/cdefs.h>
41 #include "opt_netgraph.h"
43 #include <sys/param.h>
45 #include <sys/eventhandler.h>
46 #include <sys/fcntl.h>
50 #include <sys/malloc.h>
52 #include <sys/mutex.h>
56 #include <sys/signalvar.h>
57 #include <sys/filio.h>
58 #include <sys/sockio.h>
59 #include <sys/ttycom.h>
61 #include <sys/sysent.h>
62 #include <sys/systm.h>
64 #include <sys/event.h>
69 #include <sys/socket.h>
76 #include <net/if_var.h>
77 #include <net/if_private.h>
78 #include <net/if_vlan_var.h>
79 #include <net/if_dl.h>
81 #include <net/bpf_buffer.h>
83 #include <net/bpf_jitter.h>
85 #include <net/bpf_zerocopy.h>
86 #include <net/bpfdesc.h>
87 #include <net/route.h>
90 #include <netinet/in.h>
91 #include <netinet/if_ether.h>
92 #include <sys/kernel.h>
93 #include <sys/sysctl.h>
95 #include <net80211/ieee80211_freebsd.h>
97 #include <security/mac/mac_framework.h>
99 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
101 static const struct bpf_if_ext dead_bpf_if = {
102 .bif_dlist = CK_LIST_HEAD_INITIALIZER()
106 #define bif_next bif_ext.bif_next
107 #define bif_dlist bif_ext.bif_dlist
108 struct bpf_if_ext bif_ext; /* public members */
109 u_int bif_dlt; /* link layer type */
110 u_int bif_hdrlen; /* length of link header */
111 struct bpfd_list bif_wlist; /* writer-only list */
112 struct ifnet *bif_ifp; /* corresponding interface */
113 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
114 volatile u_int bif_refcnt;
115 struct epoch_context epoch_ctx;
118 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
120 struct bpf_program_buffer {
121 struct epoch_context epoch_ctx;
123 bpf_jit_filter *func;
128 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
130 #define PRINET 26 /* interruptible */
131 #define BPF_PRIO_MAX 7
133 #define SIZEOF_BPF_HDR(type) \
134 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
136 #ifdef COMPAT_FREEBSD32
137 #include <sys/mount.h>
138 #include <compat/freebsd32/freebsd32.h>
139 #define BPF_ALIGNMENT32 sizeof(int32_t)
140 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
144 * 32-bit version of structure prepended to each packet. We use this header
145 * instead of the standard one for 32-bit streams. We mark the a stream as
146 * 32-bit the first time we see a 32-bit compat ioctl request.
149 struct timeval32 bh_tstamp; /* time stamp */
150 uint32_t bh_caplen; /* length of captured portion */
151 uint32_t bh_datalen; /* original length of packet */
152 uint16_t bh_hdrlen; /* length of bpf header (this struct
153 plus alignment padding) */
157 struct bpf_program32 {
162 struct bpf_dltlist32 {
167 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
168 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
169 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
170 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
171 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
172 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
175 #define BPF_LOCK() sx_xlock(&bpf_sx)
176 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
177 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
179 * bpf_iflist is a list of BPF interface structures, each corresponding to a
180 * specific DLT. The same network interface might have several BPF interface
181 * structures registered by different layers in the stack (i.e., 802.11
182 * frames, ethernet frames, etc).
184 CK_LIST_HEAD(bpf_iflist, bpf_if);
185 static struct bpf_iflist bpf_iflist;
186 static struct sx bpf_sx; /* bpf global lock */
187 static int bpf_bpfd_cnt;
189 static void bpfif_ref(struct bpf_if *);
190 static void bpfif_rele(struct bpf_if *);
192 static void bpfd_ref(struct bpf_d *);
193 static void bpfd_rele(struct bpf_d *);
194 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
195 static void bpf_detachd(struct bpf_d *);
196 static void bpf_detachd_locked(struct bpf_d *, bool);
197 static void bpfd_free(epoch_context_t);
198 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
199 struct sockaddr *, int *, struct bpf_d *);
200 static int bpf_setif(struct bpf_d *, struct ifreq *);
201 static void bpf_timed_out(void *);
203 bpf_wakeup(struct bpf_d *);
204 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
205 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
207 static void reset_d(struct bpf_d *);
208 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
209 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
210 static int bpf_setdlt(struct bpf_d *, u_int);
211 static void filt_bpfdetach(struct knote *);
212 static int filt_bpfread(struct knote *, long);
213 static int filt_bpfwrite(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,
258 static struct filterops bpfwrite_filtops = {
260 .f_detach = filt_bpfdetach,
261 .f_event = filt_bpfwrite,
265 * LOCKING MODEL USED BY BPF
268 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters,
269 * every bpf_iflist changes, serializes ioctl access to bpf descriptors.
270 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various
271 * structure fields used by bpf_*tap* code.
273 * Lock order: global lock, then descriptor lock.
275 * There are several possible consumers:
277 * 1. The kernel registers interface pointer with bpfattach().
278 * Each call allocates new bpf_if structure, references ifnet pointer
279 * and links bpf_if into bpf_iflist chain. This is protected with global
282 * 2. An userland application uses ioctl() call to bpf_d descriptor.
283 * All such call are serialized with global lock. BPF filters can be
284 * changed, but pointer to old filter will be freed using NET_EPOCH_CALL().
285 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to
286 * filter pointers, even if change will happen during bpf_tap execution.
287 * Destroying of bpf_d descriptor also is doing using NET_EPOCH_CALL().
289 * 3. An userland application can write packets into bpf_d descriptor.
290 * There we need to be sure, that ifnet won't disappear during bpfwrite().
292 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to
293 * bif_dlist is protected with net_epoch_preempt section. So, it should
294 * be safe to make access to bpf_d descriptor inside the section.
296 * 5. The kernel invokes bpfdetach() on interface destroying. All lists
297 * are modified with global lock held and actual free() is done using
302 bpfif_free(epoch_context_t ctx)
306 bp = __containerof(ctx, struct bpf_if, epoch_ctx);
307 if_rele(bp->bif_ifp);
312 bpfif_ref(struct bpf_if *bp)
315 refcount_acquire(&bp->bif_refcnt);
319 bpfif_rele(struct bpf_if *bp)
322 if (!refcount_release(&bp->bif_refcnt))
324 NET_EPOCH_CALL(bpfif_free, &bp->epoch_ctx);
328 bpfd_ref(struct bpf_d *d)
331 refcount_acquire(&d->bd_refcnt);
335 bpfd_rele(struct bpf_d *d)
338 if (!refcount_release(&d->bd_refcnt))
340 NET_EPOCH_CALL(bpfd_free, &d->epoch_ctx);
343 static struct bpf_program_buffer*
344 bpf_program_buffer_alloc(size_t size, int flags)
347 return (malloc(sizeof(struct bpf_program_buffer) + size,
352 bpf_program_buffer_free(epoch_context_t ctx)
354 struct bpf_program_buffer *ptr;
356 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx);
358 if (ptr->func != NULL)
359 bpf_destroy_jit_filter(ptr->func);
365 * Wrapper functions for various buffering methods. If the set of buffer
366 * modes expands, we will probably want to introduce a switch data structure
367 * similar to protosw, et.
370 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
376 switch (d->bd_bufmode) {
377 case BPF_BUFMODE_BUFFER:
378 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
380 case BPF_BUFMODE_ZBUF:
381 counter_u64_add(d->bd_zcopy, 1);
382 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
385 panic("bpf_buf_append_bytes");
390 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
396 switch (d->bd_bufmode) {
397 case BPF_BUFMODE_BUFFER:
398 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
400 case BPF_BUFMODE_ZBUF:
401 counter_u64_add(d->bd_zcopy, 1);
402 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
405 panic("bpf_buf_append_mbuf");
410 * This function gets called when the free buffer is re-assigned.
