bpf.c revision 244090
1/*- 2 * Copyright (c) 1990, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 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 8 * Berkeley Laboratory. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 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. 21 * 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 32 * SUCH DAMAGE. 33 * 34 * @(#)bpf.c 8.4 (Berkeley) 1/9/95 35 */ 36 37#include <sys/cdefs.h> 38__FBSDID("$FreeBSD: head/sys/net/bpf.c 244090 2012-12-10 16:14:44Z ghelmer $"); 39 40#include "opt_bpf.h" 41#include "opt_compat.h" 42#include "opt_netgraph.h" 43 44#include <sys/types.h> 45#include <sys/param.h> 46#include <sys/lock.h> 47#include <sys/rwlock.h> 48#include <sys/systm.h> 49#include <sys/conf.h> 50#include <sys/fcntl.h> 51#include <sys/jail.h> 52#include <sys/malloc.h> 53#include <sys/mbuf.h> 54#include <sys/time.h> 55#include <sys/priv.h> 56#include <sys/proc.h> 57#include <sys/signalvar.h> 58#include <sys/filio.h> 59#include <sys/sockio.h> 60#include <sys/ttycom.h> 61#include <sys/uio.h> 62 63#include <sys/event.h> 64#include <sys/file.h> 65#include <sys/poll.h> 66#include <sys/proc.h> 67 68#include <sys/socket.h> 69 70#include <net/if.h> 71#define BPF_INTERNAL 72#include <net/bpf.h> 73#include <net/bpf_buffer.h> 74#ifdef BPF_JITTER 75#include <net/bpf_jitter.h> 76#endif 77#include <net/bpf_zerocopy.h> 78#include <net/bpfdesc.h> 79#include <net/vnet.h> 80 81#include <netinet/in.h> 82#include <netinet/if_ether.h> 83#include <sys/kernel.h> 84#include <sys/sysctl.h> 85 86#include <net80211/ieee80211_freebsd.h> 87 88#include <security/mac/mac_framework.h> 89 90MALLOC_DEFINE(M_BPF, "BPF", "BPF data"); 91 92#if defined(DEV_BPF) || defined(NETGRAPH_BPF) 93 94#define PRINET 26 /* interruptible */ 95 96#define SIZEOF_BPF_HDR(type) \ 97 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) 98 99#ifdef COMPAT_FREEBSD32 100#include <sys/mount.h> 101#include <compat/freebsd32/freebsd32.h> 102#define BPF_ALIGNMENT32 sizeof(int32_t) 103#define BPF_WORDALIGN32(x) (((x)+(BPF_ALIGNMENT32-1))&~(BPF_ALIGNMENT32-1)) 104 105#ifndef BURN_BRIDGES 106/* 107 * 32-bit version of structure prepended to each packet. We use this header 108 * instead of the standard one for 32-bit streams. We mark the a stream as 109 * 32-bit the first time we see a 32-bit compat ioctl request. 110 */ 111struct bpf_hdr32 { 112 struct timeval32 bh_tstamp; /* time stamp */ 113 uint32_t bh_caplen; /* length of captured portion */ 114 uint32_t bh_datalen; /* original length of packet */ 115 uint16_t bh_hdrlen; /* length of bpf header (this struct 116 plus alignment padding) */ 117}; 118#endif 119 120struct bpf_program32 { 121 u_int bf_len; 122 uint32_t bf_insns; 123}; 124 125struct bpf_dltlist32 { 126 u_int bfl_len; 127 u_int bfl_list; 128}; 129 130#define BIOCSETF32 _IOW('B', 103, struct bpf_program32) 131#define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) 132#define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) 133#define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) 134#define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) 135#define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) 136#endif 137 138/* 139 * bpf_iflist is a list of BPF interface structures, each corresponding to a 140 * specific DLT. The same network interface might have several BPF interface 141 * structures registered by different layers in the stack (i.e., 802.11 142 * frames, ethernet frames, etc). 143 */ 144static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist; 145static struct mtx bpf_mtx; /* bpf global lock */ 146static int bpf_bpfd_cnt; 147 148static void bpf_attachd(struct bpf_d *, struct bpf_if *); 149static void bpf_detachd(struct bpf_d *); 150static void bpf_detachd_locked(struct bpf_d *); 151static void bpf_freed(struct bpf_d *); 152static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, 153 struct sockaddr *, int *, struct bpf_insn *); 154static int bpf_setif(struct bpf_d *, struct ifreq *); 155static void bpf_timed_out(void *); 156static __inline void 157 bpf_wakeup(struct bpf_d *); 158static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 159 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), 160 struct bintime *); 161static void reset_d(struct bpf_d *); 162static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); 163static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 164static int bpf_setdlt(struct bpf_d *, u_int); 165static void filt_bpfdetach(struct knote *); 166static int filt_bpfread(struct knote *, long); 167static void bpf_drvinit(void *); 168static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); 169 170SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); 171int bpf_maxinsns = BPF_MAXINSNS; 172SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, 173 &bpf_maxinsns, 0, "Maximum bpf program instructions"); 174static int bpf_zerocopy_enable = 0; 175SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, 176 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); 177static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, 178 bpf_stats_sysctl, "bpf statistics portal"); 179 180static VNET_DEFINE(int, bpf_optimize_writers) = 0; 181#define V_bpf_optimize_writers VNET(bpf_optimize_writers) 182SYSCTL_VNET_INT(_net_bpf, OID_AUTO, optimize_writers, 183 CTLFLAG_RW, &VNET_NAME(bpf_optimize_writers), 0, 184 "Do not send packets until BPF program is set"); 185 186static d_open_t bpfopen; 187static d_read_t bpfread; 188static d_write_t bpfwrite; 189static d_ioctl_t bpfioctl; 190static d_poll_t bpfpoll; 191static d_kqfilter_t bpfkqfilter; 192 193static struct cdevsw bpf_cdevsw = { 194 .d_version = D_VERSION, 195 .d_open = bpfopen, 196 .d_read = bpfread, 197 .d_write = bpfwrite, 198 .d_ioctl = bpfioctl, 199 .d_poll = bpfpoll, 200 .d_name = "bpf", 201 .d_kqfilter = bpfkqfilter, 202}; 203 204static struct filterops bpfread_filtops = { 205 .f_isfd = 1, 206 .f_detach = filt_bpfdetach, 207 .f_event = filt_bpfread, 208}; 209 210eventhandler_tag bpf_ifdetach_cookie = NULL; 211 212/* 213 * LOCKING MODEL USED BY BPF: 214 * Locks: 215 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal, 216 * some global counters and every bpf_if reference. 217 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters. 218 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields 219 * used by bpf_mtap code. 220 * 221 * Lock order: 222 * 223 * Global lock, interface lock, descriptor lock 224 * 225 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2] 226 * working model. In many places (like bpf_detachd) we start with BPF descriptor 227 * (and we need to at least rlock it to get reliable interface pointer). This 228 * gives us potential LOR. As a result, we use global lock to protect from bpf_if 229 * change in every such place. 230 * 231 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and 232 * 3) descriptor main wlock. 233 * Reading bd_bif can be protected by any of these locks, typically global lock. 234 * 235 * Changing read/write BPF filter is protected by the same three locks, 236 * the same applies for reading. 237 * 238 * Sleeping in global lock is not allowed due to bpfdetach() using it. 239 */ 240 241/* 242 * Wrapper functions for various buffering methods. If the set of buffer 243 * modes expands, we will probably want to introduce a switch data structure 244 * similar to protosw, et. 245 */ 246static void 247bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 248 u_int len) 249{ 250 251 BPFD_LOCK_ASSERT(d); 252 253 switch (d->bd_bufmode) { 254 case BPF_BUFMODE_BUFFER: 255 return (bpf_buffer_append_bytes(d, buf, offset, src, len)); 256 257 case BPF_BUFMODE_ZBUF: 258 d->bd_zcopy++; 259 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); 260 261 default: 262 panic("bpf_buf_append_bytes"); 263 } 264} 265 266static void 267bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 268 u_int len) 269{ 270 271 BPFD_LOCK_ASSERT(d); 272 273 switch (d->bd_bufmode) { 274 case BPF_BUFMODE_BUFFER: 275 return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); 276 277 case BPF_BUFMODE_ZBUF: 278 d->bd_zcopy++; 279 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); 280 281 default: 282 panic("bpf_buf_append_mbuf"); 283 } 284} 285 286/* 287 * This function gets called when the free buffer is re-assigned. 288 */ 289static void 290bpf_buf_reclaimed(struct bpf_d *d) 291{ 292 293 BPFD_LOCK_ASSERT(d); 294 295 switch (d->bd_bufmode) { 296 case BPF_BUFMODE_BUFFER: 297 return; 298 299 case BPF_BUFMODE_ZBUF: 300 bpf_zerocopy_buf_reclaimed(d); 301 return; 302 303 default: 304 panic("bpf_buf_reclaimed"); 305 } 306} 307 308/* 309 * If the buffer mechanism has a way to decide that a held buffer can be made 310 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is 311 * returned if the buffer can be discarded, (0) is returned if it cannot. 