1/* $NetBSD: bpf.c,v 1.252 2023/07/31 17:41:18 christos Exp $ */ 2 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 * static char rcsid[] = 38 * "Header: bpf.c,v 1.67 96/09/26 22:00:52 leres Exp "; 39 */ 40 41#include <sys/cdefs.h> 42__KERNEL_RCSID(0, "$NetBSD: bpf.c,v 1.252 2023/07/31 17:41:18 christos Exp $"); 43 44#if defined(_KERNEL_OPT) 45#include "opt_bpf.h" 46#include "sl.h" 47#include "opt_net_mpsafe.h" 48#endif 49 50#include <sys/param.h> 51#include <sys/systm.h> 52#include <sys/mbuf.h> 53#include <sys/buf.h> 54#include <sys/time.h> 55#include <sys/proc.h> 56#include <sys/ioctl.h> 57#include <sys/conf.h> 58#include <sys/vnode.h> 59#include <sys/queue.h> 60#include <sys/stat.h> 61#include <sys/module.h> 62#include <sys/atomic.h> 63#include <sys/cpu.h> 64 65#include <sys/file.h> 66#include <sys/filedesc.h> 67#include <sys/tty.h> 68#include <sys/uio.h> 69 70#include <sys/protosw.h> 71#include <sys/socket.h> 72#include <sys/errno.h> 73#include <sys/kernel.h> 74#include <sys/poll.h> 75#include <sys/sysctl.h> 76#include <sys/kauth.h> 77#include <sys/syslog.h> 78#include <sys/percpu.h> 79#include <sys/pserialize.h> 80#include <sys/lwp.h> 81#include <sys/xcall.h> 82 83#include <net/if.h> 84#include <net/slip.h> 85 86#include <net/bpf.h> 87#include <net/bpfdesc.h> 88#include <net/bpfjit.h> 89 90#include <net/if_arc.h> 91#include <net/if_ether.h> 92#include <net/if_types.h> 93 94#include <netinet/in.h> 95#include <netinet/if_inarp.h> 96 97 98#include <compat/sys/sockio.h> 99 100#ifndef BPF_BUFSIZE 101/* 102 * 4096 is too small for FDDI frames. 8192 is too small for gigabit Ethernet 103 * jumbos (circa 9k), ATM, or Intel gig/10gig ethernet jumbos (16k). 104 */ 105# define BPF_BUFSIZE 32768 106#endif 107 108#define PRINET 26 /* interruptible */ 109 110/* 111 * The default read buffer size, and limit for BIOCSBLEN, is sysctl'able. 112 * XXX the default values should be computed dynamically based 113 * on available memory size and available mbuf clusters. 114 */ 115static int bpf_bufsize = BPF_BUFSIZE; 116static int bpf_maxbufsize = BPF_DFLTBUFSIZE; /* XXX set dynamically, see above */ 117static bool bpf_jit = false; 118 119struct bpfjit_ops bpfjit_module_ops = { 120 .bj_generate_code = NULL, 121 .bj_free_code = NULL 122}; 123 124/* 125 * Global BPF statistics returned by net.bpf.stats sysctl. 126 */ 127static struct percpu *bpf_gstats_percpu; /* struct bpf_stat */ 128 129#define BPF_STATINC(id) \ 130 { \ 131 struct bpf_stat *__stats = \ 132 percpu_getref(bpf_gstats_percpu); \ 133 __stats->bs_##id++; \ 134 percpu_putref(bpf_gstats_percpu); \ 135 } 136 137/* 138 * Locking notes: 139 * - bpf_mtx (adaptive mutex) protects: 140 * - Gobal lists: bpf_iflist and bpf_dlist 141 * - struct bpf_if 142 * - bpf_close 143 * - bpf_psz (pserialize) 144 * - struct bpf_d has two mutexes: 145 * - bd_buf_mtx (spin mutex) protects the buffers that can be accessed 146 * on packet tapping 147 * - bd_mtx (adaptive mutex) protects member variables other than the buffers 148 * - Locking order: bpf_mtx => bpf_d#bd_mtx => bpf_d#bd_buf_mtx 149 * - struct bpf_d obtained via fp->f_bpf in bpf_read and bpf_write is 150 * never freed because struct bpf_d is only freed in bpf_close and 151 * bpf_close never be called while executing bpf_read and bpf_write 152 * - A filter that is assigned to bpf_d can be replaced with another filter 153 * while tapping packets, so it needs to be done atomically 154 * - struct bpf_d is iterated on bpf_dlist with psz 155 * - struct bpf_if is iterated on bpf_iflist with psz or psref 156 */ 157/* 158 * Use a mutex to avoid a race condition between gathering the stats/peers 159 * and opening/closing the device. 160 */ 161static kmutex_t bpf_mtx; 162 163static struct psref_class *bpf_psref_class __read_mostly; 164static pserialize_t bpf_psz; 165 166static inline void 167bpf_if_acquire(struct bpf_if *bp, struct psref *psref) 168{ 169 170 psref_acquire(psref, &bp->bif_psref, bpf_psref_class); 171} 172 173static inline void 174bpf_if_release(struct bpf_if *bp, struct psref *psref) 175{ 176 177 psref_release(psref, &bp->bif_psref, bpf_psref_class); 178} 179 180/* 181 * bpf_iflist is the list of interfaces; each corresponds to an ifnet 182 * bpf_dtab holds the descriptors, indexed by minor device # 183 */ 184static struct pslist_head bpf_iflist; 185static struct pslist_head bpf_dlist; 186 187/* Macros for bpf_d on bpf_dlist */ 188#define BPF_DLIST_WRITER_INSERT_HEAD(__d) \ 189 PSLIST_WRITER_INSERT_HEAD(&bpf_dlist, (__d), bd_bpf_dlist_entry) 190#define BPF_DLIST_READER_FOREACH(__d) \ 191 PSLIST_READER_FOREACH((__d), &bpf_dlist, struct bpf_d, \ 192 bd_bpf_dlist_entry) 193#define BPF_DLIST_WRITER_FOREACH(__d) \ 194 PSLIST_WRITER_FOREACH((__d), &bpf_dlist, struct bpf_d, \ 195 bd_bpf_dlist_entry) 196#define BPF_DLIST_ENTRY_INIT(__d) \ 197 PSLIST_ENTRY_INIT((__d), bd_bpf_dlist_entry) 198#define BPF_DLIST_WRITER_REMOVE(__d) \ 199 PSLIST_WRITER_REMOVE((__d), bd_bpf_dlist_entry) 200#define BPF_DLIST_ENTRY_DESTROY(__d) \ 201 PSLIST_ENTRY_DESTROY((__d), bd_bpf_dlist_entry) 202 203/* Macros for bpf_if on bpf_iflist */ 204#define BPF_IFLIST_WRITER_INSERT_HEAD(__bp) \ 205 PSLIST_WRITER_INSERT_HEAD(&bpf_iflist, (__bp), bif_iflist_entry) 206#define BPF_IFLIST_READER_FOREACH(__bp) \ 207 PSLIST_READER_FOREACH((__bp), &bpf_iflist, struct bpf_if, \ 208 bif_iflist_entry) 209#define BPF_IFLIST_WRITER_FOREACH(__bp) \ 210 PSLIST_WRITER_FOREACH((__bp), &bpf_iflist, struct bpf_if, \ 211 bif_iflist_entry) 212#define BPF_IFLIST_WRITER_REMOVE(__bp) \ 213 PSLIST_WRITER_REMOVE((__bp), bif_iflist_entry) 214#define BPF_IFLIST_ENTRY_INIT(__bp) \ 215 PSLIST_ENTRY_INIT((__bp), bif_iflist_entry) 216#define BPF_IFLIST_ENTRY_DESTROY(__bp) \ 217 PSLIST_ENTRY_DESTROY((__bp), bif_iflist_entry) 218 219/* Macros for bpf_d on bpf_if#bif_dlist_pslist */ 220#define BPFIF_DLIST_READER_FOREACH(__d, __bp) \ 221 PSLIST_READER_FOREACH((__d), &(__bp)->bif_dlist_head, struct bpf_d, \ 222 bd_bif_dlist_entry) 223#define BPFIF_DLIST_WRITER_INSERT_HEAD(__bp, __d) \ 224 PSLIST_WRITER_INSERT_HEAD(&(__bp)->bif_dlist_head, (__d), \ 225 bd_bif_dlist_entry) 226#define BPFIF_DLIST_WRITER_REMOVE(__d) \ 227 PSLIST_WRITER_REMOVE((__d), bd_bif_dlist_entry) 228#define BPFIF_DLIST_ENTRY_INIT(__d) \ 229 PSLIST_ENTRY_INIT((__d), bd_bif_dlist_entry) 230#define BPFIF_DLIST_READER_EMPTY(__bp) \ 231 (PSLIST_READER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d, \ 232 bd_bif_dlist_entry) == NULL) 233#define BPFIF_DLIST_WRITER_EMPTY(__bp) \ 234 (PSLIST_WRITER_FIRST(&(__bp)->bif_dlist_head, struct bpf_d, \ 235 bd_bif_dlist_entry) == NULL) 236#define BPFIF_DLIST_ENTRY_DESTROY(__d) \ 237 PSLIST_ENTRY_DESTROY((__d), bd_bif_dlist_entry) 238 239static int bpf_allocbufs(struct bpf_d *); 240static u_int bpf_xfilter(struct bpf_filter **, void *, u_int, u_int); 241static void bpf_deliver(struct bpf_if *, 242 void *(*cpfn)(void *, const void *, size_t), 243 void *, u_int, u_int, const u_int); 244static void bpf_freed(struct bpf_d *); 245static void bpf_free_filter(struct bpf_filter *); 246static void bpf_ifname(struct ifnet *, struct ifreq *); 247static void *bpf_mcpy(void *, const void *, size_t); 248static int bpf_movein(struct ifnet *, struct uio *, int, uint64_t, 249 struct mbuf **, struct sockaddr *, 250 struct bpf_filter **); 251static void bpf_attachd(struct bpf_d *, struct bpf_if *); 252static void bpf_detachd(struct bpf_d *); 253static int bpf_setif(struct bpf_d *, struct ifreq *); 254static int bpf_setf(struct bpf_d *, struct bpf_program *, u_long); 255static void bpf_timed_out(void *); 256static inline void 257 bpf_wakeup(struct bpf_d *); 258static int bpf_hdrlen(struct bpf_d *); 259static void catchpacket(struct bpf_d *, u_char *, u_int, u_int, 260 void *(*)(void *, const void *, size_t), struct timespec *); 261static void reset_d(struct bpf_d *); 262static int bpf_getdltlist(struct bpf_d *, struct bpf_dltlist *); 263static int bpf_setdlt(struct bpf_d *, u_int); 264 265static int bpf_read(struct file *, off_t *, struct uio *, kauth_cred_t, 266 int); 267static int bpf_write(struct file *, off_t *, struct uio *, kauth_cred_t, 268 int); 269static int bpf_ioctl(struct file *, u_long, void *); 270static int bpf_poll(struct file *, int); 271static int bpf_stat(struct file *, struct stat *); 272static int bpf_close(struct file *); 273static int bpf_kqfilter(struct file *, struct knote *); 274 275static const struct fileops bpf_fileops = { 276 .fo_name = "bpf", 277 .fo_read = bpf_read, 278 .fo_write = bpf_write, 279 .fo_ioctl = bpf_ioctl, 280 .fo_fcntl = fnullop_fcntl, 281 .fo_poll = bpf_poll, 282 .fo_stat = bpf_stat, 283 .fo_close = bpf_close, 284 .fo_kqfilter = bpf_kqfilter, 285 .fo_restart = fnullop_restart, 286}; 287 288dev_type_open(bpfopen); 289 290const struct cdevsw bpf_cdevsw = { 291 .d_open = bpfopen, 292 .d_close = noclose, 293 .d_read = noread, 294 .d_write = nowrite, 295 .d_ioctl = noioctl, 296 .d_stop = nostop, 297 .d_tty = notty, 298 .d_poll = nopoll, 299 .d_mmap = nommap, 300 .d_kqfilter = nokqfilter, 301 .d_discard = nodiscard, 302 .d_flag = D_OTHER | D_MPSAFE 303}; 304 305bpfjit_func_t 306bpf_jit_generate(bpf_ctx_t *bc, void *code, size_t size) 307{ 308 struct bpfjit_ops *ops = &bpfjit_module_ops; 309 bpfjit_func_t (*generate_code)(const bpf_ctx_t *, 310 const struct bpf_insn *, size_t); 311 312 generate_code = atomic_load_acquire(&ops->bj_generate_code); 313 if (generate_code != NULL) { 314 return generate_code(bc, code, size); 315 } 316 return NULL; 317} 318 319void 320bpf_jit_freecode(bpfjit_func_t jcode) 321{ 322 KASSERT(bpfjit_module_ops.bj_free_code != NULL); 323 bpfjit_module_ops.bj_free_code(jcode); 324} 325 326static int 327bpf_movein(struct ifnet *ifp, struct uio *uio, int linktype, uint64_t mtu, struct mbuf **mp, 328 struct sockaddr *sockp, struct bpf_filter **wfilter) 329{ 330 struct mbuf *m, *m0, *n; 331 int error; 332 size_t len; 333 size_t hlen; 334 size_t align; 335 u_int slen; 336 337 /* 338 * Build a sockaddr based on the data link layer type. 339 * We do this at this level because the ethernet header 340 * is copied directly into the data field of the sockaddr. 341 * In the case of SLIP, there is no header and the packet 342 * is forwarded as is. 343 * Also, we are careful to leave room at the front of the mbuf 344 * for the link level header. 345 */ 346 switch (linktype) { 347 348 case DLT_SLIP: 349 sockp->sa_family = AF_INET; 350 hlen = 0; 351 align = 0; 352 break; 353 354 case DLT_PPP: 355 sockp->sa_family = AF_UNSPEC; 356 hlen = 0; 357 align = 0; 358 break; 359 360 case DLT_EN10MB: 361 sockp->sa_family = AF_UNSPEC; 362 /* XXX Would MAXLINKHDR be better? */ 363 /* 6(dst)+6(src)+2(type) */ 364 hlen = sizeof(struct ether_header); 365 align = 2; 366 break; 367 368 case DLT_ARCNET: 369 sockp->sa_family = AF_UNSPEC; 370 hlen = ARC_HDRLEN; 371 align = 5; 372 break; 373 374 case DLT_FDDI: 375 sockp->sa_family = AF_LINK; 376 /* XXX 4(FORMAC)+6(dst)+6(src) */ 377 hlen = 16; 378 align = 0; 379 break; 380 381 case DLT_ECONET: 382 sockp->sa_family = AF_UNSPEC; 383 hlen = 6; 384 align = 2; 385 break; 386 387 case DLT_NULL: 388 sockp->sa_family = AF_UNSPEC; 389 if (ifp->if_type == IFT_LOOP) { 390 /* Set here to apply the following validations */ 391 hlen = sizeof(uint32_t); 392 } else 393 hlen = 0; 394 align = 0; 395 break; 396 397 default: 398 return (EIO); 399 } 400 401 len = uio->uio_resid; 402 /* 403 * If there aren't enough bytes for a link level header or the 404 * packet length exceeds the interface mtu, return an error. 405 */ 406 if (len - hlen > mtu) 407 return (EMSGSIZE); 408 409 m0 = m = m_gethdr(M_WAIT, MT_DATA); 410 m_reset_rcvif(m); 411 m->m_pkthdr.len = (int)(len - hlen); 412 if (len + align > MHLEN) { 413 m_clget(m, M_WAIT); 414 if ((m->m_flags & M_EXT) == 0) { 415 error = ENOBUFS; 416 goto bad; 417 } 418 } 419 420 /* Ensure the data is properly aligned */ 421 if (align > 0) 422 m->m_data += align; 423 424 for (;;) { 425 len = M_TRAILINGSPACE(m); 426 if (len > uio->uio_resid) 427 len = uio->uio_resid; 428 error = uiomove(mtod(m, void *), len, uio); 429 if (error) 430 goto bad; 431 m->m_len = len; 432 433 if (uio->uio_resid == 0) 434 break; 435 436 n = m_get(M_WAIT, MT_DATA); 437 m_clget(n, M_WAIT); /* if fails, there is no problem */ 438 m->m_next = n; 439 m = n; 440 } 441 442 slen = bpf_xfilter(wfilter, mtod(m, u_char *), len, len); 443 if (slen == 0) { 444 error = EPERM; 445 goto bad; 446 } 447 448 if (hlen != 0) { 449 if (linktype == DLT_NULL && ifp->if_type == IFT_LOOP) { 450 uint32_t af; 451 /* the link header indicates the address family */ 452 memcpy(&af, mtod(m0, void *), sizeof(af)); 453 sockp->sa_family = af; 454 } else { 455 /* move link level header in the top of mbuf to sa_data */ 456 memcpy(sockp->sa_data, mtod(m0, void *), hlen); 457 } 458 m0->m_data += hlen; 459 m0->m_len -= hlen; 460 } 461 462 *mp = m0; 463 return (0); 464 465bad: 466 m_freem(m0); 467 return (error); 468} 469 470/* 471 * Attach file to the bpf interface, i.e. make d listen on bp. 472 */ 473static void 474bpf_attachd(struct bpf_d *d, struct bpf_if *bp) 475{ 476 struct bpf_event_tracker *t; 477 478 KASSERT(mutex_owned(&bpf_mtx)); 479 KASSERT(mutex_owned(d->bd_mtx)); 480 /* 481 * Point d at bp, and add d to the interface's list of listeners. 482 * Finally, point the driver's bpf cookie at the interface so 483 * it will divert packets to bpf. 484 */ 485 d->bd_bif = bp; 486 BPFIF_DLIST_WRITER_INSERT_HEAD(bp, d); 487 488 *bp->bif_driverp = bp; 489 490 SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) { 491 t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt, 492 BPF_TRACK_EVENT_ATTACH); 493 } 494} 495 496/* 497 * Detach a file from its interface. 498 */ 499static void 500bpf_detachd(struct bpf_d *d) 501{ 502 struct bpf_if *bp; 503 struct bpf_event_tracker *t; 504 505 KASSERT(mutex_owned(&bpf_mtx)); 506 KASSERT(mutex_owned(d->bd_mtx)); 507 508 bp = d->bd_bif; 509 /* 510 * Check if this descriptor had requested promiscuous mode. 511 * If so, turn it off. 512 */ 513 if (d->bd_promisc) { 514 int error __diagused; 515 516 d->bd_promisc = 0; 517 /* 518 * Take device out of promiscuous mode. Since we were 519 * able to enter promiscuous mode, we should be able 520 * to turn it off. But we can get an error if 521 * the interface was configured down, so only panic 522 * if we don't get an unexpected error. 523 */ 524 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 525 error = ifpromisc(bp->bif_ifp, 0); 526 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 527#ifdef DIAGNOSTIC 528 if (error) 529 printf("%s: ifpromisc failed: %d", __func__, error); 530#endif 531 } 532 533 /* Remove d from the interface's descriptor list. */ 534 BPFIF_DLIST_WRITER_REMOVE(d); 535 536 pserialize_perform(bpf_psz); 537 538 if (BPFIF_DLIST_WRITER_EMPTY(bp)) { 539 /* 540 * Let the driver know that there are no more listeners. 541 */ 542 *d->bd_bif->bif_driverp = NULL; 543 } 544 545 d->bd_bif = NULL; 546 547 SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) { 548 t->bet_notify(bp, bp->bif_ifp, bp->bif_dlt, 549 BPF_TRACK_EVENT_DETACH); 550 } 551} 552 553static void 554bpf_init(void) 555{ 556 557 mutex_init(&bpf_mtx, MUTEX_DEFAULT, IPL_NONE); 558 bpf_psz = pserialize_create(); 559 bpf_psref_class = psref_class_create("bpf", IPL_SOFTNET); 560 561 PSLIST_INIT(&bpf_iflist); 562 PSLIST_INIT(&bpf_dlist); 563 564 bpf_gstats_percpu = percpu_alloc(sizeof(struct bpf_stat)); 565 566 return; 567} 568 569/* 570 * bpfilterattach() is called at boot time. We don't need to do anything 571 * here, since any initialization will happen as part of module init code. 572 */ 573/* ARGSUSED */ 574void 575bpfilterattach(int n) 576{ 577 578} 579 580/* 581 * Open ethernet device. Clones. 582 */ 583/* ARGSUSED */ 584int 585bpfopen(dev_t dev, int flag, int mode, struct lwp *l) 586{ 587 struct bpf_d *d; 588 struct file *fp; 589 int error, fd; 590 591 /* falloc() will fill in the descriptor for us. */ 592 if ((error = fd_allocfile(&fp, &fd)) != 0) 593 return error; 594 595 d = kmem_zalloc(sizeof(*d), KM_SLEEP); 596 d->bd_bufsize = bpf_bufsize; 597 d->bd_direction = BPF_D_INOUT; 598 d->bd_feedback = 0; 599 d->bd_pid = l->l_proc->p_pid; 600#ifdef _LP64 601 if (curproc->p_flag & PK_32) 602 d->bd_compat32 = 1; 603#endif 604 getnanotime(&d->bd_btime); 605 d->bd_atime = d->bd_mtime = d->bd_btime; 606 callout_init(&d->bd_callout, CALLOUT_MPSAFE); 607 selinit(&d->bd_sel); 608 d->bd_jitcode = NULL; 609 d->bd_rfilter = NULL; 610 d->bd_wfilter = NULL; 611 d->bd_locked = 0; 612 BPF_DLIST_ENTRY_INIT(d); 613 BPFIF_DLIST_ENTRY_INIT(d); 614 d->bd_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SOFTNET); 615 d->bd_buf_mtx = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NET); 616 cv_init(&d->bd_cv, "bpf"); 617 618 mutex_enter(&bpf_mtx); 619 BPF_DLIST_WRITER_INSERT_HEAD(d); 620 mutex_exit(&bpf_mtx); 621 622 return fd_clone(fp, fd, flag, &bpf_fileops, d); 623} 624 625/* 626 * Close the descriptor by detaching it from its interface, 627 * deallocating its buffers, and marking it free. 628 */ 629/* ARGSUSED */ 630static int 631bpf_close(struct file *fp) 632{ 633 struct bpf_d *d; 634 635 mutex_enter(&bpf_mtx); 636 637 if ((d = fp->f_bpf) == NULL) { 638 mutex_exit(&bpf_mtx); 639 return 0; 640 } 641 642 /* 643 * Refresh the PID associated with this bpf file. 644 */ 645 d->bd_pid = curproc->p_pid; 646 647 mutex_enter(d->bd_mtx); 648 if (d->bd_state == BPF_WAITING) 649 callout_halt(&d->bd_callout, d->bd_mtx); 650 d->bd_state = BPF_IDLE; 651 if (d->bd_bif) 652 bpf_detachd(d); 653 mutex_exit(d->bd_mtx); 654 655 BPF_DLIST_WRITER_REMOVE(d); 656 657 pserialize_perform(bpf_psz); 658 mutex_exit(&bpf_mtx); 659 660 BPFIF_DLIST_ENTRY_DESTROY(d); 661 BPF_DLIST_ENTRY_DESTROY(d); 662 fp->f_bpf = NULL; 663 bpf_freed(d); 664 callout_destroy(&d->bd_callout); 665 seldestroy(&d->bd_sel); 666 mutex_obj_free(d->bd_mtx); 667 mutex_obj_free(d->bd_buf_mtx); 668 cv_destroy(&d->bd_cv); 669 670 kmem_free(d, sizeof(*d)); 671 672 return (0); 673} 674 675/* 676 * Rotate the packet buffers in descriptor d. Move the store buffer 677 * into the hold slot, and the free buffer into the store slot. 678 * Zero the length of the new store buffer. 