1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28 /* $FreeBSD$ */ 29 #include "opt_inet.h" 30 #include "opt_inet6.h" 31 32 #include <sys/param.h> 33 #include <sys/module.h> 34 #include <sys/errno.h> 35 #include <sys/eventhandler.h> 36 #include <sys/jail.h> 37 #include <sys/poll.h> /* POLLIN, POLLOUT */ 38 #include <sys/kernel.h> /* types used in module initialization */ 39 #include <sys/conf.h> /* DEV_MODULE_ORDERED */ 40 #include <sys/endian.h> 41 #include <sys/syscallsubr.h> /* kern_ioctl() */ 42 43 #include <sys/rwlock.h> 44 45 #include <vm/vm.h> /* vtophys */ 46 #include <vm/pmap.h> /* vtophys */ 47 #include <vm/vm_param.h> 48 #include <vm/vm_object.h> 49 #include <vm/vm_page.h> 50 #include <vm/vm_pager.h> 51 #include <vm/uma.h> 52 53 54 #include <sys/malloc.h> 55 #include <sys/socket.h> /* sockaddrs */ 56 #include <sys/selinfo.h> 57 #include <sys/kthread.h> /* kthread_add() */ 58 #include <sys/proc.h> /* PROC_LOCK() */ 59 #include <sys/unistd.h> /* RFNOWAIT */ 60 #include <sys/sched.h> /* sched_bind() */ 61 #include <sys/smp.h> /* mp_maxid */ 62 #include <sys/taskqueue.h> /* taskqueue_enqueue(), taskqueue_create(), ... */ 63 #include <net/if.h> 64 #include <net/if_var.h> 65 #include <net/if_types.h> /* IFT_ETHER */ 66 #include <net/ethernet.h> /* ether_ifdetach */ 67 #include <net/if_dl.h> /* LLADDR */ 68 #include <machine/bus.h> /* bus_dmamap_* */ 69 #include <netinet/in.h> /* in6_cksum_pseudo() */ 70 #include <machine/in_cksum.h> /* in_pseudo(), in_cksum_hdr() */ 71 72 #include <net/netmap.h> 73 #include <dev/netmap/netmap_kern.h> 74 #include <net/netmap_virt.h> 75 #include <dev/netmap/netmap_mem2.h> 76 77 78 /* ======================== FREEBSD-SPECIFIC ROUTINES ================== */ 79 80 static void 81 nm_kqueue_notify(void *opaque, int pending) 82 { 83 struct nm_selinfo *si = opaque; 84 85 /* We use a non-zero hint to distinguish this notification call 86 * from the call done in kqueue_scan(), which uses hint=0. 87 */ 88 KNOTE_UNLOCKED(&si->si.si_note, /*hint=*/0x100); 89 } 90 91 int nm_os_selinfo_init(NM_SELINFO_T *si, const char *name) { 92 int err; 93 94 TASK_INIT(&si->ntfytask, 0, nm_kqueue_notify, si); 95 si->ntfytq = taskqueue_create(name, M_NOWAIT, 96 taskqueue_thread_enqueue, &si->ntfytq); 97 if (si->ntfytq == NULL) 98 return -ENOMEM; 99 err = taskqueue_start_threads(&si->ntfytq, 1, PI_NET, "tq %s", name); 100 if (err) { 101 taskqueue_free(si->ntfytq); 102 si->ntfytq = NULL; 103 return err; 104 } 105 106 snprintf(si->mtxname, sizeof(si->mtxname), "nmkl%s", name); 107 mtx_init(&si->m, si->mtxname, NULL, MTX_DEF); 108 knlist_init_mtx(&si->si.si_note, &si->m); 109 si->kqueue_users = 0; 110 111 return (0); 112 } 113 114 void 115 nm_os_selinfo_uninit(NM_SELINFO_T *si) 116 { 117 if (si->ntfytq == NULL) { 118 return; /* si was not initialized */ 119 } 120 taskqueue_drain(si->ntfytq, &si->ntfytask); 121 taskqueue_free(si->ntfytq); 122 si->ntfytq = NULL; 123 knlist_delete(&si->si.si_note, curthread, /*islocked=*/0); 124 knlist_destroy(&si->si.si_note); 125 /* now we don't need the mutex anymore */ 126 mtx_destroy(&si->m); 127 } 128 129 void * 130 nm_os_malloc(size_t size) 131 { 132 return malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO); 133 } 134 135 void * 136 nm_os_realloc(void *addr, size_t new_size, size_t old_size __unused) 137 { 138 return realloc(addr, new_size, M_DEVBUF, M_NOWAIT | M_ZERO); 139 } 140 141 void 142 nm_os_free(void *addr) 143 { 144 free(addr, M_DEVBUF); 145 } 146 147 void 148 nm_os_ifnet_lock(void) 149 { 150 IFNET_RLOCK(); 151 } 152 153 void 154 nm_os_ifnet_unlock(void) 155 { 156 IFNET_RUNLOCK(); 157 } 158 159 static int netmap_use_count = 0; 160 161 void 162 nm_os_get_module(void) 163 { 164 netmap_use_count++; 165 } 166 167 void 168 nm_os_put_module(void) 169 { 170 netmap_use_count--; 171 } 172 173 static void 174 netmap_ifnet_arrival_handler(void *arg __unused, struct ifnet *ifp) 175 { 176 netmap_undo_zombie(ifp); 177 } 178 179 static void 180 netmap_ifnet_departure_handler(void *arg __unused, struct ifnet *ifp) 181 { 182 netmap_make_zombie(ifp); 183 } 184 185 static eventhandler_tag nm_ifnet_ah_tag; 186 static eventhandler_tag nm_ifnet_dh_tag; 187 188 int 189 nm_os_ifnet_init(void) 190 { 191 nm_ifnet_ah_tag = 192 EVENTHANDLER_REGISTER(ifnet_arrival_event, 193 netmap_ifnet_arrival_handler, 194 NULL, EVENTHANDLER_PRI_ANY); 195 nm_ifnet_dh_tag = 196 EVENTHANDLER_REGISTER(ifnet_departure_event, 197 netmap_ifnet_departure_handler, 198 NULL, EVENTHANDLER_PRI_ANY); 199 return 0; 200 } 201 202 void 203 nm_os_ifnet_fini(void) 204 { 205 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 206 nm_ifnet_ah_tag); 207 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 208 nm_ifnet_dh_tag); 209 } 210 211 unsigned 212 nm_os_ifnet_mtu(struct ifnet *ifp) 213 { 214 return ifp->if_mtu; 215 } 216 217 rawsum_t 218 nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum) 219 { 220 /* TODO XXX please use the FreeBSD implementation for this. */ 221 uint16_t *words = (uint16_t *)data; 222 int nw = len / 2; 223 int i; 224 225 for (i = 0; i < nw; i++) 226 cur_sum += be16toh(words[i]); 227 228 if (len & 1) 229 cur_sum += (data[len-1] << 8); 230 231 return cur_sum; 232 } 233 234 /* Fold a raw checksum: 'cur_sum' is in host byte order, while the 235 * return value is in network byte order. 