312 */ 313static int 314bpf_canfreebuf(struct bpf_d *d) 315{ 316 317 BPFD_LOCK_ASSERT(d); 318 319 switch (d->bd_bufmode) { 320 case BPF_BUFMODE_ZBUF: 321 return (bpf_zerocopy_canfreebuf(d)); 322 } 323 return (0); 324} 325 326/* 327 * Allow the buffer model to indicate that the current store buffer is 328 * immutable, regardless of the appearance of space. Return (1) if the 329 * buffer is writable, and (0) if not. 330 */ 331static int 332bpf_canwritebuf(struct bpf_d *d) 333{ 334 BPFD_LOCK_ASSERT(d); 335 336 switch (d->bd_bufmode) { 337 case BPF_BUFMODE_ZBUF: 338 return (bpf_zerocopy_canwritebuf(d)); 339 } 340 return (1); 341} 342 343/* 344 * Notify buffer model that an attempt to write to the store buffer has 345 * resulted in a dropped packet, in which case the buffer may be considered 346 * full. 347 */ 348static void 349bpf_buffull(struct bpf_d *d) 350{ 351 352 BPFD_LOCK_ASSERT(d); 353 354 switch (d->bd_bufmode) { 355 case BPF_BUFMODE_ZBUF: 356 bpf_zerocopy_buffull(d); 357 break; 358 } 359} 360 361/* 362 * Notify the buffer model that a buffer has moved into the hold position. 363 */ 364void 365bpf_bufheld(struct bpf_d *d) 366{ 367 368 BPFD_LOCK_ASSERT(d); 369 370 switch (d->bd_bufmode) { 371 case BPF_BUFMODE_ZBUF: 372 bpf_zerocopy_bufheld(d); 373 break; 374 } 375} 376 377static void 378bpf_free(struct bpf_d *d) 379{ 380 381 switch (d->bd_bufmode) { 382 case BPF_BUFMODE_BUFFER: 383 return (bpf_buffer_free(d)); 384 385 case BPF_BUFMODE_ZBUF: 386 return (bpf_zerocopy_free(d)); 387 388 default: 389 panic("bpf_buf_free"); 390 } 391} 392 393static int 394bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) 395{ 396 397 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 398 return (EOPNOTSUPP); 399 return (bpf_buffer_uiomove(d, buf, len, uio)); 400} 401 402static int 403bpf_ioctl_sblen(struct bpf_d *d, u_int *i) 404{ 405 406 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 407 return (EOPNOTSUPP); 408 return (bpf_buffer_ioctl_sblen(d, i)); 409} 410 411static int 412bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) 413{ 414 415 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 416 return (EOPNOTSUPP); 417 return (bpf_zerocopy_ioctl_getzmax(td, d, i)); 418} 419 420static int 421bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 422{ 423 424 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 425 return (EOPNOTSUPP); 426 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); 427} 428 429static int 430bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 431{ 432 433 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 434 return (EOPNOTSUPP); 435 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); 436} 437 438/* 439 * General BPF functions. 440 */ 441static int 442bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, 443 struct sockaddr *sockp, int *hdrlen, struct bpf_insn *wfilter) 444{ 445 const struct ieee80211_bpf_params *p; 446 struct ether_header *eh; 447 struct mbuf *m; 448 int error; 449 int len; 450 int hlen; 451 int slen; 452 453 /* 454 * Build a sockaddr based on the data link layer type. 455 * We do this at this level because the ethernet header 456 * is copied directly into the data field of the sockaddr. 457 * In the case of SLIP, there is no header and the packet 458 * is forwarded as is. 459 * Also, we are careful to leave room at the front of the mbuf 460 * for the link level header. 461 */ 462 switch (linktype) { 463 464 case DLT_SLIP: 465 sockp->sa_family = AF_INET; 466 hlen = 0; 467 break; 468 469 case DLT_EN10MB: 470 sockp->sa_family = AF_UNSPEC; 471 /* XXX Would MAXLINKHDR be better? */ 472 hlen = ETHER_HDR_LEN; 473 break; 474 475 case DLT_FDDI: 476 sockp->sa_family = AF_IMPLINK; 477 hlen = 0; 478 break; 479 480 case DLT_RAW: 481 sockp->sa_family = AF_UNSPEC; 482 hlen = 0; 483 break; 484 485 case DLT_NULL: 486 /* 487 * null interface types require a 4 byte pseudo header which 488 * corresponds to the address family of the packet. 489 */ 490 sockp->sa_family = AF_UNSPEC; 491 hlen = 4; 492 break; 493 494 case DLT_ATM_RFC1483: 495 /* 496 * en atm driver requires 4-byte atm pseudo header. 497 * though it isn't standard, vpi:vci needs to be 498 * specified anyway. 499 */ 500 sockp->sa_family = AF_UNSPEC; 501 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ 502 break; 503 504 case DLT_PPP: 505 sockp->sa_family = AF_UNSPEC; 506 hlen = 4; /* This should match PPP_HDRLEN */ 507 break; 508 509 case DLT_IEEE802_11: /* IEEE 802.11 wireless */ 510 sockp->sa_family = AF_IEEE80211; 511 hlen = 0; 512 break; 513 514 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ 515 sockp->sa_family = AF_IEEE80211; 516 sockp->sa_len = 12; /* XXX != 0 */ 517 hlen = sizeof(struct ieee80211_bpf_params); 518 break; 519 520 default: 521 return (EIO); 522 } 523 524 len = uio->uio_resid; 525 526 if (len - hlen > ifp->if_mtu) 527 return (EMSGSIZE); 528 529 if ((unsigned)len > MJUM16BYTES) 530 return (EIO); 531 532 if (len <= MHLEN) 533 MGETHDR(m, M_WAITOK, MT_DATA); 534 else if (len <= MCLBYTES) 535 m = m_getcl(M_WAITOK, MT_DATA, M_PKTHDR); 536 else 537 m = m_getjcl(M_WAITOK, MT_DATA, M_PKTHDR, 538#if (MJUMPAGESIZE > MCLBYTES) 539 len <= MJUMPAGESIZE ? MJUMPAGESIZE : 540#endif 541 (len <= MJUM9BYTES ? MJUM9BYTES : MJUM16BYTES)); 542 m->m_pkthdr.len = m->m_len = len; 543 m->m_pkthdr.rcvif = NULL; 544 *mp = m; 545 546 if (m->m_len < hlen) { 547 error = EPERM; 548 goto bad; 549 } 550 551 error = uiomove(mtod(m, u_char *), len, uio); 552 if (error) 553 goto bad; 554 555 slen = bpf_filter(wfilter, mtod(m, u_char *), len, len); 556 if (slen == 0) { 557 error = EPERM; 558 goto bad; 559 } 560 561 /* Check for multicast destination */ 562 switch (linktype) { 563 case DLT_EN10MB: 564 eh = mtod(m, struct ether_header *); 565 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 566 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 567 ETHER_ADDR_LEN) == 0) 568 m->m_flags |= M_BCAST; 569 else 570 m->m_flags |= M_MCAST; 571 } 572 break; 573 } 574 575 /* 576 * Make room for link header, and copy it to sockaddr 577 */ 578 if (hlen != 0) { 579 if (sockp->sa_family == AF_IEEE80211) { 580 /* 581 * Collect true length from the parameter header 582 * NB: sockp is known to be zero'd so if we do a 583 * short copy unspecified parameters will be 584 * zero. 585 * NB: packet may not be aligned after stripping 586 * bpf params 587 * XXX check ibp_vers 588 */ 589 p = mtod(m, const struct ieee80211_bpf_params *); 590 hlen = p->ibp_len; 591 if (hlen > sizeof(sockp->sa_data)) { 592 error = EINVAL; 593 goto bad; 594 } 595 } 596 bcopy(m->m_data, sockp->sa_data, hlen); 597 } 598 *hdrlen = hlen; 599 600 return (0); 601bad: 602 m_freem(m); 603 return (error); 604} 605 606/* 607 * Attach file to the bpf interface, i.e. make d listen on bp. 608 */ 609static void 610bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 611{ 612 int op_w; 613 614 BPF_LOCK_ASSERT(); 615 616 /* 617 * Save sysctl value to protect from sysctl change 618 * between reads 619 */ 620 op_w = V_bpf_optimize_writers; 621 622 if (d->bd_bif != NULL) 623 bpf_detachd_locked(d); 624 /* 625 * Point d at bp, and add d to the interface's list. 626 * Since there are many applicaiotns using BPF for 627 * sending raw packets only (dhcpd, cdpd are good examples) 628 * we can delay adding d to the list of active listeners until 629 * some filter is configured. 630 */ 631 632 BPFIF_WLOCK(bp); 633 BPFD_LOCK(d); 634 635 d->bd_bif = bp; 636 637 if (op_w != 0) { 638 /* Add to writers-only list */ 639 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); 640 /* 641 * We decrement bd_writer on every filter set operation. 642 * First BIOCSETF is done by pcap_open_live() to set up 643 * snap length. After that appliation usually sets its own filter 644 */ 645 d->bd_writer = 2; 646 } else 647 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 648 649 BPFD_UNLOCK(d); 650 BPFIF_WUNLOCK(bp); 651 652 bpf_bpfd_cnt++; 653 654 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", 655 __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); 656 657 if (op_w == 0) 658 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 659} 660 661/* 662 * Add d to the list of active bp filters. 663 * Reuqires bpf_attachd() to be called before 664 */ 665static void 666bpf_upgraded(struct bpf_d *d) 667{ 668 struct bpf_if *bp; 669 670 BPF_LOCK_ASSERT(); 671 672 bp = d->bd_bif; 673 674 /* 675 * Filter can be set several times without specifying interface. 676 * Mark d as reader and exit. 677 */ 678 if (bp == NULL) { 679 BPFD_LOCK(d); 680 d->bd_writer = 0; 681 BPFD_UNLOCK(d); 682 return; 683 } 684 685 BPFIF_WLOCK(bp); 686 BPFD_LOCK(d); 687 688 /* Remove from writers-only list */ 689 LIST_REMOVE(d, bd_next); 690 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 691 /* Mark d as reader */ 692 d->bd_writer = 0; 693 694 BPFD_UNLOCK(d); 695 BPFIF_WUNLOCK(bp); 696 697 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); 698 699 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 700} 701 702/* 703 * Detach a file from its interface. 