1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1990, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from the Stanford/CMU enet packet filter, 8 * (net/enet.c) distributed as part of 4.3BSD, and code contributed 9 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence 10 * Berkeley Laboratory. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)bpf.c 8.4 (Berkeley) 1/9/95 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 41 42 #include "opt_bpf.h" 43 #include "opt_compat.h" 44 #include "opt_ddb.h" 45 #include "opt_netgraph.h" 46 47 #include <sys/types.h> 48 #include <sys/param.h> 49 #include <sys/lock.h> 50 #include <sys/rwlock.h> 51 #include <sys/systm.h> 52 #include <sys/conf.h> 53 #include <sys/fcntl.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/time.h> 58 #include <sys/priv.h> 59 #include <sys/proc.h> 60 #include <sys/signalvar.h> 61 #include <sys/filio.h> 62 #include <sys/sockio.h> 63 #include <sys/ttycom.h> 64 #include <sys/uio.h> 65 #include <sys/sysent.h> 66 67 #include <sys/event.h> 68 #include <sys/file.h> 69 #include <sys/poll.h> 70 #include <sys/proc.h> 71 72 #include <sys/socket.h> 73 74 #ifdef DDB 75 #include <ddb/ddb.h> 76 #endif 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/if_dl.h> 81 #include <net/bpf.h> 82 #include <net/bpf_buffer.h> 83 #ifdef BPF_JITTER 84 #include <net/bpf_jitter.h> 85 #endif 86 #include <net/bpf_zerocopy.h> 87 #include <net/bpfdesc.h> 88 #include <net/route.h> 89 #include <net/vnet.h> 90 91 #include <netinet/in.h> 92 #include <netinet/if_ether.h> 93 #include <sys/kernel.h> 94 #include <sys/sysctl.h> 95 96 #include <net80211/ieee80211_freebsd.h> 97 98 #include <security/mac/mac_framework.h> 99 100 MALLOC_DEFINE(M_BPF, "BPF", "BPF data"); 101 102 struct bpf_if { 103 #define bif_next bif_ext.bif_next 104 #define bif_dlist bif_ext.bif_dlist 105 struct bpf_if_ext bif_ext; /* public members */ 106 u_int bif_dlt; /* link layer type */ 107 u_int bif_hdrlen; /* length of link header */ 108 struct ifnet *bif_ifp; /* corresponding interface */ 109 struct rwlock bif_lock; /* interface lock */ 110 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */ 111 int bif_flags; /* Interface flags */ 112 struct bpf_if **bif_bpf; /* Pointer to pointer to us */ 113 }; 114 115 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0); 116 117 #if defined(DEV_BPF) || defined(NETGRAPH_BPF) 118 119 #define PRINET 26 /* interruptible */ 120 121 #define SIZEOF_BPF_HDR(type) \ 122 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) 123 124 #ifdef COMPAT_FREEBSD32 125 #include <sys/mount.h> 126 #include <compat/freebsd32/freebsd32.h> 127 #define BPF_ALIGNMENT32 sizeof(int32_t) 128 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32) 129 130 #ifndef BURN_BRIDGES 131 /* 132 * 32-bit version of structure prepended to each packet. We use this header 133 * instead of the standard one for 32-bit streams. We mark the a stream as 134 * 32-bit the first time we see a 32-bit compat ioctl request. 135 */ 136 struct bpf_hdr32 { 137 struct timeval32 bh_tstamp; /* time stamp */ 138 uint32_t bh_caplen; /* length of captured portion */ 139 uint32_t bh_datalen; /* original length of packet */ 140 uint16_t bh_hdrlen; /* length of bpf header (this struct 141 plus alignment padding) */ 142 }; 143 #endif 144 145 struct bpf_program32 { 146 u_int bf_len; 147 uint32_t bf_insns; 148 }; 149 150 struct bpf_dltlist32 { 151 u_int bfl_len; 152 u_int bfl_list; 153 }; 154 155 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32) 156 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) 157 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) 158 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) 159 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) 160 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) 161 #endif 162 163 /* 164 * bpf_iflist is a list of BPF interface structures, each corresponding to a 165 * specific DLT. The same network interface might have several BPF interface 166 * structures registered by different layers in the stack (i.e., 802.11 167 * frames, ethernet frames, etc). 168 */ 169 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist; 170 static struct mtx bpf_mtx; /* bpf global lock */ 171 static int bpf_bpfd_cnt; 172 173 static void bpf_attachd(struct bpf_d *, struct bpf_if *); 174 static void bpf_detachd(struct bpf_d *); 175 static void bpf_detachd_locked(struct bpf_d *); 176 static void bpf_freed(struct bpf_d *); 177 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, 178 struct sockaddr *, int *, struct bpf_d *); 179 static int bpf_setif(struct bpf_d *, struct ifreq *); 180 static void bpf_timed_out(void *); 181 static __inline void 182 bpf_wakeup(struct bpf_d *); 183 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 184 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), 185 struct bintime *); 186 static void reset_d(struct bpf_d *); 187 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); 188 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 189 static int bpf_setdlt(struct bpf_d *, u_int); 190 static void filt_bpfdetach(struct knote *); 191 static int filt_bpfread(struct knote *, long); 192 static void bpf_drvinit(void *); 193 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); 194 195 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); 196 int bpf_maxinsns = BPF_MAXINSNS; 197 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, 198 &bpf_maxinsns, 0, "Maximum bpf program instructions"); 199 static int bpf_zerocopy_enable = 0; 200 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, 201 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); 202 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, 203 bpf_stats_sysctl, "bpf statistics portal"); 204 205 static VNET_DEFINE(int, bpf_optimize_writers) = 0; 206 #define V_bpf_optimize_writers VNET(bpf_optimize_writers) 207 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(bpf_optimize_writers), 0, 209 "Do not send packets until BPF program is set"); 210 211 static d_open_t bpfopen; 212 static d_read_t bpfread; 213 static d_write_t bpfwrite; 214 static d_ioctl_t bpfioctl; 215 static d_poll_t bpfpoll; 216 static d_kqfilter_t bpfkqfilter; 217 218 static struct cdevsw bpf_cdevsw = { 219 .d_version = D_VERSION, 220 .d_open = bpfopen, 221 .d_read = bpfread, 222 .d_write = bpfwrite, 223 .d_ioctl = bpfioctl, 224 .d_poll = bpfpoll, 225 .d_name = "bpf", 226 .d_kqfilter = bpfkqfilter, 227 }; 228 229 static struct filterops bpfread_filtops = { 230 .f_isfd = 1, 231 .f_detach = filt_bpfdetach, 232 .f_event = filt_bpfread, 233 }; 234 235 eventhandler_tag bpf_ifdetach_cookie = NULL; 236 237 /* 238 * LOCKING MODEL USED BY BPF: 239 * Locks: 240 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal, 241 * some global counters and every bpf_if reference. 242 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters. 243 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields 244 * used by bpf_mtap code. 245 * 246 * Lock order: 247 * 248 * Global lock, interface lock, descriptor lock 249 * 250 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2] 251 * working model. In many places (like bpf_detachd) we start with BPF descriptor 252 * (and we need to at least rlock it to get reliable interface pointer). This 253 * gives us potential LOR. As a result, we use global lock to protect from bpf_if 254 * change in every such place. 255 * 256 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and 257 * 3) descriptor main wlock. 258 * Reading bd_bif can be protected by any of these locks, typically global lock. 259 * 260 * Changing read/write BPF filter is protected by the same three locks, 261 * the same applies for reading. 262 * 263 * Sleeping in global lock is not allowed due to bpfdetach() using it. 264 */ 265 266 /* 267 * Wrapper functions for various buffering methods. If the set of buffer 268 * modes expands, we will probably want to introduce a switch data structure 269 * similar to protosw, et. 270 */ 271 static void 272 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 273 u_int len) 274 { 275 276 BPFD_LOCK_ASSERT(d); 277 278 switch (d->bd_bufmode) { 279 case BPF_BUFMODE_BUFFER: 280 return (bpf_buffer_append_bytes(d, buf, offset, src, len)); 281 282 case BPF_BUFMODE_ZBUF: 283 d->bd_zcopy++; 284 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); 285 286 default: 287 panic("bpf_buf_append_bytes"); 288 } 289 } 290 291 static void 292 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 293 u_int len) 294 { 295 296 BPFD_LOCK_ASSERT(d); 297 298 switch (d->bd_bufmode) { 299 case BPF_BUFMODE_BUFFER: 300 return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); 301 302 case BPF_BUFMODE_ZBUF: 303 d->bd_zcopy++; 304 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); 305 306 default: 307 panic("bpf_buf_append_mbuf"); 308 } 309 } 310 311 /* 312 * This function gets called when the free buffer is re-assigned. 313 */ 314 static void 315 bpf_buf_reclaimed(struct bpf_d *d) 316 { 317 318 BPFD_LOCK_ASSERT(d); 319 320 switch (d->bd_bufmode) { 321 case BPF_BUFMODE_BUFFER: 322 return; 323 324 case BPF_BUFMODE_ZBUF: 325 bpf_zerocopy_buf_reclaimed(d); 326 return; 327 328 default: 329 panic("bpf_buf_reclaimed"); 330 } 331 } 332 333 /* 334 * If the buffer mechanism has a way to decide that a held buffer can be made 335 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is 336 * returned if the buffer can be discarded, (0) is returned if it cannot. 