236 */ 237 uint16_t 238 nm_os_csum_fold(rawsum_t cur_sum) 239 { 240 /* TODO XXX please use the FreeBSD implementation for this. */ 241 while (cur_sum >> 16) 242 cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16); 243 244 return htobe16((~cur_sum) & 0xFFFF); 245 } 246 247 uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph) 248 { 249 #if 0 250 return in_cksum_hdr((void *)iph); 251 #else 252 return nm_os_csum_fold(nm_os_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0)); 253 #endif 254 } 255 256 void 257 nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 258 size_t datalen, uint16_t *check) 259 { 260 #ifdef INET 261 uint16_t pseudolen = datalen + iph->protocol; 262 263 /* Compute and insert the pseudo-header checksum. */ 264 *check = in_pseudo(iph->saddr, iph->daddr, 265 htobe16(pseudolen)); 266 /* Compute the checksum on TCP/UDP header + payload 267 * (includes the pseudo-header). 268 */ 269 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 270 #else 271 static int notsupported = 0; 272 if (!notsupported) { 273 notsupported = 1; 274 nm_prerr("inet4 segmentation not supported"); 275 } 276 #endif 277 } 278 279 void 280 nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 281 size_t datalen, uint16_t *check) 282 { 283 #ifdef INET6 284 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0); 285 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 286 #else 287 static int notsupported = 0; 288 if (!notsupported) { 289 notsupported = 1; 290 nm_prerr("inet6 segmentation not supported"); 291 } 292 #endif 293 } 294 295 /* on FreeBSD we send up one packet at a time */ 296 void * 297 nm_os_send_up(struct ifnet *ifp, struct mbuf *m, struct mbuf *prev) 298 { 299 NA(ifp)->if_input(ifp, m); 300 return NULL; 301 } 302 303 int 304 nm_os_mbuf_has_csum_offld(struct mbuf *m) 305 { 306 return m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP | CSUM_SCTP | 307 CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | 308 CSUM_SCTP_IPV6); 309 } 310 311 int 312 nm_os_mbuf_has_seg_offld(struct mbuf *m) 313 { 314 return m->m_pkthdr.csum_flags & CSUM_TSO; 315 } 316 317 static void 318 freebsd_generic_rx_handler(struct ifnet *ifp, struct mbuf *m) 319 { 320 int stolen; 321 322 if (unlikely(!NM_NA_VALID(ifp))) { 323 nm_prlim(1, "Warning: RX packet intercepted, but no" 324 " emulated adapter"); 325 return; 326 } 327 328 stolen = generic_rx_handler(ifp, m); 329 if (!stolen) { 330 struct netmap_generic_adapter *gna = 331 (struct netmap_generic_adapter *)NA(ifp); 332 gna->save_if_input(ifp, m); 333 } 334 } 335 336 /* 337 * Intercept the rx routine in the standard device driver. 338 * Second argument is non-zero to intercept, 0 to restore 339 */ 340 int 341 nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept) 342 { 343 struct netmap_adapter *na = &gna->up.up; 344 struct ifnet *ifp = na->ifp; 345 int ret = 0; 346 347 nm_os_ifnet_lock(); 348 if (intercept) { 349 if (gna->save_if_input) { 350 nm_prerr("RX on %s already intercepted", na->name); 351 ret = EBUSY; /* already set */ 352 goto out; 353 } 354 355 ifp->if_capenable |= IFCAP_NETMAP; 356 gna->save_if_input = ifp->if_input; 357 ifp->if_input = freebsd_generic_rx_handler; 358 } else { 359 if (!gna->save_if_input) { 360 nm_prerr("Failed to undo RX intercept on %s", 361 na->name); 362 ret = EINVAL; /* not saved */ 363 goto out; 364 } 365 366 ifp->if_capenable &= ~IFCAP_NETMAP; 367 ifp->if_input = gna->save_if_input; 368 gna->save_if_input = NULL; 369 } 370 out: 371 nm_os_ifnet_unlock(); 372 373 return ret; 374 } 375 376 377 /* 378 * Intercept the packet steering routine in the tx path, 379 * so that we can decide which queue is used for an mbuf. 380 * Second argument is non-zero to intercept, 0 to restore. 381 * On freebsd we just intercept if_transmit. 382 */ 383 int 384 nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept) 385 { 386 struct netmap_adapter *na = &gna->up.up; 387 struct ifnet *ifp = netmap_generic_getifp(gna); 388 389 nm_os_ifnet_lock(); 390 if (intercept) { 391 na->if_transmit = ifp->if_transmit; 392 ifp->if_transmit = netmap_transmit; 393 } else { 394 ifp->if_transmit = na->if_transmit; 395 } 396 nm_os_ifnet_unlock(); 397 398 return 0; 399 } 400 401 402 /* 403 * Transmit routine used by generic_netmap_txsync(). Returns 0 on success 404 * and non-zero on error (which may be packet drops or other errors). 405 * addr and len identify the netmap buffer, m is the (preallocated) 406 * mbuf to use for transmissions. 407 * 408 * We should add a reference to the mbuf so the m_freem() at the end 409 * of the transmission does not consume resources. 