413 bpf_buf_reclaimed(struct bpf_d *d)
418 switch (d->bd_bufmode) {
419 case BPF_BUFMODE_BUFFER:
422 case BPF_BUFMODE_ZBUF:
423 bpf_zerocopy_buf_reclaimed(d);
427 panic("bpf_buf_reclaimed");
432 * If the buffer mechanism has a way to decide that a held buffer can be made
433 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
434 * returned if the buffer can be discarded, (0) is returned if it cannot.
437 bpf_canfreebuf(struct bpf_d *d)
442 switch (d->bd_bufmode) {
443 case BPF_BUFMODE_ZBUF:
444 return (bpf_zerocopy_canfreebuf(d));
450 * Allow the buffer model to indicate that the current store buffer is
451 * immutable, regardless of the appearance of space. Return (1) if the
452 * buffer is writable, and (0) if not.
455 bpf_canwritebuf(struct bpf_d *d)
459 switch (d->bd_bufmode) {
460 case BPF_BUFMODE_ZBUF:
461 return (bpf_zerocopy_canwritebuf(d));
467 * Notify buffer model that an attempt to write to the store buffer has
468 * resulted in a dropped packet, in which case the buffer may be considered
472 bpf_buffull(struct bpf_d *d)
477 switch (d->bd_bufmode) {
478 case BPF_BUFMODE_ZBUF:
479 bpf_zerocopy_buffull(d);
485 * Notify the buffer model that a buffer has moved into the hold position.
488 bpf_bufheld(struct bpf_d *d)
493 switch (d->bd_bufmode) {
494 case BPF_BUFMODE_ZBUF:
495 bpf_zerocopy_bufheld(d);
501 bpf_free(struct bpf_d *d)
504 switch (d->bd_bufmode) {
505 case BPF_BUFMODE_BUFFER:
506 return (bpf_buffer_free(d));
508 case BPF_BUFMODE_ZBUF:
509 return (bpf_zerocopy_free(d));
512 panic("bpf_buf_free");
517 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
520 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
522 return (bpf_buffer_uiomove(d, buf, len, uio));
526 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
529 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
531 return (bpf_buffer_ioctl_sblen(d, i));
535 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
538 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
540 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
544 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
547 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
549 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
553 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
556 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
558 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
562 * General BPF functions.
565 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
566 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
568 const struct ieee80211_bpf_params *p;
569 struct ether_header *eh;
577 * Build a sockaddr based on the data link layer type.
578 * We do this at this level because the ethernet header
579 * is copied directly into the data field of the sockaddr.
580 * In the case of SLIP, there is no header and the packet
581 * is forwarded as is.
582 * Also, we are careful to leave room at the front of the mbuf
583 * for the link level header.
587 sockp->sa_family = AF_INET;
592 sockp->sa_family = AF_UNSPEC;
593 /* XXX Would MAXLINKHDR be better? */
594 hlen = ETHER_HDR_LEN;
598 sockp->sa_family = AF_IMPLINK;
603 sockp->sa_family = AF_UNSPEC;
609 * null interface types require a 4 byte pseudo header which
610 * corresponds to the address family of the packet.
612 sockp->sa_family = AF_UNSPEC;
616 case DLT_ATM_RFC1483:
618 * en atm driver requires 4-byte atm pseudo header.
619 * though it isn't standard, vpi:vci needs to be
622 sockp->sa_family = AF_UNSPEC;
623 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
627 sockp->sa_family = AF_UNSPEC;
628 hlen = 4; /* This should match PPP_HDRLEN */
631 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
632 sockp->sa_family = AF_IEEE80211;
636 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
637 sockp->sa_family = AF_IEEE80211;
638 sockp->sa_len = 12; /* XXX != 0 */
639 hlen = sizeof(struct ieee80211_bpf_params);
646 len = uio->uio_resid;
647 if (len < hlen || len - hlen > ifp->if_mtu)
650 /* Allocate a mbuf for our write, since m_get2 fails if len >= to MJUMPAGESIZE, use m_getjcl for bigger buffers */
651 m = m_get3(len, M_WAITOK, MT_DATA, M_PKTHDR);
654 m->m_pkthdr.len = m->m_len = len;
657 error = uiomove(mtod(m, u_char *), len, uio);
661 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
667 /* Check for multicast destination */
670 eh = mtod(m, struct ether_header *);
671 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
672 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
673 ETHER_ADDR_LEN) == 0)
674 m->m_flags |= M_BCAST;
676 m->m_flags |= M_MCAST;
678 if (d->bd_hdrcmplt == 0) {
679 memcpy(eh->ether_shost, IF_LLADDR(ifp),
680 sizeof(eh->ether_shost));
686 * Make room for link header, and copy it to sockaddr
689 if (sockp->sa_family == AF_IEEE80211) {
691 * Collect true length from the parameter header
692 * NB: sockp is known to be zero'd so if we do a
693 * short copy unspecified parameters will be
695 * NB: packet may not be aligned after stripping
699 p = mtod(m, const struct ieee80211_bpf_params *);
701 if (hlen > sizeof(sockp->sa_data)) {
706 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
717 * Attach descriptor to the bpf interface, i.e. make d listen on bp,
718 * then reset its buffers and counters with reset_d().
721 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
728 * Save sysctl value to protect from sysctl change
731 op_w = V_bpf_optimize_writers || d->bd_writer;
733 if (d->bd_bif != NULL)
734 bpf_detachd_locked(d, false);
736 * Point d at bp, and add d to the interface's list.
737 * Since there are many applications using BPF for
738 * sending raw packets only (dhcpd, cdpd are good examples)
739 * we can delay adding d to the list of active listeners until
740 * some filter is configured.
745 * Hold reference to bpif while descriptor uses this interface.
750 /* Add to writers-only list */
751 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
753 * We decrement bd_writer on every filter set operation.
754 * First BIOCSETF is done by pcap_open_live() to set up
755 * snap length. After that appliation usually sets its own
760 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
764 /* Trigger EVFILT_WRITE events. */
770 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
771 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
774 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
778 * Check if we need to upgrade our descriptor @d from write-only mode.
781 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode,
784 int is_snap, need_upgrade;
787 * Check if we've already upgraded or new filter is empty.
789 if (d->bd_writer == 0 || fcode == NULL)
795 * Check if cmd looks like snaplen setting from
796 * pcap_bpf.c:pcap_open_live().
797 * Note we're not checking .k value here:
798 * while pcap_open_live() definitely sets to non-zero value,
799 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
800 * do not consider upgrading immediately
802 if (cmd == BIOCSETF && flen == 1 &&
803 fcode[0].code == (BPF_RET | BPF_K))
810 * We're setting first filter and it doesn't look like
811 * setting snaplen. We're probably using bpf directly.
812 * Upgrade immediately.
817 * Do not require upgrade by first BIOCSETF
818 * (used to set snaplen) by pcap_open_live().
821 if (--d->bd_writer == 0) {
823 * First snaplen filter has already
824 * been set. This is probably catch-all
832 "%s: filter function set by pid %d, "
833 "bd_writer counter %d, snap %d upgrade %d",
834 __func__, d->bd_pid, d->bd_writer,
835 is_snap, need_upgrade);
837 return (need_upgrade);
841 * Detach a file from its interface.
844 bpf_detachd(struct bpf_d *d)
847 bpf_detachd_locked(d, false);
852 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp)
859 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
861 /* Check if descriptor is attached */
862 if ((bp = d->bd_bif) == NULL)
866 /* Remove d from the interface's descriptor list. */
867 CK_LIST_REMOVE(d, bd_next);
868 /* Save bd_writer value */
869 error = d->bd_writer;
874 * Notify descriptor as it's detached, so that any
875 * sleepers wake up and get ENXIO.
