2 * Copyright (c) 1990, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from the Stanford/CMU enet packet filter,
6 * (net/enet.c) distributed as part of 4.3BSD, and code contributed
7 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)bpf.c 8.4 (Berkeley) 1/9/95
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
41 #include "opt_compat.h"
43 #include "opt_netgraph.h"
45 #include <sys/types.h>
46 #include <sys/param.h>
48 #include <sys/rwlock.h>
49 #include <sys/systm.h>
51 #include <sys/fcntl.h>
53 #include <sys/malloc.h>
58 #include <sys/signalvar.h>
59 #include <sys/filio.h>
60 #include <sys/sockio.h>
61 #include <sys/ttycom.h>
64 #include <sys/event.h>
69 #include <sys/socket.h>
76 #include <net/if_var.h>
77 #include <net/if_dl.h>
79 #include <net/bpf_buffer.h>
81 #include <net/bpf_jitter.h>
83 #include <net/bpf_zerocopy.h>
84 #include <net/bpfdesc.h>
85 #include <net/route.h>
88 #include <netinet/in.h>
89 #include <netinet/if_ether.h>
90 #include <sys/kernel.h>
91 #include <sys/sysctl.h>
93 #include <net80211/ieee80211_freebsd.h>
95 #include <security/mac/mac_framework.h>
97 MALLOC_DEFINE(M_BPF, "BPF", "BPF data");
99 static struct bpf_if_ext dead_bpf_if = {
100 .bif_dlist = LIST_HEAD_INITIALIZER()
104 #define bif_next bif_ext.bif_next
105 #define bif_dlist bif_ext.bif_dlist
106 struct bpf_if_ext bif_ext; /* public members */
107 u_int bif_dlt; /* link layer type */
108 u_int bif_hdrlen; /* length of link header */
109 struct ifnet *bif_ifp; /* corresponding interface */
110 struct rwlock bif_lock; /* interface lock */
111 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */
112 int bif_flags; /* Interface flags */
113 struct bpf_if **bif_bpf; /* Pointer to pointer to us */
116 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0);
118 #if defined(DEV_BPF) || defined(NETGRAPH_BPF)
120 #define PRINET 26 /* interruptible */
122 #define SIZEOF_BPF_HDR(type) \
123 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen))
125 #ifdef COMPAT_FREEBSD32
126 #include <sys/mount.h>
127 #include <compat/freebsd32/freebsd32.h>
128 #define BPF_ALIGNMENT32 sizeof(int32_t)
129 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32)
133 * 32-bit version of structure prepended to each packet. We use this header
134 * instead of the standard one for 32-bit streams. We mark the a stream as
135 * 32-bit the first time we see a 32-bit compat ioctl request.
138 struct timeval32 bh_tstamp; /* time stamp */
139 uint32_t bh_caplen; /* length of captured portion */
140 uint32_t bh_datalen; /* original length of packet */
141 uint16_t bh_hdrlen; /* length of bpf header (this struct
142 plus alignment padding) */
146 struct bpf_program32 {
151 struct bpf_dltlist32 {
156 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32)
157 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32)
158 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32)
159 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32)
160 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32)
161 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32)
164 #define BPF_LOCK() sx_xlock(&bpf_sx)
165 #define BPF_UNLOCK() sx_xunlock(&bpf_sx)
166 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED)
168 * bpf_iflist is a list of BPF interface structures, each corresponding to a
169 * specific DLT. The same network interface might have several BPF interface
170 * structures registered by different layers in the stack (i.e., 802.11
171 * frames, ethernet frames, etc).
173 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist;
174 static struct sx bpf_sx; /* bpf global lock */
175 static int bpf_bpfd_cnt;
177 static void bpf_attachd(struct bpf_d *, struct bpf_if *);
178 static void bpf_detachd(struct bpf_d *);
179 static void bpf_detachd_locked(struct bpf_d *);
180 static void bpf_freed(struct bpf_d *);
181 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **,
182 struct sockaddr *, int *, struct bpf_d *);
183 static int bpf_setif(struct bpf_d *, struct ifreq *);
184 static void bpf_timed_out(void *);
186 bpf_wakeup(struct bpf_d *);
187 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int,
188 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int),
190 static void reset_d(struct bpf_d *);
191 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd);
192 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *);
193 static int bpf_setdlt(struct bpf_d *, u_int);
194 static void filt_bpfdetach(struct knote *);
195 static int filt_bpfread(struct knote *, long);
196 static void bpf_drvinit(void *);
197 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS);
199 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl");
200 int bpf_maxinsns = BPF_MAXINSNS;
201 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW,
202 &bpf_maxinsns, 0, "Maximum bpf program instructions");
203 static int bpf_zerocopy_enable = 0;
204 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW,
205 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions");
206 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW,
207 bpf_stats_sysctl, "bpf statistics portal");
209 static VNET_DEFINE(int, bpf_optimize_writers) = 0;
210 #define V_bpf_optimize_writers VNET(bpf_optimize_writers)
211 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(bpf_optimize_writers), 0,
213 "Do not send packets until BPF program is set");
215 static d_open_t bpfopen;
216 static d_read_t bpfread;
217 static d_write_t bpfwrite;
218 static d_ioctl_t bpfioctl;
219 static d_poll_t bpfpoll;
220 static d_kqfilter_t bpfkqfilter;
222 static struct cdevsw bpf_cdevsw = {
223 .d_version = D_VERSION,
230 .d_kqfilter = bpfkqfilter,
233 static struct filterops bpfread_filtops = {
235 .f_detach = filt_bpfdetach,
236 .f_event = filt_bpfread,
239 eventhandler_tag bpf_ifdetach_cookie = NULL;
242 * LOCKING MODEL USED BY BPF:
244 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal,
245 * some global counters and every bpf_if reference.
246 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters.
247 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields
248 * used by bpf_mtap code.
252 * Global lock, interface lock, descriptor lock
254 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2]
255 * working model. In many places (like bpf_detachd) we start with BPF descriptor
256 * (and we need to at least rlock it to get reliable interface pointer). This
257 * gives us potential LOR. As a result, we use global lock to protect from bpf_if
258 * change in every such place.
260 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and
261 * 3) descriptor main wlock.
262 * Reading bd_bif can be protected by any of these locks, typically global lock.
264 * Changing read/write BPF filter is protected by the same three locks,
265 * the same applies for reading.
267 * Sleeping in global lock is not allowed due to bpfdetach() using it.
271 * Wrapper functions for various buffering methods. If the set of buffer
272 * modes expands, we will probably want to introduce a switch data structure
273 * similar to protosw, et.
276 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
282 switch (d->bd_bufmode) {
283 case BPF_BUFMODE_BUFFER:
284 return (bpf_buffer_append_bytes(d, buf, offset, src, len));
286 case BPF_BUFMODE_ZBUF:
288 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len));
291 panic("bpf_buf_append_bytes");
296 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src,
302 switch (d->bd_bufmode) {
303 case BPF_BUFMODE_BUFFER:
304 return (bpf_buffer_append_mbuf(d, buf, offset, src, len));
306 case BPF_BUFMODE_ZBUF:
308 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len));
311 panic("bpf_buf_append_mbuf");
316 * This function gets called when the free buffer is re-assigned.
