bpf.c revision 299615
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: stable/10/sys/net/bpf.c 299615 2016-05-13 08:04:57Z ngie $"); 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 if (len < hlen || len - hlen > ifp->if_mtu) 526 return (EMSGSIZE); 527 528 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); 529 if (m == NULL) 530 return (EIO); 531 m->m_pkthdr.len = m->m_len = len; 532 *mp = m; 533 534 error = uiomove(mtod(m, u_char *), len, uio); 535 if (error) 536 goto bad; 537 538 slen = bpf_filter(wfilter, mtod(m, u_char *), len, len); 539 if (slen == 0) { 540 error = EPERM; 541 goto bad; 542 } 543 544 /* Check for multicast destination */ 545 switch (linktype) { 546 case DLT_EN10MB: 547 eh = mtod(m, struct ether_header *); 548 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 549 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 550 ETHER_ADDR_LEN) == 0) 551 m->m_flags |= M_BCAST; 552 else 553 m->m_flags |= M_MCAST; 554 } 555 break; 556 } 557 558 /* 559 * Make room for link header, and copy it to sockaddr 560 */ 561 if (hlen != 0) { 562 if (sockp->sa_family == AF_IEEE80211) { 563 /* 564 * Collect true length from the parameter header 565 * NB: sockp is known to be zero'd so if we do a 566 * short copy unspecified parameters will be 567 * zero. 568 * NB: packet may not be aligned after stripping 569 * bpf params 570 * XXX check ibp_vers 571 */ 572 p = mtod(m, const struct ieee80211_bpf_params *); 573 hlen = p->ibp_len; 574 if (hlen > sizeof(sockp->sa_data)) { 575 error = EINVAL; 576 goto bad; 577 } 578 } 579 bcopy(m->m_data, sockp->sa_data, hlen); 580 } 581 *hdrlen = hlen; 582 583 return (0); 584bad: 585 m_freem(m); 586 return (error); 587} 588 589/* 590 * Attach file to the bpf interface, i.e. make d listen on bp. 591 */ 592static void 593bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 594{ 595 int op_w; 596 597 BPF_LOCK_ASSERT(); 598 599 /* 600 * Save sysctl value to protect from sysctl change 601 * between reads 602 */ 603 op_w = V_bpf_optimize_writers || d->bd_writer; 604 605 if (d->bd_bif != NULL) 606 bpf_detachd_locked(d); 607 /* 608 * Point d at bp, and add d to the interface's list. 609 * Since there are many applications using BPF for 610 * sending raw packets only (dhcpd, cdpd are good examples) 611 * we can delay adding d to the list of active listeners until 612 * some filter is configured. 613 */ 614 615 BPFIF_WLOCK(bp); 616 BPFD_LOCK(d); 617 618 d->bd_bif = bp; 619 620 if (op_w != 0) { 621 /* Add to writers-only list */ 622 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); 623 /* 624 * We decrement bd_writer on every filter set operation. 625 * First BIOCSETF is done by pcap_open_live() to set up 626 * snap length. After that appliation usually sets its own filter 627 */ 628 d->bd_writer = 2; 629 } else 630 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 631 632 BPFD_UNLOCK(d); 633 BPFIF_WUNLOCK(bp); 634 635 bpf_bpfd_cnt++; 636 637 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", 638 __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); 639 640 if (op_w == 0) 641 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 642} 643 644/* 645 * Add d to the list of active bp filters. 646 * Requires bpf_attachd() to be called before. 647 */ 648static void 649bpf_upgraded(struct bpf_d *d) 650{ 651 struct bpf_if *bp; 652 653 BPF_LOCK_ASSERT(); 654 655 bp = d->bd_bif; 656 657 /* 658 * Filter can be set several times without specifying interface. 659 * Mark d as reader and exit. 660 */ 661 if (bp == NULL) { 662 BPFD_LOCK(d); 663 d->bd_writer = 0; 664 BPFD_UNLOCK(d); 665 return; 666 } 667 668 BPFIF_WLOCK(bp); 669 BPFD_LOCK(d); 670 671 /* Remove from writers-only list */ 672 LIST_REMOVE(d, bd_next); 673 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 674 /* Mark d as reader */ 675 d->bd_writer = 0; 676 677 BPFD_UNLOCK(d); 678 BPFIF_WUNLOCK(bp); 679 680 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); 681 682 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 683} 684 685/* 686 * Detach a file from its interface. 687 */ 688static void 689bpf_detachd(struct bpf_d *d) 690{ 691 BPF_LOCK(); 692 bpf_detachd_locked(d); 693 BPF_UNLOCK(); 694} 695 696static void 697bpf_detachd_locked(struct bpf_d *d) 698{ 699 int error; 700 struct bpf_if *bp; 701 struct ifnet *ifp; 702 703 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); 704 705 BPF_LOCK_ASSERT(); 706 707 /* Check if descriptor is attached */ 708 if ((bp = d->bd_bif) == NULL) 709 return; 710 711 BPFIF_WLOCK(bp); 712 BPFD_LOCK(d); 713 714 /* Save bd_writer value */ 715 error = d->bd_writer; 716 717 /* 718 * Remove d from the interface's descriptor list. 719 */ 720 LIST_REMOVE(d, bd_next); 721 722 ifp = bp->bif_ifp; 723 d->bd_bif = NULL; 724 BPFD_UNLOCK(d); 725 BPFIF_WUNLOCK(bp); 726 727 bpf_bpfd_cnt--; 728 729 /* Call event handler iff d is attached */ 730 if (error == 0) 731 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); 732 733 /* 734 * Check if this descriptor had requested promiscuous mode. 735 * If so, turn it off. 736 */ 737 if (d->bd_promisc) { 738 d->bd_promisc = 0; 739 CURVNET_SET(ifp->if_vnet); 740 error = ifpromisc(ifp, 0); 741 CURVNET_RESTORE(); 742 if (error != 0 && error != ENXIO) { 743 /* 744 * ENXIO can happen if a pccard is unplugged 745 * Something is really wrong if we were able to put 746 * the driver into promiscuous mode, but can't 747 * take it out. 748 */ 749 if_printf(bp->bif_ifp, 750 "bpf_detach: ifpromisc failed (%d)\n", error); 751 } 752 } 753} 754 755/* 756 * Close the descriptor by detaching it from its interface, 757 * deallocating its buffers, and marking it free. 758 */ 759static void 760bpf_dtor(void *data) 761{ 762 struct bpf_d *d = data; 763 764 BPFD_LOCK(d); 765 if (d->bd_state == BPF_WAITING) 766 callout_stop(&d->bd_callout); 767 d->bd_state = BPF_IDLE; 768 BPFD_UNLOCK(d); 769 funsetown(&d->bd_sigio); 770 bpf_detachd(d); 771#ifdef MAC 772 mac_bpfdesc_destroy(d); 773#endif /* MAC */ 774 seldrain(&d->bd_sel); 775 knlist_destroy(&d->bd_sel.si_note); 776 callout_drain(&d->bd_callout); 777 bpf_freed(d); 778 free(d, M_BPF); 779} 780 781/* 782 * Open ethernet device. Returns ENXIO for illegal minor device number, 783 * EBUSY if file is open by another process. 784 */ 785/* ARGSUSED */ 786static int 787bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) 788{ 789 struct bpf_d *d; 790 int error; 791 792 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); 793 error = devfs_set_cdevpriv(d, bpf_dtor); 794 if (error != 0) { 795 free(d, M_BPF); 796 return (error); 797 } 798 799 /* 800 * For historical reasons, perform a one-time initialization call to 801 * the buffer routines, even though we're not yet committed to a 802 * particular buffer method. 803 */ 804 bpf_buffer_init(d); 805 if ((flags & FREAD) == 0) 806 d->bd_writer = 2; 807 d->bd_hbuf_in_use = 0; 808 d->bd_bufmode = BPF_BUFMODE_BUFFER; 809 d->bd_sig = SIGIO; 810 d->bd_direction = BPF_D_INOUT; 811 BPF_PID_REFRESH(d, td); 812#ifdef MAC 813 mac_bpfdesc_init(d); 814 mac_bpfdesc_create(td->td_ucred, d); 815#endif 816 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 817 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 818 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 819 820 return (0); 821} 822 823/* 824 * bpfread - read next chunk of packets from buffers 825 */ 826static int 827bpfread(struct cdev *dev, struct uio *uio, int ioflag) 828{ 829 struct bpf_d *d; 830 int error; 831 int non_block; 832 int timed_out; 833 834 error = devfs_get_cdevpriv((void **)&d); 835 if (error != 0) 836 return (error); 837 838 /* 839 * Restrict application to use a buffer the same size as 840 * as kernel buffers. 