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 * Copyright (c) 2019 Andrey V. Elsukov <ae@FreeBSD.org> 7 * 8 * This code is derived from the Stanford/CMU enet packet filter, 9 * (net/enet.c) distributed as part of 4.3BSD, and code contributed 10 * to Berkeley by Steven McCanne and Van Jacobson both of Lawrence 11 * Berkeley Laboratory. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 3. Neither the name of the University nor the names of its contributors 22 * may be used to endorse or promote products derived from this software 23 * without specific prior written permission. 24 * 25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 35 * SUCH DAMAGE. 36 * 37 * @(#)bpf.c 8.4 (Berkeley) 1/9/95 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_bpf.h" 44 #include "opt_ddb.h" 45 #include "opt_netgraph.h" 46 47 #include <sys/types.h> 48 #include <sys/param.h> 49 #include <sys/lock.h> 50 #include <sys/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 = CK_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 bpfd_list bif_wlist; /* writer-only list */ 112 struct ifnet *bif_ifp; /* corresponding interface */ 113 struct bpf_if **bif_bpf; /* Pointer to pointer to us */ 114 volatile u_int bif_refcnt; 115 struct epoch_context epoch_ctx; 116 }; 117 118 CTASSERT(offsetof(struct bpf_if, bif_ext) == 0); 119 120 struct bpf_program_buffer { 121 struct epoch_context epoch_ctx; 122 #ifdef BPF_JITTER 123 bpf_jit_filter *func; 124 #endif 125 void *buffer[0]; 126 }; 127 128 #if defined(DEV_BPF) || defined(NETGRAPH_BPF) 129 130 #define PRINET 26 /* interruptible */ 131 132 #define SIZEOF_BPF_HDR(type) \ 133 (offsetof(type, bh_hdrlen) + sizeof(((type *)0)->bh_hdrlen)) 134 135 #ifdef COMPAT_FREEBSD32 136 #include <sys/mount.h> 137 #include <compat/freebsd32/freebsd32.h> 138 #define BPF_ALIGNMENT32 sizeof(int32_t) 139 #define BPF_WORDALIGN32(x) roundup2(x, BPF_ALIGNMENT32) 140 141 #ifndef BURN_BRIDGES 142 /* 143 * 32-bit version of structure prepended to each packet. We use this header 144 * instead of the standard one for 32-bit streams. We mark the a stream as 145 * 32-bit the first time we see a 32-bit compat ioctl request. 146 */ 147 struct bpf_hdr32 { 148 struct timeval32 bh_tstamp; /* time stamp */ 149 uint32_t bh_caplen; /* length of captured portion */ 150 uint32_t bh_datalen; /* original length of packet */ 151 uint16_t bh_hdrlen; /* length of bpf header (this struct 152 plus alignment padding) */ 153 }; 154 #endif 155 156 struct bpf_program32 { 157 u_int bf_len; 158 uint32_t bf_insns; 159 }; 160 161 struct bpf_dltlist32 { 162 u_int bfl_len; 163 u_int bfl_list; 164 }; 165 166 #define BIOCSETF32 _IOW('B', 103, struct bpf_program32) 167 #define BIOCSRTIMEOUT32 _IOW('B', 109, struct timeval32) 168 #define BIOCGRTIMEOUT32 _IOR('B', 110, struct timeval32) 169 #define BIOCGDLTLIST32 _IOWR('B', 121, struct bpf_dltlist32) 170 #define BIOCSETWF32 _IOW('B', 123, struct bpf_program32) 171 #define BIOCSETFNR32 _IOW('B', 130, struct bpf_program32) 172 #endif 173 174 #define BPF_LOCK() sx_xlock(&bpf_sx) 175 #define BPF_UNLOCK() sx_xunlock(&bpf_sx) 176 #define BPF_LOCK_ASSERT() sx_assert(&bpf_sx, SA_XLOCKED) 177 /* 178 * bpf_iflist is a list of BPF interface structures, each corresponding to a 179 * specific DLT. The same network interface might have several BPF interface 180 * structures registered by different layers in the stack (i.e., 802.11 181 * frames, ethernet frames, etc). 182 */ 183 CK_LIST_HEAD(bpf_iflist, bpf_if); 184 static struct bpf_iflist bpf_iflist; 185 static struct sx bpf_sx; /* bpf global lock */ 186 static int bpf_bpfd_cnt; 187 188 static void bpfif_ref(struct bpf_if *); 189 static void bpfif_rele(struct bpf_if *); 190 191 static void bpfd_ref(struct bpf_d *); 192 static void bpfd_rele(struct bpf_d *); 193 static void bpf_attachd(struct bpf_d *, struct bpf_if *); 194 static void bpf_detachd(struct bpf_d *); 195 static void bpf_detachd_locked(struct bpf_d *, bool); 196 static void bpfd_free(epoch_context_t); 197 static int bpf_movein(struct uio *, int, struct ifnet *, struct mbuf **, 198 struct sockaddr *, int *, struct bpf_d *); 199 static int bpf_setif(struct bpf_d *, struct ifreq *); 200 static void bpf_timed_out(void *); 201 static __inline void 202 bpf_wakeup(struct bpf_d *); 203 static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 204 void (*)(struct bpf_d *, caddr_t, u_int, void *, u_int), 205 struct bintime *); 206 static void reset_d(struct bpf_d *); 207 static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long cmd); 208 static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 209 static int bpf_setdlt(struct bpf_d *, u_int); 210 static void filt_bpfdetach(struct knote *); 211 static int filt_bpfread(struct knote *, long); 212 static void bpf_drvinit(void *); 213 static int bpf_stats_sysctl(SYSCTL_HANDLER_ARGS); 214 215 SYSCTL_NODE(_net, OID_AUTO, bpf, CTLFLAG_RW, 0, "bpf sysctl"); 216 int bpf_maxinsns = BPF_MAXINSNS; 217 SYSCTL_INT(_net_bpf, OID_AUTO, maxinsns, CTLFLAG_RW, 218 &bpf_maxinsns, 0, "Maximum bpf program instructions"); 219 static int bpf_zerocopy_enable = 0; 220 SYSCTL_INT(_net_bpf, OID_AUTO, zerocopy_enable, CTLFLAG_RW, 221 &bpf_zerocopy_enable, 0, "Enable new zero-copy BPF buffer sessions"); 222 static SYSCTL_NODE(_net_bpf, OID_AUTO, stats, CTLFLAG_MPSAFE | CTLFLAG_RW, 223 bpf_stats_sysctl, "bpf statistics portal"); 224 225 VNET_DEFINE_STATIC(int, bpf_optimize_writers) = 0; 226 #define V_bpf_optimize_writers VNET(bpf_optimize_writers) 227 SYSCTL_INT(_net_bpf, OID_AUTO, optimize_writers, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(bpf_optimize_writers), 0, 229 "Do not send packets until BPF program is set"); 230 231 static d_open_t bpfopen; 232 static d_read_t bpfread; 233 static d_write_t bpfwrite; 234 static d_ioctl_t bpfioctl; 235 static d_poll_t bpfpoll; 236 static d_kqfilter_t bpfkqfilter; 237 238 static struct cdevsw bpf_cdevsw = { 239 .d_version = D_VERSION, 240 .d_open = bpfopen, 241 .d_read = bpfread, 242 .d_write = bpfwrite, 243 .d_ioctl = bpfioctl, 244 .d_poll = bpfpoll, 245 .d_name = "bpf", 246 .d_kqfilter = bpfkqfilter, 247 }; 248 249 static struct filterops bpfread_filtops = { 250 .f_isfd = 1, 251 .f_detach = filt_bpfdetach, 252 .f_event = filt_bpfread, 253 }; 254 255 /* 256 * LOCKING MODEL USED BY BPF 257 * 258 * Locks: 259 * 1) global lock (BPF_LOCK). Sx, used to protect some global counters, 260 * every bpf_iflist changes, serializes ioctl access to bpf descriptors. 261 * 2) Descriptor lock. Mutex, used to protect BPF buffers and various 262 * structure fields used by bpf_*tap* code. 263 * 264 * Lock order: global lock, then descriptor lock. 265 * 266 * There are several possible consumers: 267 * 268 * 1. The kernel registers interface pointer with bpfattach(). 269 * Each call allocates new bpf_if structure, references ifnet pointer 270 * and links bpf_if into bpf_iflist chain. This is protected with global 271 * lock. 272 * 273 * 2. An userland application uses ioctl() call to bpf_d descriptor. 274 * All such call are serialized with global lock. BPF filters can be 275 * changed, but pointer to old filter will be freed using epoch_call(). 276 * Thus it should be safe for bpf_tap/bpf_mtap* code to do access to 277 * filter pointers, even if change will happen during bpf_tap execution. 278 * Destroying of bpf_d descriptor also is doing using epoch_call(). 279 * 280 * 3. An userland application can write packets into bpf_d descriptor. 281 * There we need to be sure, that ifnet won't disappear during bpfwrite(). 282 * 283 * 4. The kernel invokes bpf_tap/bpf_mtap* functions. The access to 284 * bif_dlist is protected with net_epoch_preempt section. So, it should 285 * be safe to make access to bpf_d descriptor inside the section. 286 * 287 * 5. The kernel invokes bpfdetach() on interface destroying. All lists 288 * are modified with global lock held and actual free() is done using 289 * epoch_call(). 290 */ 291 292 static void 293 bpfif_free(epoch_context_t ctx) 294 { 295 struct bpf_if *bp; 296 297 bp = __containerof(ctx, struct bpf_if, epoch_ctx); 298 if_rele(bp->bif_ifp); 299 free(bp, M_BPF); 300 } 301 302 static void 303 bpfif_ref(struct bpf_if *bp) 304 { 305 306 refcount_acquire(&bp->bif_refcnt); 307 } 308 309 static void 310 bpfif_rele(struct bpf_if *bp) 311 { 312 313 if (!refcount_release(&bp->bif_refcnt)) 314 return; 315 epoch_call(net_epoch_preempt, &bp->epoch_ctx, bpfif_free); 316 } 317 318 static void 319 bpfd_ref(struct bpf_d *d) 320 { 321 322 refcount_acquire(&d->bd_refcnt); 323 } 324 325 static void 326 bpfd_rele(struct bpf_d *d) 327 { 328 329 if (!refcount_release(&d->bd_refcnt)) 330 return; 331 epoch_call(net_epoch_preempt, &d->epoch_ctx, bpfd_free); 332 } 333 334 static struct bpf_program_buffer* 335 bpf_program_buffer_alloc(size_t size, int flags) 336 { 337 338 return (malloc(sizeof(struct bpf_program_buffer) + size, 339 M_BPF, flags)); 340 } 341 342 static void 343 bpf_program_buffer_free(epoch_context_t ctx) 344 { 345 struct bpf_program_buffer *ptr; 346 347 ptr = __containerof(ctx, struct bpf_program_buffer, epoch_ctx); 348 #ifdef BPF_JITTER 349 if (ptr->func != NULL) 350 bpf_destroy_jit_filter(ptr->func); 351 #endif 352 free(ptr, M_BPF); 353 } 354 355 /* 356 * Wrapper functions for various buffering methods. If the set of buffer 357 * modes expands, we will probably want to introduce a switch data structure 358 * similar to protosw, et. 359 */ 360 static void 361 bpf_append_bytes(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 362 u_int len) 363 { 364 365 BPFD_LOCK_ASSERT(d); 366 367 switch (d->bd_bufmode) { 368 case BPF_BUFMODE_BUFFER: 369 return (bpf_buffer_append_bytes(d, buf, offset, src, len)); 370 371 case BPF_BUFMODE_ZBUF: 372 counter_u64_add(d->bd_zcopy, 1); 373 return (bpf_zerocopy_append_bytes(d, buf, offset, src, len)); 374 375 default: 376 panic("bpf_buf_append_bytes"); 377 } 378 } 379 380 static void 381 bpf_append_mbuf(struct bpf_d *d, caddr_t buf, u_int offset, void *src, 382 u_int len) 383 { 384 385 BPFD_LOCK_ASSERT(d); 386 387 switch (d->bd_bufmode) { 388 case BPF_BUFMODE_BUFFER: 389 return (bpf_buffer_append_mbuf(d, buf, offset, src, len)); 390 391 case BPF_BUFMODE_ZBUF: 392 counter_u64_add(d->bd_zcopy, 1); 393 return (bpf_zerocopy_append_mbuf(d, buf, offset, src, len)); 394 395 default: 396 panic("bpf_buf_append_mbuf"); 397 } 398 } 399 400 /* 401 * This function gets called when the free buffer is re-assigned. 