882 /* Call event handler iff d is attached */
884 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
887 * Check if this descriptor had requested promiscuous mode.
888 * If so and ifnet is not detached, turn it off.
890 if (d->bd_promisc && !detached_ifp) {
892 CURVNET_SET(ifp->if_vnet);
893 error = ifpromisc(ifp, 0);
895 if (error != 0 && error != ENXIO) {
897 * ENXIO can happen if a pccard is unplugged
898 * Something is really wrong if we were able to put
899 * the driver into promiscuous mode, but can't
902 if_printf(bp->bif_ifp,
903 "bpf_detach: ifpromisc failed (%d)\n", error);
910 * Close the descriptor by detaching it from its interface,
911 * deallocating its buffers, and marking it free.
916 struct bpf_d *d = data;
919 if (d->bd_state == BPF_WAITING)
920 callout_stop(&d->bd_callout);
921 d->bd_state = BPF_IDLE;
923 funsetown(&d->bd_sigio);
926 mac_bpfdesc_destroy(d);
928 seldrain(&d->bd_sel);
929 knlist_destroy(&d->bd_sel.si_note);
930 callout_drain(&d->bd_callout);
935 * Open ethernet device. Returns ENXIO for illegal minor device number,
936 * EBUSY if file is open by another process.
940 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
945 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
946 error = devfs_set_cdevpriv(d, bpf_dtor);
953 d->bd_rcount = counter_u64_alloc(M_WAITOK);
954 d->bd_dcount = counter_u64_alloc(M_WAITOK);
955 d->bd_fcount = counter_u64_alloc(M_WAITOK);
956 d->bd_wcount = counter_u64_alloc(M_WAITOK);
957 d->bd_wfcount = counter_u64_alloc(M_WAITOK);
958 d->bd_wdcount = counter_u64_alloc(M_WAITOK);
959 d->bd_zcopy = counter_u64_alloc(M_WAITOK);
962 * For historical reasons, perform a one-time initialization call to
963 * the buffer routines, even though we're not yet committed to a
964 * particular buffer method.
967 if ((flags & FREAD) == 0)
969 d->bd_hbuf_in_use = 0;
970 d->bd_bufmode = BPF_BUFMODE_BUFFER;
972 d->bd_direction = BPF_D_INOUT;
973 refcount_init(&d->bd_refcnt, 1);
974 BPF_PID_REFRESH(d, td);
977 mac_bpfdesc_create(td->td_ucred, d);
979 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
980 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
981 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
983 /* Disable VLAN pcp tagging. */
990 * bpfread - read next chunk of packets from buffers
993 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
1000 error = devfs_get_cdevpriv((void **)&d);
1005 * Restrict application to use a buffer the same size as
1006 * as kernel buffers.
1008 if (uio->uio_resid != d->bd_bufsize)
1011 non_block = ((ioflag & O_NONBLOCK) != 0);
1014 BPF_PID_REFRESH_CUR(d);
1015 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
1017 return (EOPNOTSUPP);
1019 if (d->bd_state == BPF_WAITING)
1020 callout_stop(&d->bd_callout);
1021 timed_out = (d->bd_state == BPF_TIMED_OUT);
1022 d->bd_state = BPF_IDLE;
1023 while (d->bd_hbuf_in_use) {
1024 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1025 PRINET|PCATCH, "bd_hbuf", 0);
1032 * If the hold buffer is empty, then do a timed sleep, which
1033 * ends when the timeout expires or when enough packets
1034 * have arrived to fill the store buffer.
1036 while (d->bd_hbuf == NULL) {
1037 if (d->bd_slen != 0) {
1039 * A packet(s) either arrived since the previous
1040 * read or arrived while we were asleep.
1042 if (d->bd_immediate || non_block || timed_out) {
1044 * Rotate the buffers and return what's here
1045 * if we are in immediate mode, non-blocking
1046 * flag is set, or this descriptor timed out.
1054 * No data is available, check to see if the bpf device
1055 * is still pointed at a real interface. If not, return
1056 * ENXIO so that the userland process knows to rebind
1057 * it before using it again.
1059 if (d->bd_bif == NULL) {
1066 return (EWOULDBLOCK);
1068 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
1069 "bpf", d->bd_rtout);
1070 if (error == EINTR || error == ERESTART) {
1074 if (error == EWOULDBLOCK) {
1076 * On a timeout, return what's in the buffer,
1077 * which may be nothing. If there is something
1078 * in the store buffer, we can rotate the buffers.
1082 * We filled up the buffer in between
1083 * getting the timeout and arriving
1084 * here, so we don't need to rotate.
1088 if (d->bd_slen == 0) {
1097 * At this point, we know we have something in the hold slot.
1099 d->bd_hbuf_in_use = 1;
1103 * Move data from hold buffer into user space.
1104 * We know the entire buffer is transferred since
1105 * we checked above that the read buffer is bpf_bufsize bytes.
1107 * We do not have to worry about simultaneous reads because
1108 * we waited for sole access to the hold buffer above.
1110 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1113 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1114 d->bd_fbuf = d->bd_hbuf;
1117 bpf_buf_reclaimed(d);
1118 d->bd_hbuf_in_use = 0;
1119 wakeup(&d->bd_hbuf_in_use);
1126 * If there are processes sleeping on this descriptor, wake them up.
1128 static __inline void
1129 bpf_wakeup(struct bpf_d *d)
1132 BPFD_LOCK_ASSERT(d);
1133 if (d->bd_state == BPF_WAITING) {
1134 callout_stop(&d->bd_callout);
1135 d->bd_state = BPF_IDLE;
1138 if (d->bd_async && d->bd_sig && d->bd_sigio)
1139 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1141 selwakeuppri(&d->bd_sel, PRINET);
1142 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1146 bpf_timed_out(void *arg)
1148 struct bpf_d *d = (struct bpf_d *)arg;
1150 BPFD_LOCK_ASSERT(d);
1152 if (callout_pending(&d->bd_callout) ||
1153 !callout_active(&d->bd_callout))
1155 if (d->bd_state == BPF_WAITING) {
1156 d->bd_state = BPF_TIMED_OUT;
1157 if (d->bd_slen != 0)
1163 bpf_ready(struct bpf_d *d)
1166 BPFD_LOCK_ASSERT(d);
1168 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1170 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1177 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1180 struct sockaddr dst;
1181 struct epoch_tracker et;
1185 struct mbuf *m, *mc;
1188 error = devfs_get_cdevpriv((void **)&d);
1192 NET_EPOCH_ENTER(et);
1194 BPF_PID_REFRESH_CUR(d);
1195 counter_u64_add(d->bd_wcount, 1);
1196 if ((bp = d->bd_bif) == NULL) {
1202 if ((ifp->if_flags & IFF_UP) == 0) {
1207 if (uio->uio_resid == 0)
1210 bzero(&dst, sizeof(dst));
1215 * Take extra reference, unlock d and exit from epoch section,
1216 * since bpf_movein() can sleep.
1222 error = bpf_movein(uio, (int)bp->bif_dlt, ifp,
1223 &m, &dst, &hlen, d);
1226 counter_u64_add(d->bd_wdcount, 1);
1233 * Check that descriptor is still attached to the interface.
1234 * This can happen on bpfdetach(). To avoid access to detached
1235 * ifnet, free mbuf and return ENXIO.
