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_ddb.h" 44 #include "opt_netgraph.h" 45 46 #include <sys/types.h> 47 #include <sys/param.h> 48 #include <sys/lock.h> 49 #include <sys/rwlock.h> 50 #include <sys/systm.h> 51 #include <sys/conf.h> 52 #include <sys/fcntl.h> 53 #include <sys/jail.h> 54 #include <sys/malloc.h> 55 #include <sys/mbuf.h> 56 #include <sys/time.h> 57 #include <sys/priv.h> 58 #include <sys/proc.h> 59 #include <sys/signalvar.h> 60 #include <sys/filio.h> 61 #include <sys/sockio.h> 62 #include <sys/ttycom.h> 63 #include <sys/uio.h> 64 #include <sys/sysent.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 counter_u64_add(d->bd_zcopy, 1); 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 counter_u64_add(d->bd_zcopy, 1); 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 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 /* Setup counters */ 889 d->bd_rcount = counter_u64_alloc(M_WAITOK); 890 d->bd_dcount = counter_u64_alloc(M_WAITOK); 891 d->bd_fcount = counter_u64_alloc(M_WAITOK); 892 d->bd_wcount = counter_u64_alloc(M_WAITOK); 893 d->bd_wfcount = counter_u64_alloc(M_WAITOK); 894 d->bd_wdcount = counter_u64_alloc(M_WAITOK); 895 d->bd_zcopy = counter_u64_alloc(M_WAITOK); 896 897 /* 898 * For historical reasons, perform a one-time initialization call to 899 * the buffer routines, even though we're not yet committed to a 900 * particular buffer method. 901 */ 902 bpf_buffer_init(d); 903 if ((flags & FREAD) == 0) 904 d->bd_writer = 2; 905 d->bd_hbuf_in_use = 0; 906 d->bd_bufmode = BPF_BUFMODE_BUFFER; 907 d->bd_sig = SIGIO; 908 d->bd_direction = BPF_D_INOUT; 909 BPF_PID_REFRESH(d, td); 910 #ifdef MAC 911 mac_bpfdesc_init(d); 912 mac_bpfdesc_create(td->td_ucred, d); 913 #endif 914 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 915 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 916 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 917 918 return (0); 919 } 920 921 /* 922 * bpfread - read next chunk of packets from buffers 923 */ 924 static int 925 bpfread(struct cdev *dev, struct uio *uio, int ioflag) 926 { 927 struct bpf_d *d; 928 int error; 929 int non_block; 930 int timed_out; 931 932 error = devfs_get_cdevpriv((void **)&d); 933 if (error != 0) 934 return (error); 935 936 /* 937 * Restrict application to use a buffer the same size as 938 * as kernel buffers. 939 */ 940 if (uio->uio_resid != d->bd_bufsize) 941 return (EINVAL); 942 943 non_block = ((ioflag & O_NONBLOCK) != 0); 944 945 BPFD_LOCK(d); 946 BPF_PID_REFRESH_CUR(d); 947 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 948 BPFD_UNLOCK(d); 949 return (EOPNOTSUPP); 950 } 951 if (d->bd_state == BPF_WAITING) 952 callout_stop(&d->bd_callout); 953 timed_out = (d->bd_state == BPF_TIMED_OUT); 954 d->bd_state = BPF_IDLE; 955 while (d->bd_hbuf_in_use) { 956 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 957 PRINET|PCATCH, "bd_hbuf", 0); 958 if (error != 0) { 959 BPFD_UNLOCK(d); 960 return (error); 961 } 962 } 963 /* 964 * If the hold buffer is empty, then do a timed sleep, which 965 * ends when the timeout expires or when enough packets 966 * have arrived to fill the store buffer. 967 */ 968 while (d->bd_hbuf == NULL) { 969 if (d->bd_slen != 0) { 970 /* 971 * A packet(s) either arrived since the previous 972 * read or arrived while we were asleep. 973 */ 974 if (d->bd_immediate || non_block || timed_out) { 975 /* 976 * Rotate the buffers and return what's here 977 * if we are in immediate mode, non-blocking 978 * flag is set, or this descriptor timed out. 979 */ 980 ROTATE_BUFFERS(d); 981 break; 982 } 983 } 984 985 /* 986 * No data is available, check to see if the bpf device 987 * is still pointed at a real interface. If not, return 988 * ENXIO so that the userland process knows to rebind 989 * it before using it again. 990 */ 991 if (d->bd_bif == NULL) { 992 BPFD_UNLOCK(d); 993 return (ENXIO); 994 } 995 996 if (non_block) { 997 BPFD_UNLOCK(d); 998 return (EWOULDBLOCK); 999 } 1000 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 1001 "bpf", d->bd_rtout); 1002 if (error == EINTR || error == ERESTART) { 1003 BPFD_UNLOCK(d); 1004 return (error); 1005 } 1006 if (error == EWOULDBLOCK) { 1007 /* 1008 * On a timeout, return what's in the buffer, 1009 * which may be nothing. If there is something 1010 * in the store buffer, we can rotate the buffers. 1011 */ 1012 if (d->bd_hbuf) 1013 /* 1014 * We filled up the buffer in between 1015 * getting the timeout and arriving 1016 * here, so we don't need to rotate. 1017 */ 1018 break; 1019 1020 if (d->bd_slen == 0) { 1021 BPFD_UNLOCK(d); 1022 return (0); 1023 } 1024 ROTATE_BUFFERS(d); 1025 break; 1026 } 1027 } 1028 /* 1029 * At this point, we know we have something in the hold slot. 1030 */ 1031 d->bd_hbuf_in_use = 1; 1032 BPFD_UNLOCK(d); 1033 1034 /* 1035 * Move data from hold buffer into user space. 1036 * We know the entire buffer is transferred since 1037 * we checked above that the read buffer is bpf_bufsize bytes. 1038 * 1039 * We do not have to worry about simultaneous reads because 1040 * we waited for sole access to the hold buffer above. 1041 */ 1042 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 1043 1044 BPFD_LOCK(d); 1045 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 1046 d->bd_fbuf = d->bd_hbuf; 1047 d->bd_hbuf = NULL; 1048 d->bd_hlen = 0; 1049 bpf_buf_reclaimed(d); 1050 d->bd_hbuf_in_use = 0; 1051 wakeup(&d->bd_hbuf_in_use); 1052 BPFD_UNLOCK(d); 1053 1054 return (error); 1055 } 1056 1057 /* 1058 * If there are processes sleeping on this descriptor, wake them up. 1059 */ 1060 static __inline void 1061 bpf_wakeup(struct bpf_d *d) 1062 { 1063 1064 BPFD_LOCK_ASSERT(d); 1065 if (d->bd_state == BPF_WAITING) { 1066 callout_stop(&d->bd_callout); 1067 d->bd_state = BPF_IDLE; 1068 } 1069 wakeup(d); 1070 if (d->bd_async && d->bd_sig && d->bd_sigio) 1071 pgsigio(&d->bd_sigio, d->bd_sig, 0); 1072 1073 selwakeuppri(&d->bd_sel, PRINET); 1074 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 1075 } 1076 1077 static void 1078 bpf_timed_out(void *arg) 1079 { 1080 struct bpf_d *d = (struct bpf_d *)arg; 1081 1082 BPFD_LOCK_ASSERT(d); 1083 1084 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) 1085 return; 1086 if (d->bd_state == BPF_WAITING) { 1087 d->bd_state = BPF_TIMED_OUT; 1088 if (d->bd_slen != 0) 1089 bpf_wakeup(d); 1090 } 1091 } 1092 1093 static int 1094 bpf_ready(struct bpf_d *d) 1095 { 1096 1097 BPFD_LOCK_ASSERT(d); 1098 1099 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1100 return (1); 1101 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1102 d->bd_slen != 0) 1103 return (1); 1104 return (0); 1105 } 1106 1107 static int 1108 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1109 { 1110 struct bpf_d *d; 1111 struct ifnet *ifp; 1112 struct mbuf *m, *mc; 1113 struct sockaddr dst; 1114 struct route ro; 1115 int error, hlen; 1116 1117 error = devfs_get_cdevpriv((void **)&d); 1118 if (error != 0) 1119 return (error); 1120 1121 BPF_PID_REFRESH_CUR(d); 1122 counter_u64_add(d->bd_wcount, 1); 1123 /* XXX: locking required */ 1124 if (d->bd_bif == NULL) { 1125 counter_u64_add(d->bd_wdcount, 