402 */ 403 static void 404 bpf_buf_reclaimed(struct bpf_d *d) 405 { 406 407 BPFD_LOCK_ASSERT(d); 408 409 switch (d->bd_bufmode) { 410 case BPF_BUFMODE_BUFFER: 411 return; 412 413 case BPF_BUFMODE_ZBUF: 414 bpf_zerocopy_buf_reclaimed(d); 415 return; 416 417 default: 418 panic("bpf_buf_reclaimed"); 419 } 420 } 421 422 /* 423 * If the buffer mechanism has a way to decide that a held buffer can be made 424 * free, then it is exposed via the bpf_canfreebuf() interface. (1) is 425 * returned if the buffer can be discarded, (0) is returned if it cannot. 426 */ 427 static int 428 bpf_canfreebuf(struct bpf_d *d) 429 { 430 431 BPFD_LOCK_ASSERT(d); 432 433 switch (d->bd_bufmode) { 434 case BPF_BUFMODE_ZBUF: 435 return (bpf_zerocopy_canfreebuf(d)); 436 } 437 return (0); 438 } 439 440 /* 441 * Allow the buffer model to indicate that the current store buffer is 442 * immutable, regardless of the appearance of space. Return (1) if the 443 * buffer is writable, and (0) if not. 444 */ 445 static int 446 bpf_canwritebuf(struct bpf_d *d) 447 { 448 BPFD_LOCK_ASSERT(d); 449 450 switch (d->bd_bufmode) { 451 case BPF_BUFMODE_ZBUF: 452 return (bpf_zerocopy_canwritebuf(d)); 453 } 454 return (1); 455 } 456 457 /* 458 * Notify buffer model that an attempt to write to the store buffer has 459 * resulted in a dropped packet, in which case the buffer may be considered 460 * full. 461 */ 462 static void 463 bpf_buffull(struct bpf_d *d) 464 { 465 466 BPFD_LOCK_ASSERT(d); 467 468 switch (d->bd_bufmode) { 469 case BPF_BUFMODE_ZBUF: 470 bpf_zerocopy_buffull(d); 471 break; 472 } 473 } 474 475 /* 476 * Notify the buffer model that a buffer has moved into the hold position. 477 */ 478 void 479 bpf_bufheld(struct bpf_d *d) 480 { 481 482 BPFD_LOCK_ASSERT(d); 483 484 switch (d->bd_bufmode) { 485 case BPF_BUFMODE_ZBUF: 486 bpf_zerocopy_bufheld(d); 487 break; 488 } 489 } 490 491 static void 492 bpf_free(struct bpf_d *d) 493 { 494 495 switch (d->bd_bufmode) { 496 case BPF_BUFMODE_BUFFER: 497 return (bpf_buffer_free(d)); 498 499 case BPF_BUFMODE_ZBUF: 500 return (bpf_zerocopy_free(d)); 501 502 default: 503 panic("bpf_buf_free"); 504 } 505 } 506 507 static int 508 bpf_uiomove(struct bpf_d *d, caddr_t buf, u_int len, struct uio *uio) 509 { 510 511 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 512 return (EOPNOTSUPP); 513 return (bpf_buffer_uiomove(d, buf, len, uio)); 514 } 515 516 static int 517 bpf_ioctl_sblen(struct bpf_d *d, u_int *i) 518 { 519 520 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) 521 return (EOPNOTSUPP); 522 return (bpf_buffer_ioctl_sblen(d, i)); 523 } 524 525 static int 526 bpf_ioctl_getzmax(struct thread *td, struct bpf_d *d, size_t *i) 527 { 528 529 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 530 return (EOPNOTSUPP); 531 return (bpf_zerocopy_ioctl_getzmax(td, d, i)); 532 } 533 534 static int 535 bpf_ioctl_rotzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 536 { 537 538 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 539 return (EOPNOTSUPP); 540 return (bpf_zerocopy_ioctl_rotzbuf(td, d, bz)); 541 } 542 543 static int 544 bpf_ioctl_setzbuf(struct thread *td, struct bpf_d *d, struct bpf_zbuf *bz) 545 { 546 547 if (d->bd_bufmode != BPF_BUFMODE_ZBUF) 548 return (EOPNOTSUPP); 549 return (bpf_zerocopy_ioctl_setzbuf(td, d, bz)); 550 } 551 552 /* 553 * General BPF functions. 554 */ 555 static int 556 bpf_movein(struct uio *uio, int linktype, struct ifnet *ifp, struct mbuf **mp, 557 struct sockaddr *sockp, int *hdrlen, struct bpf_d *d) 558 { 559 const struct ieee80211_bpf_params *p; 560 struct ether_header *eh; 561 struct mbuf *m; 562 int error; 563 int len; 564 int hlen; 565 int slen; 566 567 /* 568 * Build a sockaddr based on the data link layer type. 569 * We do this at this level because the ethernet header 570 * is copied directly into the data field of the sockaddr. 571 * In the case of SLIP, there is no header and the packet 572 * is forwarded as is. 573 * Also, we are careful to leave room at the front of the mbuf 574 * for the link level header. 575 */ 576 switch (linktype) { 577 578 case DLT_SLIP: 579 sockp->sa_family = AF_INET; 580 hlen = 0; 581 break; 582 583 case DLT_EN10MB: 584 sockp->sa_family = AF_UNSPEC; 585 /* XXX Would MAXLINKHDR be better? */ 586 hlen = ETHER_HDR_LEN; 587 break; 588 589 case DLT_FDDI: 590 sockp->sa_family = AF_IMPLINK; 591 hlen = 0; 592 break; 593 594 case DLT_RAW: 595 sockp->sa_family = AF_UNSPEC; 596 hlen = 0; 597 break; 598 599 case DLT_NULL: 600 /* 601 * null interface types require a 4 byte pseudo header which 602 * corresponds to the address family of the packet. 603 */ 604 sockp->sa_family = AF_UNSPEC; 605 hlen = 4; 606 break; 607 608 case DLT_ATM_RFC1483: 609 /* 610 * en atm driver requires 4-byte atm pseudo header. 611 * though it isn't standard, vpi:vci needs to be 612 * specified anyway. 613 */ 614 sockp->sa_family = AF_UNSPEC; 615 hlen = 12; /* XXX 4(ATM_PH) + 3(LLC) + 5(SNAP) */ 616 break; 617 618 case DLT_PPP: 619 sockp->sa_family = AF_UNSPEC; 620 hlen = 4; /* This should match PPP_HDRLEN */ 621 break; 622 623 case DLT_IEEE802_11: /* IEEE 802.11 wireless */ 624 sockp->sa_family = AF_IEEE80211; 625 hlen = 0; 626 break; 627 628 case DLT_IEEE802_11_RADIO: /* IEEE 802.11 wireless w/ phy params */ 629 sockp->sa_family = AF_IEEE80211; 630 sockp->sa_len = 12; /* XXX != 0 */ 631 hlen = sizeof(struct ieee80211_bpf_params); 632 break; 633 634 default: 635 return (EIO); 636 } 637 638 len = uio->uio_resid; 639 if (len < hlen || len - hlen > ifp->if_mtu) 640 return (EMSGSIZE); 641 642 m = m_get2(len, M_WAITOK, MT_DATA, M_PKTHDR); 643 if (m == NULL) 644 return (EIO); 645 m->m_pkthdr.len = m->m_len = len; 646 *mp = m; 647 648 error = uiomove(mtod(m, u_char *), len, uio); 649 if (error) 650 goto bad; 651 652 slen = bpf_filter(d->bd_wfilter, mtod(m, u_char *), len, len); 653 if (slen == 0) { 654 error = EPERM; 655 goto bad; 656 } 657 658 /* Check for multicast destination */ 659 switch (linktype) { 660 case DLT_EN10MB: 661 eh = mtod(m, struct ether_header *); 662 if (ETHER_IS_MULTICAST(eh->ether_dhost)) { 663 if (bcmp(ifp->if_broadcastaddr, eh->ether_dhost, 664 ETHER_ADDR_LEN) == 0) 665 m->m_flags |= M_BCAST; 666 else 667 m->m_flags |= M_MCAST; 668 } 669 if (d->bd_hdrcmplt == 0) { 670 memcpy(eh->ether_shost, IF_LLADDR(ifp), 671 sizeof(eh->ether_shost)); 672 } 673 break; 674 } 675 676 /* 677 * Make room for link header, and copy it to sockaddr 678 */ 679 if (hlen != 0) { 680 if (sockp->sa_family == AF_IEEE80211) { 681 /* 682 * Collect true length from the parameter header 683 * NB: sockp is known to be zero'd so if we do a 684 * short copy unspecified parameters will be 685 * zero. 686 * NB: packet may not be aligned after stripping 687 * bpf params 688 * XXX check ibp_vers 689 */ 690 p = mtod(m, const struct ieee80211_bpf_params *); 691 hlen = p->ibp_len; 692 if (hlen > sizeof(sockp->sa_data)) { 693 error = EINVAL; 694 goto bad; 695 } 696 } 697 bcopy(mtod(m, const void *), sockp->sa_data, hlen); 698 } 699 *hdrlen = hlen; 700 701 return (0); 702 bad: 703 m_freem(m); 704 return (error); 705 } 706 707 /* 708 * Attach descriptor to the bpf interface, i.e. make d listen on bp, 709 * then reset its buffers and counters with reset_d(). 710 */ 711 static void 712 bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 713 { 714 int op_w; 715 716 BPF_LOCK_ASSERT(); 717 718 /* 719 * Save sysctl value to protect from sysctl change 720 * between reads 721 */ 722 op_w = V_bpf_optimize_writers || d->bd_writer; 723 724 if (d->bd_bif != NULL) 725 bpf_detachd_locked(d, false); 726 /* 727 * Point d at bp, and add d to the interface's list. 728 * Since there are many applications using BPF for 729 * sending raw packets only (dhcpd, cdpd are good examples) 730 * we can delay adding d to the list of active listeners until 731 * some filter is configured. 732 */ 733 734 BPFD_LOCK(d); 735 /* 736 * Hold reference to bpif while descriptor uses this interface. 737 */ 738 bpfif_ref(bp); 739 d->bd_bif = bp; 740 if (op_w != 0) { 741 /* Add to writers-only list */ 742 CK_LIST_INSERT_HEAD(&bp->bif_wlist, d, bd_next); 743 /* 744 * We decrement bd_writer on every filter set operation. 745 * First BIOCSETF is done by pcap_open_live() to set up 746 * snap length. After that appliation usually sets its own 747 * filter. 748 */ 749 d->bd_writer = 2; 750 } else 751 CK_LIST_INSERT_HEAD(&bp->bif_dlist, d, bd_next); 752 753 reset_d(d); 754 BPFD_UNLOCK(d); 755 bpf_bpfd_cnt++; 756 757 CTR3(KTR_NET, "%s: bpf_attach called by pid %d, adding to %s list", 758 __func__, d->bd_pid, d->bd_writer ? "writer" : "active"); 759 760 if (op_w == 0) 761 EVENTHANDLER_INVOKE(bpf_track, bp->bif_ifp, bp->bif_dlt, 1); 762 } 763 764 /* 765 * Check if we need to upgrade our descriptor @d from write-only mode. 766 */ 767 static int 768 bpf_check_upgrade(u_long cmd, struct bpf_d *d, struct bpf_insn *fcode, 769 int flen) 770 { 771 int is_snap, need_upgrade; 772 773 /* 774 * Check if we've already upgraded or new filter is empty. 775 */ 776 if (d->bd_writer == 0 || fcode == NULL) 777 return (0); 778 779 need_upgrade = 0; 780 781 /* 782 * Check if cmd looks like snaplen setting from 783 * pcap_bpf.c:pcap_open_live(). 