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 static struct bpf_if_ext dead_bpf_if = { 102 .bif_dlist = LIST_HEAD_INITIALIZER() 103 }; 104 105 struct bpf_if { 106 #define bif_next bif_ext.bif_next 107 #define bif_dlist bif_ext.bif_dlist 108 struct bpf_if_ext bif_ext; /* public members */ 109 u_int bif_dlt; /* link layer type */ 110 u_int bif_hdrlen; /* length of link header */ 111 struct ifnet *bif_ifp; /* corresponding interface */ 112 struct rwlock bif_lock; /* interface lock */ 113 LIST_HEAD(, bpf_d) bif_wlist; /* writer-only list */ 114 int bif_flags; /* Interface flags */ 115 struct bpf_if **bif_bpf; /* Pointer to pointer to us */ 116 }; 117 118 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0); 119 120 #if defined(DEV_BPF) || defined(NETGRAPH_BPF) 121 122 #define PRINET 26 /* interruptible */ 123 124 #define SIZEOF_BPF_HDR(type) \ 125 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) 126 127 #ifdef COMPAT_FREEBSD32 128 #include <sys/mount.h> 129 #include <compat/freebsd32/freebsd32.h> 130 #define BPF_ALIGNMENT32 sizeof(int32_t) 131 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32) 132 133 #ifndef BURN_BRIDGES 134 /* 135 * 32-bit version of structure prepended to each packet. We use this header 136 * instead of the standard one for 32-bit streams. We mark the a stream as 137 * 32-bit the first time we see a 32-bit compat ioctl request. 138 */ 139 struct bpf_hdr32 { 140 struct timeval32 bh_tstamp; /* time stamp */ 141 uint32_t bh_caplen; /* length of captured portion */ 142 uint32_t bh_datalen; /* original length of packet */ 143 uint16_t bh_hdrlen; /* length of bpf header (this struct 144 plus alignment padding) */ 145 }; 146 #endif 147 148 struct bpf_program32 { 149 u_int bf_len; 150 uint32_t bf_insns; 151 }; 152 153 struct bpf_dltlist32 { 154 u_int bfl_len; 155 u_int bfl_list; 156 }; 157 158 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32) 159 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) 160 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) 161 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) 162 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) 163 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) 164 #endif 165 166 #define BPF_LOCK() sx_xlock(&bpf_sx) 167 #define BPF_UNLOCK() sx_xunlock(&bpf_sx) 168 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED) 169 /* 170 * bpf_iflist is a list of BPF interface structures, each corresponding to a 171 * specific DLT. The same network interface might have several BPF interface 172 * structures registered by different layers in the stack (i.e., 802.11 173 * frames, ethernet frames, etc). 174 */ 175 static LIST_HEAD(, bpf_if) bpf_iflist, bpf_freelist; 176 static struct sx bpf_sx; /* bpf global lock */ 177 static int bpf_bpfd_cnt; 178 179 static void bpf_attachd(struct bpf_d *, struct bpf_if *); 180 static void bpf_detachd(struct bpf_d *); 181 static void bpf_detachd_locked(struct bpf_d *); 182 static void bpf_freed(struct bpf_d *); 183 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, 184 struct sockaddr *, int *, struct bpf_d *); 185 static int bpf_setif(struct bpf_d *, struct ifreq *); 186 static void bpf_timed_out(void *); 187 static __inline void 188 bpf_wakeup(struct bpf_d *); 189 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 190 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), 191 struct bintime *); 192 static void reset_d(struct bpf_d *); 193 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); 194 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 195 static int bpf_setdlt(struct bpf_d *, u_int); 196 static void filt_bpfdetach(struct knote *); 197 static int filt_bpfread(struct knote *, long); 198 static void bpf_drvinit(void *); 199 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); 200 201 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); 202 int bpf_maxinsns = BPF_MAXINSNS; 203 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, 204 &bpf_maxinsns, 0, "Maximum bpf program instructions"); 205 static int bpf_zerocopy_enable = 0; 206 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, 207 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); 208 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, 209 bpf_stats_sysctl, "bpf statistics portal"); 210 211 static VNET_DEFINE(int, bpf_optimize_writers) = 0; 212 #define V_bpf_optimize_writers VNET(bpf_optimize_writers) 213 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(bpf_optimize_writers), 0, 215 "Do not send packets until BPF program is set"); 216 217 static d_open_t bpfopen; 218 static d_read_t bpfread; 219 static d_write_t bpfwrite; 220 static d_ioctl_t bpfioctl; 221 static d_poll_t bpfpoll; 222 static d_kqfilter_t bpfkqfilter; 223 224 static struct cdevsw bpf_cdevsw = { 225 .d_version = D_VERSION, 226 .d_open = bpfopen, 227 .d_read = bpfread, 228 .d_write = bpfwrite, 229 .d_ioctl = bpfioctl, 230 .d_poll = bpfpoll, 231 .d_name = "bpf", 232 .d_kqfilter = bpfkqfilter, 233 }; 234 235 static struct filterops bpfread_filtops = { 236 .f_isfd = 1, 237 .f_detach = filt_bpfdetach, 238 .f_event = filt_bpfread, 239 }; 240 241 eventhandler_tag bpf_ifdetach_cookie = NULL; 242 243 /* 244 * LOCKING MODEL USED BY BPF: 245 * Locks: 246 * 1) global lock (BPF_LOCK). Mutex, used to protect interface addition/removal, 247 * some global counters and every bpf_if reference. 248 * 2) Interface lock. Rwlock, used to protect list of BPF descriptors and their filters. 249 * 3) Descriptor lock. Mutex, used to protect BPF buffers and various structure fields 250 * used by bpf_mtap code. 251 * 252 * Lock order: 253 * 254 * Global lock, interface lock, descriptor lock 255 * 256 * We have to acquire interface lock before descriptor main lock due to BPF_MTAP[2] 257 * working model. In many places (like bpf_detachd) we start with BPF descriptor 258 * (and we need to at least rlock it to get reliable interface pointer). This 259 * gives us potential LOR. As a result, we use global lock to protect from bpf_if 260 * change in every such place. 261 * 262 * Changing d->bd_bif is protected by 1) global lock, 2) interface lock and 263 * 3) descriptor main wlock. 264 * Reading bd_bif can be protected by any of these locks, typically global lock. 265 * 266 * Changing read/write BPF filter is protected by the same three locks, 267 * the same applies for reading. 268 * 269 * Sleeping in global lock is not allowed due to bpfdetach() using it. 270 */ 271 272 /* 273 * Wrapper functions for various buffering methods. If the set of buffer 274 * modes expands, we will probably want to introduce a switch data structure 275 * similar to protosw, et. 276 */ 277 static void 278 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 279 u_int len) 280 { 281 282 BPFD_LOCK_ASSERT(d); 283 284 switch (d->bd_bufmode) { 285 case BPF_BUFMODE_BUFFER: 286 return (bpf_buffer_append_bytes(d, buf, offset, src, len)); 287 288 case BPF_BUFMODE_ZBUF: 289 counter_u64_add(d->bd_zcopy, 1); 290 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); 291 292 default: 293 panic("bpf_buf_append_bytes"); 294 } 295 } 296 297 static void 298 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 299 u_int len) 300 { 301 302 BPFD_LOCK_ASSERT(d); 303 304 switch (d->bd_bufmode) { 305 case BPF_BUFMODE_BUFFER: 306 return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); 307 308 case BPF_BUFMODE_ZBUF: 309 counter_u64_add(d->bd_zcopy, 1); 310 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); 311 312 default: 313 panic("bpf_buf_append_mbuf"); 314 } 315 } 316 317 /* 318 * This function gets called when the free buffer is re-assigned. 319 */ 320 static void 321 bpf_buf_reclaimed(struct bpf_d *d) 322 { 323 324 BPFD_LOCK_ASSERT(d); 325 326 switch (d->bd_bufmode) { 327 case BPF_BUFMODE_BUFFER: 328 return; 329 330 case BPF_BUFMODE_ZBUF: 331 bpf_zerocopy_buf_reclaimed(d); 332 return; 333 334 default: 335 panic("bpf_buf_reclaimed"); 336 } 337 } 338 339 /* 340 * If the buffer mechanism has a way to decide that a held buffer can be made 341 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is 342 * returned if the buffer can be discarded, (0) is returned if it cannot. 343 */ 344 static int 345 bpf_canfreebuf(struct bpf_d *d) 346 { 347 348 BPFD_LOCK_ASSERT(d); 349 350 switch (d->bd_bufmode) { 351 case BPF_BUFMODE_ZBUF: 352 return (bpf_zerocopy_canfreebuf(d)); 353 } 354 return (0); 355 } 356 357 /* 358 * Allow the buffer model to indicate that the current store buffer is 359 * immutable, regardless of the appearance of space. Return (1) if the 360 * buffer is writable, and (0) if not. 361 */ 362 static int 363 bpf_canwritebuf(struct bpf_d *d) 364 { 365 BPFD_LOCK_ASSERT(d); 366 367 switch (d->bd_bufmode) { 368 case BPF_BUFMODE_ZBUF: 369 return (bpf_zerocopy_canwritebuf(d)); 370 } 371 return (1); 372 } 373 374 /* 375 * Notify buffer model that an attempt to write to the store buffer has 376 * resulted in a dropped packet, in which case the buffer may be considered 377 * full. 378 */ 379 static void 380 bpf_buffull(struct bpf_d *d) 381 { 382 383 BPFD_LOCK_ASSERT(d); 384 385 switch (d->bd_bufmode) { 386 case BPF_BUFMODE_ZBUF: 387 bpf_zerocopy_buffull(d); 388 break; 389 } 390 } 391 392 /* 393 * Notify the buffer model that a buffer has moved into the hold position. 