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