337 */ 338 static int 339 bpf_canfreebuf(struct bpf_d *d) 340 { 341 342 BPFD_LOCK_ASSERT(d); 343 344 switch (d->bd_bufmode) { 345 case BPF_BUFMODE_ZBUF: 346 return (bpf_zerocopy_canfreebuf(d)); 347 } 348 return (0); 349 } 350 351 /* 352 * Allow the buffer model to indicate that the current store buffer is 353 * immutable, regardless of the appearance of space. Return (1) if the 354 * buffer is writable, and (0) if not. 355 */ 356 static int 357 bpf_canwritebuf(struct bpf_d *d) 358 { 359 BPFD_LOCK_ASSERT(d); 360 361 switch (d->bd_bufmode) { 362 case BPF_BUFMODE_ZBUF: 363 return (bpf_zerocopy_canwritebuf(d)); 364 } 365 return (1); 366 } 367 368 /* 369 * Notify buffer model that an attempt to write to the store buffer has 370 * resulted in a dropped packet, in which case the buffer may be considered 371 * full. 372 */ 373 static void 374 bpf_buffull(struct bpf_d *d) 375 { 376 377 BPFD_LOCK_ASSERT(d); 378 379 switch (d->bd_bufmode) { 380 case BPF_BUFMODE_ZBUF: 381 bpf_zerocopy_buffull(d); 382 break; 383 } 384 } 385 386 /* 387 * Notify the buffer model that a buffer has moved into the hold position. 388 */ 389 void 390 bpf_bufheld(struct bpf_d *d) 391 { 392 393 BPFD_LOCK_ASSERT(d); 394 395 switch (d->bd_bufmode) { 396 case BPF_BUFMODE_ZBUF: 397 bpf_zerocopy_bufheld(d); 398 break; 399 } 400 } 401 402 static void 403 bpf_free(struct bpf_d *d) 404 { 405 406 switch (d->bd_bufmode) { 407 case BPF_BUFMODE_BUFFER: 408 return (bpf_buffer_free(d)); 409 410 case BPF_BUFMODE_ZBUF: 411 return (bpf_zerocopy_free(d)); 412 413 default: 414 panic("bpf_buf_free"); 415 } 416 } 417 418 static int 419 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) 420 { 421 422 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 423 return (EOPNOTSUPP); 424 return (bpf_buffer_uiomove(d, buf, len, uio)); 425 } 426 427 static int 428 bpf_ioctl_sblen(struct bpf_d *d, u_int *i) 429 { 430 431 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 432 return (EOPNOTSUPP); 433 return (bpf_buffer_ioctl_sblen(d, i)); 434 } 435 436 static int 437 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) 438 { 439 440 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 441 return (EOPNOTSUPP); 442 return (bpf_zerocopy_ioctl_getzmax(td, d, i)); 443 } 444 445 static int 446 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 447 { 448 449 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 450 return (EOPNOTSUPP); 451 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); 452 } 453 454 static int 455 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 456 { 457 458 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 459 return (EOPNOTSUPP); 460 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); 461 } 462 463 /* 464 * General BPF functions. 465 */ 466 static int 467 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, 468 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d) 469 { 470 const struct ieee80211_bpf_params *p; 471 struct ether_header *eh; 472 struct mbuf *m; 473 int error; 474 int len; 475 int hlen; 476 int slen; 477 478 /* 479 * Build a sockaddr based on the data link layer type. 480 * We do this at this level because the ethernet header 481 * is copied directly into the data field of the sockaddr. 482 * In the case of SLIP, there is no header and the packet 483 * is forwarded as is. 484 * Also, we are careful to leave room at the front of the mbuf 485 * for the link level header. 486 */ 487 switch (linktype) { 488 489 case DLT_SLIP: 490 sockp->sa_family = AF_INET; 491 hlen = 0; 492 break; 493 494 case DLT_EN10MB: 495 sockp->sa_family = AF_UNSPEC; 496 /* XXX Would MAXLINKHDR be better? */ 497 hlen = ETHER_HDR_LEN; 498 break; 499 500 case DLT_FDDI: 501 sockp->sa_family = AF_IMPLINK; 502 hlen = 0; 503 break; 504 505 case DLT_RAW: 506 sockp->sa_family = AF_UNSPEC; 507 hlen = 0; 508 break; 509 510 case DLT_NULL: 511 /* 512 * null interface types require a 4 byte pseudo header which 513 * corresponds to the address family of the packet. 514 */ 515 sockp->sa_family = AF_UNSPEC; 516 hlen = 4; 517 break; 518 519 case DLT_ATM_RFC1483: 520 /* 521 * en atm driver requires 4-byte atm pseudo header. 522 * though it isn't standard, vpi:vci needs to be 523 * specified anyway. 524 */ 525 sockp->sa_family = AF_UNSPEC; 526 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ 527 break; 528 529 case DLT_PPP: 530 sockp->sa_family = AF_UNSPEC; 531 hlen = 4; /* This should match PPP_HDRLEN */ 532 break; 533 534 case DLT_IEEE802_11: /* IEEE 802.11 wireless */ 535 sockp->sa_family = AF_IEEE80211; 536 hlen = 0; 537 break; 538 539 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ 540 sockp->sa_family = AF_IEEE80211; 541 sockp->sa_len = 12; /* XXX != 0 */ 542 hlen = sizeof(struct ieee80211_bpf_params); 543 break; 544 545 default: 546 return (EIO); 547 } 548 549 len = uio->uio_resid; 550 if (len < hlen || len - hlen > ifp->if_mtu) 551 return (EMSGSIZE); 552 553 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); 554 if (m == NULL) 555 return (EIO); 556 m->m_pkthdr.len = m->m_len = len; 557 *mp = m; 558 559 error = uiomove(mtod(m, u_char *), len, uio); 560 if (error) 561 goto bad; 562 563 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len); 564 if (slen == 0) { 565 error = EPERM; 566 goto bad; 567 } 568 569 /* Check for multicast destination */ 570 switch (linktype) { 571 case DLT_EN10MB: 572 eh = mtod(m, struct ether_header *); 573 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 574 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 575 ETHER_ADDR_LEN) == 0) 576 m->m_flags |= M_BCAST; 577 else 578 m->m_flags |= M_MCAST; 579 } 580 if (d->bd_hdrcmplt == 0) { 581 memcpy(eh->ether_shost, IF_LLADDR(ifp), 582 sizeof(eh->ether_shost)); 583 } 584 break; 585 } 586 587 /* 588 * Make room for link header, and copy it to sockaddr 589 */ 590 if (hlen != 0) { 591 if (sockp->sa_family == AF_IEEE80211) { 592 /* 593 * Collect true length from the parameter header 594 * NB: sockp is known to be zero'd so if we do a 595 * short copy unspecified parameters will be 596 * zero. 597 * NB: packet may not be aligned after stripping 598 * bpf params 599 * XXX check ibp_vers 600 */ 601 p = mtod(m, const struct ieee80211_bpf_params *); 602 hlen = p->ibp_len; 603 if (hlen > sizeof(sockp->sa_data)) { 604 error = EINVAL; 605 goto bad; 606 } 607 } 608 bcopy(mtod(m, const void *), sockp->sa_data, hlen); 609 } 610 *hdrlen = hlen; 611 612 return (0); 613 bad: 614 m_freem(m); 615 return (error); 616 } 617 618 /* 619 * Attach file to the bpf interface, i.e. make d listen on bp. 620 */ 621 static void 622 bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 623 { 624 int op_w; 625 626 BPF_LOCK_ASSERT(); 627 628 /* 629 * Save sysctl value to protect from sysctl change 630 * between reads 631 */ 632 op_w = V_bpf_optimize_writers || d->bd_writer; 633 634 if (d->bd_bif != NULL) 635 bpf_detachd_locked(d); 636 /* 637 * Point d at bp, and add d to the interface's list. 638 * Since there are many applications using BPF for 639 * sending raw packets only (dhcpd, cdpd are good examples) 640 * we can delay adding d to the list of active listeners until 641 * some filter is configured. 642 */ 643 644 BPFIF_WLOCK(bp); 645 BPFD_LOCK(d); 646 647 d->bd_bif = bp; 648 649 if (op_w != 0) { 650 /* Add to writers-only list */ 651 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); 652 /* 653 * We decrement bd_writer on every filter set operation. 654 * First BIOCSETF is done by pcap_open_live() to set up 655 * snap length. After that appliation usually sets its own filter 656 */ 657 d->bd_writer = 2; 658 } else 659 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 660 661 BPFD_UNLOCK(d); 662 BPFIF_WUNLOCK(bp); 663 664 bpf_bpfd_cnt++; 665 666 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", 667 __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); 668 669 if (op_w == 0) 670 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 671 } 672 673 /* 674 * Check if we need to upgrade our descriptor @d from write-only mode. 675 */ 676 static int 677 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen) 678 { 679 int is_snap, need_upgrade; 680 681 /* 682 * Check if we've already upgraded or new filter is empty. 683 */ 684 if (d->bd_writer == 0 || fcode == NULL) 685 return (0); 686 687 need_upgrade = 0; 688 689 /* 690 * Check if cmd looks like snaplen setting from 691 * pcap_bpf.c:pcap_open_live(). 692 * Note we're not checking .k value here: 693 * while pcap_open_live() definitely sets to non-zero value, 694 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g. 695 * do not consider upgrading immediately 696 */ 697 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K)) 698 is_snap = 1; 699 else 700 is_snap = 0; 701 702 if (is_snap == 0) { 703 /* 704 * We're setting first filter and it doesn't look like 705 * setting snaplen. We're probably using bpf directly. 706 * Upgrade immediately. 707 */ 708 need_upgrade = 1; 709 } else { 710 /* 711 * Do not require upgrade by first BIOCSETF 712 * (used to set snaplen) by pcap_open_live(). 713 */ 714 715 if (--d->bd_writer == 0) { 716 /* 717 * First snaplen filter has already 718 * been set. This is probably catch-all 719 * filter 720 */ 721 need_upgrade = 1; 722 } 723 } 724 725 CTR5(KTR_NET, 726 "%s: filter function set by pid %d, " 727 "bd_writer counter %d, snap %d upgrade %d", 728 __func__, d->bd_pid, d->bd_writer, 729 is_snap, need_upgrade); 730 731 return (need_upgrade); 732 } 733 734 /* 735 * Add d to the list of active bp filters. 736 * Requires bpf_attachd() to be called before. 737 */ 738 static void 739 bpf_upgraded(struct bpf_d *d) 740 { 741 struct bpf_if *bp; 742 743 BPF_LOCK_ASSERT(); 744 745 bp = d->bd_bif; 746 747 /* 748 * Filter can be set several times without specifying interface. 