319 bpf_buf_reclaimed(struct bpf_d *d)
324 switch (d->bd_bufmode) {
325 case BPF_BUFMODE_BUFFER:
328 case BPF_BUFMODE_ZBUF:
329 bpf_zerocopy_buf_reclaimed(d);
333 panic("bpf_buf_reclaimed");
338 * If the buffer mechanism has a way to decide that a held buffer can be made
339 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is
340 * returned if the buffer can be discarded, (0) is returned if it cannot.
343 bpf_canfreebuf(struct bpf_d *d)
348 switch (d->bd_bufmode) {
349 case BPF_BUFMODE_ZBUF:
350 return (bpf_zerocopy_canfreebuf(d));
356 * Allow the buffer model to indicate that the current store buffer is
357 * immutable, regardless of the appearance of space. Return (1) if the
358 * buffer is writable, and (0) if not.
361 bpf_canwritebuf(struct bpf_d *d)
365 switch (d->bd_bufmode) {
366 case BPF_BUFMODE_ZBUF:
367 return (bpf_zerocopy_canwritebuf(d));
373 * Notify buffer model that an attempt to write to the store buffer has
374 * resulted in a dropped packet, in which case the buffer may be considered
378 bpf_buffull(struct bpf_d *d)
383 switch (d->bd_bufmode) {
384 case BPF_BUFMODE_ZBUF:
385 bpf_zerocopy_buffull(d);
391 * Notify the buffer model that a buffer has moved into the hold position.
394 bpf_bufheld(struct bpf_d *d)
399 switch (d->bd_bufmode) {
400 case BPF_BUFMODE_ZBUF:
401 bpf_zerocopy_bufheld(d);
407 bpf_free(struct bpf_d *d)
410 switch (d->bd_bufmode) {
411 case BPF_BUFMODE_BUFFER:
412 return (bpf_buffer_free(d));
414 case BPF_BUFMODE_ZBUF:
415 return (bpf_zerocopy_free(d));
418 panic("bpf_buf_free");
423 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio)
426 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
428 return (bpf_buffer_uiomove(d, buf, len, uio));
432 bpf_ioctl_sblen(struct bpf_d *d, u_int *i)
435 if (d->bd_bufmode != BPF_BUFMODE_BUFFER)
437 return (bpf_buffer_ioctl_sblen(d, i));
441 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i)
444 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
446 return (bpf_zerocopy_ioctl_getzmax(td, d, i));
450 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
453 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
455 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz));
459 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz)
462 if (d->bd_bufmode != BPF_BUFMODE_ZBUF)
464 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz));
468 * General BPF functions.
471 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp,
472 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d)
474 const struct ieee80211_bpf_params *p;
475 struct ether_header *eh;
483 * Build a sockaddr based on the data link layer type.
484 * We do this at this level because the ethernet header
485 * is copied directly into the data field of the sockaddr.
486 * In the case of SLIP, there is no header and the packet
487 * is forwarded as is.
488 * Also, we are careful to leave room at the front of the mbuf
489 * for the link level header.
494 sockp->sa_family = AF_INET;
499 sockp->sa_family = AF_UNSPEC;
500 /* XXX Would MAXLINKHDR be better? */
501 hlen = ETHER_HDR_LEN;
505 sockp->sa_family = AF_IMPLINK;
510 sockp->sa_family = AF_UNSPEC;
516 * null interface types require a 4 byte pseudo header which
517 * corresponds to the address family of the packet.
519 sockp->sa_family = AF_UNSPEC;
523 case DLT_ATM_RFC1483:
525 * en atm driver requires 4-byte atm pseudo header.
526 * though it isn't standard, vpi:vci needs to be
529 sockp->sa_family = AF_UNSPEC;
530 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */
534 sockp->sa_family = AF_UNSPEC;
535 hlen = 4; /* This should match PPP_HDRLEN */
538 case DLT_IEEE802_11: /* IEEE 802.11 wireless */
539 sockp->sa_family = AF_IEEE80211;
543 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */
544 sockp->sa_family = AF_IEEE80211;
545 sockp->sa_len = 12; /* XXX != 0 */
546 hlen = sizeof(struct ieee80211_bpf_params);
553 len = uio->uio_resid;
554 if (len < hlen || len - hlen > ifp->if_mtu)
557 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR);
560 m->m_pkthdr.len = m->m_len = len;
563 error = uiomove(mtod(m, u_char *), len, uio);
567 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len);
573 /* Check for multicast destination */
576 eh = mtod(m, struct ether_header *);
577 if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
578 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost,
579 ETHER_ADDR_LEN) == 0)
580 m->m_flags |= M_BCAST;
582 m->m_flags |= M_MCAST;
584 if (d->bd_hdrcmplt == 0) {
585 memcpy(eh->ether_shost, IF_LLADDR(ifp),
586 sizeof(eh->ether_shost));
592 * Make room for link header, and copy it to sockaddr
595 if (sockp->sa_family == AF_IEEE80211) {
597 * Collect true length from the parameter header
598 * NB: sockp is known to be zero'd so if we do a
599 * short copy unspecified parameters will be
601 * NB: packet may not be aligned after stripping
605 p = mtod(m, const struct ieee80211_bpf_params *);
607 if (hlen > sizeof(sockp->sa_data)) {
612 bcopy(mtod(m, const void *), sockp->sa_data, hlen);
623 * Attach file to the bpf interface, i.e. make d listen on bp.
626 bpf_attachd(struct bpf_d *d, struct bpf_if *bp)
633 * Save sysctl value to protect from sysctl change
636 op_w = V_bpf_optimize_writers || d->bd_writer;
638 if (d->bd_bif != NULL)
639 bpf_detachd_locked(d);
641 * Point d at bp, and add d to the interface's list.
642 * Since there are many applications using BPF for
643 * sending raw packets only (dhcpd, cdpd are good examples)
644 * we can delay adding d to the list of active listeners until
645 * some filter is configured.
654 /* Add to writers-only list */
655 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next);
657 * We decrement bd_writer on every filter set operation.
658 * First BIOCSETF is done by pcap_open_live() to set up
659 * snap length. After that appliation usually sets its own filter
663 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
670 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list",
671 __func__, d->bd_pid, d->bd_writer ? "writer" : "active");
674 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
678 * Check if we need to upgrade our descriptor @d from write-only mode.
681 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen)
683 int is_snap, need_upgrade;
686 * Check if we've already upgraded or new filter is empty.
688 if (d->bd_writer == 0 || fcode == NULL)
694 * Check if cmd looks like snaplen setting from
695 * pcap_bpf.c:pcap_open_live().
696 * Note we're not checking .k value here:
697 * while pcap_open_live() definitely sets to non-zero value,
698 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g.
699 * do not consider upgrading immediately
701 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K))
708 * We're setting first filter and it doesn't look like
709 * setting snaplen. We're probably using bpf directly.
710 * Upgrade immediately.
715 * Do not require upgrade by first BIOCSETF
716 * (used to set snaplen) by pcap_open_live().
719 if (--d->bd_writer == 0) {
721 * First snaplen filter has already
722 * been set. This is probably catch-all
730 "%s: filter function set by pid %d, "
731 "bd_writer counter %d, snap %d upgrade %d",
732 __func__, d->bd_pid, d->bd_writer,
733 is_snap, need_upgrade);
735 return (need_upgrade);
739 * Add d to the list of active bp filters.