1237 if (d->bd_bif == NULL) {
1238 counter_u64_add(d->bd_wdcount, 1);
1244 counter_u64_add(d->bd_wfcount, 1);
1246 dst.sa_family = pseudo_AF_HDRCMPLT;
1248 if (d->bd_feedback) {
1249 mc = m_dup(m, M_NOWAIT);
1251 mc->m_pkthdr.rcvif = ifp;
1252 /* Set M_PROMISC for outgoing packets to be discarded. */
1253 if (d->bd_direction == BPF_D_INOUT)
1254 m->m_flags |= M_PROMISC;
1258 m->m_pkthdr.len -= hlen;
1260 m->m_data += hlen; /* XXX */
1262 CURVNET_SET(ifp->if_vnet);
1264 mac_bpfdesc_create_mbuf(d, m);
1266 mac_bpfdesc_create_mbuf(d, mc);
1269 bzero(&ro, sizeof(ro));
1271 ro.ro_prepend = (u_char *)&dst.sa_data;
1273 ro.ro_flags = RT_HAS_HEADER;
1277 vlan_set_pcp(m, d->bd_pcp);
1279 /* Avoid possible recursion on BPFD_LOCK(). */
1280 NET_EPOCH_ENTER(et);
1282 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1284 counter_u64_add(d->bd_wdcount, 1);
1288 (*ifp->if_input)(ifp, mc);
1298 counter_u64_add(d->bd_wdcount, 1);
1305 * Reset a descriptor by flushing its packet buffer and clearing the receive
1306 * and drop counts. This is doable for kernel-only buffers, but with
1307 * zero-copy buffers, we can't write to (or rotate) buffers that are
1308 * currently owned by userspace. It would be nice if we could encapsulate
1309 * this logic in the buffer code rather than here.
1312 reset_d(struct bpf_d *d)
1315 BPFD_LOCK_ASSERT(d);
1317 while (d->bd_hbuf_in_use)
1318 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1320 if ((d->bd_hbuf != NULL) &&
1321 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1322 /* Free the hold buffer. */
1323 d->bd_fbuf = d->bd_hbuf;
1326 bpf_buf_reclaimed(d);
1328 if (bpf_canwritebuf(d))
1330 counter_u64_zero(d->bd_rcount);
1331 counter_u64_zero(d->bd_dcount);
1332 counter_u64_zero(d->bd_fcount);
1333 counter_u64_zero(d->bd_wcount);
1334 counter_u64_zero(d->bd_wfcount);
1335 counter_u64_zero(d->bd_wdcount);
1336 counter_u64_zero(d->bd_zcopy);
1340 * FIONREAD Check for read packet available.
1341 * BIOCGBLEN Get buffer len [for read()].
1342 * BIOCSETF Set read filter.
1343 * BIOCSETFNR Set read filter without resetting descriptor.
1344 * BIOCSETWF Set write filter.
1345 * BIOCFLUSH Flush read packet buffer.
1346 * BIOCPROMISC Put interface into promiscuous mode.
1347 * BIOCGDLT Get link layer type.
1348 * BIOCGETIF Get interface name.
1349 * BIOCSETIF Set interface.
1350 * BIOCSRTIMEOUT Set read timeout.
1351 * BIOCGRTIMEOUT Get read timeout.
1352 * BIOCGSTATS Get packet stats.
1353 * BIOCIMMEDIATE Set immediate mode.
1354 * BIOCVERSION Get filter language version.
1355 * BIOCGHDRCMPLT Get "header already complete" flag
1356 * BIOCSHDRCMPLT Set "header already complete" flag
1357 * BIOCGDIRECTION Get packet direction flag
1358 * BIOCSDIRECTION Set packet direction flag
1359 * BIOCGTSTAMP Get time stamp format and resolution.
1360 * BIOCSTSTAMP Set time stamp format and resolution.
1361 * BIOCLOCK Set "locked" flag
1362 * BIOCFEEDBACK Set packet feedback mode.
1363 * BIOCSETZBUF Set current zero-copy buffer locations.
1364 * BIOCGETZMAX Get maximum zero-copy buffer size.
1365 * BIOCROTZBUF Force rotation of zero-copy buffer
1366 * BIOCSETBUFMODE Set buffer mode.
1367 * BIOCGETBUFMODE Get current buffer mode.
1368 * BIOCSETVLANPCP Set VLAN PCP tag.
1372 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1378 error = devfs_get_cdevpriv((void **)&d);
1383 * Refresh PID associated with this descriptor.
1386 BPF_PID_REFRESH(d, td);
1387 if (d->bd_state == BPF_WAITING)
1388 callout_stop(&d->bd_callout);
1389 d->bd_state = BPF_IDLE;
1392 if (d->bd_locked == 1) {
1398 #ifdef COMPAT_FREEBSD32
1399 case BIOCGDLTLIST32:
1403 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1404 case BIOCGRTIMEOUT32:
1415 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1416 case BIOCSRTIMEOUT32:
1426 #ifdef COMPAT_FREEBSD32
1428 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1429 * that it will get 32-bit packet headers.
1435 case BIOCGDLTLIST32:
1436 case BIOCGRTIMEOUT32:
1437 case BIOCSRTIMEOUT32:
1438 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1446 CURVNET_SET(TD_TO_VNET(td));
1453 * Check for read packet available.
1461 while (d->bd_hbuf_in_use)
1462 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1463 PRINET, "bd_hbuf", 0);
1473 * Get buffer len [for read()].
1477 *(u_int *)addr = d->bd_bufsize;
1482 * Set buffer length.
1485 error = bpf_ioctl_sblen(d, (u_int *)addr);
1489 * Set link layer read filter.
1494 #ifdef COMPAT_FREEBSD32
1499 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1503 * Flush read packet buffer.
1512 * Put interface into promiscuous mode.
1516 if (d->bd_bif == NULL) {
1518 * No interface attached yet.
1521 } else if (d->bd_promisc == 0) {
1522 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1530 * Get current data link type.
1534 if (d->bd_bif == NULL)
1537 *(u_int *)addr = d->bd_bif->bif_dlt;
1542 * Get a list of supported data link types.
1544 #ifdef COMPAT_FREEBSD32
1545 case BIOCGDLTLIST32:
1547 struct bpf_dltlist32 *list32;
1548 struct bpf_dltlist dltlist;
1550 list32 = (struct bpf_dltlist32 *)addr;
1551 dltlist.bfl_len = list32->bfl_len;
1552 dltlist.bfl_list = PTRIN(list32->bfl_list);
1554 if (d->bd_bif == NULL)
1557 error = bpf_getdltlist(d, &dltlist);
1559 list32->bfl_len = dltlist.bfl_len;
1568 if (d->bd_bif == NULL)
1571 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1576 * Set data link type.
1580 if (d->bd_bif == NULL)
1583 error = bpf_setdlt(d, *(u_int *)addr);
1588 * Get interface name.
1592 if (d->bd_bif == NULL)
1595 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1596 struct ifreq *const ifr = (struct ifreq *)addr;
1598 strlcpy(ifr->ifr_name, ifp->if_xname,
1599 sizeof(ifr->ifr_name));
1609 int alloc_buf, size;
1612 * Behavior here depends on the buffering model. If
1613 * we're using kernel memory buffers, then we can
1614 * allocate them here. If we're using zero-copy,
1615 * then the user process must have registered buffers
1616 * by the time we get here.