841 */ 842 if (uio->uio_resid != d->bd_bufsize) 843 return (EINVAL); 844 845 non_block = ((ioflag & O_NONBLOCK) != 0); 846 847 BPFD_LOCK(d); 848 BPF_PID_REFRESH_CUR(d); 849 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 850 BPFD_UNLOCK(d); 851 return (EOPNOTSUPP); 852 } 853 if (d->bd_state == BPF_WAITING) 854 callout_stop(&d->bd_callout); 855 timed_out = (d->bd_state == BPF_TIMED_OUT); 856 d->bd_state = BPF_IDLE; 857 while (d->bd_hbuf_in_use) { 858 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 859 PRINET|PCATCH, "bd_hbuf", 0); 860 if (error != 0) { 861 BPFD_UNLOCK(d); 862 return (error); 863 } 864 } 865 /* 866 * If the hold buffer is empty, then do a timed sleep, which 867 * ends when the timeout expires or when enough packets 868 * have arrived to fill the store buffer. 869 */ 870 while (d->bd_hbuf == NULL) { 871 if (d->bd_slen != 0) { 872 /* 873 * A packet(s) either arrived since the previous 874 * read or arrived while we were asleep. 875 */ 876 if (d->bd_immediate || non_block || timed_out) { 877 /* 878 * Rotate the buffers and return what's here 879 * if we are in immediate mode, non-blocking 880 * flag is set, or this descriptor timed out. 881 */ 882 ROTATE_BUFFERS(d); 883 break; 884 } 885 } 886 887 /* 888 * No data is available, check to see if the bpf device 889 * is still pointed at a real interface. If not, return 890 * ENXIO so that the userland process knows to rebind 891 * it before using it again. 892 */ 893 if (d->bd_bif == NULL) { 894 BPFD_UNLOCK(d); 895 return (ENXIO); 896 } 897 898 if (non_block) { 899 BPFD_UNLOCK(d); 900 return (EWOULDBLOCK); 901 } 902 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 903 "bpf", d->bd_rtout); 904 if (error == EINTR || error == ERESTART) { 905 BPFD_UNLOCK(d); 906 return (error); 907 } 908 if (error == EWOULDBLOCK) { 909 /* 910 * On a timeout, return what's in the buffer, 911 * which may be nothing. If there is something 912 * in the store buffer, we can rotate the buffers. 913 */ 914 if (d->bd_hbuf) 915 /* 916 * We filled up the buffer in between 917 * getting the timeout and arriving 918 * here, so we don't need to rotate. 919 */ 920 break; 921 922 if (d->bd_slen == 0) { 923 BPFD_UNLOCK(d); 924 return (0); 925 } 926 ROTATE_BUFFERS(d); 927 break; 928 } 929 } 930 /* 931 * At this point, we know we have something in the hold slot. 932 */ 933 d->bd_hbuf_in_use = 1; 934 BPFD_UNLOCK(d); 935 936 /* 937 * Move data from hold buffer into user space. 938 * We know the entire buffer is transferred since 939 * we checked above that the read buffer is bpf_bufsize bytes. 940 * 941 * We do not have to worry about simultaneous reads because 942 * we waited for sole access to the hold buffer above. 943 */ 944 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 945 946 BPFD_LOCK(d); 947 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 948 d->bd_fbuf = d->bd_hbuf; 949 d->bd_hbuf = NULL; 950 d->bd_hlen = 0; 951 bpf_buf_reclaimed(d); 952 d->bd_hbuf_in_use = 0; 953 wakeup(&d->bd_hbuf_in_use); 954 BPFD_UNLOCK(d); 955 956 return (error); 957} 958 959/* 960 * If there are processes sleeping on this descriptor, wake them up. 961 */ 962static __inline void 963bpf_wakeup(struct bpf_d *d) 964{ 965 966 BPFD_LOCK_ASSERT(d); 967 if (d->bd_state == BPF_WAITING) { 968 callout_stop(&d->bd_callout); 969 d->bd_state = BPF_IDLE; 970 } 971 wakeup(d); 972 if (d->bd_async && d->bd_sig && d->bd_sigio) 973 pgsigio(&d->bd_sigio, d->bd_sig, 0); 974 975 selwakeuppri(&d->bd_sel, PRINET); 976 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 977} 978 979static void 980bpf_timed_out(void *arg) 981{ 982 struct bpf_d *d = (struct bpf_d *)arg; 983 984 BPFD_LOCK_ASSERT(d); 985 986 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) 987 return; 988 if (d->bd_state == BPF_WAITING) { 989 d->bd_state = BPF_TIMED_OUT; 990 if (d->bd_slen != 0) 991 bpf_wakeup(d); 992 } 993} 994 995static int 996bpf_ready(struct bpf_d *d) 997{ 998 999 BPFD_LOCK_ASSERT(d); 1000 1001 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1002 return (1); 1003 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1004 d->bd_slen != 0) 1005 return (1); 1006 return (0); 1007} 1008 1009static int 1010bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1011{ 1012 struct bpf_d *d; 1013 struct ifnet *ifp; 1014 struct mbuf *m, *mc; 1015 struct sockaddr dst; 1016 int error, hlen; 1017 1018 error = devfs_get_cdevpriv((void **)&d); 1019 if (error != 0) 1020 return (error); 1021 1022 BPF_PID_REFRESH_CUR(d); 1023 d->bd_wcount++; 1024 /* XXX: locking required */ 1025 if (d->bd_bif == NULL) { 1026 d->bd_wdcount++; 1027 return (ENXIO); 1028 } 1029 1030 ifp = d->bd_bif->bif_ifp; 1031 1032 if ((ifp->if_flags & IFF_UP) == 0) { 1033 d->bd_wdcount++; 1034 return (ENETDOWN); 1035 } 1036 1037 if (uio->uio_resid == 0) { 1038 d->bd_wdcount++; 1039 return (0); 1040 } 1041 1042 bzero(&dst, sizeof(dst)); 1043 m = NULL; 1044 hlen = 0; 1045 /* XXX: bpf_movein() can sleep */ 1046 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, 1047 &m, &dst, &hlen, d->bd_wfilter); 1048 if (error) { 1049 d->bd_wdcount++; 1050 return (error); 1051 } 1052 d->bd_wfcount++; 1053 if (d->bd_hdrcmplt) 1054 dst.sa_family = pseudo_AF_HDRCMPLT; 1055 1056 if (d->bd_feedback) { 1057 mc = m_dup(m, M_NOWAIT); 1058 if (mc != NULL) 1059 mc->m_pkthdr.rcvif = ifp; 1060 /* Set M_PROMISC for outgoing packets to be discarded. */ 1061 if (d->bd_direction == BPF_D_INOUT) 1062 m->m_flags |= M_PROMISC; 1063 } else 1064 mc = NULL; 1065 1066 m->m_pkthdr.len -= hlen; 1067 m->m_len -= hlen; 1068 m->m_data += hlen; /* XXX */ 1069 1070 CURVNET_SET(ifp->if_vnet); 1071#ifdef MAC 1072 BPFD_LOCK(d); 1073 mac_bpfdesc_create_mbuf(d, m); 1074 if (mc != NULL) 1075 mac_bpfdesc_create_mbuf(d, mc); 1076 BPFD_UNLOCK(d); 1077#endif 1078 1079 error = (*ifp->if_output)(ifp, m, &dst, NULL); 1080 if (error) 1081 d->bd_wdcount++; 1082 1083 if (mc != NULL) { 1084 if (error == 0) 1085 (*ifp->if_input)(ifp, mc); 1086 else 1087 m_freem(mc); 1088 } 1089 CURVNET_RESTORE(); 1090 1091 return (error); 1092} 1093 1094/* 1095 * Reset a descriptor by flushing its packet buffer and clearing the receive 1096 * and drop counts. This is doable for kernel-only buffers, but with 1097 * zero-copy buffers, we can't write to (or rotate) buffers that are 1098 * currently owned by userspace. It would be nice if we could encapsulate 1099 * this logic in the buffer code rather than here. 1100 */ 1101static void 1102reset_d(struct bpf_d *d) 1103{ 1104 1105 BPFD_LOCK_ASSERT(d); 1106 1107 while (d->bd_hbuf_in_use) 1108 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1109 "bd_hbuf", 0); 1110 if ((d->bd_hbuf != NULL) && 1111 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1112 /* Free the hold buffer. */ 1113 d->bd_fbuf = d->bd_hbuf; 1114 d->bd_hbuf = NULL; 1115 d->bd_hlen = 0; 1116 bpf_buf_reclaimed(d); 1117 } 1118 if (bpf_canwritebuf(d)) 1119 d->bd_slen = 0; 1120 d->bd_rcount = 0; 1121 d->bd_dcount = 0; 1122 d->bd_fcount = 0; 1123 d->bd_wcount = 0; 1124 d->bd_wfcount = 0; 1125 d->bd_wdcount = 0; 1126 d->bd_zcopy = 0; 1127} 1128 1129/* 1130 * FIONREAD Check for read packet available. 1131 * SIOCGIFADDR Get interface address - convenient hook to driver. 1132 * BIOCGBLEN Get buffer len [for read()]. 1133 * BIOCSETF Set read filter. 