394 */ 395 void 396 bpf_bufheld(struct bpf_d *d) 397 { 398 399 BPFD_LOCK_ASSERT(d); 400 401 switch (d->bd_bufmode) { 402 case BPF_BUFMODE_ZBUF: 403 bpf_zerocopy_bufheld(d); 404 break; 405 } 406 } 407 408 static void 409 bpf_free(struct bpf_d *d) 410 { 411 412 switch (d->bd_bufmode) { 413 case BPF_BUFMODE_BUFFER: 414 return (bpf_buffer_free(d)); 415 416 case BPF_BUFMODE_ZBUF: 417 return (bpf_zerocopy_free(d)); 418 419 default: 420 panic("bpf_buf_free"); 421 } 422 } 423 424 static int 425 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) 426 { 427 428 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 429 return (EOPNOTSUPP); 430 return (bpf_buffer_uiomove(d, buf, len, uio)); 431 } 432 433 static int 434 bpf_ioctl_sblen(struct bpf_d *d, u_int *i) 435 { 436 437 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 438 return (EOPNOTSUPP); 439 return (bpf_buffer_ioctl_sblen(d, i)); 440 } 441 442 static int 443 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) 444 { 445 446 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 447 return (EOPNOTSUPP); 448 return (bpf_zerocopy_ioctl_getzmax(td, d, i)); 449 } 450 451 static int 452 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 453 { 454 455 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 456 return (EOPNOTSUPP); 457 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); 458 } 459 460 static int 461 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 462 { 463 464 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 465 return (EOPNOTSUPP); 466 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); 467 } 468 469 /* 470 * General BPF functions. 471 */ 472 static int 473 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, 474 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d) 475 { 476 const struct ieee80211_bpf_params *p; 477 struct ether_header *eh; 478 struct mbuf *m; 479 int error; 480 int len; 481 int hlen; 482 int slen; 483 484 /* 485 * Build a sockaddr based on the data link layer type. 486 * We do this at this level because the ethernet header 487 * is copied directly into the data field of the sockaddr. 488 * In the case of SLIP, there is no header and the packet 489 * is forwarded as is. 490 * Also, we are careful to leave room at the front of the mbuf 491 * for the link level header. 492 */ 493 switch (linktype) { 494 495 case DLT_SLIP: 496 sockp->sa_family = AF_INET; 497 hlen = 0; 498 break; 499 500 case DLT_EN10MB: 501 sockp->sa_family = AF_UNSPEC; 502 /* XXX Would MAXLINKHDR be better? */ 503 hlen = ETHER_HDR_LEN; 504 break; 505 506 case DLT_FDDI: 507 sockp->sa_family = AF_IMPLINK; 508 hlen = 0; 509 break; 510 511 case DLT_RAW: 512 sockp->sa_family = AF_UNSPEC; 513 hlen = 0; 514 break; 515 516 case DLT_NULL: 517 /* 518 * null interface types require a 4 byte pseudo header which 519 * corresponds to the address family of the packet. 520 */ 521 sockp->sa_family = AF_UNSPEC; 522 hlen = 4; 523 break; 524 525 case DLT_ATM_RFC1483: 526 /* 527 * en atm driver requires 4-byte atm pseudo header. 528 * though it isn't standard, vpi:vci needs to be 529 * specified anyway. 530 */ 531 sockp->sa_family = AF_UNSPEC; 532 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ 533 break; 534 535 case DLT_PPP: 536 sockp->sa_family = AF_UNSPEC; 537 hlen = 4; /* This should match PPP_HDRLEN */ 538 break; 539 540 case DLT_IEEE802_11: /* IEEE 802.11 wireless */ 541 sockp->sa_family = AF_IEEE80211; 542 hlen = 0; 543 break; 544 545 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ 546 sockp->sa_family = AF_IEEE80211; 547 sockp->sa_len = 12; /* XXX != 0 */ 548 hlen = sizeof(struct ieee80211_bpf_params); 549 break; 550 551 default: 552 return (EIO); 553 } 554 555 len = uio->uio_resid; 556 if (len < hlen || len - hlen > ifp->if_mtu) 557 return (EMSGSIZE); 558 559 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); 560 if (m == NULL) 561 return (EIO); 562 m->m_pkthdr.len = m->m_len = len; 563 *mp = m; 564 565 error = uiomove(mtod(m, u_char *), len, uio); 566 if (error) 567 goto bad; 568 569 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len); 570 if (slen == 0) { 571 error = EPERM; 572 goto bad; 573 } 574 575 /* Check for multicast destination */ 576 switch (linktype) { 577 case DLT_EN10MB: 578 eh = mtod(m, struct ether_header *); 579 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 580 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 581 ETHER_ADDR_LEN) == 0) 582 m->m_flags |= M_BCAST; 583 else 584 m->m_flags |= M_MCAST; 585 } 586 if (d->bd_hdrcmplt == 0) { 587 memcpy(eh->ether_shost, IF_LLADDR(ifp), 588 sizeof(eh->ether_shost)); 589 } 590 break; 591 } 592 593 /* 594 * Make room for link header, and copy it to sockaddr 595 */ 596 if (hlen != 0) { 597 if (sockp->sa_family == AF_IEEE80211) { 598 /* 599 * Collect true length from the parameter header 600 * NB: sockp is known to be zero'd so if we do a 601 * short copy unspecified parameters will be 602 * zero. 603 * NB: packet may not be aligned after stripping 604 * bpf params 605 * XXX check ibp_vers 606 */ 607 p = mtod(m, const struct ieee80211_bpf_params *); 608 hlen = p->ibp_len; 609 if (hlen > sizeof(sockp->sa_data)) { 610 error = EINVAL; 611 goto bad; 612 } 613 } 614 bcopy(mtod(m, const void *), sockp->sa_data, hlen); 615 } 616 *hdrlen = hlen; 617 618 return (0); 619 bad: 620 m_freem(m); 621 return (error); 622 } 623 624 /* 625 * Attach file to the bpf interface, i.e. make d listen on bp. 626 */ 627 static void 628 bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 629 { 630 int op_w; 631 632 BPF_LOCK_ASSERT(); 633 634 /* 635 * Save sysctl value to protect from sysctl change 636 * between reads 637 */ 638 op_w = V_bpf_optimize_writers || d->bd_writer; 639 640 if (d->bd_bif != NULL) 641 bpf_detachd_locked(d); 642 /* 643 * Point d at bp, and add d to the interface's list. 644 * Since there are many applications using BPF for 645 * sending raw packets only (dhcpd, cdpd are good examples) 646 * we can delay adding d to the list of active listeners until 647 * some filter is configured. 648 */ 649 650 BPFIF_WLOCK(bp); 651 BPFD_LOCK(d); 652 653 d->bd_bif = bp; 654 655 if (op_w != 0) { 656 /* Add to writers-only list */ 657 LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); 658 /* 659 * We decrement bd_writer on every filter set operation. 660 * First BIOCSETF is done by pcap_open_live() to set up 661 * snap length. After that appliation usually sets its own filter 662 */ 663 d->bd_writer = 2; 664 } else 665 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 666 667 BPFD_UNLOCK(d); 668 BPFIF_WUNLOCK(bp); 669 670 bpf_bpfd_cnt++; 671 672 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", 673 __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); 674 675 if (op_w == 0) 676 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 677 } 678 679 /* 680 * Check if we need to upgrade our descriptor @d from write-only mode. 681 */ 682 static int 683 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, int flen) 684 { 685 int is_snap, need_upgrade; 686 687 /* 688 * Check if we've already upgraded or new filter is empty. 689 */ 690 if (d->bd_writer == 0 || fcode == NULL) 691 return (0); 692 693 need_upgrade = 0; 694 695 /* 696 * Check if cmd looks like snaplen setting from 697 * pcap_bpf.c:pcap_open_live(). 698 * Note we're not checking .k value here: 699 * while pcap_open_live() definitely sets to non-zero value, 700 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g. 701 * do not consider upgrading immediately 702 */ 703 if (cmd == BIOCSETF && flen == 1 && fcode[0].code == (BPF_RET | BPF_K)) 704 is_snap = 1; 705 else 706 is_snap = 0; 707 708 if (is_snap == 0) { 709 /* 710 * We're setting first filter and it doesn't look like 711 * setting snaplen. We're probably using bpf directly. 712 * Upgrade immediately. 713 */ 714 need_upgrade = 1; 715 } else { 716 /* 717 * Do not require upgrade by first BIOCSETF 718 * (used to set snaplen) by pcap_open_live(). 719 */ 720 721 if (--d->bd_writer == 0) { 722 /* 723 * First snaplen filter has already 724 * been set. This is probably catch-all 725 * filter 726 */ 727 need_upgrade = 1; 728 } 729 } 730 731 CTR5(KTR_NET, 732 "%s: filter function set by pid %d, " 733 "bd_writer counter %d, snap %d upgrade %d", 734 __func__, d->bd_pid, d->bd_writer, 735 is_snap, need_upgrade); 736 737 return (need_upgrade); 738 } 739 740 /* 741 * Add d to the list of active bp filters. 742 * Requires bpf_attachd() to be called before. 743 */ 744 static void 745 bpf_upgraded(struct bpf_d *d) 746 { 747 struct bpf_if *bp; 748 749 BPF_LOCK_ASSERT(); 750 751 bp = d->bd_bif; 752 753 /* 754 * Filter can be set several times without specifying interface. 755 * Mark d as reader and exit. 756 */ 757 if (bp == NULL) { 758 BPFD_LOCK(d); 759 d->bd_writer = 0; 760 BPFD_UNLOCK(d); 761 return; 762 } 763 764 BPFIF_WLOCK(bp); 765 BPFD_LOCK(d); 766 767 /* Remove from writers-only list */ 768 LIST_REMOVE(d, bd_next); 769 LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 770 /* Mark d as reader */ 771 d->bd_writer = 0; 772 773 BPFD_UNLOCK(d); 774 BPFIF_WUNLOCK(bp); 775 776 CTR2(KTR_NET, "%s: upgrade required by pid %d", __func__, d->bd_pid); 777 778 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 779 } 780 781 /* 782 * Detach a file from its interface. 