410 * 411 * On FreeBSD, and on multiqueue cards, we can force the queue using 412 * if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 413 * i = m->m_pkthdr.flowid % adapter->num_queues; 414 * else 415 * i = curcpu % adapter->num_queues; 416 * 417 */ 418 int 419 nm_os_generic_xmit_frame(struct nm_os_gen_arg *a) 420 { 421 int ret; 422 u_int len = a->len; 423 struct ifnet *ifp = a->ifp; 424 struct mbuf *m = a->m; 425 426 /* Link the external storage to 427 * the netmap buffer, so that no copy is necessary. */ 428 m->m_ext.ext_buf = m->m_data = a->addr; 429 m->m_ext.ext_size = len; 430 431 m->m_flags |= M_PKTHDR; 432 m->m_len = m->m_pkthdr.len = len; 433 434 /* mbuf refcnt is not contended, no need to use atomic 435 * (a memory barrier is enough). */ 436 SET_MBUF_REFCNT(m, 2); 437 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 438 m->m_pkthdr.flowid = a->ring_nr; 439 m->m_pkthdr.rcvif = ifp; /* used for tx notification */ 440 CURVNET_SET(ifp->if_vnet); 441 ret = NA(ifp)->if_transmit(ifp, m); 442 CURVNET_RESTORE(); 443 return ret ? -1 : 0; 444 } 445 446 447 struct netmap_adapter * 448 netmap_getna(if_t ifp) 449 { 450 return (NA((struct ifnet *)ifp)); 451 } 452 453 /* 454 * The following two functions are empty until we have a generic 455 * way to extract the info from the ifp 456 */ 457 int 458 nm_os_generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx) 459 { 460 return 0; 461 } 462 463 464 void 465 nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq) 466 { 467 unsigned num_rings = netmap_generic_rings ? netmap_generic_rings : 1; 468 469 *txq = num_rings; 470 *rxq = num_rings; 471 } 472 473 void 474 nm_os_generic_set_features(struct netmap_generic_adapter *gna) 475 { 476 477 gna->rxsg = 1; /* Supported through m_copydata. */ 478 gna->txqdisc = 0; /* Not supported. */ 479 } 480 481 void 482 nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na) 483 { 484 mit->mit_pending = 0; 485 mit->mit_ring_idx = idx; 486 mit->mit_na = na; 487 } 488 489 490 void 491 nm_os_mitigation_start(struct nm_generic_mit *mit) 492 { 493 } 494 495 496 void 497 nm_os_mitigation_restart(struct nm_generic_mit *mit) 498 { 499 } 500 501 502 int 503 nm_os_mitigation_active(struct nm_generic_mit *mit) 504 { 505 506 return 0; 507 } 508 509 510 void 511 nm_os_mitigation_cleanup(struct nm_generic_mit *mit) 512 { 513 } 514 515 static int 516 nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr) 517 { 518 519 return EINVAL; 520 } 521 522 static void 523 nm_vi_start(struct ifnet *ifp) 524 { 525 panic("nm_vi_start() must not be called"); 526 } 527 528 /* 529 * Index manager of persistent virtual interfaces. 530 * It is used to decide the lowest byte of the MAC address. 531 * We use the same algorithm with management of bridge port index. 532 */ 533 #define NM_VI_MAX 255 534 static struct { 535 uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */ 536 uint8_t active; 537 struct mtx lock; 538 } nm_vi_indices; 539 540 void 541 nm_os_vi_init_index(void) 542 { 543 int i; 544 for (i = 0; i < NM_VI_MAX; i++) 545 nm_vi_indices.index[i] = i; 546 nm_vi_indices.active = 0; 547 mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF); 548 } 549 550 /* return -1 if no index available */ 551 static int 552 nm_vi_get_index(void) 553 { 554 int ret; 555 556 mtx_lock(&nm_vi_indices.lock); 557 ret = nm_vi_indices.active == NM_VI_MAX ? -1 : 558 nm_vi_indices.index[nm_vi_indices.active++]; 559 mtx_unlock(&nm_vi_indices.lock); 560 return ret; 561 } 562 563 static void 564 nm_vi_free_index(uint8_t val) 565 { 566 int i, lim; 567 568 mtx_lock(&nm_vi_indices.lock); 569 lim = nm_vi_indices.active; 570 for (i = 0; i < lim; i++) { 571 if (nm_vi_indices.index[i] == val) { 572 /* swap index[lim-1] and j */ 573 int tmp = nm_vi_indices.index[lim-1]; 574 nm_vi_indices.index[lim-1] = val; 575 nm_vi_indices.index[i] = tmp; 576 nm_vi_indices.active--; 577 break; 578 } 579 } 580 if (lim == nm_vi_indices.active) 581 nm_prerr("Index %u not found", val); 582 mtx_unlock(&nm_vi_indices.lock); 583 } 584 #undef NM_VI_MAX 585 586 /* 587 * Implementation of a netmap-capable virtual interface that 588 * registered to the system. 589 * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9. 590 * 591 * Note: Linux sets refcount to 0 on allocation of net_device, 592 * then increments it on registration to the system. 593 * FreeBSD sets refcount to 1 on if_alloc(), and does not 594 * increment this refcount on if_attach(). 595 */ 596 int 597 nm_os_vi_persist(const char *name, struct ifnet **ret) 598 { 599 struct ifnet *ifp; 600 u_short macaddr_hi; 601 uint32_t macaddr_mid; 602 u_char eaddr[6]; 603 int unit = nm_vi_get_index(); /* just to decide MAC address */ 604 605 if (unit < 0) 606 return EBUSY; 607 /* 608 * We use the same MAC address generation method with tap 609 * except for the highest octet is 00:be instead of 00:bd 610 */ 611 macaddr_hi = htons(0x00be); /* XXX tap + 1 */ 612 macaddr_mid = (uint32_t) ticks; 613 bcopy(&macaddr_hi, eaddr, sizeof(short)); 614 bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t)); 615 eaddr[5] = (uint8_t)unit; 616 617 ifp = if_alloc(IFT_ETHER); 618 if (ifp == NULL) { 619 nm_prerr("if_alloc failed"); 620 return ENOMEM; 621 } 622 if_initname(ifp, name, IF_DUNIT_NONE); 623 ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; 624 ifp->if_init = (void *)nm_vi_dummy; 625 ifp->if_ioctl = nm_vi_dummy; 626 ifp->if_start = nm_vi_start; 627 ifp->if_mtu = ETHERMTU; 628 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 629 ifp->if_capabilities |= IFCAP_LINKSTATE; 630 ifp->if_capenable |= IFCAP_LINKSTATE; 631 632 ether_ifattach(ifp, eaddr); 633 *ret = ifp; 634 return 0; 635 } 636 637 /* unregister from the system and drop the final refcount */ 638 void 639 nm_os_vi_detach(struct ifnet *ifp) 640 { 641 nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]); 642 ether_ifdetach(ifp); 643 if_free(ifp); 644 } 645 646 #ifdef WITH_EXTMEM 647 #include <vm/vm_map.h> 648 #include <vm/vm_extern.h> 649 #include <vm/vm_kern.h> 650 struct nm_os_extmem { 651 vm_object_t obj; 652 vm_offset_t kva; 653 vm_offset_t size; 654 uintptr_t scan; 655 }; 656 657 void 658 nm_os_extmem_delete(struct nm_os_extmem *e) 659 { 660 nm_prinf("freeing %zx bytes", (size_t)e->size); 661 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 662 nm_os_free(e); 663 } 664 665 char * 666 nm_os_extmem_nextpage(struct nm_os_extmem *e) 667 { 668 char *rv = NULL; 669 if (e->scan < e->kva + e->size) { 670 rv = (char *)e->scan; 671 e->scan += PAGE_SIZE; 672 } 673 return rv; 674 } 675 676 int 677 nm_os_extmem_isequal(struct nm_os_extmem *e1, struct nm_os_extmem *e2) 678 { 679 return (e1->obj == e2->obj); 680 } 681 682 int 683 nm_os_extmem_nr_pages(struct nm_os_extmem *e) 684 { 685 return e->size >> PAGE_SHIFT; 686 } 687 688 struct nm_os_extmem * 689 nm_os_extmem_create(unsigned long p, struct nmreq_pools_info *pi, int *perror) 690 { 691 vm_map_t map; 692 vm_map_entry_t entry; 693 vm_object_t obj; 694 vm_prot_t prot; 695 vm_pindex_t index; 696 boolean_t wired; 697 struct nm_os_extmem *e = NULL; 698 int rv, error = 0; 699 700 e = nm_os_malloc(sizeof(*e)); 701 if (e == NULL) { 702 error = ENOMEM; 703 goto out; 704 } 705 706 map = &curthread->td_proc->p_vmspace->vm_map; 707 rv = vm_map_lookup(&map, p, VM_PROT_RW, &entry, 708 &obj, &index, &prot, &wired); 709 if (rv != KERN_SUCCESS) { 710 nm_prerr("address %lx not found", p); 711 error = vm_mmap_to_errno(rv); 712 goto out_free; 713 } 714 vm_object_reference(obj); 715 716 /* check that we are given the whole vm_object ? */ 717 vm_map_lookup_done(map, entry); 718 719 e->obj = obj; 720 /* Wire the memory and add the vm_object to the kernel map, 721 * to make sure that it is not freed even if all the processes 722 * that are mmap()ing should munmap() it. 723 */ 724 e->kva = vm_map_min(kernel_map); 725 e->size = obj->size << PAGE_SHIFT; 726 rv = vm_map_find(kernel_map, obj, 0, &e->kva, e->size, 0, 727 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 728 VM_PROT_READ | VM_PROT_WRITE, 0); 729 if (rv != KERN_SUCCESS) { 730 nm_prerr("vm_map_find(%zx) failed", (size_t)e->size); 731 error = vm_mmap_to_errno(rv); 732 goto out_rel; 733 } 734 rv = vm_map_wire(kernel_map, e->kva, e->kva + e->size, 735 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 736 if (rv != KERN_SUCCESS) { 737 nm_prerr("vm_map_wire failed"); 738 error = vm_mmap_to_errno(rv); 739 goto out_rem; 740 } 741 742 e->scan = e->kva; 743 744 return e; 745 746 out_rem: 747 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 748 out_rel: 749 vm_object_deallocate(e->obj); 750 e->obj = NULL; 751 out_free: 752 nm_os_free(e); 753 out: 754 if (perror) 755 *perror = error; 756 return NULL; 757 } 758 #endif /* WITH_EXTMEM */ 759 760 /* ================== PTNETMAP GUEST SUPPORT ==================== */ 761 762 #ifdef WITH_PTNETMAP 763 #include <sys/bus.h> 764 #include <sys/rman.h> 765 #include <machine/bus.h> /* bus_dmamap_* */ 766 #include <machine/resource.h> 767 #include <dev/pci/pcivar.h> 768 #include <dev/pci/pcireg.h> 769 /* 770 * ptnetmap memory device (memdev) for freebsd guest, 771 * ssed to expose host netmap memory to the guest through a PCI BAR. 772 */ 773 774 /* 775 * ptnetmap memdev private data structure 776 */ 777 struct ptnetmap_memdev { 778 device_t dev; 779 struct resource *pci_io; 780 struct resource *pci_mem; 781 struct netmap_mem_d *nm_mem; 782 }; 783 784 static int ptn_memdev_probe(device_t); 785 static int ptn_memdev_attach(device_t); 786 static int ptn_memdev_detach(device_t); 787 static int ptn_memdev_shutdown(device_t); 788 789 static device_method_t ptn_memdev_methods[] = { 790 DEVMETHOD(device_probe, ptn_memdev_probe), 791 DEVMETHOD(device_attach, ptn_memdev_attach), 792 DEVMETHOD(device_detach, ptn_memdev_detach), 793 DEVMETHOD(device_shutdown, ptn_memdev_shutdown), 794 DEVMETHOD_END 795 }; 796 797 static driver_t ptn_memdev_driver = { 798 PTNETMAP_MEMDEV_NAME, 799 ptn_memdev_methods, 800 sizeof(struct ptnetmap_memdev), 801 }; 802 803 /* We use (SI_ORDER_MIDDLE+1) here, see DEV_MODULE_ORDERED() invocation 804 * below. */ 805 DRIVER_MODULE_ORDERED(ptn_memdev, pci, ptn_memdev_driver, NULL, NULL, 806 SI_ORDER_MIDDLE + 1); 807 808 /* 809 * Map host netmap memory through PCI-BAR in the guest OS, 810 * returning physical (nm_paddr) and virtual (nm_addr) addresses 811 * of the netmap memory mapped in the guest. 