749 * Mark d as reader and exit. 750 */ 751 if (bp == NULL) { 752 BPFD_LOCK(d); 753 d->bd_writer = 0; 754 BPFD_UNLOCK(d); 755 return; 756 } 757 758 BPFIF_WLOCK(bp); 759 BPFD_LOCK(d); 760 761 /* Remove from writers-only list */ 762 LIST_REMOVE(d, bd_next); 763 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 764 /* Mark d as reader */ 765 d->bd_writer = 0; 766 767 BPFD_UNLOCK(d); 768 BPFIF_WUNLOCK(bp); 769 770 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); 771 772 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 773 } 774 775 /* 776 * Detach a file from its interface. 777 */ 778 static void 779 bpf_detachd(struct bpf_d *d) 780 { 781 BPF_LOCK(); 782 bpf_detachd_locked(d); 783 BPF_UNLOCK(); 784 } 785 786 static void 787 bpf_detachd_locked(struct bpf_d *d) 788 { 789 int error; 790 struct bpf_if *bp; 791 struct ifnet *ifp; 792 793 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); 794 795 BPF_LOCK_ASSERT(); 796 797 /* Check if descriptor is attached */ 798 if ((bp = d->bd_bif) == NULL) 799 return; 800 801 BPFIF_WLOCK(bp); 802 BPFD_LOCK(d); 803 804 /* Save bd_writer value */ 805 error = d->bd_writer; 806 807 /* 808 * Remove d from the interface's descriptor list. 809 */ 810 LIST_REMOVE(d, bd_next); 811 812 ifp = bp->bif_ifp; 813 d->bd_bif = NULL; 814 BPFD_UNLOCK(d); 815 BPFIF_WUNLOCK(bp); 816 817 bpf_bpfd_cnt--; 818 819 /* Call event handler iff d is attached */ 820 if (error == 0) 821 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); 822 823 /* 824 * Check if this descriptor had requested promiscuous mode. 825 * If so, turn it off. 826 */ 827 if (d->bd_promisc) { 828 d->bd_promisc = 0; 829 CURVNET_SET(ifp->if_vnet); 830 error = ifpromisc(ifp, 0); 831 CURVNET_RESTORE(); 832 if (error != 0 && error != ENXIO) { 833 /* 834 * ENXIO can happen if a pccard is unplugged 835 * Something is really wrong if we were able to put 836 * the driver into promiscuous mode, but can't 837 * take it out. 838 */ 839 if_printf(bp->bif_ifp, 840 "bpf_detach: ifpromisc failed (%d)\n", error); 841 } 842 } 843 } 844 845 /* 846 * Close the descriptor by detaching it from its interface, 847 * deallocating its buffers, and marking it free. 848 */ 849 static void 850 bpf_dtor(void *data) 851 { 852 struct bpf_d *d = data; 853 854 BPFD_LOCK(d); 855 if (d->bd_state == BPF_WAITING) 856 callout_stop(&d->bd_callout); 857 d->bd_state = BPF_IDLE; 858 BPFD_UNLOCK(d); 859 funsetown(&d->bd_sigio); 860 bpf_detachd(d); 861 #ifdef MAC 862 mac_bpfdesc_destroy(d); 863 #endif /* MAC */ 864 seldrain(&d->bd_sel); 865 knlist_destroy(&d->bd_sel.si_note); 866 callout_drain(&d->bd_callout); 867 bpf_freed(d); 868 free(d, M_BPF); 869 } 870 871 /* 872 * Open ethernet device. Returns ENXIO for illegal minor device number, 873 * EBUSY if file is open by another process. 874 */ 875 /* ARGSUSED */ 876 static int 877 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) 878 { 879 struct bpf_d *d; 880 int error; 881 882 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); 883 error = devfs_set_cdevpriv(d, bpf_dtor); 884 if (error != 0) { 885 free(d, M_BPF); 886 return (error); 887 } 888 889 /* 890 * For historical reasons, perform a one-time initialization call to 891 * the buffer routines, even though we're not yet committed to a 892 * particular buffer method. 893 */ 894 bpf_buffer_init(d); 895 if ((flags & FREAD) == 0) 896 d->bd_writer = 2; 897 d->bd_hbuf_in_use = 0; 898 d->bd_bufmode = BPF_BUFMODE_BUFFER; 899 d->bd_sig = SIGIO; 900 d->bd_direction = BPF_D_INOUT; 901 BPF_PID_REFRESH(d, td); 902 #ifdef MAC 903 mac_bpfdesc_init(d); 904 mac_bpfdesc_create(td->td_ucred, d); 905 #endif 906 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 907 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 908 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 909 910 return (0); 911 } 912 913 /* 914 * bpfread - read next chunk of packets from buffers 915 */ 916 static int 917 bpfread(struct cdev *dev, struct uio *uio, int ioflag) 918 { 919 struct bpf_d *d; 920 int error; 921 int non_block; 922 int timed_out; 923 924 error = devfs_get_cdevpriv((void **)&d); 925 if (error != 0) 926 return (error); 927 928 /* 929 * Restrict application to use a buffer the same size as 930 * as kernel buffers. 931 */ 932 if (uio->uio_resid != d->bd_bufsize) 933 return (EINVAL); 934 935 non_block = ((ioflag & O_NONBLOCK) != 0); 936 937 BPFD_LOCK(d); 938 BPF_PID_REFRESH_CUR(d); 939 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 940 BPFD_UNLOCK(d); 941 return (EOPNOTSUPP); 942 } 943 if (d->bd_state == BPF_WAITING) 944 callout_stop(&d->bd_callout); 945 timed_out = (d->bd_state == BPF_TIMED_OUT); 946 d->bd_state = BPF_IDLE; 947 while (d->bd_hbuf_in_use) { 948 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 949 PRINET|PCATCH, "bd_hbuf", 0); 950 if (error != 0) { 951 BPFD_UNLOCK(d); 952 return (error); 953 } 954 } 955 /* 956 * If the hold buffer is empty, then do a timed sleep, which 957 * ends when the timeout expires or when enough packets 958 * have arrived to fill the store buffer. 959 */ 960 while (d->bd_hbuf == NULL) { 961 if (d->bd_slen != 0) { 962 /* 963 * A packet(s) either arrived since the previous 964 * read or arrived while we were asleep. 965 */ 966 if (d->bd_immediate || non_block || timed_out) { 967 /* 968 * Rotate the buffers and return what's here 969 * if we are in immediate mode, non-blocking 970 * flag is set, or this descriptor timed out. 971 */ 972 ROTATE_BUFFERS(d); 973 break; 974 } 975 } 976 977 /* 978 * No data is available, check to see if the bpf device 979 * is still pointed at a real interface. If not, return 980 * ENXIO so that the userland process knows to rebind 981 * it before using it again. 982 */ 983 if (d->bd_bif == NULL) { 984 BPFD_UNLOCK(d); 985 return (ENXIO); 986 } 987 988 if (non_block) { 989 BPFD_UNLOCK(d); 990 return (EWOULDBLOCK); 991 } 992 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 993 "bpf", d->bd_rtout); 994 if (error == EINTR || error == ERESTART) { 995 BPFD_UNLOCK(d); 996 return (error); 997 } 998 if (error == EWOULDBLOCK) { 999 /* 1000 * On a timeout, return what's in the buffer, 1001 * which may be nothing. If there is something 1002 * in the store buffer, we can rotate the buffers. 1003 */ 1004 if (d->bd_hbuf) 1005 /* 1006 * We filled up the buffer in between 1007 * getting the timeout and arriving 1008 * here, so we don't need to rotate. 1009 */ 1010 break; 1011 1012 if (d->bd_slen == 0) { 1013 BPFD_UNLOCK(d); 1014 return (0); 1015 } 1016 ROTATE_BUFFERS(d); 1017 break; 1018 } 1019 } 1020 /* 1021 * At this point, we know we have something in the hold slot. 1022 */ 1023 d->bd_hbuf_in_use = 1; 1024 BPFD_UNLOCK(d); 1025 1026 /* 1027 * Move data from hold buffer into user space. 1028 * We know the entire buffer is transferred since 1029 * we checked above that the read buffer is bpf_bufsize bytes. 1030 * 1031 * We do not have to worry about simultaneous reads because 1032 * we waited for sole access to the hold buffer above. 1033 */ 1034 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 1035 1036 BPFD_LOCK(d); 1037 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 1038 d->bd_fbuf = d->bd_hbuf; 1039 d->bd_hbuf = NULL; 1040 d->bd_hlen = 0; 1041 bpf_buf_reclaimed(d); 1042 d->bd_hbuf_in_use = 0; 1043 wakeup(&d->bd_hbuf_in_use); 1044 BPFD_UNLOCK(d); 1045 1046 return (error); 1047 } 1048 1049 /* 1050 * If there are processes sleeping on this descriptor, wake them up. 1051 */ 1052 static __inline void 1053 bpf_wakeup(struct bpf_d *d) 1054 { 1055 1056 BPFD_LOCK_ASSERT(d); 1057 if (d->bd_state == BPF_WAITING) { 1058 callout_stop(&d->bd_callout); 1059 d->bd_state = BPF_IDLE; 1060 } 1061 wakeup(d); 1062 if (d->bd_async && d->bd_sig && d->bd_sigio) 1063 pgsigio(&d->bd_sigio, d->bd_sig, 0); 1064 1065 selwakeuppri(&d->bd_sel, PRINET); 1066 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 1067 } 1068 1069 static void 1070 bpf_timed_out(void *arg) 1071 { 1072 struct bpf_d *d = (struct bpf_d *)arg; 1073 1074 BPFD_LOCK_ASSERT(d); 1075 1076 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) 1077 return; 1078 if (d->bd_state == BPF_WAITING) { 1079 d->bd_state = BPF_TIMED_OUT; 1080 if (d->bd_slen != 0) 1081 bpf_wakeup(d); 1082 } 1083 } 1084 1085 static int 1086 bpf_ready(struct bpf_d *d) 1087 { 1088 1089 BPFD_LOCK_ASSERT(d); 1090 1091 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1092 return (1); 1093 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1094 d->bd_slen != 0) 1095 return (1); 1096 return (0); 1097 } 1098 1099 static int 1100 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1101 { 1102 struct bpf_d *d; 1103 struct ifnet *ifp; 1104 struct mbuf *m, *mc; 1105 struct sockaddr dst; 1106 struct route ro; 1107 int error, hlen; 1108 1109 error = devfs_get_cdevpriv((void **)&d); 1110 if (error != 0) 1111 return (error); 1112 1113 BPF_PID_REFRESH_CUR(d); 1114 d->bd_wcount++; 1115 /* XXX: locking required */ 1116 if (d->bd_bif == NULL) { 1117 d->bd_wdcount++; 1118 return (ENXIO); 1119 } 1120 1121 ifp = d->bd_bif->bif_ifp; 1122 1123 if ((ifp->if_flags & IFF_UP) == 0) { 1124 d->bd_wdcount++; 1125 return (ENETDOWN); 1126 } 1127 1128 if (uio->uio_resid == 0) { 1129 d->bd_wdcount++; 1130 return (0); 1131 } 1132 1133 bzero(&dst, sizeof(dst)); 1134 m = NULL; 1135 hlen = 0; 1136 /* XXX: bpf_movein() can sleep */ 1137 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, 1138 &m, &dst, &hlen, d); 1139 if (error) { 1140 d->bd_wdcount++; 1141 return (error); 1142 } 1143 d->bd_wfcount++; 1144 if (d->bd_hdrcmplt) 1145 dst.sa_family = pseudo_AF_HDRCMPLT; 1146 1147 if (d->bd_feedback) { 1148 mc = m_dup(m, M_NOWAIT); 1149 if (mc != NULL) 1150 mc->m_pkthdr.rcvif = ifp; 1151 /* Set M_PROMISC for