740 * Requires bpf_attachd() to be called before.
743 bpf_upgraded(struct bpf_d *d)
752 * Filter can be set several times without specifying interface.
753 * Mark d as reader and exit.
765 /* Remove from writers-only list */
766 LIST_REMOVE(d, bd_next);
767 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next);
768 /* Mark d as reader */
774 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid);
776 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1);
780 * Detach a file from its interface.
783 bpf_detachd(struct bpf_d *d)
786 bpf_detachd_locked(d);
791 bpf_detachd_locked(struct bpf_d *d)
797 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid);
801 /* Check if descriptor is attached */
802 if ((bp = d->bd_bif) == NULL)
808 /* Save bd_writer value */
809 error = d->bd_writer;
812 * Remove d from the interface's descriptor list.
814 LIST_REMOVE(d, bd_next);
823 /* Call event handler iff d is attached */
825 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0);
828 * Check if this descriptor had requested promiscuous mode.
829 * If so, turn it off.
833 CURVNET_SET(ifp->if_vnet);
834 error = ifpromisc(ifp, 0);
836 if (error != 0 && error != ENXIO) {
838 * ENXIO can happen if a pccard is unplugged
839 * Something is really wrong if we were able to put
840 * the driver into promiscuous mode, but can't
843 if_printf(bp->bif_ifp,
844 "bpf_detach: ifpromisc failed (%d)\n", error);
850 * Close the descriptor by detaching it from its interface,
851 * deallocating its buffers, and marking it free.
856 struct bpf_d *d = data;
859 if (d->bd_state == BPF_WAITING)
860 callout_stop(&d->bd_callout);
861 d->bd_state = BPF_IDLE;
863 funsetown(&d->bd_sigio);
866 mac_bpfdesc_destroy(d);
868 seldrain(&d->bd_sel);
869 knlist_destroy(&d->bd_sel.si_note);
870 callout_drain(&d->bd_callout);
876 * Open ethernet device. Returns ENXIO for illegal minor device number,
877 * EBUSY if file is open by another process.
881 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td)
886 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO);
887 error = devfs_set_cdevpriv(d, bpf_dtor);
894 * For historical reasons, perform a one-time initialization call to
895 * the buffer routines, even though we're not yet committed to a
896 * particular buffer method.
899 if ((flags & FREAD) == 0)
901 d->bd_hbuf_in_use = 0;
902 d->bd_bufmode = BPF_BUFMODE_BUFFER;
904 d->bd_direction = BPF_D_INOUT;
905 BPF_PID_REFRESH(d, td);
908 mac_bpfdesc_create(td->td_ucred, d);
910 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF);
911 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0);
912 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock);
918 * bpfread - read next chunk of packets from buffers
921 bpfread(struct cdev *dev, struct uio *uio, int ioflag)
928 error = devfs_get_cdevpriv((void **)&d);
933 * Restrict application to use a buffer the same size as
936 if (uio->uio_resid != d->bd_bufsize)
939 non_block = ((ioflag & O_NONBLOCK) != 0);
942 BPF_PID_REFRESH_CUR(d);
943 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) {
947 if (d->bd_state == BPF_WAITING)
948 callout_stop(&d->bd_callout);
949 timed_out = (d->bd_state == BPF_TIMED_OUT);
950 d->bd_state = BPF_IDLE;
951 while (d->bd_hbuf_in_use) {
952 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
953 PRINET|PCATCH, "bd_hbuf", 0);
960 * If the hold buffer is empty, then do a timed sleep, which
961 * ends when the timeout expires or when enough packets
962 * have arrived to fill the store buffer.
964 while (d->bd_hbuf == NULL) {
965 if (d->bd_slen != 0) {
967 * A packet(s) either arrived since the previous
968 * read or arrived while we were asleep.
970 if (d->bd_immediate || non_block || timed_out) {
972 * Rotate the buffers and return what's here
973 * if we are in immediate mode, non-blocking
974 * flag is set, or this descriptor timed out.
982 * No data is available, check to see if the bpf device
983 * is still pointed at a real interface. If not, return
984 * ENXIO so that the userland process knows to rebind
985 * it before using it again.
987 if (d->bd_bif == NULL) {
994 return (EWOULDBLOCK);
996 error = msleep(d, &d->bd_lock, PRINET|PCATCH,
998 if (error == EINTR || error == ERESTART) {
1002 if (error == EWOULDBLOCK) {
1004 * On a timeout, return what's in the buffer,
1005 * which may be nothing. If there is something
1006 * in the store buffer, we can rotate the buffers.
1010 * We filled up the buffer in between
1011 * getting the timeout and arriving
1012 * here, so we don't need to rotate.
1016 if (d->bd_slen == 0) {
1025 * At this point, we know we have something in the hold slot.
1027 d->bd_hbuf_in_use = 1;
1031 * Move data from hold buffer into user space.
1032 * We know the entire buffer is transferred since
1033 * we checked above that the read buffer is bpf_bufsize bytes.
1035 * We do not have to worry about simultaneous reads because
1036 * we waited for sole access to the hold buffer above.
1038 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio);
1041 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf"));
1042 d->bd_fbuf = d->bd_hbuf;
1045 bpf_buf_reclaimed(d);
1046 d->bd_hbuf_in_use = 0;
1047 wakeup(&d->bd_hbuf_in_use);
1054 * If there are processes sleeping on this descriptor, wake them up.
1056 static __inline void
1057 bpf_wakeup(struct bpf_d *d)
1060 BPFD_LOCK_ASSERT(d);
1061 if (d->bd_state == BPF_WAITING) {
1062 callout_stop(&d->bd_callout);
1063 d->bd_state = BPF_IDLE;
1066 if (d->bd_async && d->bd_sig && d->bd_sigio)
1067 pgsigio(&d->bd_sigio, d->bd_sig, 0);
1069 selwakeuppri(&d->bd_sel, PRINET);
1070 KNOTE_LOCKED(&d->bd_sel.si_note, 0);
1074 bpf_timed_out(void *arg)
1076 struct bpf_d *d = (struct bpf_d *)arg;
1078 BPFD_LOCK_ASSERT(d);
1080 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout))
1082 if (d->bd_state == BPF_WAITING) {
1083 d->bd_state = BPF_TIMED_OUT;
1084 if (d->bd_slen != 0)
1090 bpf_ready(struct bpf_d *d)
1093 BPFD_LOCK_ASSERT(d);
1095 if (!bpf_canfreebuf(d) && d->bd_hlen != 0)
1097 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) &&
1104 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag)
1108 struct mbuf *m, *mc;
1109 struct sockaddr dst;
1113 error = devfs_get_cdevpriv((void **)&d);
1117 BPF_PID_REFRESH_CUR(d);
1119 /* XXX: locking required */
1120 if (d->bd_bif == NULL) {
1125 ifp = d->bd_bif->bif_ifp;
1127 if ((ifp->if_flags & IFF_UP) == 0) {
1132 if (uio->uio_resid == 0) {
1137 bzero(&dst, sizeof(dst));
1140 /* XXX: bpf_movein() can sleep */
1141 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp,
1142 &m, &dst, &hlen, d);
1149 dst.sa_family = pseudo_AF_HDRCMPLT;
1151 if (d->bd_feedback) {
1152 mc = m_dup(m, M_NOWAIT);
1154 mc->m_pkthdr.rcvif = ifp;
1155 /* Set M_PROMISC for outgoing packets to be discarded. */
1156 if (d->bd_direction == BPF_D_INOUT)
1157 m->m_flags |= M_PROMISC;
1161 m->m_pkthdr.len -= hlen;
1163 m->m_data += hlen; /* XXX */
1165 CURVNET_SET(ifp->if_vnet);
1168 mac_bpfdesc_create_mbuf(d, m);
1170 mac_bpfdesc_create_mbuf(d, mc);
1174 bzero(&ro, sizeof(ro));
1176 ro.ro_prepend = (u_char *)&dst.sa_data;
1178 ro.ro_flags = RT_HAS_HEADER;
1181 error = (*ifp->if_output)(ifp, m, &dst, &ro);
1187 (*ifp->if_input)(ifp, mc);
1197 * Reset a descriptor by flushing its packet buffer and clearing the receive
1198 * and drop counts. This is doable for kernel-only buffers, but with
1199 * zero-copy buffers, we can't write to (or rotate) buffers that are
1200 * currently owned by userspace. It would be nice if we could encapsulate
1201 * this logic in the buffer code rather than here.