812 */ 813 int 814 nm_os_pt_memdev_iomap(struct ptnetmap_memdev *ptn_dev, vm_paddr_t *nm_paddr, 815 void **nm_addr, uint64_t *mem_size) 816 { 817 int rid; 818 819 nm_prinf("ptn_memdev_driver iomap"); 820 821 rid = PCIR_BAR(PTNETMAP_MEM_PCI_BAR); 822 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_HI); 823 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_LO) | 824 (*mem_size << 32); 825 826 /* map memory allocator */ 827 ptn_dev->pci_mem = bus_alloc_resource(ptn_dev->dev, SYS_RES_MEMORY, 828 &rid, 0, ~0, *mem_size, RF_ACTIVE); 829 if (ptn_dev->pci_mem == NULL) { 830 *nm_paddr = 0; 831 *nm_addr = NULL; 832 return ENOMEM; 833 } 834 835 *nm_paddr = rman_get_start(ptn_dev->pci_mem); 836 *nm_addr = rman_get_virtual(ptn_dev->pci_mem); 837 838 nm_prinf("=== BAR %d start %lx len %lx mem_size %lx ===", 839 PTNETMAP_MEM_PCI_BAR, 840 (unsigned long)(*nm_paddr), 841 (unsigned long)rman_get_size(ptn_dev->pci_mem), 842 (unsigned long)*mem_size); 843 return (0); 844 } 845 846 uint32_t 847 nm_os_pt_memdev_ioread(struct ptnetmap_memdev *ptn_dev, unsigned int reg) 848 { 849 return bus_read_4(ptn_dev->pci_io, reg); 850 } 851 852 /* Unmap host netmap memory. */ 853 void 854 nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *ptn_dev) 855 { 856 nm_prinf("ptn_memdev_driver iounmap"); 857 858 if (ptn_dev->pci_mem) { 859 bus_release_resource(ptn_dev->dev, SYS_RES_MEMORY, 860 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 861 ptn_dev->pci_mem = NULL; 862 } 863 } 864 865 /* Device identification routine, return BUS_PROBE_DEFAULT on success, 866 * positive on failure */ 867 static int 868 ptn_memdev_probe(device_t dev) 869 { 870 char desc[256]; 871 872 if (pci_get_vendor(dev) != PTNETMAP_PCI_VENDOR_ID) 873 return (ENXIO); 874 if (pci_get_device(dev) != PTNETMAP_PCI_DEVICE_ID) 875 return (ENXIO); 876 877 snprintf(desc, sizeof(desc), "%s PCI adapter", 878 PTNETMAP_MEMDEV_NAME); 879 device_set_desc_copy(dev, desc); 880 881 return (BUS_PROBE_DEFAULT); 882 } 883 884 /* Device initialization routine. */ 885 static int 886 ptn_memdev_attach(device_t dev) 887 { 888 struct ptnetmap_memdev *ptn_dev; 889 int rid; 890 uint16_t mem_id; 891 892 ptn_dev = device_get_softc(dev); 893 ptn_dev->dev = dev; 894 895 pci_enable_busmaster(dev); 896 897 rid = PCIR_BAR(PTNETMAP_IO_PCI_BAR); 898 ptn_dev->pci_io = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid, 899 RF_ACTIVE); 900 if (ptn_dev->pci_io == NULL) { 901 device_printf(dev, "cannot map I/O space\n"); 902 return (ENXIO); 903 } 904 905 mem_id = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMID); 906 907 /* create guest allocator */ 908 ptn_dev->nm_mem = netmap_mem_pt_guest_attach(ptn_dev, mem_id); 909 if (ptn_dev->nm_mem == NULL) { 910 ptn_memdev_detach(dev); 911 return (ENOMEM); 912 } 913 netmap_mem_get(ptn_dev->nm_mem); 914 915 nm_prinf("ptnetmap memdev attached, host memid: %u", mem_id); 916 917 return (0); 918 } 919 920 /* Device removal routine. */ 921 static int 922 ptn_memdev_detach(device_t dev) 923 { 924 struct ptnetmap_memdev *ptn_dev; 925 926 ptn_dev = device_get_softc(dev); 927 928 if (ptn_dev->nm_mem) { 929 nm_prinf("ptnetmap memdev detached, host memid %u", 930 netmap_mem_get_id(ptn_dev->nm_mem)); 931 netmap_mem_put(ptn_dev->nm_mem); 932 ptn_dev->nm_mem = NULL; 933 } 934 if (ptn_dev->pci_mem) { 935 bus_release_resource(dev, SYS_RES_MEMORY, 936 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 937 ptn_dev->pci_mem = NULL; 938 } 939 if (ptn_dev->pci_io) { 940 bus_release_resource(dev, SYS_RES_IOPORT, 941 PCIR_BAR(PTNETMAP_IO_PCI_BAR), ptn_dev->pci_io); 942 ptn_dev->pci_io = NULL; 943 } 944 945 return (0); 946 } 947 948 static int 949 ptn_memdev_shutdown(device_t dev) 950 { 951 return bus_generic_shutdown(dev); 952 } 953 954 #endif /* WITH_PTNETMAP */ 955 956 /* 957 * In order to track whether pages are still mapped, we hook into 958 * the standard cdev_pager and intercept the constructor and 959 * destructor. 960 */ 961 962 struct netmap_vm_handle_t { 963 struct cdev *dev; 964 struct netmap_priv_d *priv; 965 }; 966 967 968 static int 969 netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 970 vm_ooffset_t foff, struct ucred *cred, u_short *color) 971 { 972 struct netmap_vm_handle_t *vmh = handle; 973 974 if (netmap_verbose) 975 nm_prinf("handle %p size %jd prot %d foff %jd", 976 handle, (intmax_t)size, prot, (intmax_t)foff); 977 if (color) 978 *color = 0; 979 dev_ref(vmh->dev); 980 return 0; 981 } 982 983 984 static void 985 netmap_dev_pager_dtor(void *handle) 986 { 987 struct netmap_vm_handle_t *vmh = handle; 988 struct cdev *dev = vmh->dev; 989 struct netmap_priv_d *priv = vmh->priv; 990 991 if (netmap_verbose) 992 nm_prinf("handle %p", handle); 993 netmap_dtor(priv); 994 free(vmh, M_DEVBUF); 995 dev_rel(dev); 996 } 997 998 999 static int 1000 netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, 1001 int prot, vm_page_t *mres) 1002 { 1003 struct netmap_vm_handle_t *vmh = object->handle; 1004 struct netmap_priv_d *priv = vmh->priv; 1005 struct netmap_adapter *na = priv->np_na; 1006 vm_paddr_t paddr; 1007 vm_page_t page; 1008 vm_memattr_t memattr; 1009 1010 nm_prdis("object %p offset %jd prot %d mres %p", 1011 object, (intmax_t)offset, prot, mres); 1012 memattr = object->memattr; 1013 paddr = netmap_mem_ofstophys(na->nm_mem, offset); 1014 if (paddr == 0) 1015 return VM_PAGER_FAIL; 1016 1017 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 1018 /* 1019 * If the passed in result page is a fake page, update it with 1020 * the new physical address. 1021 */ 1022 page = *mres; 1023 vm_page_updatefake(page, paddr, memattr); 1024 } else { 1025 /* 1026 * Replace the passed in reqpage page with our own fake page and 1027 * free up the all of the original pages. 