1134 * BIOCSETFNR Set read filter without resetting descriptor. 1135 * BIOCSETWF Set write filter. 1136 * BIOCFLUSH Flush read packet buffer. 1137 * BIOCPROMISC Put interface into promiscuous mode. 1138 * BIOCGDLT Get link layer type. 1139 * BIOCGETIF Get interface name. 1140 * BIOCSETIF Set interface. 1141 * BIOCSRTIMEOUT Set read timeout. 1142 * BIOCGRTIMEOUT Get read timeout. 1143 * BIOCGSTATS Get packet stats. 1144 * BIOCIMMEDIATE Set immediate mode. 1145 * BIOCVERSION Get filter language version. 1146 * BIOCGHDRCMPLT Get "header already complete" flag 1147 * BIOCSHDRCMPLT Set "header already complete" flag 1148 * BIOCGDIRECTION Get packet direction flag 1149 * BIOCSDIRECTION Set packet direction flag 1150 * BIOCGTSTAMP Get time stamp format and resolution. 1151 * BIOCSTSTAMP Set time stamp format and resolution. 1152 * BIOCLOCK Set "locked" flag 1153 * BIOCFEEDBACK Set packet feedback mode. 1154 * BIOCSETZBUF Set current zero-copy buffer locations. 1155 * BIOCGETZMAX Get maximum zero-copy buffer size. 1156 * BIOCROTZBUF Force rotation of zero-copy buffer 1157 * BIOCSETBUFMODE Set buffer mode. 1158 * BIOCGETBUFMODE Get current buffer mode. 1159 */ 1160/* ARGSUSED */ 1161static int 1162bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1163 struct thread *td) 1164{ 1165 struct bpf_d *d; 1166 int error; 1167 1168 error = devfs_get_cdevpriv((void **)&d); 1169 if (error != 0) 1170 return (error); 1171 1172 /* 1173 * Refresh PID associated with this descriptor. 1174 */ 1175 BPFD_LOCK(d); 1176 BPF_PID_REFRESH(d, td); 1177 if (d->bd_state == BPF_WAITING) 1178 callout_stop(&d->bd_callout); 1179 d->bd_state = BPF_IDLE; 1180 BPFD_UNLOCK(d); 1181 1182 if (d->bd_locked == 1) { 1183 switch (cmd) { 1184 case BIOCGBLEN: 1185 case BIOCFLUSH: 1186 case BIOCGDLT: 1187 case BIOCGDLTLIST: 1188#ifdef COMPAT_FREEBSD32 1189 case BIOCGDLTLIST32: 1190#endif 1191 case BIOCGETIF: 1192 case BIOCGRTIMEOUT: 1193#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1194 case BIOCGRTIMEOUT32: 1195#endif 1196 case BIOCGSTATS: 1197 case BIOCVERSION: 1198 case BIOCGRSIG: 1199 case BIOCGHDRCMPLT: 1200 case BIOCSTSTAMP: 1201 case BIOCFEEDBACK: 1202 case FIONREAD: 1203 case BIOCLOCK: 1204 case BIOCSRTIMEOUT: 1205#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1206 case BIOCSRTIMEOUT32: 1207#endif 1208 case BIOCIMMEDIATE: 1209 case TIOCGPGRP: 1210 case BIOCROTZBUF: 1211 break; 1212 default: 1213 return (EPERM); 1214 } 1215 } 1216#ifdef COMPAT_FREEBSD32 1217 /* 1218 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1219 * that it will get 32-bit packet headers. 1220 */ 1221 switch (cmd) { 1222 case BIOCSETF32: 1223 case BIOCSETFNR32: 1224 case BIOCSETWF32: 1225 case BIOCGDLTLIST32: 1226 case BIOCGRTIMEOUT32: 1227 case BIOCSRTIMEOUT32: 1228 BPFD_LOCK(d); 1229 d->bd_compat32 = 1; 1230 BPFD_UNLOCK(d); 1231 } 1232#endif 1233 1234 CURVNET_SET(TD_TO_VNET(td)); 1235 switch (cmd) { 1236 1237 default: 1238 error = EINVAL; 1239 break; 1240 1241 /* 1242 * Check for read packet available. 1243 */ 1244 case FIONREAD: 1245 { 1246 int n; 1247 1248 BPFD_LOCK(d); 1249 n = d->bd_slen; 1250 while (d->bd_hbuf_in_use) 1251 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1252 PRINET, "bd_hbuf", 0); 1253 if (d->bd_hbuf) 1254 n += d->bd_hlen; 1255 BPFD_UNLOCK(d); 1256 1257 *(int *)addr = n; 1258 break; 1259 } 1260 1261 case SIOCGIFADDR: 1262 { 1263 struct ifnet *ifp; 1264 1265 if (d->bd_bif == NULL) 1266 error = EINVAL; 1267 else { 1268 ifp = d->bd_bif->bif_ifp; 1269 error = (*ifp->if_ioctl)(ifp, cmd, addr); 1270 } 1271 break; 1272 } 1273 1274 /* 1275 * Get buffer len [for read()]. 1276 */ 1277 case BIOCGBLEN: 1278 BPFD_LOCK(d); 1279 *(u_int *)addr = d->bd_bufsize; 1280 BPFD_UNLOCK(d); 1281 break; 1282 1283 /* 1284 * Set buffer length. 1285 */ 1286 case BIOCSBLEN: 1287 error = bpf_ioctl_sblen(d, (u_int *)addr); 1288 break; 1289 1290 /* 1291 * Set link layer read filter. 1292 */ 1293 case BIOCSETF: 1294 case BIOCSETFNR: 1295 case BIOCSETWF: 1296#ifdef COMPAT_FREEBSD32 1297 case BIOCSETF32: 1298 case BIOCSETFNR32: 1299 case BIOCSETWF32: 1300#endif 1301 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1302 break; 1303 1304 /* 1305 * Flush read packet buffer. 1306 */ 1307 case BIOCFLUSH: 1308 BPFD_LOCK(d); 1309 reset_d(d); 1310 BPFD_UNLOCK(d); 1311 break; 1312 1313 /* 1314 * Put interface into promiscuous mode. 1315 */ 1316 case BIOCPROMISC: 1317 if (d->bd_bif == NULL) { 1318 /* 1319 * No interface attached yet. 1320 */ 1321 error = EINVAL; 1322 break; 1323 } 1324 if (d->bd_promisc == 0) { 1325 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1326 if (error == 0) 1327 d->bd_promisc = 1; 1328 } 1329 break; 1330 1331 /* 1332 * Get current data link type. 1333 */ 1334 case BIOCGDLT: 1335 BPF_LOCK(); 1336 if (d->bd_bif == NULL) 1337 error = EINVAL; 1338 else 1339 *(u_int *)addr = d->bd_bif->bif_dlt; 1340 BPF_UNLOCK(); 1341 break; 1342 1343 /* 1344 * Get a list of supported data link types. 1345 */ 1346#ifdef COMPAT_FREEBSD32 1347 case BIOCGDLTLIST32: 1348 { 1349 struct bpf_dltlist32 *list32; 1350 struct bpf_dltlist dltlist; 1351 1352 list32 = (struct bpf_dltlist32 *)addr; 1353 dltlist.bfl_len = list32->bfl_len; 1354 dltlist.bfl_list = PTRIN(list32->bfl_list); 1355 BPF_LOCK(); 1356 if (d->bd_bif == NULL) 1357 error = EINVAL; 1358 else { 1359 error = bpf_getdltlist(d, &dltlist); 1360 if (error == 0) 1361 list32->bfl_len = dltlist.bfl_len; 1362 } 1363 BPF_UNLOCK(); 1364 break; 1365 } 1366#endif 1367 1368 case BIOCGDLTLIST: 1369 BPF_LOCK(); 1370 if (d->bd_bif == NULL) 1371 error = EINVAL; 1372 else 1373 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1374 BPF_UNLOCK(); 1375 break; 1376 1377 /* 1378 * Set data link type. 1379 */ 1380 case BIOCSDLT: 1381 BPF_LOCK(); 1382 if (d->bd_bif == NULL) 1383 error = EINVAL; 1384 else 1385 error = bpf_setdlt(d, *(u_int *)addr); 1386 BPF_UNLOCK(); 1387 break; 1388 1389 /* 1390 * Get interface name. 1391 */ 1392 case BIOCGETIF: 1393 BPF_LOCK(); 1394 if (d->bd_bif == NULL) 1395 error = EINVAL; 1396 else { 1397 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1398 struct ifreq *const ifr = (struct ifreq *)addr; 1399 1400 strlcpy(ifr->ifr_name, ifp->if_xname, 1401 sizeof(ifr->ifr_name)); 1402 } 1403 BPF_UNLOCK(); 1404 break; 1405 1406 /* 1407 * Set interface. 1408 */ 1409 case BIOCSETIF: 1410 { 1411 int alloc_buf, size; 1412 1413 /* 1414 * Behavior here depends on the buffering model. If 1415 * we're using kernel memory buffers, then we can 1416 * allocate them here. If we're using zero-copy, 1417 * then the user process must have registered buffers 1418 * by the time we get here. 1419 */ 1420 alloc_buf = 0; 1421 BPFD_LOCK(d); 1422 if (d->bd_bufmode == BPF_BUFMODE_BUFFER && 1423 d->bd_sbuf == NULL) 1424 alloc_buf = 1; 1425 BPFD_UNLOCK(d); 1426 if (alloc_buf) { 1427 size = d->bd_bufsize; 1428 error = bpf_buffer_ioctl_sblen(d, &size); 1429 if (error != 0) 1430 break; 1431 } 1432 BPF_LOCK(); 1433 error = bpf_setif(d, (struct ifreq *)addr); 1434 BPF_UNLOCK(); 1435 break; 1436 } 1437 1438 /* 1439 * Set read timeout. 1440 */ 1441 case BIOCSRTIMEOUT: 1442#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1443 case BIOCSRTIMEOUT32: 1444#endif 1445 { 1446 struct timeval *tv = (struct timeval *)addr; 1447#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1448 struct timeval32 *tv32; 1449 struct timeval tv64; 1450 1451 if (cmd == BIOCSRTIMEOUT32) { 1452 tv32 = (struct timeval32 *)addr; 1453 tv = &tv64; 1454 tv->tv_sec = tv32->tv_sec; 1455 tv->tv_usec = tv32->tv_usec; 1456 } else 1457#endif 1458 tv = (struct timeval *)addr; 1459 1460 /* 1461 * Subtract 1 tick from tvtohz() since this isn't 1462 * a one-shot timer. 