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 80static void 81nm_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 91int 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 114void 115nm_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 129void * 130nm_os_malloc(size_t size) 131{ 132 return malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO); 133} 134 135void * 136nm_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 141void 142nm_os_free(void *addr) 143{ 144 free(addr, M_DEVBUF); 145} 146 147void 148nm_os_ifnet_lock(void) 149{ 150 IFNET_RLOCK(); 151} 152 153void 154nm_os_ifnet_unlock(void) 155{ 156 IFNET_RUNLOCK(); 157} 158 159static int netmap_use_count = 0; 160 161void 162nm_os_get_module(void) 163{ 164 netmap_use_count++; 165} 166 167void 168nm_os_put_module(void) 169{ 170 netmap_use_count--; 171} 172 173static void 174netmap_ifnet_arrival_handler(void *arg __unused, struct ifnet *ifp) 175{ 176 netmap_undo_zombie(ifp); 177} 178 179static void 180netmap_ifnet_departure_handler(void *arg __unused, struct ifnet *ifp) 181{ 182 netmap_make_zombie(ifp); 183} 184 185static eventhandler_tag nm_ifnet_ah_tag; 186static eventhandler_tag nm_ifnet_dh_tag; 187 188int 189nm_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 202void 203nm_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 211unsigned 212nm_os_ifnet_mtu(struct ifnet *ifp) 213{ 214#if __FreeBSD_version < 1100030 215 return ifp->if_data.ifi_mtu; 216#else /* __FreeBSD_version >= 1100030 */ 217 return ifp->if_mtu; 218#endif 219} 220 221rawsum_t 222nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum) 223{ 224 /* TODO XXX please use the FreeBSD implementation for this. */ 225 uint16_t *words = (uint16_t *)data; 226 int nw = len / 2; 227 int i; 228 229 for (i = 0; i < nw; i++) 230 cur_sum += be16toh(words[i]); 231 232 if (len & 1) 233 cur_sum += (data[len-1] << 8); 234 235 return cur_sum; 236} 237 238/* Fold a raw checksum: 'cur_sum' is in host byte order, while the 239 * return value is in network byte order. 240 */ 241uint16_t 242nm_os_csum_fold(rawsum_t cur_sum) 243{ 244 /* TODO XXX please use the FreeBSD implementation for this. */ 245 while (cur_sum >> 16) 246 cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16); 247 248 return htobe16((~cur_sum) & 0xFFFF); 249} 250 251uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph) 252{ 253#if 0 254 return in_cksum_hdr((void *)iph); 255#else 256 return nm_os_csum_fold(nm_os_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0)); 257#endif 258} 259 260void 261nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 262 size_t datalen, uint16_t *check) 263{ 264#ifdef INET 265 uint16_t pseudolen = datalen + iph->protocol; 266 267 /* Compute and insert the pseudo-header cheksum. */ 268 *check = in_pseudo(iph->saddr, iph->daddr, 269 htobe16(pseudolen)); 270 /* Compute the checksum on TCP/UDP header + payload 271 * (includes the pseudo-header). 272 */ 273 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 274#else 275 static int notsupported = 0; 276 if (!notsupported) { 277 notsupported = 1; 278 nm_prerr("inet4 segmentation not supported"); 279 } 280#endif 281} 282 283void 284nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 285 size_t datalen, uint16_t *check) 286{ 287#ifdef INET6 288 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0); 289 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 290#else 291 static int notsupported = 0; 292 if (!notsupported) { 293 notsupported = 1; 294 nm_prerr("inet6 segmentation not supported"); 295 } 296#endif 297} 298 299/* on FreeBSD we send up one packet at a time */ 300void * 301nm_os_send_up(struct ifnet *ifp, struct mbuf *m, struct mbuf *prev) 302{ 303 NA(ifp)->if_input(ifp, m); 304 return NULL; 305} 306 307int 308nm_os_mbuf_has_csum_offld(struct mbuf *m) 309{ 310 return m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP | CSUM_SCTP | 311 CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | 312 CSUM_SCTP_IPV6); 