704 */ 705static void 706bpf_detachd(struct bpf_d *d) 707{ 708 BPF_LOCK(); 709 bpf_detachd_locked(d); 710 BPF_UNLOCK(); 711} 712 713static void 714bpf_detachd_locked(struct bpf_d *d) 715{ 716 int error; 717 struct bpf_if *bp; 718 struct ifnet *ifp; 719 720 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); 721 722 BPF_LOCK_ASSERT(); 723 724 /* Check if descriptor is attached */ 725 if ((bp = d->bd_bif) == NULL) 726 return; 727 728 BPFIF_WLOCK(bp); 729 BPFD_LOCK(d); 730 731 /* Save bd_writer value */ 732 error = d->bd_writer; 733 734 /* 735 * Remove d from the interface's descriptor list. 736 */ 737 LIST_REMOVE(d, bd_next); 738 739 ifp = bp->bif_ifp; 740 d->bd_bif = NULL; 741 BPFD_UNLOCK(d); 742 BPFIF_WUNLOCK(bp); 743 744 bpf_bpfd_cnt--; 745 746 /* Call event handler iff d is attached */ 747 if (error == 0) 748 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); 749 750 /* 751 * Check if this descriptor had requested promiscuous mode. 752 * If so, turn it off. 753 */ 754 if (d->bd_promisc) { 755 d->bd_promisc = 0; 756 CURVNET_SET(ifp->if_vnet); 757 error = ifpromisc(ifp, 0); 758 CURVNET_RESTORE(); 759 if (error != 0 && error != ENXIO) { 760 /* 761 * ENXIO can happen if a pccard is unplugged 762 * Something is really wrong if we were able to put 763 * the driver into promiscuous mode, but can't 764 * take it out. 765 */ 766 if_printf(bp->bif_ifp, 767 "bpf_detach: ifpromisc failed (%d)\n", error); 768 } 769 } 770} 771 772/* 773 * Close the descriptor by detaching it from its interface, 774 * deallocating its buffers, and marking it free. 775 */ 776static void 777bpf_dtor(void *data) 778{ 779 struct bpf_d *d = data; 780 781 BPFD_LOCK(d); 782 if (d->bd_state == BPF_WAITING) 783 callout_stop(&d->bd_callout); 784 d->bd_state = BPF_IDLE; 785 BPFD_UNLOCK(d); 786 funsetown(&d->bd_sigio); 787 bpf_detachd(d); 788#ifdef MAC 789 mac_bpfdesc_destroy(d); 790#endif /* MAC */ 791 seldrain(&d->bd_sel); 792 knlist_destroy(&d->bd_sel.si_note); 793 callout_drain(&d->bd_callout); 794 bpf_freed(d); 795 free(d, M_BPF); 796} 797 798/* 799 * Open ethernet device. Returns ENXIO for illegal minor device number, 800 * EBUSY if file is open by another process. 801 */ 802/* ARGSUSED */ 803static int 804bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) 805{ 806 struct bpf_d *d; 807 int error, size; 808 809 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); 810 error = devfs_set_cdevpriv(d, bpf_dtor); 811 if (error != 0) { 812 free(d, M_BPF); 813 return (error); 814 } 815 816 /* 817 * For historical reasons, perform a one-time initialization call to 818 * the buffer routines, even though we're not yet committed to a 819 * particular buffer method. 820 */ 821 bpf_buffer_init(d); 822 d->bd_hbuf_in_use = 0; 823 d->bd_bufmode = BPF_BUFMODE_BUFFER; 824 d->bd_sig = SIGIO; 825 d->bd_direction = BPF_D_INOUT; 826 BPF_PID_REFRESH(d, td); 827#ifdef MAC 828 mac_bpfdesc_init(d); 829 mac_bpfdesc_create(td->td_ucred, d); 830#endif 831 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 832 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 833 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 834 835 /* Allocate default buffers */ 836 size = d->bd_bufsize; 837 bpf_buffer_ioctl_sblen(d, &size); 838 839 return (0); 840} 841 842/* 843 * bpfread - read next chunk of packets from buffers 844 */ 845static int 846bpfread(struct cdev *dev, struct uio *uio, int ioflag) 847{ 848 struct bpf_d *d; 849 int error; 850 int non_block; 851 int timed_out; 852 853 error = devfs_get_cdevpriv((void **)&d); 854 if (error != 0) 855 return (error); 856 857 /* 858 * Restrict application to use a buffer the same size as 859 * as kernel buffers. 860 */ 861 if (uio->uio_resid != d->bd_bufsize) 862 return (EINVAL); 863 864 non_block = ((ioflag & O_NONBLOCK) != 0); 865 866 BPFD_LOCK(d); 867 BPF_PID_REFRESH_CUR(d); 868 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 869 BPFD_UNLOCK(d); 870 return (EOPNOTSUPP); 871 } 872 if (d->bd_state == BPF_WAITING) 873 callout_stop(&d->bd_callout); 874 timed_out = (d->bd_state == BPF_TIMED_OUT); 875 d->bd_state = BPF_IDLE; 876 while (d->bd_hbuf_in_use) 877 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 878 PRINET|PCATCH, "bd_hbuf", 0); 879 /* 880 * If the hold buffer is empty, then do a timed sleep, which 881 * ends when the timeout expires or when enough packets 882 * have arrived to fill the store buffer. 883 */ 884 while (d->bd_hbuf == NULL) { 885 if (d->bd_slen != 0) { 886 /* 887 * A packet(s) either arrived since the previous 888 * read or arrived while we were asleep. 889 */ 890 if (d->bd_immediate || non_block || timed_out) { 891 /* 892 * Rotate the buffers and return what's here 893 * if we are in immediate mode, non-blocking 894 * flag is set, or this descriptor timed out. 895 */ 896 ROTATE_BUFFERS(d); 897 break; 898 } 899 } 900 901 /* 902 * No data is available, check to see if the bpf device 903 * is still pointed at a real interface. If not, return 904 * ENXIO so that the userland process knows to rebind 905 * it before using it again. 906 */ 907 if (d->bd_bif == NULL) { 908 BPFD_UNLOCK(d); 909 return (ENXIO); 910 } 911 912 if (non_block) { 913 BPFD_UNLOCK(d); 914 return (EWOULDBLOCK); 915 } 916 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 917 "bpf", d->bd_rtout); 918 if (error == EINTR || error == ERESTART) { 919 BPFD_UNLOCK(d); 920 return (error); 921 } 922 if (error == EWOULDBLOCK) { 923 /* 924 * On a timeout, return what's in the buffer, 925 * which may be nothing. If there is something 926 * in the store buffer, we can rotate the buffers. 927 */ 928 if (d->bd_hbuf) 929 /* 930 * We filled up the buffer in between 931 * getting the timeout and arriving 932 * here, so we don't need to rotate. 933 */ 934 break; 935 936 if (d->bd_slen == 0) { 937 BPFD_UNLOCK(d); 938 return (0); 939 } 940 ROTATE_BUFFERS(d); 941 break; 942 } 943 } 944 /* 945 * At this point, we know we have something in the hold slot. 946 */ 947 d->bd_hbuf_in_use = 1; 948 BPFD_UNLOCK(d); 949 950 /* 951 * Move data from hold buffer into user space. 952 * We know the entire buffer is transferred since 953 * we checked above that the read buffer is bpf_bufsize bytes. 954 * 955 * We do not have to worry about simultaneous reads because 956 * we waited for sole access to the hold buffer above. 957 */ 958 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 959 960 BPFD_LOCK(d); 961 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 962 d->bd_fbuf = d->bd_hbuf; 963 d->bd_hbuf = NULL; 964 d->bd_hlen = 0; 965 bpf_buf_reclaimed(d); 966 d->bd_hbuf_in_use = 0; 967 wakeup(&d->bd_hbuf_in_use); 968 BPFD_UNLOCK(d); 969 970 return (error); 971} 972 973/* 974 * If there are processes sleeping on this descriptor, wake them up. 975 */ 976static __inline void 977bpf_wakeup(struct bpf_d *d) 978{ 979 980 BPFD_LOCK_ASSERT(d); 981 if (d->bd_state == BPF_WAITING) { 982 callout_stop(&d->bd_callout); 983 d->bd_state = BPF_IDLE; 984 } 985 wakeup(d); 986 if (d->bd_async && d->bd_sig && d->bd_sigio) 987 pgsigio(&d->bd_sigio, d->bd_sig, 0); 988 989 selwakeuppri(&d->bd_sel, PRINET); 990 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 991} 992 993static void 994bpf_timed_out(void *arg) 995{ 996 struct bpf_d *d = (struct bpf_d *)arg; 997 998 BPFD_LOCK_ASSERT(d); 999 1000 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) 1001 return; 1002 if (d->bd_state == BPF_WAITING) { 1003 d->bd_state = BPF_TIMED_OUT; 1004 if (d->bd_slen != 0) 1005 bpf_wakeup(d); 1006 } 1007} 1008 1009static int 1010bpf_ready(struct bpf_d *d) 1011{ 1012 1013 BPFD_LOCK_ASSERT(d); 1014 1015 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1016 return (1); 1017 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1018 d->bd_slen != 0) 1019 return (1); 1020 return (0); 1021} 1022 1023static int 1024bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1025{ 1026 struct bpf_d *d; 1027 struct ifnet *ifp; 1028 struct mbuf *m, *mc; 1029 struct sockaddr dst; 1030 int error, hlen; 1031 1032 error = devfs_get_cdevpriv((void **)&d); 1033 if (error != 0) 1034 return (error); 1035 1036 BPF_PID_REFRESH_CUR(d); 1037 d->bd_wcount++; 1038 /* XXX: locking required */ 1039 if (d->bd_bif == NULL) { 1040 d->bd_wdcount++; 1041 return (ENXIO); 1042 } 1043 1044 ifp = d->bd_bif->bif_ifp; 1045 1046 if ((ifp->if_flags & IFF_UP) == 0) { 1047 d->bd_wdcount++; 1048 return (ENETDOWN); 1049 } 1050 1051 if (uio->uio_resid == 0) { 1052 d->bd_wdcount++; 1053 return (0); 1054 } 1055 1056 bzero(&dst, sizeof(dst)); 1057 m = NULL; 1058 hlen = 0; 1059 /* XXX: bpf_movein() can sleep */ 1060 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, 1061 &m, &dst, &hlen, d->bd_wfilter); 1062 if (error) { 1063 d->bd_wdcount++; 1064 return (error); 1065 } 1066 d->bd_wfcount++; 1067 if (d->bd_hdrcmplt) 1068 dst.sa_family = pseudo_AF_HDRCMPLT; 1069 1070 if (d->bd_feedback) { 1071 mc = m_dup(m, M_NOWAIT); 1072 if (mc != NULL) 1073 mc->m_pkthdr.rcvif = ifp; 1074 /* Set M_PROMISC for outgoing packets to be discarded. */ 1075 if (d->bd_direction == BPF_D_INOUT) 1076 m->m_flags |= M_PROMISC; 1077 } else 1078 mc = NULL; 1079 1080 m->m_pkthdr.len -= hlen; 1081 m->m_len -= hlen; 1082 m->m_data += hlen; /* XXX */ 1083 1084 CURVNET_SET(ifp->if_vnet); 1085#ifdef MAC 1086 BPFD_LOCK(d); 1087 mac_bpfdesc_create_mbuf(d, m); 1088 if (mc != NULL) 1089 mac_bpfdesc_create_mbuf(d, mc); 1090 BPFD_UNLOCK(d); 1091#endif 1092 1093 error = (*ifp->if_output)(ifp, m, &dst, NULL); 1094 if (error) 1095 d->bd_wdcount++; 1096 1097 if (mc != NULL) { 1098 if (error == 0) 1099 (*ifp->if_input)(ifp, mc); 1100 else 1101 m_freem(mc); 1102 } 1103 CURVNET_RESTORE(); 1104 1105 return (error); 1106} 1107 1108/* 1109 * Reset a descriptor by flushing its packet buffer and clearing the receive 1110 * and drop counts. This is doable for kernel-only buffers, but with 1111 * zero-copy buffers, we can't write to (or rotate) buffers that are 1112 * currently owned by userspace. It would be nice if we could encapsulate 1113 * this logic in the buffer code rather than here. 1114 */ 1115static void 1116reset_d(struct bpf_d *d) 1117{ 1118 1119 BPFD_LOCK_ASSERT(d); 1120 1121 while (d->bd_hbuf_in_use) 1122 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1123 "bd_hbuf", 0); 1124 if ((d->bd_hbuf != NULL) && 1125 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1126 /* Free the hold buffer. */ 1127 d->bd_fbuf = d->bd_hbuf; 1128 d->bd_hbuf = NULL; 1129 d->bd_hlen = 0; 1130 bpf_buf_reclaimed(d); 1131 } 1132 if (bpf_canwritebuf(d)) 1133 d->bd_slen = 0; 1134 d->bd_rcount = 0; 1135 d->bd_dcount = 0; 1136 d->bd_fcount = 0; 1137 d->bd_wcount = 0; 1138 d->bd_wfcount = 0; 1139 d->bd_wdcount = 0; 1140 d->bd_zcopy = 0; 1141} 1142 1143/* 1144 * FIONREAD Check for read packet available. 