1463 */ 1464 if ((error = itimerfix(tv)) == 0) 1465 d->bd_rtout = tvtohz(tv) - 1; 1466 break; 1467 } 1468 1469 /* 1470 * Get read timeout. 1471 */ 1472 case BIOCGRTIMEOUT: 1473#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1474 case BIOCGRTIMEOUT32: 1475#endif 1476 { 1477 struct timeval *tv; 1478#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1479 struct timeval32 *tv32; 1480 struct timeval tv64; 1481 1482 if (cmd == BIOCGRTIMEOUT32) 1483 tv = &tv64; 1484 else 1485#endif 1486 tv = (struct timeval *)addr; 1487 1488 tv->tv_sec = d->bd_rtout / hz; 1489 tv->tv_usec = (d->bd_rtout % hz) * tick; 1490#if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1491 if (cmd == BIOCGRTIMEOUT32) { 1492 tv32 = (struct timeval32 *)addr; 1493 tv32->tv_sec = tv->tv_sec; 1494 tv32->tv_usec = tv->tv_usec; 1495 } 1496#endif 1497 1498 break; 1499 } 1500 1501 /* 1502 * Get packet stats. 1503 */ 1504 case BIOCGSTATS: 1505 { 1506 struct bpf_stat *bs = (struct bpf_stat *)addr; 1507 1508 /* XXXCSJP overflow */ 1509 bs->bs_recv = d->bd_rcount; 1510 bs->bs_drop = d->bd_dcount; 1511 break; 1512 } 1513 1514 /* 1515 * Set immediate mode. 1516 */ 1517 case BIOCIMMEDIATE: 1518 BPFD_LOCK(d); 1519 d->bd_immediate = *(u_int *)addr; 1520 BPFD_UNLOCK(d); 1521 break; 1522 1523 case BIOCVERSION: 1524 { 1525 struct bpf_version *bv = (struct bpf_version *)addr; 1526 1527 bv->bv_major = BPF_MAJOR_VERSION; 1528 bv->bv_minor = BPF_MINOR_VERSION; 1529 break; 1530 } 1531 1532 /* 1533 * Get "header already complete" flag 1534 */ 1535 case BIOCGHDRCMPLT: 1536 BPFD_LOCK(d); 1537 *(u_int *)addr = d->bd_hdrcmplt; 1538 BPFD_UNLOCK(d); 1539 break; 1540 1541 /* 1542 * Set "header already complete" flag 1543 */ 1544 case BIOCSHDRCMPLT: 1545 BPFD_LOCK(d); 1546 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1547 BPFD_UNLOCK(d); 1548 break; 1549 1550 /* 1551 * Get packet direction flag 1552 */ 1553 case BIOCGDIRECTION: 1554 BPFD_LOCK(d); 1555 *(u_int *)addr = d->bd_direction; 1556 BPFD_UNLOCK(d); 1557 break; 1558 1559 /* 1560 * Set packet direction flag 1561 */ 1562 case BIOCSDIRECTION: 1563 { 1564 u_int direction; 1565 1566 direction = *(u_int *)addr; 1567 switch (direction) { 1568 case BPF_D_IN: 1569 case BPF_D_INOUT: 1570 case BPF_D_OUT: 1571 BPFD_LOCK(d); 1572 d->bd_direction = direction; 1573 BPFD_UNLOCK(d); 1574 break; 1575 default: 1576 error = EINVAL; 1577 } 1578 } 1579 break; 1580 1581 /* 1582 * Get packet timestamp format and resolution. 1583 */ 1584 case BIOCGTSTAMP: 1585 BPFD_LOCK(d); 1586 *(u_int *)addr = d->bd_tstamp; 1587 BPFD_UNLOCK(d); 1588 break; 1589 1590 /* 1591 * Set packet timestamp format and resolution. 1592 */ 1593 case BIOCSTSTAMP: 1594 { 1595 u_int func; 1596 1597 func = *(u_int *)addr; 1598 if (BPF_T_VALID(func)) 1599 d->bd_tstamp = func; 1600 else 1601 error = EINVAL; 1602 } 1603 break; 1604 1605 case BIOCFEEDBACK: 1606 BPFD_LOCK(d); 1607 d->bd_feedback = *(u_int *)addr; 1608 BPFD_UNLOCK(d); 1609 break; 1610 1611 case BIOCLOCK: 1612 BPFD_LOCK(d); 1613 d->bd_locked = 1; 1614 BPFD_UNLOCK(d); 1615 break; 1616 1617 case FIONBIO: /* Non-blocking I/O */ 1618 break; 1619 1620 case FIOASYNC: /* Send signal on receive packets */ 1621 BPFD_LOCK(d); 1622 d->bd_async = *(int *)addr; 1623 BPFD_UNLOCK(d); 1624 break; 1625 1626 case FIOSETOWN: 1627 /* 1628 * XXX: Add some sort of locking here? 1629 * fsetown() can sleep. 1630 */ 1631 error = fsetown(*(int *)addr, &d->bd_sigio); 1632 break; 1633 1634 case FIOGETOWN: 1635 BPFD_LOCK(d); 1636 *(int *)addr = fgetown(&d->bd_sigio); 1637 BPFD_UNLOCK(d); 1638 break; 1639 1640 /* This is deprecated, FIOSETOWN should be used instead. */ 1641 case TIOCSPGRP: 1642 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1643 break; 1644 1645 /* This is deprecated, FIOGETOWN should be used instead. */ 1646 case TIOCGPGRP: 1647 *(int *)addr = -fgetown(&d->bd_sigio); 1648 break; 1649 1650 case BIOCSRSIG: /* Set receive signal */ 1651 { 1652 u_int sig; 1653 1654 sig = *(u_int *)addr; 1655 1656 if (sig >= NSIG) 1657 error = EINVAL; 1658 else { 1659 BPFD_LOCK(d); 1660 d->bd_sig = sig; 1661 BPFD_UNLOCK(d); 1662 } 1663 break; 1664 } 1665 case BIOCGRSIG: 1666 BPFD_LOCK(d); 1667 *(u_int *)addr = d->bd_sig; 1668 BPFD_UNLOCK(d); 1669 break; 1670 1671 case BIOCGETBUFMODE: 1672 BPFD_LOCK(d); 1673 *(u_int *)addr = d->bd_bufmode; 1674 BPFD_UNLOCK(d); 1675 break; 1676 1677 case BIOCSETBUFMODE: 1678 /* 1679 * Allow the buffering mode to be changed as long as we 1680 * haven't yet committed to a particular mode. Our 1681 * definition of commitment, for now, is whether or not a 1682 * buffer has been allocated or an interface attached, since 1683 * that's the point where things get tricky. 1684 */ 1685 switch (*(u_int *)addr) { 1686 case BPF_BUFMODE_BUFFER: 1687 break; 1688 1689 case BPF_BUFMODE_ZBUF: 1690 if (bpf_zerocopy_enable) 1691 break; 1692 /* FALLSTHROUGH */ 1693 1694 default: 1695 CURVNET_RESTORE(); 1696 return (EINVAL); 1697 } 1698 1699 BPFD_LOCK(d); 1700 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1701 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1702 BPFD_UNLOCK(d); 1703 CURVNET_RESTORE(); 1704 return (EBUSY); 1705 } 1706 d->bd_bufmode = *(u_int *)addr; 1707 BPFD_UNLOCK(d); 1708 break; 1709 1710 case BIOCGETZMAX: 1711 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1712 break; 1713 1714 case BIOCSETZBUF: 1715 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1716 break; 1717 1718 case BIOCROTZBUF: 1719 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1720 break; 1721 } 1722 CURVNET_RESTORE(); 1723 return (error); 1724} 1725 1726/* 1727 * Set d's packet filter program to fp. If this file already has a filter, 1728 * free it and replace it. Returns EINVAL for bogus requests. 1729 * 1730 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls 1731 * since reading d->bd_bif can't be protected by d or interface lock due to 1732 * lock order. 1733 * 1734 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses 1735 * interface read lock to read all filers. 1736 * 1737 */ 1738static int 1739bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1740{ 1741#ifdef COMPAT_FREEBSD32 1742 struct bpf_program fp_swab; 1743 struct bpf_program32 *fp32; 1744#endif 1745 struct bpf_insn *fcode, *old; 1746#ifdef BPF_JITTER 1747 bpf_jit_filter *jfunc, *ofunc; 1748#endif 1749 size_t size; 1750 u_int flen; 1751 int need_upgrade; 1752 1753#ifdef COMPAT_FREEBSD32 1754 switch (cmd) { 1755 case BIOCSETF32: 1756 case BIOCSETWF32: 1757 case BIOCSETFNR32: 1758 fp32 = (struct bpf_program32 *)fp; 1759 fp_swab.bf_len = fp32->bf_len; 1760 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1761 fp = &fp_swab; 1762 switch (cmd) { 1763 case BIOCSETF32: 1764 cmd = BIOCSETF; 1765 break; 1766 case BIOCSETWF32: 1767 cmd = BIOCSETWF; 1768 break; 1769 } 1770 break; 1771 } 1772#endif 1773 1774 fcode = NULL; 1775#ifdef BPF_JITTER 1776 jfunc = ofunc = NULL; 1777#endif 1778 need_upgrade = 0; 1779 1780 /* 1781 * Check new filter validness before acquiring any locks. 1782 * Allocate memory for new filter, if needed. 1783 */ 1784 flen = fp->bf_len; 1785 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1786 return (EINVAL); 1787 size = flen * sizeof(*fp->bf_insns); 1788 if (size > 0) { 1789 /* We're setting up new filter. Copy and check actual data. */ 1790 fcode = malloc(size, M_BPF, M_WAITOK); 1791 if (copyin(fp->bf_insns, fcode, size) != 0 || 1792 !bpf_validate(fcode, flen)) { 1793 free(fcode, M_BPF); 1794 return (EINVAL); 1795 } 1796#ifdef BPF_JITTER 1797 /* Filter is copied inside fcode and is perfectly valid. */ 1798 jfunc = bpf_jitter(fcode, flen); 1799#endif 1800 } 1801 1802 BPF_LOCK(); 1803 1804 /* 1805 * Set up new filter. 