783 */ 784 static void 785 bpf_detachd(struct bpf_d *d) 786 { 787 BPF_LOCK(); 788 bpf_detachd_locked(d); 789 BPF_UNLOCK(); 790 } 791 792 static void 793 bpf_detachd_locked(struct bpf_d *d) 794 { 795 int error; 796 struct bpf_if *bp; 797 struct ifnet *ifp; 798 799 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); 800 801 BPF_LOCK_ASSERT(); 802 803 /* Check if descriptor is attached */ 804 if ((bp = d->bd_bif) == NULL) 805 return; 806 807 BPFIF_WLOCK(bp); 808 BPFD_LOCK(d); 809 810 /* Save bd_writer value */ 811 error = d->bd_writer; 812 813 /* 814 * Remove d from the interface's descriptor list. 815 */ 816 LIST_REMOVE(d, bd_next); 817 818 ifp = bp->bif_ifp; 819 d->bd_bif = NULL; 820 BPFD_UNLOCK(d); 821 BPFIF_WUNLOCK(bp); 822 823 bpf_bpfd_cnt--; 824 825 /* Call event handler iff d is attached */ 826 if (error == 0) 827 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); 828 829 /* 830 * Check if this descriptor had requested promiscuous mode. 831 * If so, turn it off. 832 */ 833 if (d->bd_promisc) { 834 d->bd_promisc = 0; 835 CURVNET_SET(ifp->if_vnet); 836 error = ifpromisc(ifp, 0); 837 CURVNET_RESTORE(); 838 if (error != 0 && error != ENXIO) { 839 /* 840 * ENXIO can happen if a pccard is unplugged 841 * Something is really wrong if we were able to put 842 * the driver into promiscuous mode, but can't 843 * take it out. 844 */ 845 if_printf(bp->bif_ifp, 846 "bpf_detach: ifpromisc failed (%d)\n", error); 847 } 848 } 849 } 850 851 /* 852 * Close the descriptor by detaching it from its interface, 853 * deallocating its buffers, and marking it free. 854 */ 855 static void 856 bpf_dtor(void *data) 857 { 858 struct bpf_d *d = data; 859 860 BPFD_LOCK(d); 861 if (d->bd_state == BPF_WAITING) 862 callout_stop(&d->bd_callout); 863 d->bd_state = BPF_IDLE; 864 BPFD_UNLOCK(d); 865 funsetown(&d->bd_sigio); 866 bpf_detachd(d); 867 #ifdef MAC 868 mac_bpfdesc_destroy(d); 869 #endif /* MAC */ 870 seldrain(&d->bd_sel); 871 knlist_destroy(&d->bd_sel.si_note); 872 callout_drain(&d->bd_callout); 873 bpf_freed(d); 874 free(d, M_BPF); 875 } 876 877 /* 878 * Open ethernet device. Returns ENXIO for illegal minor device number, 879 * EBUSY if file is open by another process. 880 */ 881 /* ARGSUSED */ 882 static int 883 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) 884 { 885 struct bpf_d *d; 886 int error; 887 888 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); 889 error = devfs_set_cdevpriv(d, bpf_dtor); 890 if (error != 0) { 891 free(d, M_BPF); 892 return (error); 893 } 894 895 /* Setup counters */ 896 d->bd_rcount = counter_u64_alloc(M_WAITOK); 897 d->bd_dcount = counter_u64_alloc(M_WAITOK); 898 d->bd_fcount = counter_u64_alloc(M_WAITOK); 899 d->bd_wcount = counter_u64_alloc(M_WAITOK); 900 d->bd_wfcount = counter_u64_alloc(M_WAITOK); 901 d->bd_wdcount = counter_u64_alloc(M_WAITOK); 902 d->bd_zcopy = counter_u64_alloc(M_WAITOK); 903 904 /* 905 * For historical reasons, perform a one-time initialization call to 906 * the buffer routines, even though we're not yet committed to a 907 * particular buffer method. 908 */ 909 bpf_buffer_init(d); 910 if ((flags & FREAD) == 0) 911 d->bd_writer = 2; 912 d->bd_hbuf_in_use = 0; 913 d->bd_bufmode = BPF_BUFMODE_BUFFER; 914 d->bd_sig = SIGIO; 915 d->bd_direction = BPF_D_INOUT; 916 BPF_PID_REFRESH(d, td); 917 #ifdef MAC 918 mac_bpfdesc_init(d); 919 mac_bpfdesc_create(td->td_ucred, d); 920 #endif 921 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 922 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 923 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 924 925 return (0); 926 } 927 928 /* 929 * bpfread - read next chunk of packets from buffers 930 */ 931 static int 932 bpfread(struct cdev *dev, struct uio *uio, int ioflag) 933 { 934 struct bpf_d *d; 935 int error; 936 int non_block; 937 int timed_out; 938 939 error = devfs_get_cdevpriv((void **)&d); 940 if (error != 0) 941 return (error); 942 943 /* 944 * Restrict application to use a buffer the same size as 945 * as kernel buffers. 946 */ 947 if (uio->uio_resid != d->bd_bufsize) 948 return (EINVAL); 949 950 non_block = ((ioflag & O_NONBLOCK) != 0); 951 952 BPFD_LOCK(d); 953 BPF_PID_REFRESH_CUR(d); 954 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 955 BPFD_UNLOCK(d); 956 return (EOPNOTSUPP); 957 } 958 if (d->bd_state == BPF_WAITING) 959 callout_stop(&d->bd_callout); 960 timed_out = (d->bd_state == BPF_TIMED_OUT); 961 d->bd_state = BPF_IDLE; 962 while (d->bd_hbuf_in_use) { 963 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 964 PRINET|PCATCH, "bd_hbuf", 0); 965 if (error != 0) { 966 BPFD_UNLOCK(d); 967 return (error); 968 } 969 } 970 /* 971 * If the hold buffer is empty, then do a timed sleep, which 972 * ends when the timeout expires or when enough packets 973 * have arrived to fill the store buffer. 974 */ 975 while (d->bd_hbuf == NULL) { 976 if (d->bd_slen != 0) { 977 /* 978 * A packet(s) either arrived since the previous 979 * read or arrived while we were asleep. 980 */ 981 if (d->bd_immediate || non_block || timed_out) { 982 /* 983 * Rotate the buffers and return what's here 984 * if we are in immediate mode, non-blocking 985 * flag is set, or this descriptor timed out. 986 */ 987 ROTATE_BUFFERS(d); 988 break; 989 } 990 } 991 992 /* 993 * No data is available, check to see if the bpf device 994 * is still pointed at a real interface. If not, return 995 * ENXIO so that the userland process knows to rebind 996 * it before using it again. 997 */ 998 if (d->bd_bif == NULL) { 999 BPFD_UNLOCK(d); 1000 return (ENXIO); 1001 } 1002 1003 if (non_block) { 1004 BPFD_UNLOCK(d); 1005 return (EWOULDBLOCK); 1006 } 1007 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 1008 "bpf", d->bd_rtout); 1009 if (error == EINTR || error == ERESTART) { 1010 BPFD_UNLOCK(d); 1011 return (error); 1012 } 1013 if (error == EWOULDBLOCK) { 1014 /* 1015 * On a timeout, return what's in the buffer, 1016 * which may be nothing. If there is something 1017 * in the store buffer, we can rotate the buffers. 1018 */ 1019 if (d->bd_hbuf) 1020 /* 1021 * We filled up the buffer in between 1022 * getting the timeout and arriving 1023 * here, so we don't need to rotate. 1024 */ 1025 break; 1026 1027 if (d->bd_slen == 0) { 1028 BPFD_UNLOCK(d); 1029 return (0); 1030 } 1031 ROTATE_BUFFERS(d); 1032 break; 1033 } 1034 } 1035 /* 1036 * At this point, we know we have something in the hold slot. 1037 */ 1038 d->bd_hbuf_in_use = 1; 1039 BPFD_UNLOCK(d); 1040 1041 /* 1042 * Move data from hold buffer into user space. 1043 * We know the entire buffer is transferred since 1044 * we checked above that the read buffer is bpf_bufsize bytes. 1045 * 1046 * We do not have to worry about simultaneous reads because 1047 * we waited for sole access to the hold buffer above. 1048 */ 1049 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 1050 1051 BPFD_LOCK(d); 1052 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 1053 d->bd_fbuf = d->bd_hbuf; 1054 d->bd_hbuf = NULL; 1055 d->bd_hlen = 0; 1056 bpf_buf_reclaimed(d); 1057 d->bd_hbuf_in_use = 0; 1058 wakeup(&d->bd_hbuf_in_use); 1059 BPFD_UNLOCK(d); 1060 1061 return (error); 1062 } 1063 1064 /* 1065 * If there are processes sleeping on this descriptor, wake them up. 1066 */ 1067 static __inline void 1068 bpf_wakeup(struct bpf_d *d) 1069 { 1070 1071 BPFD_LOCK_ASSERT(d); 1072 if (d->bd_state == BPF_WAITING) { 1073 callout_stop(&d->bd_callout); 1074 d->bd_state = BPF_IDLE; 1075 } 1076 wakeup(d); 1077 if (d->bd_async && d->bd_sig && d->bd_sigio) 1078 pgsigio(&d->bd_sigio, d->bd_sig, 0); 1079 1080 selwakeuppri(&d->bd_sel, PRINET); 1081 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 1082 } 1083 1084 static void 1085 bpf_timed_out(void *arg) 1086 { 1087 struct bpf_d *d = (struct bpf_d *)arg; 1088 1089 BPFD_LOCK_ASSERT(d); 1090 1091 if (callout_pending(&d->bd_callout) || !callout_active(&d->bd_callout)) 1092 return; 1093 if (d->bd_state == BPF_WAITING) { 1094 d->bd_state = BPF_TIMED_OUT; 1095 if (d->bd_slen != 0) 1096 bpf_wakeup(d); 1097 } 1098 } 1099 1100 static int 1101 bpf_ready(struct bpf_d *d) 1102 { 1103 1104 BPFD_LOCK_ASSERT(d); 1105 1106 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1107 return (1); 1108 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1109 d->bd_slen != 0) 1110 return (1); 1111 return (0); 1112 } 1113 1114 static int 1115 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1116 { 1117 struct bpf_d *d; 1118 struct ifnet *ifp; 1119 struct mbuf *m, *mc; 1120 struct sockaddr dst; 1121 struct route ro; 1122 int error, hlen; 1123 1124 error = devfs_get_cdevpriv((void **)&d); 1125 if (error != 0) 1126 return (error); 1127 1128 BPF_PID_REFRESH_CUR(d); 1129 counter_u64_add(d->bd_wcount, 1); 1130 /* XXX: locking required */ 1131 if (d->bd_bif == NULL) { 1132 counter_u64_add(d->bd_wdcount, 1); 1133 return (ENXIO); 1134 } 1135 1136 ifp = d->bd_bif->bif_ifp; 1137 1138 if ((ifp->if_flags & IFF_UP) == 0) { 1139 counter_u64_add(d->bd_wdcount, 1); 1140 return (ENETDOWN); 1141 } 1142 1143 if (uio->uio_resid == 0) { 1144 counter_u64_add(d->bd_wdcount, 1); 1145 return (0); 1146 } 1147 1148 bzero(&dst, sizeof(dst)); 1149 m = NULL; 1150 hlen = 0; 1151 /* XXX: bpf_movein() can sleep */ 1152 error = bpf_movein(uio, (int)d->bd_bif->bif_dlt, ifp, 1153 &m, &dst, &hlen, d); 1154 if (error) { 1155 counter_u64_add(d->bd_wdcount, 1); 1156 return (error); 1157 } 1158 counter_u64_add(d->bd_wfcount, 1); 1159 if (d->bd_hdrcmplt) 1160 dst.sa_family = pseudo_AF_HDRCMPLT; 1161 1162 if (d->bd_feedback) { 1163 mc = m_dup(m, M_NOWAIT); 1164 if (mc != NULL) 1165 mc->m_pkthdr.rcvif = ifp; 1166 /* Set M_PROMISC for outgoing packets to be discarded. */ 1167 if (d->bd_direction == BPF_D_INOUT) 1168 m->m_flags |= M_PROMISC; 1169 } else 1170 mc = NULL; 1171 1172 m->m_pkthdr.len -= hlen; 1173 m->m_len -= hlen; 1174 m->m_data += hlen; /* XXX */ 1175 1176 CURVNET_SET(ifp->if_vnet); 1177 #ifdef MAC 1178 BPFD_LOCK(d); 1179 mac_bpfdesc_create_mbuf(d, m); 1180 if (mc != NULL) 1181 mac_bpfdesc_create_mbuf(d, mc); 1182 BPFD_UNLOCK(d); 1183 #endif 1184 1185 bzero(&ro, sizeof(ro)); 1186 if (hlen != 0) { 1187 ro.ro_prepend = (u_char *)&dst.sa_data; 1188 ro.ro_plen = hlen; 1189 ro.ro_flags = RT_HAS_HEADER; 1190 } 1191 1192 error = (*ifp->if_output)(ifp, m, &dst, &ro); 1193 if (error) 1194 counter_u64_add(d->bd_wdcount, 1); 1195 1196 if (mc != NULL) { 1197 if (error == 0) 1198 (*ifp->if_input)(ifp, mc); 1199 else 1200 m_freem(mc); 1201 } 1202 CURVNET_RESTORE(); 1203 1204 return (error); 1205 } 1206 1207 /* 1208 * Reset a descriptor by flushing its packet buffer and clearing the receive 1209 * and drop counts. This is doable for kernel-only buffers, but with 1210 * zero-copy buffers, we can't write to (or rotate) buffers that are 1211 * currently owned by userspace. It would be nice if we could encapsulate 1212 * this logic in the buffer code rather than here. 1213 */ 1214 static void 1215 reset_d(struct bpf_d *d) 1216 { 1217 1218 BPFD_LOCK_ASSERT(d); 1219 1220 while (d->bd_hbuf_in_use) 1221 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1222 "bd_hbuf", 0); 1223 if ((d->bd_hbuf != NULL) && 1224 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1225 /* Free the hold buffer. */ 1226 d->bd_fbuf = d->bd_hbuf; 1227 d->bd_hbuf = NULL; 1228 d->bd_hlen = 0; 1229 bpf_buf_reclaimed(d); 1230 } 1231 if (bpf_canwritebuf(d)) 1232 d->bd_slen = 0; 1233 counter_u64_zero(d->bd_rcount); 1234 counter_u64_zero(d->bd_dcount); 1235 counter_u64_zero(d->bd_fcount); 1236 counter_u64_zero(d->bd_wcount); 1237 counter_u64_zero(d->bd_wfcount); 1238 counter_u64_zero(d->bd_wdcount); 1239 counter_u64_zero(d->bd_zcopy); 1240 } 1241 1242 /* 1243 * FIONREAD Check for read packet available. 