784 * Note we're not checking .k value here: 785 * while pcap_open_live() definitely sets to non-zero value, 786 * we'd prefer to treat k=0 (deny ALL) case the same way: e.g. 787 * do not consider upgrading immediately 788 */ 789 if (cmd == BIOCSETF && flen == 1 && 790 fcode[0].code == (BPF_RET | BPF_K)) 791 is_snap = 1; 792 else 793 is_snap = 0; 794 795 if (is_snap == 0) { 796 /* 797 * We're setting first filter and it doesn't look like 798 * setting snaplen. We're probably using bpf directly. 799 * Upgrade immediately. 800 */ 801 need_upgrade = 1; 802 } else { 803 /* 804 * Do not require upgrade by first BIOCSETF 805 * (used to set snaplen) by pcap_open_live(). 806 */ 807 808 if (--d->bd_writer == 0) { 809 /* 810 * First snaplen filter has already 811 * been set. This is probably catch-all 812 * filter 813 */ 814 need_upgrade = 1; 815 } 816 } 817 818 CTR5(KTR_NET, 819 "%s: filter function set by pid %d, " 820 "bd_writer counter %d, snap %d upgrade %d", 821 __func__, d->bd_pid, d->bd_writer, 822 is_snap, need_upgrade); 823 824 return (need_upgrade); 825 } 826 827 /* 828 * Detach a file from its interface. 829 */ 830 static void 831 bpf_detachd(struct bpf_d *d) 832 { 833 BPF_LOCK(); 834 bpf_detachd_locked(d, false); 835 BPF_UNLOCK(); 836 } 837 838 static void 839 bpf_detachd_locked(struct bpf_d *d, bool detached_ifp) 840 { 841 struct bpf_if *bp; 842 struct ifnet *ifp; 843 int error; 844 845 BPF_LOCK_ASSERT(); 846 CTR2(KTR_NET, "%s: detach required by pid %d", __func__, d->bd_pid); 847 848 /* Check if descriptor is attached */ 849 if ((bp = d->bd_bif) == NULL) 850 return; 851 852 BPFD_LOCK(d); 853 /* Remove d from the interface's descriptor list. */ 854 CK_LIST_REMOVE(d, bd_next); 855 /* Save bd_writer value */ 856 error = d->bd_writer; 857 ifp = bp->bif_ifp; 858 d->bd_bif = NULL; 859 if (detached_ifp) { 860 /* 861 * Notify descriptor as it's detached, so that any 862 * sleepers wake up and get ENXIO. 863 */ 864 bpf_wakeup(d); 865 } 866 BPFD_UNLOCK(d); 867 bpf_bpfd_cnt--; 868 869 /* Call event handler iff d is attached */ 870 if (error == 0) 871 EVENTHANDLER_INVOKE(bpf_track, ifp, bp->bif_dlt, 0); 872 873 /* 874 * Check if this descriptor had requested promiscuous mode. 875 * If so and ifnet is not detached, turn it off. 876 */ 877 if (d->bd_promisc && !detached_ifp) { 878 d->bd_promisc = 0; 879 CURVNET_SET(ifp->if_vnet); 880 error = ifpromisc(ifp, 0); 881 CURVNET_RESTORE(); 882 if (error != 0 && error != ENXIO) { 883 /* 884 * ENXIO can happen if a pccard is unplugged 885 * Something is really wrong if we were able to put 886 * the driver into promiscuous mode, but can't 887 * take it out. 888 */ 889 if_printf(bp->bif_ifp, 890 "bpf_detach: ifpromisc failed (%d)\n", error); 891 } 892 } 893 bpfif_rele(bp); 894 } 895 896 /* 897 * Close the descriptor by detaching it from its interface, 898 * deallocating its buffers, and marking it free. 899 */ 900 static void 901 bpf_dtor(void *data) 902 { 903 struct bpf_d *d = data; 904 905 BPFD_LOCK(d); 906 if (d->bd_state == BPF_WAITING) 907 callout_stop(&d->bd_callout); 908 d->bd_state = BPF_IDLE; 909 BPFD_UNLOCK(d); 910 funsetown(&d->bd_sigio); 911 bpf_detachd(d); 912 #ifdef MAC 913 mac_bpfdesc_destroy(d); 914 #endif /* MAC */ 915 seldrain(&d->bd_sel); 916 knlist_destroy(&d->bd_sel.si_note); 917 callout_drain(&d->bd_callout); 918 bpfd_rele(d); 919 } 920 921 /* 922 * Open ethernet device. Returns ENXIO for illegal minor device number, 923 * EBUSY if file is open by another process. 924 */ 925 /* ARGSUSED */ 926 static int 927 bpfopen(struct cdev *dev, int flags, int fmt, struct thread *td) 928 { 929 struct bpf_d *d; 930 int error; 931 932 d = malloc(sizeof(*d), M_BPF, M_WAITOK | M_ZERO); 933 error = devfs_set_cdevpriv(d, bpf_dtor); 934 if (error != 0) { 935 free(d, M_BPF); 936 return (error); 937 } 938 939 /* Setup counters */ 940 d->bd_rcount = counter_u64_alloc(M_WAITOK); 941 d->bd_dcount = counter_u64_alloc(M_WAITOK); 942 d->bd_fcount = counter_u64_alloc(M_WAITOK); 943 d->bd_wcount = counter_u64_alloc(M_WAITOK); 944 d->bd_wfcount = counter_u64_alloc(M_WAITOK); 945 d->bd_wdcount = counter_u64_alloc(M_WAITOK); 946 d->bd_zcopy = counter_u64_alloc(M_WAITOK); 947 948 /* 949 * For historical reasons, perform a one-time initialization call to 950 * the buffer routines, even though we're not yet committed to a 951 * particular buffer method. 952 */ 953 bpf_buffer_init(d); 954 if ((flags & FREAD) == 0) 955 d->bd_writer = 2; 956 d->bd_hbuf_in_use = 0; 957 d->bd_bufmode = BPF_BUFMODE_BUFFER; 958 d->bd_sig = SIGIO; 959 d->bd_direction = BPF_D_INOUT; 960 d->bd_refcnt = 1; 961 BPF_PID_REFRESH(d, td); 962 #ifdef MAC 963 mac_bpfdesc_init(d); 964 mac_bpfdesc_create(td->td_ucred, d); 965 #endif 966 mtx_init(&d->bd_lock, devtoname(dev), "bpf cdev lock", MTX_DEF); 967 callout_init_mtx(&d->bd_callout, &d->bd_lock, 0); 968 knlist_init_mtx(&d->bd_sel.si_note, &d->bd_lock); 969 970 return (0); 971 } 972 973 /* 974 * bpfread - read next chunk of packets from buffers 975 */ 976 static int 977 bpfread(struct cdev *dev, struct uio *uio, int ioflag) 978 { 979 struct bpf_d *d; 980 int error; 981 int non_block; 982 int timed_out; 983 984 error = devfs_get_cdevpriv((void **)&d); 985 if (error != 0) 986 return (error); 987 988 /* 989 * Restrict application to use a buffer the same size as 990 * as kernel buffers. 991 */ 992 if (uio->uio_resid != d->bd_bufsize) 993 return (EINVAL); 994 995 non_block = ((ioflag & O_NONBLOCK) != 0); 996 997 BPFD_LOCK(d); 998 BPF_PID_REFRESH_CUR(d); 999 if (d->bd_bufmode != BPF_BUFMODE_BUFFER) { 1000 BPFD_UNLOCK(d); 1001 return (EOPNOTSUPP); 1002 } 1003 if (d->bd_state == BPF_WAITING) 1004 callout_stop(&d->bd_callout); 1005 timed_out = (d->bd_state == BPF_TIMED_OUT); 1006 d->bd_state = BPF_IDLE; 1007 while (d->bd_hbuf_in_use) { 1008 error = mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1009 PRINET|PCATCH, "bd_hbuf", 0); 1010 if (error != 0) { 1011 BPFD_UNLOCK(d); 1012 return (error); 1013 } 1014 } 1015 /* 1016 * If the hold buffer is empty, then do a timed sleep, which 1017 * ends when the timeout expires or when enough packets 1018 * have arrived to fill the store buffer. 1019 */ 1020 while (d->bd_hbuf == NULL) { 1021 if (d->bd_slen != 0) { 1022 /* 1023 * A packet(s) either arrived since the previous 1024 * read or arrived while we were asleep. 1025 */ 1026 if (d->bd_immediate || non_block || timed_out) { 1027 /* 1028 * Rotate the buffers and return what's here 1029 * if we are in immediate mode, non-blocking 1030 * flag is set, or this descriptor timed out. 1031 */ 1032 ROTATE_BUFFERS(d); 1033 break; 1034 } 1035 } 1036 1037 /* 1038 * No data is available, check to see if the bpf device 1039 * is still pointed at a real interface. If not, return 1040 * ENXIO so that the userland process knows to rebind 1041 * it before using it again. 1042 */ 1043 if (d->bd_bif == NULL) { 1044 BPFD_UNLOCK(d); 1045 return (ENXIO); 1046 } 1047 1048 if (non_block) { 1049 BPFD_UNLOCK(d); 1050 return (EWOULDBLOCK); 1051 } 1052 error = msleep(d, &d->bd_lock, PRINET|PCATCH, 1053 "bpf", d->bd_rtout); 1054 if (error == EINTR || error == ERESTART) { 1055 BPFD_UNLOCK(d); 1056 return (error); 1057 } 1058 if (error == EWOULDBLOCK) { 1059 /* 1060 * On a timeout, return what's in the buffer, 1061 * which may be nothing. If there is something 1062 * in the store buffer, we can rotate the buffers. 1063 */ 1064 if (d->bd_hbuf) 1065 /* 1066 * We filled up the buffer in between 1067 * getting the timeout and arriving 1068 * here, so we don't need to rotate. 1069 */ 1070 break; 1071 1072 if (d->bd_slen == 0) { 1073 BPFD_UNLOCK(d); 1074 return (0); 1075 } 1076 ROTATE_BUFFERS(d); 1077 break; 1078 } 1079 } 1080 /* 1081 * At this point, we know we have something in the hold slot. 1082 */ 1083 d->bd_hbuf_in_use = 1; 1084 BPFD_UNLOCK(d); 1085 1086 /* 1087 * Move data from hold buffer into user space. 1088 * We know the entire buffer is transferred since 1089 * we checked above that the read buffer is bpf_bufsize bytes. 1090 * 1091 * We do not have to worry about simultaneous reads because 1092 * we waited for sole access to the hold buffer above. 1093 */ 1094 error = bpf_uiomove(d, d->bd_hbuf, d->bd_hlen, uio); 1095 1096 BPFD_LOCK(d); 1097 KASSERT(d->bd_hbuf != NULL, ("bpfread: lost bd_hbuf")); 1098 d->bd_fbuf = d->bd_hbuf; 1099 d->bd_hbuf = NULL; 1100 d->bd_hlen = 0; 1101 bpf_buf_reclaimed(d); 1102 d->bd_hbuf_in_use = 0; 1103 wakeup(&d->bd_hbuf_in_use); 1104 BPFD_UNLOCK(d); 1105 1106 return (error); 1107 } 1108 1109 /* 1110 * If there are processes sleeping on this descriptor, wake them up. 1111 */ 1112 static __inline void 1113 bpf_wakeup(struct bpf_d *d) 1114 { 1115 1116 BPFD_LOCK_ASSERT(d); 1117 if (d->bd_state == BPF_WAITING) { 1118 callout_stop(&d->bd_callout); 1119 d->bd_state = BPF_IDLE; 1120 } 1121 wakeup(d); 1122 if (d->bd_async && d->bd_sig && d->bd_sigio) 1123 pgsigio(&d->bd_sigio, d->bd_sig, 0); 1124 1125 selwakeuppri(&d->bd_sel, PRINET); 1126 KNOTE_LOCKED(&d->bd_sel.si_note, 0); 1127 } 1128 1129 static void 1130 bpf_timed_out(void *arg) 1131 { 1132 struct bpf_d *d = (struct bpf_d *)arg; 1133 1134 BPFD_LOCK_ASSERT(d); 1135 1136 if (callout_pending(&d->bd_callout) || 1137 !callout_active(&d->bd_callout)) 1138 return; 1139 if (d->bd_state == BPF_WAITING) { 1140 d->bd_state = BPF_TIMED_OUT; 1141 if (d->bd_slen != 0) 1142 bpf_wakeup(d); 1143 } 1144 } 1145 1146 static int 1147 bpf_ready(struct bpf_d *d) 1148 { 1149 1150 BPFD_LOCK_ASSERT(d); 1151 1152 if (!bpf_canfreebuf(d) && d->bd_hlen != 0) 1153 return (1); 1154 if ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1155 d->bd_slen != 0) 1156 return (1); 1157 return (0); 1158 } 1159 1160 static int 1161 bpfwrite(struct cdev *dev, struct uio *uio, int ioflag) 1162 { 1163 struct route ro; 1164 struct sockaddr dst; 1165 struct epoch_tracker et; 1166 struct bpf_if *bp; 1167 struct bpf_d *d; 1168 struct ifnet *ifp; 1169 struct mbuf *m, *mc; 1170 int error, hlen; 1171 1172 error = devfs_get_cdevpriv((void **)&d); 1173 if (error != 0) 1174 return (error); 1175 1176 NET_EPOCH_ENTER(et); 1177 BPFD_LOCK(d); 1178 BPF_PID_REFRESH_CUR(d); 1179 counter_u64_add(d->bd_wcount, 1); 1180 if ((bp = d->bd_bif) == NULL) { 1181 error = ENXIO; 1182 goto out_locked; 1183 } 1184 1185 ifp = bp->bif_ifp; 1186 if ((ifp->if_flags & IFF_UP) == 0) { 1187 error = ENETDOWN; 1188 goto out_locked; 1189 } 1190 1191 if (uio->uio_resid == 0) 1192 goto out_locked; 1193 1194 bzero(&dst, sizeof(dst)); 1195 m = NULL; 1196 hlen = 0; 1197 1198 /* 1199 * Take extra reference, unlock d and exit from epoch section, 1200 * since bpf_movein() can sleep. 