1204 reset_d(struct bpf_d *d)
1207 BPFD_LOCK_ASSERT(d);
1209 while (d->bd_hbuf_in_use)
1210 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET,
1212 if ((d->bd_hbuf != NULL) &&
1213 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) {
1214 /* Free the hold buffer. */
1215 d->bd_fbuf = d->bd_hbuf;
1218 bpf_buf_reclaimed(d);
1220 if (bpf_canwritebuf(d))
1232 * FIONREAD Check for read packet available.
1233 * BIOCGBLEN Get buffer len [for read()].
1234 * BIOCSETF Set read filter.
1235 * BIOCSETFNR Set read filter without resetting descriptor.
1236 * BIOCSETWF Set write filter.
1237 * BIOCFLUSH Flush read packet buffer.
1238 * BIOCPROMISC Put interface into promiscuous mode.
1239 * BIOCGDLT Get link layer type.
1240 * BIOCGETIF Get interface name.
1241 * BIOCSETIF Set interface.
1242 * BIOCSRTIMEOUT Set read timeout.
1243 * BIOCGRTIMEOUT Get read timeout.
1244 * BIOCGSTATS Get packet stats.
1245 * BIOCIMMEDIATE Set immediate mode.
1246 * BIOCVERSION Get filter language version.
1247 * BIOCGHDRCMPLT Get "header already complete" flag
1248 * BIOCSHDRCMPLT Set "header already complete" flag
1249 * BIOCGDIRECTION Get packet direction flag
1250 * BIOCSDIRECTION Set packet direction flag
1251 * BIOCGTSTAMP Get time stamp format and resolution.
1252 * BIOCSTSTAMP Set time stamp format and resolution.
1253 * BIOCLOCK Set "locked" flag
1254 * BIOCFEEDBACK Set packet feedback mode.
1255 * BIOCSETZBUF Set current zero-copy buffer locations.
1256 * BIOCGETZMAX Get maximum zero-copy buffer size.
1257 * BIOCROTZBUF Force rotation of zero-copy buffer
1258 * BIOCSETBUFMODE Set buffer mode.
1259 * BIOCGETBUFMODE Get current buffer mode.
1263 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags,
1269 error = devfs_get_cdevpriv((void **)&d);
1274 * Refresh PID associated with this descriptor.
1277 BPF_PID_REFRESH(d, td);
1278 if (d->bd_state == BPF_WAITING)
1279 callout_stop(&d->bd_callout);
1280 d->bd_state = BPF_IDLE;
1283 if (d->bd_locked == 1) {
1289 #ifdef COMPAT_FREEBSD32
1290 case BIOCGDLTLIST32:
1294 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1295 case BIOCGRTIMEOUT32:
1306 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1307 case BIOCSRTIMEOUT32:
1317 #ifdef COMPAT_FREEBSD32
1319 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so
1320 * that it will get 32-bit packet headers.
1326 case BIOCGDLTLIST32:
1327 case BIOCGRTIMEOUT32:
1328 case BIOCSRTIMEOUT32:
1335 CURVNET_SET(TD_TO_VNET(td));
1343 * Check for read packet available.
1351 while (d->bd_hbuf_in_use)
1352 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock,
1353 PRINET, "bd_hbuf", 0);
1363 * Get buffer len [for read()].
1367 *(u_int *)addr = d->bd_bufsize;
1372 * Set buffer length.
1375 error = bpf_ioctl_sblen(d, (u_int *)addr);
1379 * Set link layer read filter.
1384 #ifdef COMPAT_FREEBSD32
1389 error = bpf_setf(d, (struct bpf_program *)addr, cmd);
1393 * Flush read packet buffer.
1402 * Put interface into promiscuous mode.
1405 if (d->bd_bif == NULL) {
1407 * No interface attached yet.
1412 if (d->bd_promisc == 0) {
1413 error = ifpromisc(d->bd_bif->bif_ifp, 1);
1420 * Get current data link type.
1424 if (d->bd_bif == NULL)
1427 *(u_int *)addr = d->bd_bif->bif_dlt;
1432 * Get a list of supported data link types.
1434 #ifdef COMPAT_FREEBSD32
1435 case BIOCGDLTLIST32:
1437 struct bpf_dltlist32 *list32;
1438 struct bpf_dltlist dltlist;
1440 list32 = (struct bpf_dltlist32 *)addr;
1441 dltlist.bfl_len = list32->bfl_len;
1442 dltlist.bfl_list = PTRIN(list32->bfl_list);
1444 if (d->bd_bif == NULL)
1447 error = bpf_getdltlist(d, &dltlist);
1449 list32->bfl_len = dltlist.bfl_len;
1458 if (d->bd_bif == NULL)
1461 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr);
1466 * Set data link type.
1470 if (d->bd_bif == NULL)
1473 error = bpf_setdlt(d, *(u_int *)addr);
1478 * Get interface name.
1482 if (d->bd_bif == NULL)
1485 struct ifnet *const ifp = d->bd_bif->bif_ifp;
1486 struct ifreq *const ifr = (struct ifreq *)addr;
1488 strlcpy(ifr->ifr_name, ifp->if_xname,
1489 sizeof(ifr->ifr_name));
1499 int alloc_buf, size;
1502 * Behavior here depends on the buffering model. If
1503 * we're using kernel memory buffers, then we can
1504 * allocate them here. If we're using zero-copy,
1505 * then the user process must have registered buffers
1506 * by the time we get here.