1806 * Protect filter change by interface lock. 1807 * Additionally, we are protected by global lock here. 1808 */ 1809 if (d->bd_bif != NULL) 1810 BPFIF_WLOCK(d->bd_bif); 1811 BPFD_LOCK(d); 1812 if (cmd == BIOCSETWF) { 1813 old = d->bd_wfilter; 1814 d->bd_wfilter = fcode; 1815 } else { 1816 old = d->bd_rfilter; 1817 d->bd_rfilter = fcode; 1818#ifdef BPF_JITTER 1819 ofunc = d->bd_bfilter; 1820 d->bd_bfilter = jfunc; 1821#endif 1822 if (cmd == BIOCSETF) 1823 reset_d(d); 1824 1825 if (fcode != NULL) { 1826 /* 1827 * Do not require upgrade by first BIOCSETF 1828 * (used to set snaplen) by pcap_open_live(). 1829 */ 1830 if (d->bd_writer != 0 && --d->bd_writer == 0) 1831 need_upgrade = 1; 1832 CTR4(KTR_NET, "%s: filter function set by pid %d, " 1833 "bd_writer counter %d, need_upgrade %d", 1834 __func__, d->bd_pid, d->bd_writer, need_upgrade); 1835 } 1836 } 1837 BPFD_UNLOCK(d); 1838 if (d->bd_bif != NULL) 1839 BPFIF_WUNLOCK(d->bd_bif); 1840 if (old != NULL) 1841 free(old, M_BPF); 1842#ifdef BPF_JITTER 1843 if (ofunc != NULL) 1844 bpf_destroy_jit_filter(ofunc); 1845#endif 1846 1847 /* Move d to active readers list. */ 1848 if (need_upgrade) 1849 bpf_upgraded(d); 1850 1851 BPF_UNLOCK(); 1852 return (0); 1853} 1854 1855/* 1856 * Detach a file from its current interface (if attached at all) and attach 1857 * to the interface indicated by the name stored in ifr. 1858 * Return an errno or 0. 1859 */ 1860static int 1861bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1862{ 1863 struct bpf_if *bp; 1864 struct ifnet *theywant; 1865 1866 BPF_LOCK_ASSERT(); 1867 1868 theywant = ifunit(ifr->ifr_name); 1869 if (theywant == NULL || theywant->if_bpf == NULL) 1870 return (ENXIO); 1871 1872 bp = theywant->if_bpf; 1873 1874 /* Check if interface is not being detached from BPF */ 1875 BPFIF_RLOCK(bp); 1876 if (bp->flags & BPFIF_FLAG_DYING) { 1877 BPFIF_RUNLOCK(bp); 1878 return (ENXIO); 1879 } 1880 BPFIF_RUNLOCK(bp); 1881 1882 /* 1883 * At this point, we expect the buffer is already allocated. If not, 1884 * return an error. 1885 */ 1886 switch (d->bd_bufmode) { 1887 case BPF_BUFMODE_BUFFER: 1888 case BPF_BUFMODE_ZBUF: 1889 if (d->bd_sbuf == NULL) 1890 return (EINVAL); 1891 break; 1892 1893 default: 1894 panic("bpf_setif: bufmode %d", d->bd_bufmode); 1895 } 1896 if (bp != d->bd_bif) 1897 bpf_attachd(d, bp); 1898 BPFD_LOCK(d); 1899 reset_d(d); 1900 BPFD_UNLOCK(d); 1901 return (0); 1902} 1903 1904/* 1905 * Support for select() and poll() system calls 1906 * 1907 * Return true iff the specific operation will not block indefinitely. 1908 * Otherwise, return false but make a note that a selwakeup() must be done. 1909 */ 1910static int 1911bpfpoll(struct cdev *dev, int events, struct thread *td) 1912{ 1913 struct bpf_d *d; 1914 int revents; 1915 1916 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 1917 return (events & 1918 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 1919 1920 /* 1921 * Refresh PID associated with this descriptor. 1922 */ 1923 revents = events & (POLLOUT | POLLWRNORM); 1924 BPFD_LOCK(d); 1925 BPF_PID_REFRESH(d, td); 1926 if (events & (POLLIN | POLLRDNORM)) { 1927 if (bpf_ready(d)) 1928 revents |= events & (POLLIN | POLLRDNORM); 1929 else { 1930 selrecord(td, &d->bd_sel); 1931 /* Start the read timeout if necessary. */ 1932 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1933 callout_reset(&d->bd_callout, d->bd_rtout, 1934 bpf_timed_out, d); 1935 d->bd_state = BPF_WAITING; 1936 } 1937 } 1938 } 1939 BPFD_UNLOCK(d); 1940 return (revents); 1941} 1942 1943/* 1944 * Support for kevent() system call. Register EVFILT_READ filters and 1945 * reject all others. 1946 */ 1947int 1948bpfkqfilter(struct cdev *dev, struct knote *kn) 1949{ 1950 struct bpf_d *d; 1951 1952 if (devfs_get_cdevpriv((void **)&d) != 0 || 1953 kn->kn_filter != EVFILT_READ) 1954 return (1); 1955 1956 /* 1957 * Refresh PID associated with this descriptor. 1958 */ 1959 BPFD_LOCK(d); 1960 BPF_PID_REFRESH_CUR(d); 1961 kn->kn_fop = &bpfread_filtops; 1962 kn->kn_hook = d; 1963 knlist_add(&d->bd_sel.si_note, kn, 1); 1964 BPFD_UNLOCK(d); 1965 1966 return (0); 1967} 1968 1969static void 1970filt_bpfdetach(struct knote *kn) 1971{ 1972 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 1973 1974 knlist_remove(&d->bd_sel.si_note, kn, 0); 1975} 1976 1977static int 1978filt_bpfread(struct knote *kn, long hint) 1979{ 1980 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 1981 int ready; 1982 1983 BPFD_LOCK_ASSERT(d); 1984 ready = bpf_ready(d); 1985 if (ready) { 1986 kn->kn_data = d->bd_slen; 1987 /* 1988 * Ignore the hold buffer if it is being copied to user space. 1989 */ 1990 if (!d->bd_hbuf_in_use && d->bd_hbuf) 1991 kn->kn_data += d->bd_hlen; 1992 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1993 callout_reset(&d->bd_callout, d->bd_rtout, 1994 bpf_timed_out, d); 1995 d->bd_state = BPF_WAITING; 1996 } 1997 1998 return (ready); 1999} 2000 2001#define BPF_TSTAMP_NONE 0 2002#define BPF_TSTAMP_FAST 1 2003#define BPF_TSTAMP_NORMAL 2 2004#define BPF_TSTAMP_EXTERN 3 2005 2006static int 2007bpf_ts_quality(int tstype) 2008{ 2009 2010 if (tstype == BPF_T_NONE) 2011 return (BPF_TSTAMP_NONE); 2012 if ((tstype & BPF_T_FAST) != 0) 2013 return (BPF_TSTAMP_FAST); 2014 2015 return (BPF_TSTAMP_NORMAL); 2016} 2017 2018static int 2019bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2020{ 2021 struct m_tag *tag; 2022 int quality; 2023 2024 quality = bpf_ts_quality(tstype); 2025 if (quality == BPF_TSTAMP_NONE) 2026 return (quality); 2027 2028 if (m != NULL) { 2029 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2030 if (tag != NULL) { 2031 *bt = *(struct bintime *)(tag + 1); 2032 return (BPF_TSTAMP_EXTERN); 2033 } 2034 } 2035 if (quality == BPF_TSTAMP_NORMAL) 2036 binuptime(bt); 2037 else 2038 getbinuptime(bt); 2039 2040 return (quality); 2041} 2042 2043/* 2044 * Incoming linkage from device drivers. Process the packet pkt, of length 2045 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2046 * by each process' filter, and if accepted, stashed into the corresponding 2047 * buffer. 2048 */ 2049void 2050bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2051{ 2052 struct bintime bt; 2053 struct bpf_d *d; 2054#ifdef BPF_JITTER 2055 bpf_jit_filter *bf; 2056#endif 2057 u_int slen; 2058 int gottime; 2059 2060 gottime = BPF_TSTAMP_NONE; 2061 2062 BPFIF_RLOCK(bp); 2063 2064 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2065 /* 2066 * We are not using any locks for d here because: 2067 * 1) any filter change is protected by interface 2068 * write lock 2069 * 2) destroying/detaching d is protected by interface 2070 * write lock, too 2071 */ 2072 2073 /* XXX: Do not protect counter for the sake of performance. */ 2074 ++d->bd_rcount; 2075 /* 2076 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no 2077 * way for the caller to indiciate to us whether this packet 2078 * is inbound or outbound. In the bpf_mtap() routines, we use 2079 * the interface pointers on the mbuf to figure it out. 2080 */ 2081#ifdef BPF_JITTER 2082 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2083 if (bf != NULL) 2084 slen = (*(bf->func))(pkt, pktlen, pktlen); 2085 else 2086#endif 2087 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2088 if (slen != 0) { 2089 /* 2090 * Filter matches. Let's to acquire write lock. 2091 */ 2092 BPFD_LOCK(d); 2093 2094 d->bd_fcount++; 2095 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2096 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); 2097#ifdef MAC 2098 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2099#endif 2100 catchpacket(d, pkt, pktlen, slen, 2101 bpf_append_bytes, &bt); 2102 BPFD_UNLOCK(d); 2103 } 2104 } 2105 BPFIF_RUNLOCK(bp); 2106} 2107 2108#define BPF_CHECK_DIRECTION(d, r, i) \ 2109 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2110 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2111 2112/* 2113 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2114 * Locking model is explained in bpf_tap(). 