1201 */ 1202 bpfd_ref(d); 1203 NET_EPOCH_EXIT(et); 1204 BPFD_UNLOCK(d); 1205 1206 error = bpf_movein(uio, (int)bp->bif_dlt, ifp, 1207 &m, &dst, &hlen, d); 1208 1209 if (error != 0) { 1210 counter_u64_add(d->bd_wdcount, 1); 1211 bpfd_rele(d); 1212 return (error); 1213 } 1214 1215 BPFD_LOCK(d); 1216 /* 1217 * Check that descriptor is still attached to the interface. 1218 * This can happen on bpfdetach(). To avoid access to detached 1219 * ifnet, free mbuf and return ENXIO. 1220 */ 1221 if (d->bd_bif == NULL) { 1222 counter_u64_add(d->bd_wdcount, 1); 1223 BPFD_UNLOCK(d); 1224 bpfd_rele(d); 1225 m_freem(m); 1226 return (ENXIO); 1227 } 1228 counter_u64_add(d->bd_wfcount, 1); 1229 if (d->bd_hdrcmplt) 1230 dst.sa_family = pseudo_AF_HDRCMPLT; 1231 1232 if (d->bd_feedback) { 1233 mc = m_dup(m, M_NOWAIT); 1234 if (mc != NULL) 1235 mc->m_pkthdr.rcvif = ifp; 1236 /* Set M_PROMISC for outgoing packets to be discarded. */ 1237 if (d->bd_direction == BPF_D_INOUT) 1238 m->m_flags |= M_PROMISC; 1239 } else 1240 mc = NULL; 1241 1242 m->m_pkthdr.len -= hlen; 1243 m->m_len -= hlen; 1244 m->m_data += hlen; /* XXX */ 1245 1246 CURVNET_SET(ifp->if_vnet); 1247 #ifdef MAC 1248 mac_bpfdesc_create_mbuf(d, m); 1249 if (mc != NULL) 1250 mac_bpfdesc_create_mbuf(d, mc); 1251 #endif 1252 1253 bzero(&ro, sizeof(ro)); 1254 if (hlen != 0) { 1255 ro.ro_prepend = (u_char *)&dst.sa_data; 1256 ro.ro_plen = hlen; 1257 ro.ro_flags = RT_HAS_HEADER; 1258 } 1259 1260 error = (*ifp->if_output)(ifp, m, &dst, &ro); 1261 if (error) 1262 counter_u64_add(d->bd_wdcount, 1); 1263 1264 if (mc != NULL) { 1265 if (error == 0) 1266 (*ifp->if_input)(ifp, mc); 1267 else 1268 m_freem(mc); 1269 } 1270 CURVNET_RESTORE(); 1271 BPFD_UNLOCK(d); 1272 bpfd_rele(d); 1273 return (error); 1274 1275 out_locked: 1276 counter_u64_add(d->bd_wdcount, 1); 1277 NET_EPOCH_EXIT(et); 1278 BPFD_UNLOCK(d); 1279 return (error); 1280 } 1281 1282 /* 1283 * Reset a descriptor by flushing its packet buffer and clearing the receive 1284 * and drop counts. This is doable for kernel-only buffers, but with 1285 * zero-copy buffers, we can't write to (or rotate) buffers that are 1286 * currently owned by userspace. It would be nice if we could encapsulate 1287 * this logic in the buffer code rather than here. 1288 */ 1289 static void 1290 reset_d(struct bpf_d *d) 1291 { 1292 1293 BPFD_LOCK_ASSERT(d); 1294 1295 while (d->bd_hbuf_in_use) 1296 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, PRINET, 1297 "bd_hbuf", 0); 1298 if ((d->bd_hbuf != NULL) && 1299 (d->bd_bufmode != BPF_BUFMODE_ZBUF || bpf_canfreebuf(d))) { 1300 /* Free the hold buffer. */ 1301 d->bd_fbuf = d->bd_hbuf; 1302 d->bd_hbuf = NULL; 1303 d->bd_hlen = 0; 1304 bpf_buf_reclaimed(d); 1305 } 1306 if (bpf_canwritebuf(d)) 1307 d->bd_slen = 0; 1308 counter_u64_zero(d->bd_rcount); 1309 counter_u64_zero(d->bd_dcount); 1310 counter_u64_zero(d->bd_fcount); 1311 counter_u64_zero(d->bd_wcount); 1312 counter_u64_zero(d->bd_wfcount); 1313 counter_u64_zero(d->bd_wdcount); 1314 counter_u64_zero(d->bd_zcopy); 1315 } 1316 1317 /* 1318 * FIONREAD Check for read packet available. 1319 * BIOCGBLEN Get buffer len [for read()]. 1320 * BIOCSETF Set read filter. 1321 * BIOCSETFNR Set read filter without resetting descriptor. 1322 * BIOCSETWF Set write filter. 1323 * BIOCFLUSH Flush read packet buffer. 1324 * BIOCPROMISC Put interface into promiscuous mode. 1325 * BIOCGDLT Get link layer type. 1326 * BIOCGETIF Get interface name. 1327 * BIOCSETIF Set interface. 1328 * BIOCSRTIMEOUT Set read timeout. 1329 * BIOCGRTIMEOUT Get read timeout. 1330 * BIOCGSTATS Get packet stats. 1331 * BIOCIMMEDIATE Set immediate mode. 1332 * BIOCVERSION Get filter language version. 1333 * BIOCGHDRCMPLT Get "header already complete" flag 1334 * BIOCSHDRCMPLT Set "header already complete" flag 1335 * BIOCGDIRECTION Get packet direction flag 1336 * BIOCSDIRECTION Set packet direction flag 1337 * BIOCGTSTAMP Get time stamp format and resolution. 1338 * BIOCSTSTAMP Set time stamp format and resolution. 1339 * BIOCLOCK Set "locked" flag 1340 * BIOCFEEDBACK Set packet feedback mode. 1341 * BIOCSETZBUF Set current zero-copy buffer locations. 1342 * BIOCGETZMAX Get maximum zero-copy buffer size. 1343 * BIOCROTZBUF Force rotation of zero-copy buffer 1344 * BIOCSETBUFMODE Set buffer mode. 1345 * BIOCGETBUFMODE Get current buffer mode. 1346 */ 1347 /* ARGSUSED */ 1348 static int 1349 bpfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, 1350 struct thread *td) 1351 { 1352 struct bpf_d *d; 1353 int error; 1354 1355 error = devfs_get_cdevpriv((void **)&d); 1356 if (error != 0) 1357 return (error); 1358 1359 /* 1360 * Refresh PID associated with this descriptor. 1361 */ 1362 BPFD_LOCK(d); 1363 BPF_PID_REFRESH(d, td); 1364 if (d->bd_state == BPF_WAITING) 1365 callout_stop(&d->bd_callout); 1366 d->bd_state = BPF_IDLE; 1367 BPFD_UNLOCK(d); 1368 1369 if (d->bd_locked == 1) { 1370 switch (cmd) { 1371 case BIOCGBLEN: 1372 case BIOCFLUSH: 1373 case BIOCGDLT: 1374 case BIOCGDLTLIST: 1375 #ifdef COMPAT_FREEBSD32 1376 case BIOCGDLTLIST32: 1377 #endif 1378 case BIOCGETIF: 1379 case BIOCGRTIMEOUT: 1380 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1381 case BIOCGRTIMEOUT32: 1382 #endif 1383 case BIOCGSTATS: 1384 case BIOCVERSION: 1385 case BIOCGRSIG: 1386 case BIOCGHDRCMPLT: 1387 case BIOCSTSTAMP: 1388 case BIOCFEEDBACK: 1389 case FIONREAD: 1390 case BIOCLOCK: 1391 case BIOCSRTIMEOUT: 1392 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1393 case BIOCSRTIMEOUT32: 1394 #endif 1395 case BIOCIMMEDIATE: 1396 case TIOCGPGRP: 1397 case BIOCROTZBUF: 1398 break; 1399 default: 1400 return (EPERM); 1401 } 1402 } 1403 #ifdef COMPAT_FREEBSD32 1404 /* 1405 * If we see a 32-bit compat ioctl, mark the stream as 32-bit so 1406 * that it will get 32-bit packet headers. 1407 */ 1408 switch (cmd) { 1409 case BIOCSETF32: 1410 case BIOCSETFNR32: 1411 case BIOCSETWF32: 1412 case BIOCGDLTLIST32: 1413 case BIOCGRTIMEOUT32: 1414 case BIOCSRTIMEOUT32: 1415 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) { 1416 BPFD_LOCK(d); 1417 d->bd_compat32 = 1; 1418 BPFD_UNLOCK(d); 1419 } 1420 } 1421 #endif 1422 1423 CURVNET_SET(TD_TO_VNET(td)); 1424 switch (cmd) { 1425 1426 default: 1427 error = EINVAL; 1428 break; 1429 1430 /* 1431 * Check for read packet available. 1432 */ 1433 case FIONREAD: 1434 { 1435 int n; 1436 1437 BPFD_LOCK(d); 1438 n = d->bd_slen; 1439 while (d->bd_hbuf_in_use) 1440 mtx_sleep(&d->bd_hbuf_in_use, &d->bd_lock, 1441 PRINET, "bd_hbuf", 0); 1442 if (d->bd_hbuf) 1443 n += d->bd_hlen; 1444 BPFD_UNLOCK(d); 1445 1446 *(int *)addr = n; 1447 break; 1448 } 1449 1450 /* 1451 * Get buffer len [for read()]. 1452 */ 1453 case BIOCGBLEN: 1454 BPFD_LOCK(d); 1455 *(u_int *)addr = d->bd_bufsize; 1456 BPFD_UNLOCK(d); 1457 break; 1458 1459 /* 1460 * Set buffer length. 1461 */ 1462 case BIOCSBLEN: 1463 error = bpf_ioctl_sblen(d, (u_int *)addr); 1464 break; 1465 1466 /* 1467 * Set link layer read filter. 1468 */ 1469 case BIOCSETF: 1470 case BIOCSETFNR: 1471 case BIOCSETWF: 1472 #ifdef COMPAT_FREEBSD32 1473 case BIOCSETF32: 1474 case BIOCSETFNR32: 1475 case BIOCSETWF32: 1476 #endif 1477 error = bpf_setf(d, (struct bpf_program *)addr, cmd); 1478 break; 1479 1480 /* 1481 * Flush read packet buffer. 1482 */ 1483 case BIOCFLUSH: 1484 BPFD_LOCK(d); 1485 reset_d(d); 1486 BPFD_UNLOCK(d); 1487 break; 1488 1489 /* 1490 * Put interface into promiscuous mode. 1491 */ 1492 case BIOCPROMISC: 1493 if (d->bd_bif == NULL) { 1494 /* 1495 * No interface attached yet. 1496 */ 1497 error = EINVAL; 1498 break; 1499 } 1500 if (d->bd_promisc == 0) { 1501 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1502 if (error == 0) 1503 d->bd_promisc = 1; 1504 } 1505 break; 1506 1507 /* 1508 * Get current data link type. 1509 */ 1510 case BIOCGDLT: 1511 BPF_LOCK(); 1512 if (d->bd_bif == NULL) 1513 error = EINVAL; 1514 else 1515 *(u_int *)addr = d->bd_bif->bif_dlt; 1516 BPF_UNLOCK(); 1517 break; 1518 1519 /* 1520 * Get a list of supported data link types. 1521 */ 1522 #ifdef COMPAT_FREEBSD32 1523 case BIOCGDLTLIST32: 1524 { 1525 struct bpf_dltlist32 *list32; 1526 struct bpf_dltlist dltlist; 1527 1528 list32 = (struct bpf_dltlist32 *)addr; 1529 dltlist.bfl_len = list32->bfl_len; 1530 dltlist.bfl_list = PTRIN(list32->bfl_list); 1531 BPF_LOCK(); 1532 if (d->bd_bif == NULL) 1533 error = EINVAL; 1534 else { 1535 error = bpf_getdltlist(d, &dltlist); 1536 if (error == 0) 1537 list32->bfl_len = dltlist.bfl_len; 1538 } 1539 BPF_UNLOCK(); 1540 break; 1541 } 1542 #endif 1543 1544 case BIOCGDLTLIST: 1545 BPF_LOCK(); 1546 if (d->bd_bif == NULL) 1547 error = EINVAL; 1548 else 1549 error = bpf_getdltlist(d, (struct bpf_dltlist *)addr); 1550 BPF_UNLOCK(); 1551 break; 1552 1553 /* 1554 * Set data link type. 1555 */ 1556 case BIOCSDLT: 1557 BPF_LOCK(); 1558 if (d->bd_bif == NULL) 1559 error = EINVAL; 1560 else 1561 error = bpf_setdlt(d, *(u_int *)addr); 1562 BPF_UNLOCK(); 1563 break; 1564 1565 /* 1566 * Get interface name. 1567 */ 1568 case BIOCGETIF: 1569 BPF_LOCK(); 1570 if (d->bd_bif == NULL) 1571 error = EINVAL; 1572 else { 1573 struct ifnet *const ifp = d->bd_bif->bif_ifp; 1574 struct ifreq *const ifr = (struct ifreq *)addr; 1575 1576 strlcpy(ifr->ifr_name, ifp->if_xname, 1577 sizeof(ifr->ifr_name)); 1578 } 1579 BPF_UNLOCK(); 1580 break; 1581 1582 /* 1583 * Set interface. 