1510 if (d->bd_bufmode == BPF_BUFMODE_BUFFER &&
1515 size = d->bd_bufsize;
1516 error = bpf_buffer_ioctl_sblen(d, &size);
1521 error = bpf_setif(d, (struct ifreq *)addr);
1530 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1531 case BIOCSRTIMEOUT32:
1534 struct timeval *tv = (struct timeval *)addr;
1535 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1536 struct timeval32 *tv32;
1537 struct timeval tv64;
1539 if (cmd == BIOCSRTIMEOUT32) {
1540 tv32 = (struct timeval32 *)addr;
1542 tv->tv_sec = tv32->tv_sec;
1543 tv->tv_usec = tv32->tv_usec;
1546 tv = (struct timeval *)addr;
1549 * Subtract 1 tick from tvtohz() since this isn't
1552 if ((error = itimerfix(tv)) == 0)
1553 d->bd_rtout = tvtohz(tv) - 1;
1561 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1562 case BIOCGRTIMEOUT32:
1566 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1567 struct timeval32 *tv32;
1568 struct timeval tv64;
1570 if (cmd == BIOCGRTIMEOUT32)
1574 tv = (struct timeval *)addr;
1576 tv->tv_sec = d->bd_rtout / hz;
1577 tv->tv_usec = (d->bd_rtout % hz) * tick;
1578 #if defined(COMPAT_FREEBSD32) && !defined(__mips__)
1579 if (cmd == BIOCGRTIMEOUT32) {
1580 tv32 = (struct timeval32 *)addr;
1581 tv32->tv_sec = tv->tv_sec;
1582 tv32->tv_usec = tv->tv_usec;
1594 struct bpf_stat *bs = (struct bpf_stat *)addr;
1596 /* XXXCSJP overflow */
1597 bs->bs_recv = d->bd_rcount;
1598 bs->bs_drop = d->bd_dcount;
1603 * Set immediate mode.
1607 d->bd_immediate = *(u_int *)addr;
1613 struct bpf_version *bv = (struct bpf_version *)addr;
1615 bv->bv_major = BPF_MAJOR_VERSION;
1616 bv->bv_minor = BPF_MINOR_VERSION;
1621 * Get "header already complete" flag
1625 *(u_int *)addr = d->bd_hdrcmplt;
1630 * Set "header already complete" flag
1634 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0;
1639 * Get packet direction flag
1641 case BIOCGDIRECTION:
1643 *(u_int *)addr = d->bd_direction;
1648 * Set packet direction flag
1650 case BIOCSDIRECTION:
1654 direction = *(u_int *)addr;
1655 switch (direction) {
1660 d->bd_direction = direction;
1670 * Get packet timestamp format and resolution.
1674 *(u_int *)addr = d->bd_tstamp;
1679 * Set packet timestamp format and resolution.
1685 func = *(u_int *)addr;
1686 if (BPF_T_VALID(func))
1687 d->bd_tstamp = func;
1695 d->bd_feedback = *(u_int *)addr;
1705 case FIONBIO: /* Non-blocking I/O */
1708 case FIOASYNC: /* Send signal on receive packets */
1710 d->bd_async = *(int *)addr;
1716 * XXX: Add some sort of locking here?
1717 * fsetown() can sleep.
1719 error = fsetown(*(int *)addr, &d->bd_sigio);
1724 *(int *)addr = fgetown(&d->bd_sigio);
1728 /* This is deprecated, FIOSETOWN should be used instead. */
1730 error = fsetown(-(*(int *)addr), &d->bd_sigio);
1733 /* This is deprecated, FIOGETOWN should be used instead. */
1735 *(int *)addr = -fgetown(&d->bd_sigio);
1738 case BIOCSRSIG: /* Set receive signal */
1742 sig = *(u_int *)addr;
1755 *(u_int *)addr = d->bd_sig;
1759 case BIOCGETBUFMODE:
1761 *(u_int *)addr = d->bd_bufmode;
1765 case BIOCSETBUFMODE:
1767 * Allow the buffering mode to be changed as long as we
1768 * haven't yet committed to a particular mode. Our
1769 * definition of commitment, for now, is whether or not a
1770 * buffer has been allocated or an interface attached, since
1771 * that's the point where things get tricky.
1773 switch (*(u_int *)addr) {
1774 case BPF_BUFMODE_BUFFER:
1777 case BPF_BUFMODE_ZBUF:
1778 if (bpf_zerocopy_enable)
1788 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL ||
1789 d->bd_fbuf != NULL || d->bd_bif != NULL) {
1794 d->bd_bufmode = *(u_int *)addr;
1799 error = bpf_ioctl_getzmax(td, d, (size_t *)addr);
1803 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr);
1807 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr);
1815 * Set d's packet filter program to fp. If this file already has a filter,
1816 * free it and replace it. Returns EINVAL for bogus requests.
1818 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls
1819 * since reading d->bd_bif can't be protected by d or interface lock due to
1822 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses
1823 * interface read lock to read all filers.
1827 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd)
1829 #ifdef COMPAT_FREEBSD32
1830 struct bpf_program fp_swab;
1831 struct bpf_program32 *fp32;
1833 struct bpf_insn *fcode, *old;
1835 bpf_jit_filter *jfunc, *ofunc;
1841 #ifdef COMPAT_FREEBSD32
1846 fp32 = (struct bpf_program32 *)fp;
1847 fp_swab.bf_len = fp32->bf_len;
1848 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns;
1864 jfunc = ofunc = NULL;
1869 * Check new filter validness before acquiring any locks.
1870 * Allocate memory for new filter, if needed.
1873 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0))
1875 size = flen * sizeof(*fp->bf_insns);
1877 /* We're setting up new filter. Copy and check actual data. */
1878 fcode = malloc(size, M_BPF, M_WAITOK);
1879 if (copyin(fp->bf_insns, fcode, size) != 0 ||
1880 !bpf_validate(fcode, flen)) {
1885 /* Filter is copied inside fcode and is perfectly valid. */
1886 jfunc = bpf_jitter(fcode, flen);
1893 * Set up new filter.
1894 * Protect filter change by interface lock.
1895 * Additionally, we are protected by global lock here.
1897 if (d->bd_bif != NULL)
1898 BPFIF_WLOCK(d->bd_bif);
1900 if (cmd == BIOCSETWF) {
1901 old = d->bd_wfilter;
1902 d->bd_wfilter = fcode;
1904 old = d->bd_rfilter;
1905 d->bd_rfilter = fcode;
1907 ofunc = d->bd_bfilter;
1908 d->bd_bfilter = jfunc;
1910 if (cmd == BIOCSETF)
1913 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen);
1916 if (d->bd_bif != NULL)
1917 BPFIF_WUNLOCK(d->bd_bif);
1922 bpf_destroy_jit_filter(ofunc);
1925 /* Move d to active readers list. */
1926 if (need_upgrade != 0)
1934 * Detach a file from its current interface (if attached at all) and attach
1935 * to the interface indicated by the name stored in ifr.
1936 * Return an errno or 0.
1939 bpf_setif(struct bpf_d *d, struct ifreq *ifr)
1942 struct ifnet *theywant;
1946 theywant = ifunit(ifr->ifr_name);
1947 if (theywant == NULL || theywant->if_bpf == NULL)
1950 bp = theywant->if_bpf;
1952 /* Check if interface is not being detached from BPF */
1954 if (bp->bif_flags & BPFIF_FLAG_DYING) {
1961 * At this point, we expect the buffer is already allocated. If not,
1964 switch (d->bd_bufmode) {
1965 case BPF_BUFMODE_BUFFER:
1966 case BPF_BUFMODE_ZBUF:
1967 if (d->bd_sbuf == NULL)
1972 panic("bpf_setif: bufmode %d", d->bd_bufmode);
1974 if (bp != d->bd_bif)
1983 * Support for select() and poll() system calls
1985 * Return true iff the specific operation will not block indefinitely.