1620 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1625 size = d->bd_bufsize;
1626 error = bpf_buffer_ioctl_sblen(d, &size);
1631 error = bpf_setif(d, (struct ifreq *)addr);
1640 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1641 case BIOCSRTIMEOUT32:
1644 struct timeval *tv = (struct timeval *)addr;
1645 #if defined(COMPAT_FREEBSD32)
1646 struct timeval32 *tv32;
1647 struct timeval tv64;
1649 if (cmd == BIOCSRTIMEOUT32) {
1650 tv32 = (struct timeval32 *)addr;
1652 tv->tv_sec = tv32->tv_sec;
1653 tv->tv_usec = tv32->tv_usec;
1656 tv = (struct timeval *)addr;
1659 * Subtract 1 tick from tvtohz() since this isn't
1662 if ((error = itimerfix(tv)) == 0)
1663 d->bd_rtout = tvtohz(tv) - 1;
1671 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1672 case BIOCGRTIMEOUT32:
1676 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1677 struct timeval32 *tv32;
1678 struct timeval tv64;
1680 if (cmd == BIOCGRTIMEOUT32)
1684 tv = (struct timeval *)addr;
1686 tv->tv_sec = d->bd_rtout / hz;
1687 tv->tv_usec = (d->bd_rtout % hz) * tick;
1688 #if defined(COMPAT_FREEBSD32) && defined(__amd64__)
1689 if (cmd == BIOCGRTIMEOUT32) {
1690 tv32 = (struct timeval32 *)addr;
1691 tv32->tv_sec = tv->tv_sec;
1692 tv32->tv_usec = tv->tv_usec;
1704 struct bpf_stat *bs = (struct bpf_stat *)addr;
1706 /* XXXCSJP overflow */
1707 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount);
1708 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount);
1713 * Set immediate mode.
1717 d->bd_immediate = *(u_int *)addr;
1723 struct bpf_version *bv = (struct bpf_version *)addr;
1725 bv->bv_major = BPF_MAJOR_VERSION;
1726 bv->bv_minor = BPF_MINOR_VERSION;
1731 * Get "header already complete" flag
1735 *(u_int *)addr = d->bd_hdrcmplt;
1740 * Set "header already complete" flag
1744 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1749 * Get packet direction flag
1751 case BIOCGDIRECTION:
1753 *(u_int *)addr = d->bd_direction;
1758 * Set packet direction flag
1760 case BIOCSDIRECTION:
1764 direction = *(u_int *)addr;
1765 switch (direction) {
1770 d->bd_direction = direction;
1780 * Get packet timestamp format and resolution.
1784 *(u_int *)addr = d->bd_tstamp;
1789 * Set packet timestamp format and resolution.
1795 func = *(u_int *)addr;
1796 if (BPF_T_VALID(func))
1797 d->bd_tstamp = func;
1805 d->bd_feedback = *(u_int *)addr;
1815 case FIONBIO: /* Non-blocking I/O */
1818 case FIOASYNC: /* Send signal on receive packets */
1820 d->bd_async = *(int *)addr;
1826 * XXX: Add some sort of locking here?
1827 * fsetown() can sleep.
1829 error = fsetown(*(int *)addr, &d->bd_sigio);
1834 *(int *)addr = fgetown(&d->bd_sigio);
1838 /* This is deprecated, FIOSETOWN should be used instead. */
1840 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1843 /* This is deprecated, FIOGETOWN should be used instead. */
1845 *(int *)addr = -fgetown(&d->bd_sigio);
1848 case BIOCSRSIG: /* Set receive signal */
1852 sig = *(u_int *)addr;
1865 *(u_int *)addr = d->bd_sig;
1869 case BIOCGETBUFMODE:
1871 *(u_int *)addr = d->bd_bufmode;
1875 case BIOCSETBUFMODE:
1877 * Allow the buffering mode to be changed as long as we
1878 * haven't yet committed to a particular mode. Our
1879 * definition of commitment, for now, is whether or not a
1880 * buffer has been allocated or an interface attached, since
1881 * that's the point where things get tricky.
1883 switch (*(u_int *)addr) {
1884 case BPF_BUFMODE_BUFFER:
1887 case BPF_BUFMODE_ZBUF:
1888 if (bpf_zerocopy_enable)
1898 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1899 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1904 d->bd_bufmode = *(u_int *)addr;
1909 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1913 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1917 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1920 case BIOCSETVLANPCP:
1924 pcp = *(u_int *)addr;
1925 if (pcp > BPF_PRIO_MAX || pcp < 0) {
1938 * Set d's packet filter program to fp. If this file already has a filter,
1939 * free it and replace it. Returns EINVAL for bogus requests.
1941 * Note we use global lock here to serialize bpf_setf() and bpf_setif()
1945 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1947 #ifdef COMPAT_FREEBSD32
1948 struct bpf_program fp_swab;
1949 struct bpf_program32 *fp32;
1951 struct bpf_program_buffer *fcode;
1952 struct bpf_insn *filter;
1954 bpf_jit_filter *jfunc;
1960 #ifdef COMPAT_FREEBSD32
1965 fp32 = (struct bpf_program32 *)fp;
1966 fp_swab.bf_len = fp32->bf_len;
1968 (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1987 * Check new filter validness before acquiring any locks.
1988 * Allocate memory for new filter, if needed.
1991 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1993 size = flen * sizeof(*fp->bf_insns);
1995 /* We're setting up new filter. Copy and check actual data. */
1996 fcode = bpf_program_buffer_alloc(size, M_WAITOK);
1997 filter = (struct bpf_insn *)fcode->buffer;
1998 if (copyin(fp->bf_insns, filter, size) != 0 ||
1999 !bpf_validate(filter, flen)) {
2004 if (cmd != BIOCSETWF) {
2006 * Filter is copied inside fcode and is
2009 jfunc = bpf_jitter(filter, flen);
2014 track_event = false;
2019 /* Set up new filter. */
2020 if (cmd == BIOCSETWF) {
2021 if (d->bd_wfilter != NULL) {
2022 fcode = __containerof((void *)d->bd_wfilter,
2023 struct bpf_program_buffer, buffer);
2028 d->bd_wfilter = filter;
2030 if (d->bd_rfilter != NULL) {
2031 fcode = __containerof((void *)d->bd_rfilter,
2032 struct bpf_program_buffer, buffer);
2034 fcode->func = d->bd_bfilter;
2037 d->bd_rfilter = filter;
2039 d->bd_bfilter = jfunc;
2041 if (cmd == BIOCSETF)
2044 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) {
2046 * Filter can be set several times without
2047 * specifying interface. In this case just mark d
2051 if (d->bd_bif != NULL) {
2053 * Remove descriptor from writers-only list
2054 * and add it to active readers list.
2056 CK_LIST_REMOVE(d, bd_next);
2057 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist,
2060 "%s: upgrade required by pid %d",
2061 __func__, d->bd_pid);
2069 NET_EPOCH_CALL(bpf_program_buffer_free, &fcode->epoch_ctx);
2072 EVENTHANDLER_INVOKE(bpf_track,
2073 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1);
2080 * Detach a file from its current interface (if attached at all) and attach
2081 * to the interface indicated by the name stored in ifr.
2082 * Return an errno or 0.
2085 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
2088 struct ifnet *theywant;
2092 theywant = ifunit(ifr->ifr_name);
2093 if (theywant == NULL || theywant->if_bpf == NULL)
2096 bp = theywant->if_bpf;
2098 * At this point, we expect the buffer is already allocated. If not,
2101 switch (d->bd_bufmode) {
2102 case BPF_BUFMODE_BUFFER:
2103 case BPF_BUFMODE_ZBUF:
2104 if (d->bd_sbuf == NULL)
2109 panic("bpf_setif: bufmode %d", d->bd_bufmode);
2111 if (bp != d->bd_bif)
2122 * Support for select() and poll() system calls
2124 * Return true iff the specific operation will not block indefinitely.
2125 * Otherwise, return false but make a note that a selwakeup() must be done.
2128 bpfpoll(struct cdev *dev, int events, struct thread *td)
2133 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
2135 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
2138 * Refresh PID associated with this descriptor.