2115 */ 2116void 2117bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2118{ 2119 struct bintime bt; 2120 struct bpf_d *d; 2121#ifdef BPF_JITTER 2122 bpf_jit_filter *bf; 2123#endif 2124 u_int pktlen, slen; 2125 int gottime; 2126 2127 /* Skip outgoing duplicate packets. */ 2128 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2129 m->m_flags &= ~M_PROMISC; 2130 return; 2131 } 2132 2133 pktlen = m_length(m, NULL); 2134 gottime = BPF_TSTAMP_NONE; 2135 2136 BPFIF_RLOCK(bp); 2137 2138 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2139 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2140 continue; 2141 ++d->bd_rcount; 2142#ifdef BPF_JITTER 2143 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2144 /* XXX We cannot handle multiple mbufs. */ 2145 if (bf != NULL && m->m_next == NULL) 2146 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); 2147 else 2148#endif 2149 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2150 if (slen != 0) { 2151 BPFD_LOCK(d); 2152 2153 d->bd_fcount++; 2154 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2155 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2156#ifdef MAC 2157 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2158#endif 2159 catchpacket(d, (u_char *)m, pktlen, slen, 2160 bpf_append_mbuf, &bt); 2161 BPFD_UNLOCK(d); 2162 } 2163 } 2164 BPFIF_RUNLOCK(bp); 2165} 2166 2167/* 2168 * Incoming linkage from device drivers, when packet is in 2169 * an mbuf chain and to be prepended by a contiguous header. 2170 */ 2171void 2172bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2173{ 2174 struct bintime bt; 2175 struct mbuf mb; 2176 struct bpf_d *d; 2177 u_int pktlen, slen; 2178 int gottime; 2179 2180 /* Skip outgoing duplicate packets. */ 2181 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2182 m->m_flags &= ~M_PROMISC; 2183 return; 2184 } 2185 2186 pktlen = m_length(m, NULL); 2187 /* 2188 * Craft on-stack mbuf suitable for passing to bpf_filter. 2189 * Note that we cut corners here; we only setup what's 2190 * absolutely needed--this mbuf should never go anywhere else. 2191 */ 2192 mb.m_next = m; 2193 mb.m_data = data; 2194 mb.m_len = dlen; 2195 pktlen += dlen; 2196 2197 gottime = BPF_TSTAMP_NONE; 2198 2199 BPFIF_RLOCK(bp); 2200 2201 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2202 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2203 continue; 2204 ++d->bd_rcount; 2205 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2206 if (slen != 0) { 2207 BPFD_LOCK(d); 2208 2209 d->bd_fcount++; 2210 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2211 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2212#ifdef MAC 2213 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2214#endif 2215 catchpacket(d, (u_char *)&mb, pktlen, slen, 2216 bpf_append_mbuf, &bt); 2217 BPFD_UNLOCK(d); 2218 } 2219 } 2220 BPFIF_RUNLOCK(bp); 2221} 2222 2223#undef BPF_CHECK_DIRECTION 2224 2225#undef BPF_TSTAMP_NONE 2226#undef BPF_TSTAMP_FAST 2227#undef BPF_TSTAMP_NORMAL 2228#undef BPF_TSTAMP_EXTERN 2229 2230static int 2231bpf_hdrlen(struct bpf_d *d) 2232{ 2233 int hdrlen; 2234 2235 hdrlen = d->bd_bif->bif_hdrlen; 2236#ifndef BURN_BRIDGES 2237 if (d->bd_tstamp == BPF_T_NONE || 2238 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2239#ifdef COMPAT_FREEBSD32 2240 if (d->bd_compat32) 2241 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2242 else 2243#endif 2244 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2245 else 2246#endif 2247 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2248#ifdef COMPAT_FREEBSD32 2249 if (d->bd_compat32) 2250 hdrlen = BPF_WORDALIGN32(hdrlen); 2251 else 2252#endif 2253 hdrlen = BPF_WORDALIGN(hdrlen); 2254 2255 return (hdrlen - d->bd_bif->bif_hdrlen); 2256} 2257 2258static void 2259bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2260{ 2261 struct bintime bt2; 2262 struct timeval tsm; 2263 struct timespec tsn; 2264 2265 if ((tstype & BPF_T_MONOTONIC) == 0) { 2266 bt2 = *bt; 2267 bintime_add(&bt2, &boottimebin); 2268 bt = &bt2; 2269 } 2270 switch (BPF_T_FORMAT(tstype)) { 2271 case BPF_T_MICROTIME: 2272 bintime2timeval(bt, &tsm); 2273 ts->bt_sec = tsm.tv_sec; 2274 ts->bt_frac = tsm.tv_usec; 2275 break; 2276 case BPF_T_NANOTIME: 2277 bintime2timespec(bt, &tsn); 2278 ts->bt_sec = tsn.tv_sec; 2279 ts->bt_frac = tsn.tv_nsec; 2280 break; 2281 case BPF_T_BINTIME: 2282 ts->bt_sec = bt->sec; 2283 ts->bt_frac = bt->frac; 2284 break; 2285 } 2286} 2287 2288/* 2289 * Move the packet data from interface memory (pkt) into the 2290 * store buffer. "cpfn" is the routine called to do the actual data 2291 * transfer. bcopy is passed in to copy contiguous chunks, while 2292 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2293 * pkt is really an mbuf. 2294 */ 2295static void 2296catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2297 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2298 struct bintime *bt) 2299{ 2300 struct bpf_xhdr hdr; 2301#ifndef BURN_BRIDGES 2302 struct bpf_hdr hdr_old; 2303#ifdef COMPAT_FREEBSD32 2304 struct bpf_hdr32 hdr32_old; 2305#endif 2306#endif 2307 int caplen, curlen, hdrlen, totlen; 2308 int do_wakeup = 0; 2309 int do_timestamp; 2310 int tstype; 2311 2312 BPFD_LOCK_ASSERT(d); 2313 2314 /* 2315 * Detect whether user space has released a buffer back to us, and if 2316 * so, move it from being a hold buffer to a free buffer. This may 2317 * not be the best place to do it (for example, we might only want to 2318 * run this check if we need the space), but for now it's a reliable 2319 * spot to do it. 2320 */ 2321 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2322 d->bd_fbuf = d->bd_hbuf; 2323 d->bd_hbuf = NULL; 2324 d->bd_hlen = 0; 2325 bpf_buf_reclaimed(d); 2326 } 2327 2328 /* 2329 * Figure out how many bytes to move. If the packet is 2330 * greater or equal to the snapshot length, transfer that 2331 * much. Otherwise, transfer the whole packet (unless 2332 * we hit the buffer size limit). 2333 */ 2334 hdrlen = bpf_hdrlen(d); 2335 totlen = hdrlen + min(snaplen, pktlen); 2336 if (totlen > d->bd_bufsize) 2337 totlen = d->bd_bufsize; 2338 2339 /* 2340 * Round up the end of the previous packet to the next longword. 2341 * 2342 * Drop the packet if there's no room and no hope of room 2343 * If the packet would overflow the storage buffer or the storage 2344 * buffer is considered immutable by the buffer model, try to rotate 2345 * the buffer and wakeup pending processes. 2346 */ 2347#ifdef COMPAT_FREEBSD32 2348 if (d->bd_compat32) 2349 curlen = BPF_WORDALIGN32(d->bd_slen); 2350 else 2351#endif 2352 curlen = BPF_WORDALIGN(d->bd_slen); 2353 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2354 if (d->bd_fbuf == NULL) { 2355 /* 2356 * There's no room in the store buffer, and no 2357 * prospect of room, so drop the packet. Notify the 2358 * buffer model. 