1028 */ 1029 #ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */ 1030 #define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK 1031 #define VM_OBJECT_WLOCK VM_OBJECT_LOCK 1032 #endif /* VM_OBJECT_WUNLOCK */ 1033 1034 VM_OBJECT_WUNLOCK(object); 1035 page = vm_page_getfake(paddr, memattr); 1036 VM_OBJECT_WLOCK(object); 1037 vm_page_replace(page, object, (*mres)->pindex, *mres); 1038 *mres = page; 1039 } 1040 page->valid = VM_PAGE_BITS_ALL; 1041 return (VM_PAGER_OK); 1042 } 1043 1044 1045 static struct cdev_pager_ops netmap_cdev_pager_ops = { 1046 .cdev_pg_ctor = netmap_dev_pager_ctor, 1047 .cdev_pg_dtor = netmap_dev_pager_dtor, 1048 .cdev_pg_fault = netmap_dev_pager_fault, 1049 }; 1050 1051 1052 static int 1053 netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff, 1054 vm_size_t objsize, vm_object_t *objp, int prot) 1055 { 1056 int error; 1057 struct netmap_vm_handle_t *vmh; 1058 struct netmap_priv_d *priv; 1059 vm_object_t obj; 1060 1061 if (netmap_verbose) 1062 nm_prinf("cdev %p foff %jd size %jd objp %p prot %d", cdev, 1063 (intmax_t )*foff, (intmax_t )objsize, objp, prot); 1064 1065 vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF, 1066 M_NOWAIT | M_ZERO); 1067 if (vmh == NULL) 1068 return ENOMEM; 1069 vmh->dev = cdev; 1070 1071 NMG_LOCK(); 1072 error = devfs_get_cdevpriv((void**)&priv); 1073 if (error) 1074 goto err_unlock; 1075 if (priv->np_nifp == NULL) { 1076 error = EINVAL; 1077 goto err_unlock; 1078 } 1079 vmh->priv = priv; 1080 priv->np_refs++; 1081 NMG_UNLOCK(); 1082 1083 obj = cdev_pager_allocate(vmh, OBJT_DEVICE, 1084 &netmap_cdev_pager_ops, objsize, prot, 1085 *foff, NULL); 1086 if (obj == NULL) { 1087 nm_prerr("cdev_pager_allocate failed"); 1088 error = EINVAL; 1089 goto err_deref; 1090 } 1091 1092 *objp = obj; 1093 return 0; 1094 1095 err_deref: 1096 NMG_LOCK(); 1097 priv->np_refs--; 1098 err_unlock: 1099 NMG_UNLOCK(); 1100 // err: 1101 free(vmh, M_DEVBUF); 1102 return error; 1103 } 1104 1105 /* 1106 * On FreeBSD the close routine is only called on the last close on 1107 * the device (/dev/netmap) so we cannot do anything useful. 1108 * To track close() on individual file descriptors we pass netmap_dtor() to 1109 * devfs_set_cdevpriv() on open(). The FreeBSD kernel will call the destructor 1110 * when the last fd pointing to the device is closed. 1111 * 1112 * Note that FreeBSD does not even munmap() on close() so we also have 1113 * to track mmap() ourselves, and postpone the call to 1114 * netmap_dtor() is called when the process has no open fds and no active 1115 * memory maps on /dev/netmap, as in linux. 1116 */ 1117 static int 1118 netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 1119 { 1120 if (netmap_verbose) 1121 nm_prinf("dev %p fflag 0x%x devtype %d td %p", 1122 dev, fflag, devtype, td); 1123 return 0; 1124 } 1125 1126 1127 static int 1128 netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 1129 { 1130 struct netmap_priv_d *priv; 1131 int error; 1132 1133 (void)dev; 1134 (void)oflags; 1135 (void)devtype; 1136 (void)td; 1137 1138 NMG_LOCK(); 1139 priv = netmap_priv_new(); 1140 if (priv == NULL) { 1141 error = ENOMEM; 1142 goto out; 1143 } 1144 error = devfs_set_cdevpriv(priv, netmap_dtor); 1145 if (error) { 1146 netmap_priv_delete(priv); 1147 } 1148 out: 1149 NMG_UNLOCK(); 1150 return error; 1151 } 1152 1153 /******************** kthread wrapper ****************/ 1154 #include <sys/sysproto.h> 1155 u_int 1156 nm_os_ncpus(void) 1157 { 1158 return mp_maxid + 1; 1159 } 1160 1161 struct nm_kctx_ctx { 1162 /* Userspace thread (kthread creator). */ 1163 struct thread *user_td; 1164 1165 /* worker function and parameter */ 1166 nm_kctx_worker_fn_t worker_fn; 1167 void *worker_private; 1168 1169 struct nm_kctx *nmk; 1170 1171 /* integer to manage multiple worker contexts (e.g., RX or TX on ptnetmap) */ 1172 long type; 1173 }; 1174 1175 struct nm_kctx { 1176 struct thread *worker; 1177 struct mtx worker_lock; 1178 struct nm_kctx_ctx worker_ctx; 1179 int run; /* used to stop kthread */ 1180 int attach_user; /* kthread attached to user_process */ 1181 int affinity; 1182 }; 1183 1184 static void 1185 nm_kctx_worker(void *data) 1186 { 1187 struct nm_kctx *nmk = data; 1188 struct nm_kctx_ctx *ctx = &nmk->worker_ctx; 1189 1190 if (nmk->affinity >= 0) { 1191 thread_lock(curthread); 1192 sched_bind(curthread, nmk->affinity); 1193 thread_unlock(curthread); 1194 } 1195 1196 while (nmk->run) { 1197 /* 1198 * check if the parent process dies 1199 * (when kthread is attached to user process) 1200 */ 1201 if (ctx->user_td) { 1202 PROC_LOCK(curproc); 1203 thread_suspend_check(0); 1204 PROC_UNLOCK(curproc); 1205 } else { 1206 kthread_suspend_check(); 1207 } 1208 1209 /* Continuously execute worker process. */ 1210 ctx->worker_fn(ctx->worker_private); /* worker body */ 1211 } 1212 1213 kthread_exit(); 1214 } 1215 1216 void 1217 nm_os_kctx_worker_setaff(struct nm_kctx *nmk, int affinity) 1218 { 1219 nmk->affinity = affinity; 1220 } 1221 1222 struct nm_kctx * 1223 nm_os_kctx_create(struct nm_kctx_cfg *cfg, void *opaque) 1224 { 1225 struct nm_kctx *nmk = NULL; 1226 1227 nmk = malloc(sizeof(*nmk), M_DEVBUF, M_NOWAIT | M_ZERO); 1228 if (!nmk) 1229 return NULL; 1230 1231 mtx_init(&nmk->worker_lock, "nm_kthread lock", NULL, MTX_DEF); 1232 nmk->worker_ctx.worker_fn = cfg->worker_fn; 1233 nmk->worker_ctx.worker_private = cfg->worker_private; 1234 nmk->worker_ctx.type = cfg->type; 1235 nmk->affinity = -1; 1236 1237 /* attach kthread to user process (ptnetmap) */ 1238 nmk->attach_user = cfg->attach_user; 1239 1240 return nmk; 1241 } 1242 1243 int 1244 nm_os_kctx_worker_start(struct nm_kctx *nmk) 1245 { 1246 struct proc *p = NULL; 1247 int error = 0; 1248 1249 /* Temporarily disable this function as it is currently broken 1250 * and causes kernel crashes. The failure can be triggered by 1251 * the "vale_polling_enable_disable" test in ctrl-api-test.c. */ 1252 return EOPNOTSUPP; 1253 1254 if (nmk->worker) 1255 return EBUSY; 1256 1257 /* check if we want to attach kthread to user process */ 1258 if (nmk->attach_user) { 1259 nmk->worker_ctx.user_td = curthread; 1260 p = curthread->td_proc; 1261 } 1262 1263 /* enable kthread main loop */ 1264 nmk->run = 1; 1265 /* create kthread */ 1266 if((error = kthread_add(nm_kctx_worker, nmk, p, 1267 &nmk->worker, RFNOWAIT /* to be checked */, 0, "nm-kthread-%ld", 1268 nmk->worker_ctx.type))) { 1269 goto err; 1270 } 1271 1272 nm_prinf("nm_kthread started td %p", nmk->worker); 1273 1274 return 0; 1275 err: 1276 nm_prerr("nm_kthread start failed err %d", error); 1277 nmk->worker = NULL; 1278 return error; 1279 } 1280 1281 void 1282 nm_os_kctx_worker_stop(struct nm_kctx *nmk) 1283 { 1284 if (!nmk->worker) 1285 return; 1286 1287 /* tell to kthread to exit from main loop */ 1288 nmk->run = 0; 1289 1290 /* wake up kthread if it sleeps */ 1291 kthread_resume(nmk->worker); 1292 1293 nmk->worker = NULL; 1294 } 1295 1296 void 1297 nm_os_kctx_destroy(struct nm_kctx *nmk) 1298 { 1299 if (!nmk) 1300 return; 1301 1302 if (nmk->worker) 1303 nm_os_kctx_worker_stop(nmk); 1304 1305 free(nmk, M_DEVBUF); 1306 } 1307 1308 /******************** kqueue support ****************/ 1309 1310 /* 1311 * In addition to calling selwakeuppri(), nm_os_selwakeup() also 1312 * needs to call knote() to wake up kqueue listeners. 1313 * This operation is deferred to a taskqueue in order to avoid possible 1314 * lock order reversals; these may happen because knote() grabs a 1315 * private lock associated to the 'si' (see struct selinfo, 1316 * struct nm_selinfo, and nm_os_selinfo_init), and nm_os_selwakeup() 1317 * can be called while holding the lock associated to a different 1318 * 'si'. 1319 * When calling knote() we use a non-zero 'hint' argument to inform 1320 * the netmap_knrw() function that it is being called from 1321 * 'nm_os_selwakeup'; this is necessary because when netmap_knrw() is 1322 * called by the kevent subsystem (i.e. kevent_scan()) we also need to 1323 * call netmap_poll(). 1324 * 1325 * The netmap_kqfilter() function registers one or another f_event 1326 * depending on read or write mode. A pointer to the struct 1327 * 'netmap_priv_d' is stored into kn->kn_hook, so that it can later 1328 * be passed to netmap_poll(). We pass NULL as a third argument to 1329 * netmap_poll(), so that the latter only runs the txsync/rxsync 1330 * (if necessary), and skips the nm_os_selrecord() calls. 1331 */ 1332 1333 1334 void 1335 nm_os_selwakeup(struct nm_selinfo *si) 1336 { 1337 selwakeuppri(&si->si, PI_NET); 1338 if (si->kqueue_users > 0) { 1339 taskqueue_enqueue(si->ntfytq, &si->ntfytask); 1340 } 1341 } 1342 1343 void 1344 nm_os_selrecord(struct thread *td, struct nm_selinfo *si) 1345 { 1346 selrecord(td, &si->si); 1347 } 1348 1349 static void 1350 netmap_knrdetach(struct knote *kn) 1351 { 1352 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1353 struct nm_selinfo *si = priv->np_si[NR_RX]; 1354 1355 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1356 NMG_LOCK(); 1357 KASSERT(si->kqueue_users > 0, ("kqueue_user underflow on %s", 1358 si->mtxname)); 1359 si->kqueue_users--; 1360 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1361 NMG_UNLOCK(); 1362 } 1363 1364 static void 1365 netmap_knwdetach(struct knote *kn) 1366 { 1367 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1368 struct nm_selinfo *si = priv->np_si[NR_TX]; 1369 1370 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1371 NMG_LOCK(); 1372 si->kqueue_users--; 1373 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1374 NMG_UNLOCK(); 1375 } 1376 1377 /* 1378 * Callback triggered by netmap notifications (see netmap_notify()), 1379 * and by the application calling kevent(). In the former case we 1380 * just return 1 (events ready), since we are not able to do better. 1381 * In the latter case we use netmap_poll() to see which events are 1382 * ready. 1383 */ 1384 static int 1385 netmap_knrw(struct knote *kn, long hint, int events) 1386 { 1387 struct netmap_priv_d *priv; 1388 int revents; 1389 1390 if (hint != 0) { 1391 /* Called from netmap_notify(), typically from a 1392 * thread different from the one issuing kevent(). 1393 * Assume we are ready. */ 1394 return 1; 1395 } 1396 1397 /* Called from kevent(). */ 1398 priv = kn->kn_hook; 1399 revents = netmap_poll(priv, events, /*thread=*/NULL); 1400 1401 return (events & revents) ? 