1584 */ 1585 case BIOCSETIF: 1586 { 1587 int alloc_buf, size; 1588 1589 /* 1590 * Behavior here depends on the buffering model. If 1591 * we're using kernel memory buffers, then we can 1592 * allocate them here. If we're using zero-copy, 1593 * then the user process must have registered buffers 1594 * by the time we get here. 1595 */ 1596 alloc_buf = 0; 1597 BPFD_LOCK(d); 1598 if (d->bd_bufmode == BPF_BUFMODE_BUFFER && 1599 d->bd_sbuf == NULL) 1600 alloc_buf = 1; 1601 BPFD_UNLOCK(d); 1602 if (alloc_buf) { 1603 size = d->bd_bufsize; 1604 error = bpf_buffer_ioctl_sblen(d, &size); 1605 if (error != 0) 1606 break; 1607 } 1608 BPF_LOCK(); 1609 error = bpf_setif(d, (struct ifreq *)addr); 1610 BPF_UNLOCK(); 1611 break; 1612 } 1613 1614 /* 1615 * Set read timeout. 1616 */ 1617 case BIOCSRTIMEOUT: 1618 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1619 case BIOCSRTIMEOUT32: 1620 #endif 1621 { 1622 struct timeval *tv = (struct timeval *)addr; 1623 #if defined(COMPAT_FREEBSD32) && !defined(__mips__) 1624 struct timeval32 *tv32; 1625 struct timeval tv64; 1626 1627 if (cmd == BIOCSRTIMEOUT32) { 1628 tv32 = (struct timeval32 *)addr; 1629 tv = &tv64; 1630 tv->tv_sec = tv32->tv_sec; 1631 tv->tv_usec = tv32->tv_usec; 1632 } else 1633 #endif 1634 tv = (struct timeval *)addr; 1635 1636 /* 1637 * Subtract 1 tick from tvtohz() since this isn't 1638 * a one-shot timer. 1639 */ 1640 if ((error = itimerfix(tv)) == 0) 1641 d->bd_rtout = tvtohz(tv) - 1; 1642 break; 1643 } 1644 1645 /* 1646 * Get read timeout. 1647 */ 1648 case BIOCGRTIMEOUT: 1649 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1650 case BIOCGRTIMEOUT32: 1651 #endif 1652 { 1653 struct timeval *tv; 1654 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1655 struct timeval32 *tv32; 1656 struct timeval tv64; 1657 1658 if (cmd == BIOCGRTIMEOUT32) 1659 tv = &tv64; 1660 else 1661 #endif 1662 tv = (struct timeval *)addr; 1663 1664 tv->tv_sec = d->bd_rtout / hz; 1665 tv->tv_usec = (d->bd_rtout % hz) * tick; 1666 #if defined(COMPAT_FREEBSD32) && defined(__amd64__) 1667 if (cmd == BIOCGRTIMEOUT32) { 1668 tv32 = (struct timeval32 *)addr; 1669 tv32->tv_sec = tv->tv_sec; 1670 tv32->tv_usec = tv->tv_usec; 1671 } 1672 #endif 1673 1674 break; 1675 } 1676 1677 /* 1678 * Get packet stats. 1679 */ 1680 case BIOCGSTATS: 1681 { 1682 struct bpf_stat *bs = (struct bpf_stat *)addr; 1683 1684 /* XXXCSJP overflow */ 1685 bs->bs_recv = (u_int)counter_u64_fetch(d->bd_rcount); 1686 bs->bs_drop = (u_int)counter_u64_fetch(d->bd_dcount); 1687 break; 1688 } 1689 1690 /* 1691 * Set immediate mode. 1692 */ 1693 case BIOCIMMEDIATE: 1694 BPFD_LOCK(d); 1695 d->bd_immediate = *(u_int *)addr; 1696 BPFD_UNLOCK(d); 1697 break; 1698 1699 case BIOCVERSION: 1700 { 1701 struct bpf_version *bv = (struct bpf_version *)addr; 1702 1703 bv->bv_major = BPF_MAJOR_VERSION; 1704 bv->bv_minor = BPF_MINOR_VERSION; 1705 break; 1706 } 1707 1708 /* 1709 * Get "header already complete" flag 1710 */ 1711 case BIOCGHDRCMPLT: 1712 BPFD_LOCK(d); 1713 *(u_int *)addr = d->bd_hdrcmplt; 1714 BPFD_UNLOCK(d); 1715 break; 1716 1717 /* 1718 * Set "header already complete" flag 1719 */ 1720 case BIOCSHDRCMPLT: 1721 BPFD_LOCK(d); 1722 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1723 BPFD_UNLOCK(d); 1724 break; 1725 1726 /* 1727 * Get packet direction flag 1728 */ 1729 case BIOCGDIRECTION: 1730 BPFD_LOCK(d); 1731 *(u_int *)addr = d->bd_direction; 1732 BPFD_UNLOCK(d); 1733 break; 1734 1735 /* 1736 * Set packet direction flag 1737 */ 1738 case BIOCSDIRECTION: 1739 { 1740 u_int direction; 1741 1742 direction = *(u_int *)addr; 1743 switch (direction) { 1744 case BPF_D_IN: 1745 case BPF_D_INOUT: 1746 case BPF_D_OUT: 1747 BPFD_LOCK(d); 1748 d->bd_direction = direction; 1749 BPFD_UNLOCK(d); 1750 break; 1751 default: 1752 error = EINVAL; 1753 } 1754 } 1755 break; 1756 1757 /* 1758 * Get packet timestamp format and resolution. 1759 */ 1760 case BIOCGTSTAMP: 1761 BPFD_LOCK(d); 1762 *(u_int *)addr = d->bd_tstamp; 1763 BPFD_UNLOCK(d); 1764 break; 1765 1766 /* 1767 * Set packet timestamp format and resolution. 1768 */ 1769 case BIOCSTSTAMP: 1770 { 1771 u_int func; 1772 1773 func = *(u_int *)addr; 1774 if (BPF_T_VALID(func)) 1775 d->bd_tstamp = func; 1776 else 1777 error = EINVAL; 1778 } 1779 break; 1780 1781 case BIOCFEEDBACK: 1782 BPFD_LOCK(d); 1783 d->bd_feedback = *(u_int *)addr; 1784 BPFD_UNLOCK(d); 1785 break; 1786 1787 case BIOCLOCK: 1788 BPFD_LOCK(d); 1789 d->bd_locked = 1; 1790 BPFD_UNLOCK(d); 1791 break; 1792 1793 case FIONBIO: /* Non-blocking I/O */ 1794 break; 1795 1796 case FIOASYNC: /* Send signal on receive packets */ 1797 BPFD_LOCK(d); 1798 d->bd_async = *(int *)addr; 1799 BPFD_UNLOCK(d); 1800 break; 1801 1802 case FIOSETOWN: 1803 /* 1804 * XXX: Add some sort of locking here? 1805 * fsetown() can sleep. 1806 */ 1807 error = fsetown(*(int *)addr, &d->bd_sigio); 1808 break; 1809 1810 case FIOGETOWN: 1811 BPFD_LOCK(d); 1812 *(int *)addr = fgetown(&d->bd_sigio); 1813 BPFD_UNLOCK(d); 1814 break; 1815 1816 /* This is deprecated, FIOSETOWN should be used instead. */ 1817 case TIOCSPGRP: 1818 error = fsetown(-(*(int *)addr), &d->bd_sigio); 1819 break; 1820 1821 /* This is deprecated, FIOGETOWN should be used instead. */ 1822 case TIOCGPGRP: 1823 *(int *)addr = -fgetown(&d->bd_sigio); 1824 break; 1825 1826 case BIOCSRSIG: /* Set receive signal */ 1827 { 1828 u_int sig; 1829 1830 sig = *(u_int *)addr; 1831 1832 if (sig >= NSIG) 1833 error = EINVAL; 1834 else { 1835 BPFD_LOCK(d); 1836 d->bd_sig = sig; 1837 BPFD_UNLOCK(d); 1838 } 1839 break; 1840 } 1841 case BIOCGRSIG: 1842 BPFD_LOCK(d); 1843 *(u_int *)addr = d->bd_sig; 1844 BPFD_UNLOCK(d); 1845 break; 1846 1847 case BIOCGETBUFMODE: 1848 BPFD_LOCK(d); 1849 *(u_int *)addr = d->bd_bufmode; 1850 BPFD_UNLOCK(d); 1851 break; 1852 1853 case BIOCSETBUFMODE: 1854 /* 1855 * Allow the buffering mode to be changed as long as we 1856 * haven't yet committed to a particular mode. Our 1857 * definition of commitment, for now, is whether or not a 1858 * buffer has been allocated or an interface attached, since 1859 * that's the point where things get tricky. 1860 */ 1861 switch (*(u_int *)addr) { 1862 case BPF_BUFMODE_BUFFER: 1863 break; 1864 1865 case BPF_BUFMODE_ZBUF: 1866 if (bpf_zerocopy_enable) 1867 break; 1868 /* FALLSTHROUGH */ 1869 1870 default: 1871 CURVNET_RESTORE(); 1872 return (EINVAL); 1873 } 1874 1875 BPFD_LOCK(d); 1876 if (d->bd_sbuf != NULL || d->bd_hbuf != NULL || 1877 d->bd_fbuf != NULL || d->bd_bif != NULL) { 1878 BPFD_UNLOCK(d); 1879 CURVNET_RESTORE(); 1880 return (EBUSY); 1881 } 1882 d->bd_bufmode = *(u_int *)addr; 1883 BPFD_UNLOCK(d); 1884 break; 1885 1886 case BIOCGETZMAX: 1887 error = bpf_ioctl_getzmax(td, d, (size_t *)addr); 1888 break; 1889 1890 case BIOCSETZBUF: 1891 error = bpf_ioctl_setzbuf(td, d, (struct bpf_zbuf *)addr); 1892 break; 1893 1894 case BIOCROTZBUF: 1895 error = bpf_ioctl_rotzbuf(td, d, (struct bpf_zbuf *)addr); 1896 break; 1897 } 1898 CURVNET_RESTORE(); 1899 return (error); 1900 } 1901 1902 /* 1903 * Set d's packet filter program to fp. If this file already has a filter, 1904 * free it and replace it. Returns EINVAL for bogus requests. 1905 * 1906 * Note we use global lock here to serialize bpf_setf() and bpf_setif() 1907 * calls. 1908 */ 1909 static int 1910 bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1911 { 1912 #ifdef COMPAT_FREEBSD32 1913 struct bpf_program fp_swab; 1914 struct bpf_program32 *fp32; 1915 #endif 1916 struct bpf_program_buffer *fcode; 1917 struct bpf_insn *filter; 1918 #ifdef BPF_JITTER 1919 bpf_jit_filter *jfunc; 1920 #endif 1921 size_t size; 1922 u_int flen; 1923 bool track_event; 1924 1925 #ifdef COMPAT_FREEBSD32 1926 switch (cmd) { 1927 case BIOCSETF32: 1928 case BIOCSETWF32: 1929 case BIOCSETFNR32: 1930 fp32 = (struct bpf_program32 *)fp; 1931 fp_swab.bf_len = fp32->bf_len; 1932 fp_swab.bf_insns = 1933 (struct bpf_insn *)(uintptr_t)fp32->bf_insns; 1934 fp = &fp_swab; 1935 switch (cmd) { 1936 case BIOCSETF32: 1937 cmd = BIOCSETF; 1938 break; 1939 case BIOCSETWF32: 1940 cmd = BIOCSETWF; 1941 break; 1942 } 1943 break; 1944 } 1945 #endif 1946 1947 filter = NULL; 1948 #ifdef BPF_JITTER 1949 jfunc = NULL; 1950 #endif 1951 /* 1952 * Check new filter validness before acquiring any locks. 1953 * Allocate memory for new filter, if needed. 1954 */ 1955 flen = fp->bf_len; 1956 if (flen > bpf_maxinsns || (fp->bf_insns == NULL && flen != 0)) 1957 return (EINVAL); 1958 size = flen * sizeof(*fp->bf_insns); 1959 if (size > 0) { 1960 /* We're setting up new filter. Copy and check actual data. */ 1961 fcode = bpf_program_buffer_alloc(size, M_WAITOK); 1962 filter = (struct bpf_insn *)fcode->buffer; 1963 if (copyin(fp->bf_insns, filter, size) != 0 || 1964 !bpf_validate(filter, flen)) { 1965 free(fcode, M_BPF); 1966 return (EINVAL); 1967 } 1968 #ifdef BPF_JITTER 1969 if (cmd != BIOCSETWF) { 1970 /* 1971 * Filter is copied inside fcode and is 1972 * perfectly valid. 1973 */ 1974 jfunc = bpf_jitter(filter, flen); 1975 } 1976 #endif 1977 } 1978 1979 track_event = false; 1980 fcode = NULL; 1981 1982 BPF_LOCK(); 1983 BPFD_LOCK(d); 1984 /* Set up new filter. */ 1985 if (cmd == BIOCSETWF) { 1986 if (d->bd_wfilter != NULL) { 1987 fcode = __containerof((void *)d->bd_wfilter, 1988 struct bpf_program_buffer, buffer); 1989 #ifdef BPF_JITTER 1990 fcode->func = NULL; 1991 #endif 1992 } 1993 d->bd_wfilter = filter; 1994 } else { 1995 if (d->bd_rfilter != NULL) { 1996 fcode = __containerof((void *)d->bd_rfilter, 1997 struct bpf_program_buffer, buffer); 1998 #ifdef BPF_JITTER 1999 fcode->func = d->bd_bfilter; 2000 #endif 2001 } 2002 d->bd_rfilter = filter; 2003 #ifdef BPF_JITTER 2004 d->bd_bfilter = jfunc; 2005 #endif 2006 if (cmd == BIOCSETF) 2007 reset_d(d); 2008 2009 if (bpf_check_upgrade(cmd, d, filter, flen) != 0) { 2010 /* 2011 * Filter can be set several times without 2012 * specifying interface. In this case just mark d 2013 * as reader. 