1); 1126 return (ENXIO); 1127 } 1128 1129 ifp = d->bd_bif->bif_ifp; 1130 1131 if ((ifp->if_flags & IFF_UP) == 0) { 1132 counter_u64_add(d->bd_wdcount, 1); 1133 return (ENETDOWN); 1134 } 1135 1136 if (uio->uio_resid == 0) { 1137 counter_u64_add(d->bd_wdcount, 1); 1138 return (0); 1139 } 1140 1141 bzero(&dst, sizeof(dst)); 1142 m = NULL; 1143 hlen = 0; 1144 /* XXX: bpf_movein() can sleep */ 1145 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, 1146 &m, &dst, &hlen, d); 1147 if (error) { 1148 counter_u64_add(d->bd_wdcount, 1); 1149 return (error); 1150 } 1151 counter_u64_add(d->bd_wfcount, 1); 1152 if (d->bd_hdrcmplt) 1153 dst.sa_family = pseudo_AF_HDRCMPLT; 1154 1155 if (d->bd_feedback) { 1156 mc = m_dup(m, M_NOWAIT); 1157 if (mc != NULL) 1158 mc->m_pkthdr.rcvif = ifp; 1159 /* Set M_PROMISC for outgoing packets to be discarded. */ 1160 if (d->bd_direction == BPF_D_INOUT) 1161 m->m_flags |= M_PROMISC; 1162 } else 1163 mc = NULL; 1164 1165 m->m_pkthdr.len -= hlen; 1166 m->m_len -= hlen; 1167 m->m_data += hlen; /* XXX */ 1168 1169 CURVNET_SET(ifp->if_vnet); 1170 #ifdef MAC 1171 BPFD_LOCK(d); 1172 mac_bpfdesc_create_mbuf(d, m); 1173 if (mc != NULL) 1174 mac_bpfdesc_create_mbuf(d, mc); 1175 BPFD_UNLOCK(d); 1176 #endif 1177 1178 bzero(&ro, sizeof(ro)); 1179 if (hlen != 0) { 1180 ro.ro_prepend = (u_char *)&dst.sa_data; 1181 ro.ro_plen = hlen; 1182 ro.ro_flags = RT_HAS_HEADER; 1183 } 1184 1185 error = (*ifp->if_output)(ifp, m, &dst, &ro); 1186 if (error) 1187 counter_u64_add(d->bd_wdcount, 1); 1188 1189 if (mc != NULL) { 1190 if (error == 0) 1191 (*ifp->if_input)(ifp, mc); 1192 else 1193 m_freem(mc); 1194 } 1195 CURVNET_RESTORE(); 1196 1197 return (error); 1198 } 1199 1200 /* 1201 * Reset a descriptor by flushing its packet buffer and clearing the receive 1202 * and drop counts. This is doable for kernel-only buffers, but with 1203 * zero-copy buffers, we can't write to (or rotate) buffers that are 1204 * currently owned by userspace. It would be nice if we could encapsulate 1205 * this logic in the buffer code rather than here. 1206 */ 1207 static void 1208 reset_d(struct bpf_d *d) 1209 { 1210 1211 BPFD_LOCK_ASSERT(d); 1212 1213 while (d->bd_hbuf_in_use) 1214 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1215 "bd_hbuf", 0); 1216 if ((d->bd_hbuf != NULL) && 1217 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1218 /* Free the hold buffer. */ 1219 d->bd_fbuf = d->bd_hbuf; 1220 d->bd_hbuf = NULL; 1221 d->bd_hlen = 0; 1222 bpf_buf_reclaimed(d); 1223 } 1224 if (bpf_canwritebuf(d)) 1225 d->bd_slen = 0; 1226 counter_u64_zero(d->bd_rcount); 1227 counter_u64_zero(d->bd_dcount); 1228 counter_u64_zero(d->bd_fcount); 1229 counter_u64_zero(d->bd_wcount); 1230 counter_u64_zero(d->bd_wfcount); 1231 counter_u64_zero(d->bd_wdcount); 1232 counter_u64_zero(d->bd_zcopy); 1233 } 1234 1235 /* 1236 * FIONREAD Check for read packet available. 1237 * BIOCGBLEN Get buffer len [for read()]. 1238 * BIOCSETF Set read filter. 1239 * BIOCSETFNR Set read filter without resetting descriptor. 1240 * BIOCSETWF Set write filter. 1241 * BIOCFLUSH Flush read packet buffer. 1242 * BIOCPROMISC Put interface into promiscuous mode. 1243 * BIOCGDLT Get link layer type. 1244 * BIOCGETIF Get interface name. 1245 * BIOCSETIF Set interface. 1246 * BIOCSRTIMEOUT Set read timeout. 1247 * BIOCGRTIMEOUT Get read timeout. 1248 * BIOCGSTATS Get packet stats. 1249 * BIOCIMMEDIATE Set immediate mode. 1250 * BIOCVERSION Get filter language version. 1251 * BIOCGHDRCMPLT Get "header already complete" flag 1252 * BIOCSHDRCMPLT Set "header already complete" flag 1253 * BIOCGDIRECTION Get packet direction flag 1254 * BIOCSDIRECTION Set packet direction flag 1255 * BIOCGTSTAMP Get time stamp format and resolution. 1256 * BIOCSTSTAMP Set time stamp format and resolution. 1257 * BIOCLOCK Set "locked" flag 1258 * BIOCFEEDBACK Set packet feedback mode. 1259 * BIOCSETZBUF Set current zero-copy buffer locations. 1260 * BIOCGETZMAX Get maximum zero-copy buffer size. 1261 * BIOCROTZBUF Force rotation of zero-copy buffer 1262 * BIOCSETBUFMODE Set buffer mode. 1263 * BIOCGETBUFMODE Get current buffer mode. 1264 */ 1265 /* ARGSUSED */ 1266 static int 1267 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1268 struct thread *td) 1269 { 1270 struct bpf_d *d; 1271 int error; 1272 1273 error = devfs_get_cdevpriv((void **)&d); 1274 if (error != 0) 1275 return (error); 1276 1277 /* 1278 * Refresh PID associated with this descriptor. 1279 */ 1280 BPFD_LOCK(d); 1281 BPF_PID_REFRESH(d, td); 1282 if (d->bd_state == BPF_WAITING) 1283 callout_stop(&d->bd_callout); 1284 d->bd_state = BPF_IDLE; 1285 BPFD_UNLOCK(d); 1286 1287 if (d->bd_locked == 1) { 1288 switch (cmd) { 1289 case BIOCGBLEN: 1290 case BIOCFLUSH: 1291 case BIOCGDLT: 1292 case BIOCGDLTLIST: 1293 #ifdef COMPAT_FREEBSD32 1294 case BIOCGDLTLIST32: 1295 #endif 1296 case BIOCGETIF: 1297 case BIOCGRTIMEOUT: 1298 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1299 case BIOCGRTIMEOUT32: 1300 #endif 1301 case BIOCGSTATS: 1302 case BIOCVERSION: 1303 case BIOCGRSIG: 1304 case BIOCGHDRCMPLT: 1305 case BIOCSTSTAMP: 1306 case BIOCFEEDBACK: 1307 case FIONREAD: 1308 case BIOCLOCK: 1309 case BIOCSRTIMEOUT: 1310 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1311 case BIOCSRTIMEOUT32: 1312 #endif 1313 case BIOCIMMEDIATE: 1314 case TIOCGPGRP: 1315 case BIOCROTZBUF: 1316 break; 1317 default: 1318 return (EPERM); 1319 } 1320 } 1321 #ifdef COMPAT_FREEBSD32 1322 /* 1323 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1324 * that it will get 32-bit packet headers. 1325 */ 1326 switch (cmd) { 1327 case BIOCSETF32: 1328 case BIOCSETFNR32: 1329 case BIOCSETWF32: 1330 case BIOCGDLTLIST32: 1331 case BIOCGRTIMEOUT32: 1332 case BIOCSRTIMEOUT32: 1333 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 1334 BPFD_LOCK(d); 1335 d->bd_compat32 = 1; 1336 BPFD_UNLOCK(d); 1337 } 1338 } 1339 #endif 1340 1341 CURVNET_SET(TD_TO_VNET(td)); 1342 switch (cmd) { 1343 1344 default: 1345 error = EINVAL; 1346 break; 1347 1348 /* 1349 * Check for read packet available. 1350 */ 1351 case FIONREAD: 1352 { 1353 int n; 1354 1355 BPFD_LOCK(d); 1356 n = d->bd_slen; 1357 while (d->bd_hbuf_in_use) 1358 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1359 PRINET, "bd_hbuf", 0); 1360 if (d->bd_hbuf) 1361 n += d->bd_hlen; 1362 BPFD_UNLOCK(d); 1363 1364 *(int *)addr = n; 1365 break; 1366 } 1367 1368 /* 1369 * Get buffer len [for read()]. 1370 */ 1371 case BIOCGBLEN: 1372 BPFD_LOCK(d); 1373 *(u_int *)addr = d->bd_bufsize; 1374 BPFD_UNLOCK(d); 1375 break; 1376 1377 /* 1378 * Set buffer length. 1379 */ 1380 case BIOCSBLEN: 1381 error = bpf_ioctl_sblen(d, (u_int *)addr); 1382 break; 1383 1384 /* 1385 * Set link layer read filter. 1386 */ 1387 case BIOCSETF: 1388 case BIOCSETFNR: 1389 case BIOCSETWF: 1390 #ifdef COMPAT_FREEBSD32 1391 case BIOCSETF32: 1392 case BIOCSETFNR32: 1393 case BIOCSETWF32: 1394 #endif 1395 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1396 break; 1397 1398 /* 1399 * Flush read packet buffer. 