1 : 0; 1402 } 1403 1404 static int 1405 netmap_knread(struct knote *kn, long hint) 1406 { 1407 return netmap_knrw(kn, hint, POLLIN); 1408 } 1409 1410 static int 1411 netmap_knwrite(struct knote *kn, long hint) 1412 { 1413 return netmap_knrw(kn, hint, POLLOUT); 1414 } 1415 1416 static struct filterops netmap_rfiltops = { 1417 .f_isfd = 1, 1418 .f_detach = netmap_knrdetach, 1419 .f_event = netmap_knread, 1420 }; 1421 1422 static struct filterops netmap_wfiltops = { 1423 .f_isfd = 1, 1424 .f_detach = netmap_knwdetach, 1425 .f_event = netmap_knwrite, 1426 }; 1427 1428 1429 /* 1430 * This is called when a thread invokes kevent() to record 1431 * a change in the configuration of the kqueue(). 1432 * The 'priv' is the one associated to the open netmap device. 1433 */ 1434 static int 1435 netmap_kqfilter(struct cdev *dev, struct knote *kn) 1436 { 1437 struct netmap_priv_d *priv; 1438 int error; 1439 struct netmap_adapter *na; 1440 struct nm_selinfo *si; 1441 int ev = kn->kn_filter; 1442 1443 if (ev != EVFILT_READ && ev != EVFILT_WRITE) { 1444 nm_prerr("bad filter request %d", ev); 1445 return 1; 1446 } 1447 error = devfs_get_cdevpriv((void**)&priv); 1448 if (error) { 1449 nm_prerr("device not yet setup"); 1450 return 1; 1451 } 1452 na = priv->np_na; 1453 if (na == NULL) { 1454 nm_prerr("no netmap adapter for this file descriptor"); 1455 return 1; 1456 } 1457 /* the si is indicated in the priv */ 1458 si = priv->np_si[(ev == EVFILT_WRITE) ? NR_TX : NR_RX]; 1459 kn->kn_fop = (ev == EVFILT_WRITE) ? 1460 &netmap_wfiltops : &netmap_rfiltops; 1461 kn->kn_hook = priv; 1462 NMG_LOCK(); 1463 si->kqueue_users++; 1464 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1465 NMG_UNLOCK(); 1466 knlist_add(&si->si.si_note, kn, /*islocked=*/0); 1467 1468 return 0; 1469 } 1470 1471 static int 1472 freebsd_netmap_poll(struct cdev *cdevi __unused, int events, struct thread *td) 1473 { 1474 struct netmap_priv_d *priv; 1475 if (devfs_get_cdevpriv((void **)&priv)) { 1476 return POLLERR; 1477 } 1478 return netmap_poll(priv, events, td); 1479 } 1480 1481 static int 1482 freebsd_netmap_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t data, 1483 int ffla __unused, struct thread *td) 1484 { 1485 int error; 1486 struct netmap_priv_d *priv; 1487 1488 CURVNET_SET(TD_TO_VNET(td)); 1489 error = devfs_get_cdevpriv((void **)&priv); 1490 if (error) { 1491 /* XXX ENOENT should be impossible, since the priv 1492 * is now created in the open */ 1493 if (error == ENOENT) 1494 error = ENXIO; 1495 goto out; 1496 } 1497 error = netmap_ioctl(priv, cmd, data, td, /*nr_body_is_user=*/1); 1498 out: 1499 CURVNET_RESTORE(); 1500 1501 return error; 1502 } 1503 1504 void 1505 nm_os_onattach(struct ifnet *ifp) 1506 { 1507 ifp->if_capabilities |= IFCAP_NETMAP; 1508 } 1509 1510 void 1511 nm_os_onenter(struct ifnet *ifp) 1512 { 1513 struct netmap_adapter *na = NA(ifp); 1514 1515 na->if_transmit = ifp->if_transmit; 1516 ifp->if_transmit = netmap_transmit; 1517 ifp->if_capenable |= IFCAP_NETMAP; 1518 } 1519 1520 void 1521 nm_os_onexit(struct ifnet *ifp) 1522 { 1523 struct netmap_adapter *na = NA(ifp); 1524 1525 ifp->if_transmit = na->if_transmit; 1526 ifp->if_capenable &= ~IFCAP_NETMAP; 1527 } 1528 1529 extern struct cdevsw netmap_cdevsw; /* XXX used in netmap.c, should go elsewhere */ 1530 struct cdevsw netmap_cdevsw = { 1531 .d_version = D_VERSION, 1532 .d_name = "netmap", 1533 .d_open = netmap_open, 1534 .d_mmap_single = netmap_mmap_single, 1535 .d_ioctl = freebsd_netmap_ioctl, 1536 .d_poll = freebsd_netmap_poll, 1537 .d_kqfilter = netmap_kqfilter, 1538 .d_close = netmap_close, 1539 }; 1540 /*--- end of kqueue support ----*/ 1541 1542 /* 1543 * Kernel entry point. 1544 * 1545 * Initialize/finalize the module and return. 1546 * 1547 * Return 0 on success, errno on failure. 1548 */ 1549 static int 1550 netmap_loader(__unused struct module *module, int event, __unused void *arg) 1551 { 1552 int error = 0; 1553 1554 switch (event) { 1555 case MOD_LOAD: 1556 error = netmap_init(); 1557 break; 1558 1559 case MOD_UNLOAD: 1560 /* 1561 * if some one is still using netmap, 1562 * then the module can not be unloaded. 1563 */ 1564 if (netmap_use_count) { 1565 nm_prerr("netmap module can not be unloaded - netmap_use_count: %d", 1566 netmap_use_count); 1567 error = EBUSY; 1568 break; 1569 } 1570 netmap_fini(); 1571 break; 1572 1573 default: 1574 error = EOPNOTSUPP; 1575 break; 1576 } 1577 1578 return (error); 1579 } 1580 1581 #ifdef DEV_MODULE_ORDERED 1582 /* 1583 * The netmap module contains three drivers: (i) the netmap character device 1584 * driver; (ii) the ptnetmap memdev PCI device driver, (iii) the ptnet PCI 1585 * device driver. The attach() routines of both (ii) and (iii) need the 1586 * lock of the global allocator, and such lock is initialized in netmap_init(), 1587 * which is part of (i). 1588 * Therefore, we make sure that (i) is loaded before (ii) and (iii), using 1589 * the 'order' parameter of driver declaration macros. For (i), we specify 1590 * SI_ORDER_MIDDLE, while higher orders are used with the DRIVER_MODULE_ORDERED 1591 * macros for (ii) and (iii). 1592 */ 1593 DEV_MODULE_ORDERED(netmap, netmap_loader, NULL, SI_ORDER_MIDDLE); 1594 #else /* !DEV_MODULE_ORDERED */ 1595 DEV_MODULE(netmap, netmap_loader, NULL); 1596 #endif /* DEV_MODULE_ORDERED */ 1597 MODULE_DEPEND(netmap, pci, 1, 1, 1); 1598 MODULE_VERSION(netmap, 1); 1599 /* reduce conditional code */ 1600 // linux API, use for the knlist in FreeBSD 1601 /* use a private mutex for the knlist */ 1602