313} 314 315int 316nm_os_mbuf_has_seg_offld(struct mbuf *m) 317{ 318 return m->m_pkthdr.csum_flags & CSUM_TSO; 319} 320 321static void 322freebsd_generic_rx_handler(struct ifnet *ifp, struct mbuf *m) 323{ 324 int stolen; 325 326 if (unlikely(!NM_NA_VALID(ifp))) { 327 nm_prlim(1, "Warning: RX packet intercepted, but no" 328 " emulated adapter"); 329 return; 330 } 331 332 stolen = generic_rx_handler(ifp, m); 333 if (!stolen) { 334 struct netmap_generic_adapter *gna = 335 (struct netmap_generic_adapter *)NA(ifp); 336 gna->save_if_input(ifp, m); 337 } 338} 339 340/* 341 * Intercept the rx routine in the standard device driver. 342 * Second argument is non-zero to intercept, 0 to restore 343 */ 344int 345nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept) 346{ 347 struct netmap_adapter *na = &gna->up.up; 348 struct ifnet *ifp = na->ifp; 349 int ret = 0; 350 351 nm_os_ifnet_lock(); 352 if (intercept) { 353 if (gna->save_if_input) { 354 nm_prerr("RX on %s already intercepted", na->name); 355 ret = EBUSY; /* already set */ 356 goto out; 357 } 358 gna->save_if_input = ifp->if_input; 359 ifp->if_input = freebsd_generic_rx_handler; 360 } else { 361 if (!gna->save_if_input) { 362 nm_prerr("Failed to undo RX intercept on %s", 363 na->name); 364 ret = EINVAL; /* not saved */ 365 goto out; 366 } 367 ifp->if_input = gna->save_if_input; 368 gna->save_if_input = NULL; 369 } 370out: 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 */ 383int 384nm_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 */ 418int 419nm_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#if __FreeBSD_version < 1100000 427 /* 428 * Old FreeBSD versions. The mbuf has a cluster attached, 429 * we need to copy from the cluster to the netmap buffer. 430 */ 431 if (MBUF_REFCNT(m) != 1) { 432 nm_prerr("invalid refcnt %d for %p", MBUF_REFCNT(m), m); 433 panic("in generic_xmit_frame"); 434 } 435 if (m->m_ext.ext_size < len) { 436 nm_prlim(2, "size %d < len %d", m->m_ext.ext_size, len); 437 len = m->m_ext.ext_size; 438 } 439 bcopy(a->addr, m->m_data, len); 440#else /* __FreeBSD_version >= 1100000 */ 441 /* New FreeBSD versions. Link the external storage to 442 * the netmap buffer, so that no copy is necessary. */ 443 m->m_ext.ext_buf = m->m_data = a->addr; 444 m->m_ext.ext_size = len; 445#endif /* __FreeBSD_version >= 1100000 */ 446 447 m->m_flags |= M_PKTHDR; 448 m->m_len = m->m_pkthdr.len = len; 449 450 /* mbuf refcnt is not contended, no need to use atomic 451 * (a memory barrier is enough). */ 452 SET_MBUF_REFCNT(m, 2); 453 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 454 m->m_pkthdr.flowid = a->ring_nr; 455 m->m_pkthdr.rcvif = ifp; /* used for tx notification */ 456 CURVNET_SET(ifp->if_vnet); 457 ret = NA(ifp)->if_transmit(ifp, m); 458 CURVNET_RESTORE(); 459 return ret ? -1 : 0; 460} 461 462 463#if __FreeBSD_version >= 1100005 464struct netmap_adapter * 465netmap_getna(if_t ifp) 466{ 467 return (NA((struct ifnet *)ifp)); 468} 469#endif /* __FreeBSD_version >= 1100005 */ 470 471/* 472 * The following two functions are empty until we have a generic 473 * way to extract the info from the ifp 474 */ 475int 476nm_os_generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx) 477{ 478 return 0; 479} 480 481 482void 483nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq) 484{ 485 unsigned num_rings = netmap_generic_rings ? netmap_generic_rings : 1; 486 487 *txq = num_rings; 488 *rxq = num_rings; 489} 490 491void 492nm_os_generic_set_features(struct netmap_generic_adapter *gna) 493{ 494 495 gna->rxsg = 1; /* Supported through m_copydata. */ 496 gna->txqdisc = 0; /* Not supported. */ 497} 498 499void 500nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na) 501{ 502 mit->mit_pending = 0; 503 mit->mit_ring_idx = idx; 504 mit->mit_na = na; 505} 506 507 508void 509nm_os_mitigation_start(struct nm_generic_mit *mit) 510{ 511} 512 513 514void 515nm_os_mitigation_restart(struct nm_generic_mit *mit) 516{ 517} 518 519 520int 521nm_os_mitigation_active(struct nm_generic_mit *mit) 522{ 523 524 return 0; 525} 526 527 528void 529nm_os_mitigation_cleanup(struct nm_generic_mit *mit) 530{ 531} 532 533static int 534nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr) 535{ 536 537 return EINVAL; 538} 539 540static void 541nm_vi_start(struct ifnet *ifp) 542{ 543 panic("nm_vi_start() must not be called"); 544} 545 546/* 547 * Index manager of persistent virtual interfaces. 