1986 * Otherwise, return false but make a note that a selwakeup() must be done.
1989 bpfpoll(struct cdev *dev, int events, struct thread *td)
1994 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL)
1996 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM));
1999 * Refresh PID associated with this descriptor.
2001 revents = events & (POLLOUT | POLLWRNORM);
2003 BPF_PID_REFRESH(d, td);
2004 if (events & (POLLIN | POLLRDNORM)) {
2006 revents |= events & (POLLIN | POLLRDNORM);
2008 selrecord(td, &d->bd_sel);
2009 /* Start the read timeout if necessary. */
2010 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2011 callout_reset(&d->bd_callout, d->bd_rtout,
2013 d->bd_state = BPF_WAITING;
2022 * Support for kevent() system call. Register EVFILT_READ filters and
2023 * reject all others.
2026 bpfkqfilter(struct cdev *dev, struct knote *kn)
2030 if (devfs_get_cdevpriv((void **)&d) != 0 ||
2031 kn->kn_filter != EVFILT_READ)
2035 * Refresh PID associated with this descriptor.
2038 BPF_PID_REFRESH_CUR(d);
2039 kn->kn_fop = &bpfread_filtops;
2041 knlist_add(&d->bd_sel.si_note, kn, 1);
2048 filt_bpfdetach(struct knote *kn)
2050 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2052 knlist_remove(&d->bd_sel.si_note, kn, 0);
2056 filt_bpfread(struct knote *kn, long hint)
2058 struct bpf_d *d = (struct bpf_d *)kn->kn_hook;
2061 BPFD_LOCK_ASSERT(d);
2062 ready = bpf_ready(d);
2064 kn->kn_data = d->bd_slen;
2066 * Ignore the hold buffer if it is being copied to user space.
2068 if (!d->bd_hbuf_in_use && d->bd_hbuf)
2069 kn->kn_data += d->bd_hlen;
2070 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) {
2071 callout_reset(&d->bd_callout, d->bd_rtout,
2073 d->bd_state = BPF_WAITING;
2079 #define BPF_TSTAMP_NONE 0
2080 #define BPF_TSTAMP_FAST 1
2081 #define BPF_TSTAMP_NORMAL 2
2082 #define BPF_TSTAMP_EXTERN 3
2085 bpf_ts_quality(int tstype)
2088 if (tstype == BPF_T_NONE)
2089 return (BPF_TSTAMP_NONE);
2090 if ((tstype & BPF_T_FAST) != 0)
2091 return (BPF_TSTAMP_FAST);
2093 return (BPF_TSTAMP_NORMAL);
2097 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m)
2102 quality = bpf_ts_quality(tstype);
2103 if (quality == BPF_TSTAMP_NONE)
2107 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL);
2109 *bt = *(struct bintime *)(tag + 1);
2110 return (BPF_TSTAMP_EXTERN);
2113 if (quality == BPF_TSTAMP_NORMAL)
2122 * Incoming linkage from device drivers. Process the packet pkt, of length
2123 * pktlen, which is stored in a contiguous buffer. The packet is parsed
2124 * by each process' filter, and if accepted, stashed into the corresponding
2128 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2138 gottime = BPF_TSTAMP_NONE;
2142 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2144 * We are not using any locks for d here because:
2145 * 1) any filter change is protected by interface
2147 * 2) destroying/detaching d is protected by interface
2151 /* XXX: Do not protect counter for the sake of performance. */
2154 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no
2155 * way for the caller to indiciate to us whether this packet
2156 * is inbound or outbound. In the bpf_mtap() routines, we use
2157 * the interface pointers on the mbuf to figure it out.
2160 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2162 slen = (*(bf->func))(pkt, pktlen, pktlen);
2165 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen);
2168 * Filter matches. Let's to acquire write lock.
2173 if (gottime < bpf_ts_quality(d->bd_tstamp))
2174 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL);
2176 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2178 catchpacket(d, pkt, pktlen, slen,
2179 bpf_append_bytes, &bt);
2186 #define BPF_CHECK_DIRECTION(d, r, i) \
2187 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \
2188 ((d)->bd_direction == BPF_D_OUT && (r) == (i)))
2191 * Incoming linkage from device drivers, when packet is in an mbuf chain.
2192 * Locking model is explained in bpf_tap().
2195 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2205 /* Skip outgoing duplicate packets. */
2206 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2207 m->m_flags &= ~M_PROMISC;
2211 pktlen = m_length(m, NULL);
2212 gottime = BPF_TSTAMP_NONE;
2216 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2217 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2221 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL;
2222 /* XXX We cannot handle multiple mbufs. */
2223 if (bf != NULL && m->m_next == NULL)
2224 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen);
2227 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0);
2232 if (gottime < bpf_ts_quality(d->bd_tstamp))
2233 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2235 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2237 catchpacket(d, (u_char *)m, pktlen, slen,
2238 bpf_append_mbuf, &bt);
2246 * Incoming linkage from device drivers, when packet is in
2247 * an mbuf chain and to be prepended by a contiguous header.
2250 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m)
2258 /* Skip outgoing duplicate packets. */
2259 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) {
2260 m->m_flags &= ~M_PROMISC;
2264 pktlen = m_length(m, NULL);
2266 * Craft on-stack mbuf suitable for passing to bpf_filter.
2267 * Note that we cut corners here; we only setup what's
2268 * absolutely needed--this mbuf should never go anywhere else.
2275 gottime = BPF_TSTAMP_NONE;
2279 LIST_FOREACH(d, &bp->bif_dlist, bd_next) {
2280 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp))
2283 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0);
2288 if (gottime < bpf_ts_quality(d->bd_tstamp))
2289 gottime = bpf_gettime(&bt, d->bd_tstamp, m);
2291 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0)
2293 catchpacket(d, (u_char *)&mb, pktlen, slen,
2294 bpf_append_mbuf, &bt);
2301 #undef BPF_CHECK_DIRECTION
2303 #undef BPF_TSTAMP_NONE
2304 #undef BPF_TSTAMP_FAST
2305 #undef BPF_TSTAMP_NORMAL
2306 #undef BPF_TSTAMP_EXTERN
2309 bpf_hdrlen(struct bpf_d *d)
2313 hdrlen = d->bd_bif->bif_hdrlen;
2314 #ifndef BURN_BRIDGES
2315 if (d->bd_tstamp == BPF_T_NONE ||
2316 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME)
2317 #ifdef COMPAT_FREEBSD32
2319 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32);
2322 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr);
2325 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr);
2326 #ifdef COMPAT_FREEBSD32
2328 hdrlen = BPF_WORDALIGN32(hdrlen);
2331 hdrlen = BPF_WORDALIGN(hdrlen);
2333 return (hdrlen - d->bd_bif->bif_hdrlen);
2337 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype)
2339 struct bintime bt2, boottimebin;
2341 struct timespec tsn;
2343 if ((tstype & BPF_T_MONOTONIC) == 0) {
2345 getboottimebin(&boottimebin);
2346 bintime_add(&bt2, &boottimebin);
2349 switch (BPF_T_FORMAT(tstype)) {
2350 case BPF_T_MICROTIME:
2351 bintime2timeval(bt, &tsm);
2352 ts->bt_sec = tsm.tv_sec;
2353 ts->bt_frac = tsm.tv_usec;
2355 case BPF_T_NANOTIME:
2356 bintime2timespec(bt, &tsn);
2357 ts->bt_sec = tsn.tv_sec;
2358 ts->bt_frac = tsn.tv_nsec;
2361 ts->bt_sec = bt->sec;
2362 ts->bt_frac = bt->frac;
2368 * Move the packet data from interface memory (pkt) into the
2369 * store buffer. "cpfn" is the routine called to do the actual data
2370 * transfer. bcopy is passed in to copy contiguous chunks, while
2371 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case,
2372 * pkt is really an mbuf.