1400 */ 1401 case BIOCFLUSH: 1402 BPFD_LOCK(d); 1403 reset_d(d); 1404 BPFD_UNLOCK(d); 1405 break; 1406 1407 /* 1408 * Put interface into promiscuous mode. 1409 */ 1410 case BIOCPROMISC: 1411 if (d->bd_bif == NULL) { 1412 /* 1413 * No interface attached yet. 1414 */ 1415 error = EINVAL; 1416 break; 1417 } 1418 if (d->bd_promisc == 0) { 1419 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1420 if (error == 0) 1421 d->bd_promisc = 1; 1422 } 1423 break; 1424 1425 /* 1426 * Get current data link type. 1427 */ 1428 case BIOCGDLT: 1429 BPF_LOCK(); 1430 if (d->bd_bif == NULL) 1431 error = EINVAL; 1432 else 1433 *(u_int *)addr = d->bd_bif->bif_dlt; 1434 BPF_UNLOCK(); 1435 break; 1436 1437 /* 1438 * Get a list of supported data link types. 1439 */ 1440 #ifdef COMPAT_FREEBSD32 1441 case BIOCGDLTLIST32: 1442 { 1443 struct bpf_dltlist32 *list32; 1444 struct bpf_dltlist dltlist; 1445 1446 list32 = (struct bpf_dltlist32 *)addr; 1447 dltlist.bfl_len = list32->bfl_len; 1448 dltlist.bfl_list = PTRIN(list32->bfl_list); 1449 BPF_LOCK(); 1450 if (d->bd_bif == NULL) 1451 error = EINVAL; 1452 else { 1453 error = bpf_getdltlist(d, &dltlist); 1454 if (error == 0) 1455 list32->bfl_len = dltlist.bfl_len; 1456 } 1457 BPF_UNLOCK(); 1458 break; 1459 } 1460 #endif 1461 1462 case BIOCGDLTLIST: 1463 BPF_LOCK(); 1464 if (d->bd_bif == NULL) 1465 error = EINVAL; 1466 else 1467 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1468 BPF_UNLOCK(); 1469 break; 1470 1471 /* 1472 * Set data link type. 1473 */ 1474 case BIOCSDLT: 1475 BPF_LOCK(); 1476 if (d->bd_bif == NULL) 1477 error = EINVAL; 1478 else 1479 error = bpf_setdlt(d, *(u_int *)addr); 1480 BPF_UNLOCK(); 1481 break; 1482 1483 /* 1484 * Get interface name. 1485 */ 1486 case BIOCGETIF: 1487 BPF_LOCK(); 1488 if (d->bd_bif == NULL) 1489 error = EINVAL; 1490 else { 1491 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1492 struct ifreq *const ifr = (struct ifreq *)addr; 1493 1494 strlcpy(ifr->ifr_name, ifp->if_xname, 1495 sizeof(ifr->ifr_name)); 1496 } 1497 BPF_UNLOCK(); 1498 break; 1499 1500 /* 1501 * Set interface. 1502 */ 1503 case BIOCSETIF: 1504 { 1505 int alloc_buf, size; 1506 1507 /* 1508 * Behavior here depends on the buffering model. If 1509 * we're using kernel memory buffers, then we can 1510 * allocate them here. If we're using zero-copy, 1511 * then the user process must have registered buffers 1512 * by the time we get here. 1513 */ 1514 alloc_buf = 0; 1515 BPFD_LOCK(d); 1516 if (d->bd_bufmode == BPF_BUFMODE_BUFFER && 1517 d->bd_sbuf == NULL) 1518 alloc_buf = 1; 1519 BPFD_UNLOCK(d); 1520 if (alloc_buf) { 1521 size = d->bd_bufsize; 1522 error = bpf_buffer_ioctl_sblen(d, &size); 1523 if (error != 0) 1524 break; 1525 } 1526 BPF_LOCK(); 1527 error = bpf_setif(d, (struct ifreq *)addr); 1528 BPF_UNLOCK(); 1529 break; 1530 } 1531 1532 /* 1533 * Set read timeout. 1534 */ 1535 case BIOCSRTIMEOUT: 1536 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1537 case BIOCSRTIMEOUT32: 1538 #endif 1539 { 1540 struct timeval *tv = (struct timeval *)addr; 1541 #if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1542 struct timeval32 *tv32; 1543 struct timeval tv64; 1544 1545 if (cmd == BIOCSRTIMEOUT32) { 1546 tv32 = (struct timeval32 *)addr; 1547 tv = &tv64; 1548 tv->tv_sec = tv32->tv_sec; 1549 tv->tv_usec = tv32->tv_usec; 1550 } else 1551 #endif 1552 tv = (struct timeval *)addr; 1553 1554 /* 1555 * Subtract 1 tick from tvtohz() since this isn't 1556 * a one-shot timer. 1557 */ 1558 if ((error = itimerfix(tv)) == 0) 1559 d->bd_rtout = tvtohz(tv) - 1; 1560 break; 1561 } 1562 1563 /* 1564 * Get read timeout. 1565 */ 1566 case BIOCGRTIMEOUT: 1567 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1568 case BIOCGRTIMEOUT32: 1569 #endif 1570 { 1571 struct timeval *tv; 1572 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1573 struct timeval32 *tv32; 1574 struct timeval tv64; 1575 1576 if (cmd == BIOCGRTIMEOUT32) 1577 tv = &tv64; 1578 else 1579 #endif 1580 tv = (struct timeval *)addr; 1581 1582 tv->tv_sec = d->bd_rtout / hz; 1583 tv->tv_usec = (d->bd_rtout % hz) * tick; 1584 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1585 if (cmd == BIOCGRTIMEOUT32) { 1586 tv32 = (struct timeval32 *)addr; 1587 tv32->tv_sec = tv->tv_sec; 1588 tv32->tv_usec = tv->tv_usec; 1589 } 1590 #endif 1591 1592 break; 1593 } 1594 1595 /* 1596 * Get packet stats. 1597 */ 1598 case BIOCGSTATS: 1599 { 1600 struct bpf_stat *bs = (struct bpf_stat *)addr; 1601 1602 /* XXXCSJP overflow */ 1603 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount); 1604 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount); 1605 break; 1606 } 1607 1608 /* 1609 * Set immediate mode. 1610 */ 1611 case BIOCIMMEDIATE: 1612 BPFD_LOCK(d); 1613 d->bd_immediate = *(u_int *)addr; 1614 BPFD_UNLOCK(d); 1615 break; 1616 1617 case BIOCVERSION: 1618 { 1619 struct bpf_version *bv = (struct bpf_version *)addr; 1620 1621 bv->bv_major = BPF_MAJOR_VERSION; 1622 bv->bv_minor = BPF_MINOR_VERSION; 1623 break; 1624 } 1625 1626 /* 1627 * Get "header already complete" flag 1628 */ 1629 case BIOCGHDRCMPLT: 1630 BPFD_LOCK(d); 1631 *(u_int *)addr = d->bd_hdrcmplt; 1632 BPFD_UNLOCK(d); 1633 break; 1634 1635 /* 1636 * Set "header already complete" flag 1637 */ 1638 case BIOCSHDRCMPLT: 1639 BPFD_LOCK(d); 1640 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1641 BPFD_UNLOCK(d); 1642 break; 1643 1644 /* 1645 * Get packet direction flag 1646 */ 1647 case BIOCGDIRECTION: 1648 BPFD_LOCK(d); 1649 *(u_int *)addr = d->bd_direction; 1650 BPFD_UNLOCK(d); 1651 break; 1652 1653 /* 1654 * Set packet direction flag 1655 */ 1656 case BIOCSDIRECTION: 1657 { 1658 u_int direction; 1659 1660 direction = *(u_int *)addr; 1661 switch (direction) { 1662 case BPF_D_IN: 1663 case BPF_D_INOUT: 1664 case BPF_D_OUT: 1665 BPFD_LOCK(d); 1666 d->bd_direction = direction; 1667 BPFD_UNLOCK(d); 1668 break; 1669 default: 1670 error = EINVAL; 1671 } 1672 } 1673 break; 1674 1675 /* 1676 * Get packet timestamp format and resolution. 1677 */ 1678 case BIOCGTSTAMP: 1679 BPFD_LOCK(d); 1680 *(u_int *)addr = d->bd_tstamp; 1681 BPFD_UNLOCK(d); 1682 break; 1683 1684 /* 1685 * Set packet timestamp format and resolution. 1686 */ 1687 case BIOCSTSTAMP: 1688 { 1689 u_int func; 1690 1691 func = *(u_int *)addr; 1692 if (BPF_T_VALID(func)) 1693 d->bd_tstamp = func; 1694 else 1695 error = EINVAL; 1696 } 1697 break; 1698 1699 case BIOCFEEDBACK: 1700 BPFD_LOCK(d); 1701 d->bd_feedback = *(u_int *)addr; 1702 BPFD_UNLOCK(d); 1703 break; 1704 1705 case BIOCLOCK: 1706 BPFD_LOCK(d); 1707 d->bd_locked = 1; 1708 BPFD_UNLOCK(d); 1709 break; 1710 1711 case FIONBIO: /* Non-blocking I/O */ 1712 break; 1713 1714 case FIOASYNC: /* Send signal on receive packets */ 1715 BPFD_LOCK(d); 1716 d->bd_async = *(int *)addr; 1717 BPFD_UNLOCK(d); 1718 break; 1719 1720 case FIOSETOWN: 1721 /* 1722 * XXX: Add some sort of locking here? 1723 * fsetown() can sleep. 