outgoing packets to be discarded. */ 1152 if (d->bd_direction == BPF_D_INOUT) 1153 m->m_flags |= M_PROMISC; 1154 } else 1155 mc = NULL; 1156 1157 m->m_pkthdr.len -= hlen; 1158 m->m_len -= hlen; 1159 m->m_data += hlen; /* XXX */ 1160 1161 CURVNET_SET(ifp->if_vnet); 1162 #ifdef MAC 1163 BPFD_LOCK(d); 1164 mac_bpfdesc_create_mbuf(d, m); 1165 if (mc != NULL) 1166 mac_bpfdesc_create_mbuf(d, mc); 1167 BPFD_UNLOCK(d); 1168 #endif 1169 1170 bzero(&ro, sizeof(ro)); 1171 if (hlen != 0) { 1172 ro.ro_prepend = (u_char *)&dst.sa_data; 1173 ro.ro_plen = hlen; 1174 ro.ro_flags = RT_HAS_HEADER; 1175 } 1176 1177 error = (*ifp->if_output)(ifp, m, &dst, &ro); 1178 if (error) 1179 d->bd_wdcount++; 1180 1181 if (mc != NULL) { 1182 if (error == 0) 1183 (*ifp->if_input)(ifp, mc); 1184 else 1185 m_freem(mc); 1186 } 1187 CURVNET_RESTORE(); 1188 1189 return (error); 1190 } 1191 1192 /* 1193 * Reset a descriptor by flushing its packet buffer and clearing the receive 1194 * and drop counts. This is doable for kernel-only buffers, but with 1195 * zero-copy buffers, we can't write to (or rotate) buffers that are 1196 * currently owned by userspace. It would be nice if we could encapsulate 1197 * this logic in the buffer code rather than here. 1198 */ 1199 static void 1200 reset_d(struct bpf_d *d) 1201 { 1202 1203 BPFD_LOCK_ASSERT(d); 1204 1205 while (d->bd_hbuf_in_use) 1206 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1207 "bd_hbuf", 0); 1208 if ((d->bd_hbuf != NULL) && 1209 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1210 /* Free the hold buffer. */ 1211 d->bd_fbuf = d->bd_hbuf; 1212 d->bd_hbuf = NULL; 1213 d->bd_hlen = 0; 1214 bpf_buf_reclaimed(d); 1215 } 1216 if (bpf_canwritebuf(d)) 1217 d->bd_slen = 0; 1218 d->bd_rcount = 0; 1219 d->bd_dcount = 0; 1220 d->bd_fcount = 0; 1221 d->bd_wcount = 0; 1222 d->bd_wfcount = 0; 1223 d->bd_wdcount = 0; 1224 d->bd_zcopy = 0; 1225 } 1226 1227 /* 1228 * FIONREAD Check for read packet available. 1229 * BIOCGBLEN Get buffer len [for read()]. 1230 * BIOCSETF Set read filter. 1231 * BIOCSETFNR Set read filter without resetting descriptor. 1232 * BIOCSETWF Set write filter. 1233 * BIOCFLUSH Flush read packet buffer. 1234 * BIOCPROMISC Put interface into promiscuous mode. 1235 * BIOCGDLT Get link layer type. 1236 * BIOCGETIF Get interface name. 1237 * BIOCSETIF Set interface. 1238 * BIOCSRTIMEOUT Set read timeout. 1239 * BIOCGRTIMEOUT Get read timeout. 1240 * BIOCGSTATS Get packet stats. 1241 * BIOCIMMEDIATE Set immediate mode. 1242 * BIOCVERSION Get filter language version. 1243 * BIOCGHDRCMPLT Get "header already complete" flag 1244 * BIOCSHDRCMPLT Set "header already complete" flag 1245 * BIOCGDIRECTION Get packet direction flag 1246 * BIOCSDIRECTION Set packet direction flag 1247 * BIOCGTSTAMP Get time stamp format and resolution. 1248 * BIOCSTSTAMP Set time stamp format and resolution. 1249 * BIOCLOCK Set "locked" flag 1250 * BIOCFEEDBACK Set packet feedback mode. 1251 * BIOCSETZBUF Set current zero-copy buffer locations. 1252 * BIOCGETZMAX Get maximum zero-copy buffer size. 1253 * BIOCROTZBUF Force rotation of zero-copy buffer 1254 * BIOCSETBUFMODE Set buffer mode. 1255 * BIOCGETBUFMODE Get current buffer mode. 1256 */ 1257 /* ARGSUSED */ 1258 static int 1259 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1260 struct thread *td) 1261 { 1262 struct bpf_d *d; 1263 int error; 1264 1265 error = devfs_get_cdevpriv((void **)&d); 1266 if (error != 0) 1267 return (error); 1268 1269 /* 1270 * Refresh PID associated with this descriptor. 1271 */ 1272 BPFD_LOCK(d); 1273 BPF_PID_REFRESH(d, td); 1274 if (d->bd_state == BPF_WAITING) 1275 callout_stop(&d->bd_callout); 1276 d->bd_state = BPF_IDLE; 1277 BPFD_UNLOCK(d); 1278 1279 if (d->bd_locked == 1) { 1280 switch (cmd) { 1281 case BIOCGBLEN: 1282 case BIOCFLUSH: 1283 case BIOCGDLT: 1284 case BIOCGDLTLIST: 1285 #ifdef COMPAT_FREEBSD32 1286 case BIOCGDLTLIST32: 1287 #endif 1288 case BIOCGETIF: 1289 case BIOCGRTIMEOUT: 1290 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1291 case BIOCGRTIMEOUT32: 1292 #endif 1293 case BIOCGSTATS: 1294 case BIOCVERSION: 1295 case BIOCGRSIG: 1296 case BIOCGHDRCMPLT: 1297 case BIOCSTSTAMP: 1298 case BIOCFEEDBACK: 1299 case FIONREAD: 1300 case BIOCLOCK: 1301 case BIOCSRTIMEOUT: 1302 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1303 case BIOCSRTIMEOUT32: 1304 #endif 1305 case BIOCIMMEDIATE: 1306 case TIOCGPGRP: 1307 case BIOCROTZBUF: 1308 break; 1309 default: 1310 return (EPERM); 1311 } 1312 } 1313 #ifdef COMPAT_FREEBSD32 1314 /* 1315 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1316 * that it will get 32-bit packet headers. 1317 */ 1318 switch (cmd) { 1319 case BIOCSETF32: 1320 case BIOCSETFNR32: 1321 case BIOCSETWF32: 1322 case BIOCGDLTLIST32: 1323 case BIOCGRTIMEOUT32: 1324 case BIOCSRTIMEOUT32: 1325 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 1326 BPFD_LOCK(d); 1327 d->bd_compat32 = 1; 1328 BPFD_UNLOCK(d); 1329 } 1330 } 1331 #endif 1332 1333 CURVNET_SET(TD_TO_VNET(td)); 1334 switch (cmd) { 1335 1336 default: 1337 error = EINVAL; 1338 break; 1339 1340 /* 1341 * Check for read packet available. 1342 */ 1343 case FIONREAD: 1344 { 1345 int n; 1346 1347 BPFD_LOCK(d); 1348 n = d->bd_slen; 1349 while (d->bd_hbuf_in_use) 1350 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1351 PRINET, "bd_hbuf", 0); 1352 if (d->bd_hbuf) 1353 n += d->bd_hlen; 1354 BPFD_UNLOCK(d); 1355 1356 *(int *)addr = n; 1357 break; 1358 } 1359 1360 /* 1361 * Get buffer len [for read()]. 1362 */ 1363 case BIOCGBLEN: 1364 BPFD_LOCK(d); 1365 *(u_int *)addr = d->bd_bufsize; 1366 BPFD_UNLOCK(d); 1367 break; 1368 1369 /* 1370 * Set buffer length. 1371 */ 1372 case BIOCSBLEN: 1373 error = bpf_ioctl_sblen(d, (u_int *)addr); 1374 break; 1375 1376 /* 1377 * Set link layer read filter. 1378 */ 1379 case BIOCSETF: 1380 case BIOCSETFNR: 1381 case BIOCSETWF: 1382 #ifdef COMPAT_FREEBSD32 1383 case BIOCSETF32: 1384 case BIOCSETFNR32: 1385 case BIOCSETWF32: 1386 #endif 1387 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1388 break; 1389 1390 /* 1391 * Flush read packet buffer. 1392 */ 1393 case BIOCFLUSH: 1394 BPFD_LOCK(d); 1395 reset_d(d); 1396 BPFD_UNLOCK(d); 1397 break; 1398 1399 /* 1400 * Put interface into promiscuous mode. 1401 */ 1402 case BIOCPROMISC: 1403 if (d->bd_bif == NULL) { 1404 /* 1405 * No interface attached yet. 1406 */ 1407 error = EINVAL; 1408 break; 1409 } 1410 if (d->bd_promisc == 0) { 1411 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1412 if (error == 0) 1413 d->bd_promisc = 1; 1414 } 1415 break; 1416 1417 /* 1418 * Get current data link type. 1419 */ 1420 case BIOCGDLT: 1421 BPF_LOCK(); 1422 if (d->bd_bif == NULL) 1423 error = EINVAL; 1424 else 1425 *(u_int *)addr = d->bd_bif->bif_dlt; 1426 BPF_UNLOCK(); 1427 break; 1428 1429 /* 1430 * Get a list of supported data link types. 1431 */ 1432 #ifdef COMPAT_FREEBSD32 1433 case BIOCGDLTLIST32: 1434 { 1435 struct bpf_dltlist32 *list32; 1436 struct bpf_dltlist dltlist; 1437 1438 list32 = (struct bpf_dltlist32 *)addr; 1439 dltlist.bfl_len = list32->bfl_len; 1440 dltlist.bfl_list = PTRIN(list32->bfl_list); 1441 BPF_LOCK(); 1442 if (d->bd_bif == NULL) 1443 error = EINVAL; 1444 else { 1445 error = bpf_getdltlist(d, &dltlist); 1446 if (error == 0) 1447 list32->bfl_len = dltlist.bfl_len; 1448 } 1449 BPF_UNLOCK(); 1450 break; 1451 } 1452 #endif 1453 1454 case BIOCGDLTLIST: 1455 BPF_LOCK(); 1456 if (d->bd_bif == NULL) 1457 error = EINVAL; 1458 else 1459 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1460 BPF_UNLOCK(); 1461 break; 1462 1463 /* 1464 * Set data link type. 1465 */ 1466 case BIOCSDLT: 1467 BPF_LOCK(); 1468 if (d->bd_bif == NULL) 1469 error = EINVAL; 1470 else 1471 error = bpf_setdlt(d, *(u_int *)addr); 1472 BPF_UNLOCK(); 1473 break; 1474 1475 /* 1476 * Get interface name. 1477 */ 1478 case BIOCGETIF: 1479 BPF_LOCK(); 1480 if (d->bd_bif == NULL) 1481 error = EINVAL; 1482 else { 1483 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1484 struct ifreq *const ifr = (struct ifreq *)addr; 1485 1486 strlcpy(ifr->ifr_name, ifp->if_xname, 1487 sizeof(ifr->ifr_name)); 1488 } 1489 BPF_UNLOCK(); 1490 break; 1491 1492 /* 1493 * Set interface. 1494 */ 1495 case BIOCSETIF: 1496 { 1497 int alloc_buf, size; 1498 1499 /* 1500 * Behavior here depends on the buffering model. If 1501 * we're using kernel memory buffers, then we can 1502 * allocate them here. If we're using zero-copy, 1503 * then the user process must have registered buffers 1504 * by the time we get here. 1505 */ 1506 alloc_buf = 0; 1507 BPFD_LOCK(d); 1508 if (d->bd_bufmode == BPF_BUFMODE_BUFFER && 1509 d->bd_sbuf == NULL) 1510 alloc_buf = 1; 1511 BPFD_UNLOCK(d); 1512 if (alloc_buf) { 1513 size = d->bd_bufsize; 1514 error = bpf_buffer_ioctl_sblen(d, &size); 1515 if (error != 0) 1516 break; 1517 } 1518 BPF_LOCK(); 1519 error = bpf_setif(d, (struct ifreq *)addr); 1520 BPF_UNLOCK(); 1521 break; 1522 } 1523 1524 /* 1525 * Set read timeout. 1526 */ 1527 case BIOCSRTIMEOUT: 1528 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1529 case BIOCSRTIMEOUT32: 1530 #endif 1531 { 1532 struct timeval *tv = (struct timeval *)addr; 1533 #if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1534 struct timeval32 *tv32; 1535 struct timeval tv64; 1536 1537 if (cmd == BIOCSRTIMEOUT32) { 1538 tv32 = (struct timeval32 *)addr; 1539 tv = &tv64; 1540 tv->tv_sec = tv32->tv_sec; 1541 tv->tv_usec = tv32->tv_usec; 1542 } else 1543 #endif 1544 tv = (struct timeval *)addr; 1545 1546 /* 1547 * Subtract 1 tick from tvtohz() since this isn't 1548 * a one-shot timer. 1549 */ 1550 if ((error = itimerfix(tv)) == 0) 1551 d->bd_rtout = tvtohz(tv) - 1; 1552 break; 1553 } 1554 1555 /* 1556 * Get read timeout. 