1244 * BIOCGBLEN Get buffer len [for read()]. 1245 * BIOCSETF Set read filter. 1246 * BIOCSETFNR Set read filter without resetting descriptor. 1247 * BIOCSETWF Set write filter. 1248 * BIOCFLUSH Flush read packet buffer. 1249 * BIOCPROMISC Put interface into promiscuous mode. 1250 * BIOCGDLT Get link layer type. 1251 * BIOCGETIF Get interface name. 1252 * BIOCSETIF Set interface. 1253 * BIOCSRTIMEOUT Set read timeout. 1254 * BIOCGRTIMEOUT Get read timeout. 1255 * BIOCGSTATS Get packet stats. 1256 * BIOCIMMEDIATE Set immediate mode. 1257 * BIOCVERSION Get filter language version. 1258 * BIOCGHDRCMPLT Get "header already complete" flag 1259 * BIOCSHDRCMPLT Set "header already complete" flag 1260 * BIOCGDIRECTION Get packet direction flag 1261 * BIOCSDIRECTION Set packet direction flag 1262 * BIOCGTSTAMP Get time stamp format and resolution. 1263 * BIOCSTSTAMP Set time stamp format and resolution. 1264 * BIOCLOCK Set "locked" flag 1265 * BIOCFEEDBACK Set packet feedback mode. 1266 * BIOCSETZBUF Set current zero-copy buffer locations. 1267 * BIOCGETZMAX Get maximum zero-copy buffer size. 1268 * BIOCROTZBUF Force rotation of zero-copy buffer 1269 * BIOCSETBUFMODE Set buffer mode. 1270 * BIOCGETBUFMODE Get current buffer mode. 1271 */ 1272 /* ARGSUSED */ 1273 static int 1274 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1275 struct thread *td) 1276 { 1277 struct bpf_d *d; 1278 int error; 1279 1280 error = devfs_get_cdevpriv((void **)&d); 1281 if (error != 0) 1282 return (error); 1283 1284 /* 1285 * Refresh PID associated with this descriptor. 1286 */ 1287 BPFD_LOCK(d); 1288 BPF_PID_REFRESH(d, td); 1289 if (d->bd_state == BPF_WAITING) 1290 callout_stop(&d->bd_callout); 1291 d->bd_state = BPF_IDLE; 1292 BPFD_UNLOCK(d); 1293 1294 if (d->bd_locked == 1) { 1295 switch (cmd) { 1296 case BIOCGBLEN: 1297 case BIOCFLUSH: 1298 case BIOCGDLT: 1299 case BIOCGDLTLIST: 1300 #ifdef COMPAT_FREEBSD32 1301 case BIOCGDLTLIST32: 1302 #endif 1303 case BIOCGETIF: 1304 case BIOCGRTIMEOUT: 1305 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1306 case BIOCGRTIMEOUT32: 1307 #endif 1308 case BIOCGSTATS: 1309 case BIOCVERSION: 1310 case BIOCGRSIG: 1311 case BIOCGHDRCMPLT: 1312 case BIOCSTSTAMP: 1313 case BIOCFEEDBACK: 1314 case FIONREAD: 1315 case BIOCLOCK: 1316 case BIOCSRTIMEOUT: 1317 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1318 case BIOCSRTIMEOUT32: 1319 #endif 1320 case BIOCIMMEDIATE: 1321 case TIOCGPGRP: 1322 case BIOCROTZBUF: 1323 break; 1324 default: 1325 return (EPERM); 1326 } 1327 } 1328 #ifdef COMPAT_FREEBSD32 1329 /* 1330 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1331 * that it will get 32-bit packet headers. 1332 */ 1333 switch (cmd) { 1334 case BIOCSETF32: 1335 case BIOCSETFNR32: 1336 case BIOCSETWF32: 1337 case BIOCGDLTLIST32: 1338 case BIOCGRTIMEOUT32: 1339 case BIOCSRTIMEOUT32: 1340 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 1341 BPFD_LOCK(d); 1342 d->bd_compat32 = 1; 1343 BPFD_UNLOCK(d); 1344 } 1345 } 1346 #endif 1347 1348 CURVNET_SET(TD_TO_VNET(td)); 1349 switch (cmd) { 1350 1351 default: 1352 error = EINVAL; 1353 break; 1354 1355 /* 1356 * Check for read packet available. 1357 */ 1358 case FIONREAD: 1359 { 1360 int n; 1361 1362 BPFD_LOCK(d); 1363 n = d->bd_slen; 1364 while (d->bd_hbuf_in_use) 1365 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1366 PRINET, "bd_hbuf", 0); 1367 if (d->bd_hbuf) 1368 n += d->bd_hlen; 1369 BPFD_UNLOCK(d); 1370 1371 *(int *)addr = n; 1372 break; 1373 } 1374 1375 /* 1376 * Get buffer len [for read()]. 1377 */ 1378 case BIOCGBLEN: 1379 BPFD_LOCK(d); 1380 *(u_int *)addr = d->bd_bufsize; 1381 BPFD_UNLOCK(d); 1382 break; 1383 1384 /* 1385 * Set buffer length. 1386 */ 1387 case BIOCSBLEN: 1388 error = bpf_ioctl_sblen(d, (u_int *)addr); 1389 break; 1390 1391 /* 1392 * Set link layer read filter. 1393 */ 1394 case BIOCSETF: 1395 case BIOCSETFNR: 1396 case BIOCSETWF: 1397 #ifdef COMPAT_FREEBSD32 1398 case BIOCSETF32: 1399 case BIOCSETFNR32: 1400 case BIOCSETWF32: 1401 #endif 1402 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1403 break; 1404 1405 /* 1406 * Flush read packet buffer. 1407 */ 1408 case BIOCFLUSH: 1409 BPFD_LOCK(d); 1410 reset_d(d); 1411 BPFD_UNLOCK(d); 1412 break; 1413 1414 /* 1415 * Put interface into promiscuous mode. 1416 */ 1417 case BIOCPROMISC: 1418 if (d->bd_bif == NULL) { 1419 /* 1420 * No interface attached yet. 1421 */ 1422 error = EINVAL; 1423 break; 1424 } 1425 if (d->bd_promisc == 0) { 1426 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1427 if (error == 0) 1428 d->bd_promisc = 1; 1429 } 1430 break; 1431 1432 /* 1433 * Get current data link type. 1434 */ 1435 case BIOCGDLT: 1436 BPF_LOCK(); 1437 if (d->bd_bif == NULL) 1438 error = EINVAL; 1439 else 1440 *(u_int *)addr = d->bd_bif->bif_dlt; 1441 BPF_UNLOCK(); 1442 break; 1443 1444 /* 1445 * Get a list of supported data link types. 1446 */ 1447 #ifdef COMPAT_FREEBSD32 1448 case BIOCGDLTLIST32: 1449 { 1450 struct bpf_dltlist32 *list32; 1451 struct bpf_dltlist dltlist; 1452 1453 list32 = (struct bpf_dltlist32 *)addr; 1454 dltlist.bfl_len = list32->bfl_len; 1455 dltlist.bfl_list = PTRIN(list32->bfl_list); 1456 BPF_LOCK(); 1457 if (d->bd_bif == NULL) 1458 error = EINVAL; 1459 else { 1460 error = bpf_getdltlist(d, &dltlist); 1461 if (error == 0) 1462 list32->bfl_len = dltlist.bfl_len; 1463 } 1464 BPF_UNLOCK(); 1465 break; 1466 } 1467 #endif 1468 1469 case BIOCGDLTLIST: 1470 BPF_LOCK(); 1471 if (d->bd_bif == NULL) 1472 error = EINVAL; 1473 else 1474 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1475 BPF_UNLOCK(); 1476 break; 1477 1478 /* 1479 * Set data link type. 1480 */ 1481 case BIOCSDLT: 1482 BPF_LOCK(); 1483 if (d->bd_bif == NULL) 1484 error = EINVAL; 1485 else 1486 error = bpf_setdlt(d, *(u_int *)addr); 1487 BPF_UNLOCK(); 1488 break; 1489 1490 /* 1491 * Get interface name. 1492 */ 1493 case BIOCGETIF: 1494 BPF_LOCK(); 1495 if (d->bd_bif == NULL) 1496 error = EINVAL; 1497 else { 1498 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1499 struct ifreq *const ifr = (struct ifreq *)addr; 1500 1501 strlcpy(ifr->ifr_name, ifp->if_xname, 1502 sizeof(ifr->ifr_name)); 1503 } 1504 BPF_UNLOCK(); 1505 break; 1506 1507 /* 1508 * Set interface. 1509 */ 1510 case BIOCSETIF: 1511 { 1512 int alloc_buf, size; 1513 1514 /* 1515 * Behavior here depends on the buffering model. If 1516 * we're using kernel memory buffers, then we can 1517 * allocate them here. If we're using zero-copy, 1518 * then the user process must have registered buffers 1519 * by the time we get here. 1520 */ 1521 alloc_buf = 0; 1522 BPFD_LOCK(d); 1523 if (d->bd_bufmode == BPF_BUFMODE_BUFFER && 1524 d->bd_sbuf == NULL) 1525 alloc_buf = 1; 1526 BPFD_UNLOCK(d); 1527 if (alloc_buf) { 1528 size = d->bd_bufsize; 1529 error = bpf_buffer_ioctl_sblen(d, &size); 1530 if (error != 0) 1531 break; 1532 } 1533 BPF_LOCK(); 1534 error = bpf_setif(d, (struct ifreq *)addr); 1535 BPF_UNLOCK(); 1536 break; 1537 } 1538 1539 /* 1540 * Set read timeout. 1541 */ 1542 case BIOCSRTIMEOUT: 1543 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1544 case BIOCSRTIMEOUT32: 1545 #endif 1546 { 1547 struct timeval *tv = (struct timeval *)addr; 1548 #if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1549 struct timeval32 *tv32; 1550 struct timeval tv64; 1551 1552 if (cmd == BIOCSRTIMEOUT32) { 1553 tv32 = (struct timeval32 *)addr; 1554 tv = &tv64; 1555 tv->tv_sec = tv32->tv_sec; 1556 tv->tv_usec = tv32->tv_usec; 1557 } else 1558 #endif 1559 tv = (struct timeval *)addr; 1560 1561 /* 1562 * Subtract 1 tick from tvtohz() since this isn't 1563 * a one-shot timer. 1564 */ 1565 if ((error = itimerfix(tv)) == 0) 1566 d->bd_rtout = tvtohz(tv) - 1; 1567 break; 1568 } 1569 1570 /* 1571 * Get read timeout. 1572 */ 1573 case BIOCGRTIMEOUT: 1574 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1575 case BIOCGRTIMEOUT32: 1576 #endif 1577 { 1578 struct timeval *tv; 1579 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1580 struct timeval32 *tv32; 1581 struct timeval tv64; 1582 1583 if (cmd == BIOCGRTIMEOUT32) 1584 tv = &tv64; 1585 else 1586 #endif 1587 tv = (struct timeval *)addr; 1588 1589 tv->tv_sec = d->bd_rtout / hz; 1590 tv->tv_usec = (d->bd_rtout % hz) * tick; 1591 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1592 if (cmd == BIOCGRTIMEOUT32) { 1593 tv32 = (struct timeval32 *)addr; 1594 tv32->tv_sec = tv->tv_sec; 1595 tv32->tv_usec = tv->tv_usec; 1596 } 1597 #endif 1598 1599 break; 1600 } 1601 1602 /* 1603 * Get packet stats. 