1145 * SIOCGIFADDR Get interface address - convenient hook to driver. 1146 * BIOCGBLEN Get buffer len [for read()]. 1147 * BIOCSETF Set read filter. 1148 * BIOCSETFNR Set read filter without resetting descriptor. 1149 * BIOCSETWF Set write filter. 1150 * BIOCFLUSH Flush read packet buffer. 1151 * BIOCPROMISC Put interface into promiscuous mode. 1152 * BIOCGDLT Get link layer type. 1153 * BIOCGETIF Get interface name. 1154 * BIOCSETIF Set interface. 1155 * BIOCSRTIMEOUT Set read timeout. 1156 * BIOCGRTIMEOUT Get read timeout. 1157 * BIOCGSTATS Get packet stats. 1158 * BIOCIMMEDIATE Set immediate mode. 1159 * BIOCVERSION Get filter language version. 1160 * BIOCGHDRCMPLT Get "header already complete" flag 1161 * BIOCSHDRCMPLT Set "header already complete" flag 1162 * BIOCGDIRECTION Get packet direction flag 1163 * BIOCSDIRECTION Set packet direction flag 1164 * BIOCGTSTAMP Get time stamp format and resolution. 1165 * BIOCSTSTAMP Set time stamp format and resolution. 1166 * BIOCLOCK Set "locked" flag 1167 * BIOCFEEDBACK Set packet feedback mode. 1168 * BIOCSETZBUF Set current zero-copy buffer locations. 1169 * BIOCGETZMAX Get maximum zero-copy buffer size. 1170 * BIOCROTZBUF Force rotation of zero-copy buffer 1171 * BIOCSETBUFMODE Set buffer mode. 1172 * BIOCGETBUFMODE Get current buffer mode. 1173 */ 1174/* ARGSUSED */ 1175static int 1176bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1177 struct thread *td) 1178{ 1179 struct bpf_d *d; 1180 int error; 1181 1182 error = devfs_get_cdevpriv((void **)&d); 1183 if (error != 0) 1184 return (error); 1185 1186 /* 1187 * Refresh PID associated with this descriptor. 1188 */ 1189 BPFD_LOCK(d); 1190 BPF_PID_REFRESH(d, td); 1191 if (d->bd_state == BPF_WAITING) 1192 callout_stop(&d->bd_callout); 1193 d->bd_state = BPF_IDLE; 1194 BPFD_UNLOCK(d); 1195 1196 if (d->bd_locked == 1) { 1197 switch (cmd) { 1198 case BIOCGBLEN: 1199 case BIOCFLUSH: 1200 case BIOCGDLT: 1201 case BIOCGDLTLIST: 1202#ifdef COMPAT_FREEBSD32 1203 case BIOCGDLTLIST32: 1204#endif 1205 case BIOCGETIF: 1206 case BIOCGRTIMEOUT: 1207#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1208 case BIOCGRTIMEOUT32: 1209#endif 1210 case BIOCGSTATS: 1211 case BIOCVERSION: 1212 case BIOCGRSIG: 1213 case BIOCGHDRCMPLT: 1214 case BIOCSTSTAMP: 1215 case BIOCFEEDBACK: 1216 case FIONREAD: 1217 case BIOCLOCK: 1218 case BIOCSRTIMEOUT: 1219#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1220 case BIOCSRTIMEOUT32: 1221#endif 1222 case BIOCIMMEDIATE: 1223 case TIOCGPGRP: 1224 case BIOCROTZBUF: 1225 break; 1226 default: 1227 return (EPERM); 1228 } 1229 } 1230#ifdef COMPAT_FREEBSD32 1231 /* 1232 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1233 * that it will get 32-bit packet headers. 1234 */ 1235 switch (cmd) { 1236 case BIOCSETF32: 1237 case BIOCSETFNR32: 1238 case BIOCSETWF32: 1239 case BIOCGDLTLIST32: 1240 case BIOCGRTIMEOUT32: 1241 case BIOCSRTIMEOUT32: 1242 BPFD_LOCK(d); 1243 d->bd_compat32 = 1; 1244 BPFD_UNLOCK(d); 1245 } 1246#endif 1247 1248 CURVNET_SET(TD_TO_VNET(td)); 1249 switch (cmd) { 1250 1251 default: 1252 error = EINVAL; 1253 break; 1254 1255 /* 1256 * Check for read packet available. 1257 */ 1258 case FIONREAD: 1259 { 1260 int n; 1261 1262 BPFD_LOCK(d); 1263 n = d->bd_slen; 1264 while (d->bd_hbuf_in_use) 1265 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1266 PRINET, "bd_hbuf", 0); 1267 if (d->bd_hbuf) 1268 n += d->bd_hlen; 1269 BPFD_UNLOCK(d); 1270 1271 *(int *)addr = n; 1272 break; 1273 } 1274 1275 case SIOCGIFADDR: 1276 { 1277 struct ifnet *ifp; 1278 1279 if (d->bd_bif == NULL) 1280 error = EINVAL; 1281 else { 1282 ifp = d->bd_bif->bif_ifp; 1283 error = (*ifp->if_ioctl)(ifp, cmd, addr); 1284 } 1285 break; 1286 } 1287 1288 /* 1289 * Get buffer len [for read()]. 1290 */ 1291 case BIOCGBLEN: 1292 BPFD_LOCK(d); 1293 *(u_int *)addr = d->bd_bufsize; 1294 BPFD_UNLOCK(d); 1295 break; 1296 1297 /* 1298 * Set buffer length. 1299 */ 1300 case BIOCSBLEN: 1301 error = bpf_ioctl_sblen(d, (u_int *)addr); 1302 break; 1303 1304 /* 1305 * Set link layer read filter. 1306 */ 1307 case BIOCSETF: 1308 case BIOCSETFNR: 1309 case BIOCSETWF: 1310#ifdef COMPAT_FREEBSD32 1311 case BIOCSETF32: 1312 case BIOCSETFNR32: 1313 case BIOCSETWF32: 1314#endif 1315 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1316 break; 1317 1318 /* 1319 * Flush read packet buffer. 1320 */ 1321 case BIOCFLUSH: 1322 BPFD_LOCK(d); 1323 reset_d(d); 1324 BPFD_UNLOCK(d); 1325 break; 1326 1327 /* 1328 * Put interface into promiscuous mode. 1329 */ 1330 case BIOCPROMISC: 1331 if (d->bd_bif == NULL) { 1332 /* 1333 * No interface attached yet. 1334 */ 1335 error = EINVAL; 1336 break; 1337 } 1338 if (d->bd_promisc == 0) { 1339 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1340 if (error == 0) 1341 d->bd_promisc = 1; 1342 } 1343 break; 1344 1345 /* 1346 * Get current data link type. 1347 */ 1348 case BIOCGDLT: 1349 BPF_LOCK(); 1350 if (d->bd_bif == NULL) 1351 error = EINVAL; 1352 else 1353 *(u_int *)addr = d->bd_bif->bif_dlt; 1354 BPF_UNLOCK(); 1355 break; 1356 1357 /* 1358 * Get a list of supported data link types. 1359 */ 1360#ifdef COMPAT_FREEBSD32 1361 case BIOCGDLTLIST32: 1362 { 1363 struct bpf_dltlist32 *list32; 1364 struct bpf_dltlist dltlist; 1365 1366 list32 = (struct bpf_dltlist32 *)addr; 1367 dltlist.bfl_len = list32->bfl_len; 1368 dltlist.bfl_list = PTRIN(list32->bfl_list); 1369 BPF_LOCK(); 1370 if (d->bd_bif == NULL) 1371 error = EINVAL; 1372 else { 1373 error = bpf_getdltlist(d, &dltlist); 1374 if (error == 0) 1375 list32->bfl_len = dltlist.bfl_len; 1376 } 1377 BPF_UNLOCK(); 1378 break; 1379 } 1380#endif 1381 1382 case BIOCGDLTLIST: 1383 BPF_LOCK(); 1384 if (d->bd_bif == NULL) 1385 error = EINVAL; 1386 else 1387 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1388 BPF_UNLOCK(); 1389 break; 1390 1391 /* 1392 * Set data link type. 1393 */ 1394 case BIOCSDLT: 1395 BPF_LOCK(); 1396 if (d->bd_bif == NULL) 1397 error = EINVAL; 1398 else 1399 error = bpf_setdlt(d, *(u_int *)addr); 1400 BPF_UNLOCK(); 1401 break; 1402 1403 /* 1404 * Get interface name. 1405 */ 1406 case BIOCGETIF: 1407 BPF_LOCK(); 1408 if (d->bd_bif == NULL) 1409 error = EINVAL; 1410 else { 1411 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1412 struct ifreq *const ifr = (struct ifreq *)addr; 1413 1414 strlcpy(ifr->ifr_name, ifp->if_xname, 1415 sizeof(ifr->ifr_name)); 1416 } 1417 BPF_UNLOCK(); 1418 break; 1419 1420 /* 1421 * Set interface. 1422 */ 1423 case BIOCSETIF: 1424 BPF_LOCK(); 1425 error = bpf_setif(d, (struct ifreq *)addr); 1426 BPF_UNLOCK(); 1427 break; 1428 1429 /* 1430 * Set read timeout. 1431 */ 1432 case BIOCSRTIMEOUT: 1433#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1434 case BIOCSRTIMEOUT32: 1435#endif 1436 { 1437 struct timeval *tv = (struct timeval *)addr; 1438#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1439 struct timeval32 *tv32; 1440 struct timeval tv64; 1441 1442 if (cmd == BIOCSRTIMEOUT32) { 1443 tv32 = (struct timeval32 *)addr; 1444 tv = &tv64; 1445 tv->tv_sec = tv32->tv_sec; 1446 tv->tv_usec = tv32->tv_usec; 1447 } else 1448#endif 1449 tv = (struct timeval *)addr; 1450 1451 /* 1452 * Subtract 1 tick from tvtohz() since this isn't 1453 * a one-shot timer. 1454 */ 1455 if ((error = itimerfix(tv)) == 0) 1456 d->bd_rtout = tvtohz(tv) - 1; 1457 break; 1458 } 1459 1460 /* 1461 * Get read timeout. 1462 */ 1463 case BIOCGRTIMEOUT: 1464#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1465 case BIOCGRTIMEOUT32: 1466#endif 1467 { 1468 struct timeval *tv; 1469#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1470 struct timeval32 *tv32; 1471 struct timeval tv64; 1472 1473 if (cmd == BIOCGRTIMEOUT32) 1474 tv = &tv64; 1475 else 1476#endif 1477 tv = (struct timeval *)addr; 1478 1479 tv->tv_sec = d->bd_rtout / hz; 1480 tv->tv_usec = (d->bd_rtout % hz) * tick; 1481#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1482 if (cmd == BIOCGRTIMEOUT32) { 1483 tv32 = (struct timeval32 *)addr; 1484 tv32->tv_sec = tv->tv_sec; 1485 tv32->tv_usec = tv->tv_usec; 1486 } 1487#endif 1488 1489 break; 1490 } 1491 1492 /* 1493 * Get packet stats. 