1557 */ 1558 case BIOCGRTIMEOUT: 1559 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1560 case BIOCGRTIMEOUT32: 1561 #endif 1562 { 1563 struct timeval *tv; 1564 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1565 struct timeval32 *tv32; 1566 struct timeval tv64; 1567 1568 if (cmd == BIOCGRTIMEOUT32) 1569 tv = &tv64; 1570 else 1571 #endif 1572 tv = (struct timeval *)addr; 1573 1574 tv->tv_sec = d->bd_rtout / hz; 1575 tv->tv_usec = (d->bd_rtout % hz) * tick; 1576 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1577 if (cmd == BIOCGRTIMEOUT32) { 1578 tv32 = (struct timeval32 *)addr; 1579 tv32->tv_sec = tv->tv_sec; 1580 tv32->tv_usec = tv->tv_usec; 1581 } 1582 #endif 1583 1584 break; 1585 } 1586 1587 /* 1588 * Get packet stats. 1589 */ 1590 case BIOCGSTATS: 1591 { 1592 struct bpf_stat *bs = (struct bpf_stat *)addr; 1593 1594 /* XXXCSJP overflow */ 1595 bs->bs_recv = d->bd_rcount; 1596 bs->bs_drop = d->bd_dcount; 1597 break; 1598 } 1599 1600 /* 1601 * Set immediate mode. 1602 */ 1603 case BIOCIMMEDIATE: 1604 BPFD_LOCK(d); 1605 d->bd_immediate = *(u_int *)addr; 1606 BPFD_UNLOCK(d); 1607 break; 1608 1609 case BIOCVERSION: 1610 { 1611 struct bpf_version *bv = (struct bpf_version *)addr; 1612 1613 bv->bv_major = BPF_MAJOR_VERSION; 1614 bv->bv_minor = BPF_MINOR_VERSION; 1615 break; 1616 } 1617 1618 /* 1619 * Get "header already complete" flag 1620 */ 1621 case BIOCGHDRCMPLT: 1622 BPFD_LOCK(d); 1623 *(u_int *)addr = d->bd_hdrcmplt; 1624 BPFD_UNLOCK(d); 1625 break; 1626 1627 /* 1628 * Set "header already complete" flag 1629 */ 1630 case BIOCSHDRCMPLT: 1631 BPFD_LOCK(d); 1632 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1633 BPFD_UNLOCK(d); 1634 break; 1635 1636 /* 1637 * Get packet direction flag 1638 */ 1639 case BIOCGDIRECTION: 1640 BPFD_LOCK(d); 1641 *(u_int *)addr = d->bd_direction; 1642 BPFD_UNLOCK(d); 1643 break; 1644 1645 /* 1646 * Set packet direction flag 1647 */ 1648 case BIOCSDIRECTION: 1649 { 1650 u_int direction; 1651 1652 direction = *(u_int *)addr; 1653 switch (direction) { 1654 case BPF_D_IN: 1655 case BPF_D_INOUT: 1656 case BPF_D_OUT: 1657 BPFD_LOCK(d); 1658 d->bd_direction = direction; 1659 BPFD_UNLOCK(d); 1660 break; 1661 default: 1662 error = EINVAL; 1663 } 1664 } 1665 break; 1666 1667 /* 1668 * Get packet timestamp format and resolution. 1669 */ 1670 case BIOCGTSTAMP: 1671 BPFD_LOCK(d); 1672 *(u_int *)addr = d->bd_tstamp; 1673 BPFD_UNLOCK(d); 1674 break; 1675 1676 /* 1677 * Set packet timestamp format and resolution. 1678 */ 1679 case BIOCSTSTAMP: 1680 { 1681 u_int func; 1682 1683 func = *(u_int *)addr; 1684 if (BPF_T_VALID(func)) 1685 d->bd_tstamp = func; 1686 else 1687 error = EINVAL; 1688 } 1689 break; 1690 1691 case BIOCFEEDBACK: 1692 BPFD_LOCK(d); 1693 d->bd_feedback = *(u_int *)addr; 1694 BPFD_UNLOCK(d); 1695 break; 1696 1697 case BIOCLOCK: 1698 BPFD_LOCK(d); 1699 d->bd_locked = 1; 1700 BPFD_UNLOCK(d); 1701 break; 1702 1703 case FIONBIO: /* Non-blocking I/O */ 1704 break; 1705 1706 case FIOASYNC: /* Send signal on receive packets */ 1707 BPFD_LOCK(d); 1708 d->bd_async = *(int *)addr; 1709 BPFD_UNLOCK(d); 1710 break; 1711 1712 case FIOSETOWN: 1713 /* 1714 * XXX: Add some sort of locking here? 1715 * fsetown() can sleep. 1716 */ 1717 error = fsetown(*(int *)addr, &d->bd_sigio); 1718 break; 1719 1720 case FIOGETOWN: 1721 BPFD_LOCK(d); 1722 *(int *)addr = fgetown(&d->bd_sigio); 1723 BPFD_UNLOCK(d); 1724 break; 1725 1726 /* This is deprecated, FIOSETOWN should be used instead. */ 1727 case TIOCSPGRP: 1728 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1729 break; 1730 1731 /* This is deprecated, FIOGETOWN should be used instead. */ 1732 case TIOCGPGRP: 1733 *(int *)addr = -fgetown(&d->bd_sigio); 1734 break; 1735 1736 case BIOCSRSIG: /* Set receive signal */ 1737 { 1738 u_int sig; 1739 1740 sig = *(u_int *)addr; 1741 1742 if (sig >= NSIG) 1743 error = EINVAL; 1744 else { 1745 BPFD_LOCK(d); 1746 d->bd_sig = sig; 1747 BPFD_UNLOCK(d); 1748 } 1749 break; 1750 } 1751 case BIOCGRSIG: 1752 BPFD_LOCK(d); 1753 *(u_int *)addr = d->bd_sig; 1754 BPFD_UNLOCK(d); 1755 break; 1756 1757 case BIOCGETBUFMODE: 1758 BPFD_LOCK(d); 1759 *(u_int *)addr = d->bd_bufmode; 1760 BPFD_UNLOCK(d); 1761 break; 1762 1763 case BIOCSETBUFMODE: 1764 /* 1765 * Allow the buffering mode to be changed as long as we 1766 * haven't yet committed to a particular mode. Our 1767 * definition of commitment, for now, is whether or not a 1768 * buffer has been allocated or an interface attached, since 1769 * that's the point where things get tricky. 1770 */ 1771 switch (*(u_int *)addr) { 1772 case BPF_BUFMODE_BUFFER: 1773 break; 1774 1775 case BPF_BUFMODE_ZBUF: 1776 if (bpf_zerocopy_enable) 1777 break; 1778 /* FALLSTHROUGH */ 1779 1780 default: 1781 CURVNET_RESTORE(); 1782 return (EINVAL); 1783 } 1784 1785 BPFD_LOCK(d); 1786 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1787 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1788 BPFD_UNLOCK(d); 1789 CURVNET_RESTORE(); 1790 return (EBUSY); 1791 } 1792 d->bd_bufmode = *(u_int *)addr; 1793 BPFD_UNLOCK(d); 1794 break; 1795 1796 case BIOCGETZMAX: 1797 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1798 break; 1799 1800 case BIOCSETZBUF: 1801 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1802 break; 1803 1804 case BIOCROTZBUF: 1805 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1806 break; 1807 } 1808 CURVNET_RESTORE(); 1809 return (error); 1810 } 1811 1812 /* 1813 * Set d's packet filter program to fp. If this file already has a filter, 1814 * free it and replace it. Returns EINVAL for bogus requests. 1815 * 1816 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls 1817 * since reading d->bd_bif can't be protected by d or interface lock due to 1818 * lock order. 1819 * 1820 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses 1821 * interface read lock to read all filers. 1822 * 1823 */ 1824 static int 1825 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1826 { 1827 #ifdef COMPAT_FREEBSD32 1828 struct bpf_program fp_swab; 1829 struct bpf_program32 *fp32; 1830 #endif 1831 struct bpf_insn *fcode, *old; 1832 #ifdef BPF_JITTER 1833 bpf_jit_filter *jfunc, *ofunc; 1834 #endif 1835 size_t size; 1836 u_int flen; 1837 int need_upgrade; 1838 1839 #ifdef COMPAT_FREEBSD32 1840 switch (cmd) { 1841 case BIOCSETF32: 1842 case BIOCSETWF32: 1843 case BIOCSETFNR32: 1844 fp32 = (struct bpf_program32 *)fp; 1845 fp_swab.bf_len = fp32->bf_len; 1846 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1847 fp = &fp_swab; 1848 switch (cmd) { 1849 case BIOCSETF32: 1850 cmd = BIOCSETF; 1851 break; 1852 case BIOCSETWF32: 1853 cmd = BIOCSETWF; 1854 break; 1855 } 1856 break; 1857 } 1858 #endif 1859 1860 fcode = NULL; 1861 #ifdef BPF_JITTER 1862 jfunc = ofunc = NULL; 1863 #endif 1864 need_upgrade = 0; 1865 1866 /* 1867 * Check new filter validness before acquiring any locks. 1868 * Allocate memory for new filter, if needed. 1869 */ 1870 flen = fp->bf_len; 1871 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1872 return (EINVAL); 1873 size = flen * sizeof(*fp->bf_insns); 1874 if (size > 0) { 1875 /* We're setting up new filter. Copy and check actual data. */ 1876 fcode = malloc(size, M_BPF, M_WAITOK); 1877 if (copyin(fp->bf_insns, fcode, size) != 0 || 1878 !bpf_validate(fcode, flen)) { 1879 free(fcode, M_BPF); 1880 return (EINVAL); 1881 } 1882 #ifdef BPF_JITTER 1883 /* Filter is copied inside fcode and is perfectly valid. */ 1884 jfunc = bpf_jitter(fcode, flen); 1885 #endif 1886 } 1887 1888 BPF_LOCK(); 1889 1890 /* 1891 * Set up new filter. 1892 * Protect filter change by interface lock. 1893 * Additionally, we are protected by global lock here. 1894 */ 1895 if (d->bd_bif != NULL) 1896 BPFIF_WLOCK(d->bd_bif); 1897 BPFD_LOCK(d); 1898 if (cmd == BIOCSETWF) { 1899 old = d->bd_wfilter; 1900 d->bd_wfilter = fcode; 1901 } else { 1902 old = d->bd_rfilter; 1903 d->bd_rfilter = fcode; 1904 #ifdef BPF_JITTER 1905 ofunc = d->bd_bfilter; 1906 d->bd_bfilter = jfunc; 1907 #endif 1908 if (cmd == BIOCSETF) 1909 reset_d(d); 1910 1911 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen); 1912 } 1913 BPFD_UNLOCK(d); 1914 if (d->bd_bif != NULL) 1915 BPFIF_WUNLOCK(d->bd_bif); 1916 if (old != NULL) 1917 free(old, M_BPF); 1918 #ifdef BPF_JITTER 1919 if (ofunc != NULL) 1920 bpf_destroy_jit_filter(ofunc); 1921 #endif 1922 1923 /* Move d to active readers list. */ 1924 if (need_upgrade != 0) 1925 bpf_upgraded(d); 1926 1927 BPF_UNLOCK(); 1928 return (0); 1929 } 1930 1931 /* 1932 * Detach a file from its current interface (if attached at all) and attach 1933 * to the interface indicated by the name stored in ifr. 1934 * Return an errno or 0. 1935 */ 1936 static int 1937 bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1938 { 1939 struct bpf_if *bp; 1940 struct ifnet *theywant; 1941 1942 BPF_LOCK_ASSERT(); 1943 1944 theywant = ifunit(ifr->ifr_name); 1945 if (theywant == NULL || theywant->if_bpf == NULL) 1946 return (ENXIO); 1947 1948 bp = theywant->if_bpf; 1949 1950 /* Check if interface is not being detached from BPF */ 1951 BPFIF_RLOCK(bp); 1952 if (bp->bif_flags & BPFIF_FLAG_DYING) { 1953 BPFIF_RUNLOCK(bp); 1954 return (ENXIO); 1955 } 1956 BPFIF_RUNLOCK(bp); 1957 1958 /* 1959 * At this point, we expect the buffer is already allocated. If not, 1960 * return an error. 1961 */ 1962 switch (d->bd_bufmode) { 1963 case BPF_BUFMODE_BUFFER: 1964 case BPF_BUFMODE_ZBUF: 1965 if (d->bd_sbuf == NULL) 1966 return (EINVAL); 1967 break; 1968 1969 default: 1970 panic("bpf_setif: bufmode %d", d->bd_bufmode); 1971 } 1972 if (bp != d->bd_bif) 1973 bpf_attachd(d, bp); 1974 BPFD_LOCK(d); 1975 reset_d(d); 1976 BPFD_UNLOCK(d); 1977 return (0); 1978 } 1979 1980 /* 1981 * Support for select() and poll() system calls 1982 * 1983 * Return true iff the specific operation will not block indefinitely. 