679 */ 680#define ROTATE_BUFFERS(d) \ 681 (d)->bd_hbuf = (d)->bd_sbuf; \ 682 (d)->bd_hlen = (d)->bd_slen; \ 683 (d)->bd_sbuf = (d)->bd_fbuf; \ 684 (d)->bd_slen = 0; \ 685 (d)->bd_fbuf = NULL; 686/* 687 * bpfread - read next chunk of packets from buffers 688 */ 689static int 690bpf_read(struct file *fp, off_t *offp, struct uio *uio, 691 kauth_cred_t cred, int flags) 692{ 693 struct bpf_d *d = fp->f_bpf; 694 int timed_out; 695 int error; 696 697 /* 698 * Refresh the PID associated with this bpf file. 699 */ 700 d->bd_pid = curproc->p_pid; 701 702 getnanotime(&d->bd_atime); 703 /* 704 * Restrict application to use a buffer the same size as 705 * the kernel buffers. 706 */ 707 if (uio->uio_resid != d->bd_bufsize) 708 return (EINVAL); 709 710 mutex_enter(d->bd_mtx); 711 if (d->bd_state == BPF_WAITING) 712 callout_halt(&d->bd_callout, d->bd_mtx); 713 timed_out = (d->bd_state == BPF_TIMED_OUT); 714 d->bd_state = BPF_IDLE; 715 mutex_exit(d->bd_mtx); 716 /* 717 * If the hold buffer is empty, then do a timed sleep, which 718 * ends when the timeout expires or when enough packets 719 * have arrived to fill the store buffer. 720 */ 721 mutex_enter(d->bd_buf_mtx); 722 while (d->bd_hbuf == NULL) { 723 if (fp->f_flag & FNONBLOCK) { 724 if (d->bd_slen == 0) { 725 error = EWOULDBLOCK; 726 goto out; 727 } 728 ROTATE_BUFFERS(d); 729 break; 730 } 731 732 if ((d->bd_immediate || timed_out) && d->bd_slen != 0) { 733 /* 734 * A packet(s) either arrived since the previous 735 * read or arrived while we were asleep. 736 * Rotate the buffers and return what's here. 737 */ 738 ROTATE_BUFFERS(d); 739 break; 740 } 741 742 error = cv_timedwait_sig(&d->bd_cv, d->bd_buf_mtx, d->bd_rtout); 743 744 if (error == EINTR || error == ERESTART) 745 goto out; 746 747 if (error == EWOULDBLOCK) { 748 /* 749 * On a timeout, return what's in the buffer, 750 * which may be nothing. If there is something 751 * in the store buffer, we can rotate the buffers. 752 */ 753 if (d->bd_hbuf) 754 /* 755 * We filled up the buffer in between 756 * getting the timeout and arriving 757 * here, so we don't need to rotate. 758 */ 759 break; 760 761 if (d->bd_slen == 0) { 762 error = 0; 763 goto out; 764 } 765 ROTATE_BUFFERS(d); 766 break; 767 } 768 if (error != 0) 769 goto out; 770 } 771 /* 772 * At this point, we know we have something in the hold slot. 773 */ 774 mutex_exit(d->bd_buf_mtx); 775 776 /* 777 * Move data from hold buffer into user space. 778 * We know the entire buffer is transferred since 779 * we checked above that the read buffer is bpf_bufsize bytes. 780 */ 781 error = uiomove(d->bd_hbuf, d->bd_hlen, uio); 782 783 mutex_enter(d->bd_buf_mtx); 784 d->bd_fbuf = d->bd_hbuf; 785 d->bd_hbuf = NULL; 786 d->bd_hlen = 0; 787out: 788 mutex_exit(d->bd_buf_mtx); 789 return (error); 790} 791 792 793/* 794 * If there are processes sleeping on this descriptor, wake them up. 795 */ 796static inline void 797bpf_wakeup(struct bpf_d *d) 798{ 799 800 mutex_enter(d->bd_buf_mtx); 801 cv_broadcast(&d->bd_cv); 802 mutex_exit(d->bd_buf_mtx); 803 804 if (d->bd_async) 805 fownsignal(d->bd_pgid, SIGIO, 0, 0, NULL); 806 selnotify(&d->bd_sel, 0, 0); 807} 808 809static void 810bpf_timed_out(void *arg) 811{ 812 struct bpf_d *d = arg; 813 814 mutex_enter(d->bd_mtx); 815 if (d->bd_state == BPF_WAITING) { 816 d->bd_state = BPF_TIMED_OUT; 817 if (d->bd_slen != 0) 818 bpf_wakeup(d); 819 } 820 mutex_exit(d->bd_mtx); 821} 822 823 824static int 825bpf_write(struct file *fp, off_t *offp, struct uio *uio, 826 kauth_cred_t cred, int flags) 827{ 828 struct bpf_d *d = fp->f_bpf; 829 struct bpf_if *bp; 830 struct ifnet *ifp; 831 struct mbuf *m, *mc; 832 int error; 833 static struct sockaddr_storage dst; 834 struct psref psref; 835 int bound; 836 837 /* 838 * Refresh the PID associated with this bpf file. 839 */ 840 d->bd_pid = curproc->p_pid; 841 842 m = NULL; /* XXX gcc */ 843 844 bound = curlwp_bind(); 845 mutex_enter(d->bd_mtx); 846 bp = d->bd_bif; 847 if (bp == NULL) { 848 mutex_exit(d->bd_mtx); 849 error = ENXIO; 850 goto out_bindx; 851 } 852 bpf_if_acquire(bp, &psref); 853 mutex_exit(d->bd_mtx); 854 855 getnanotime(&d->bd_mtime); 856 857 ifp = bp->bif_ifp; 858 if (if_is_deactivated(ifp)) { 859 error = ENXIO; 860 goto out; 861 } 862 863 if (uio->uio_resid == 0) { 864 error = 0; 865 goto out; 866 } 867 868 error = bpf_movein(ifp, uio, (int)bp->bif_dlt, ifp->if_mtu, &m, 869 (struct sockaddr *) &dst, &d->bd_wfilter); 870 if (error) 871 goto out; 872 873 if (m->m_pkthdr.len > ifp->if_mtu) { 874 m_freem(m); 875 error = EMSGSIZE; 876 goto out; 877 } 878 879 /* 880 * If writing to a loopback interface, the address family has 881 * already been specially computed in bpf_movein(), so don't 882 * clobber it, or the loopback will reject it in looutput(). 883 */ 884 if (d->bd_hdrcmplt && ifp->if_type != IFT_LOOP) 885 dst.ss_family = pseudo_AF_HDRCMPLT; 886 887 if (d->bd_feedback) { 888 mc = m_dup(m, 0, M_COPYALL, M_NOWAIT); 889 if (mc != NULL) 890 m_set_rcvif(mc, ifp); 891 /* Set M_PROMISC for outgoing packets to be discarded. */ 892 if (1 /*d->bd_direction == BPF_D_INOUT*/) 893 m->m_flags |= M_PROMISC; 894 } else 895 mc = NULL; 896 897 error = if_output_lock(ifp, ifp, m, (struct sockaddr *) &dst, NULL); 898 899 if (mc != NULL) { 900 if (error == 0) { 901 int s = splsoftnet(); 902 KERNEL_LOCK_UNLESS_IFP_MPSAFE(ifp); 903 ifp->_if_input(ifp, mc); 904 KERNEL_UNLOCK_UNLESS_IFP_MPSAFE(ifp); 905 splx(s); 906 } else 907 m_freem(mc); 908 } 909 /* 910 * The driver frees the mbuf. 911 */ 912out: 913 bpf_if_release(bp, &psref); 914out_bindx: 915 curlwp_bindx(bound); 916 return error; 917} 918 919/* 920 * Reset a descriptor by flushing its packet buffer and clearing the 921 * receive and drop counts. 922 */ 923static void 924reset_d(struct bpf_d *d) 925{ 926 927 KASSERT(mutex_owned(d->bd_mtx)); 928 929 mutex_enter(d->bd_buf_mtx); 930 if (d->bd_hbuf) { 931 /* Free the hold buffer. */ 932 d->bd_fbuf = d->bd_hbuf; 933 d->bd_hbuf = NULL; 934 } 935 d->bd_slen = 0; 936 d->bd_hlen = 0; 937 d->bd_rcount = 0; 938 d->bd_dcount = 0; 939 d->bd_ccount = 0; 940 mutex_exit(d->bd_buf_mtx); 941} 942 943/* 944 * FIONREAD Check for read packet available. 945 * BIOCGBLEN Get buffer len [for read()]. 946 * BIOCSETF Set ethernet read filter. 947 * BIOCFLUSH Flush read packet buffer. 948 * BIOCPROMISC Put interface into promiscuous mode. 949 * BIOCGDLT Get link layer type. 950 * BIOCGETIF Get interface name. 951 * BIOCSETIF Set interface. 952 * BIOCSRTIMEOUT Set read timeout. 953 * BIOCGRTIMEOUT Get read timeout. 954 * BIOCGSTATS Get packet stats. 955 * BIOCIMMEDIATE Set immediate mode. 956 * BIOCVERSION Get filter language version. 957 * BIOCGHDRCMPLT Get "header already complete" flag. 958 * BIOCSHDRCMPLT Set "header already complete" flag. 959 * BIOCSFEEDBACK Set packet feedback mode. 960 * BIOCGFEEDBACK Get packet feedback mode. 961 * BIOCGDIRECTION Get packet direction flag 962 * BIOCSDIRECTION Set packet direction flag 963 */ 964/* ARGSUSED */ 965static int 966bpf_ioctl(struct file *fp, u_long cmd, void *addr) 967{ 968 struct bpf_d *d = fp->f_bpf; 969 int error = 0; 970 971 /* 972 * Refresh the PID associated with this bpf file. 973 */ 974 d->bd_pid = curproc->p_pid; 975#ifdef _LP64 976 if (curproc->p_flag & PK_32) 977 d->bd_compat32 = 1; 978 else 979 d->bd_compat32 = 0; 980#endif 981 982 mutex_enter(d->bd_mtx); 983 if (d->bd_state == BPF_WAITING) 984 callout_halt(&d->bd_callout, d->bd_mtx); 985 d->bd_state = BPF_IDLE; 986 mutex_exit(d->bd_mtx); 987 988 if (d->bd_locked) { 989 switch (cmd) { 990 case BIOCGBLEN: /* FALLTHROUGH */ 991 case BIOCFLUSH: /* FALLTHROUGH */ 992 case BIOCGDLT: /* FALLTHROUGH */ 993 case BIOCGDLTLIST: /* FALLTHROUGH */ 994 case BIOCGETIF: /* FALLTHROUGH */ 995 case BIOCGRTIMEOUT: /* FALLTHROUGH */ 996 case BIOCGSTATS: /* FALLTHROUGH */ 997 case BIOCVERSION: /* FALLTHROUGH */ 998 case BIOCGHDRCMPLT: /* FALLTHROUGH */ 999 case FIONREAD: /* FALLTHROUGH */ 1000 case BIOCLOCK: /* FALLTHROUGH */ 1001 case BIOCSRTIMEOUT: /* FALLTHROUGH */ 1002 case BIOCIMMEDIATE: /* FALLTHROUGH */ 1003 case TIOCGPGRP: 1004 break; 1005 default: 1006 return EPERM; 1007 } 1008 } 1009 1010 switch (cmd) { 1011 1012 default: 1013 error = EINVAL; 1014 break; 1015 1016 /* 1017 * Check for read packet available. 