2014 */ 2015 d->bd_writer = 0; 2016 if (d->bd_bif != NULL) { 2017 /* 2018 * Remove descriptor from writers-only list 2019 * and add it to active readers list. 2020 */ 2021 CK_LIST_REMOVE(d, bd_next); 2022 CK_LIST_INSERT_HEAD(&d->bd_bif->bif_dlist, 2023 d, bd_next); 2024 CTR2(KTR_NET, 2025 "%s: upgrade required by pid %d", 2026 __func__, d->bd_pid); 2027 track_event = true; 2028 } 2029 } 2030 } 2031 BPFD_UNLOCK(d); 2032 2033 if (fcode != NULL) 2034 epoch_call(net_epoch_preempt, &fcode->epoch_ctx, 2035 bpf_program_buffer_free); 2036 2037 if (track_event) 2038 EVENTHANDLER_INVOKE(bpf_track, 2039 d->bd_bif->bif_ifp, d->bd_bif->bif_dlt, 1); 2040 2041 BPF_UNLOCK(); 2042 return (0); 2043 } 2044 2045 /* 2046 * Detach a file from its current interface (if attached at all) and attach 2047 * to the interface indicated by the name stored in ifr. 2048 * Return an errno or 0. 2049 */ 2050 static int 2051 bpf_setif(struct bpf_d *d, struct ifreq *ifr) 2052 { 2053 struct bpf_if *bp; 2054 struct ifnet *theywant; 2055 2056 BPF_LOCK_ASSERT(); 2057 2058 theywant = ifunit(ifr->ifr_name); 2059 if (theywant == NULL || theywant->if_bpf == NULL) 2060 return (ENXIO); 2061 2062 bp = theywant->if_bpf; 2063 /* 2064 * At this point, we expect the buffer is already allocated. If not, 2065 * return an error. 2066 */ 2067 switch (d->bd_bufmode) { 2068 case BPF_BUFMODE_BUFFER: 2069 case BPF_BUFMODE_ZBUF: 2070 if (d->bd_sbuf == NULL) 2071 return (EINVAL); 2072 break; 2073 2074 default: 2075 panic("bpf_setif: bufmode %d", d->bd_bufmode); 2076 } 2077 if (bp != d->bd_bif) 2078 bpf_attachd(d, bp); 2079 else { 2080 BPFD_LOCK(d); 2081 reset_d(d); 2082 BPFD_UNLOCK(d); 2083 } 2084 return (0); 2085 } 2086 2087 /* 2088 * Support for select() and poll() system calls 2089 * 2090 * Return true iff the specific operation will not block indefinitely. 2091 * Otherwise, return false but make a note that a selwakeup() must be done. 2092 */ 2093 static int 2094 bpfpoll(struct cdev *dev, int events, struct thread *td) 2095 { 2096 struct bpf_d *d; 2097 int revents; 2098 2099 if (devfs_get_cdevpriv((void **)&d) != 0 || d->bd_bif == NULL) 2100 return (events & 2101 (POLLHUP|POLLIN|POLLRDNORM|POLLOUT|POLLWRNORM)); 2102 2103 /* 2104 * Refresh PID associated with this descriptor. 2105 */ 2106 revents = events & (POLLOUT | POLLWRNORM); 2107 BPFD_LOCK(d); 2108 BPF_PID_REFRESH(d, td); 2109 if (events & (POLLIN | POLLRDNORM)) { 2110 if (bpf_ready(d)) 2111 revents |= events & (POLLIN | POLLRDNORM); 2112 else { 2113 selrecord(td, &d->bd_sel); 2114 /* Start the read timeout if necessary. */ 2115 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2116 callout_reset(&d->bd_callout, d->bd_rtout, 2117 bpf_timed_out, d); 2118 d->bd_state = BPF_WAITING; 2119 } 2120 } 2121 } 2122 BPFD_UNLOCK(d); 2123 return (revents); 2124 } 2125 2126 /* 2127 * Support for kevent() system call. Register EVFILT_READ filters and 2128 * reject all others. 2129 */ 2130 int 2131 bpfkqfilter(struct cdev *dev, struct knote *kn) 2132 { 2133 struct bpf_d *d; 2134 2135 if (devfs_get_cdevpriv((void **)&d) != 0 || 2136 kn->kn_filter != EVFILT_READ) 2137 return (1); 2138 2139 /* 2140 * Refresh PID associated with this descriptor. 2141 */ 2142 BPFD_LOCK(d); 2143 BPF_PID_REFRESH_CUR(d); 2144 kn->kn_fop = &bpfread_filtops; 2145 kn->kn_hook = d; 2146 knlist_add(&d->bd_sel.si_note, kn, 1); 2147 BPFD_UNLOCK(d); 2148 2149 return (0); 2150 } 2151 2152 static void 2153 filt_bpfdetach(struct knote *kn) 2154 { 2155 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2156 2157 knlist_remove(&d->bd_sel.si_note, kn, 0); 2158 } 2159 2160 static int 2161 filt_bpfread(struct knote *kn, long hint) 2162 { 2163 struct bpf_d *d = (struct bpf_d *)kn->kn_hook; 2164 int ready; 2165 2166 BPFD_LOCK_ASSERT(d); 2167 ready = bpf_ready(d); 2168 if (ready) { 2169 kn->kn_data = d->bd_slen; 2170 /* 2171 * Ignore the hold buffer if it is being copied to user space. 2172 */ 2173 if (!d->bd_hbuf_in_use && d->bd_hbuf) 2174 kn->kn_data += d->bd_hlen; 2175 } else if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 2176 callout_reset(&d->bd_callout, d->bd_rtout, 2177 bpf_timed_out, d); 2178 d->bd_state = BPF_WAITING; 2179 } 2180 2181 return (ready); 2182 } 2183 2184 #define BPF_TSTAMP_NONE 0 2185 #define BPF_TSTAMP_FAST 1 2186 #define BPF_TSTAMP_NORMAL 2 2187 #define BPF_TSTAMP_EXTERN 3 2188 2189 static int 2190 bpf_ts_quality(int tstype) 2191 { 2192 2193 if (tstype == BPF_T_NONE) 2194 return (BPF_TSTAMP_NONE); 2195 if ((tstype & BPF_T_FAST) != 0) 2196 return (BPF_TSTAMP_FAST); 2197 2198 return (BPF_TSTAMP_NORMAL); 2199 } 2200 2201 static int 2202 bpf_gettime(struct bintime *bt, int tstype, struct mbuf *m) 2203 { 2204 struct m_tag *tag; 2205 int quality; 2206 2207 quality = bpf_ts_quality(tstype); 2208 if (quality == BPF_TSTAMP_NONE) 2209 return (quality); 2210 2211 if (m != NULL) { 2212 tag = m_tag_locate(m, MTAG_BPF, MTAG_BPF_TIMESTAMP, NULL); 2213 if (tag != NULL) { 2214 *bt = *(struct bintime *)(tag + 1); 2215 return (BPF_TSTAMP_EXTERN); 2216 } 2217 } 2218 if (quality == BPF_TSTAMP_NORMAL) 2219 binuptime(bt); 2220 else 2221 getbinuptime(bt); 2222 2223 return (quality); 2224 } 2225 2226 /* 2227 * Incoming linkage from device drivers. Process the packet pkt, of length 2228 * pktlen, which is stored in a contiguous buffer. The packet is parsed 2229 * by each process' filter, and if accepted, stashed into the corresponding 2230 * buffer. 2231 */ 2232 void 2233 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2234 { 2235 struct epoch_tracker et; 2236 struct bintime bt; 2237 struct bpf_d *d; 2238 #ifdef BPF_JITTER 2239 bpf_jit_filter *bf; 2240 #endif 2241 u_int slen; 2242 int gottime; 2243 2244 gottime = BPF_TSTAMP_NONE; 2245 NET_EPOCH_ENTER(et); 2246 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2247 counter_u64_add(d->bd_rcount, 1); 2248 /* 2249 * NB: We dont call BPF_CHECK_DIRECTION() here since there 2250 * is no way for the caller to indiciate to us whether this 2251 * packet is inbound or outbound. In the bpf_mtap() routines, 2252 * we use the interface pointers on the mbuf to figure it out. 2253 */ 2254 #ifdef BPF_JITTER 2255 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2256 if (bf != NULL) 2257 slen = (*(bf->func))(pkt, pktlen, pktlen); 2258 else 2259 #endif 2260 slen = bpf_filter(d->bd_rfilter, pkt, pktlen, pktlen); 2261 if (slen != 0) { 2262 /* 2263 * Filter matches. Let's to acquire write lock. 2264 */ 2265 BPFD_LOCK(d); 2266 counter_u64_add(d->bd_fcount, 1); 2267 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2268 gottime = bpf_gettime(&bt, d->bd_tstamp, 2269 NULL); 2270 #ifdef MAC 2271 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2272 #endif 2273 catchpacket(d, pkt, pktlen, slen, 2274 bpf_append_bytes, &bt); 2275 BPFD_UNLOCK(d); 2276 } 2277 } 2278 NET_EPOCH_EXIT(et); 2279 } 2280 2281 #define BPF_CHECK_DIRECTION(d, r, i) \ 2282 (((d)->bd_direction == BPF_D_IN && (r) != (i)) || \ 2283 ((d)->bd_direction == BPF_D_OUT && (r) == (i))) 2284 2285 /* 2286 * Incoming linkage from device drivers, when packet is in an mbuf chain. 2287 * Locking model is explained in bpf_tap(). 2288 */ 2289 void 2290 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 2291 { 2292 struct epoch_tracker et; 2293 struct bintime bt; 2294 struct bpf_d *d; 2295 #ifdef BPF_JITTER 2296 bpf_jit_filter *bf; 2297 #endif 2298 u_int pktlen, slen; 2299 int gottime; 2300 2301 /* Skip outgoing duplicate packets. */ 2302 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2303 m->m_flags &= ~M_PROMISC; 2304 return; 2305 } 2306 2307 pktlen = m_length(m, NULL); 2308 gottime = BPF_TSTAMP_NONE; 2309 2310 NET_EPOCH_ENTER(et); 2311 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2312 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2313 continue; 2314 counter_u64_add(d->bd_rcount, 1); 2315 #ifdef BPF_JITTER 2316 bf = bpf_jitter_enable != 0 ? d->bd_bfilter : NULL; 2317 /* XXX We cannot handle multiple mbufs. */ 2318 if (bf != NULL && m->m_next == NULL) 2319 slen = (*(bf->func))(mtod(m, u_char *), pktlen, 2320 pktlen); 2321 else 2322 #endif 2323 slen = bpf_filter(d->bd_rfilter, (u_char *)m, pktlen, 0); 2324 if (slen != 0) { 2325 BPFD_LOCK(d); 2326 2327 counter_u64_add(d->bd_fcount, 1); 2328 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2329 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2330 #ifdef MAC 2331 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2332 #endif 2333 catchpacket(d, (u_char *)m, pktlen, slen, 2334 bpf_append_mbuf, &bt); 2335 BPFD_UNLOCK(d); 2336 } 2337 } 2338 NET_EPOCH_EXIT(et); 2339 } 2340 2341 /* 2342 * Incoming linkage from device drivers, when packet is in 2343 * an mbuf chain and to be prepended by a contiguous header. 2344 */ 2345 void 2346 bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m) 2347 { 2348 struct epoch_tracker et; 2349 struct bintime bt; 2350 struct mbuf mb; 2351 struct bpf_d *d; 2352 u_int pktlen, slen; 2353 int gottime; 2354 2355 /* Skip outgoing duplicate packets. */ 2356 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif == NULL) { 2357 m->m_flags &= ~M_PROMISC; 2358 return; 2359 } 2360 2361 pktlen = m_length(m, NULL); 2362 /* 2363 * Craft on-stack mbuf suitable for passing to bpf_filter. 