1494 */ 1495 case BIOCGSTATS: 1496 { 1497 struct bpf_stat *bs = (struct bpf_stat *)addr; 1498 1499 /* XXXCSJP overflow */ 1500 bs->bs_recv = d->bd_rcount; 1501 bs->bs_drop = d->bd_dcount; 1502 break; 1503 } 1504 1505 /* 1506 * Set immediate mode. 1507 */ 1508 case BIOCIMMEDIATE: 1509 BPFD_LOCK(d); 1510 d->bd_immediate = *(u_int *)addr; 1511 BPFD_UNLOCK(d); 1512 break; 1513 1514 case BIOCVERSION: 1515 { 1516 struct bpf_version *bv = (struct bpf_version *)addr; 1517 1518 bv->bv_major = BPF_MAJOR_VERSION; 1519 bv->bv_minor = BPF_MINOR_VERSION; 1520 break; 1521 } 1522 1523 /* 1524 * Get "header already complete" flag 1525 */ 1526 case BIOCGHDRCMPLT: 1527 BPFD_LOCK(d); 1528 *(u_int *)addr = d->bd_hdrcmplt; 1529 BPFD_UNLOCK(d); 1530 break; 1531 1532 /* 1533 * Set "header already complete" flag 1534 */ 1535 case BIOCSHDRCMPLT: 1536 BPFD_LOCK(d); 1537 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1538 BPFD_UNLOCK(d); 1539 break; 1540 1541 /* 1542 * Get packet direction flag 1543 */ 1544 case BIOCGDIRECTION: 1545 BPFD_LOCK(d); 1546 *(u_int *)addr = d->bd_direction; 1547 BPFD_UNLOCK(d); 1548 break; 1549 1550 /* 1551 * Set packet direction flag 1552 */ 1553 case BIOCSDIRECTION: 1554 { 1555 u_int direction; 1556 1557 direction = *(u_int *)addr; 1558 switch (direction) { 1559 case BPF_D_IN: 1560 case BPF_D_INOUT: 1561 case BPF_D_OUT: 1562 BPFD_LOCK(d); 1563 d->bd_direction = direction; 1564 BPFD_UNLOCK(d); 1565 break; 1566 default: 1567 error = EINVAL; 1568 } 1569 } 1570 break; 1571 1572 /* 1573 * Get packet timestamp format and resolution. 1574 */ 1575 case BIOCGTSTAMP: 1576 BPFD_LOCK(d); 1577 *(u_int *)addr = d->bd_tstamp; 1578 BPFD_UNLOCK(d); 1579 break; 1580 1581 /* 1582 * Set packet timestamp format and resolution. 1583 */ 1584 case BIOCSTSTAMP: 1585 { 1586 u_int func; 1587 1588 func = *(u_int *)addr; 1589 if (BPF_T_VALID(func)) 1590 d->bd_tstamp = func; 1591 else 1592 error = EINVAL; 1593 } 1594 break; 1595 1596 case BIOCFEEDBACK: 1597 BPFD_LOCK(d); 1598 d->bd_feedback = *(u_int *)addr; 1599 BPFD_UNLOCK(d); 1600 break; 1601 1602 case BIOCLOCK: 1603 BPFD_LOCK(d); 1604 d->bd_locked = 1; 1605 BPFD_UNLOCK(d); 1606 break; 1607 1608 case FIONBIO: /* Non-blocking I/O */ 1609 break; 1610 1611 case FIOASYNC: /* Send signal on receive packets */ 1612 BPFD_LOCK(d); 1613 d->bd_async = *(int *)addr; 1614 BPFD_UNLOCK(d); 1615 break; 1616 1617 case FIOSETOWN: 1618 /* 1619 * XXX: Add some sort of locking here? 1620 * fsetown() can sleep. 1621 */ 1622 error = fsetown(*(int *)addr, &d->bd_sigio); 1623 break; 1624 1625 case FIOGETOWN: 1626 BPFD_LOCK(d); 1627 *(int *)addr = fgetown(&d->bd_sigio); 1628 BPFD_UNLOCK(d); 1629 break; 1630 1631 /* This is deprecated, FIOSETOWN should be used instead. */ 1632 case TIOCSPGRP: 1633 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1634 break; 1635 1636 /* This is deprecated, FIOGETOWN should be used instead. */ 1637 case TIOCGPGRP: 1638 *(int *)addr = -fgetown(&d->bd_sigio); 1639 break; 1640 1641 case BIOCSRSIG: /* Set receive signal */ 1642 { 1643 u_int sig; 1644 1645 sig = *(u_int *)addr; 1646 1647 if (sig >= NSIG) 1648 error = EINVAL; 1649 else { 1650 BPFD_LOCK(d); 1651 d->bd_sig = sig; 1652 BPFD_UNLOCK(d); 1653 } 1654 break; 1655 } 1656 case BIOCGRSIG: 1657 BPFD_LOCK(d); 1658 *(u_int *)addr = d->bd_sig; 1659 BPFD_UNLOCK(d); 1660 break; 1661 1662 case BIOCGETBUFMODE: 1663 BPFD_LOCK(d); 1664 *(u_int *)addr = d->bd_bufmode; 1665 BPFD_UNLOCK(d); 1666 break; 1667 1668 case BIOCSETBUFMODE: 1669 /* 1670 * Allow the buffering mode to be changed as long as we 1671 * haven't yet committed to a particular mode. Our 1672 * definition of commitment, for now, is whether or not a 1673 * buffer has been allocated or an interface attached, since 1674 * that's the point where things get tricky. 1675 */ 1676 switch (*(u_int *)addr) { 1677 case BPF_BUFMODE_BUFFER: 1678 break; 1679 1680 case BPF_BUFMODE_ZBUF: 1681 if (bpf_zerocopy_enable) 1682 break; 1683 /* FALLSTHROUGH */ 1684 1685 default: 1686 CURVNET_RESTORE(); 1687 return (EINVAL); 1688 } 1689 1690 BPFD_LOCK(d); 1691 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1692 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1693 BPFD_UNLOCK(d); 1694 CURVNET_RESTORE(); 1695 return (EBUSY); 1696 } 1697 d->bd_bufmode = *(u_int *)addr; 1698 BPFD_UNLOCK(d); 1699 break; 1700 1701 case BIOCGETZMAX: 1702 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1703 break; 1704 1705 case BIOCSETZBUF: 1706 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1707 break; 1708 1709 case BIOCROTZBUF: 1710 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1711 break; 1712 } 1713 CURVNET_RESTORE(); 1714 return (error); 1715} 1716 1717/* 1718 * Set d's packet filter program to fp. If this file already has a filter, 1719 * free it and replace it. Returns EINVAL for bogus requests. 1720 * 1721 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls 1722 * since reading d->bd_bif can't be protected by d or interface lock due to 1723 * lock order. 1724 * 1725 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses 1726 * interface read lock to read all filers. 1727 * 1728 */ 1729static int 1730bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1731{ 1732#ifdef COMPAT_FREEBSD32 1733 struct bpf_program fp_swab; 1734 struct bpf_program32 *fp32; 1735#endif 1736 struct bpf_insn *fcode, *old; 1737#ifdef BPF_JITTER 1738 bpf_jit_filter *jfunc, *ofunc; 1739#endif 1740 size_t size; 1741 u_int flen; 1742 int need_upgrade; 1743 1744#ifdef COMPAT_FREEBSD32 1745 switch (cmd) { 1746 case BIOCSETF32: 1747 case BIOCSETWF32: 1748 case BIOCSETFNR32: 1749 fp32 = (struct bpf_program32 *)fp; 1750 fp_swab.bf_len = fp32->bf_len; 1751 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1752 fp = &fp_swab; 1753 switch (cmd) { 1754 case BIOCSETF32: 1755 cmd = BIOCSETF; 1756 break; 1757 case BIOCSETWF32: 1758 cmd = BIOCSETWF; 1759 break; 1760 } 1761 break; 1762 } 1763#endif 1764 1765 fcode = NULL; 1766#ifdef BPF_JITTER 1767 jfunc = ofunc = NULL; 1768#endif 1769 need_upgrade = 0; 1770 1771 /* 1772 * Check new filter validness before acquiring any locks. 1773 * Allocate memory for new filter, if needed. 1774 */ 1775 flen = fp->bf_len; 1776 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1777 return (EINVAL); 1778 size = flen * sizeof(*fp->bf_insns); 1779 if (size > 0) { 1780 /* We're setting up new filter. Copy and check actual data. */ 1781 fcode = malloc(size, M_BPF, M_WAITOK); 1782 if (copyin(fp->bf_insns, fcode, size) != 0 || 1783 !bpf_validate(fcode, flen)) { 1784 free(fcode, M_BPF); 1785 return (EINVAL); 1786 } 1787#ifdef BPF_JITTER 1788 /* Filter is copied inside fcode and is perfectly valid. */ 1789 jfunc = bpf_jitter(fcode, flen); 1790#endif 1791 } 1792 1793 BPF_LOCK(); 1794 1795 /* 1796 * Set up new filter. 1797 * Protect filter change by interface lock. 1798 * Additionally, we are protected by global lock here. 1799 */ 1800 if (d->bd_bif != NULL) 1801 BPFIF_WLOCK(d->bd_bif); 1802 BPFD_LOCK(d); 1803 if (cmd == BIOCSETWF) { 1804 old = d->bd_wfilter; 1805 d->bd_wfilter = fcode; 1806 } else { 1807 old = d->bd_rfilter; 1808 d->bd_rfilter = fcode; 1809#ifdef BPF_JITTER 1810 ofunc = d->bd_bfilter; 1811 d->bd_bfilter = jfunc; 1812#endif 1813 if (cmd == BIOCSETF) 1814 reset_d(d); 1815 1816 if (fcode != NULL) { 1817 /* 1818 * Do not require upgrade by first BIOCSETF 1819 * (used to set snaplen) by pcap_open_live(). 1820 */ 1821 if (d->bd_writer != 0 && --d->bd_writer == 0) 1822 need_upgrade = 1; 1823 CTR4(KTR_NET, "%s: filter function set by pid %d, " 1824 "bd_writer counter %d, need_upgrade %d", 1825 __func__, d->bd_pid, d->bd_writer, need_upgrade); 1826 } 1827 } 1828 BPFD_UNLOCK(d); 1829 if (d->bd_bif != NULL) 1830 BPFIF_WUNLOCK(d->bd_bif); 1831 if (old != NULL) 1832 free(old, M_BPF); 1833#ifdef BPF_JITTER 1834 if (ofunc != NULL) 1835 bpf_destroy_jit_filter(ofunc); 1836#endif 1837 1838 /* Move d to active readers list. */ 1839 if (need_upgrade) 1840 bpf_upgraded(d); 1841 1842 BPF_UNLOCK(); 1843 return (0); 1844} 1845 1846/* 1847 * Detach a file from its current interface (if attached at all) and attach 1848 * to the interface indicated by the name stored in ifr. 1849 * Return an errno or 0. 1850 */ 1851static int 1852bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1853{ 1854 struct bpf_if *bp; 1855 struct ifnet *theywant; 1856 1857 BPF_LOCK_ASSERT(); 1858 1859 theywant = ifunit(ifr->ifr_name); 1860 if (theywant == NULL || theywant->if_bpf == NULL) 1861 return (ENXIO); 1862 1863 bp = theywant->if_bpf; 1864 1865 /* Check if interface is not being detached from BPF */ 1866 BPFIF_RLOCK(bp); 1867 if (bp->flags & BPFIF_FLAG_DYING) { 1868 BPFIF_RUNLOCK(bp); 1869 return (ENXIO); 1870 } 1871 BPFIF_RUNLOCK(bp); 1872 1873 /* 1874 * Behavior here depends on the buffering model. If we're using 1875 * kernel memory buffers, then we can allocate them here. If we're 1876 * using zero-copy, then the user process must have registered 1877 * buffers by the time we get here. If not, return an error. 