1018 */ 1019 case FIONREAD: 1020 { 1021 int n; 1022 1023 mutex_enter(d->bd_buf_mtx); 1024 n = d->bd_slen; 1025 if (d->bd_hbuf) 1026 n += d->bd_hlen; 1027 mutex_exit(d->bd_buf_mtx); 1028 1029 *(int *)addr = n; 1030 break; 1031 } 1032 1033 /* 1034 * Get buffer len [for read()]. 1035 */ 1036 case BIOCGBLEN: 1037 *(u_int *)addr = d->bd_bufsize; 1038 break; 1039 1040 /* 1041 * Set buffer length. 1042 */ 1043 case BIOCSBLEN: 1044 /* 1045 * Forbid to change the buffer length if buffers are already 1046 * allocated. 1047 */ 1048 mutex_enter(d->bd_mtx); 1049 mutex_enter(d->bd_buf_mtx); 1050 if (d->bd_bif != NULL || d->bd_sbuf != NULL) 1051 error = EINVAL; 1052 else { 1053 u_int size = *(u_int *)addr; 1054 1055 if (size > bpf_maxbufsize) 1056 *(u_int *)addr = size = bpf_maxbufsize; 1057 else if (size < BPF_MINBUFSIZE) 1058 *(u_int *)addr = size = BPF_MINBUFSIZE; 1059 d->bd_bufsize = size; 1060 } 1061 mutex_exit(d->bd_buf_mtx); 1062 mutex_exit(d->bd_mtx); 1063 break; 1064 1065 /* 1066 * Set link layer read filter. 1067 */ 1068 case BIOCSETF: /* FALLTHROUGH */ 1069 case BIOCSETWF: 1070 error = bpf_setf(d, addr, cmd); 1071 break; 1072 1073 case BIOCLOCK: 1074 d->bd_locked = 1; 1075 break; 1076 1077 /* 1078 * Flush read packet buffer. 1079 */ 1080 case BIOCFLUSH: 1081 mutex_enter(d->bd_mtx); 1082 reset_d(d); 1083 mutex_exit(d->bd_mtx); 1084 break; 1085 1086 /* 1087 * Put interface into promiscuous mode. 1088 */ 1089 case BIOCPROMISC: 1090 mutex_enter(d->bd_mtx); 1091 if (d->bd_bif == NULL) { 1092 mutex_exit(d->bd_mtx); 1093 /* 1094 * No interface attached yet. 1095 */ 1096 error = EINVAL; 1097 break; 1098 } 1099 if (d->bd_promisc == 0) { 1100 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 1101 error = ifpromisc(d->bd_bif->bif_ifp, 1); 1102 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 1103 if (error == 0) 1104 d->bd_promisc = 1; 1105 } 1106 mutex_exit(d->bd_mtx); 1107 break; 1108 1109 /* 1110 * Get device parameters. 1111 */ 1112 case BIOCGDLT: 1113 mutex_enter(d->bd_mtx); 1114 if (d->bd_bif == NULL) 1115 error = EINVAL; 1116 else 1117 *(u_int *)addr = d->bd_bif->bif_dlt; 1118 mutex_exit(d->bd_mtx); 1119 break; 1120 1121 /* 1122 * Get a list of supported device parameters. 1123 */ 1124 case BIOCGDLTLIST: 1125 mutex_enter(d->bd_mtx); 1126 if (d->bd_bif == NULL) 1127 error = EINVAL; 1128 else 1129 error = bpf_getdltlist(d, addr); 1130 mutex_exit(d->bd_mtx); 1131 break; 1132 1133 /* 1134 * Set device parameters. 1135 */ 1136 case BIOCSDLT: 1137 mutex_enter(&bpf_mtx); 1138 mutex_enter(d->bd_mtx); 1139 if (d->bd_bif == NULL) 1140 error = EINVAL; 1141 else 1142 error = bpf_setdlt(d, *(u_int *)addr); 1143 mutex_exit(d->bd_mtx); 1144 mutex_exit(&bpf_mtx); 1145 break; 1146 1147 /* 1148 * Set interface name. 1149 */ 1150#ifdef OBIOCGETIF 1151 case OBIOCGETIF: 1152#endif 1153 case BIOCGETIF: 1154 mutex_enter(d->bd_mtx); 1155 if (d->bd_bif == NULL) 1156 error = EINVAL; 1157 else 1158 bpf_ifname(d->bd_bif->bif_ifp, addr); 1159 mutex_exit(d->bd_mtx); 1160 break; 1161 1162 /* 1163 * Set interface. 1164 */ 1165#ifdef OBIOCSETIF 1166 case OBIOCSETIF: 1167#endif 1168 case BIOCSETIF: 1169 mutex_enter(&bpf_mtx); 1170 error = bpf_setif(d, addr); 1171 mutex_exit(&bpf_mtx); 1172 break; 1173 1174 /* 1175 * Set read timeout. 1176 */ 1177 case BIOCSRTIMEOUT: 1178 { 1179 struct timeval *tv = addr; 1180 1181 /* Compute number of ticks. */ 1182 if (tv->tv_sec < 0 || 1183 tv->tv_usec < 0 || tv->tv_usec >= 1000000) { 1184 error = EINVAL; 1185 break; 1186 } else if (tv->tv_sec > INT_MAX/hz - 1) { 1187 d->bd_rtout = INT_MAX; 1188 } else { 1189 d->bd_rtout = tv->tv_sec * hz 1190 + tv->tv_usec / tick; 1191 } 1192 if ((d->bd_rtout == 0) && (tv->tv_usec != 0)) 1193 d->bd_rtout = 1; 1194 break; 1195 } 1196 1197#ifdef BIOCGORTIMEOUT 1198 /* 1199 * Get read timeout. 1200 */ 1201 case BIOCGORTIMEOUT: 1202 { 1203 struct timeval50 *tv = addr; 1204 1205 tv->tv_sec = d->bd_rtout / hz; 1206 tv->tv_usec = (d->bd_rtout % hz) * tick; 1207 break; 1208 } 1209#endif 1210 1211#ifdef BIOCSORTIMEOUT 1212 /* 1213 * Set read timeout. 1214 */ 1215 case BIOCSORTIMEOUT: 1216 { 1217 struct timeval50 *tv = addr; 1218 1219 /* Compute number of ticks. */ 1220 if (tv->tv_sec < 0 || 1221 tv->tv_usec < 0 || tv->tv_usec >= 1000000) { 1222 error = EINVAL; 1223 break; 1224 } else if (tv->tv_sec > INT_MAX/hz - 1) { 1225 d->bd_rtout = INT_MAX; 1226 } else { 1227 d->bd_rtout = tv->tv_sec * hz 1228 + tv->tv_usec / tick; 1229 } 1230 if ((d->bd_rtout == 0) && (tv->tv_usec != 0)) 1231 d->bd_rtout = 1; 1232 break; 1233 } 1234#endif 1235 1236 /* 1237 * Get read timeout. 1238 */ 1239 case BIOCGRTIMEOUT: 1240 { 1241 struct timeval *tv = addr; 1242 1243 tv->tv_sec = d->bd_rtout / hz; 1244 tv->tv_usec = (d->bd_rtout % hz) * tick; 1245 break; 1246 } 1247 /* 1248 * Get packet stats. 1249 */ 1250 case BIOCGSTATS: 1251 { 1252 struct bpf_stat *bs = addr; 1253 1254 bs->bs_recv = d->bd_rcount; 1255 bs->bs_drop = d->bd_dcount; 1256 bs->bs_capt = d->bd_ccount; 1257 break; 1258 } 1259 1260 case BIOCGSTATS_30: 1261 { 1262 struct bpf_stat30 *bs = addr; 1263 1264 bs->bs_recv = d->bd_rcount; 1265 bs->bs_drop = d->bd_dcount; 1266 break; 1267 } 1268 1269 /* 1270 * Set immediate mode. 1271 */ 1272 case BIOCIMMEDIATE: 1273 d->bd_immediate = *(u_int *)addr; 1274 break; 1275 1276 case BIOCVERSION: 1277 { 1278 struct bpf_version *bv = addr; 1279 1280 bv->bv_major = BPF_MAJOR_VERSION; 1281 bv->bv_minor = BPF_MINOR_VERSION; 1282 break; 1283 } 1284 1285 case BIOCGHDRCMPLT: /* get "header already complete" flag */ 1286 *(u_int *)addr = d->bd_hdrcmplt; 1287 break; 1288 1289 case BIOCSHDRCMPLT: /* set "header already complete" flag */ 1290 d->bd_hdrcmplt = *(u_int *)addr ? 1 : 0; 1291 break; 1292 1293 /* 1294 * Get packet direction flag 1295 */ 1296 case BIOCGDIRECTION: 1297 *(u_int *)addr = d->bd_direction; 1298 break; 1299 1300 /* 1301 * Set packet direction flag 1302 */ 1303 case BIOCSDIRECTION: 1304 { 1305 u_int direction; 1306 1307 direction = *(u_int *)addr; 1308 switch (direction) { 1309 case BPF_D_IN: 1310 case BPF_D_INOUT: 1311 case BPF_D_OUT: 1312 d->bd_direction = direction; 1313 break; 1314 default: 1315 error = EINVAL; 1316 } 1317 } 1318 break; 1319 1320 /* 1321 * Set "feed packets from bpf back to input" mode 1322 */ 1323 case BIOCSFEEDBACK: 1324 d->bd_feedback = *(u_int *)addr; 1325 break; 1326 1327 /* 1328 * Get "feed packets from bpf back to input" mode 1329 */ 1330 case BIOCGFEEDBACK: 1331 *(u_int *)addr = d->bd_feedback; 1332 break; 1333 1334 case FIONBIO: /* Non-blocking I/O */ 1335 /* 1336 * No need to do anything special as we use IO_NDELAY in 1337 * bpfread() as an indication of whether or not to block 1338 * the read. 1339 */ 1340 break; 1341 1342 case FIOASYNC: /* Send signal on receive packets */ 1343 mutex_enter(d->bd_mtx); 1344 d->bd_async = *(int *)addr; 1345 mutex_exit(d->bd_mtx); 1346 break; 1347 1348 case TIOCSPGRP: /* Process or group to send signals to */ 1349 case FIOSETOWN: 1350 error = fsetown(&d->bd_pgid, cmd, addr); 1351 break; 1352 1353 case TIOCGPGRP: 1354 case FIOGETOWN: 1355 error = fgetown(d->bd_pgid, cmd, addr); 1356 break; 1357 } 1358 return (error); 1359} 1360 1361/* 1362 * Set d's packet filter program to fp. If this file already has a filter, 1363 * free it and replace it. Returns EINVAL for bogus requests. 