1604 */ 1605 case BIOCGSTATS: 1606 { 1607 struct bpf_stat *bs = (struct bpf_stat *)addr; 1608 1609 /* XXXCSJP overflow */ 1610 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount); 1611 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount); 1612 break; 1613 } 1614 1615 /* 1616 * Set immediate mode. 1617 */ 1618 case BIOCIMMEDIATE: 1619 BPFD_LOCK(d); 1620 d->bd_immediate = *(u_int *)addr; 1621 BPFD_UNLOCK(d); 1622 break; 1623 1624 case BIOCVERSION: 1625 { 1626 struct bpf_version *bv = (struct bpf_version *)addr; 1627 1628 bv->bv_major = BPF_MAJOR_VERSION; 1629 bv->bv_minor = BPF_MINOR_VERSION; 1630 break; 1631 } 1632 1633 /* 1634 * Get "header already complete" flag 1635 */ 1636 case BIOCGHDRCMPLT: 1637 BPFD_LOCK(d); 1638 *(u_int *)addr = d->bd_hdrcmplt; 1639 BPFD_UNLOCK(d); 1640 break; 1641 1642 /* 1643 * Set "header already complete" flag 1644 */ 1645 case BIOCSHDRCMPLT: 1646 BPFD_LOCK(d); 1647 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1648 BPFD_UNLOCK(d); 1649 break; 1650 1651 /* 1652 * Get packet direction flag 1653 */ 1654 case BIOCGDIRECTION: 1655 BPFD_LOCK(d); 1656 *(u_int *)addr = d->bd_direction; 1657 BPFD_UNLOCK(d); 1658 break; 1659 1660 /* 1661 * Set packet direction flag 1662 */ 1663 case BIOCSDIRECTION: 1664 { 1665 u_int direction; 1666 1667 direction = *(u_int *)addr; 1668 switch (direction) { 1669 case BPF_D_IN: 1670 case BPF_D_INOUT: 1671 case BPF_D_OUT: 1672 BPFD_LOCK(d); 1673 d->bd_direction = direction; 1674 BPFD_UNLOCK(d); 1675 break; 1676 default: 1677 error = EINVAL; 1678 } 1679 } 1680 break; 1681 1682 /* 1683 * Get packet timestamp format and resolution. 1684 */ 1685 case BIOCGTSTAMP: 1686 BPFD_LOCK(d); 1687 *(u_int *)addr = d->bd_tstamp; 1688 BPFD_UNLOCK(d); 1689 break; 1690 1691 /* 1692 * Set packet timestamp format and resolution. 1693 */ 1694 case BIOCSTSTAMP: 1695 { 1696 u_int func; 1697 1698 func = *(u_int *)addr; 1699 if (BPF_T_VALID(func)) 1700 d->bd_tstamp = func; 1701 else 1702 error = EINVAL; 1703 } 1704 break; 1705 1706 case BIOCFEEDBACK: 1707 BPFD_LOCK(d); 1708 d->bd_feedback = *(u_int *)addr; 1709 BPFD_UNLOCK(d); 1710 break; 1711 1712 case BIOCLOCK: 1713 BPFD_LOCK(d); 1714 d->bd_locked = 1; 1715 BPFD_UNLOCK(d); 1716 break; 1717 1718 case FIONBIO: /* Non-blocking I/O */ 1719 break; 1720 1721 case FIOASYNC: /* Send signal on receive packets */ 1722 BPFD_LOCK(d); 1723 d->bd_async = *(int *)addr; 1724 BPFD_UNLOCK(d); 1725 break; 1726 1727 case FIOSETOWN: 1728 /* 1729 * XXX: Add some sort of locking here? 1730 * fsetown() can sleep. 1731 */ 1732 error = fsetown(*(int *)addr, &d->bd_sigio); 1733 break; 1734 1735 case FIOGETOWN: 1736 BPFD_LOCK(d); 1737 *(int *)addr = fgetown(&d->bd_sigio); 1738 BPFD_UNLOCK(d); 1739 break; 1740 1741 /* This is deprecated, FIOSETOWN should be used instead. */ 1742 case TIOCSPGRP: 1743 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1744 break; 1745 1746 /* This is deprecated, FIOGETOWN should be used instead. */ 1747 case TIOCGPGRP: 1748 *(int *)addr = -fgetown(&d->bd_sigio); 1749 break; 1750 1751 case BIOCSRSIG: /* Set receive signal */ 1752 { 1753 u_int sig; 1754 1755 sig = *(u_int *)addr; 1756 1757 if (sig >= NSIG) 1758 error = EINVAL; 1759 else { 1760 BPFD_LOCK(d); 1761 d->bd_sig = sig; 1762 BPFD_UNLOCK(d); 1763 } 1764 break; 1765 } 1766 case BIOCGRSIG: 1767 BPFD_LOCK(d); 1768 *(u_int *)addr = d->bd_sig; 1769 BPFD_UNLOCK(d); 1770 break; 1771 1772 case BIOCGETBUFMODE: 1773 BPFD_LOCK(d); 1774 *(u_int *)addr = d->bd_bufmode; 1775 BPFD_UNLOCK(d); 1776 break; 1777 1778 case BIOCSETBUFMODE: 1779 /* 1780 * Allow the buffering mode to be changed as long as we 1781 * haven't yet committed to a particular mode. Our 1782 * definition of commitment, for now, is whether or not a 1783 * buffer has been allocated or an interface attached, since 1784 * that's the point where things get tricky. 1785 */ 1786 switch (*(u_int *)addr) { 1787 case BPF_BUFMODE_BUFFER: 1788 break; 1789 1790 case BPF_BUFMODE_ZBUF: 1791 if (bpf_zerocopy_enable) 1792 break; 1793 /* FALLSTHROUGH */ 1794 1795 default: 1796 CURVNET_RESTORE(); 1797 return (EINVAL); 1798 } 1799 1800 BPFD_LOCK(d); 1801 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1802 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1803 BPFD_UNLOCK(d); 1804 CURVNET_RESTORE(); 1805 return (EBUSY); 1806 } 1807 d->bd_bufmode = *(u_int *)addr; 1808 BPFD_UNLOCK(d); 1809 break; 1810 1811 case BIOCGETZMAX: 1812 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1813 break; 1814 1815 case BIOCSETZBUF: 1816 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1817 break; 1818 1819 case BIOCROTZBUF: 1820 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1821 break; 1822 } 1823 CURVNET_RESTORE(); 1824 return (error); 1825 } 1826 1827 /* 1828 * Set d's packet filter program to fp. If this file already has a filter, 1829 * free it and replace it. Returns EINVAL for bogus requests. 1830 * 1831 * Note we need global lock here to serialize bpf_setf() and bpf_setif() calls 1832 * since reading d->bd_bif can't be protected by d or interface lock due to 1833 * lock order. 1834 * 1835 * Additionally, we have to acquire interface write lock due to bpf_mtap() uses 1836 * interface read lock to read all filers. 1837 * 1838 */ 1839 static int 1840 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1841 { 1842 #ifdef COMPAT_FREEBSD32 1843 struct bpf_program fp_swab; 1844 struct bpf_program32 *fp32; 1845 #endif 1846 struct bpf_insn *fcode, *old; 1847 #ifdef BPF_JITTER 1848 bpf_jit_filter *jfunc, *ofunc; 1849 #endif 1850 size_t size; 1851 u_int flen; 1852 int need_upgrade; 1853 1854 #ifdef COMPAT_FREEBSD32 1855 switch (cmd) { 1856 case BIOCSETF32: 1857 case BIOCSETWF32: 1858 case BIOCSETFNR32: 1859 fp32 = (struct bpf_program32 *)fp; 1860 fp_swab.bf_len = fp32->bf_len; 1861 fp_swab.bf_insns = (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1862 fp = &fp_swab; 1863 switch (cmd) { 1864 case BIOCSETF32: 1865 cmd = BIOCSETF; 1866 break; 1867 case BIOCSETWF32: 1868 cmd = BIOCSETWF; 1869 break; 1870 } 1871 break; 1872 } 1873 #endif 1874 1875 fcode = NULL; 1876 #ifdef BPF_JITTER 1877 jfunc = ofunc = NULL; 1878 #endif 1879 need_upgrade = 0; 1880 1881 /* 1882 * Check new filter validness before acquiring any locks. 1883 * Allocate memory for new filter, if needed. 1884 */ 1885 flen = fp->bf_len; 1886 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1887 return (EINVAL); 1888 size = flen * sizeof(*fp->bf_insns); 1889 if (size > 0) { 1890 /* We're setting up new filter. Copy and check actual data. */ 1891 fcode = malloc(size, M_BPF, M_WAITOK); 1892 if (copyin(fp->bf_insns, fcode, size) != 0 || 1893 !bpf_validate(fcode, flen)) { 1894 free(fcode, M_BPF); 1895 return (EINVAL); 1896 } 1897 #ifdef BPF_JITTER 1898 /* Filter is copied inside fcode and is perfectly valid. */ 1899 jfunc = bpf_jitter(fcode, flen); 1900 #endif 1901 } 1902 1903 BPF_LOCK(); 1904 1905 /* 1906 * Set up new filter. 1907 * Protect filter change by interface lock. 1908 * Additionally, we are protected by global lock here. 1909 */ 1910 if (d->bd_bif != NULL) 1911 BPFIF_WLOCK(d->bd_bif); 1912 BPFD_LOCK(d); 1913 if (cmd == BIOCSETWF) { 1914 old = d->bd_wfilter; 1915 d->bd_wfilter = fcode; 1916 } else { 1917 old = d->bd_rfilter; 1918 d->bd_rfilter = fcode; 1919 #ifdef BPF_JITTER 1920 ofunc = d->bd_bfilter; 1921 d->bd_bfilter = jfunc; 1922 #endif 1923 if (cmd == BIOCSETF) 1924 reset_d(d); 1925 1926 need_upgrade = bpf_check_upgrade(cmd, d, fcode, flen); 1927 } 1928 BPFD_UNLOCK(d); 1929 if (d->bd_bif != NULL) 1930 BPFIF_WUNLOCK(d->bd_bif); 1931 if (old != NULL) 1932 free(old, M_BPF); 1933 #ifdef BPF_JITTER 1934 if (ofunc != NULL) 1935 bpf_destroy_jit_filter(ofunc); 1936 #endif 1937 1938 /* Move d to active readers list. */ 1939 if (need_upgrade != 0) 1940 bpf_upgraded(d); 1941 1942 BPF_UNLOCK(); 1943 return (0); 1944 } 1945 1946 /* 1947 * Detach a file from its current interface (if attached at all) and attach 1948 * to the interface indicated by the name stored in ifr. 1949 * Return an errno or 0. 1950 */ 1951 static int 1952 bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1953 { 1954 struct bpf_if *bp; 1955 struct ifnet *theywant; 1956 1957 BPF_LOCK_ASSERT(); 1958 1959 theywant = ifunit(ifr->ifr_name); 1960 if (theywant == NULL || theywant->if_bpf == NULL) 1961 return (ENXIO); 1962 1963 bp = theywant->if_bpf; 1964 1965 /* Check if interface is not being detached from BPF */ 1966 BPFIF_RLOCK(bp); 1967 if (bp->bif_flags & BPFIF_FLAG_DYING) { 1968 BPFIF_RUNLOCK(bp); 1969 return (ENXIO); 1970 } 1971 BPFIF_RUNLOCK(bp); 1972 1973 /* 1974 * At this point, we expect the buffer is already allocated. If not, 1975 * return an error. 1976 */ 1977 switch (d->bd_bufmode) { 1978 case BPF_BUFMODE_BUFFER: 1979 case BPF_BUFMODE_ZBUF: 1980 if (d->bd_sbuf == NULL) 1981 return (EINVAL); 1982 break; 1983 1984 default: 1985 panic("bpf_setif: bufmode %d", d->bd_bufmode); 1986 } 1987 if (bp != d->bd_bif) 1988 bpf_attachd(d, bp); 1989 BPFD_LOCK(d); 1990 reset_d(d); 1991 BPFD_UNLOCK(d); 1992 return (0); 1993 } 1994 1995 /* 1996 * Support for select() and poll() system calls 1997 * 1998 * Return true iff the specific operation will not block indefinitely. 1999 * Otherwise, return false but make a note that a selwakeup() must be done. 2000 */ 2001 static int 2002 bpfpoll(struct cdev *dev, int events, struct thread *td) 2003 { 2004 struct bpf_d *d; 2005 int revents; 2006 2007 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 2008 return (events & 2009 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 2010 2011 /* 2012 * Refresh PID associated with this descriptor. 