1724 */ 1725 error = fsetown(*(int *)addr, &d->bd_sigio); 1726 break; 1727 1728 case FIOGETOWN: 1729 BPFD_LOCK(d); 1730 *(int *)addr = fgetown(&d->bd_sigio); 1731 BPFD_UNLOCK(d); 1732 break; 1733 1734 /* This is deprecated, FIOSETOWN should be used instead. */ 1735 case TIOCSPGRP: 1736 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1737 break; 1738 1739 /* This is deprecated, FIOGETOWN should be used instead. */ 1740 case TIOCGPGRP: 1741 *(int *)addr = -fgetown(&d->bd_sigio); 1742 break; 1743 1744 case BIOCSRSIG: /* Set receive signal */ 1745 { 1746 u_int sig; 1747 1748 sig = *(u_int *)addr; 1749 1750 if (sig >= NSIG) 1751 error = EINVAL; 1752 else { 1753 BPFD_LOCK(d); 1754 d->bd_sig = sig; 1755 BPFD_UNLOCK(d); 1756 } 1757 break; 1758 } 1759 case BIOCGRSIG: 1760 BPFD_LOCK(d); 1761 *(u_int *)addr = d->bd_sig; 1762 BPFD_UNLOCK(d); 1763 break; 1764 1765 case BIOCGETBUFMODE: 1766 BPFD_LOCK(d); 1767 *(u_int *)addr = d->bd_bufmode; 1768 BPFD_UNLOCK(d); 1769 break; 1770 1771 case BIOCSETBUFMODE: 1772 /* 1773 * Allow the buffering mode to be changed as long as we 1774 * haven't yet committed to a particular mode. Our 1775 * definition of commitment, for now, is whether or not a 1776 * buffer has been allocated or an interface attached, since 1777 * that's the point where things get tricky. 1778 */ 1779 switch (*(u_int *)addr) { 1780 case BPF_BUFMODE_BUFFER: 1781 break; 1782 1783 case BPF_BUFMODE_ZBUF: 1784 if (bpf_zerocopy_enable) 1785 break; 1786 /* FALLSTHROUGH */ 1787 1788 default: 1789 CURVNET_RESTORE(); 1790 return (EINVAL); 1791 } 1792 1793 BPFD_LOCK(d); 1794 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1795 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1796 BPFD_UNLOCK(d); 1797 CURVNET_RESTORE(); 1798 return (EBUSY); 1799 } 1800 d->bd_bufmode = *(u_int *)addr; 1801 BPFD_UNLOCK(d); 1802 break; 1803 1804 case BIOCGETZMAX: 1805 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1806 break; 1807 1808 case BIOCSETZBUF: 1809 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1810 break; 1811 1812 case BIOCROTZBUF: 1813 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1814 break; 1815 } 1816 CURVNET_RESTORE(); 1817 return (error); 1818 } 1819 1820 /* 1821 * Set d's packet filter program to fp. If this file already has a filter, 1822 * free it and replace it. Returns EINVAL for bogus requests. 1823 * 1824 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls 1825 * since reading d->bd_bif can't be protected by d or interface lock due to 1826 * lock order. 1827 * 1828 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses 1829 * interface read lock to read all filers. 1830 * 1831 */ 1832 static int 1833 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1834 { 1835 #ifdef COMPAT_FREEBSD32 1836 struct bpf_program fp_swab; 1837 struct bpf_program32 *fp32; 1838 #endif 1839 struct bpf_insn *fcode, *old; 1840 #ifdef BPF_JITTER 1841 bpf_jit_filter *jfunc, *ofunc; 1842 #endif 1843 size_t size; 1844 u_int flen; 1845 int need_upgrade; 1846 1847 #ifdef COMPAT_FREEBSD32 1848 switch (cmd) { 1849 case BIOCSETF32: 1850 case BIOCSETWF32: 1851 case BIOCSETFNR32: 1852 fp32 = (struct bpf_program32 *)fp; 1853 fp_swab.bf_len = fp32->bf_len; 1854 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1855 fp = &fp_swab; 1856 switch (cmd) { 1857 case BIOCSETF32: 1858 cmd = BIOCSETF; 1859 break; 1860 case BIOCSETWF32: 1861 cmd = BIOCSETWF; 1862 break; 1863 } 1864 break; 1865 } 1866 #endif 1867 1868 fcode = NULL; 1869 #ifdef BPF_JITTER 1870 jfunc = ofunc = NULL; 1871 #endif 1872 need_upgrade = 0; 1873 1874 /* 1875 * Check new filter validness before acquiring any locks. 1876 * Allocate memory for new filter, if needed. 1877 */ 1878 flen = fp->bf_len; 1879 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1880 return (EINVAL); 1881 size = flen * sizeof(*fp->bf_insns); 1882 if (size > 0) { 1883 /* We're setting up new filter. Copy and check actual data. */ 1884 fcode = malloc(size, M_BPF, M_WAITOK); 1885 if (copyin(fp->bf_insns, fcode, size) != 0 || 1886 !bpf_validate(fcode, flen)) { 1887 free(fcode, M_BPF); 1888 return (EINVAL); 1889 } 1890 #ifdef BPF_JITTER 1891 /* Filter is copied inside fcode and is perfectly valid. */ 1892 jfunc = bpf_jitter(fcode, flen); 1893 #endif 1894 } 1895 1896 BPF_LOCK(); 1897 1898 /* 1899 * Set up new filter. 1900 * Protect filter change by interface lock. 1901 * Additionally, we are protected by global lock here. 1902 */ 1903 if (d->bd_bif != NULL) 1904 BPFIF_WLOCK(d->bd_bif); 1905 BPFD_LOCK(d); 1906 if (cmd == BIOCSETWF) { 1907 old = d->bd_wfilter; 1908 d->bd_wfilter = fcode; 1909 } else { 1910 old = d->bd_rfilter; 1911 d->bd_rfilter = fcode; 1912 #ifdef BPF_JITTER 1913 ofunc = d->bd_bfilter; 1914 d->bd_bfilter = jfunc; 1915 #endif 1916 if (cmd == BIOCSETF) 1917 reset_d(d); 1918 1919 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen); 1920 } 1921 BPFD_UNLOCK(d); 1922 if (d->bd_bif != NULL) 1923 BPFIF_WUNLOCK(d->bd_bif); 1924 if (old != NULL) 1925 free(old, M_BPF); 1926 #ifdef BPF_JITTER 1927 if (ofunc != NULL) 1928 bpf_destroy_jit_filter(ofunc); 1929 #endif 1930 1931 /* Move d to active readers list. */ 1932 if (need_upgrade != 0) 1933 bpf_upgraded(d); 1934 1935 BPF_UNLOCK(); 1936 return (0); 1937 } 1938 1939 /* 1940 * Detach a file from its current interface (if attached at all) and attach 1941 * to the interface indicated by the name stored in ifr. 1942 * Return an errno or 0. 1943 */ 1944 static int 1945 bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1946 { 1947 struct bpf_if *bp; 1948 struct ifnet *theywant; 1949 1950 BPF_LOCK_ASSERT(); 1951 1952 theywant = ifunit(ifr->ifr_name); 1953 if (theywant == NULL || theywant->if_bpf == NULL) 1954 return (ENXIO); 1955 1956 bp = theywant->if_bpf; 1957 1958 /* Check if interface is not being detached from BPF */ 1959 BPFIF_RLOCK(bp); 1960 if (bp->bif_flags & BPFIF_FLAG_DYING) { 1961 BPFIF_RUNLOCK(bp); 1962 return (ENXIO); 1963 } 1964 BPFIF_RUNLOCK(bp); 1965 1966 /* 1967 * At this point, we expect the buffer is already allocated. If not, 1968 * return an error. 1969 */ 1970 switch (d->bd_bufmode) { 1971 case BPF_BUFMODE_BUFFER: 1972 case BPF_BUFMODE_ZBUF: 1973 if (d->bd_sbuf == NULL) 1974 return (EINVAL); 1975 break; 1976 1977 default: 1978 panic("bpf_setif: bufmode %d", d->bd_bufmode); 1979 } 1980 if (bp != d->bd_bif) 1981 bpf_attachd(d, bp); 1982 BPFD_LOCK(d); 1983 reset_d(d); 1984 BPFD_UNLOCK(d); 1985 return (0); 1986 } 1987 1988 /* 1989 * Support for select() and poll() system calls 1990 * 1991 * Return true iff the specific operation will not block indefinitely. 1992 * Otherwise, return false but make a note that a selwakeup() must be done. 1993 */ 1994 static int 1995 bpfpoll(struct cdev *dev, int events, struct thread *td) 1996 { 1997 struct bpf_d *d; 1998 int revents; 1999 2000 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 2001 return (events & 2002 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 2003 2004 /* 2005 * Refresh PID associated with this descriptor. 2006 */ 2007 revents = events & (POLLOUT | POLLWRNORM); 2008 BPFD_LOCK(d); 2009 BPF_PID_REFRESH(d, td); 2010 if (events & (POLLIN | POLLRDNORM)) { 2011 if (bpf_ready(d)) 2012 revents |= events & (POLLIN | POLLRDNORM); 2013 else { 2014 selrecord(td, &d->bd_sel); 2015 /* Start the read timeout if necessary. */ 2016 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2017 callout_reset(&d->bd_callout, d->bd_rtout, 2018 bpf_timed_out, d); 2019 d->bd_state = BPF_WAITING; 2020 } 2021 } 2022 } 2023 BPFD_UNLOCK(d); 2024 return (revents); 2025 } 2026 2027 /* 2028 * Support for kevent() system call. Register EVFILT_READ filters and 2029 * reject all others. 