1364 */ 1365static int 1366bpf_setf(struct bpf_d *d, struct bpf_program *fp, u_long cmd) 1367{ 1368 struct bpf_insn *fcode; 1369 bpfjit_func_t jcode; 1370 size_t flen, size = 0; 1371 struct bpf_filter *oldf, *newf, **storef; 1372 1373 jcode = NULL; 1374 flen = fp->bf_len; 1375 1376 if ((fp->bf_insns == NULL && flen) || flen > BPF_MAXINSNS) { 1377 return EINVAL; 1378 } 1379 1380 if (flen) { 1381 /* 1382 * Allocate the buffer, copy the byte-code from 1383 * userspace and validate it. 1384 */ 1385 size = flen * sizeof(*fp->bf_insns); 1386 fcode = kmem_alloc(size, KM_SLEEP); 1387 if (copyin(fp->bf_insns, fcode, size) != 0 || 1388 !bpf_validate(fcode, (int)flen)) { 1389 kmem_free(fcode, size); 1390 return EINVAL; 1391 } 1392 if (bpf_jit) 1393 jcode = bpf_jit_generate(NULL, fcode, flen); 1394 } else { 1395 fcode = NULL; 1396 } 1397 1398 newf = kmem_alloc(sizeof(*newf), KM_SLEEP); 1399 newf->bf_insn = fcode; 1400 newf->bf_size = size; 1401 newf->bf_jitcode = jcode; 1402 if (cmd == BIOCSETF) 1403 d->bd_jitcode = jcode; /* XXX just for kvm(3) users */ 1404 1405 /* Need to hold bpf_mtx for pserialize_perform */ 1406 mutex_enter(&bpf_mtx); 1407 mutex_enter(d->bd_mtx); 1408 if (cmd == BIOCSETWF) { 1409 oldf = d->bd_wfilter; 1410 storef = &d->bd_wfilter; 1411 } else { 1412 oldf = d->bd_rfilter; 1413 storef = &d->bd_rfilter; 1414 } 1415 atomic_store_release(storef, newf); 1416 reset_d(d); 1417 pserialize_perform(bpf_psz); 1418 mutex_exit(d->bd_mtx); 1419 mutex_exit(&bpf_mtx); 1420 1421 if (oldf != NULL) 1422 bpf_free_filter(oldf); 1423 1424 return 0; 1425} 1426 1427/* 1428 * Detach a file from its current interface (if attached at all) and attach 1429 * to the interface indicated by the name stored in ifr. 1430 * Return an errno or 0. 1431 */ 1432static int 1433bpf_setif(struct bpf_d *d, struct ifreq *ifr) 1434{ 1435 struct bpf_if *bp; 1436 char *cp; 1437 int unit_seen, i, error; 1438 1439 KASSERT(mutex_owned(&bpf_mtx)); 1440 /* 1441 * Make sure the provided name has a unit number, and default 1442 * it to '0' if not specified. 1443 * XXX This is ugly ... do this differently? 1444 */ 1445 unit_seen = 0; 1446 cp = ifr->ifr_name; 1447 cp[sizeof(ifr->ifr_name) - 1] = '\0'; /* sanity */ 1448 while (*cp++) 1449 if (*cp >= '0' && *cp <= '9') 1450 unit_seen = 1; 1451 if (!unit_seen) { 1452 /* Make sure to leave room for the '\0'. */ 1453 for (i = 0; i < (IFNAMSIZ - 1); ++i) { 1454 if ((ifr->ifr_name[i] >= 'a' && 1455 ifr->ifr_name[i] <= 'z') || 1456 (ifr->ifr_name[i] >= 'A' && 1457 ifr->ifr_name[i] <= 'Z')) 1458 continue; 1459 ifr->ifr_name[i] = '0'; 1460 } 1461 } 1462 1463 /* 1464 * Look through attached interfaces for the named one. 1465 */ 1466 BPF_IFLIST_WRITER_FOREACH(bp) { 1467 struct ifnet *ifp = bp->bif_ifp; 1468 1469 if (ifp == NULL || 1470 strcmp(ifp->if_xname, ifr->ifr_name) != 0) 1471 continue; 1472 /* skip additional entry */ 1473 if (bp->bif_driverp != &ifp->if_bpf) 1474 continue; 1475 /* 1476 * We found the requested interface. 1477 * Allocate the packet buffers if we need to. 1478 * If we're already attached to requested interface, 1479 * just flush the buffer. 1480 */ 1481 /* 1482 * bpf_allocbufs is called only here. bpf_mtx ensures that 1483 * no race condition happen on d->bd_sbuf. 1484 */ 1485 if (d->bd_sbuf == NULL) { 1486 error = bpf_allocbufs(d); 1487 if (error != 0) 1488 return (error); 1489 } 1490 mutex_enter(d->bd_mtx); 1491 if (bp != d->bd_bif) { 1492 if (d->bd_bif) { 1493 /* 1494 * Detach if attached to something else. 1495 */ 1496 bpf_detachd(d); 1497 BPFIF_DLIST_ENTRY_INIT(d); 1498 } 1499 1500 bpf_attachd(d, bp); 1501 } 1502 reset_d(d); 1503 mutex_exit(d->bd_mtx); 1504 return (0); 1505 } 1506 /* Not found. */ 1507 return (ENXIO); 1508} 1509 1510/* 1511 * Copy the interface name to the ifreq. 1512 */ 1513static void 1514bpf_ifname(struct ifnet *ifp, struct ifreq *ifr) 1515{ 1516 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 1517} 1518 1519static int 1520bpf_stat(struct file *fp, struct stat *st) 1521{ 1522 struct bpf_d *d = fp->f_bpf; 1523 1524 (void)memset(st, 0, sizeof(*st)); 1525 mutex_enter(d->bd_mtx); 1526 st->st_dev = makedev(cdevsw_lookup_major(&bpf_cdevsw), d->bd_pid); 1527 st->st_atimespec = d->bd_atime; 1528 st->st_mtimespec = d->bd_mtime; 1529 st->st_ctimespec = st->st_birthtimespec = d->bd_btime; 1530 st->st_uid = kauth_cred_geteuid(fp->f_cred); 1531 st->st_gid = kauth_cred_getegid(fp->f_cred); 1532 st->st_mode = S_IFCHR; 1533 mutex_exit(d->bd_mtx); 1534 return 0; 1535} 1536 1537/* 1538 * Support for poll() system call 1539 * 1540 * Return true iff the specific operation will not block indefinitely - with 1541 * the assumption that it is safe to positively acknowledge a request for the 1542 * ability to write to the BPF device. 1543 * Otherwise, return false but make a note that a selnotify() must be done. 1544 */ 1545static int 1546bpf_poll(struct file *fp, int events) 1547{ 1548 struct bpf_d *d = fp->f_bpf; 1549 int revents; 1550 1551 /* 1552 * Refresh the PID associated with this bpf file. 1553 */ 1554 mutex_enter(&bpf_mtx); 1555 d->bd_pid = curproc->p_pid; 1556 1557 revents = events & (POLLOUT | POLLWRNORM); 1558 if (events & (POLLIN | POLLRDNORM)) { 1559 /* 1560 * An imitation of the FIONREAD ioctl code. 1561 */ 1562 mutex_enter(d->bd_mtx); 1563 if (d->bd_hlen != 0 || 1564 ((d->bd_immediate || d->bd_state == BPF_TIMED_OUT) && 1565 d->bd_slen != 0)) { 1566 revents |= events & (POLLIN | POLLRDNORM); 1567 } else { 1568 selrecord(curlwp, &d->bd_sel); 1569 /* Start the read timeout if necessary */ 1570 if (d->bd_rtout > 0 && d->bd_state == BPF_IDLE) { 1571 callout_reset(&d->bd_callout, d->bd_rtout, 1572 bpf_timed_out, d); 1573 d->bd_state = BPF_WAITING; 1574 } 1575 } 1576 mutex_exit(d->bd_mtx); 1577 } 1578 1579 mutex_exit(&bpf_mtx); 1580 return (revents); 1581} 1582 1583static void 1584filt_bpfrdetach(struct knote *kn) 1585{ 1586 struct bpf_d *d = kn->kn_hook; 1587 1588 mutex_enter(d->bd_buf_mtx); 1589 selremove_knote(&d->bd_sel, kn); 1590 mutex_exit(d->bd_buf_mtx); 1591} 1592 1593static int 1594filt_bpfread(struct knote *kn, long hint) 1595{ 1596 struct bpf_d *d = kn->kn_hook; 1597 int rv; 1598 1599 /* 1600 * Refresh the PID associated with this bpf file. 1601 */ 1602 d->bd_pid = curproc->p_pid; 1603 1604 mutex_enter(d->bd_buf_mtx); 1605 kn->kn_data = d->bd_hlen; 1606 if (d->bd_immediate) 1607 kn->kn_data += d->bd_slen; 1608 rv = (kn->kn_data > 0); 1609 mutex_exit(d->bd_buf_mtx); 1610 return rv; 1611} 1612 1613static const struct filterops bpfread_filtops = { 1614 .f_flags = FILTEROP_ISFD, 1615 .f_attach = NULL, 1616 .f_detach = filt_bpfrdetach, 1617 .f_event = filt_bpfread, 1618}; 1619 1620static int 1621bpf_kqfilter(struct file *fp, struct knote *kn) 1622{ 1623 struct bpf_d *d = fp->f_bpf; 1624 1625 switch (kn->kn_filter) { 1626 case EVFILT_READ: 1627 kn->kn_fop = &bpfread_filtops; 1628 break; 1629 1630 default: 1631 return (EINVAL); 1632 } 1633 1634 kn->kn_hook = d; 1635 1636 mutex_enter(d->bd_buf_mtx); 1637 selrecord_knote(&d->bd_sel, kn); 1638 mutex_exit(d->bd_buf_mtx); 1639 1640 return (0); 1641} 1642 1643/* 1644 * Copy data from an mbuf chain into a buffer. This code is derived 1645 * from m_copydata in sys/uipc_mbuf.c. 1646 */ 1647static void * 1648bpf_mcpy(void *dst_arg, const void *src_arg, size_t len) 1649{ 1650 const struct mbuf *m; 1651 u_int count; 1652 u_char *dst; 1653 1654 m = src_arg; 1655 dst = dst_arg; 1656 while (len > 0) { 1657 if (m == NULL) 1658 panic("bpf_mcpy"); 1659 count = uimin(m->m_len, len); 1660 memcpy(dst, mtod(m, const void *), count); 1661 m = m->m_next; 1662 dst += count; 1663 len -= count; 1664 } 1665 return dst_arg; 1666} 1667 1668static inline u_int 1669bpf_xfilter(struct bpf_filter **filter, void *pkt, u_int pktlen, u_int buflen) 1670{ 1671 struct bpf_filter *filt; 1672 uint32_t mem[BPF_MEMWORDS]; 1673 bpf_args_t args = { 1674 .pkt = (const uint8_t *)pkt, 1675 .wirelen = pktlen, 1676 .buflen = buflen, 1677 .mem = mem, 1678 .arg = NULL 1679 }; 1680 u_int slen; 1681 1682 filt = atomic_load_consume(filter); 1683 if (filt == NULL) /* No filter means accept all. */ 1684 return (u_int)-1; 1685 1686 if (filt->bf_jitcode != NULL) 1687 slen = filt->bf_jitcode(NULL, &args); 1688 else 1689 slen = bpf_filter_ext(NULL, filt->bf_insn, &args); 1690 return slen; 1691} 1692 1693/* 1694 * Dispatch a packet to all the listeners on interface bp. 1695 * 1696 * pkt pointer to the packet, either a data buffer or an mbuf chain 1697 * buflen buffer length, if pkt is a data buffer 1698 * cpfn a function that can copy pkt into the listener's buffer 1699 * pktlen length of the packet 1700 * direction BPF_D_IN or BPF_D_OUT 1701 */ 1702static inline void 1703bpf_deliver(struct bpf_if *bp, void *(*cpfn)(void *, const void *, size_t), 1704 void *pkt, u_int pktlen, u_int buflen, const u_int direction) 1705{ 1706 bool gottime = false; 1707 struct timespec ts; 1708 struct bpf_d *d; 1709 int s; 1710 u_int slen; 1711 1712 KASSERT(!cpu_intr_p()); 1713 1714 /* 1715 * Note that the IPL does not have to be raised at this point. 