2013 */ 2014 revents = events & (POLLOUT | POLLWRNORM); 2015 BPFD_LOCK(d); 2016 BPF_PID_REFRESH(d, td); 2017 if (events & (POLLIN | POLLRDNORM)) { 2018 if (bpf_ready(d)) 2019 revents |= events & (POLLIN | POLLRDNORM); 2020 else { 2021 selrecord(td, &d->bd_sel); 2022 /* Start the read timeout if necessary. */ 2023 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2024 callout_reset(&d->bd_callout, d->bd_rtout, 2025 bpf_timed_out, d); 2026 d->bd_state = BPF_WAITING; 2027 } 2028 } 2029 } 2030 BPFD_UNLOCK(d); 2031 return (revents); 2032 } 2033 2034 /* 2035 * Support for kevent() system call. Register EVFILT_READ filters and 2036 * reject all others. 2037 */ 2038 int 2039 bpfkqfilter(struct cdev *dev, struct knote *kn) 2040 { 2041 struct bpf_d *d; 2042 2043 if (devfs_get_cdevpriv((void **)&d) != 0 || 2044 kn->kn_filter != EVFILT_READ) 2045 return (1); 2046 2047 /* 2048 * Refresh PID associated with this descriptor. 2049 */ 2050 BPFD_LOCK(d); 2051 BPF_PID_REFRESH_CUR(d); 2052 kn->kn_fop = &bpfread_filtops; 2053 kn->kn_hook = d; 2054 knlist_add(&d->bd_sel.si_note, kn, 1); 2055 BPFD_UNLOCK(d); 2056 2057 return (0); 2058 } 2059 2060 static void 2061 filt_bpfdetach(struct knote *kn) 2062 { 2063 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2064 2065 knlist_remove(&d->bd_sel.si_note, kn, 0); 2066 } 2067 2068 static int 2069 filt_bpfread(struct knote *kn, long hint) 2070 { 2071 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2072 int ready; 2073 2074 BPFD_LOCK_ASSERT(d); 2075 ready = bpf_ready(d); 2076 if (ready) { 2077 kn->kn_data = d->bd_slen; 2078 /* 2079 * Ignore the hold buffer if it is being copied to user space. 2080 */ 2081 if (!d->bd_hbuf_in_use && d->bd_hbuf) 2082 kn->kn_data += d->bd_hlen; 2083 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2084 callout_reset(&d->bd_callout, d->bd_rtout, 2085 bpf_timed_out, d); 2086 d->bd_state = BPF_WAITING; 2087 } 2088 2089 return (ready); 2090 } 2091 2092 #define BPF_TSTAMP_NONE 0 2093 #define BPF_TSTAMP_FAST 1 2094 #define BPF_TSTAMP_NORMAL 2 2095 #define BPF_TSTAMP_EXTERN 3 2096 2097 static int 2098 bpf_ts_quality(int tstype) 2099 { 2100 2101 if (tstype == BPF_T_NONE) 2102 return (BPF_TSTAMP_NONE); 2103 if ((tstype & BPF_T_FAST) != 0) 2104 return (BPF_TSTAMP_FAST); 2105 2106 return (BPF_TSTAMP_NORMAL); 2107 } 2108 2109 static int 2110 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2111 { 2112 struct m_tag *tag; 2113 int quality; 2114 2115 quality = bpf_ts_quality(tstype); 2116 if (quality == BPF_TSTAMP_NONE) 2117 return (quality); 2118 2119 if (m != NULL) { 2120 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2121 if (tag != NULL) { 2122 *bt = *(struct bintime *)(tag + 1); 2123 return (BPF_TSTAMP_EXTERN); 2124 } 2125 } 2126 if (quality == BPF_TSTAMP_NORMAL) 2127 binuptime(bt); 2128 else 2129 getbinuptime(bt); 2130 2131 return (quality); 2132 } 2133 2134 /* 2135 * Incoming linkage from device drivers. Process the packet pkt, of length 2136 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2137 * by each process' filter, and if accepted, stashed into the corresponding 2138 * buffer. 2139 */ 2140 void 2141 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2142 { 2143 struct bintime bt; 2144 struct bpf_d *d; 2145 #ifdef BPF_JITTER 2146 bpf_jit_filter *bf; 2147 #endif 2148 u_int slen; 2149 int gottime; 2150 2151 gottime = BPF_TSTAMP_NONE; 2152 2153 BPFIF_RLOCK(bp); 2154 2155 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2156 /* 2157 * We are not using any locks for d here because: 2158 * 1) any filter change is protected by interface 2159 * write lock 2160 * 2) destroying/detaching d is protected by interface 2161 * write lock, too 2162 */ 2163 2164 counter_u64_add(d->bd_rcount, 1); 2165 /* 2166 * NB: We dont call BPF_CHECK_DIRECTION() here since there is no 2167 * way for the caller to indiciate to us whether this packet 2168 * is inbound or outbound. In the bpf_mtap() routines, we use 2169 * the interface pointers on the mbuf to figure it out. 2170 */ 2171 #ifdef BPF_JITTER 2172 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2173 if (bf != NULL) 2174 slen = (*(bf->func))(pkt, pktlen, pktlen); 2175 else 2176 #endif 2177 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2178 if (slen != 0) { 2179 /* 2180 * Filter matches. Let's to acquire write lock. 2181 */ 2182 BPFD_LOCK(d); 2183 2184 counter_u64_add(d->bd_fcount, 1); 2185 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2186 gottime = bpf_gettime(&bt, d->bd_tstamp, NULL); 2187 #ifdef MAC 2188 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2189 #endif 2190 catchpacket(d, pkt, pktlen, slen, 2191 bpf_append_bytes, &bt); 2192 BPFD_UNLOCK(d); 2193 } 2194 } 2195 BPFIF_RUNLOCK(bp); 2196 } 2197 2198 #define BPF_CHECK_DIRECTION(d, r, i) \ 2199 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2200 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2201 2202 /* 2203 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2204 * Locking model is explained in bpf_tap(). 2205 */ 2206 void 2207 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2208 { 2209 struct bintime bt; 2210 struct bpf_d *d; 2211 #ifdef BPF_JITTER 2212 bpf_jit_filter *bf; 2213 #endif 2214 u_int pktlen, slen; 2215 int gottime; 2216 2217 /* Skip outgoing duplicate packets. */ 2218 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2219 m->m_flags &= ~M_PROMISC; 2220 return; 2221 } 2222 2223 pktlen = m_length(m, NULL); 2224 gottime = BPF_TSTAMP_NONE; 2225 2226 BPFIF_RLOCK(bp); 2227 2228 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2229 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2230 continue; 2231 counter_u64_add(d->bd_rcount, 1); 2232 #ifdef BPF_JITTER 2233 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2234 /* XXX We cannot handle multiple mbufs. */ 2235 if (bf != NULL && m->m_next == NULL) 2236 slen = (*(bf->func))(mtod(m, u_char *), pktlen, pktlen); 2237 else 2238 #endif 2239 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2240 if (slen != 0) { 2241 BPFD_LOCK(d); 2242 2243 counter_u64_add(d->bd_fcount, 1); 2244 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2245 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2246 #ifdef MAC 2247 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2248 #endif 2249 catchpacket(d, (u_char *)m, pktlen, slen, 2250 bpf_append_mbuf, &bt); 2251 BPFD_UNLOCK(d); 2252 } 2253 } 2254 BPFIF_RUNLOCK(bp); 2255 } 2256 2257 /* 2258 * Incoming linkage from device drivers, when packet is in 2259 * an mbuf chain and to be prepended by a contiguous header. 2260 */ 2261 void 2262 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2263 { 2264 struct bintime bt; 2265 struct mbuf mb; 2266 struct bpf_d *d; 2267 u_int pktlen, slen; 2268 int gottime; 2269 2270 /* Skip outgoing duplicate packets. */ 2271 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2272 m->m_flags &= ~M_PROMISC; 2273 return; 2274 } 2275 2276 pktlen = m_length(m, NULL); 2277 /* 2278 * Craft on-stack mbuf suitable for passing to bpf_filter. 2279 * Note that we cut corners here; we only setup what's 2280 * absolutely needed--this mbuf should never go anywhere else. 2281 */ 2282 mb.m_next = m; 2283 mb.m_data = data; 2284 mb.m_len = dlen; 2285 pktlen += dlen; 2286 2287 gottime = BPF_TSTAMP_NONE; 2288 2289 BPFIF_RLOCK(bp); 2290 2291 LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2292 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2293 continue; 2294 counter_u64_add(d->bd_rcount, 1); 2295 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2296 if (slen != 0) { 2297 BPFD_LOCK(d); 2298 2299 counter_u64_add(d->bd_fcount, 1); 2300 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2301 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2302 #ifdef MAC 2303 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2304 #endif 2305 catchpacket(d, (u_char *)&mb, pktlen, slen, 2306 bpf_append_mbuf, &bt); 2307 BPFD_UNLOCK(d); 2308 } 2309 } 2310 BPFIF_RUNLOCK(bp); 2311 } 2312 2313 #undef BPF_CHECK_DIRECTION 2314 2315 #undef BPF_TSTAMP_NONE 2316 #undef BPF_TSTAMP_FAST 2317 #undef BPF_TSTAMP_NORMAL 2318 #undef BPF_TSTAMP_EXTERN 2319 2320 static int 2321 bpf_hdrlen(struct bpf_d *d) 2322 { 2323 int hdrlen; 2324 2325 hdrlen = d->bd_bif->bif_hdrlen; 2326 #ifndef BURN_BRIDGES 2327 if (d->bd_tstamp == BPF_T_NONE || 2328 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2329 #ifdef COMPAT_FREEBSD32 2330 if (d->bd_compat32) 2331 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2332 else 2333 #endif 2334 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2335 else 2336 #endif 2337 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2338 #ifdef COMPAT_FREEBSD32 2339 if (d->bd_compat32) 2340 hdrlen = BPF_WORDALIGN32(hdrlen); 2341 else 2342 #endif 2343 hdrlen = BPF_WORDALIGN(hdrlen); 2344 2345 return (hdrlen - d->bd_bif->bif_hdrlen); 2346 } 2347 2348 static void 2349 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2350 { 2351 struct bintime bt2, boottimebin; 2352 struct timeval tsm; 2353 struct timespec tsn; 2354 2355 if ((tstype & BPF_T_MONOTONIC) == 0) { 2356 bt2 = *bt; 2357 getboottimebin(&boottimebin); 2358 bintime_add(&bt2, &boottimebin); 2359 bt = &bt2; 2360 } 2361 switch (BPF_T_FORMAT(tstype)) { 2362 case BPF_T_MICROTIME: 2363 bintime2timeval(bt, &tsm); 2364 ts->bt_sec = tsm.tv_sec; 2365 ts->bt_frac = tsm.tv_usec; 2366 break; 2367 case BPF_T_NANOTIME: 2368 bintime2timespec(bt, &tsn); 2369 ts->bt_sec = tsn.tv_sec; 2370 ts->bt_frac = tsn.tv_nsec; 2371 break; 2372 case BPF_T_BINTIME: 2373 ts->bt_sec = bt->sec; 2374 ts->bt_frac = bt->frac; 2375 break; 2376 } 2377 } 2378 2379 /* 2380 * Move the packet data from interface memory (pkt) into the 2381 * store buffer. "cpfn" is the routine called to do the actual data 2382 * transfer. bcopy is passed in to copy contiguous chunks, while 2383 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2384 * pkt is really an mbuf. 2385 */ 2386 static void 2387 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2388 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2389 struct bintime *bt) 2390 { 2391 struct bpf_xhdr hdr; 2392 #ifndef BURN_BRIDGES 2393 struct bpf_hdr hdr_old; 2394 #ifdef COMPAT_FREEBSD32 2395 struct bpf_hdr32 hdr32_old; 2396 #endif 2397 #endif 2398 int caplen, curlen, hdrlen, totlen; 2399 int do_wakeup = 0; 2400 int do_timestamp; 2401 int tstype; 2402 2403 BPFD_LOCK_ASSERT(d); 2404 2405 /* 2406 * Detect whether user space has released a buffer back to us, and if 2407 * so, move it from being a hold buffer to a free buffer. This may 2408 * not be the best place to do it (for example, we might only want to 2409 * run this check if we need the space), but for now it's a reliable 2410 * spot to do it. 2411 */ 2412 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2413 d->bd_fbuf = d->bd_hbuf; 2414 d->bd_hbuf = NULL; 2415 d->bd_hlen = 0; 2416 bpf_buf_reclaimed(d); 2417 } 2418 2419 /* 2420 * Figure out how many bytes to move. If the packet is 2421 * greater or equal to the snapshot length, transfer that 2422 * much. Otherwise, transfer the whole packet (unless 2423 * we hit the buffer size limit). 2424 */ 2425 hdrlen = bpf_hdrlen(d); 2426 totlen = hdrlen + min(snaplen, pktlen); 2427 if (totlen > d->bd_bufsize) 2428 totlen = d->bd_bufsize; 2429 2430 /* 2431 * Round up the end of the previous packet to the next longword. 2432 * 2433 * Drop the packet if there's no room and no hope of room 2434 * If the packet would overflow the storage buffer or the storage 2435 * buffer is considered immutable by the buffer model, try to rotate 2436 * the buffer and wakeup pending processes. 2437 */ 2438 #ifdef COMPAT_FREEBSD32 2439 if (d->bd_compat32) 2440 curlen = BPF_WORDALIGN32(d->bd_slen); 2441 else 2442 #endif 2443 curlen = BPF_WORDALIGN(d->bd_slen); 2444 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2445 if (d->bd_fbuf == NULL) { 2446 /* 2447 * There's no room in the store buffer, and no 2448 * prospect of room, so drop the packet. Notify the 2449 * buffer model. 2450 */ 2451 bpf_buffull(d); 2452 counter_u64_add(d->bd_dcount, 1); 2453 return; 2454 } 2455 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); 2456 ROTATE_BUFFERS(d); 2457 do_wakeup = 1; 2458 curlen = 0; 2459 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2460 /* 2461 * Immediate mode is set, or the read timeout has already 2462 * expired during a select call. A packet arrived, so the 2463 * reader should be woken up. 2464 */ 2465 do_wakeup = 1; 2466 caplen = totlen - hdrlen; 2467 tstype = d->bd_tstamp; 2468 do_timestamp = tstype != BPF_T_NONE; 2469 #ifndef BURN_BRIDGES 2470 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2471 struct bpf_ts ts; 2472 if (do_timestamp) 2473 bpf_bintime2ts(bt, &ts, tstype); 2474 #ifdef COMPAT_FREEBSD32 2475 if (d->bd_compat32) { 2476 bzero(&hdr32_old, sizeof(hdr32_old)); 2477 if (do_timestamp) { 2478 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2479 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2480 } 2481 hdr32_old.bh_datalen = pktlen; 2482 hdr32_old.bh_hdrlen = hdrlen; 2483 hdr32_old.bh_caplen = caplen; 2484 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2485 sizeof(hdr32_old)); 2486 goto copy; 2487 } 2488 #endif 2489 bzero(&hdr_old, sizeof(hdr_old)); 2490 if (do_timestamp) { 2491 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2492 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2493 } 2494 hdr_old.bh_datalen = pktlen; 2495 hdr_old.bh_hdrlen = hdrlen; 2496 hdr_old.bh_caplen = caplen; 2497 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2498 sizeof(hdr_old)); 2499 goto copy; 2500 } 2501 #endif 2502 2503 /* 2504 * Append the bpf header. Note we append the actual header size, but 2505 * move forward the length of the header plus padding. 2506 */ 2507 bzero(&hdr, sizeof(hdr)); 2508 if (do_timestamp) 2509 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2510 hdr.bh_datalen = pktlen; 2511 hdr.bh_hdrlen = hdrlen; 2512 hdr.bh_caplen = caplen; 2513 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2514 2515 /* 2516 * Copy the packet data into the store buffer and update its length. 2517 */ 2518 #ifndef BURN_BRIDGES 2519 copy: 2520 #endif 2521 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2522 d->bd_slen = curlen + totlen; 2523 2524 if (do_wakeup) 2525 bpf_wakeup(d); 2526 } 2527 2528 /* 2529 * Free buffers currently in use by a descriptor. 2530 * Called on close. 2531 */ 2532 static void 2533 bpf_freed(struct bpf_d *d) 2534 { 2535 2536 /* 2537 * We don't need to lock out interrupts since this descriptor has 2538 * been detached from its interface and it yet hasn't been marked 2539 * free. 2540 */ 2541 bpf_free(d); 2542 if (d->bd_rfilter != NULL) { 2543 free((caddr_t)d->bd_rfilter, M_BPF); 2544 #ifdef BPF_JITTER 2545 if (d->bd_bfilter != NULL) 2546 bpf_destroy_jit_filter(d->bd_bfilter); 2547 #endif 2548 } 2549 if (d->bd_wfilter != NULL) 2550 free((caddr_t)d->bd_wfilter, M_BPF); 2551 mtx_destroy(&d->bd_lock); 2552 2553 counter_u64_free(d->bd_rcount); 2554 counter_u64_free(d->bd_dcount); 2555 counter_u64_free(d->bd_fcount); 2556 counter_u64_free(d->bd_wcount); 2557 counter_u64_free(d->bd_wfcount); 2558 counter_u64_free(d->bd_wdcount); 2559 counter_u64_free(d->bd_zcopy); 2560 2561 } 2562 2563 /* 2564 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2565 * fixed size of the link header (variable length headers not yet supported). 2566 */ 2567 void 2568 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2569 { 2570 2571 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2572 } 2573 2574 /* 2575 * Attach an interface to bpf. ifp is a pointer to the structure 2576 * defining the interface to be attached, dlt is the link layer type, 2577 * and hdrlen is the fixed size of the link header (variable length 2578 * headers are not yet supporrted). 2579 */ 2580 void 2581 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2582 { 2583 struct bpf_if *bp; 2584 2585 bp = malloc(sizeof(*bp), M_BPF, M_NOWAIT | M_ZERO); 2586 if (bp == NULL) 2587 panic("bpfattach"); 2588 2589 LIST_INIT(&bp->bif_dlist); 2590 LIST_INIT(&bp->bif_wlist); 2591 bp->bif_ifp = ifp; 2592 bp->bif_dlt = dlt; 2593 rw_init(&bp->bif_lock, "bpf interface lock"); 2594 KASSERT(*driverp == NULL, ("bpfattach2: driverp already initialized")); 2595 bp->bif_bpf = driverp; 2596 *driverp = bp; 2597 2598 BPF_LOCK(); 2599 LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2600 BPF_UNLOCK(); 2601 2602 bp->bif_hdrlen = hdrlen; 2603 2604 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 2605 if_printf(ifp, "bpf attached\n"); 2606 } 2607 2608 #ifdef VIMAGE 2609 /* 2610 * When moving interfaces between vnet instances we need a way to 2611 * query the dlt and hdrlen before detach so we can re-attch the if_bpf 2612 * after the vmove. We unfortunately have no device driver infrastructure 2613 * to query the interface for these values after creation/attach, thus 2614 * add this as a workaround. 2615 */ 2616 int 2617 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen) 2618 { 2619 2620 if (bp == NULL) 2621 return (ENXIO); 2622 if (bif_dlt == NULL && bif_hdrlen == NULL) 2623 return (0); 2624 2625 if (bif_dlt != NULL) 2626 *bif_dlt = bp->bif_dlt; 2627 if (bif_hdrlen != NULL) 2628 *bif_hdrlen = bp->bif_hdrlen; 2629 2630 return (0); 2631 } 2632 #endif 2633 2634 /* 2635 * Detach bpf from an interface. This involves detaching each descriptor 2636 * associated with the interface. Notify each descriptor as it's detached 2637 * so that any sleepers wake up and get ENXIO. 2638 */ 2639 void 2640 bpfdetach(struct ifnet *ifp) 2641 { 2642 struct bpf_if *bp, *bp_temp; 2643 struct bpf_d *d; 2644 int ndetached; 2645 2646 ndetached = 0; 2647 2648 BPF_LOCK(); 2649 /* Find all bpf_if struct's which reference ifp and detach them. */ 2650 LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2651 if (ifp != bp->bif_ifp) 2652 continue; 2653 2654 LIST_REMOVE(bp, bif_next); 2655 /* Add to to-be-freed list */ 2656 LIST_INSERT_HEAD(&bpf_freelist, bp, bif_next); 2657 2658 ndetached++; 2659 /* 2660 * Delay freeing bp till interface is detached 2661 * and all routes through this interface are removed. 2662 * Mark bp as detached to restrict new consumers. 2663 */ 2664 BPFIF_WLOCK(bp); 2665 bp->bif_flags |= BPFIF_FLAG_DYING; 2666 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if; 2667 BPFIF_WUNLOCK(bp); 2668 2669 CTR4(KTR_NET, "%s: sheduling free for encap %d (%p) for if %p", 2670 __func__, bp->bif_dlt, bp, ifp); 2671 2672 /* Free common descriptors */ 2673 while ((d = LIST_FIRST(&bp->bif_dlist)) != NULL) { 2674 bpf_detachd_locked(d); 2675 BPFD_LOCK(d); 2676 bpf_wakeup(d); 2677 BPFD_UNLOCK(d); 2678 } 2679 2680 /* Free writer-only descriptors */ 2681 while ((d = LIST_FIRST(&bp->bif_wlist)) != NULL) { 2682 bpf_detachd_locked(d); 2683 BPFD_LOCK(d); 2684 bpf_wakeup(d); 2685 BPFD_UNLOCK(d); 2686 } 2687 } 2688 BPF_UNLOCK(); 2689 2690 #ifdef INVARIANTS 2691 if (ndetached == 0) 2692 printf("bpfdetach: %s was not attached\n", ifp->if_xname); 2693 #endif 2694 } 2695 2696 /* 2697 * Interface departure handler. 