2030 */ 2031 int 2032 bpfkqfilter(struct cdev *dev, struct knote *kn) 2033 { 2034 struct bpf_d *d; 2035 2036 if (devfs_get_cdevpriv((void **)&d) != 0 || 2037 kn->kn_filter != EVFILT_READ) 2038 return (1); 2039 2040 /* 2041 * Refresh PID associated with this descriptor. 2042 */ 2043 BPFD_LOCK(d); 2044 BPF_PID_REFRESH_CUR(d); 2045 kn->kn_fop = &bpfread_filtops; 2046 kn->kn_hook = d; 2047 knlist_add(&d->bd_sel.si_note, kn, 1); 2048 BPFD_UNLOCK(d); 2049 2050 return (0); 2051 } 2052 2053 static void 2054 filt_bpfdetach(struct knote *kn) 2055 { 2056 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2057 2058 knlist_remove(&d->bd_sel.si_note, kn, 0); 2059 } 2060 2061 static int 2062 filt_bpfread(struct knote *kn, long hint) 2063 { 2064 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2065 int ready; 2066 2067 BPFD_LOCK_ASSERT(d); 2068 ready = bpf_ready(d); 2069 if (ready) { 2070 kn->kn_data = d->bd_slen; 2071 /* 2072 * Ignore the hold buffer if it is being copied to user space. 2073 */ 2074 if (!d->bd_hbuf_in_use && d->bd_hbuf) 2075 kn->kn_data += d->bd_hlen; 2076 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2077 callout_reset(&d->bd_callout, d->bd_rtout, 2078 bpf_timed_out, d); 2079 d->bd_state = BPF_WAITING; 2080 } 2081 2082 return (ready); 2083 } 2084 2085 #define BPF_TSTAMP_NONE 0 2086 #define BPF_TSTAMP_FAST 1 2087 #define BPF_TSTAMP_NORMAL 2 2088 #define BPF_TSTAMP_EXTERN 3 2089 2090 static int 2091 bpf_ts_quality(int tstype) 2092 { 2093 2094 if (tstype == BPF_T_NONE) 2095 return (BPF_TSTAMP_NONE); 2096 if ((tstype & BPF_T_FAST) != 0) 2097 return (BPF_TSTAMP_FAST); 2098 2099 return (BPF_TSTAMP_NORMAL); 2100 } 2101 2102 static int 2103 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2104 { 2105 struct m_tag *tag; 2106 int quality; 2107 2108 quality = bpf_ts_quality(tstype); 2109 if (quality == BPF_TSTAMP_NONE) 2110 return (quality); 2111 2112 if (m != NULL) { 2113 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2114 if (tag != NULL) { 2115 *bt = *(struct bintime *)(tag + 1); 2116 return (BPF_TSTAMP_EXTERN); 2117 } 2118 } 2119 if (quality == BPF_TSTAMP_NORMAL) 2120 binuptime(bt); 2121 else 2122 getbinuptime(bt); 2123 2124 return (quality); 2125 } 2126 2127 /* 2128 * Incoming linkage from device drivers. Process the packet pkt, of length 2129 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2130 * by each process' filter, and if accepted, stashed into the corresponding 2131 * buffer. 2132 */ 2133 void 2134 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2135 { 2136 struct bintime bt; 2137 struct bpf_d *d; 2138 #ifdef BPF_JITTER 2139 bpf_jit_filter *bf; 2140 #endif 2141 u_int slen; 2142 int gottime; 2143 2144 gottime = BPF_TSTAMP_NONE; 2145 2146 BPFIF_RLOCK(bp); 2147 2148 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2149 /* 2150 * We are not using any locks for d here because: 2151 * 1) any filter change is protected by interface 2152 * write lock 2153 * 2) destroying/detaching d is protected by interface 2154 * write lock, too 2155 */ 2156 2157 counter_u64_add(d->bd_rcount, 1); 2158 /* 2159 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no 2160 * way for the caller to indiciate to us whether this packet 2161 * is inbound or outbound. In the bpf_mtap() routines, we use 2162 * the interface pointers on the mbuf to figure it out. 2163 */ 2164 #ifdef BPF_JITTER 2165 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2166 if (bf != NULL) 2167 slen = (*(bf->func))(pkt, pktlen, pktlen); 2168 else 2169 #endif 2170 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2171 if (slen != 0) { 2172 /* 2173 * Filter matches. Let's to acquire write lock. 2174 */ 2175 BPFD_LOCK(d); 2176 2177 counter_u64_add(d->bd_fcount, 1); 2178 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2179 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); 2180 #ifdef MAC 2181 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2182 #endif 2183 catchpacket(d, pkt, pktlen, slen, 2184 bpf_append_bytes, &bt); 2185 BPFD_UNLOCK(d); 2186 } 2187 } 2188 BPFIF_RUNLOCK(bp); 2189 } 2190 2191 #define BPF_CHECK_DIRECTION(d, r, i) \ 2192 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2193 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2194 2195 /* 2196 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2197 * Locking model is explained in bpf_tap(). 2198 */ 2199 void 2200 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2201 { 2202 struct bintime bt; 2203 struct bpf_d *d; 2204 #ifdef BPF_JITTER 2205 bpf_jit_filter *bf; 2206 #endif 2207 u_int pktlen, slen; 2208 int gottime; 2209 2210 /* Skip outgoing duplicate packets. */ 2211 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2212 m->m_flags &= ~M_PROMISC; 2213 return; 2214 } 2215 2216 pktlen = m_length(m, NULL); 2217 gottime = BPF_TSTAMP_NONE; 2218 2219 BPFIF_RLOCK(bp); 2220 2221 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2222 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2223 continue; 2224 counter_u64_add(d->bd_rcount, 1); 2225 #ifdef BPF_JITTER 2226 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2227 /* XXX We cannot handle multiple mbufs. */ 2228 if (bf != NULL && m->m_next == NULL) 2229 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); 2230 else 2231 #endif 2232 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2233 if (slen != 0) { 2234 BPFD_LOCK(d); 2235 2236 counter_u64_add(d->bd_fcount, 1); 2237 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2238 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2239 #ifdef MAC 2240 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2241 #endif 2242 catchpacket(d, (u_char *)m, pktlen, slen, 2243 bpf_append_mbuf, &bt); 2244 BPFD_UNLOCK(d); 2245 } 2246 } 2247 BPFIF_RUNLOCK(bp); 2248 } 2249 2250 /* 2251 * Incoming linkage from device drivers, when packet is in 2252 * an mbuf chain and to be prepended by a contiguous header. 2253 */ 2254 void 2255 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2256 { 2257 struct bintime bt; 2258 struct mbuf mb; 2259 struct bpf_d *d; 2260 u_int pktlen, slen; 2261 int gottime; 2262 2263 /* Skip outgoing duplicate packets. */ 2264 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2265 m->m_flags &= ~M_PROMISC; 2266 return; 2267 } 2268 2269 pktlen = m_length(m, NULL); 2270 /* 2271 * Craft on-stack mbuf suitable for passing to bpf_filter. 