2140 revents = events & (POLLOUT | POLLWRNORM);
2142 BPF_PID_REFRESH(d, td);
2143 if (events & (POLLIN | POLLRDNORM)) {
2145 revents |= events & (POLLIN | POLLRDNORM);
2147 selrecord(td, &d->bd_sel);
2148 /* Start the read timeout if necessary. */
2149 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2150 callout_reset(&d->bd_callout, d->bd_rtout,
2152 d->bd_state = BPF_WAITING;
2161 * Support for kevent() system call. Register EVFILT_READ filters and
2162 * reject all others.
2165 bpfkqfilter(struct cdev *dev, struct knote *kn)
2169 if (devfs_get_cdevpriv((void **)&d) != 0)
2172 switch (kn->kn_filter) {
2174 kn->kn_fop = &bpfread_filtops;
2178 kn->kn_fop = &bpfwrite_filtops;
2186 * Refresh PID associated with this descriptor.
2189 BPF_PID_REFRESH_CUR(d);
2191 knlist_add(&d->bd_sel.si_note, kn, 1);
2198 filt_bpfdetach(struct knote *kn)
2200 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2202 knlist_remove(&d->bd_sel.si_note, kn, 0);
2206 filt_bpfread(struct knote *kn, long hint)
2208 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2211 BPFD_LOCK_ASSERT(d);
2212 ready = bpf_ready(d);
2214 kn->kn_data = d->bd_slen;
2216 * Ignore the hold buffer if it is being copied to user space.
2218 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2219 kn->kn_data += d->bd_hlen;
2220 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2221 callout_reset(&d->bd_callout, d->bd_rtout,
2223 d->bd_state = BPF_WAITING;
2230 filt_bpfwrite(struct knote *kn, long hint)
2232 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2234 BPFD_LOCK_ASSERT(d);
2236 if (d->bd_bif == NULL) {
2240 kn->kn_data = d->bd_bif->bif_ifp->if_mtu;
2245 #define BPF_TSTAMP_NONE 0
2246 #define BPF_TSTAMP_FAST 1
2247 #define BPF_TSTAMP_NORMAL 2
2248 #define BPF_TSTAMP_EXTERN 3
2251 bpf_ts_quality(int tstype)
2254 if (tstype == BPF_T_NONE)
2255 return (BPF_TSTAMP_NONE);
2256 if ((tstype & BPF_T_FAST) != 0)
2257 return (BPF_TSTAMP_FAST);
2259 return (BPF_TSTAMP_NORMAL);
2263 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2269 quality = bpf_ts_quality(tstype);
2270 if (quality == BPF_TSTAMP_NONE)
2274 if ((m->m_flags & (M_PKTHDR | M_TSTMP)) == (M_PKTHDR | M_TSTMP)) {
2275 mbuf_tstmp2timespec(m, &ts);
2276 timespec2bintime(&ts, bt);
2277 return (BPF_TSTAMP_EXTERN);
2279 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2281 *bt = *(struct bintime *)(tag + 1);
2282 return (BPF_TSTAMP_EXTERN);
2285 if (quality == BPF_TSTAMP_NORMAL)
2294 * Incoming linkage from device drivers. Process the packet pkt, of length
2295 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2296 * by each process' filter, and if accepted, stashed into the corresponding
2300 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2302 struct epoch_tracker et;
2311 gottime = BPF_TSTAMP_NONE;
2312 NET_EPOCH_ENTER(et);
2313 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2314 counter_u64_add(d->bd_rcount, 1);
2316 * NB: We dont call BPF_CHECK_DIRECTION() here since there
2317 * is no way for the caller to indiciate to us whether this
2318 * packet is inbound or outbound. In the bpf_mtap() routines,
2319 * we use the interface pointers on the mbuf to figure it out.
2322 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2324 slen = (*(bf->func))(pkt, pktlen, pktlen);
2327 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2330 * Filter matches. Let's to acquire write lock.
2333 counter_u64_add(d->bd_fcount, 1);
2334 if (gottime < bpf_ts_quality(d->bd_tstamp))
2335 gottime = bpf_gettime(&bt, d->bd_tstamp,
2338 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2340 catchpacket(d, pkt, pktlen, slen,
2341 bpf_append_bytes, &bt);
2349 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
2351 if (bpf_peers_present(ifp->if_bpf))
2352 bpf_tap(ifp->if_bpf, pkt, pktlen);
2355 #define BPF_CHECK_DIRECTION(d, r, i) \
2356 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2357 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2360 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2361 * Locking model is explained in bpf_tap().
2364 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2366 struct epoch_tracker et;
2375 /* Skip outgoing duplicate packets. */
2376 if ((m->m_flags & M_PROMISC) != 0 && m_rcvif(m) == NULL) {
2377 m->m_flags &= ~M_PROMISC;
2381 pktlen = m_length(m, NULL);
2382 gottime = BPF_TSTAMP_NONE;
2384 NET_EPOCH_ENTER(et);
2385 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2386 if (BPF_CHECK_DIRECTION(d, m_rcvif(m), bp->bif_ifp))
2388 counter_u64_add(d->bd_rcount, 1);
2390 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2391 /* XXX We cannot handle multiple mbufs. */
2392 if (bf != NULL && m->m_next == NULL)
2393 slen = (*(bf->func))(mtod(m, u_char *), pktlen,
2397 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2401 counter_u64_add(d->bd_fcount, 1);
2402 if (gottime < bpf_ts_quality(d->bd_tstamp))
2403 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2405 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2407 catchpacket(d, (u_char *)m, pktlen, slen,
2408 bpf_append_mbuf, &bt);
2416 bpf_mtap_if(if_t ifp, struct mbuf *m)
2418 if (bpf_peers_present(ifp->if_bpf)) {
2420 bpf_mtap(ifp->if_bpf, m);
2425 * Incoming linkage from device drivers, when packet is in
2426 * an mbuf chain and to be prepended by a contiguous header.
2429 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2431 struct epoch_tracker et;
2438 /* Skip outgoing duplicate packets. */
2439 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2440 m->m_flags &= ~M_PROMISC;
2444 pktlen = m_length(m, NULL);
2446 * Craft on-stack mbuf suitable for passing to bpf_filter.
2447 * Note that we cut corners here; we only setup what's
2448 * absolutely needed--this mbuf should never go anywhere else.
2456 gottime = BPF_TSTAMP_NONE;
2458 NET_EPOCH_ENTER(et);
2459 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2460 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2462 counter_u64_add(d->bd_rcount, 1);
2463 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2467 counter_u64_add(d->bd_fcount, 1);
2468 if (gottime < bpf_ts_quality(d->bd_tstamp))
2469 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2471 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2473 catchpacket(d, (u_char *)&mb, pktlen, slen,
2474 bpf_append_mbuf, &bt);
2482 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
2484 if (bpf_peers_present(ifp->if_bpf)) {
2486 bpf_mtap2(ifp->if_bpf, data, dlen, m);
2490 #undef BPF_CHECK_DIRECTION
2491 #undef BPF_TSTAMP_NONE
2492 #undef BPF_TSTAMP_FAST
2493 #undef BPF_TSTAMP_NORMAL
2494 #undef BPF_TSTAMP_EXTERN
2497 bpf_hdrlen(struct bpf_d *d)
2501 hdrlen = d->bd_bif->bif_hdrlen;
2502 #ifndef BURN_BRIDGES
2503 if (d->bd_tstamp == BPF_T_NONE ||
2504 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2505 #ifdef COMPAT_FREEBSD32
2507 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2510 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2513 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2514 #ifdef COMPAT_FREEBSD32
2516 hdrlen = BPF_WORDALIGN32(hdrlen);
2519 hdrlen = BPF_WORDALIGN(hdrlen);
2521 return (hdrlen - d->bd_bif->bif_hdrlen);
2525 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2527 struct bintime bt2, boottimebin;
2529 struct timespec tsn;
2531 if ((tstype & BPF_T_MONOTONIC) == 0) {
2533 getboottimebin(&boottimebin);
2534 bintime_add(&bt2, &boottimebin);
2537 switch (BPF_T_FORMAT(tstype)) {
2538 case BPF_T_MICROTIME:
2539 bintime2timeval(bt, &tsm);
2540 ts->bt_sec = tsm.tv_sec;
2541 ts->bt_frac = tsm.tv_usec;
2543 case BPF_T_NANOTIME:
2544 bintime2timespec(bt, &tsn);
2545 ts->bt_sec = tsn.tv_sec;
2546 ts->bt_frac = tsn.tv_nsec;
2549 ts->bt_sec = bt->sec;
2550 ts->bt_frac = bt->frac;
2556 * Move the packet data from interface memory (pkt) into the
2557 * store buffer. "cpfn" is the routine called to do the actual data
2558 * transfer. bcopy is passed in to copy contiguous chunks, while
2559 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2560 * pkt is really an mbuf.