1984 * Otherwise, return false but make a note that a selwakeup() must be done. 1985 */ 1986 static int 1987 bpfpoll(struct cdev *dev, int events, struct thread *td) 1988 { 1989 struct bpf_d *d; 1990 int revents; 1991 1992 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 1993 return (events & 1994 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 1995 1996 /* 1997 * Refresh PID associated with this descriptor. 1998 */ 1999 revents = events & (POLLOUT | POLLWRNORM); 2000 BPFD_LOCK(d); 2001 BPF_PID_REFRESH(d, td); 2002 if (events & (POLLIN | POLLRDNORM)) { 2003 if (bpf_ready(d)) 2004 revents |= events & (POLLIN | POLLRDNORM); 2005 else { 2006 selrecord(td, &d->bd_sel); 2007 /* Start the read timeout if necessary. */ 2008 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2009 callout_reset(&d->bd_callout, d->bd_rtout, 2010 bpf_timed_out, d); 2011 d->bd_state = BPF_WAITING; 2012 } 2013 } 2014 } 2015 BPFD_UNLOCK(d); 2016 return (revents); 2017 } 2018 2019 /* 2020 * Support for kevent() system call. Register EVFILT_READ filters and 2021 * reject all others. 2022 */ 2023 int 2024 bpfkqfilter(struct cdev *dev, struct knote *kn) 2025 { 2026 struct bpf_d *d; 2027 2028 if (devfs_get_cdevpriv((void **)&d) != 0 || 2029 kn->kn_filter != EVFILT_READ) 2030 return (1); 2031 2032 /* 2033 * Refresh PID associated with this descriptor. 2034 */ 2035 BPFD_LOCK(d); 2036 BPF_PID_REFRESH_CUR(d); 2037 kn->kn_fop = &bpfread_filtops; 2038 kn->kn_hook = d; 2039 knlist_add(&d->bd_sel.si_note, kn, 1); 2040 BPFD_UNLOCK(d); 2041 2042 return (0); 2043 } 2044 2045 static void 2046 filt_bpfdetach(struct knote *kn) 2047 { 2048 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2049 2050 knlist_remove(&d->bd_sel.si_note, kn, 0); 2051 } 2052 2053 static int 2054 filt_bpfread(struct knote *kn, long hint) 2055 { 2056 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2057 int ready; 2058 2059 BPFD_LOCK_ASSERT(d); 2060 ready = bpf_ready(d); 2061 if (ready) { 2062 kn->kn_data = d->bd_slen; 2063 /* 2064 * Ignore the hold buffer if it is being copied to user space. 2065 */ 2066 if (!d->bd_hbuf_in_use && d->bd_hbuf) 2067 kn->kn_data += d->bd_hlen; 2068 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2069 callout_reset(&d->bd_callout, d->bd_rtout, 2070 bpf_timed_out, d); 2071 d->bd_state = BPF_WAITING; 2072 } 2073 2074 return (ready); 2075 } 2076 2077 #define BPF_TSTAMP_NONE 0 2078 #define BPF_TSTAMP_FAST 1 2079 #define BPF_TSTAMP_NORMAL 2 2080 #define BPF_TSTAMP_EXTERN 3 2081 2082 static int 2083 bpf_ts_quality(int tstype) 2084 { 2085 2086 if (tstype == BPF_T_NONE) 2087 return (BPF_TSTAMP_NONE); 2088 if ((tstype & BPF_T_FAST) != 0) 2089 return (BPF_TSTAMP_FAST); 2090 2091 return (BPF_TSTAMP_NORMAL); 2092 } 2093 2094 static int 2095 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2096 { 2097 struct m_tag *tag; 2098 int quality; 2099 2100 quality = bpf_ts_quality(tstype); 2101 if (quality == BPF_TSTAMP_NONE) 2102 return (quality); 2103 2104 if (m != NULL) { 2105 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2106 if (tag != NULL) { 2107 *bt = *(struct bintime *)(tag + 1); 2108 return (BPF_TSTAMP_EXTERN); 2109 } 2110 } 2111 if (quality == BPF_TSTAMP_NORMAL) 2112 binuptime(bt); 2113 else 2114 getbinuptime(bt); 2115 2116 return (quality); 2117 } 2118 2119 /* 2120 * Incoming linkage from device drivers. Process the packet pkt, of length 2121 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2122 * by each process' filter, and if accepted, stashed into the corresponding 2123 * buffer. 2124 */ 2125 void 2126 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2127 { 2128 struct bintime bt; 2129 struct bpf_d *d; 2130 #ifdef BPF_JITTER 2131 bpf_jit_filter *bf; 2132 #endif 2133 u_int slen; 2134 int gottime; 2135 2136 gottime = BPF_TSTAMP_NONE; 2137 2138 BPFIF_RLOCK(bp); 2139 2140 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2141 /* 2142 * We are not using any locks for d here because: 2143 * 1) any filter change is protected by interface 2144 * write lock 2145 * 2) destroying/detaching d is protected by interface 2146 * write lock, too 2147 */ 2148 2149 /* XXX: Do not protect counter for the sake of performance. */ 2150 ++d->bd_rcount; 2151 /* 2152 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no 2153 * way for the caller to indiciate to us whether this packet 2154 * is inbound or outbound. In the bpf_mtap() routines, we use 2155 * the interface pointers on the mbuf to figure it out. 2156 */ 2157 #ifdef BPF_JITTER 2158 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2159 if (bf != NULL) 2160 slen = (*(bf->func))(pkt, pktlen, pktlen); 2161 else 2162 #endif 2163 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2164 if (slen != 0) { 2165 /* 2166 * Filter matches. Let's to acquire write lock. 2167 */ 2168 BPFD_LOCK(d); 2169 2170 d->bd_fcount++; 2171 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2172 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); 2173 #ifdef MAC 2174 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2175 #endif 2176 catchpacket(d, pkt, pktlen, slen, 2177 bpf_append_bytes, &bt); 2178 BPFD_UNLOCK(d); 2179 } 2180 } 2181 BPFIF_RUNLOCK(bp); 2182 } 2183 2184 #define BPF_CHECK_DIRECTION(d, r, i) \ 2185 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2186 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2187 2188 /* 2189 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2190 * Locking model is explained in bpf_tap(). 2191 */ 2192 void 2193 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2194 { 2195 struct bintime bt; 2196 struct bpf_d *d; 2197 #ifdef BPF_JITTER 2198 bpf_jit_filter *bf; 2199 #endif 2200 u_int pktlen, slen; 2201 int gottime; 2202 2203 /* Skip outgoing duplicate packets. */ 2204 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2205 m->m_flags &= ~M_PROMISC; 2206 return; 2207 } 2208 2209 pktlen = m_length(m, NULL); 2210 gottime = BPF_TSTAMP_NONE; 2211 2212 BPFIF_RLOCK(bp); 2213 2214 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2215 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2216 continue; 2217 ++d->bd_rcount; 2218 #ifdef BPF_JITTER 2219 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2220 /* XXX We cannot handle multiple mbufs. */ 2221 if (bf != NULL && m->m_next == NULL) 2222 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); 2223 else 2224 #endif 2225 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2226 if (slen != 0) { 2227 BPFD_LOCK(d); 2228 2229 d->bd_fcount++; 2230 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2231 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2232 #ifdef MAC 2233 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2234 #endif 2235 catchpacket(d, (u_char *)m, pktlen, slen, 2236 bpf_append_mbuf, &bt); 2237 BPFD_UNLOCK(d); 2238 } 2239 } 2240 BPFIF_RUNLOCK(bp); 2241 } 2242 2243 /* 2244 * Incoming linkage from device drivers, when packet is in 2245 * an mbuf chain and to be prepended by a contiguous header. 2246 */ 2247 void 2248 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2249 { 2250 struct bintime bt; 2251 struct mbuf mb; 2252 struct bpf_d *d; 2253 u_int pktlen, slen; 2254 int gottime; 2255 2256 /* Skip outgoing duplicate packets. */ 2257 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2258 m->m_flags &= ~M_PROMISC; 2259 return; 2260 } 2261 2262 pktlen = m_length(m, NULL); 2263 /* 2264 * Craft on-stack mbuf suitable for passing to bpf_filter. 2265 * Note that we cut corners here; we only setup what's 2266 * absolutely needed--this mbuf should never go anywhere else. 2267 */ 2268 mb.m_next = m; 2269 mb.m_data = data; 2270 mb.m_len = dlen; 2271 pktlen += dlen; 2272 2273 gottime = BPF_TSTAMP_NONE; 2274 2275 BPFIF_RLOCK(bp); 2276 2277 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2278 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2279 continue; 2280 ++d->bd_rcount; 2281 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2282 if (slen != 0) { 2283 BPFD_LOCK(d); 2284 2285 d->bd_fcount++; 2286 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2287 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2288 #ifdef MAC 2289 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2290 #endif 2291 catchpacket(d, (u_char *)&mb, pktlen, slen, 2292 bpf_append_mbuf, &bt); 2293 BPFD_UNLOCK(d); 2294 } 2295 } 2296 BPFIF_RUNLOCK(bp); 2297 } 2298 2299 #undef BPF_CHECK_DIRECTION 2300 2301 #undef BPF_TSTAMP_NONE 2302 #undef BPF_TSTAMP_FAST 2303 #undef BPF_TSTAMP_NORMAL 2304 #undef BPF_TSTAMP_EXTERN 2305 2306 static int 2307 bpf_hdrlen(struct bpf_d *d) 2308 { 2309 int hdrlen; 2310 2311 hdrlen = d->bd_bif->bif_hdrlen; 2312 #ifndef BURN_BRIDGES 2313 if (d->bd_tstamp == BPF_T_NONE || 2314 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2315 #ifdef COMPAT_FREEBSD32 2316 if (d->bd_compat32) 2317 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2318 else 2319 #endif 2320 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2321 else 2322 #endif 2323 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2324 #ifdef COMPAT_FREEBSD32 2325 if (d->bd_compat32) 2326 hdrlen = BPF_WORDALIGN32(hdrlen); 2327 else 2328 #endif 2329 hdrlen = BPF_WORDALIGN(hdrlen); 2330 2331 return (hdrlen - d->bd_bif->bif_hdrlen); 2332 } 2333 2334 static void 2335 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2336 { 2337 struct bintime bt2, boottimebin; 2338 struct timeval tsm; 2339 struct timespec tsn; 2340 2341 if ((tstype & BPF_T_MONOTONIC) == 