2272 * Note that we cut corners here; we only setup what's 2273 * absolutely needed--this mbuf should never go anywhere else. 2274 */ 2275 mb.m_next = m; 2276 mb.m_data = data; 2277 mb.m_len = dlen; 2278 pktlen += dlen; 2279 2280 gottime = BPF_TSTAMP_NONE; 2281 2282 BPFIF_RLOCK(bp); 2283 2284 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2285 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2286 continue; 2287 counter_u64_add(d->bd_rcount, 1); 2288 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2289 if (slen != 0) { 2290 BPFD_LOCK(d); 2291 2292 counter_u64_add(d->bd_fcount, 1); 2293 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2294 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2295 #ifdef MAC 2296 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2297 #endif 2298 catchpacket(d, (u_char *)&mb, pktlen, slen, 2299 bpf_append_mbuf, &bt); 2300 BPFD_UNLOCK(d); 2301 } 2302 } 2303 BPFIF_RUNLOCK(bp); 2304 } 2305 2306 #undef BPF_CHECK_DIRECTION 2307 2308 #undef BPF_TSTAMP_NONE 2309 #undef BPF_TSTAMP_FAST 2310 #undef BPF_TSTAMP_NORMAL 2311 #undef BPF_TSTAMP_EXTERN 2312 2313 static int 2314 bpf_hdrlen(struct bpf_d *d) 2315 { 2316 int hdrlen; 2317 2318 hdrlen = d->bd_bif->bif_hdrlen; 2319 #ifndef BURN_BRIDGES 2320 if (d->bd_tstamp == BPF_T_NONE || 2321 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2322 #ifdef COMPAT_FREEBSD32 2323 if (d->bd_compat32) 2324 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2325 else 2326 #endif 2327 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2328 else 2329 #endif 2330 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2331 #ifdef COMPAT_FREEBSD32 2332 if (d->bd_compat32) 2333 hdrlen = BPF_WORDALIGN32(hdrlen); 2334 else 2335 #endif 2336 hdrlen = BPF_WORDALIGN(hdrlen); 2337 2338 return (hdrlen - d->bd_bif->bif_hdrlen); 2339 } 2340 2341 static void 2342 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2343 { 2344 struct bintime bt2, boottimebin; 2345 struct timeval tsm; 2346 struct timespec tsn; 2347 2348 if ((tstype & BPF_T_MONOTONIC) == 0) { 2349 bt2 = *bt; 2350 getboottimebin(&boottimebin); 2351 bintime_add(&bt2, &boottimebin); 2352 bt = &bt2; 2353 } 2354 switch (BPF_T_FORMAT(tstype)) { 2355 case BPF_T_MICROTIME: 2356 bintime2timeval(bt, &tsm); 2357 ts->bt_sec = tsm.tv_sec; 2358 ts->bt_frac = tsm.tv_usec; 2359 break; 2360 case BPF_T_NANOTIME: 2361 bintime2timespec(bt, &tsn); 2362 ts->bt_sec = tsn.tv_sec; 2363 ts->bt_frac = tsn.tv_nsec; 2364 break; 2365 case BPF_T_BINTIME: 2366 ts->bt_sec = bt->sec; 2367 ts->bt_frac = bt->frac; 2368 break; 2369 } 2370 } 2371 2372 /* 2373 * Move the packet data from interface memory (pkt) into the 2374 * store buffer. "cpfn" is the routine called to do the actual data 2375 * transfer. bcopy is passed in to copy contiguous chunks, while 2376 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2377 * pkt is really an mbuf. 2378 */ 2379 static void 2380 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2381 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2382 struct bintime *bt) 2383 { 2384 struct bpf_xhdr hdr; 2385 #ifndef BURN_BRIDGES 2386 struct bpf_hdr hdr_old; 2387 #ifdef COMPAT_FREEBSD32 2388 struct bpf_hdr32 hdr32_old; 2389 #endif 2390 #endif 2391 int caplen, curlen, hdrlen, totlen; 2392 int do_wakeup = 0; 2393 int do_timestamp; 2394 int tstype; 2395 2396 BPFD_LOCK_ASSERT(d); 2397 2398 /* 2399 * Detect whether user space has released a buffer back to us, and if 2400 * so, move it from being a hold buffer to a free buffer. This may 2401 * not be the best place to do it (for example, we might only want to 2402 * run this check if we need the space), but for now it's a reliable 2403 * spot to do it. 2404 */ 2405 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2406 d->bd_fbuf = d->bd_hbuf; 2407 d->bd_hbuf = NULL; 2408 d->bd_hlen = 0; 2409 bpf_buf_reclaimed(d); 2410 } 2411 2412 /* 2413 * Figure out how many bytes to move. If the packet is 2414 * greater or equal to the snapshot length, transfer that 2415 * much. Otherwise, transfer the whole packet (unless 2416 * we hit the buffer size limit). 2417 */ 2418 hdrlen = bpf_hdrlen(d); 2419 totlen = hdrlen + min(snaplen, pktlen); 2420 if (totlen > d->bd_bufsize) 2421 totlen = d->bd_bufsize; 2422 2423 /* 2424 * Round up the end of the previous packet to the next longword. 2425 * 2426 * Drop the packet if there's no room and no hope of room 2427 * If the packet would overflow the storage buffer or the storage 2428 * buffer is considered immutable by the buffer model, try to rotate 2429 * the buffer and wakeup pending processes. 2430 */ 2431 #ifdef COMPAT_FREEBSD32 2432 if (d->bd_compat32) 2433 curlen = BPF_WORDALIGN32(d->bd_slen); 2434 else 2435 #endif 2436 curlen = BPF_WORDALIGN(d->bd_slen); 2437 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2438 if (d->bd_fbuf == NULL) { 2439 /* 2440 * There's no room in the store buffer, and no 2441 * prospect of room, so drop the packet. Notify the 2442 * buffer model. 2443 */ 2444 bpf_buffull(d); 2445 counter_u64_add(d->bd_dcount, 1); 2446 return; 2447 } 2448 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); 2449 ROTATE_BUFFERS(d); 2450 do_wakeup = 1; 2451 curlen = 0; 2452 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2453 /* 2454 * Immediate mode is set, or the read timeout has already 2455 * expired during a select call. A packet arrived, so the 2456 * reader should be woken up. 2457 */ 2458 do_wakeup = 1; 2459 caplen = totlen - hdrlen; 2460 tstype = d->bd_tstamp; 2461 do_timestamp = tstype != BPF_T_NONE; 2462 #ifndef BURN_BRIDGES 2463 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2464 struct bpf_ts ts; 2465 if (do_timestamp) 2466 bpf_bintime2ts(bt, &ts, tstype); 2467 #ifdef COMPAT_FREEBSD32 2468 if (d->bd_compat32) { 2469 bzero(&hdr32_old, sizeof(hdr32_old)); 2470 if (do_timestamp) { 2471 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2472 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2473 } 2474 hdr32_old.bh_datalen = pktlen; 2475 hdr32_old.bh_hdrlen = hdrlen; 2476 hdr32_old.bh_caplen = caplen; 2477 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2478 sizeof(hdr32_old)); 2479 goto copy; 2480 } 2481 #endif 2482 bzero(&hdr_old, sizeof(hdr_old)); 2483 if (do_timestamp) { 2484 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2485 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2486 } 2487 hdr_old.bh_datalen = pktlen; 2488 hdr_old.bh_hdrlen = hdrlen; 2489 hdr_old.bh_caplen = caplen; 2490 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2491 sizeof(hdr_old)); 2492 goto copy; 2493 } 2494 #endif 2495 2496 /* 2497 * Append the bpf header. Note we append the actual header size, but 2498 * move forward the length of the header plus padding. 2499 */ 2500 bzero(&hdr, sizeof(hdr)); 2501 if (do_timestamp) 2502 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2503 hdr.bh_datalen = pktlen; 2504 hdr.bh_hdrlen = hdrlen; 2505 hdr.bh_caplen = caplen; 2506 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2507 2508 /* 2509 * Copy the packet data into the store buffer and update its length. 2510 */ 2511 #ifndef BURN_BRIDGES 2512 copy: 2513 #endif 2514 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2515 d->bd_slen = curlen + totlen; 2516 2517 if (do_wakeup) 2518 bpf_wakeup(d); 2519 } 2520 2521 /* 2522 * Free buffers currently in use by a descriptor. 2523 * Called on close. 