2563 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2564 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2567 static char zeroes[BPF_ALIGNMENT];
2568 struct bpf_xhdr hdr;
2569 #ifndef BURN_BRIDGES
2570 struct bpf_hdr hdr_old;
2571 #ifdef COMPAT_FREEBSD32
2572 struct bpf_hdr32 hdr32_old;
2575 int caplen, curlen, hdrlen, pad, totlen;
2580 BPFD_LOCK_ASSERT(d);
2581 if (d->bd_bif == NULL) {
2582 /* Descriptor was detached in concurrent thread */
2583 counter_u64_add(d->bd_dcount, 1);
2588 * Detect whether user space has released a buffer back to us, and if
2589 * so, move it from being a hold buffer to a free buffer. This may
2590 * not be the best place to do it (for example, we might only want to
2591 * run this check if we need the space), but for now it's a reliable
2594 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2595 d->bd_fbuf = d->bd_hbuf;
2598 bpf_buf_reclaimed(d);
2602 * Figure out how many bytes to move. If the packet is
2603 * greater or equal to the snapshot length, transfer that
2604 * much. Otherwise, transfer the whole packet (unless
2605 * we hit the buffer size limit).
2607 hdrlen = bpf_hdrlen(d);
2608 totlen = hdrlen + min(snaplen, pktlen);
2609 if (totlen > d->bd_bufsize)
2610 totlen = d->bd_bufsize;
2613 * Round up the end of the previous packet to the next longword.
2615 * Drop the packet if there's no room and no hope of room
2616 * If the packet would overflow the storage buffer or the storage
2617 * buffer is considered immutable by the buffer model, try to rotate
2618 * the buffer and wakeup pending processes.
2620 #ifdef COMPAT_FREEBSD32
2622 curlen = BPF_WORDALIGN32(d->bd_slen);
2625 curlen = BPF_WORDALIGN(d->bd_slen);
2626 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2627 if (d->bd_fbuf == NULL) {
2629 * There's no room in the store buffer, and no
2630 * prospect of room, so drop the packet. Notify the
2634 counter_u64_add(d->bd_dcount, 1);
2637 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2642 if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) {
2644 * Immediate mode is set, or the read timeout has
2645 * already expired during a select call. A packet
2646 * arrived, so the reader should be woken up.
2650 pad = curlen - d->bd_slen;
2651 KASSERT(pad >= 0 && pad <= sizeof(zeroes),
2652 ("%s: invalid pad byte count %d", __func__, pad));
2654 /* Zero pad bytes. */
2655 bpf_append_bytes(d, d->bd_sbuf, d->bd_slen, zeroes,
2660 caplen = totlen - hdrlen;
2661 tstype = d->bd_tstamp;
2662 do_timestamp = tstype != BPF_T_NONE;
2663 #ifndef BURN_BRIDGES
2664 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2667 bpf_bintime2ts(bt, &ts, tstype);
2668 #ifdef COMPAT_FREEBSD32
2669 if (d->bd_compat32) {
2670 bzero(&hdr32_old, sizeof(hdr32_old));
2672 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2673 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2675 hdr32_old.bh_datalen = pktlen;
2676 hdr32_old.bh_hdrlen = hdrlen;
2677 hdr32_old.bh_caplen = caplen;
2678 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2683 bzero(&hdr_old, sizeof(hdr_old));
2685 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2686 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2688 hdr_old.bh_datalen = pktlen;
2689 hdr_old.bh_hdrlen = hdrlen;
2690 hdr_old.bh_caplen = caplen;
2691 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2698 * Append the bpf header. Note we append the actual header size, but
2699 * move forward the length of the header plus padding.
2701 bzero(&hdr, sizeof(hdr));
2703 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2704 hdr.bh_datalen = pktlen;
2705 hdr.bh_hdrlen = hdrlen;
2706 hdr.bh_caplen = caplen;
2707 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2710 * Copy the packet data into the store buffer and update its length.
2712 #ifndef BURN_BRIDGES
2715 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2716 d->bd_slen = curlen + totlen;
2723 * Free buffers currently in use by a descriptor.
2727 bpfd_free(epoch_context_t ctx)
2730 struct bpf_program_buffer *p;
2733 * We don't need to lock out interrupts since this descriptor has
2734 * been detached from its interface and it yet hasn't been marked
2737 d = __containerof(ctx, struct bpf_d, epoch_ctx);
2739 if (d->bd_rfilter != NULL) {
2740 p = __containerof((void *)d->bd_rfilter,
2741 struct bpf_program_buffer, buffer);
2743 p->func = d->bd_bfilter;
2745 bpf_program_buffer_free(&p->epoch_ctx);
2747 if (d->bd_wfilter != NULL) {
2748 p = __containerof((void *)d->bd_wfilter,
2749 struct bpf_program_buffer, buffer);
2753 bpf_program_buffer_free(&p->epoch_ctx);
2756 mtx_destroy(&d->bd_lock);
2757 counter_u64_free(d->bd_rcount);
2758 counter_u64_free(d->bd_dcount);
2759 counter_u64_free(d->bd_fcount);
2760 counter_u64_free(d->bd_wcount);
2761 counter_u64_free(d->bd_wfcount);
2762 counter_u64_free(d->bd_wdcount);
2763 counter_u64_free(d->bd_zcopy);
2768 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2769 * fixed size of the link header (variable length headers not yet supported).
2772 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2775 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2779 * Attach an interface to bpf. ifp is a pointer to the structure
2780 * defining the interface to be attached, dlt is the link layer type,
2781 * and hdrlen is the fixed size of the link header (variable length
2782 * headers are not yet supporrted).
2785 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen,
2786 struct bpf_if **driverp)
2790 KASSERT(*driverp == NULL,
2791 ("bpfattach2: driverp already initialized"));
2793 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO);
2795 CK_LIST_INIT(&bp->bif_dlist);
2796 CK_LIST_INIT(&bp->bif_wlist);
2799 bp->bif_hdrlen = hdrlen;
2800 bp->bif_bpf = driverp;
2801 refcount_init(&bp->bif_refcnt, 1);
2804 * Reference ifnet pointer, so it won't freed until
2809 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2812 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2813 if_printf(ifp, "bpf attached\n");
2818 * When moving interfaces between vnet instances we need a way to
2819 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2820 * after the vmove. We unfortunately have no device driver infrastructure
2821 * to query the interface for these values after creation/attach, thus
2822 * add this as a workaround.