2375 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen,
2376 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int),
2379 struct bpf_xhdr hdr;
2380 #ifndef BURN_BRIDGES
2381 struct bpf_hdr hdr_old;
2382 #ifdef COMPAT_FREEBSD32
2383 struct bpf_hdr32 hdr32_old;
2386 int caplen, curlen, hdrlen, totlen;
2391 BPFD_LOCK_ASSERT(d);
2394 * Detect whether user space has released a buffer back to us, and if
2395 * so, move it from being a hold buffer to a free buffer. This may
2396 * not be the best place to do it (for example, we might only want to
2397 * run this check if we need the space), but for now it's a reliable
2400 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) {
2401 d->bd_fbuf = d->bd_hbuf;
2404 bpf_buf_reclaimed(d);
2408 * Figure out how many bytes to move. If the packet is
2409 * greater or equal to the snapshot length, transfer that
2410 * much. Otherwise, transfer the whole packet (unless
2411 * we hit the buffer size limit).
2413 hdrlen = bpf_hdrlen(d);
2414 totlen = hdrlen + min(snaplen, pktlen);
2415 if (totlen > d->bd_bufsize)
2416 totlen = d->bd_bufsize;
2419 * Round up the end of the previous packet to the next longword.
2421 * Drop the packet if there's no room and no hope of room
2422 * If the packet would overflow the storage buffer or the storage
2423 * buffer is considered immutable by the buffer model, try to rotate
2424 * the buffer and wakeup pending processes.
2426 #ifdef COMPAT_FREEBSD32
2428 curlen = BPF_WORDALIGN32(d->bd_slen);
2431 curlen = BPF_WORDALIGN(d->bd_slen);
2432 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) {
2433 if (d->bd_fbuf == NULL) {
2435 * There's no room in the store buffer, and no
2436 * prospect of room, so drop the packet. Notify the
2443 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use"));
2447 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT)
2449 * Immediate mode is set, or the read timeout has already
2450 * expired during a select call. A packet arrived, so the
2451 * reader should be woken up.
2454 caplen = totlen - hdrlen;
2455 tstype = d->bd_tstamp;
2456 do_timestamp = tstype != BPF_T_NONE;
2457 #ifndef BURN_BRIDGES
2458 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) {
2461 bpf_bintime2ts(bt, &ts, tstype);
2462 #ifdef COMPAT_FREEBSD32
2463 if (d->bd_compat32) {
2464 bzero(&hdr32_old, sizeof(hdr32_old));
2466 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec;
2467 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac;
2469 hdr32_old.bh_datalen = pktlen;
2470 hdr32_old.bh_hdrlen = hdrlen;
2471 hdr32_old.bh_caplen = caplen;
2472 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old,
2477 bzero(&hdr_old, sizeof(hdr_old));
2479 hdr_old.bh_tstamp.tv_sec = ts.bt_sec;
2480 hdr_old.bh_tstamp.tv_usec = ts.bt_frac;
2482 hdr_old.bh_datalen = pktlen;
2483 hdr_old.bh_hdrlen = hdrlen;
2484 hdr_old.bh_caplen = caplen;
2485 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old,
2492 * Append the bpf header. Note we append the actual header size, but
2493 * move forward the length of the header plus padding.
2495 bzero(&hdr, sizeof(hdr));
2497 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype);
2498 hdr.bh_datalen = pktlen;
2499 hdr.bh_hdrlen = hdrlen;
2500 hdr.bh_caplen = caplen;
2501 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr));
2504 * Copy the packet data into the store buffer and update its length.
2506 #ifndef BURN_BRIDGES
2509 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen);
2510 d->bd_slen = curlen + totlen;
2517 * Free buffers currently in use by a descriptor.
2521 bpf_freed(struct bpf_d *d)
2525 * We don't need to lock out interrupts since this descriptor has
2526 * been detached from its interface and it yet hasn't been marked
2530 if (d->bd_rfilter != NULL) {
2531 free((caddr_t)d->bd_rfilter, M_BPF);
2533 if (d->bd_bfilter != NULL)
2534 bpf_destroy_jit_filter(d->bd_bfilter);
2537 if (d->bd_wfilter != NULL)
2538 free((caddr_t)d->bd_wfilter, M_BPF);
2539 mtx_destroy(&d->bd_lock);
2543 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the
2544 * fixed size of the link header (variable length headers not yet supported).
2547 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2550 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2554 * Attach an interface to bpf. ifp is a pointer to the structure
2555 * defining the interface to be attached, dlt is the link layer type,
2556 * and hdrlen is the fixed size of the link header (variable length
2557 * headers are not yet supporrted).
2560 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2564 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO);
2568 LIST_INIT(&bp->bif_dlist);
2569 LIST_INIT(&bp->bif_wlist);
2572 rw_init(&bp->bif_lock, "bpf interface lock");
2573 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized"));
2574 bp->bif_bpf = driverp;
2578 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next);
2581 bp->bif_hdrlen = hdrlen;
2583 if (bootverbose && IS_DEFAULT_VNET(curvnet))
2584 if_printf(ifp, "bpf attached\n");
2589 * When moving interfaces between vnet instances we need a way to
2590 * query the dlt and hdrlen before detach so we can re-attch the if_bpf
2591 * after the vmove. We unfortunately have no device driver infrastructure
2592 * to query the interface for these values after creation/attach, thus
2593 * add this as a workaround.
2596 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen)
2601 if (bif_dlt == NULL && bif_hdrlen == NULL)
2604 if (bif_dlt != NULL)
2605 *bif_dlt = bp->bif_dlt;
2606 if (bif_hdrlen != NULL)
2607 *bif_hdrlen = bp->bif_hdrlen;
2614 * Detach bpf from an interface. This involves detaching each descriptor
2615 * associated with the interface. Notify each descriptor as it's detached
2616 * so that any sleepers wake up and get ENXIO.
2619 bpfdetach(struct ifnet *ifp)
2621 struct bpf_if *bp, *bp_temp;
2628 /* Find all bpf_if struct's which reference ifp and detach them. */
2629 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) {
2630 if (ifp != bp->bif_ifp)
2633 LIST_REMOVE(bp, bif_next);
2634 /* Add to to-be-freed list */
2635 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next);
2639 * Delay freeing bp till interface is detached
2640 * and all routes through this interface are removed.
2641 * Mark bp as detached to restrict new consumers.
2644 bp->bif_flags |= BPFIF_FLAG_DYING;
2645 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if;
2648 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p",
2649 __func__, bp->bif_dlt, bp, ifp);
2651 /* Free common descriptors */
2652 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) {
2653 bpf_detachd_locked(d);
2659 /* Free writer-only descriptors */
2660 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) {
2661 bpf_detachd_locked(d);
2671 printf("bpfdetach: %s was not attached\n", ifp->if_xname);
2676 * Interface departure handler.