1716 * The only problem that could arise here is that if two different 1717 * interfaces shared any data. This is not the case. 1718 */ 1719 s = pserialize_read_enter(); 1720 BPFIF_DLIST_READER_FOREACH(d, bp) { 1721 if (direction == BPF_D_IN) { 1722 if (d->bd_direction == BPF_D_OUT) 1723 continue; 1724 } else { /* BPF_D_OUT */ 1725 if (d->bd_direction == BPF_D_IN) 1726 continue; 1727 } 1728 1729 atomic_inc_ulong(&d->bd_rcount); 1730 BPF_STATINC(recv); 1731 1732 slen = bpf_xfilter(&d->bd_rfilter, pkt, pktlen, buflen); 1733 if (slen == 0) 1734 continue; 1735 1736 if (!gottime) { 1737 gottime = true; 1738 nanotime(&ts); 1739 } 1740 /* Assume catchpacket doesn't sleep */ 1741 catchpacket(d, pkt, pktlen, slen, cpfn, &ts); 1742 } 1743 pserialize_read_exit(s); 1744} 1745 1746/* 1747 * Incoming linkage from device drivers, when the head of the packet is in 1748 * a buffer, and the tail is in an mbuf chain. 1749 */ 1750static void 1751_bpf_mtap2(struct bpf_if *bp, void *data, u_int dlen, struct mbuf *m, 1752 u_int direction) 1753{ 1754 u_int pktlen; 1755 struct mbuf mb; 1756 1757 /* Skip outgoing duplicate packets. */ 1758 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) { 1759 m->m_flags &= ~M_PROMISC; 1760 return; 1761 } 1762 1763 pktlen = m_length(m) + dlen; 1764 1765 /* 1766 * Craft on-stack mbuf suitable for passing to bpf_filter. 1767 * Note that we cut corners here; we only set up what's 1768 * absolutely needed--this mbuf should never go anywhere else. 1769 */ 1770 (void)memset(&mb, 0, sizeof(mb)); 1771 mb.m_type = MT_DATA; 1772 mb.m_next = m; 1773 mb.m_data = data; 1774 mb.m_len = dlen; 1775 1776 bpf_deliver(bp, bpf_mcpy, &mb, pktlen, 0, direction); 1777} 1778 1779/* 1780 * Incoming linkage from device drivers, when packet is in an mbuf chain. 1781 */ 1782static void 1783_bpf_mtap(struct bpf_if *bp, struct mbuf *m, u_int direction) 1784{ 1785 void *(*cpfn)(void *, const void *, size_t); 1786 u_int pktlen, buflen; 1787 void *marg; 1788 1789 /* Skip outgoing duplicate packets. */ 1790 if ((m->m_flags & M_PROMISC) != 0 && m->m_pkthdr.rcvif_index == 0) { 1791 m->m_flags &= ~M_PROMISC; 1792 return; 1793 } 1794 1795 pktlen = m_length(m); 1796 1797 /* Skip zero-sized packets. */ 1798 if (__predict_false(pktlen == 0)) { 1799 return; 1800 } 1801 1802 if (pktlen == m->m_len) { 1803 cpfn = (void *)memcpy; 1804 marg = mtod(m, void *); 1805 buflen = pktlen; 1806 KASSERT(buflen != 0); 1807 } else { 1808 cpfn = bpf_mcpy; 1809 marg = m; 1810 buflen = 0; 1811 } 1812 1813 bpf_deliver(bp, cpfn, marg, pktlen, buflen, direction); 1814} 1815 1816/* 1817 * We need to prepend the address family as 1818 * a four byte field. Cons up a dummy header 1819 * to pacify bpf. This is safe because bpf 1820 * will only read from the mbuf (i.e., it won't 1821 * try to free it or keep a pointer a to it). 1822 */ 1823static void 1824_bpf_mtap_af(struct bpf_if *bp, uint32_t af, struct mbuf *m, u_int direction) 1825{ 1826 struct mbuf m0; 1827 1828 m0.m_type = MT_DATA; 1829 m0.m_flags = 0; 1830 m0.m_next = m; 1831 m0.m_nextpkt = NULL; 1832 m0.m_owner = NULL; 1833 m0.m_len = 4; 1834 m0.m_data = (char *)⁡ 1835 1836 _bpf_mtap(bp, &m0, direction); 1837} 1838 1839/* 1840 * Put the SLIP pseudo-"link header" in place. 1841 * Note this M_PREPEND() should never fail, 1842 * since we know we always have enough space 1843 * in the input buffer. 1844 */ 1845static void 1846_bpf_mtap_sl_in(struct bpf_if *bp, u_char *chdr, struct mbuf **m) 1847{ 1848 u_char *hp; 1849 1850 M_PREPEND(*m, SLIP_HDRLEN, M_DONTWAIT); 1851 if (*m == NULL) 1852 return; 1853 1854 hp = mtod(*m, u_char *); 1855 hp[SLX_DIR] = SLIPDIR_IN; 1856 (void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN); 1857 1858 _bpf_mtap(bp, *m, BPF_D_IN); 1859 1860 m_adj(*m, SLIP_HDRLEN); 1861} 1862 1863/* 1864 * Put the SLIP pseudo-"link header" in 1865 * place. The compressed header is now 1866 * at the beginning of the mbuf. 1867 */ 1868static void 1869_bpf_mtap_sl_out(struct bpf_if *bp, u_char *chdr, struct mbuf *m) 1870{ 1871 struct mbuf m0; 1872 u_char *hp; 1873 1874 m0.m_type = MT_DATA; 1875 m0.m_flags = 0; 1876 m0.m_next = m; 1877 m0.m_nextpkt = NULL; 1878 m0.m_owner = NULL; 1879 m0.m_data = m0.m_dat; 1880 m0.m_len = SLIP_HDRLEN; 1881 1882 hp = mtod(&m0, u_char *); 1883 1884 hp[SLX_DIR] = SLIPDIR_OUT; 1885 (void)memcpy(&hp[SLX_CHDR], chdr, CHDR_LEN); 1886 1887 _bpf_mtap(bp, &m0, BPF_D_OUT); 1888 m_freem(m); 1889} 1890 1891static struct mbuf * 1892bpf_mbuf_enqueue(struct bpf_if *bp, struct mbuf *m) 1893{ 1894 struct mbuf *dup; 1895 1896 dup = m_dup(m, 0, M_COPYALL, M_NOWAIT); 1897 if (dup == NULL) 1898 return NULL; 1899 1900 if (bp->bif_mbuf_tail != NULL) { 1901 bp->bif_mbuf_tail->m_nextpkt = dup; 1902 } else { 1903 bp->bif_mbuf_head = dup; 1904 } 1905 bp->bif_mbuf_tail = dup; 1906#ifdef BPF_MTAP_SOFTINT_DEBUG 1907 log(LOG_DEBUG, "%s: enqueued mbuf=%p to %s\n", 1908 __func__, dup, bp->bif_ifp->if_xname); 1909#endif 1910 1911 return dup; 1912} 1913 1914static struct mbuf * 1915bpf_mbuf_dequeue(struct bpf_if *bp) 1916{ 1917 struct mbuf *m; 1918 int s; 1919 1920 /* XXX NOMPSAFE: assumed running on one CPU */ 1921 s = splnet(); 1922 m = bp->bif_mbuf_head; 1923 if (m != NULL) { 1924 bp->bif_mbuf_head = m->m_nextpkt; 1925 m->m_nextpkt = NULL; 1926 1927 if (bp->bif_mbuf_head == NULL) 1928 bp->bif_mbuf_tail = NULL; 1929#ifdef BPF_MTAP_SOFTINT_DEBUG 1930 log(LOG_DEBUG, "%s: dequeued mbuf=%p from %s\n", 1931 __func__, m, bp->bif_ifp->if_xname); 1932#endif 1933 } 1934 splx(s); 1935 1936 return m; 1937} 1938 1939static void 1940bpf_mtap_si(void *arg) 1941{ 1942 struct bpf_if *bp = arg; 1943 struct mbuf *m; 1944 1945 while ((m = bpf_mbuf_dequeue(bp)) != NULL) { 1946#ifdef BPF_MTAP_SOFTINT_DEBUG 1947 log(LOG_DEBUG, "%s: tapping mbuf=%p on %s\n", 1948 __func__, m, bp->bif_ifp->if_xname); 1949#endif 1950 bpf_ops->bpf_mtap(bp, m, BPF_D_IN); 1951 m_freem(m); 1952 } 1953} 1954 1955static void 1956_bpf_mtap_softint(struct ifnet *ifp, struct mbuf *m) 1957{ 1958 struct bpf_if *bp = ifp->if_bpf; 1959 struct mbuf *dup; 1960 1961 KASSERT(cpu_intr_p()); 1962 1963 /* To avoid extra invocations of the softint */ 1964 if (BPFIF_DLIST_READER_EMPTY(bp)) 1965 return; 1966 KASSERT(bp->bif_si != NULL); 1967 1968 dup = bpf_mbuf_enqueue(bp, m); 1969 if (dup != NULL) 1970 softint_schedule(bp->bif_si); 1971} 1972 1973static int 1974bpf_hdrlen(struct bpf_d *d) 1975{ 1976 int hdrlen = d->bd_bif->bif_hdrlen; 1977 /* 1978 * Compute the length of the bpf header. This is not necessarily 1979 * equal to SIZEOF_BPF_HDR because we want to insert spacing such 1980 * that the network layer header begins on a longword boundary (for 1981 * performance reasons and to alleviate alignment restrictions). 1982 */ 1983#ifdef _LP64 1984 if (d->bd_compat32) 1985 return (BPF_WORDALIGN32(hdrlen + SIZEOF_BPF_HDR32) - hdrlen); 1986 else 1987#endif 1988 return (BPF_WORDALIGN(hdrlen + SIZEOF_BPF_HDR) - hdrlen); 1989} 1990 1991/* 1992 * Move the packet data from interface memory (pkt) into the 1993 * store buffer. Call the wakeup functions if it's time to wake up 1994 * a listener (buffer full), "cpfn" is the routine called to do the 1995 * actual data transfer. memcpy is passed in to copy contiguous chunks, 1996 * while bpf_mcpy is passed in to copy mbuf chains. In the latter case, 1997 * pkt is really an mbuf. 1998 */ 1999static void 2000catchpacket(struct bpf_d *d, u_char *pkt, u_int pktlen, u_int snaplen, 2001 void *(*cpfn)(void *, const void *, size_t), struct timespec *ts) 2002{ 2003 char *h; 2004 int totlen, curlen, caplen; 2005 int hdrlen = bpf_hdrlen(d); 2006 int do_wakeup = 0; 2007 2008 atomic_inc_ulong(&d->bd_ccount); 2009 BPF_STATINC(capt); 2010 /* 2011 * Figure out how many bytes to move. If the packet is 2012 * greater or equal to the snapshot length, transfer that 2013 * much. Otherwise, transfer the whole packet (unless 2014 * we hit the buffer size limit). 2015 */ 2016 totlen = hdrlen + uimin(snaplen, pktlen); 2017 if (totlen > d->bd_bufsize) 2018 totlen = d->bd_bufsize; 2019 /* 2020 * If we adjusted totlen to fit the bufsize, it could be that 2021 * totlen is smaller than hdrlen because of the link layer header. 2022 */ 2023 caplen = totlen - hdrlen; 2024 if (caplen < 0) 2025 caplen = 0; 2026 2027 mutex_enter(d->bd_buf_mtx); 2028 /* 2029 * Round up the end of the previous packet to the next longword. 2030 */ 2031#ifdef _LP64 2032 if (d->bd_compat32) 2033 curlen = BPF_WORDALIGN32(d->bd_slen); 2034 else 2035#endif 2036 curlen = BPF_WORDALIGN(d->bd_slen); 2037 if (curlen + totlen > d->bd_bufsize) { 2038 /* 2039 * This packet will overflow the storage buffer. 2040 * Rotate the buffers if we can, then wakeup any 2041 * pending reads. 2042 */ 2043 if (d->bd_fbuf == NULL) { 2044 mutex_exit(d->bd_buf_mtx); 2045 /* 2046 * We haven't completed the previous read yet, 2047 * so drop the packet. 