2825 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2830 if (bif_dlt == NULL && bif_hdrlen == NULL)
2833 if (bif_dlt != NULL)
2834 *bif_dlt = bp->bif_dlt;
2835 if (bif_hdrlen != NULL)
2836 *bif_hdrlen = bp->bif_hdrlen;
2843 * Detach bpf from an interface. This involves detaching each descriptor
2844 * associated with the interface. Notify each descriptor as it's detached
2845 * so that any sleepers wake up and get ENXIO.
2848 bpfdetach(struct ifnet *ifp)
2850 struct bpf_if *bp, *bp_temp;
2854 /* Find all bpf_if struct's which reference ifp and detach them. */
2855 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2856 if (ifp != bp->bif_ifp)
2859 CK_LIST_REMOVE(bp, bif_next);
2860 *bp->bif_bpf = __DECONST(struct bpf_if *, &dead_bpf_if);
2863 "%s: sheduling free for encap %d (%p) for if %p",
2864 __func__, bp->bif_dlt, bp, ifp);
2866 /* Detach common descriptors */
2867 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) {
2868 bpf_detachd_locked(d, true);
2871 /* Detach writer-only descriptors */
2872 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) {
2873 bpf_detachd_locked(d, true);
2881 bpf_peers_present_if(struct ifnet *ifp)
2883 struct bpf_if *bp = ifp->if_bpf;
2885 return (bpf_peers_present(bp) > 0);
2889 * Get a list of available data link type of the interface.
2892 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2901 ifp = d->bd_bif->bif_ifp;
2903 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2904 if (bp->bif_ifp == ifp)
2907 if (bfl->bfl_list == NULL) {
2911 if (n1 > bfl->bfl_len)
2914 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2916 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2917 if (bp->bif_ifp != ifp)
2919 lst[n++] = bp->bif_dlt;
2921 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2928 * Set the data link type of a BPF instance.
2931 bpf_setdlt(struct bpf_d *d, u_int dlt)
2933 int error, opromisc;
2938 MPASS(d->bd_bif != NULL);
2941 * It is safe to check bd_bif without BPFD_LOCK, it can not be
2942 * changed while we hold global lock.
2944 if (d->bd_bif->bif_dlt == dlt)
2947 ifp = d->bd_bif->bif_ifp;
2948 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2949 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2955 opromisc = d->bd_promisc;
2958 error = ifpromisc(bp->bif_ifp, 1);
2960 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n",
2969 bpf_drvinit(void *unused)
2973 sx_init(&bpf_sx, "bpf global lock");
2974 CK_LIST_INIT(&bpf_iflist);
2976 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2977 /* For compatibility */
2978 make_dev_alias(dev, "bpf0");
2982 * Zero out the various packet counters associated with all of the bpf
2983 * descriptors. At some point, we will probably want to get a bit more
2984 * granular and allow the user to specify descriptors to be zeroed.
2987 bpf_zero_counters(void)
2994 * We are protected by global lock here, interfaces and
2995 * descriptors can not be deleted while we hold it.
2997 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2998 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2999 counter_u64_zero(bd->bd_rcount);
3000 counter_u64_zero(bd->bd_dcount);
3001 counter_u64_zero(bd->bd_fcount);
3002 counter_u64_zero(bd->bd_wcount);
3003 counter_u64_zero(bd->bd_wfcount);
3004 counter_u64_zero(bd->bd_zcopy);
3011 * Fill filter statistics
3014 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
3018 bzero(d, sizeof(*d));
3019 d->bd_structsize = sizeof(*d);
3020 d->bd_immediate = bd->bd_immediate;
3021 d->bd_promisc = bd->bd_promisc;
3022 d->bd_hdrcmplt = bd->bd_hdrcmplt;
3023 d->bd_direction = bd->bd_direction;
3024 d->bd_feedback = bd->bd_feedback;
3025 d->bd_async = bd->bd_async;
3026 d->bd_rcount = counter_u64_fetch(bd->bd_rcount);
3027 d->bd_dcount = counter_u64_fetch(bd->bd_dcount);
3028 d->bd_fcount = counter_u64_fetch(bd->bd_fcount);
3029 d->bd_sig = bd->bd_sig;
3030 d->bd_slen = bd->bd_slen;
3031 d->bd_hlen = bd->bd_hlen;
3032 d->bd_bufsize = bd->bd_bufsize;
3033 d->bd_pid = bd->bd_pid;
3034 strlcpy(d->bd_ifname,
3035 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
3036 d->bd_locked = bd->bd_locked;
3037 d->bd_wcount = counter_u64_fetch(bd->bd_wcount);
3038 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount);
3039 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount);
3040 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy);
3041 d->bd_bufmode = bd->bd_bufmode;
3045 * Handle `netstat -B' stats request
3048 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
3050 static const struct xbpf_d zerostats;
3051 struct xbpf_d *xbdbuf, *xbd, tempstats;
3057 * XXX This is not technically correct. It is possible for non
3058 * privileged users to open bpf devices. It would make sense
3059 * if the users who opened the devices were able to retrieve
3060 * the statistics for them, too.
3062 error = priv_check(req->td, PRIV_NET_BPF);
3066 * Check to see if the user is requesting that the counters be
3067 * zeroed out. Explicitly check that the supplied data is zeroed,
3068 * as we aren't allowing the user to set the counters currently.
3070 if (req->newptr != NULL) {
3071 if (req->newlen != sizeof(tempstats))
3073 memset(&tempstats, 0, sizeof(tempstats));
3074 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
3077 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
3079 bpf_zero_counters();
3082 if (req->oldptr == NULL)
3083 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
3084 if (bpf_bpfd_cnt == 0)
3085 return (SYSCTL_OUT(req, 0, 0));
3086 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
3088 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
3090 free(xbdbuf, M_BPF);
3094 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) {
3095 /* Send writers-only first */
3096 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
3097 xbd = &xbdbuf[index++];
3098 bpfstats_fill_xbpf(xbd, bd);
3100 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
3101 xbd = &xbdbuf[index++];
3102 bpfstats_fill_xbpf(xbd, bd);
3106 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
3107 free(xbdbuf, M_BPF);
3111 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
3113 #else /* !DEV_BPF && !NETGRAPH_BPF */
3116 * NOP stubs to allow bpf-using drivers to load and function.
3118 * A 'better' implementation would allow the core bpf functionality
3119 * to be loaded at runtime.
3123 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
3128 bpf_tap_if(if_t ifp, u_char *pkt, u_int pktlen)
3133 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
3138 bpf_mtap_if(if_t ifp, struct mbuf *m)
3143 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
3148 bpf_mtap2_if(if_t ifp, void *data, u_int dlen, struct mbuf *m)
3153 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
3156 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
3160 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
3163 *driverp = __DECONST(struct bpf_if *, &dead_bpf_if);
3167 bpfdetach(struct ifnet *ifp)
3172 bpf_peers_present_if(struct ifnet *ifp)
3178 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3180 return -1; /* "no filter" behaviour */
3184 bpf_validate(const struct bpf_insn *f, int len)
3186 return 0; /* false */
3189 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3193 bpf_show_bpf_if(struct bpf_if *bpf_if)
3198 db_printf("%p:\n", bpf_if);
3199 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3200 #define BPF_DB_PRINTF_RAW(f, e) db_printf(" %s = " f "\n", #e, e);
3201 /* bif_ext.bif_next */
3202 /* bif_ext.bif_dlist */
3203 BPF_DB_PRINTF("%#x", bif_dlt);
3204 BPF_DB_PRINTF("%u", bif_hdrlen);
3206 BPF_DB_PRINTF("%p", bif_ifp);
3207 BPF_DB_PRINTF("%p", bif_bpf);
3208 BPF_DB_PRINTF_RAW("%u", refcount_load(&bpf_if->bif_refcnt));
3211 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3215 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3219 bpf_show_bpf_if((struct bpf_if *)addr);