0) { 2342 bt2 = *bt; 2343 getboottimebin(&boottimebin); 2344 bintime_add(&bt2, &boottimebin); 2345 bt = &bt2; 2346 } 2347 switch (BPF_T_FORMAT(tstype)) { 2348 case BPF_T_MICROTIME: 2349 bintime2timeval(bt, &tsm); 2350 ts->bt_sec = tsm.tv_sec; 2351 ts->bt_frac = tsm.tv_usec; 2352 break; 2353 case BPF_T_NANOTIME: 2354 bintime2timespec(bt, &tsn); 2355 ts->bt_sec = tsn.tv_sec; 2356 ts->bt_frac = tsn.tv_nsec; 2357 break; 2358 case BPF_T_BINTIME: 2359 ts->bt_sec = bt->sec; 2360 ts->bt_frac = bt->frac; 2361 break; 2362 } 2363 } 2364 2365 /* 2366 * Move the packet data from interface memory (pkt) into the 2367 * store buffer. "cpfn" is the routine called to do the actual data 2368 * transfer. bcopy is passed in to copy contiguous chunks, while 2369 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2370 * pkt is really an mbuf. 2371 */ 2372 static void 2373 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2374 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2375 struct bintime *bt) 2376 { 2377 struct bpf_xhdr hdr; 2378 #ifndef BURN_BRIDGES 2379 struct bpf_hdr hdr_old; 2380 #ifdef COMPAT_FREEBSD32 2381 struct bpf_hdr32 hdr32_old; 2382 #endif 2383 #endif 2384 int caplen, curlen, hdrlen, totlen; 2385 int do_wakeup = 0; 2386 int do_timestamp; 2387 int tstype; 2388 2389 BPFD_LOCK_ASSERT(d); 2390 2391 /* 2392 * Detect whether user space has released a buffer back to us, and if 2393 * so, move it from being a hold buffer to a free buffer. This may 2394 * not be the best place to do it (for example, we might only want to 2395 * run this check if we need the space), but for now it's a reliable 2396 * spot to do it. 2397 */ 2398 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2399 d->bd_fbuf = d->bd_hbuf; 2400 d->bd_hbuf = NULL; 2401 d->bd_hlen = 0; 2402 bpf_buf_reclaimed(d); 2403 } 2404 2405 /* 2406 * Figure out how many bytes to move. If the packet is 2407 * greater or equal to the snapshot length, transfer that 2408 * much. Otherwise, transfer the whole packet (unless 2409 * we hit the buffer size limit). 2410 */ 2411 hdrlen = bpf_hdrlen(d); 2412 totlen = hdrlen + min(snaplen, pktlen); 2413 if (totlen > d->bd_bufsize) 2414 totlen = d->bd_bufsize; 2415 2416 /* 2417 * Round up the end of the previous packet to the next longword. 2418 * 2419 * Drop the packet if there's no room and no hope of room 2420 * If the packet would overflow the storage buffer or the storage 2421 * buffer is considered immutable by the buffer model, try to rotate 2422 * the buffer and wakeup pending processes. 2423 */ 2424 #ifdef COMPAT_FREEBSD32 2425 if (d->bd_compat32) 2426 curlen = BPF_WORDALIGN32(d->bd_slen); 2427 else 2428 #endif 2429 curlen = BPF_WORDALIGN(d->bd_slen); 2430 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2431 if (d->bd_fbuf == NULL) { 2432 /* 2433 * There's no room in the store buffer, and no 2434 * prospect of room, so drop the packet. Notify the 2435 * buffer model. 2436 */ 2437 bpf_buffull(d); 2438 ++d->bd_dcount; 2439 return; 2440 } 2441 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); 2442 ROTATE_BUFFERS(d); 2443 do_wakeup = 1; 2444 curlen = 0; 2445 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2446 /* 2447 * Immediate mode is set, or the read timeout has already 2448 * expired during a select call. A packet arrived, so the 2449 * reader should be woken up. 2450 */ 2451 do_wakeup = 1; 2452 caplen = totlen - hdrlen; 2453 tstype = d->bd_tstamp; 2454 do_timestamp = tstype != BPF_T_NONE; 2455 #ifndef BURN_BRIDGES 2456 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2457 struct bpf_ts ts; 2458 if (do_timestamp) 2459 bpf_bintime2ts(bt, &ts, tstype); 2460 #ifdef COMPAT_FREEBSD32 2461 if (d->bd_compat32) { 2462 bzero(&hdr32_old, sizeof(hdr32_old)); 2463 if (do_timestamp) { 2464 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2465 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2466 } 2467 hdr32_old.bh_datalen = pktlen; 2468 hdr32_old.bh_hdrlen = hdrlen; 2469 hdr32_old.bh_caplen = caplen; 2470 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2471 sizeof(hdr32_old)); 2472 goto copy; 2473 } 2474 #endif 2475 bzero(&hdr_old, sizeof(hdr_old)); 2476 if (do_timestamp) { 2477 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2478 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2479 } 2480 hdr_old.bh_datalen = pktlen; 2481 hdr_old.bh_hdrlen = hdrlen; 2482 hdr_old.bh_caplen = caplen; 2483 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2484 sizeof(hdr_old)); 2485 goto copy; 2486 } 2487 #endif 2488 2489 /* 2490 * Append the bpf header. Note we append the actual header size, but 2491 * move forward the length of the header plus padding. 2492 */ 2493 bzero(&hdr, sizeof(hdr)); 2494 if (do_timestamp) 2495 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2496 hdr.bh_datalen = pktlen; 2497 hdr.bh_hdrlen = hdrlen; 2498 hdr.bh_caplen = caplen; 2499 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2500 2501 /* 2502 * Copy the packet data into the store buffer and update its length. 2503 */ 2504 #ifndef BURN_BRIDGES 2505 copy: 2506 #endif 2507 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2508 d->bd_slen = curlen + totlen; 2509 2510 if (do_wakeup) 2511 bpf_wakeup(d); 2512 } 2513 2514 /* 2515 * Free buffers currently in use by a descriptor. 2516 * Called on close. 2517 */ 2518 static void 2519 bpf_freed(struct bpf_d *d) 2520 { 2521 2522 /* 2523 * We don't need to lock out interrupts since this descriptor has 2524 * been detached from its interface and it yet hasn't been marked 2525 * free. 2526 */ 2527 bpf_free(d); 2528 if (d->bd_rfilter != NULL) { 2529 free((caddr_t)d->bd_rfilter, M_BPF); 2530 #ifdef BPF_JITTER 2531 if (d->bd_bfilter != NULL) 2532 bpf_destroy_jit_filter(d->bd_bfilter); 2533 #endif 2534 } 2535 if (d->bd_wfilter != NULL) 2536 free((caddr_t)d->bd_wfilter, M_BPF); 2537 mtx_destroy(&d->bd_lock); 2538 } 2539 2540 /* 2541 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2542 * fixed size of the link header (variable length headers not yet supported). 2543 */ 2544 void 2545 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2546 { 2547 2548 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2549 } 2550 2551 /* 2552 * Attach an interface to bpf. ifp is a pointer to the structure 2553 * defining the interface to be attached, dlt is the link layer type, 2554 * and hdrlen is the fixed size of the link header (variable length 2555 * headers are not yet supporrted). 2556 */ 2557 void 2558 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2559 { 2560 struct bpf_if *bp; 2561 2562 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); 2563 if (bp == NULL) 2564 panic("bpfattach"); 2565 2566 LIST_INIT(&bp->bif_dlist); 2567 LIST_INIT(&bp->bif_wlist); 2568 bp->bif_ifp = ifp; 2569 bp->bif_dlt = dlt; 2570 rw_init(&bp->bif_lock, "bpf interface lock"); 2571 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); 2572 bp->bif_bpf = driverp; 2573 *driverp = bp; 2574 2575 BPF_LOCK(); 2576 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2577 BPF_UNLOCK(); 2578 2579 bp->bif_hdrlen = hdrlen; 2580 2581 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 2582 if_printf(ifp, "bpf attached\n"); 2583 } 2584 2585 #ifdef VIMAGE 2586 /* 2587 * When moving interfaces between vnet instances we need a way to 2588 * query the dlt and hdrlen before detach so we can re-attch the if_bpf 2589 * after the vmove. We unfortunately have no device driver infrastructure 2590 * to query the interface for these values after creation/attach, thus 2591 * add this as a workaround. 2592 */ 2593 int 2594 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen) 2595 { 2596 2597 if (bp == NULL) 2598 return (ENXIO); 2599 if (bif_dlt == NULL && bif_hdrlen == NULL) 2600 return (0); 2601 2602 if (bif_dlt != NULL) 2603 *bif_dlt = bp->bif_dlt; 2604 if (bif_hdrlen != NULL) 2605 *bif_hdrlen = bp->bif_hdrlen; 2606 2607 return (0); 2608 } 2609 #endif 2610 2611 /* 2612 * Detach bpf from an interface. This involves detaching each descriptor 2613 * associated with the interface. Notify each descriptor as it's detached 2614 * so that any sleepers wake up and get ENXIO. 2615 */ 2616 void 2617 bpfdetach(struct ifnet *ifp) 2618 { 2619 struct bpf_if *bp, *bp_temp; 2620 struct bpf_d *d; 2621 int ndetached; 2622 2623 ndetached = 0; 2624 2625 BPF_LOCK(); 2626 /* Find all bpf_if struct's which reference ifp and detach them. */ 2627 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2628 if (ifp != bp->bif_ifp) 2629 continue; 2630 2631 LIST_REMOVE(bp, bif_next); 2632 /* Add to to-be-freed list */ 2633 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); 2634 2635 ndetached++; 2636 /* 2637 * Delay freeing bp till interface is detached 2638 * and all routes through this interface are removed. 2639 * Mark bp as detached to restrict new consumers. 2640 */ 2641 BPFIF_WLOCK(bp); 2642 bp->bif_flags |= BPFIF_FLAG_DYING; 2643 *bp->bif_bpf = NULL; 2644 BPFIF_WUNLOCK(bp); 2645 2646 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", 2647 __func__, bp->bif_dlt, bp, ifp); 2648 2649 /* Free common descriptors */ 2650 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { 2651 bpf_detachd_locked(d); 2652 BPFD_LOCK(d); 2653 bpf_wakeup(d); 2654 BPFD_UNLOCK(d); 2655 } 2656 2657 /* Free writer-only descriptors */ 2658 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { 2659 bpf_detachd_locked(d); 2660 BPFD_LOCK(d); 2661 bpf_wakeup(d); 2662 BPFD_UNLOCK(d); 2663 } 2664 } 2665 BPF_UNLOCK(); 2666 2667 #ifdef INVARIANTS 2668 if (ndetached == 0) 2669 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 2670 #endif 2671 } 2672 2673 /* 2674 * Interface departure handler. 