2048 */ 2049 atomic_inc_ulong(&d->bd_dcount); 2050 BPF_STATINC(drop); 2051 return; 2052 } 2053 ROTATE_BUFFERS(d); 2054 do_wakeup = 1; 2055 curlen = 0; 2056 } else if (d->bd_immediate || d->bd_state == BPF_TIMED_OUT) { 2057 /* 2058 * Immediate mode is set, or the read timeout has 2059 * already expired during a select call. A packet 2060 * arrived, so the reader should be woken up. 2061 */ 2062 do_wakeup = 1; 2063 } 2064 2065 /* 2066 * Append the bpf header. 2067 */ 2068 h = (char *)d->bd_sbuf + curlen; 2069#ifdef _LP64 2070 if (d->bd_compat32) { 2071 struct bpf_hdr32 *hp32; 2072 2073 hp32 = (struct bpf_hdr32 *)h; 2074 hp32->bh_tstamp.tv_sec = ts->tv_sec; 2075 hp32->bh_tstamp.tv_usec = ts->tv_nsec / 1000; 2076 hp32->bh_datalen = pktlen; 2077 hp32->bh_hdrlen = hdrlen; 2078 hp32->bh_caplen = caplen; 2079 } else 2080#endif 2081 { 2082 struct bpf_hdr *hp; 2083 2084 hp = (struct bpf_hdr *)h; 2085 hp->bh_tstamp.tv_sec = ts->tv_sec; 2086 hp->bh_tstamp.tv_usec = ts->tv_nsec / 1000; 2087 hp->bh_datalen = pktlen; 2088 hp->bh_hdrlen = hdrlen; 2089 hp->bh_caplen = caplen; 2090 } 2091 2092 /* 2093 * Copy the packet data into the store buffer and update its length. 2094 */ 2095 (*cpfn)(h + hdrlen, pkt, caplen); 2096 d->bd_slen = curlen + totlen; 2097 mutex_exit(d->bd_buf_mtx); 2098 2099 /* 2100 * Call bpf_wakeup after bd_slen has been updated so that kevent(2) 2101 * will cause filt_bpfread() to be called with it adjusted. 2102 */ 2103 if (do_wakeup) 2104 bpf_wakeup(d); 2105} 2106 2107/* 2108 * Initialize all nonzero fields of a descriptor. 2109 */ 2110static int 2111bpf_allocbufs(struct bpf_d *d) 2112{ 2113 2114 d->bd_fbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP); 2115 if (!d->bd_fbuf) 2116 return (ENOBUFS); 2117 d->bd_sbuf = kmem_zalloc(d->bd_bufsize, KM_NOSLEEP); 2118 if (!d->bd_sbuf) { 2119 kmem_free(d->bd_fbuf, d->bd_bufsize); 2120 return (ENOBUFS); 2121 } 2122 d->bd_slen = 0; 2123 d->bd_hlen = 0; 2124 return (0); 2125} 2126 2127static void 2128bpf_free_filter(struct bpf_filter *filter) 2129{ 2130 2131 KASSERT(filter != NULL); 2132 2133 if (filter->bf_insn != NULL) 2134 kmem_free(filter->bf_insn, filter->bf_size); 2135 if (filter->bf_jitcode != NULL) 2136 bpf_jit_freecode(filter->bf_jitcode); 2137 kmem_free(filter, sizeof(*filter)); 2138} 2139 2140/* 2141 * Free buffers currently in use by a descriptor. 2142 * Called on close. 2143 */ 2144static void 2145bpf_freed(struct bpf_d *d) 2146{ 2147 /* 2148 * We don't need to lock out interrupts since this descriptor has 2149 * been detached from its interface and it yet hasn't been marked 2150 * free. 2151 */ 2152 if (d->bd_sbuf != NULL) { 2153 kmem_free(d->bd_sbuf, d->bd_bufsize); 2154 if (d->bd_hbuf != NULL) 2155 kmem_free(d->bd_hbuf, d->bd_bufsize); 2156 if (d->bd_fbuf != NULL) 2157 kmem_free(d->bd_fbuf, d->bd_bufsize); 2158 } 2159 if (d->bd_rfilter != NULL) { 2160 bpf_free_filter(d->bd_rfilter); 2161 d->bd_rfilter = NULL; 2162 } 2163 if (d->bd_wfilter != NULL) { 2164 bpf_free_filter(d->bd_wfilter); 2165 d->bd_wfilter = NULL; 2166 } 2167 d->bd_jitcode = NULL; 2168} 2169 2170/* 2171 * Attach an interface to bpf. dlt is the link layer type; 2172 * hdrlen is the fixed size of the link header for the specified dlt 2173 * (variable length headers not yet supported). 2174 */ 2175static void 2176_bpfattach(struct ifnet *ifp, u_int dlt, u_int hdrlen, struct bpf_if **driverp) 2177{ 2178 struct bpf_if *bp; 2179 2180 bp = kmem_alloc(sizeof(*bp), KM_SLEEP); 2181 2182 mutex_enter(&bpf_mtx); 2183 bp->bif_driverp = driverp; 2184 bp->bif_ifp = ifp; 2185 bp->bif_dlt = dlt; 2186 bp->bif_si = NULL; 2187 BPF_IFLIST_ENTRY_INIT(bp); 2188 PSLIST_INIT(&bp->bif_dlist_head); 2189 psref_target_init(&bp->bif_psref, bpf_psref_class); 2190 SLIST_INIT(&bp->bif_trackers); 2191 2192 BPF_IFLIST_WRITER_INSERT_HEAD(bp); 2193 2194 *bp->bif_driverp = NULL; 2195 2196 bp->bif_hdrlen = hdrlen; 2197 mutex_exit(&bpf_mtx); 2198#if 0 2199 printf("bpf: %s attached with dlt %x\n", ifp->if_xname, dlt); 2200#endif 2201} 2202 2203static void 2204_bpf_mtap_softint_init(struct ifnet *ifp) 2205{ 2206 struct bpf_if *bp; 2207 2208 mutex_enter(&bpf_mtx); 2209 BPF_IFLIST_WRITER_FOREACH(bp) { 2210 if (bp->bif_ifp != ifp) 2211 continue; 2212 2213 bp->bif_mbuf_head = NULL; 2214 bp->bif_mbuf_tail = NULL; 2215 bp->bif_si = softint_establish(SOFTINT_NET, bpf_mtap_si, bp); 2216 if (bp->bif_si == NULL) 2217 panic("%s: softint_establish() failed", __func__); 2218 break; 2219 } 2220 mutex_exit(&bpf_mtx); 2221 2222 if (bp == NULL) 2223 panic("%s: no bpf_if found for %s", __func__, ifp->if_xname); 2224} 2225 2226/* 2227 * Remove an interface from bpf. 2228 */ 2229static void 2230_bpfdetach(struct ifnet *ifp) 2231{ 2232 struct bpf_if *bp; 2233 struct bpf_d *d; 2234 int s; 2235 2236 mutex_enter(&bpf_mtx); 2237 /* Nuke the vnodes for any open instances */ 2238 again_d: 2239 BPF_DLIST_WRITER_FOREACH(d) { 2240 mutex_enter(d->bd_mtx); 2241 if (d->bd_bif != NULL && d->bd_bif->bif_ifp == ifp) { 2242 /* 2243 * Detach the descriptor from an interface now. 2244 * It will be free'ed later by close routine. 2245 */ 2246 bpf_detachd(d); 2247 mutex_exit(d->bd_mtx); 2248 goto again_d; 2249 } 2250 mutex_exit(d->bd_mtx); 2251 } 2252 2253 again: 2254 BPF_IFLIST_WRITER_FOREACH(bp) { 2255 if (bp->bif_ifp == ifp) { 2256 BPF_IFLIST_WRITER_REMOVE(bp); 2257 2258 pserialize_perform(bpf_psz); 2259 psref_target_destroy(&bp->bif_psref, bpf_psref_class); 2260 2261 while (!SLIST_EMPTY(&bp->bif_trackers)) { 2262 struct bpf_event_tracker *t = 2263 SLIST_FIRST(&bp->bif_trackers); 2264 SLIST_REMOVE_HEAD(&bp->bif_trackers, 2265 bet_entries); 2266 kmem_free(t, sizeof(*t)); 2267 } 2268 2269 BPF_IFLIST_ENTRY_DESTROY(bp); 2270 if (bp->bif_si != NULL) { 2271 /* XXX NOMPSAFE: assumed running on one CPU */ 2272 s = splnet(); 2273 while (bp->bif_mbuf_head != NULL) { 2274 struct mbuf *m = bp->bif_mbuf_head; 2275 bp->bif_mbuf_head = m->m_nextpkt; 2276 m_freem(m); 2277 } 2278 splx(s); 2279 softint_disestablish(bp->bif_si); 2280 } 2281 kmem_free(bp, sizeof(*bp)); 2282 goto again; 2283 } 2284 } 2285 mutex_exit(&bpf_mtx); 2286} 2287 2288/* 2289 * Change the data link type of a interface. 2290 */ 2291static void 2292_bpf_change_type(struct ifnet *ifp, u_int dlt, u_int hdrlen) 2293{ 2294 struct bpf_if *bp; 2295 2296 mutex_enter(&bpf_mtx); 2297 BPF_IFLIST_WRITER_FOREACH(bp) { 2298 if (bp->bif_driverp == &ifp->if_bpf) 2299 break; 2300 } 2301 if (bp == NULL) 2302 panic("bpf_change_type"); 2303 2304 bp->bif_dlt = dlt; 2305 2306 bp->bif_hdrlen = hdrlen; 2307 mutex_exit(&bpf_mtx); 2308} 2309 2310/* 2311 * Get a list of available data link type of the interface. 2312 */ 2313static int 2314bpf_getdltlist(struct bpf_d *d, struct bpf_dltlist *bfl) 2315{ 2316 int n, error; 2317 struct ifnet *ifp; 2318 struct bpf_if *bp; 2319 int s, bound; 2320 2321 KASSERT(mutex_owned(d->bd_mtx)); 2322 2323 ifp = d->bd_bif->bif_ifp; 2324 n = 0; 2325 error = 0; 2326 2327 bound = curlwp_bind(); 2328 s = pserialize_read_enter(); 2329 BPF_IFLIST_READER_FOREACH(bp) { 2330 if (bp->bif_ifp != ifp) 2331 continue; 2332 if (bfl->bfl_list != NULL) { 2333 struct psref psref; 2334 2335 if (n >= bfl->bfl_len) { 2336 pserialize_read_exit(s); 2337 return ENOMEM; 2338 } 2339 2340 bpf_if_acquire(bp, &psref); 2341 pserialize_read_exit(s); 2342 2343 error = copyout(&bp->bif_dlt, 2344 bfl->bfl_list + n, sizeof(u_int)); 2345 2346 s = pserialize_read_enter(); 2347 bpf_if_release(bp, &psref); 2348 } 2349 n++; 2350 } 2351 pserialize_read_exit(s); 2352 curlwp_bindx(bound); 2353 2354 bfl->bfl_len = n; 2355 return error; 2356} 2357 2358/* 2359 * Set the data link type of a BPF instance. 