2359 */ 2360 bpf_buffull(d); 2361 ++d->bd_dcount; 2362 return; 2363 } 2364 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); 2365 ROTATE_BUFFERS(d); 2366 do_wakeup = 1; 2367 curlen = 0; 2368 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2369 /* 2370 * Immediate mode is set, or the read timeout has already 2371 * expired during a select call. A packet arrived, so the 2372 * reader should be woken up. 2373 */ 2374 do_wakeup = 1; 2375 caplen = totlen - hdrlen; 2376 tstype = d->bd_tstamp; 2377 do_timestamp = tstype != BPF_T_NONE; 2378#ifndef BURN_BRIDGES 2379 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2380 struct bpf_ts ts; 2381 if (do_timestamp) 2382 bpf_bintime2ts(bt, &ts, tstype); 2383#ifdef COMPAT_FREEBSD32 2384 if (d->bd_compat32) { 2385 bzero(&hdr32_old, sizeof(hdr32_old)); 2386 if (do_timestamp) { 2387 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2388 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2389 } 2390 hdr32_old.bh_datalen = pktlen; 2391 hdr32_old.bh_hdrlen = hdrlen; 2392 hdr32_old.bh_caplen = caplen; 2393 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2394 sizeof(hdr32_old)); 2395 goto copy; 2396 } 2397#endif 2398 bzero(&hdr_old, sizeof(hdr_old)); 2399 if (do_timestamp) { 2400 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2401 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2402 } 2403 hdr_old.bh_datalen = pktlen; 2404 hdr_old.bh_hdrlen = hdrlen; 2405 hdr_old.bh_caplen = caplen; 2406 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2407 sizeof(hdr_old)); 2408 goto copy; 2409 } 2410#endif 2411 2412 /* 2413 * Append the bpf header. Note we append the actual header size, but 2414 * move forward the length of the header plus padding. 2415 */ 2416 bzero(&hdr, sizeof(hdr)); 2417 if (do_timestamp) 2418 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2419 hdr.bh_datalen = pktlen; 2420 hdr.bh_hdrlen = hdrlen; 2421 hdr.bh_caplen = caplen; 2422 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2423 2424 /* 2425 * Copy the packet data into the store buffer and update its length. 2426 */ 2427#ifndef BURN_BRIDGES 2428copy: 2429#endif 2430 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2431 d->bd_slen = curlen + totlen; 2432 2433 if (do_wakeup) 2434 bpf_wakeup(d); 2435} 2436 2437/* 2438 * Free buffers currently in use by a descriptor. 2439 * Called on close. 2440 */ 2441static void 2442bpf_freed(struct bpf_d *d) 2443{ 2444 2445 /* 2446 * We don't need to lock out interrupts since this descriptor has 2447 * been detached from its interface and it yet hasn't been marked 2448 * free. 2449 */ 2450 bpf_free(d); 2451 if (d->bd_rfilter != NULL) { 2452 free((caddr_t)d->bd_rfilter, M_BPF); 2453#ifdef BPF_JITTER 2454 if (d->bd_bfilter != NULL) 2455 bpf_destroy_jit_filter(d->bd_bfilter); 2456#endif 2457 } 2458 if (d->bd_wfilter != NULL) 2459 free((caddr_t)d->bd_wfilter, M_BPF); 2460 mtx_destroy(&d->bd_lock); 2461} 2462 2463/* 2464 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2465 * fixed size of the link header (variable length headers not yet supported). 2466 */ 2467void 2468bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2469{ 2470 2471 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2472} 2473 2474/* 2475 * Attach an interface to bpf. ifp is a pointer to the structure 2476 * defining the interface to be attached, dlt is the link layer type, 2477 * and hdrlen is the fixed size of the link header (variable length 2478 * headers are not yet supporrted). 2479 */ 2480void 2481bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2482{ 2483 struct bpf_if *bp; 2484 2485 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); 2486 if (bp == NULL) 2487 panic("bpfattach"); 2488 2489 LIST_INIT(&bp->bif_dlist); 2490 LIST_INIT(&bp->bif_wlist); 2491 bp->bif_ifp = ifp; 2492 bp->bif_dlt = dlt; 2493 rw_init(&bp->bif_lock, "bpf interface lock"); 2494 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); 2495 *driverp = bp; 2496 2497 BPF_LOCK(); 2498 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2499 BPF_UNLOCK(); 2500 2501 bp->bif_hdrlen = hdrlen; 2502 2503 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 2504 if_printf(ifp, "bpf attached\n"); 2505} 2506 2507/* 2508 * Detach bpf from an interface. This involves detaching each descriptor 2509 * associated with the interface. Notify each descriptor as it's detached 2510 * so that any sleepers wake up and get ENXIO. 2511 */ 2512void 2513bpfdetach(struct ifnet *ifp) 2514{ 2515 struct bpf_if *bp, *bp_temp; 2516 struct bpf_d *d; 2517 int ndetached; 2518 2519 ndetached = 0; 2520 2521 BPF_LOCK(); 2522 /* Find all bpf_if struct's which reference ifp and detach them. */ 2523 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2524 if (ifp != bp->bif_ifp) 2525 continue; 2526 2527 LIST_REMOVE(bp, bif_next); 2528 /* Add to to-be-freed list */ 2529 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); 2530 2531 ndetached++; 2532 /* 2533 * Delay freeing bp till interface is detached 2534 * and all routes through this interface are removed. 2535 * Mark bp as detached to restrict new consumers. 2536 */ 2537 BPFIF_WLOCK(bp); 2538 bp->flags |= BPFIF_FLAG_DYING; 2539 BPFIF_WUNLOCK(bp); 2540 2541 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", 2542 __func__, bp->bif_dlt, bp, ifp); 2543 2544 /* Free common descriptors */ 2545 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { 2546 bpf_detachd_locked(d); 2547 BPFD_LOCK(d); 2548 bpf_wakeup(d); 2549 BPFD_UNLOCK(d); 2550 } 2551 2552 /* Free writer-only descriptors */ 2553 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { 2554 bpf_detachd_locked(d); 2555 BPFD_LOCK(d); 2556 bpf_wakeup(d); 2557 BPFD_UNLOCK(d); 2558 } 2559 } 2560 BPF_UNLOCK(); 2561 2562#ifdef INVARIANTS 2563 if (ndetached == 0) 2564 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 2565#endif 2566} 2567 2568/* 2569 * Interface departure handler. 2570 * Note departure event does not guarantee interface is going down. 2571 * Interface renaming is currently done via departure/arrival event set. 2572 * 2573 * Departure handled is called after all routes pointing to 2574 * given interface are removed and interface is in down state 2575 * restricting any packets to be sent/received. We assume it is now safe 2576 * to free data allocated by BPF. 2577 */ 2578static void 2579bpf_ifdetach(void *arg __unused, struct ifnet *ifp) 2580{ 2581 struct bpf_if *bp, *bp_temp; 2582 int nmatched = 0; 2583 2584 BPF_LOCK(); 2585 /* 2586 * Find matching entries in free list. 2587 * Nothing should be found if bpfdetach() was not called. 2588 */ 2589 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { 2590 if (ifp != bp->bif_ifp) 2591 continue; 2592 2593 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", 2594 __func__, bp, ifp); 2595 2596 LIST_REMOVE(bp, bif_next); 2597 2598 rw_destroy(&bp->bif_lock); 2599 free(bp, M_BPF); 2600 2601 nmatched++; 2602 } 2603 BPF_UNLOCK(); 2604 2605 /* 2606 * Note that we cannot zero other pointers to 2607 * custom DLTs possibly used by given interface. 2608 */ 2609 if (nmatched != 0) 2610 ifp->if_bpf = NULL; 2611} 2612 2613/* 2614 * Get a list of available data link type of the interface. 