2698 * Note departure event does not guarantee interface is going down. 2699 * Interface renaming is currently done via departure/arrival event set. 2700 * 2701 * Departure handled is called after all routes pointing to 2702 * given interface are removed and interface is in down state 2703 * restricting any packets to be sent/received. We assume it is now safe 2704 * to free data allocated by BPF. 2705 */ 2706 static void 2707 bpf_ifdetach(void *arg __unused, struct ifnet *ifp) 2708 { 2709 struct bpf_if *bp, *bp_temp; 2710 int nmatched = 0; 2711 2712 /* Ignore ifnet renaming. */ 2713 if (ifp->if_flags & IFF_RENAMING) 2714 return; 2715 2716 BPF_LOCK(); 2717 /* 2718 * Find matching entries in free list. 2719 * Nothing should be found if bpfdetach() was not called. 2720 */ 2721 LIST_FOREACH_SAFE(bp, &bpf_freelist, bif_next, bp_temp) { 2722 if (ifp != bp->bif_ifp) 2723 continue; 2724 2725 CTR3(KTR_NET, "%s: freeing BPF instance %p for interface %p", 2726 __func__, bp, ifp); 2727 2728 LIST_REMOVE(bp, bif_next); 2729 2730 rw_destroy(&bp->bif_lock); 2731 free(bp, M_BPF); 2732 2733 nmatched++; 2734 } 2735 BPF_UNLOCK(); 2736 } 2737 2738 /* 2739 * Get a list of available data link type of the interface. 2740 */ 2741 static int 2742 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2743 { 2744 struct ifnet *ifp; 2745 struct bpf_if *bp; 2746 u_int *lst; 2747 int error, n, n1; 2748 2749 BPF_LOCK_ASSERT(); 2750 2751 ifp = d->bd_bif->bif_ifp; 2752 again: 2753 n1 = 0; 2754 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2755 if (bp->bif_ifp == ifp) 2756 n1++; 2757 } 2758 if (bfl->bfl_list == NULL) { 2759 bfl->bfl_len = n1; 2760 return (0); 2761 } 2762 if (n1 > bfl->bfl_len) 2763 return (ENOMEM); 2764 BPF_UNLOCK(); 2765 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); 2766 n = 0; 2767 BPF_LOCK(); 2768 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2769 if (bp->bif_ifp != ifp) 2770 continue; 2771 if (n >= n1) { 2772 free(lst, M_TEMP); 2773 goto again; 2774 } 2775 lst[n] = bp->bif_dlt; 2776 n++; 2777 } 2778 BPF_UNLOCK(); 2779 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); 2780 free(lst, M_TEMP); 2781 BPF_LOCK(); 2782 bfl->bfl_len = n; 2783 return (error); 2784 } 2785 2786 /* 2787 * Set the data link type of a BPF instance. 2788 */ 2789 static int 2790 bpf_setdlt(struct bpf_d *d, u_int dlt) 2791 { 2792 int error, opromisc; 2793 struct ifnet *ifp; 2794 struct bpf_if *bp; 2795 2796 BPF_LOCK_ASSERT(); 2797 2798 if (d->bd_bif->bif_dlt == dlt) 2799 return (0); 2800 ifp = d->bd_bif->bif_ifp; 2801 2802 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2803 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2804 break; 2805 } 2806 2807 if (bp != NULL) { 2808 opromisc = d->bd_promisc; 2809 bpf_attachd(d, bp); 2810 BPFD_LOCK(d); 2811 reset_d(d); 2812 BPFD_UNLOCK(d); 2813 if (opromisc) { 2814 error = ifpromisc(bp->bif_ifp, 1); 2815 if (error) 2816 if_printf(bp->bif_ifp, 2817 "bpf_setdlt: ifpromisc failed (%d)\n", 2818 error); 2819 else 2820 d->bd_promisc = 1; 2821 } 2822 } 2823 return (bp == NULL ? EINVAL : 0); 2824 } 2825 2826 static void 2827 bpf_drvinit(void *unused) 2828 { 2829 struct cdev *dev; 2830 2831 sx_init(&bpf_sx, "bpf global lock"); 2832 LIST_INIT(&bpf_iflist); 2833 LIST_INIT(&bpf_freelist); 2834 2835 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2836 /* For compatibility */ 2837 make_dev_alias(dev, "bpf0"); 2838 2839 /* Register interface departure handler */ 2840 bpf_ifdetach_cookie = EVENTHANDLER_REGISTER( 2841 ifnet_departure_event, bpf_ifdetach, NULL, 2842 EVENTHANDLER_PRI_ANY); 2843 } 2844 2845 /* 2846 * Zero out the various packet counters associated with all of the bpf 2847 * descriptors. At some point, we will probably want to get a bit more 2848 * granular and allow the user to specify descriptors to be zeroed. 2849 */ 2850 static void 2851 bpf_zero_counters(void) 2852 { 2853 struct bpf_if *bp; 2854 struct bpf_d *bd; 2855 2856 BPF_LOCK(); 2857 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2858 BPFIF_RLOCK(bp); 2859 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2860 BPFD_LOCK(bd); 2861 counter_u64_zero(bd->bd_rcount); 2862 counter_u64_zero(bd->bd_dcount); 2863 counter_u64_zero(bd->bd_fcount); 2864 counter_u64_zero(bd->bd_wcount); 2865 counter_u64_zero(bd->bd_wfcount); 2866 counter_u64_zero(bd->bd_zcopy); 2867 BPFD_UNLOCK(bd); 2868 } 2869 BPFIF_RUNLOCK(bp); 2870 } 2871 BPF_UNLOCK(); 2872 } 2873 2874 /* 2875 * Fill filter statistics 2876 */ 2877 static void 2878 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2879 { 2880 2881 bzero(d, sizeof(*d)); 2882 BPFD_LOCK_ASSERT(bd); 2883 d->bd_structsize = sizeof(*d); 2884 /* XXX: reading should be protected by global lock */ 2885 d->bd_immediate = bd->bd_immediate; 2886 d->bd_promisc = bd->bd_promisc; 2887 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2888 d->bd_direction = bd->bd_direction; 2889 d->bd_feedback = bd->bd_feedback; 2890 d->bd_async = bd->bd_async; 2891 d->bd_rcount = counter_u64_fetch(bd->bd_rcount); 2892 d->bd_dcount = counter_u64_fetch(bd->bd_dcount); 2893 d->bd_fcount = counter_u64_fetch(bd->bd_fcount); 2894 d->bd_sig = bd->bd_sig; 2895 d->bd_slen = bd->bd_slen; 2896 d->bd_hlen = bd->bd_hlen; 2897 d->bd_bufsize = bd->bd_bufsize; 2898 d->bd_pid = bd->bd_pid; 2899 strlcpy(d->bd_ifname, 2900 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2901 d->bd_locked = bd->bd_locked; 2902 d->bd_wcount = counter_u64_fetch(bd->bd_wcount); 2903 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount); 2904 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount); 2905 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy); 2906 d->bd_bufmode = bd->bd_bufmode; 2907 } 2908 2909 /* 2910 * Handle `netstat -B' stats request 2911 */ 2912 static int 2913 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2914 { 2915 static const struct xbpf_d zerostats; 2916 struct xbpf_d *xbdbuf, *xbd, tempstats; 2917 int index, error; 2918 struct bpf_if *bp; 2919 struct bpf_d *bd; 2920 2921 /* 2922 * XXX This is not technically correct. It is possible for non 2923 * privileged users to open bpf devices. It would make sense 2924 * if the users who opened the devices were able to retrieve 2925 * the statistics for them, too. 2926 */ 2927 error = priv_check(req->td, PRIV_NET_BPF); 2928 if (error) 2929 return (error); 2930 /* 2931 * Check to see if the user is requesting that the counters be 2932 * zeroed out. Explicitly check that the supplied data is zeroed, 2933 * as we aren't allowing the user to set the counters currently. 2934 */ 2935 if (req->newptr != NULL) { 2936 if (req->newlen != sizeof(tempstats)) 2937 return (EINVAL); 2938 memset(&tempstats, 0, sizeof(tempstats)); 2939 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); 2940 if (error) 2941 return (error); 2942 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) 2943 return (EINVAL); 2944 bpf_zero_counters(); 2945 return (0); 2946 } 2947 if (req->oldptr == NULL) 2948 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2949 if (bpf_bpfd_cnt == 0) 2950 return (SYSCTL_OUT(req, 0, 0)); 2951 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2952 BPF_LOCK(); 2953 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2954 BPF_UNLOCK(); 2955 free(xbdbuf, M_BPF); 2956 return (ENOMEM); 2957 } 2958 index = 0; 2959 LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2960 BPFIF_RLOCK(bp); 2961 /* Send writers-only first */ 2962 LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2963 xbd = &xbdbuf[index++]; 2964 BPFD_LOCK(bd); 2965 bpfstats_fill_xbpf(xbd, bd); 2966 BPFD_UNLOCK(bd); 2967 } 2968 LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2969 xbd = &xbdbuf[index++]; 2970 BPFD_LOCK(bd); 2971 bpfstats_fill_xbpf(xbd, bd); 2972 BPFD_UNLOCK(bd); 2973 } 2974 BPFIF_RUNLOCK(bp); 2975 } 2976 BPF_UNLOCK(); 2977 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2978 free(xbdbuf, M_BPF); 2979 return (error); 2980 } 2981 2982 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2983 2984 #else /* !DEV_BPF && !NETGRAPH_BPF */ 2985 2986 /* 2987 * NOP stubs to allow bpf-using drivers to load and function. 2988 * 2989 * A 'better' implementation would allow the core bpf functionality 2990 * to be loaded at runtime. 2991 */ 2992 2993 void 2994 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2995 { 2996 } 2997 2998 void 2999 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 3000 { 3001 } 3002 3003 void 3004 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 3005 { 3006 } 3007 3008 void 3009 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 3010 { 3011 3012 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 3013 } 3014 3015 void 3016 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 3017 { 3018 3019 *driverp = (struct bpf_if *)&dead_bpf_if; 3020 } 3021 3022 void 3023 bpfdetach(struct ifnet *ifp) 3024 { 3025 } 3026 3027 u_int 3028 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 3029 { 3030 return -1; /* "no filter" behaviour */ 3031 } 3032 3033 int 3034 bpf_validate(const struct bpf_insn *f, int len) 3035 { 3036 return 0; /* false */ 3037 } 3038 3039 #endif /* !DEV_BPF && !NETGRAPH_BPF */ 3040 3041 #ifdef DDB 3042 static void 3043 bpf_show_bpf_if(struct bpf_if *bpf_if) 3044 { 3045 3046 if (bpf_if == NULL) 3047 return; 3048 db_printf("%p:\n", bpf_if); 3049 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e); 3050 /* bif_ext.bif_next */ 3051 /* bif_ext.bif_dlist */ 3052 BPF_DB_PRINTF("%#x", bif_dlt); 3053 BPF_DB_PRINTF("%u", bif_hdrlen); 3054 BPF_DB_PRINTF("%p", bif_ifp); 3055 /* bif_lock */ 3056 /* bif_wlist */ 3057 BPF_DB_PRINTF("%#x", bif_flags); 3058 } 3059 3060 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if) 3061 { 3062 3063 if (!have_addr) { 3064 db_printf("usage: show bpf_if <struct bpf_if *>\n"); 3065 return; 3066 } 3067 3068 bpf_show_bpf_if((struct bpf_if *)addr); 3069 } 3070 #endif 3071