2677 * Note departure event does not guarantee interface is going down.
2678 * Interface renaming is currently done via departure/arrival event set.
2680 * Departure handled is called after all routes pointing to
2681 * given interface are removed and interface is in down state
2682 * restricting any packets to be sent/received. We assume it is now safe
2683 * to free data allocated by BPF.
2686 bpf_ifdetach(void *arg __unused, struct ifnet *ifp)
2688 struct bpf_if *bp, *bp_temp;
2691 /* Ignore ifnet renaming. */
2692 if (ifp->if_flags & IFF_RENAMING)
2697 * Find matching entries in free list.
2698 * Nothing should be found if bpfdetach() was not called.
2700 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) {
2701 if (ifp != bp->bif_ifp)
2704 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p",
2707 LIST_REMOVE(bp, bif_next);
2709 rw_destroy(&bp->bif_lock);
2718 * Get a list of available data link type of the interface.
2721 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl)
2730 ifp = d->bd_bif->bif_ifp;
2733 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2734 if (bp->bif_ifp == ifp)
2737 if (bfl->bfl_list == NULL) {
2741 if (n1 > bfl->bfl_len)
2744 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK);
2747 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2748 if (bp->bif_ifp != ifp)
2754 lst[n] = bp->bif_dlt;
2758 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n);
2766 * Set the data link type of a BPF instance.
2769 bpf_setdlt(struct bpf_d *d, u_int dlt)
2771 int error, opromisc;
2777 if (d->bd_bif->bif_dlt == dlt)
2779 ifp = d->bd_bif->bif_ifp;
2781 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2782 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt)
2787 opromisc = d->bd_promisc;
2793 error = ifpromisc(bp->bif_ifp, 1);
2795 if_printf(bp->bif_ifp,
2796 "bpf_setdlt: ifpromisc failed (%d)\n",
2802 return (bp == NULL ? EINVAL : 0);
2806 bpf_drvinit(void *unused)
2810 sx_init(&bpf_sx, "bpf global lock");
2811 LIST_INIT(&bpf_iflist);
2812 LIST_INIT(&bpf_freelist);
2814 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf");
2815 /* For compatibility */
2816 make_dev_alias(dev, "bpf0");
2818 /* Register interface departure handler */
2819 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER(
2820 ifnet_departure_event, bpf_ifdetach, NULL,
2821 EVENTHANDLER_PRI_ANY);
2825 * Zero out the various packet counters associated with all of the bpf
2826 * descriptors. At some point, we will probably want to get a bit more
2827 * granular and allow the user to specify descriptors to be zeroed.
2830 bpf_zero_counters(void)
2836 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2838 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2854 * Fill filter statistics
2857 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd)
2860 bzero(d, sizeof(*d));
2861 BPFD_LOCK_ASSERT(bd);
2862 d->bd_structsize = sizeof(*d);
2863 /* XXX: reading should be protected by global lock */
2864 d->bd_immediate = bd->bd_immediate;
2865 d->bd_promisc = bd->bd_promisc;
2866 d->bd_hdrcmplt = bd->bd_hdrcmplt;
2867 d->bd_direction = bd->bd_direction;
2868 d->bd_feedback = bd->bd_feedback;
2869 d->bd_async = bd->bd_async;
2870 d->bd_rcount = bd->bd_rcount;
2871 d->bd_dcount = bd->bd_dcount;
2872 d->bd_fcount = bd->bd_fcount;
2873 d->bd_sig = bd->bd_sig;
2874 d->bd_slen = bd->bd_slen;
2875 d->bd_hlen = bd->bd_hlen;
2876 d->bd_bufsize = bd->bd_bufsize;
2877 d->bd_pid = bd->bd_pid;
2878 strlcpy(d->bd_ifname,
2879 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ);
2880 d->bd_locked = bd->bd_locked;
2881 d->bd_wcount = bd->bd_wcount;
2882 d->bd_wdcount = bd->bd_wdcount;
2883 d->bd_wfcount = bd->bd_wfcount;
2884 d->bd_zcopy = bd->bd_zcopy;
2885 d->bd_bufmode = bd->bd_bufmode;
2889 * Handle `netstat -B' stats request
2892 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS)
2894 static const struct xbpf_d zerostats;
2895 struct xbpf_d *xbdbuf, *xbd, tempstats;
2901 * XXX This is not technically correct. It is possible for non
2902 * privileged users to open bpf devices. It would make sense
2903 * if the users who opened the devices were able to retrieve
2904 * the statistics for them, too.
2906 error = priv_check(req->td, PRIV_NET_BPF);
2910 * Check to see if the user is requesting that the counters be
2911 * zeroed out. Explicitly check that the supplied data is zeroed,
2912 * as we aren't allowing the user to set the counters currently.
2914 if (req->newptr != NULL) {
2915 if (req->newlen != sizeof(tempstats))
2917 memset(&tempstats, 0, sizeof(tempstats));
2918 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats));
2921 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0)
2923 bpf_zero_counters();
2926 if (req->oldptr == NULL)
2927 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd)));
2928 if (bpf_bpfd_cnt == 0)
2929 return (SYSCTL_OUT(req, 0, 0));
2930 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK);
2932 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) {
2934 free(xbdbuf, M_BPF);
2938 LIST_FOREACH(bp, &bpf_iflist, bif_next) {
2940 /* Send writers-only first */
2941 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) {
2942 xbd = &xbdbuf[index++];
2944 bpfstats_fill_xbpf(xbd, bd);
2947 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) {
2948 xbd = &xbdbuf[index++];
2950 bpfstats_fill_xbpf(xbd, bd);
2956 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd));
2957 free(xbdbuf, M_BPF);
2961 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL);
2963 #else /* !DEV_BPF && !NETGRAPH_BPF */
2966 * NOP stubs to allow bpf-using drivers to load and function.
2968 * A 'better' implementation would allow the core bpf functionality
2969 * to be loaded at runtime.
2973 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen)
2978 bpf_mtap(struct bpf_if *bp, struct mbuf *m)
2983 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m)
2988 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen)
2991 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf);
2995 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp)
2998 *driverp = (struct bpf_if *)&dead_bpf_if;
3002 bpfdetach(struct ifnet *ifp)
3007 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen)
3009 return -1; /* "no filter" behaviour */
3013 bpf_validate(const struct bpf_insn *f, int len)
3015 return 0; /* false */
3018 #endif /* !DEV_BPF && !NETGRAPH_BPF */
3022 bpf_show_bpf_if(struct bpf_if *bpf_if)
3027 db_printf("%p:\n", bpf_if);
3028 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e);
3029 /* bif_ext.bif_next */
3030 /* bif_ext.bif_dlist */
3031 BPF_DB_PRINTF("%#x", bif_dlt);
3032 BPF_DB_PRINTF("%u", bif_hdrlen);
3033 BPF_DB_PRINTF("%p", bif_ifp);
3036 BPF_DB_PRINTF("%#x", bif_flags);
3039 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if)
3043 db_printf("usage: show bpf_if <struct bpf_if *>\n");
3047 bpf_show_bpf_if((struct bpf_if *)addr);