2364 * Note that we cut corners here; we only setup what's 2365 * absolutely needed--this mbuf should never go anywhere else. 2366 */ 2367 mb.m_next = m; 2368 mb.m_data = data; 2369 mb.m_len = dlen; 2370 pktlen += dlen; 2371 2372 gottime = BPF_TSTAMP_NONE; 2373 2374 NET_EPOCH_ENTER(et); 2375 CK_LIST_FOREACH(d, &bp->bif_dlist, bd_next) { 2376 if (BPF_CHECK_DIRECTION(d, m->m_pkthdr.rcvif, bp->bif_ifp)) 2377 continue; 2378 counter_u64_add(d->bd_rcount, 1); 2379 slen = bpf_filter(d->bd_rfilter, (u_char *)&mb, pktlen, 0); 2380 if (slen != 0) { 2381 BPFD_LOCK(d); 2382 2383 counter_u64_add(d->bd_fcount, 1); 2384 if (gottime < bpf_ts_quality(d->bd_tstamp)) 2385 gottime = bpf_gettime(&bt, d->bd_tstamp, m); 2386 #ifdef MAC 2387 if (mac_bpfdesc_check_receive(d, bp->bif_ifp) == 0) 2388 #endif 2389 catchpacket(d, (u_char *)&mb, pktlen, slen, 2390 bpf_append_mbuf, &bt); 2391 BPFD_UNLOCK(d); 2392 } 2393 } 2394 NET_EPOCH_EXIT(et); 2395 } 2396 2397 #undef BPF_CHECK_DIRECTION 2398 #undef BPF_TSTAMP_NONE 2399 #undef BPF_TSTAMP_FAST 2400 #undef BPF_TSTAMP_NORMAL 2401 #undef BPF_TSTAMP_EXTERN 2402 2403 static int 2404 bpf_hdrlen(struct bpf_d *d) 2405 { 2406 int hdrlen; 2407 2408 hdrlen = d->bd_bif->bif_hdrlen; 2409 #ifndef BURN_BRIDGES 2410 if (d->bd_tstamp == BPF_T_NONE || 2411 BPF_T_FORMAT(d->bd_tstamp) == BPF_T_MICROTIME) 2412 #ifdef COMPAT_FREEBSD32 2413 if (d->bd_compat32) 2414 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr32); 2415 else 2416 #endif 2417 hdrlen += SIZEOF_BPF_HDR(struct bpf_hdr); 2418 else 2419 #endif 2420 hdrlen += SIZEOF_BPF_HDR(struct bpf_xhdr); 2421 #ifdef COMPAT_FREEBSD32 2422 if (d->bd_compat32) 2423 hdrlen = BPF_WORDALIGN32(hdrlen); 2424 else 2425 #endif 2426 hdrlen = BPF_WORDALIGN(hdrlen); 2427 2428 return (hdrlen - d->bd_bif->bif_hdrlen); 2429 } 2430 2431 static void 2432 bpf_bintime2ts(struct bintime *bt, struct bpf_ts *ts, int tstype) 2433 { 2434 struct bintime bt2, boottimebin; 2435 struct timeval tsm; 2436 struct timespec tsn; 2437 2438 if ((tstype & BPF_T_MONOTONIC) == 0) { 2439 bt2 = *bt; 2440 getboottimebin(&boottimebin); 2441 bintime_add(&bt2, &boottimebin); 2442 bt = &bt2; 2443 } 2444 switch (BPF_T_FORMAT(tstype)) { 2445 case BPF_T_MICROTIME: 2446 bintime2timeval(bt, &tsm); 2447 ts->bt_sec = tsm.tv_sec; 2448 ts->bt_frac = tsm.tv_usec; 2449 break; 2450 case BPF_T_NANOTIME: 2451 bintime2timespec(bt, &tsn); 2452 ts->bt_sec = tsn.tv_sec; 2453 ts->bt_frac = tsn.tv_nsec; 2454 break; 2455 case BPF_T_BINTIME: 2456 ts->bt_sec = bt->sec; 2457 ts->bt_frac = bt->frac; 2458 break; 2459 } 2460 } 2461 2462 /* 2463 * Move the packet data from interface memory (pkt) into the 2464 * store buffer. "cpfn" is the routine called to do the actual data 2465 * transfer. bcopy is passed in to copy contiguous chunks, while 2466 * bpf_append_mbuf is passed in to copy mbuf chains. In the latter case, 2467 * pkt is really an mbuf. 2468 */ 2469 static void 2470 catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2471 void (*cpfn)(struct bpf_d *, caddr_t, u_int, void *, u_int), 2472 struct bintime *bt) 2473 { 2474 struct bpf_xhdr hdr; 2475 #ifndef BURN_BRIDGES 2476 struct bpf_hdr hdr_old; 2477 #ifdef COMPAT_FREEBSD32 2478 struct bpf_hdr32 hdr32_old; 2479 #endif 2480 #endif 2481 int caplen, curlen, hdrlen, totlen; 2482 int do_wakeup = 0; 2483 int do_timestamp; 2484 int tstype; 2485 2486 BPFD_LOCK_ASSERT(d); 2487 2488 /* 2489 * Detect whether user space has released a buffer back to us, and if 2490 * so, move it from being a hold buffer to a free buffer. This may 2491 * not be the best place to do it (for example, we might only want to 2492 * run this check if we need the space), but for now it's a reliable 2493 * spot to do it. 2494 */ 2495 if (d->bd_fbuf == NULL && bpf_canfreebuf(d)) { 2496 d->bd_fbuf = d->bd_hbuf; 2497 d->bd_hbuf = NULL; 2498 d->bd_hlen = 0; 2499 bpf_buf_reclaimed(d); 2500 } 2501 2502 /* 2503 * Figure out how many bytes to move. If the packet is 2504 * greater or equal to the snapshot length, transfer that 2505 * much. Otherwise, transfer the whole packet (unless 2506 * we hit the buffer size limit). 2507 */ 2508 hdrlen = bpf_hdrlen(d); 2509 totlen = hdrlen + min(snaplen, pktlen); 2510 if (totlen > d->bd_bufsize) 2511 totlen = d->bd_bufsize; 2512 2513 /* 2514 * Round up the end of the previous packet to the next longword. 2515 * 2516 * Drop the packet if there's no room and no hope of room 2517 * If the packet would overflow the storage buffer or the storage 2518 * buffer is considered immutable by the buffer model, try to rotate 2519 * the buffer and wakeup pending processes. 2520 */ 2521 #ifdef COMPAT_FREEBSD32 2522 if (d->bd_compat32) 2523 curlen = BPF_WORDALIGN32(d->bd_slen); 2524 else 2525 #endif 2526 curlen = BPF_WORDALIGN(d->bd_slen); 2527 if (curlen + totlen > d->bd_bufsize || !bpf_canwritebuf(d)) { 2528 if (d->bd_fbuf == NULL) { 2529 /* 2530 * There's no room in the store buffer, and no 2531 * prospect of room, so drop the packet. Notify the 2532 * buffer model. 2533 */ 2534 bpf_buffull(d); 2535 counter_u64_add(d->bd_dcount, 1); 2536 return; 2537 } 2538 KASSERT(!d->bd_hbuf_in_use, ("hold buffer is in use")); 2539 ROTATE_BUFFERS(d); 2540 do_wakeup = 1; 2541 curlen = 0; 2542 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) 2543 /* 2544 * Immediate mode is set, or the read timeout has already 2545 * expired during a select call. A packet arrived, so the 2546 * reader should be woken up. 2547 */ 2548 do_wakeup = 1; 2549 caplen = totlen - hdrlen; 2550 tstype = d->bd_tstamp; 2551 do_timestamp = tstype != BPF_T_NONE; 2552 #ifndef BURN_BRIDGES 2553 if (tstype == BPF_T_NONE || BPF_T_FORMAT(tstype) == BPF_T_MICROTIME) { 2554 struct bpf_ts ts; 2555 if (do_timestamp) 2556 bpf_bintime2ts(bt, &ts, tstype); 2557 #ifdef COMPAT_FREEBSD32 2558 if (d->bd_compat32) { 2559 bzero(&hdr32_old, sizeof(hdr32_old)); 2560 if (do_timestamp) { 2561 hdr32_old.bh_tstamp.tv_sec = ts.bt_sec; 2562 hdr32_old.bh_tstamp.tv_usec = ts.bt_frac; 2563 } 2564 hdr32_old.bh_datalen = pktlen; 2565 hdr32_old.bh_hdrlen = hdrlen; 2566 hdr32_old.bh_caplen = caplen; 2567 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr32_old, 2568 sizeof(hdr32_old)); 2569 goto copy; 2570 } 2571 #endif 2572 bzero(&hdr_old, sizeof(hdr_old)); 2573 if (do_timestamp) { 2574 hdr_old.bh_tstamp.tv_sec = ts.bt_sec; 2575 hdr_old.bh_tstamp.tv_usec = ts.bt_frac; 2576 } 2577 hdr_old.bh_datalen = pktlen; 2578 hdr_old.bh_hdrlen = hdrlen; 2579 hdr_old.bh_caplen = caplen; 2580 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr_old, 2581 sizeof(hdr_old)); 2582 goto copy; 2583 } 2584 #endif 2585 2586 /* 2587 * Append the bpf header. Note we append the actual header size, but 2588 * move forward the length of the header plus padding. 2589 */ 2590 bzero(&hdr, sizeof(hdr)); 2591 if (do_timestamp) 2592 bpf_bintime2ts(bt, &hdr.bh_tstamp, tstype); 2593 hdr.bh_datalen = pktlen; 2594 hdr.bh_hdrlen = hdrlen; 2595 hdr.bh_caplen = caplen; 2596 bpf_append_bytes(d, d->bd_sbuf, curlen, &hdr, sizeof(hdr)); 2597 2598 /* 2599 * Copy the packet data into the store buffer and update its length. 2600 */ 2601 #ifndef BURN_BRIDGES 2602 copy: 2603 #endif 2604 (*cpfn)(d, d->bd_sbuf, curlen + hdrlen, pkt, caplen); 2605 d->bd_slen = curlen + totlen; 2606 2607 if (do_wakeup) 2608 bpf_wakeup(d); 2609 } 2610 2611 /* 2612 * Free buffers currently in use by a descriptor. 2613 * Called on close. 2614 */ 2615 static void 2616 bpfd_free(epoch_context_t ctx) 2617 { 2618 struct bpf_d *d; 2619 struct bpf_program_buffer *p; 2620 2621 /* 2622 * We don't need to lock out interrupts since this descriptor has 2623 * been detached from its interface and it yet hasn't been marked 2624 * free. 2625 */ 2626 d = __containerof(ctx, struct bpf_d, epoch_ctx); 2627 bpf_free(d); 2628 if (d->bd_rfilter != NULL) { 2629 p = __containerof((void *)d->bd_rfilter, 2630 struct bpf_program_buffer, buffer); 2631 bpf_program_buffer_free(&p->epoch_ctx); 2632 } 2633 if (d->bd_wfilter != NULL) { 2634 p = __containerof((void *)d->bd_wfilter, 2635 struct bpf_program_buffer, buffer); 2636 bpf_program_buffer_free(&p->epoch_ctx); 2637 } 2638 2639 mtx_destroy(&d->bd_lock); 2640 counter_u64_free(d->bd_rcount); 2641 counter_u64_free(d->bd_dcount); 2642 counter_u64_free(d->bd_fcount); 2643 counter_u64_free(d->bd_wcount); 2644 counter_u64_free(d->bd_wfcount); 2645 counter_u64_free(d->bd_wdcount); 2646 counter_u64_free(d->bd_zcopy); 2647 free(d, M_BPF); 2648 } 2649 2650 /* 2651 * Attach an interface to bpf. dlt is the link layer type; hdrlen is the 2652 * fixed size of the link header (variable length headers not yet supported). 2653 */ 2654 void 2655 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2656 { 2657 2658 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 2659 } 2660 2661 /* 2662 * Attach an interface to bpf. ifp is a pointer to the structure 2663 * defining the interface to be attached, dlt is the link layer type, 2664 * and hdrlen is the fixed size of the link header (variable length 2665 * headers are not yet supporrted). 2666 */ 2667 void 2668 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, 2669 struct bpf_if **driverp) 2670 { 2671 struct bpf_if *bp; 2672 2673 KASSERT(*driverp == NULL, 2674 ("bpfattach2: driverp already initialized")); 2675 2676 bp = malloc(sizeof(*bp), M_BPF, M_WAITOK | M_ZERO); 2677 2678 CK_LIST_INIT(&bp->bif_dlist); 2679 CK_LIST_INIT(&bp->bif_wlist); 2680 bp->bif_ifp = ifp; 2681 bp->bif_dlt = dlt; 2682 bp->bif_hdrlen = hdrlen; 2683 bp->bif_bpf = driverp; 2684 bp->bif_refcnt = 1; 2685 *driverp = bp; 2686 /* 2687 * Reference ifnet pointer, so it won't freed until 2688 * we release it. 2689 */ 2690 if_ref(ifp); 2691 BPF_LOCK(); 2692 CK_LIST_INSERT_HEAD(&bpf_iflist, bp, bif_next); 2693 BPF_UNLOCK(); 2694 2695 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 2696 if_printf(ifp, "bpf attached\n"); 2697 } 2698 2699 #ifdef VIMAGE 2700 /* 2701 * When moving interfaces between vnet instances we need a way to 2702 * query the dlt and hdrlen before detach so we can re-attch the if_bpf 2703 * after the vmove. We unfortunately have no device driver infrastructure 2704 * to query the interface for these values after creation/attach, thus 2705 * add this as a workaround. 