2524 */ 2525 static void 2526 bpf_freed(struct bpf_d *d) 2527 { 2528 2529 /* 2530 * We don't need to lock out interrupts since this descriptor has 2531 * been detached from its interface and it yet hasn't been marked 2532 * free. 2533 */ 2534 bpf_free(d); 2535 if (d->bd_rfilter != NULL) { 2536 free((caddr_t)d->bd_rfilter, M_BPF); 2537 #ifdef BPF_JITTER 2538 if (d->bd_bfilter != NULL) 2539 bpf_destroy_jit_filter(d->bd_bfilter); 2540 #endif 2541 } 2542 if (d->bd_wfilter != NULL) 2543 free((caddr_t)d->bd_wfilter, M_BPF); 2544 mtx_destroy(&d->bd_lock); 2545 2546 counter_u64_free(d->bd_rcount); 2547 counter_u64_free(d->bd_dcount); 2548 counter_u64_free(d->bd_fcount); 2549 counter_u64_free(d->bd_wcount); 2550 counter_u64_free(d->bd_wfcount); 2551 counter_u64_free(d->bd_wdcount); 2552 counter_u64_free(d->bd_zcopy); 2553 2554 } 2555 2556 /* 2557 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2558 * fixed size of the link header (variable length headers not yet supported). 2559 */ 2560 void 2561 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2562 { 2563 2564 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2565 } 2566 2567 /* 2568 * Attach an interface to bpf. ifp is a pointer to the structure 2569 * defining the interface to be attached, dlt is the link layer type, 2570 * and hdrlen is the fixed size of the link header (variable length 2571 * headers are not yet supporrted). 2572 */ 2573 void 2574 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2575 { 2576 struct bpf_if *bp; 2577 2578 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); 2579 if (bp == NULL) 2580 panic("bpfattach"); 2581 2582 LIST_INIT(&bp->bif_dlist); 2583 LIST_INIT(&bp->bif_wlist); 2584 bp->bif_ifp = ifp; 2585 bp->bif_dlt = dlt; 2586 rw_init(&bp->bif_lock, "bpf interface lock"); 2587 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); 2588 bp->bif_bpf = driverp; 2589 *driverp = bp; 2590 2591 BPF_LOCK(); 2592 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2593 BPF_UNLOCK(); 2594 2595 bp->bif_hdrlen = hdrlen; 2596 2597 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 2598 if_printf(ifp, "bpf attached\n"); 2599 } 2600 2601 #ifdef VIMAGE 2602 /* 2603 * When moving interfaces between vnet instances we need a way to 2604 * query the dlt and hdrlen before detach so we can re-attch the if_bpf 2605 * after the vmove. We unfortunately have no device driver infrastructure 2606 * to query the interface for these values after creation/attach, thus 2607 * add this as a workaround. 2608 */ 2609 int 2610 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen) 2611 { 2612 2613 if (bp == NULL) 2614 return (ENXIO); 2615 if (bif_dlt == NULL && bif_hdrlen == NULL) 2616 return (0); 2617 2618 if (bif_dlt != NULL) 2619 *bif_dlt = bp->bif_dlt; 2620 if (bif_hdrlen != NULL) 2621 *bif_hdrlen = bp->bif_hdrlen; 2622 2623 return (0); 2624 } 2625 #endif 2626 2627 /* 2628 * Detach bpf from an interface. This involves detaching each descriptor 2629 * associated with the interface. Notify each descriptor as it's detached 2630 * so that any sleepers wake up and get ENXIO. 2631 */ 2632 void 2633 bpfdetach(struct ifnet *ifp) 2634 { 2635 struct bpf_if *bp, *bp_temp; 2636 struct bpf_d *d; 2637 int ndetached; 2638 2639 ndetached = 0; 2640 2641 BPF_LOCK(); 2642 /* Find all bpf_if struct's which reference ifp and detach them. */ 2643 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2644 if (ifp != bp->bif_ifp) 2645 continue; 2646 2647 LIST_REMOVE(bp, bif_next); 2648 /* Add to to-be-freed list */ 2649 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); 2650 2651 ndetached++; 2652 /* 2653 * Delay freeing bp till interface is detached 2654 * and all routes through this interface are removed. 2655 * Mark bp as detached to restrict new consumers. 2656 */ 2657 BPFIF_WLOCK(bp); 2658 bp->bif_flags |= BPFIF_FLAG_DYING; 2659 *bp->bif_bpf = NULL; 2660 BPFIF_WUNLOCK(bp); 2661 2662 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", 2663 __func__, bp->bif_dlt, bp, ifp); 2664 2665 /* Free common descriptors */ 2666 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { 2667 bpf_detachd_locked(d); 2668 BPFD_LOCK(d); 2669 bpf_wakeup(d); 2670 BPFD_UNLOCK(d); 2671 } 2672 2673 /* Free writer-only descriptors */ 2674 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { 2675 bpf_detachd_locked(d); 2676 BPFD_LOCK(d); 2677 bpf_wakeup(d); 2678 BPFD_UNLOCK(d); 2679 } 2680 } 2681 BPF_UNLOCK(); 2682 2683 #ifdef INVARIANTS 2684 if (ndetached == 0) 2685 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 2686 #endif 2687 } 2688 2689 /* 2690 * Interface departure handler. 2691 * Note departure event does not guarantee interface is going down. 2692 * Interface renaming is currently done via departure/arrival event set. 2693 * 2694 * Departure handled is called after all routes pointing to 2695 * given interface are removed and interface is in down state 2696 * restricting any packets to be sent/received. We assume it is now safe 2697 * to free data allocated by BPF. 2698 */ 2699 static void 2700 bpf_ifdetach(void *arg __unused, struct ifnet *ifp) 2701 { 2702 struct bpf_if *bp, *bp_temp; 2703 int nmatched = 0; 2704 2705 /* Ignore ifnet renaming. */ 2706 if (ifp->if_flags & IFF_RENAMING) 2707 return; 2708 2709 BPF_LOCK(); 2710 /* 2711 * Find matching entries in free list. 2712 * Nothing should be found if bpfdetach() was not called. 2713 */ 2714 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { 2715 if (ifp != bp->bif_ifp) 2716 continue; 2717 2718 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", 2719 __func__, bp, ifp); 2720 2721 LIST_REMOVE(bp, bif_next); 2722 2723 rw_destroy(&bp->bif_lock); 2724 free(bp, M_BPF); 2725 2726 nmatched++; 2727 } 2728 BPF_UNLOCK(); 2729 } 2730 2731 /* 2732 * Get a list of available data link type of the interface. 2733 */ 2734 static int 2735 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2736 { 2737 struct ifnet *ifp; 2738 struct bpf_if *bp; 2739 u_int *lst; 2740 int error, n, n1; 2741 2742 BPF_LOCK_ASSERT(); 2743 2744 ifp = d->bd_bif->bif_ifp; 2745 again: 2746 n1 = 0; 2747 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2748 if (bp->bif_ifp == ifp) 2749 n1++; 2750 } 2751 if (bfl->bfl_list == NULL) { 2752 bfl->bfl_len = n1; 2753 return (0); 2754 } 2755 if (n1 > bfl->bfl_len) 2756 return (ENOMEM); 2757 BPF_UNLOCK(); 2758 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); 2759 n = 0; 2760 BPF_LOCK(); 2761 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2762 if (bp->bif_ifp != ifp) 2763 continue; 2764 if (n >= n1) { 2765 free(lst, M_TEMP); 2766 goto again; 2767 } 2768 lst[n] = bp->bif_dlt; 2769 n++; 2770 } 2771 BPF_UNLOCK(); 2772 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); 2773 free(lst, M_TEMP); 2774 BPF_LOCK(); 2775 bfl->bfl_len = n; 2776 return (error); 2777 } 2778 2779 /* 2780 * Set the data link type of a BPF instance. 