2360 */ 2361static int 2362bpf_setdlt(struct bpf_d *d, u_int dlt) 2363{ 2364 int error, opromisc; 2365 struct ifnet *ifp; 2366 struct bpf_if *bp; 2367 2368 KASSERT(mutex_owned(&bpf_mtx)); 2369 KASSERT(mutex_owned(d->bd_mtx)); 2370 2371 if (d->bd_bif->bif_dlt == dlt) 2372 return 0; 2373 ifp = d->bd_bif->bif_ifp; 2374 BPF_IFLIST_WRITER_FOREACH(bp) { 2375 if (bp->bif_ifp == ifp && bp->bif_dlt == dlt) 2376 break; 2377 } 2378 if (bp == NULL) 2379 return EINVAL; 2380 opromisc = d->bd_promisc; 2381 bpf_detachd(d); 2382 BPFIF_DLIST_ENTRY_INIT(d); 2383 bpf_attachd(d, bp); 2384 reset_d(d); 2385 if (opromisc) { 2386 KERNEL_LOCK_UNLESS_NET_MPSAFE(); 2387 error = ifpromisc(bp->bif_ifp, 1); 2388 KERNEL_UNLOCK_UNLESS_NET_MPSAFE(); 2389 if (error) 2390 printf("%s: bpf_setdlt: ifpromisc failed (%d)\n", 2391 bp->bif_ifp->if_xname, error); 2392 else 2393 d->bd_promisc = 1; 2394 } 2395 return 0; 2396} 2397 2398static int 2399sysctl_net_bpf_maxbufsize(SYSCTLFN_ARGS) 2400{ 2401 int newsize, error; 2402 struct sysctlnode node; 2403 2404 node = *rnode; 2405 node.sysctl_data = &newsize; 2406 newsize = bpf_maxbufsize; 2407 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2408 if (error || newp == NULL) 2409 return (error); 2410 2411 if (newsize < BPF_MINBUFSIZE || newsize > BPF_MAXBUFSIZE) 2412 return (EINVAL); 2413 2414 bpf_maxbufsize = newsize; 2415 2416 return (0); 2417} 2418 2419#if defined(MODULAR) || defined(BPFJIT) 2420static int 2421sysctl_net_bpf_jit(SYSCTLFN_ARGS) 2422{ 2423 bool newval; 2424 int error; 2425 struct sysctlnode node; 2426 2427 node = *rnode; 2428 node.sysctl_data = &newval; 2429 newval = bpf_jit; 2430 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2431 if (error != 0 || newp == NULL) 2432 return error; 2433 2434 bpf_jit = newval; 2435 if (newval && bpfjit_module_ops.bj_generate_code == NULL) { 2436 printf("JIT compilation is postponed " 2437 "until after bpfjit module is loaded\n"); 2438 } 2439 2440 return 0; 2441} 2442#endif 2443 2444static int 2445sysctl_net_bpf_peers(SYSCTLFN_ARGS) 2446{ 2447 int error, elem_count; 2448 struct bpf_d *dp; 2449 struct bpf_d_ext dpe; 2450 size_t len, needed, elem_size, out_size; 2451 char *sp; 2452 2453 if (namelen == 1 && name[0] == CTL_QUERY) 2454 return (sysctl_query(SYSCTLFN_CALL(rnode))); 2455 2456 if (namelen != 2) 2457 return (EINVAL); 2458 2459 /* BPF peers is privileged information. */ 2460 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE, 2461 KAUTH_REQ_NETWORK_INTERFACE_GETPRIV, NULL, NULL, NULL); 2462 if (error) 2463 return (EPERM); 2464 2465 len = (oldp != NULL) ? *oldlenp : 0; 2466 sp = oldp; 2467 elem_size = name[0]; 2468 elem_count = name[1]; 2469 out_size = MIN(sizeof(dpe), elem_size); 2470 needed = 0; 2471 2472 if (elem_size < 1 || elem_count < 0) 2473 return (EINVAL); 2474 2475 mutex_enter(&bpf_mtx); 2476 BPF_DLIST_WRITER_FOREACH(dp) { 2477 if (len >= elem_size && elem_count > 0) { 2478#define BPF_EXT(field) dpe.bde_ ## field = dp->bd_ ## field 2479 BPF_EXT(bufsize); 2480 BPF_EXT(promisc); 2481 BPF_EXT(state); 2482 BPF_EXT(immediate); 2483 BPF_EXT(hdrcmplt); 2484 BPF_EXT(direction); 2485 BPF_EXT(pid); 2486 BPF_EXT(rcount); 2487 BPF_EXT(dcount); 2488 BPF_EXT(ccount); 2489#undef BPF_EXT 2490 mutex_enter(dp->bd_mtx); 2491 if (dp->bd_bif) 2492 (void)strlcpy(dpe.bde_ifname, 2493 dp->bd_bif->bif_ifp->if_xname, 2494 IFNAMSIZ - 1); 2495 else 2496 dpe.bde_ifname[0] = '\0'; 2497 dpe.bde_locked = dp->bd_locked; 2498 mutex_exit(dp->bd_mtx); 2499 2500 error = copyout(&dpe, sp, out_size); 2501 if (error) 2502 break; 2503 sp += elem_size; 2504 len -= elem_size; 2505 } 2506 needed += elem_size; 2507 if (elem_count > 0 && elem_count != INT_MAX) 2508 elem_count--; 2509 } 2510 mutex_exit(&bpf_mtx); 2511 2512 *oldlenp = needed; 2513 2514 return (error); 2515} 2516 2517static void 2518bpf_stats(void *p, void *arg, struct cpu_info *ci __unused) 2519{ 2520 struct bpf_stat *const stats = p; 2521 struct bpf_stat *sum = arg; 2522 2523 int s = splnet(); 2524 2525 sum->bs_recv += stats->bs_recv; 2526 sum->bs_drop += stats->bs_drop; 2527 sum->bs_capt += stats->bs_capt; 2528 2529 splx(s); 2530} 2531 2532static int 2533bpf_sysctl_gstats_handler(SYSCTLFN_ARGS) 2534{ 2535 struct sysctlnode node; 2536 int error; 2537 struct bpf_stat sum; 2538 2539 memset(&sum, 0, sizeof(sum)); 2540 node = *rnode; 2541 2542 percpu_foreach_xcall(bpf_gstats_percpu, XC_HIGHPRI_IPL(IPL_SOFTNET), 2543 bpf_stats, &sum); 2544 2545 node.sysctl_data = ∑ 2546 node.sysctl_size = sizeof(sum); 2547 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 2548 if (error != 0 || newp == NULL) 2549 return error; 2550 2551 return 0; 2552} 2553 2554SYSCTL_SETUP(sysctl_net_bpf_setup, "bpf sysctls") 2555{ 2556 const struct sysctlnode *node; 2557 2558 node = NULL; 2559 sysctl_createv(clog, 0, NULL, &node, 2560 CTLFLAG_PERMANENT, 2561 CTLTYPE_NODE, "bpf", 2562 SYSCTL_DESCR("BPF options"), 2563 NULL, 0, NULL, 0, 2564 CTL_NET, CTL_CREATE, CTL_EOL); 2565 if (node != NULL) { 2566#if defined(MODULAR) || defined(BPFJIT) 2567 sysctl_createv(clog, 0, NULL, NULL, 2568 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2569 CTLTYPE_BOOL, "jit", 2570 SYSCTL_DESCR("Toggle Just-In-Time compilation"), 2571 sysctl_net_bpf_jit, 0, &bpf_jit, 0, 2572 CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL); 2573#endif 2574 sysctl_createv(clog, 0, NULL, NULL, 2575 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2576 CTLTYPE_INT, "maxbufsize", 2577 SYSCTL_DESCR("Maximum size for data capture buffer"), 2578 sysctl_net_bpf_maxbufsize, 0, &bpf_maxbufsize, 0, 2579 CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL); 2580 sysctl_createv(clog, 0, NULL, NULL, 2581 CTLFLAG_PERMANENT, 2582 CTLTYPE_STRUCT, "stats", 2583 SYSCTL_DESCR("BPF stats"), 2584 bpf_sysctl_gstats_handler, 0, NULL, 0, 2585 CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL); 2586 sysctl_createv(clog, 0, NULL, NULL, 2587 CTLFLAG_PERMANENT, 2588 CTLTYPE_STRUCT, "peers", 2589 SYSCTL_DESCR("BPF peers"), 2590 sysctl_net_bpf_peers, 0, NULL, 0, 2591 CTL_NET, node->sysctl_num, CTL_CREATE, CTL_EOL); 2592 } 2593 2594} 2595 2596static int 2597_bpf_register_track_event(struct bpf_if **driverp, 2598 void (*_fun)(struct bpf_if *, struct ifnet *, int, int)) 2599{ 2600 struct bpf_if *bp; 2601 struct bpf_event_tracker *t; 2602 int ret = ENOENT; 2603 2604 t = kmem_zalloc(sizeof(*t), KM_SLEEP); 2605 if (!t) 2606 return ENOMEM; 2607 t->bet_notify = _fun; 2608 2609 mutex_enter(&bpf_mtx); 2610 BPF_IFLIST_WRITER_FOREACH(bp) { 2611 if (bp->bif_driverp != driverp) 2612 continue; 2613 SLIST_INSERT_HEAD(&bp->bif_trackers, t, bet_entries); 2614 ret = 0; 2615 break; 2616 } 2617 mutex_exit(&bpf_mtx); 2618 2619 return ret; 2620} 2621 2622static int 2623_bpf_deregister_track_event(struct bpf_if **driverp, 2624 void (*_fun)(struct bpf_if *, struct ifnet *, int, int)) 2625{ 2626 struct bpf_if *bp; 2627 struct bpf_event_tracker *t = NULL; 2628 int ret = ENOENT; 2629 2630 mutex_enter(&bpf_mtx); 2631 BPF_IFLIST_WRITER_FOREACH(bp) { 2632 if (bp->bif_driverp != driverp) 2633 continue; 2634 SLIST_FOREACH(t, &bp->bif_trackers, bet_entries) { 2635 if (t->bet_notify == _fun) { 2636 ret = 0; 2637 break; 2638 } 2639 } 2640 if (ret == 0) 2641 break; 2642 } 2643 if (ret == 0 && t && t->bet_notify == _fun) { 2644 SLIST_REMOVE(&bp->bif_trackers, t, bpf_event_tracker, 2645 bet_entries); 2646 } 2647 mutex_exit(&bpf_mtx); 2648 if (ret == 0) 2649 kmem_free(t, sizeof(*t)); 2650 return ret; 2651} 2652 2653struct bpf_ops bpf_ops_kernel = { 2654 .bpf_attach = _bpfattach, 2655 .bpf_detach = _bpfdetach, 2656 .bpf_change_type = _bpf_change_type, 2657 .bpf_register_track_event = _bpf_register_track_event, 2658 .bpf_deregister_track_event = _bpf_deregister_track_event, 2659 2660 .bpf_mtap = _bpf_mtap, 2661 .bpf_mtap2 = _bpf_mtap2, 2662 .bpf_mtap_af = _bpf_mtap_af, 2663 .bpf_mtap_sl_in = _bpf_mtap_sl_in, 2664 .bpf_mtap_sl_out = _bpf_mtap_sl_out, 2665 2666 .bpf_mtap_softint = _bpf_mtap_softint, 2667 .bpf_mtap_softint_init = _bpf_mtap_softint_init, 2668}; 2669 2670MODULE(MODULE_CLASS_DRIVER, bpf, "bpf_filter"); 2671 2672static int 2673bpf_modcmd(modcmd_t cmd, void *arg) 2674{ 2675#ifdef _MODULE 2676 devmajor_t bmajor, cmajor; 2677#endif 2678 int error = 0; 2679 2680 switch (cmd) { 2681 case MODULE_CMD_INIT: 2682 bpf_init(); 2683#ifdef _MODULE 2684 bmajor = cmajor = NODEVMAJOR; 2685 error = devsw_attach("bpf", NULL, &bmajor, 2686 &bpf_cdevsw, &cmajor); 2687 if (error) 2688 break; 2689#endif 2690 2691 bpf_ops_handover_enter(&bpf_ops_kernel); 2692 atomic_swap_ptr(&bpf_ops, &bpf_ops_kernel); 2693 bpf_ops_handover_exit(); 2694 break; 2695 2696 case MODULE_CMD_FINI: 2697 /* 2698 * While there is no reference counting for bpf callers, 2699 * unload could at least in theory be done similarly to 2700 * system call disestablishment. This should even be 2701 * a little simpler: 2702 * 2703 * 1) replace op vector with stubs 2704 * 2) post update to all cpus with xc 2705 * 3) check that nobody is in bpf anymore 2706 * (it's doubtful we'd want something like l_sysent, 2707 * but we could do something like *signed* percpu 2708 * counters. if the sum is 0, we're good). 2709 * 4) if fail, unroll changes 2710 * 2711 * NOTE: change won't be atomic to the outside. some 2712 * packets may be not captured even if unload is 2713 * not successful. I think packet capture not working 2714 * is a perfectly logical consequence of trying to 2715 * disable packet capture. 2716 */ 2717 error = EOPNOTSUPP; 2718 break; 2719 2720 default: 2721 error = ENOTTY; 2722 break; 2723 } 2724 2725 return error; 2726} 2727