1878 */ 1879 switch (d->bd_bufmode) { 1880 case BPF_BUFMODE_BUFFER: 1881 case BPF_BUFMODE_ZBUF: 1882 if (d->bd_sbuf == NULL) 1883 return (EINVAL); 1884 break; 1885 1886 default: 1887 panic("bpf_setif: bufmode %d", d->bd_bufmode); 1888 } 1889 if (bp != d->bd_bif) 1890 bpf_attachd(d, bp); 1891 BPFD_LOCK(d); 1892 reset_d(d); 1893 BPFD_UNLOCK(d); 1894 return (0); 1895} 1896 1897/* 1898 * Support for select() and poll() system calls 1899 * 1900 * Return true iff the specific operation will not block indefinitely. 1901 * Otherwise, return false but make a note that a selwakeup() must be done. 1902 */ 1903static int 1904bpfpoll(struct cdev *dev, int events, struct thread *td) 1905{ 1906 struct bpf_d *d; 1907 int revents; 1908 1909 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 1910 return (events & 1911 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 1912 1913 /* 1914 * Refresh PID associated with this descriptor. 1915 */ 1916 revents = events & (POLLOUT | POLLWRNORM); 1917 BPFD_LOCK(d); 1918 BPF_PID_REFRESH(d, td); 1919 if (events & (POLLIN | POLLRDNORM)) { 1920 if (bpf_ready(d)) 1921 revents |= events & (POLLIN | POLLRDNORM); 1922 else { 1923 selrecord(td, &d->bd_sel); 1924 /* Start the read timeout if necessary. */ 1925 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1926 callout_reset(&d->bd_callout, d->bd_rtout, 1927 bpf_timed_out, d); 1928 d->bd_state = BPF_WAITING; 1929 } 1930 } 1931 } 1932 BPFD_UNLOCK(d); 1933 return (revents); 1934} 1935 1936/* 1937 * Support for kevent() system call. Register EVFILT_READ filters and 1938 * reject all others. 1939 */ 1940int 1941bpfkqfilter(struct cdev *dev, struct knote *kn) 1942{ 1943 struct bpf_d *d; 1944 1945 if (devfs_get_cdevpriv((void **)&d) != 0 || 1946 kn->kn_filter != EVFILT_READ) 1947 return (1); 1948 1949 /* 1950 * Refresh PID associated with this descriptor. 1951 */ 1952 BPFD_LOCK(d); 1953 BPF_PID_REFRESH_CUR(d); 1954 kn->kn_fop = &bpfread_filtops; 1955 kn->kn_hook = d; 1956 knlist_add(&d->bd_sel.si_note, kn, 1); 1957 BPFD_UNLOCK(d); 1958 1959 return (0); 1960} 1961 1962static void 1963filt_bpfdetach(struct knote *kn) 1964{ 1965 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 1966 1967 knlist_remove(&d->bd_sel.si_note, kn, 0); 1968} 1969 1970static int 1971filt_bpfread(struct knote *kn, long hint) 1972{ 1973 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 1974 int ready; 1975 1976 BPFD_LOCK_ASSERT(d); 1977 ready = bpf_ready(d); 1978 if (ready) { 1979 kn->kn_data = d->bd_slen; 1980 while (d->bd_hbuf_in_use) 1981 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1982 PRINET, "bd_hbuf", 0); 1983 if (d->bd_hbuf) 1984 kn->kn_data += d->bd_hlen; 1985 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1986 callout_reset(&d->bd_callout, d->bd_rtout, 1987 bpf_timed_out, d); 1988 d->bd_state = BPF_WAITING; 1989 } 1990 1991 return (ready); 1992} 1993 1994#define BPF_TSTAMP_NONE 0 1995#define BPF_TSTAMP_FAST 1 1996#define BPF_TSTAMP_NORMAL 2 1997#define BPF_TSTAMP_EXTERN 3 1998 1999static int 2000bpf_ts_quality(int tstype) 2001{ 2002 2003 if (tstype == BPF_T_NONE) 2004 return (BPF_TSTAMP_NONE); 2005 if ((tstype & BPF_T_FAST) != 0) 2006 return (BPF_TSTAMP_FAST); 2007 2008 return (BPF_TSTAMP_NORMAL); 2009} 2010 2011static int 2012bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2013{ 2014 struct m_tag *tag; 2015 int quality; 2016 2017 quality = bpf_ts_quality(tstype); 2018 if (quality == BPF_TSTAMP_NONE) 2019 return (quality); 2020 2021 if (m != NULL) { 2022 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2023 if (tag != NULL) { 2024 *bt = *(struct bintime *)(tag + 1); 2025 return (BPF_TSTAMP_EXTERN); 2026 } 2027 } 2028 if (quality == BPF_TSTAMP_NORMAL) 2029 binuptime(bt); 2030 else 2031 getbinuptime(bt); 2032 2033 return (quality); 2034} 2035 2036/* 2037 * Incoming linkage from device drivers. Process the packet pkt, of length 2038 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2039 * by each process' filter, and if accepted, stashed into the corresponding 2040 * buffer. 2041 */ 2042void 2043bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2044{ 2045 struct bintime bt; 2046 struct bpf_d *d; 2047#ifdef BPF_JITTER 2048 bpf_jit_filter *bf; 2049#endif 2050 u_int slen; 2051 int gottime; 2052 2053 gottime = BPF_TSTAMP_NONE; 2054 2055 BPFIF_RLOCK(bp); 2056 2057 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2058 /* 2059 * We are not using any locks for d here because: 2060 * 1) any filter change is protected by interface 2061 * write lock 2062 * 2) destroying/detaching d is protected by interface 2063 * write lock, too 2064 */ 2065 2066 /* XXX: Do not protect counter for the sake of performance. */ 2067 ++d->bd_rcount; 2068 /* 2069 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no 2070 * way for the caller to indiciate to us whether this packet 2071 * is inbound or outbound. In the bpf_mtap() routines, we use 2072 * the interface pointers on the mbuf to figure it out. 2073 */ 2074#ifdef BPF_JITTER 2075 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2076 if (bf != NULL) 2077 slen = (*(bf->func))(pkt, pktlen, pktlen); 2078 else 2079#endif 2080 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2081 if (slen != 0) { 2082 /* 2083 * Filter matches. Let's to acquire write lock. 2084 */ 2085 BPFD_LOCK(d); 2086 2087 d->bd_fcount++; 2088 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2089 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); 2090#ifdef MAC 2091 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2092#endif 2093 catchpacket(d, pkt, pktlen, slen, 2094 bpf_append_bytes, &bt); 2095 BPFD_UNLOCK(d); 2096 } 2097 } 2098 BPFIF_RUNLOCK(bp); 2099} 2100 2101#define BPF_CHECK_DIRECTION(d, r, i) \ 2102 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2103 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2104 2105/* 2106 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2107 * Locking model is explained in bpf_tap(). 2108 */ 2109void 2110bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2111{ 2112 struct bintime bt; 2113 struct bpf_d *d; 2114#ifdef BPF_JITTER 2115 bpf_jit_filter *bf; 2116#endif 2117 u_int pktlen, slen; 2118 int gottime; 2119 2120 /* Skip outgoing duplicate packets. */ 2121 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2122 m->m_flags &= ~M_PROMISC; 2123 return; 2124 } 2125 2126 pktlen = m_length(m, NULL); 2127 gottime = BPF_TSTAMP_NONE; 2128 2129 BPFIF_RLOCK(bp); 2130 2131 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2132 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2133 continue; 2134 ++d->bd_rcount; 2135#ifdef BPF_JITTER 2136 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2137 /* XXX We cannot handle multiple mbufs. */ 2138 if (bf != NULL && m->m_next == NULL) 2139 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); 2140 else 2141#endif 2142 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2143 if (slen != 0) { 2144 BPFD_LOCK(d); 2145 2146 d->bd_fcount++; 2147 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2148 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2149#ifdef MAC 2150 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2151#endif 2152 catchpacket(d, (u_char *)m, pktlen, slen, 2153 bpf_append_mbuf, &bt); 2154 BPFD_UNLOCK(d); 2155 } 2156 } 2157 BPFIF_RUNLOCK(bp); 2158} 2159 2160/* 2161 * Incoming linkage from device drivers, when packet is in 2162 * an mbuf chain and to be prepended by a contiguous header. 2163 */ 2164void 2165bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2166{ 2167 struct bintime bt; 2168 struct mbuf mb; 2169 struct bpf_d *d; 2170 u_int pktlen, slen; 2171 int gottime; 2172 2173 /* Skip outgoing duplicate packets. */ 2174 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2175 m->m_flags &= ~M_PROMISC; 2176 return; 2177 } 2178 2179 pktlen = m_length(m, NULL); 2180 /* 2181 * Craft on-stack mbuf suitable for passing to bpf_filter. 