2615 */ 2616static int 2617bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2618{ 2619 struct ifnet *ifp; 2620 struct bpf_if *bp; 2621 u_int *lst; 2622 int error, n, n1; 2623 2624 BPF_LOCK_ASSERT(); 2625 2626 ifp = d->bd_bif->bif_ifp; 2627again: 2628 n1 = 0; 2629 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2630 if (bp->bif_ifp == ifp) 2631 n1++; 2632 } 2633 if (bfl->bfl_list == NULL) { 2634 bfl->bfl_len = n1; 2635 return (0); 2636 } 2637 if (n1 > bfl->bfl_len) 2638 return (ENOMEM); 2639 BPF_UNLOCK(); 2640 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); 2641 n = 0; 2642 BPF_LOCK(); 2643 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2644 if (bp->bif_ifp != ifp) 2645 continue; 2646 if (n >= n1) { 2647 free(lst, M_TEMP); 2648 goto again; 2649 } 2650 lst[n] = bp->bif_dlt; 2651 n++; 2652 } 2653 BPF_UNLOCK(); 2654 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); 2655 free(lst, M_TEMP); 2656 BPF_LOCK(); 2657 bfl->bfl_len = n; 2658 return (error); 2659} 2660 2661/* 2662 * Set the data link type of a BPF instance. 2663 */ 2664static int 2665bpf_setdlt(struct bpf_d *d, u_int dlt) 2666{ 2667 int error, opromisc; 2668 struct ifnet *ifp; 2669 struct bpf_if *bp; 2670 2671 BPF_LOCK_ASSERT(); 2672 2673 if (d->bd_bif->bif_dlt == dlt) 2674 return (0); 2675 ifp = d->bd_bif->bif_ifp; 2676 2677 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2678 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2679 break; 2680 } 2681 2682 if (bp != NULL) { 2683 opromisc = d->bd_promisc; 2684 bpf_attachd(d, bp); 2685 BPFD_LOCK(d); 2686 reset_d(d); 2687 BPFD_UNLOCK(d); 2688 if (opromisc) { 2689 error = ifpromisc(bp->bif_ifp, 1); 2690 if (error) 2691 if_printf(bp->bif_ifp, 2692 "bpf_setdlt: ifpromisc failed (%d)\n", 2693 error); 2694 else 2695 d->bd_promisc = 1; 2696 } 2697 } 2698 return (bp == NULL ? EINVAL : 0); 2699} 2700 2701static void 2702bpf_drvinit(void *unused) 2703{ 2704 struct cdev *dev; 2705 2706 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF); 2707 LIST_INIT(&bpf_iflist); 2708 LIST_INIT(&bpf_freelist); 2709 2710 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2711 /* For compatibility */ 2712 make_dev_alias(dev, "bpf0"); 2713 2714 /* Register interface departure handler */ 2715 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( 2716 ifnet_departure_event, bpf_ifdetach, NULL, 2717 EVENTHANDLER_PRI_ANY); 2718} 2719 2720/* 2721 * Zero out the various packet counters associated with all of the bpf 2722 * descriptors. At some point, we will probably want to get a bit more 2723 * granular and allow the user to specify descriptors to be zeroed. 2724 */ 2725static void 2726bpf_zero_counters(void) 2727{ 2728 struct bpf_if *bp; 2729 struct bpf_d *bd; 2730 2731 BPF_LOCK(); 2732 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2733 BPFIF_RLOCK(bp); 2734 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2735 BPFD_LOCK(bd); 2736 bd->bd_rcount = 0; 2737 bd->bd_dcount = 0; 2738 bd->bd_fcount = 0; 2739 bd->bd_wcount = 0; 2740 bd->bd_wfcount = 0; 2741 bd->bd_zcopy = 0; 2742 BPFD_UNLOCK(bd); 2743 } 2744 BPFIF_RUNLOCK(bp); 2745 } 2746 BPF_UNLOCK(); 2747} 2748 2749/* 2750 * Fill filter statistics 2751 */ 2752static void 2753bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2754{ 2755 2756 bzero(d, sizeof(*d)); 2757 BPFD_LOCK_ASSERT(bd); 2758 d->bd_structsize = sizeof(*d); 2759 /* XXX: reading should be protected by global lock */ 2760 d->bd_immediate = bd->bd_immediate; 2761 d->bd_promisc = bd->bd_promisc; 2762 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2763 d->bd_direction = bd->bd_direction; 2764 d->bd_feedback = bd->bd_feedback; 2765 d->bd_async = bd->bd_async; 2766 d->bd_rcount = bd->bd_rcount; 2767 d->bd_dcount = bd->bd_dcount; 2768 d->bd_fcount = bd->bd_fcount; 2769 d->bd_sig = bd->bd_sig; 2770 d->bd_slen = bd->bd_slen; 2771 d->bd_hlen = bd->bd_hlen; 2772 d->bd_bufsize = bd->bd_bufsize; 2773 d->bd_pid = bd->bd_pid; 2774 strlcpy(d->bd_ifname, 2775 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2776 d->bd_locked = bd->bd_locked; 2777 d->bd_wcount = bd->bd_wcount; 2778 d->bd_wdcount = bd->bd_wdcount; 2779 d->bd_wfcount = bd->bd_wfcount; 2780 d->bd_zcopy = bd->bd_zcopy; 2781 d->bd_bufmode = bd->bd_bufmode; 2782} 2783 2784/* 2785 * Handle `netstat -B' stats request 2786 */ 2787static int 2788bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2789{ 2790 static const struct xbpf_d zerostats; 2791 struct xbpf_d *xbdbuf, *xbd, tempstats; 2792 int index, error; 2793 struct bpf_if *bp; 2794 struct bpf_d *bd; 2795 2796 /* 2797 * XXX This is not technically correct. It is possible for non 2798 * privileged users to open bpf devices. It would make sense 2799 * if the users who opened the devices were able to retrieve 2800 * the statistics for them, too. 2801 */ 2802 error = priv_check(req->td, PRIV_NET_BPF); 2803 if (error) 2804 return (error); 2805 /* 2806 * Check to see if the user is requesting that the counters be 2807 * zeroed out. Explicitly check that the supplied data is zeroed, 2808 * as we aren't allowing the user to set the counters currently. 2809 */ 2810 if (req->newptr != NULL) { 2811 if (req->newlen != sizeof(tempstats)) 2812 return (EINVAL); 2813 memset(&tempstats, 0, sizeof(tempstats)); 2814 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); 2815 if (error) 2816 return (error); 2817 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) 2818 return (EINVAL); 2819 bpf_zero_counters(); 2820 return (0); 2821 } 2822 if (req->oldptr == NULL) 2823 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2824 if (bpf_bpfd_cnt == 0) 2825 return (SYSCTL_OUT(req, 0, 0)); 2826 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2827 BPF_LOCK(); 2828 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2829 BPF_UNLOCK(); 2830 free(xbdbuf, M_BPF); 2831 return (ENOMEM); 2832 } 2833 index = 0; 2834 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2835 BPFIF_RLOCK(bp); 2836 /* Send writers-only first */ 2837 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2838 xbd = &xbdbuf[index++]; 2839 BPFD_LOCK(bd); 2840 bpfstats_fill_xbpf(xbd, bd); 2841 BPFD_UNLOCK(bd); 2842 } 2843 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2844 xbd = &xbdbuf[index++]; 2845 BPFD_LOCK(bd); 2846 bpfstats_fill_xbpf(xbd, bd); 2847 BPFD_UNLOCK(bd); 2848 } 2849 BPFIF_RUNLOCK(bp); 2850 } 2851 BPF_UNLOCK(); 2852 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2853 free(xbdbuf, M_BPF); 2854 return (error); 2855} 2856 2857SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2858 2859#else /* !DEV_BPF && !NETGRAPH_BPF */ 2860/* 2861 * NOP stubs to allow bpf-using drivers to load and function. 2862 * 2863 * A 'better' implementation would allow the core bpf functionality 2864 * to be loaded at runtime. 2865 */ 2866static struct bpf_if bp_null; 2867 2868void 2869bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2870{ 2871} 2872 2873void 2874bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2875{ 2876} 2877 2878void 2879bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 2880{ 2881} 2882 2883void 2884bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2885{ 2886 2887 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2888} 2889 2890void 2891bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2892{ 2893 2894 *driverp = &bp_null; 2895} 2896 2897void 2898bpfdetach(struct ifnet *ifp) 2899{ 2900} 2901 2902u_int 2903bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 2904{ 2905 return -1; /* "no filter" behaviour */ 2906} 2907 2908int 2909bpf_validate(const struct bpf_insn *f, int len) 2910{ 2911 return 0; /* false */ 2912} 2913 2914#endif /* !DEV_BPF && !NETGRAPH_BPF */ 2915