2706 */ 2707 int 2708 bpf_get_bp_params(struct bpf_if *bp, u_int *bif_dlt, u_int *bif_hdrlen) 2709 { 2710 2711 if (bp == NULL) 2712 return (ENXIO); 2713 if (bif_dlt == NULL && bif_hdrlen == NULL) 2714 return (0); 2715 2716 if (bif_dlt != NULL) 2717 *bif_dlt = bp->bif_dlt; 2718 if (bif_hdrlen != NULL) 2719 *bif_hdrlen = bp->bif_hdrlen; 2720 2721 return (0); 2722 } 2723 #endif 2724 2725 /* 2726 * Detach bpf from an interface. This involves detaching each descriptor 2727 * associated with the interface. Notify each descriptor as it's detached 2728 * so that any sleepers wake up and get ENXIO. 2729 */ 2730 void 2731 bpfdetach(struct ifnet *ifp) 2732 { 2733 struct bpf_if *bp, *bp_temp; 2734 struct bpf_d *d; 2735 2736 BPF_LOCK(); 2737 /* Find all bpf_if struct's which reference ifp and detach them. */ 2738 CK_LIST_FOREACH_SAFE(bp, &bpf_iflist, bif_next, bp_temp) { 2739 if (ifp != bp->bif_ifp) 2740 continue; 2741 2742 CK_LIST_REMOVE(bp, bif_next); 2743 *bp->bif_bpf = (struct bpf_if *)&dead_bpf_if; 2744 2745 CTR4(KTR_NET, 2746 "%s: sheduling free for encap %d (%p) for if %p", 2747 __func__, bp->bif_dlt, bp, ifp); 2748 2749 /* Detach common descriptors */ 2750 while ((d = CK_LIST_FIRST(&bp->bif_dlist)) != NULL) { 2751 bpf_detachd_locked(d, true); 2752 } 2753 2754 /* Detach writer-only descriptors */ 2755 while ((d = CK_LIST_FIRST(&bp->bif_wlist)) != NULL) { 2756 bpf_detachd_locked(d, true); 2757 } 2758 bpfif_rele(bp); 2759 } 2760 BPF_UNLOCK(); 2761 } 2762 2763 /* 2764 * Get a list of available data link type of the interface. 2765 */ 2766 static int 2767 bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2768 { 2769 struct ifnet *ifp; 2770 struct bpf_if *bp; 2771 u_int *lst; 2772 int error, n, n1; 2773 2774 BPF_LOCK_ASSERT(); 2775 2776 ifp = d->bd_bif->bif_ifp; 2777 n1 = 0; 2778 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2779 if (bp->bif_ifp == ifp) 2780 n1++; 2781 } 2782 if (bfl->bfl_list == NULL) { 2783 bfl->bfl_len = n1; 2784 return (0); 2785 } 2786 if (n1 > bfl->bfl_len) 2787 return (ENOMEM); 2788 2789 lst = malloc(n1 * sizeof(u_int), M_TEMP, M_WAITOK); 2790 n = 0; 2791 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2792 if (bp->bif_ifp != ifp) 2793 continue; 2794 lst[n++] = bp->bif_dlt; 2795 } 2796 error = copyout(lst, bfl->bfl_list, sizeof(u_int) * n); 2797 free(lst, M_TEMP); 2798 bfl->bfl_len = n; 2799 return (error); 2800 } 2801 2802 /* 2803 * Set the data link type of a BPF instance. 2804 */ 2805 static int 2806 bpf_setdlt(struct bpf_d *d, u_int dlt) 2807 { 2808 int error, opromisc; 2809 struct ifnet *ifp; 2810 struct bpf_if *bp; 2811 2812 BPF_LOCK_ASSERT(); 2813 MPASS(d->bd_bif != NULL); 2814 2815 /* 2816 * It is safe to check bd_bif without BPFD_LOCK, it can not be 2817 * changed while we hold global lock. 2818 */ 2819 if (d->bd_bif->bif_dlt == dlt) 2820 return (0); 2821 2822 ifp = d->bd_bif->bif_ifp; 2823 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2824 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2825 break; 2826 } 2827 if (bp == NULL) 2828 return (EINVAL); 2829 2830 opromisc = d->bd_promisc; 2831 bpf_attachd(d, bp); 2832 if (opromisc) { 2833 error = ifpromisc(bp->bif_ifp, 1); 2834 if (error) 2835 if_printf(bp->bif_ifp, "%s: ifpromisc failed (%d)\n", 2836 __func__, error); 2837 else 2838 d->bd_promisc = 1; 2839 } 2840 return (0); 2841 } 2842 2843 static void 2844 bpf_drvinit(void *unused) 2845 { 2846 struct cdev *dev; 2847 2848 sx_init(&bpf_sx, "bpf global lock"); 2849 CK_LIST_INIT(&bpf_iflist); 2850 2851 dev = make_dev(&bpf_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "bpf"); 2852 /* For compatibility */ 2853 make_dev_alias(dev, "bpf0"); 2854 } 2855 2856 /* 2857 * Zero out the various packet counters associated with all of the bpf 2858 * descriptors. At some point, we will probably want to get a bit more 2859 * granular and allow the user to specify descriptors to be zeroed. 2860 */ 2861 static void 2862 bpf_zero_counters(void) 2863 { 2864 struct bpf_if *bp; 2865 struct bpf_d *bd; 2866 2867 BPF_LOCK(); 2868 /* 2869 * We are protected by global lock here, interfaces and 2870 * descriptors can not be deleted while we hold it. 2871 */ 2872 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2873 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2874 counter_u64_zero(bd->bd_rcount); 2875 counter_u64_zero(bd->bd_dcount); 2876 counter_u64_zero(bd->bd_fcount); 2877 counter_u64_zero(bd->bd_wcount); 2878 counter_u64_zero(bd->bd_wfcount); 2879 counter_u64_zero(bd->bd_zcopy); 2880 } 2881 } 2882 BPF_UNLOCK(); 2883 } 2884 2885 /* 2886 * Fill filter statistics 2887 */ 2888 static void 2889 bpfstats_fill_xbpf(struct xbpf_d *d, struct bpf_d *bd) 2890 { 2891 2892 BPF_LOCK_ASSERT(); 2893 bzero(d, sizeof(*d)); 2894 d->bd_structsize = sizeof(*d); 2895 d->bd_immediate = bd->bd_immediate; 2896 d->bd_promisc = bd->bd_promisc; 2897 d->bd_hdrcmplt = bd->bd_hdrcmplt; 2898 d->bd_direction = bd->bd_direction; 2899 d->bd_feedback = bd->bd_feedback; 2900 d->bd_async = bd->bd_async; 2901 d->bd_rcount = counter_u64_fetch(bd->bd_rcount); 2902 d->bd_dcount = counter_u64_fetch(bd->bd_dcount); 2903 d->bd_fcount = counter_u64_fetch(bd->bd_fcount); 2904 d->bd_sig = bd->bd_sig; 2905 d->bd_slen = bd->bd_slen; 2906 d->bd_hlen = bd->bd_hlen; 2907 d->bd_bufsize = bd->bd_bufsize; 2908 d->bd_pid = bd->bd_pid; 2909 strlcpy(d->bd_ifname, 2910 bd->bd_bif->bif_ifp->if_xname, IFNAMSIZ); 2911 d->bd_locked = bd->bd_locked; 2912 d->bd_wcount = counter_u64_fetch(bd->bd_wcount); 2913 d->bd_wdcount = counter_u64_fetch(bd->bd_wdcount); 2914 d->bd_wfcount = counter_u64_fetch(bd->bd_wfcount); 2915 d->bd_zcopy = counter_u64_fetch(bd->bd_zcopy); 2916 d->bd_bufmode = bd->bd_bufmode; 2917 } 2918 2919 /* 2920 * Handle `netstat -B' stats request 2921 */ 2922 static int 2923 bpf_stats_sysctl(SYSCTL_HANDLER_ARGS) 2924 { 2925 static const struct xbpf_d zerostats; 2926 struct xbpf_d *xbdbuf, *xbd, tempstats; 2927 int index, error; 2928 struct bpf_if *bp; 2929 struct bpf_d *bd; 2930 2931 /* 2932 * XXX This is not technically correct. It is possible for non 2933 * privileged users to open bpf devices. It would make sense 2934 * if the users who opened the devices were able to retrieve 2935 * the statistics for them, too. 2936 */ 2937 error = priv_check(req->td, PRIV_NET_BPF); 2938 if (error) 2939 return (error); 2940 /* 2941 * Check to see if the user is requesting that the counters be 2942 * zeroed out. Explicitly check that the supplied data is zeroed, 2943 * as we aren't allowing the user to set the counters currently. 2944 */ 2945 if (req->newptr != NULL) { 2946 if (req->newlen != sizeof(tempstats)) 2947 return (EINVAL); 2948 memset(&tempstats, 0, sizeof(tempstats)); 2949 error = SYSCTL_IN(req, &tempstats, sizeof(tempstats)); 2950 if (error) 2951 return (error); 2952 if (bcmp(&tempstats, &zerostats, sizeof(tempstats)) != 0) 2953 return (EINVAL); 2954 bpf_zero_counters(); 2955 return (0); 2956 } 2957 if (req->oldptr == NULL) 2958 return (SYSCTL_OUT(req, 0, bpf_bpfd_cnt * sizeof(*xbd))); 2959 if (bpf_bpfd_cnt == 0) 2960 return (SYSCTL_OUT(req, 0, 0)); 2961 xbdbuf = malloc(req->oldlen, M_BPF, M_WAITOK); 2962 BPF_LOCK(); 2963 if (req->oldlen < (bpf_bpfd_cnt * sizeof(*xbd))) { 2964 BPF_UNLOCK(); 2965 free(xbdbuf, M_BPF); 2966 return (ENOMEM); 2967 } 2968 index = 0; 2969 CK_LIST_FOREACH(bp, &bpf_iflist, bif_next) { 2970 /* Send writers-only first */ 2971 CK_LIST_FOREACH(bd, &bp->bif_wlist, bd_next) { 2972 xbd = &xbdbuf[index++]; 2973 bpfstats_fill_xbpf(xbd, bd); 2974 } 2975 CK_LIST_FOREACH(bd, &bp->bif_dlist, bd_next) { 2976 xbd = &xbdbuf[index++]; 2977 bpfstats_fill_xbpf(xbd, bd); 2978 } 2979 } 2980 BPF_UNLOCK(); 2981 error = SYSCTL_OUT(req, xbdbuf, index * sizeof(*xbd)); 2982 free(xbdbuf, M_BPF); 2983 return (error); 2984 } 2985 2986 SYSINIT(bpfdev,SI_SUB_DRIVERS,SI_ORDER_MIDDLE,bpf_drvinit,NULL); 2987 2988 #else /* !DEV_BPF && !NETGRAPH_BPF */ 2989 2990 /* 2991 * NOP stubs to allow bpf-using drivers to load and function. 2992 * 2993 * A 'better' implementation would allow the core bpf functionality 2994 * to be loaded at runtime. 2995 */ 2996 2997 void 2998 bpf_tap(struct bpf_if *bp, u_char *pkt, u_int pktlen) 2999 { 3000 } 3001 3002 void 3003 bpf_mtap(struct bpf_if *bp, struct mbuf *m) 3004 { 3005 } 3006 3007 void 3008 bpf_mtap2(struct bpf_if *bp, void *d, u_int l, struct mbuf *m) 3009 { 3010 } 3011 3012 void 3013 bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen) 3014 { 3015 3016 bpfattach2(ifp, dlt, hdrlen, &ifp->if_bpf); 3017 } 3018 3019 void 3020 bpfattach2(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 3021 { 3022 3023 *driverp = (struct bpf_if *)&dead_bpf_if; 3024 } 3025 3026 void 3027 bpfdetach(struct ifnet *ifp) 3028 { 3029 } 3030 3031 u_int 3032 bpf_filter(const struct bpf_insn *pc, u_char *p, u_int wirelen, u_int buflen) 3033 { 3034 return -1; /* "no filter" behaviour */ 3035 } 3036 3037 int 3038 bpf_validate(const struct bpf_insn *f, int len) 3039 { 3040 return 0; /* false */ 3041 } 3042 3043 #endif /* !DEV_BPF && !NETGRAPH_BPF */ 3044 3045 #ifdef DDB 3046 static void 3047 bpf_show_bpf_if(struct bpf_if *bpf_if) 3048 { 3049 3050 if (bpf_if == NULL) 3051 return; 3052 db_printf("%p:\n", bpf_if); 3053 #define BPF_DB_PRINTF(f, e) db_printf(" %s = " f "\n", #e, bpf_if->e); 3054 /* bif_ext.bif_next */ 3055 /* bif_ext.bif_dlist */ 3056 BPF_DB_PRINTF("%#x", bif_dlt); 3057 BPF_DB_PRINTF("%u", bif_hdrlen); 3058 /* bif_wlist */ 3059 BPF_DB_PRINTF("%p", bif_ifp); 3060 BPF_DB_PRINTF("%p", bif_bpf); 3061 BPF_DB_PRINTF("%u", bif_refcnt); 3062 } 3063 3064 DB_SHOW_COMMAND(bpf_if, db_show_bpf_if) 3065 { 3066 3067 if (!have_addr) { 3068 db_printf("usage: show bpf_if <struct bpf_if *>\n"); 3069 return; 3070 } 3071 3072 bpf_show_bpf_if((struct bpf_if *)addr); 3073 } 3074 #endif 3075