2182 * Note that we cut corners here; we only setup what's 2183 * absolutely needed--this mbuf should never go anywhere else. 2184 */ 2185 mb.m_next = m; 2186 mb.m_data = data; 2187 mb.m_len = dlen; 2188 pktlen += dlen; 2189 2190 gottime = BPF_TSTAMP_NONE; 2191 2192 BPFIF_RLOCK(bp); 2193 2194 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2195 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2196 continue; 2197 ++d->bd_rcount; 2198 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2199 if (slen != 0) { 2200 BPFD_LOCK(d); 2201 2202 d->bd_fcount++; 2203 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2204 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2205#ifdef MAC 2206 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2207#endif 2208 catchpacket(d, (u_char *)&mb, pktlen, slen, 2209 bpf_append_mbuf, &bt); 2210 BPFD_UNLOCK(d); 2211 } 2212 } 2213 BPFIF_RUNLOCK(bp); 2214} 2215 2216#undef BPF_CHECK_DIRECTION 2217 2218#undef BPF_TSTAMP_NONE 2219#undef BPF_TSTAMP_FAST 2220#undef BPF_TSTAMP_NORMAL 2221#undef BPF_TSTAMP_EXTERN 2222 2223static int 2224bpf_hdrlen(struct bpf_d *d) 2225{ 2226 int hdrlen; 2227 2228 hdrlen = d->bd_bif->bif_hdrlen; 2229#ifndef BURN_BRIDGES 2230 if (d->bd_tstamp == BPF_T_NONE || 2231 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2232#ifdef COMPAT_FREEBSD32 2233 if (d->bd_compat32) 2234 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2235 else 2236#endif 2237 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2238 else 2239#endif 2240 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2241#ifdef COMPAT_FREEBSD32 2242 if (d->bd_compat32) 2243 hdrlen = BPF_WORDALIGN32(hdrlen); 2244 else 2245#endif 2246 hdrlen = BPF_WORDALIGN(hdrlen); 2247 2248 return (hdrlen - d->bd_bif->bif_hdrlen); 2249} 2250 2251static void 2252bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2253{ 2254 struct bintime bt2; 2255 struct timeval tsm; 2256 struct timespec tsn; 2257 2258 if ((tstype & BPF_T_MONOTONIC) == 0) { 2259 bt2 = *bt; 2260 bintime_add(&bt2, &boottimebin); 2261 bt = &bt2; 2262 } 2263 switch (BPF_T_FORMAT(tstype)) { 2264 case BPF_T_MICROTIME: 2265 bintime2timeval(bt, &tsm); 2266 ts->bt_sec = tsm.tv_sec; 2267 ts->bt_frac = tsm.tv_usec; 2268 break; 2269 case BPF_T_NANOTIME: 2270 bintime2timespec(bt, &tsn); 2271 ts->bt_sec = tsn.tv_sec; 2272 ts->bt_frac = tsn.tv_nsec; 2273 break; 2274 case BPF_T_BINTIME: 2275 ts->bt_sec = bt->sec; 2276 ts->bt_frac = bt->frac; 2277 break; 2278 } 2279} 2280 2281/* 2282 * Move the packet data from interface memory (pkt) into the 2283 * store buffer. "cpfn" is the routine called to do the actual data 2284 * transfer. bcopy is passed in to copy contiguous chunks, while 2285 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2286 * pkt is really an mbuf. 2287 */ 2288static void 2289catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2290 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2291 struct bintime *bt) 2292{ 2293 struct bpf_xhdr hdr; 2294#ifndef BURN_BRIDGES 2295 struct bpf_hdr hdr_old; 2296#ifdef COMPAT_FREEBSD32 2297 struct bpf_hdr32 hdr32_old; 2298#endif 2299#endif 2300 int caplen, curlen, hdrlen, totlen; 2301 int do_wakeup = 0; 2302 int do_timestamp; 2303 int tstype; 2304 2305 BPFD_LOCK_ASSERT(d); 2306 2307 /* 2308 * Detect whether user space has released a buffer back to us, and if 2309 * so, move it from being a hold buffer to a free buffer. This may 2310 * not be the best place to do it (for example, we might only want to 2311 * run this check if we need the space), but for now it's a reliable 2312 * spot to do it. 2313 */ 2314 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2315 while (d->bd_hbuf_in_use) 2316 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 2317 PRINET, "bd_hbuf", 0); 2318 d->bd_fbuf = d->bd_hbuf; 2319 d->bd_hbuf = NULL; 2320 d->bd_hlen = 0; 2321 bpf_buf_reclaimed(d); 2322 } 2323 2324 /* 2325 * Figure out how many bytes to move. If the packet is 2326 * greater or equal to the snapshot length, transfer that 2327 * much. Otherwise, transfer the whole packet (unless 2328 * we hit the buffer size limit). 2329 */ 2330 hdrlen = bpf_hdrlen(d); 2331 totlen = hdrlen + min(snaplen, pktlen); 2332 if (totlen > d->bd_bufsize) 2333 totlen = d->bd_bufsize; 2334 2335 /* 2336 * Round up the end of the previous packet to the next longword. 2337 * 2338 * Drop the packet if there's no room and no hope of room 2339 * If the packet would overflow the storage buffer or the storage 2340 * buffer is considered immutable by the buffer model, try to rotate 2341 * the buffer and wakeup pending processes. 2342 */ 2343#ifdef COMPAT_FREEBSD32 2344 if (d->bd_compat32) 2345 curlen = BPF_WORDALIGN32(d->bd_slen); 2346 else 2347#endif 2348 curlen = BPF_WORDALIGN(d->bd_slen); 2349 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2350 if (d->bd_fbuf == NULL) { 2351 /* 2352 * There's no room in the store buffer, and no 2353 * prospect of room, so drop the packet. Notify the 2354 * buffer model. 2355 */ 2356 bpf_buffull(d); 2357 ++d->bd_dcount; 2358 return; 2359 } 2360 while (d->bd_hbuf_in_use) 2361 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 2362 PRINET, "bd_hbuf", 0); 2363 ROTATE_BUFFERS(d); 2364 do_wakeup = 1; 2365 curlen = 0; 2366 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2367 /* 2368 * Immediate mode is set, or the read timeout has already 2369 * expired during a select call. A packet arrived, so the 2370 * reader should be woken up. 2371 */ 2372 do_wakeup = 1; 2373 caplen = totlen - hdrlen; 2374 tstype = d->bd_tstamp; 2375 do_timestamp = tstype != BPF_T_NONE; 2376#ifndef BURN_BRIDGES 2377 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2378 struct bpf_ts ts; 2379 if (do_timestamp) 2380 bpf_bintime2ts(bt, &ts, tstype); 2381#ifdef COMPAT_FREEBSD32 2382 if (d->bd_compat32) { 2383 bzero(&hdr32_old, sizeof(hdr32_old)); 2384 if (do_timestamp) { 2385 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2386 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2387 } 2388 hdr32_old.bh_datalen = pktlen; 2389 hdr32_old.bh_hdrlen = hdrlen; 2390 hdr32_old.bh_caplen = caplen; 2391 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2392 sizeof(hdr32_old)); 2393 goto copy; 2394 } 2395#endif 2396 bzero(&hdr_old, sizeof(hdr_old)); 2397 if (do_timestamp) { 2398 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2399 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2400 } 2401 hdr_old.bh_datalen = pktlen; 2402 hdr_old.bh_hdrlen = hdrlen; 2403 hdr_old.bh_caplen = caplen; 2404 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2405 sizeof(hdr_old)); 2406 goto copy; 2407 } 2408#endif 2409 2410 /* 2411 * Append the bpf header. Note we append the actual header size, but 2412 * move forward the length of the header plus padding. 2413 */ 2414 bzero(&hdr, sizeof(hdr)); 2415 if (do_timestamp) 2416 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2417 hdr.bh_datalen = pktlen; 2418 hdr.bh_hdrlen = hdrlen; 2419 hdr.bh_caplen = caplen; 2420 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2421 2422 /* 2423 * Copy the packet data into the store buffer and update its length. 2424 */ 2425#ifndef BURN_BRIDGES 2426copy: 2427#endif 2428 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2429 d->bd_slen = curlen + totlen; 2430 2431 if (do_wakeup) 2432 bpf_wakeup(d); 2433} 2434 2435/* 2436 * Free buffers currently in use by a descriptor. 2437 * Called on close. 2438 */ 2439static void 2440bpf_freed(struct bpf_d *d) 2441{ 2442 2443 /* 2444 * We don't need to lock out interrupts since this descriptor has 2445 * been detached from its interface and it yet hasn't been marked 2446 * free. 2447 */ 2448 bpf_free(d); 2449 if (d->bd_rfilter != NULL) { 2450 free((caddr_t)d->bd_rfilter, M_BPF); 2451#ifdef BPF_JITTER 2452 if (d->bd_bfilter != NULL) 2453 bpf_destroy_jit_filter(d->bd_bfilter); 2454#endif 2455 } 2456 if (d->bd_wfilter != NULL) 2457 free((caddr_t)d->bd_wfilter, M_BPF); 2458 mtx_destroy(&d->bd_lock); 2459} 2460 2461/* 2462 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2463 * fixed size of the link header (variable length headers not yet supported). 2464 */ 2465void 2466bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2467{ 2468 2469 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2470} 2471 2472/* 2473 * Attach an interface to bpf. ifp is a pointer to the structure 2474 * defining the interface to be attached, dlt is the link layer type, 2475 * and hdrlen is the fixed size of the link header (variable length 2476 * headers are not yet supporrted). 2477 */ 2478void 2479bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2480{ 2481 struct bpf_if *bp; 2482 2483 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); 2484 if (bp == NULL) 2485 panic("bpfattach"); 2486 2487 LIST_INIT(&bp->bif_dlist); 2488 LIST_INIT(&bp->bif_wlist); 2489 bp->bif_ifp = ifp; 2490 bp->bif_dlt = dlt; 2491 rw_init(&bp->bif_lock, "bpf interface lock"); 2492 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); 2493 *driverp = bp; 2494 2495 BPF_LOCK(); 2496 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2497 BPF_UNLOCK(); 2498 2499 bp->bif_hdrlen = hdrlen; 2500 2501 if (bootverbose) 2502 if_printf(ifp, "bpf attached\n"); 2503} 2504 2505/* 2506 * Detach bpf from an interface. This involves detaching each descriptor 2507 * associated with the interface. Notify each descriptor as it's detached 2508 * so that any sleepers wake up and get ENXIO. 2509 */ 2510void 2511bpfdetach(struct ifnet *ifp) 2512{ 2513 struct bpf_if *bp, *bp_temp; 2514 struct bpf_d *d; 2515 int ndetached; 2516 2517 ndetached = 0; 2518 2519 BPF_LOCK(); 2520 /* Find all bpf_if struct's which reference ifp and detach them. */ 2521 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2522 if (ifp != bp->bif_ifp) 2523 continue; 2524 2525 LIST_REMOVE(bp, bif_next); 2526 /* Add to to-be-freed list */ 2527 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); 2528 2529 ndetached++; 2530 /* 2531 * Delay freeing bp till interface is detached 2532 * and all routes through this interface are removed. 2533 * Mark bp as detached to restrict new consumers. 