2781 */ 2782 static int 2783 bpf_setdlt(struct bpf_d *d, u_int dlt) 2784 { 2785 int error, opromisc; 2786 struct ifnet *ifp; 2787 struct bpf_if *bp; 2788 2789 BPF_LOCK_ASSERT(); 2790 2791 if (d->bd_bif->bif_dlt == dlt) 2792 return (0); 2793 ifp = d->bd_bif->bif_ifp; 2794 2795 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2796 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2797 break; 2798 } 2799 2800 if (bp != NULL) { 2801 opromisc = d->bd_promisc; 2802 bpf_attachd(d, bp); 2803 BPFD_LOCK(d); 2804 reset_d(d); 2805 BPFD_UNLOCK(d); 2806 if (opromisc) { 2807 error = ifpromisc(bp->bif_ifp, 1); 2808 if (error) 2809 if_printf(bp->bif_ifp, 2810 "bpf_setdlt: ifpromisc failed (%d)\n", 2811 error); 2812 else 2813 d->bd_promisc = 1; 2814 } 2815 } 2816 return (bp == NULL ? EINVAL : 0); 2817 } 2818 2819 static void 2820 bpf_drvinit(void *unused) 2821 { 2822 struct cdev *dev; 2823 2824 mtx_init(&bpf_mtx, "bpf global lock", NULL, MTX_DEF); 2825 LIST_INIT(&bpf_iflist); 2826 LIST_INIT(&bpf_freelist); 2827 2828 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2829 /* For compatibility */ 2830 make_dev_alias(dev, "bpf0"); 2831 2832 /* Register interface departure handler */ 2833 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( 2834 ifnet_departure_event, bpf_ifdetach, NULL, 2835 EVENTHANDLER_PRI_ANY); 2836 } 2837 2838 /* 2839 * Zero out the various packet counters associated with all of the bpf 2840 * descriptors. At some point, we will probably want to get a bit more 2841 * granular and allow the user to specify descriptors to be zeroed. 2842 */ 2843 static void 2844 bpf_zero_counters(void) 2845 { 2846 struct bpf_if *bp; 2847 struct bpf_d *bd; 2848 2849 BPF_LOCK(); 2850 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2851 BPFIF_RLOCK(bp); 2852 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2853 BPFD_LOCK(bd); 2854 counter_u64_zero(bd->bd_rcount); 2855 counter_u64_zero(bd->bd_dcount); 2856 counter_u64_zero(bd->bd_fcount); 2857 counter_u64_zero(bd->bd_wcount); 2858 counter_u64_zero(bd->bd_wfcount); 2859 counter_u64_zero(bd->bd_zcopy); 2860 BPFD_UNLOCK(bd); 2861 } 2862 BPFIF_RUNLOCK(bp); 2863 } 2864 BPF_UNLOCK(); 2865 } 2866 2867 /* 2868 * Fill filter statistics 2869 */ 2870 static void 2871 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2872 { 2873 2874 bzero(d, sizeof(*d)); 2875 BPFD_LOCK_ASSERT(bd); 2876 d->bd_structsize = sizeof(*d); 2877 /* XXX: reading should be protected by global lock */ 2878 d->bd_immediate = bd->bd_immediate; 2879 d->bd_promisc = bd->bd_promisc; 2880 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2881 d->bd_direction = bd->bd_direction; 2882 d->bd_feedback = bd->bd_feedback; 2883 d->bd_async = bd->bd_async; 2884 d->bd_rcount = counter_u64_fetch(bd->bd_rcount); 2885 d->bd_dcount = counter_u64_fetch(bd->bd_dcount); 2886 d->bd_fcount = counter_u64_fetch(bd->bd_fcount); 2887 d->bd_sig = bd->bd_sig; 2888 d->bd_slen = bd->bd_slen; 2889 d->bd_hlen = bd->bd_hlen; 2890 d->bd_bufsize = bd->bd_bufsize; 2891 d->bd_pid = bd->bd_pid; 2892 strlcpy(d->bd_ifname, 2893 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2894 d->bd_locked = bd->bd_locked; 2895 d->bd_wcount = counter_u64_fetch(bd->bd_wcount); 2896 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount); 2897 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount); 2898 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy); 2899 d->bd_bufmode = bd->bd_bufmode; 2900 } 2901 2902 /* 2903 * Handle `netstat -B' stats request 2904 */ 2905 static int 2906 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2907 { 2908 static const struct xbpf_d zerostats; 2909 struct xbpf_d *xbdbuf, *xbd, tempstats; 2910 int index, error; 2911 struct bpf_if *bp; 2912 struct bpf_d *bd; 2913 2914 /* 2915 * XXX This is not technically correct. It is possible for non 2916 * privileged users to open bpf devices. It would make sense 2917 * if the users who opened the devices were able to retrieve 2918 * the statistics for them, too. 2919 */ 2920 error = priv_check(req->td, PRIV_NET_BPF); 2921 if (error) 2922 return (error); 2923 /* 2924 * Check to see if the user is requesting that the counters be 2925 * zeroed out. Explicitly check that the supplied data is zeroed, 2926 * as we aren't allowing the user to set the counters currently. 2927 */ 2928 if (req->newptr != NULL) { 2929 if (req->newlen != sizeof(tempstats)) 2930 return (EINVAL); 2931 memset(&tempstats, 0, sizeof(tempstats)); 2932 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); 2933 if (error) 2934 return (error); 2935 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) 2936 return (EINVAL); 2937 bpf_zero_counters(); 2938 return (0); 2939 } 2940 if (req->oldptr == NULL) 2941 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2942 if (bpf_bpfd_cnt == 0) 2943 return (SYSCTL_OUT(req, 0, 0)); 2944 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2945 BPF_LOCK(); 2946 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2947 BPF_UNLOCK(); 2948 free(xbdbuf, M_BPF); 2949 return (ENOMEM); 2950 } 2951 index = 0; 2952 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2953 BPFIF_RLOCK(bp); 2954 /* Send writers-only first */ 2955 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2956 xbd = &xbdbuf[index++]; 2957 BPFD_LOCK(bd); 2958 bpfstats_fill_xbpf(xbd, bd); 2959 BPFD_UNLOCK(bd); 2960 } 2961 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2962 xbd = &xbdbuf[index++]; 2963 BPFD_LOCK(bd); 2964 bpfstats_fill_xbpf(xbd, bd); 2965 BPFD_UNLOCK(bd); 2966 } 2967 BPFIF_RUNLOCK(bp); 2968 } 2969 BPF_UNLOCK(); 2970 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2971 free(xbdbuf, M_BPF); 2972 return (error); 2973 } 2974 2975 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2976 2977 #else /* !DEV_BPF && !NETGRAPH_BPF */ 2978 /* 2979 * NOP stubs to allow bpf-using drivers to load and function. 2980 * 2981 * A 'better' implementation would allow the core bpf functionality 2982 * to be loaded at runtime. 2983 */ 2984 static struct bpf_if bp_null; 2985 2986 void 2987 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2988 { 2989 } 2990 2991 void 2992 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2993 { 2994 } 2995 2996 void 2997 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 2998 { 2999 } 3000 3001 void 3002 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 3003 { 3004 3005 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 3006 } 3007 3008 void 3009 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 3010 { 3011 3012 *driverp = &bp_null; 3013 } 3014 3015 void 3016 bpfdetach(struct ifnet *ifp) 3017 { 3018 } 3019 3020 u_int 3021 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 3022 { 3023 return -1; /* "no filter" behaviour */ 3024 } 3025 3026 int 3027 bpf_validate(const struct bpf_insn *f, int len) 3028 { 3029 return 0; /* false */ 3030 } 3031 3032 #endif /* !DEV_BPF && !NETGRAPH_BPF */ 3033 3034 #ifdef DDB 3035 static void 3036 bpf_show_bpf_if(struct bpf_if *bpf_if) 3037 { 3038 3039 if (bpf_if == NULL) 3040 return; 3041 db_printf("%p:\n", bpf_if); 3042 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e); 3043 /* bif_ext.bif_next */ 3044 /* bif_ext.bif_dlist */ 3045 BPF_DB_PRINTF("%#x", bif_dlt); 3046 BPF_DB_PRINTF("%u", bif_hdrlen); 3047 BPF_DB_PRINTF("%p", bif_ifp); 3048 /* bif_lock */ 3049 /* bif_wlist */ 3050 BPF_DB_PRINTF("%#x", bif_flags); 3051 } 3052 3053 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if) 3054 { 3055 3056 if (!have_addr) { 3057 db_printf("usage: show bpf_if <struct bpf_if *>\n"); 3058 return; 3059 } 3060 3061 bpf_show_bpf_if((struct bpf_if *)addr); 3062 } 3063 #endif 3064