2675 * Note departure event does not guarantee interface is going down. 2676 * Interface renaming is currently done via departure/arrival event set. 2677 * 2678 * Departure handled is called after all routes pointing to 2679 * given interface are removed and interface is in down state 2680 * restricting any packets to be sent/received. We assume it is now safe 2681 * to free data allocated by BPF. 2682 */ 2683 static void 2684 bpf_ifdetach(void *arg __unused, struct ifnet *ifp) 2685 { 2686 struct bpf_if *bp, *bp_temp; 2687 int nmatched = 0; 2688 2689 /* Ignore ifnet renaming. */ 2690 if (ifp->if_flags & IFF_RENAMING) 2691 return; 2692 2693 BPF_LOCK(); 2694 /* 2695 * Find matching entries in free list. 2696 * Nothing should be found if bpfdetach() was not called. 2697 */ 2698 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { 2699 if (ifp != bp->bif_ifp) 2700 continue; 2701 2702 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", 2703 __func__, bp, ifp); 2704 2705 LIST_REMOVE(bp, bif_next); 2706 2707 rw_destroy(&bp->bif_lock); 2708 free(bp, M_BPF); 2709 2710 nmatched++; 2711 } 2712 BPF_UNLOCK(); 2713 } 2714 2715 /* 2716 * Get a list of available data link type of the interface. 2717 */ 2718 static int 2719 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2720 { 2721 struct ifnet *ifp; 2722 struct bpf_if *bp; 2723 u_int *lst; 2724 int error, n, n1; 2725 2726 BPF_LOCK_ASSERT(); 2727 2728 ifp = d->bd_bif->bif_ifp; 2729 again: 2730 n1 = 0; 2731 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2732 if (bp->bif_ifp == ifp) 2733 n1++; 2734 } 2735 if (bfl->bfl_list == NULL) { 2736 bfl->bfl_len = n1; 2737 return (0); 2738 } 2739 if (n1 > bfl->bfl_len) 2740 return (ENOMEM); 2741 BPF_UNLOCK(); 2742 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); 2743 n = 0; 2744 BPF_LOCK(); 2745 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2746 if (bp->bif_ifp != ifp) 2747 continue; 2748 if (n >= n1) { 2749 free(lst, M_TEMP); 2750 goto again; 2751 } 2752 lst[n] = bp->bif_dlt; 2753 n++; 2754 } 2755 BPF_UNLOCK(); 2756 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); 2757 free(lst, M_TEMP); 2758 BPF_LOCK(); 2759 bfl->bfl_len = n; 2760 return (error); 2761 } 2762 2763 /* 2764 * Set the data link type of a BPF instance. 2765 */ 2766 static int 2767 bpf_setdlt(struct bpf_d *d, u_int dlt) 2768 { 2769 int error, opromisc; 2770 struct ifnet *ifp; 2771 struct bpf_if *bp; 2772 2773 BPF_LOCK_ASSERT(); 2774 2775 if (d->bd_bif->bif_dlt == dlt) 2776 return (0); 2777 ifp = d->bd_bif->bif_ifp; 2778 2779 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2780 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2781 break; 2782 } 2783 2784 if (bp != NULL) { 2785 opromisc = d->bd_promisc; 2786 bpf_attachd(d, bp); 2787 BPFD_LOCK(d); 2788 reset_d(d); 2789 BPFD_UNLOCK(d); 2790 if (opromisc) { 2791 error = ifpromisc(bp->bif_ifp, 1); 2792 if (error) 2793 if_printf(bp->bif_ifp, 2794 "bpf_setdlt: ifpromisc failed (%d)\n", 2795 error); 2796 else 2797 d->bd_promisc = 1; 2798 } 2799 } 2800 return (bp == NULL ? EINVAL : 0); 2801 } 2802 2803 static void 2804 bpf_drvinit(void *unused) 2805 { 2806 struct cdev *dev; 2807 2808 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF); 2809 LIST_INIT(&bpf_iflist); 2810 LIST_INIT(&bpf_freelist); 2811 2812 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2813 /* For compatibility */ 2814 make_dev_alias(dev, "bpf0"); 2815 2816 /* Register interface departure handler */ 2817 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( 2818 ifnet_departure_event, bpf_ifdetach, NULL, 2819 EVENTHANDLER_PRI_ANY); 2820 } 2821 2822 /* 2823 * Zero out the various packet counters associated with all of the bpf 2824 * descriptors. At some point, we will probably want to get a bit more 2825 * granular and allow the user to specify descriptors to be zeroed. 2826 */ 2827 static void 2828 bpf_zero_counters(void) 2829 { 2830 struct bpf_if *bp; 2831 struct bpf_d *bd; 2832 2833 BPF_LOCK(); 2834 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2835 BPFIF_RLOCK(bp); 2836 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2837 BPFD_LOCK(bd); 2838 bd->bd_rcount = 0; 2839 bd->bd_dcount = 0; 2840 bd->bd_fcount = 0; 2841 bd->bd_wcount = 0; 2842 bd->bd_wfcount = 0; 2843 bd->bd_zcopy = 0; 2844 BPFD_UNLOCK(bd); 2845 } 2846 BPFIF_RUNLOCK(bp); 2847 } 2848 BPF_UNLOCK(); 2849 } 2850 2851 /* 2852 * Fill filter statistics 2853 */ 2854 static void 2855 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2856 { 2857 2858 bzero(d, sizeof(*d)); 2859 BPFD_LOCK_ASSERT(bd); 2860 d->bd_structsize = sizeof(*d); 2861 /* XXX: reading should be protected by global lock */ 2862 d->bd_immediate = bd->bd_immediate; 2863 d->bd_promisc = bd->bd_promisc; 2864 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2865 d->bd_direction = bd->bd_direction; 2866 d->bd_feedback = bd->bd_feedback; 2867 d->bd_async = bd->bd_async; 2868 d->bd_rcount = bd->bd_rcount; 2869 d->bd_dcount = bd->bd_dcount; 2870 d->bd_fcount = bd->bd_fcount; 2871 d->bd_sig = bd->bd_sig; 2872 d->bd_slen = bd->bd_slen; 2873 d->bd_hlen = bd->bd_hlen; 2874 d->bd_bufsize = bd->bd_bufsize; 2875 d->bd_pid = bd->bd_pid; 2876 strlcpy(d->bd_ifname, 2877 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2878 d->bd_locked = bd->bd_locked; 2879 d->bd_wcount = bd->bd_wcount; 2880 d->bd_wdcount = bd->bd_wdcount; 2881 d->bd_wfcount = bd->bd_wfcount; 2882 d->bd_zcopy = bd->bd_zcopy; 2883 d->bd_bufmode = bd->bd_bufmode; 2884 } 2885 2886 /* 2887 * Handle `netstat -B' stats request 2888 */ 2889 static int 2890 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2891 { 2892 static const struct xbpf_d zerostats; 2893 struct xbpf_d *xbdbuf, *xbd, tempstats; 2894 int index, error; 2895 struct bpf_if *bp; 2896 struct bpf_d *bd; 2897 2898 /* 2899 * XXX This is not technically correct. It is possible for non 2900 * privileged users to open bpf devices. It would make sense 2901 * if the users who opened the devices were able to retrieve 2902 * the statistics for them, too. 2903 */ 2904 error = priv_check(req->td, PRIV_NET_BPF); 2905 if (error) 2906 return (error); 2907 /* 2908 * Check to see if the user is requesting that the counters be 2909 * zeroed out. Explicitly check that the supplied data is zeroed, 2910 * as we aren't allowing the user to set the counters currently. 2911 */ 2912 if (req->newptr != NULL) { 2913 if (req->newlen != sizeof(tempstats)) 2914 return (EINVAL); 2915 memset(&tempstats, 0, sizeof(tempstats)); 2916 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); 2917 if (error) 2918 return (error); 2919 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) 2920 return (EINVAL); 2921 bpf_zero_counters(); 2922 return (0); 2923 } 2924 if (req->oldptr == NULL) 2925 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2926 if (bpf_bpfd_cnt == 0) 2927 return (SYSCTL_OUT(req, 0, 0)); 2928 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2929 BPF_LOCK(); 2930 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2931 BPF_UNLOCK(); 2932 free(xbdbuf, M_BPF); 2933 return (ENOMEM); 2934 } 2935 index = 0; 2936 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2937 BPFIF_RLOCK(bp); 2938 /* Send writers-only first */ 2939 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2940 xbd = &xbdbuf[index++]; 2941 BPFD_LOCK(bd); 2942 bpfstats_fill_xbpf(xbd, bd); 2943 BPFD_UNLOCK(bd); 2944 } 2945 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2946 xbd = &xbdbuf[index++]; 2947 BPFD_LOCK(bd); 2948 bpfstats_fill_xbpf(xbd, bd); 2949 BPFD_UNLOCK(bd); 2950 } 2951 BPFIF_RUNLOCK(bp); 2952 } 2953 BPF_UNLOCK(); 2954 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2955 free(xbdbuf, M_BPF); 2956 return (error); 2957 } 2958 2959 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2960 2961 #else /* !DEV_BPF && !NETGRAPH_BPF */ 2962 /* 2963 * NOP stubs to allow bpf-using drivers to load and function. 2964 * 2965 * A 'better' implementation would allow the core bpf functionality 2966 * to be loaded at runtime. 2967 */ 2968 static struct bpf_if bp_null; 2969 2970 void 2971 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2972 { 2973 } 2974 2975 void 2976 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2977 { 2978 } 2979 2980 void 2981 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 2982 { 2983 } 2984 2985 void 2986 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2987 { 2988 2989 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2990 } 2991 2992 void 2993 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2994 { 2995 2996 *driverp = &bp_null; 2997 } 2998 2999 void 3000 bpfdetach(struct ifnet *ifp) 3001 { 3002 } 3003 3004 u_int 3005 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 3006 { 3007 return -1; /* "no filter" behaviour */ 3008 } 3009 3010 int 3011 bpf_validate(const struct bpf_insn *f, int len) 3012 { 3013 return 0; /* false */ 3014 } 3015 3016 #endif /* !DEV_BPF && !NETGRAPH_BPF */ 3017 3018 #ifdef DDB 3019 static void 3020 bpf_show_bpf_if(struct bpf_if *bpf_if) 3021 { 3022 3023 if (bpf_if == NULL) 3024 return; 3025 db_printf("%p:\n", bpf_if); 3026 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e); 3027 /* bif_ext.bif_next */ 3028 /* bif_ext.bif_dlist */ 3029 BPF_DB_PRINTF("%#x", bif_dlt); 3030 BPF_DB_PRINTF("%u", bif_hdrlen); 3031 BPF_DB_PRINTF("%p", bif_ifp); 3032 /* bif_lock */ 3033 /* bif_wlist */ 3034 BPF_DB_PRINTF("%#x", bif_flags); 3035 } 3036 3037 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if) 3038 { 3039 3040 if (!have_addr) { 3041 db_printf("usage: show bpf_if <struct bpf_if *>\n"); 3042 return; 3043 } 3044 3045 bpf_show_bpf_if((struct bpf_if *)addr); 3046 } 3047 #endif 3048