2534 */ 2535 BPFIF_WLOCK(bp); 2536 bp->flags |= BPFIF_FLAG_DYING; 2537 BPFIF_WUNLOCK(bp); 2538 2539 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", 2540 __func__, bp->bif_dlt, bp, ifp); 2541 2542 /* Free common descriptors */ 2543 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { 2544 bpf_detachd_locked(d); 2545 BPFD_LOCK(d); 2546 bpf_wakeup(d); 2547 BPFD_UNLOCK(d); 2548 } 2549 2550 /* Free writer-only descriptors */ 2551 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { 2552 bpf_detachd_locked(d); 2553 BPFD_LOCK(d); 2554 bpf_wakeup(d); 2555 BPFD_UNLOCK(d); 2556 } 2557 } 2558 BPF_UNLOCK(); 2559 2560#ifdef INVARIANTS 2561 if (ndetached == 0) 2562 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 2563#endif 2564} 2565 2566/* 2567 * Interface departure handler. 2568 * Note departure event does not guarantee interface is going down. 2569 * Interface renaming is currently done via departure/arrival event set. 2570 * 2571 * Departure handled is called after all routes pointing to 2572 * given interface are removed and interface is in down state 2573 * restricting any packets to be sent/received. We assume it is now safe 2574 * to free data allocated by BPF. 2575 */ 2576static void 2577bpf_ifdetach(void *arg __unused, struct ifnet *ifp) 2578{ 2579 struct bpf_if *bp, *bp_temp; 2580 int nmatched = 0; 2581 2582 BPF_LOCK(); 2583 /* 2584 * Find matching entries in free list. 2585 * Nothing should be found if bpfdetach() was not called. 2586 */ 2587 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { 2588 if (ifp != bp->bif_ifp) 2589 continue; 2590 2591 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", 2592 __func__, bp, ifp); 2593 2594 LIST_REMOVE(bp, bif_next); 2595 2596 rw_destroy(&bp->bif_lock); 2597 free(bp, M_BPF); 2598 2599 nmatched++; 2600 } 2601 BPF_UNLOCK(); 2602 2603 /* 2604 * Note that we cannot zero other pointers to 2605 * custom DLTs possibly used by given interface. 2606 */ 2607 if (nmatched != 0) 2608 ifp->if_bpf = NULL; 2609} 2610 2611/* 2612 * Get a list of available data link type of the interface. 2613 */ 2614static int 2615bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2616{ 2617 int n, error; 2618 struct ifnet *ifp; 2619 struct bpf_if *bp; 2620 2621 BPF_LOCK_ASSERT(); 2622 2623 ifp = d->bd_bif->bif_ifp; 2624 n = 0; 2625 error = 0; 2626 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2627 if (bp->bif_ifp != ifp) 2628 continue; 2629 if (bfl->bfl_list != NULL) { 2630 if (n >= bfl->bfl_len) 2631 return (ENOMEM); 2632 error = copyout(&bp->bif_dlt, 2633 bfl->bfl_list + n, sizeof(u_int)); 2634 } 2635 n++; 2636 } 2637 bfl->bfl_len = n; 2638 return (error); 2639} 2640 2641/* 2642 * Set the data link type of a BPF instance. 2643 */ 2644static int 2645bpf_setdlt(struct bpf_d *d, u_int dlt) 2646{ 2647 int error, opromisc; 2648 struct ifnet *ifp; 2649 struct bpf_if *bp; 2650 2651 BPF_LOCK_ASSERT(); 2652 2653 if (d->bd_bif->bif_dlt == dlt) 2654 return (0); 2655 ifp = d->bd_bif->bif_ifp; 2656 2657 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2658 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2659 break; 2660 } 2661 2662 if (bp != NULL) { 2663 opromisc = d->bd_promisc; 2664 bpf_attachd(d, bp); 2665 BPFD_LOCK(d); 2666 reset_d(d); 2667 BPFD_UNLOCK(d); 2668 if (opromisc) { 2669 error = ifpromisc(bp->bif_ifp, 1); 2670 if (error) 2671 if_printf(bp->bif_ifp, 2672 "bpf_setdlt: ifpromisc failed (%d)\n", 2673 error); 2674 else 2675 d->bd_promisc = 1; 2676 } 2677 } 2678 return (bp == NULL ? EINVAL : 0); 2679} 2680 2681static void 2682bpf_drvinit(void *unused) 2683{ 2684 struct cdev *dev; 2685 2686 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF); 2687 LIST_INIT(&bpf_iflist); 2688 LIST_INIT(&bpf_freelist); 2689 2690 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2691 /* For compatibility */ 2692 make_dev_alias(dev, "bpf0"); 2693 2694 /* Register interface departure handler */ 2695 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( 2696 ifnet_departure_event, bpf_ifdetach, NULL, 2697 EVENTHANDLER_PRI_ANY); 2698} 2699 2700/* 2701 * Zero out the various packet counters associated with all of the bpf 2702 * descriptors. At some point, we will probably want to get a bit more 2703 * granular and allow the user to specify descriptors to be zeroed. 2704 */ 2705static void 2706bpf_zero_counters(void) 2707{ 2708 struct bpf_if *bp; 2709 struct bpf_d *bd; 2710 2711 BPF_LOCK(); 2712 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2713 BPFIF_RLOCK(bp); 2714 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2715 BPFD_LOCK(bd); 2716 bd->bd_rcount = 0; 2717 bd->bd_dcount = 0; 2718 bd->bd_fcount = 0; 2719 bd->bd_wcount = 0; 2720 bd->bd_wfcount = 0; 2721 bd->bd_zcopy = 0; 2722 BPFD_UNLOCK(bd); 2723 } 2724 BPFIF_RUNLOCK(bp); 2725 } 2726 BPF_UNLOCK(); 2727} 2728 2729/* 2730 * Fill filter statistics 2731 */ 2732static void 2733bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2734{ 2735 2736 bzero(d, sizeof(*d)); 2737 BPFD_LOCK_ASSERT(bd); 2738 d->bd_structsize = sizeof(*d); 2739 /* XXX: reading should be protected by global lock */ 2740 d->bd_immediate = bd->bd_immediate; 2741 d->bd_promisc = bd->bd_promisc; 2742 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2743 d->bd_direction = bd->bd_direction; 2744 d->bd_feedback = bd->bd_feedback; 2745 d->bd_async = bd->bd_async; 2746 d->bd_rcount = bd->bd_rcount; 2747 d->bd_dcount = bd->bd_dcount; 2748 d->bd_fcount = bd->bd_fcount; 2749 d->bd_sig = bd->bd_sig; 2750 d->bd_slen = bd->bd_slen; 2751 d->bd_hlen = bd->bd_hlen; 2752 d->bd_bufsize = bd->bd_bufsize; 2753 d->bd_pid = bd->bd_pid; 2754 strlcpy(d->bd_ifname, 2755 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2756 d->bd_locked = bd->bd_locked; 2757 d->bd_wcount = bd->bd_wcount; 2758 d->bd_wdcount = bd->bd_wdcount; 2759 d->bd_wfcount = bd->bd_wfcount; 2760 d->bd_zcopy = bd->bd_zcopy; 2761 d->bd_bufmode = bd->bd_bufmode; 2762} 2763 2764/* 2765 * Handle `netstat -B' stats request 2766 */ 2767static int 2768bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2769{ 2770 struct xbpf_d *xbdbuf, *xbd, zerostats; 2771 int index, error; 2772 struct bpf_if *bp; 2773 struct bpf_d *bd; 2774 2775 /* 2776 * XXX This is not technically correct. It is possible for non 2777 * privileged users to open bpf devices. It would make sense 2778 * if the users who opened the devices were able to retrieve 2779 * the statistics for them, too. 2780 */ 2781 error = priv_check(req->td, PRIV_NET_BPF); 2782 if (error) 2783 return (error); 2784 /* 2785 * Check to see if the user is requesting that the counters be 2786 * zeroed out. Explicitly check that the supplied data is zeroed, 2787 * as we aren't allowing the user to set the counters currently. 2788 */ 2789 if (req->newptr != NULL) { 2790 if (req->newlen != sizeof(zerostats)) 2791 return (EINVAL); 2792 bzero(&zerostats, sizeof(zerostats)); 2793 xbd = req->newptr; 2794 if (bcmp(xbd, &zerostats, sizeof(*xbd)) != 0) 2795 return (EINVAL); 2796 bpf_zero_counters(); 2797 return (0); 2798 } 2799 if (req->oldptr == NULL) 2800 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2801 if (bpf_bpfd_cnt == 0) 2802 return (SYSCTL_OUT(req, 0, 0)); 2803 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2804 BPF_LOCK(); 2805 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2806 BPF_UNLOCK(); 2807 free(xbdbuf, M_BPF); 2808 return (ENOMEM); 2809 } 2810 index = 0; 2811 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2812 BPFIF_RLOCK(bp); 2813 /* Send writers-only first */ 2814 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2815 xbd = &xbdbuf[index++]; 2816 BPFD_LOCK(bd); 2817 bpfstats_fill_xbpf(xbd, bd); 2818 BPFD_UNLOCK(bd); 2819 } 2820 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2821 xbd = &xbdbuf[index++]; 2822 BPFD_LOCK(bd); 2823 bpfstats_fill_xbpf(xbd, bd); 2824 BPFD_UNLOCK(bd); 2825 } 2826 BPFIF_RUNLOCK(bp); 2827 } 2828 BPF_UNLOCK(); 2829 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2830 free(xbdbuf, M_BPF); 2831 return (error); 2832} 2833 2834SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2835 2836#else /* !DEV_BPF && !NETGRAPH_BPF */ 2837/* 2838 * NOP stubs to allow bpf-using drivers to load and function. 2839 * 2840 * A 'better' implementation would allow the core bpf functionality 2841 * to be loaded at runtime. 2842 */ 2843static struct bpf_if bp_null; 2844 2845void 2846bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2847{ 2848} 2849 2850void 2851bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2852{ 2853} 2854 2855void 2856bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 2857{ 2858} 2859 2860void 2861bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2862{ 2863 2864 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2865} 2866 2867void 2868bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2869{ 2870 2871 *driverp = &bp_null; 2872} 2873 2874void 2875bpfdetach(struct ifnet *ifp) 2876{ 2877} 2878 2879u_int 2880bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 2881{ 2882 return -1; /* "no filter" behaviour */ 2883} 2884 2885int 2886bpf_validate(const struct bpf_insn *f, int len) 2887{ 2888 return 0; /* false */ 2889} 2890 2891#endif /* !DEV_BPF && !NETGRAPH_BPF */ 2892