548 * It is used to decide the lowest byte of the MAC address. 549 * We use the same algorithm with management of bridge port index. 550 */ 551#define NM_VI_MAX 255 552static struct { 553 uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */ 554 uint8_t active; 555 struct mtx lock; 556} nm_vi_indices; 557 558void 559nm_os_vi_init_index(void) 560{ 561 int i; 562 for (i = 0; i < NM_VI_MAX; i++) 563 nm_vi_indices.index[i] = i; 564 nm_vi_indices.active = 0; 565 mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF); 566} 567 568/* return -1 if no index available */ 569static int 570nm_vi_get_index(void) 571{ 572 int ret; 573 574 mtx_lock(&nm_vi_indices.lock); 575 ret = nm_vi_indices.active == NM_VI_MAX ? -1 : 576 nm_vi_indices.index[nm_vi_indices.active++]; 577 mtx_unlock(&nm_vi_indices.lock); 578 return ret; 579} 580 581static void 582nm_vi_free_index(uint8_t val) 583{ 584 int i, lim; 585 586 mtx_lock(&nm_vi_indices.lock); 587 lim = nm_vi_indices.active; 588 for (i = 0; i < lim; i++) { 589 if (nm_vi_indices.index[i] == val) { 590 /* swap index[lim-1] and j */ 591 int tmp = nm_vi_indices.index[lim-1]; 592 nm_vi_indices.index[lim-1] = val; 593 nm_vi_indices.index[i] = tmp; 594 nm_vi_indices.active--; 595 break; 596 } 597 } 598 if (lim == nm_vi_indices.active) 599 nm_prerr("Index %u not found", val); 600 mtx_unlock(&nm_vi_indices.lock); 601} 602#undef NM_VI_MAX 603 604/* 605 * Implementation of a netmap-capable virtual interface that 606 * registered to the system. 607 * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9. 608 * 609 * Note: Linux sets refcount to 0 on allocation of net_device, 610 * then increments it on registration to the system. 611 * FreeBSD sets refcount to 1 on if_alloc(), and does not 612 * increment this refcount on if_attach(). 613 */ 614int 615nm_os_vi_persist(const char *name, struct ifnet **ret) 616{ 617 struct ifnet *ifp; 618 u_short macaddr_hi; 619 uint32_t macaddr_mid; 620 u_char eaddr[6]; 621 int unit = nm_vi_get_index(); /* just to decide MAC address */ 622 623 if (unit < 0) 624 return EBUSY; 625 /* 626 * We use the same MAC address generation method with tap 627 * except for the highest octet is 00:be instead of 00:bd 628 */ 629 macaddr_hi = htons(0x00be); /* XXX tap + 1 */ 630 macaddr_mid = (uint32_t) ticks; 631 bcopy(&macaddr_hi, eaddr, sizeof(short)); 632 bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t)); 633 eaddr[5] = (uint8_t)unit; 634 635 ifp = if_alloc(IFT_ETHER); 636 if (ifp == NULL) { 637 nm_prerr("if_alloc failed"); 638 return ENOMEM; 639 } 640 if_initname(ifp, name, IF_DUNIT_NONE); 641 ifp->if_mtu = 65536; 642 ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; 643 ifp->if_init = (void *)nm_vi_dummy; 644 ifp->if_ioctl = nm_vi_dummy; 645 ifp->if_start = nm_vi_start; 646 ifp->if_mtu = ETHERMTU; 647 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 648 ifp->if_capabilities |= IFCAP_LINKSTATE; 649 ifp->if_capenable |= IFCAP_LINKSTATE; 650 651 ether_ifattach(ifp, eaddr); 652 *ret = ifp; 653 return 0; 654} 655 656/* unregister from the system and drop the final refcount */ 657void 658nm_os_vi_detach(struct ifnet *ifp) 659{ 660 nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]); 661 ether_ifdetach(ifp); 662 if_free(ifp); 663} 664 665#ifdef WITH_EXTMEM 666#include <vm/vm_map.h> 667#include <vm/vm_extern.h> 668#include <vm/vm_kern.h> 669struct nm_os_extmem { 670 vm_object_t obj; 671 vm_offset_t kva; 672 vm_offset_t size; 673 uintptr_t scan; 674}; 675 676void 677nm_os_extmem_delete(struct nm_os_extmem *e) 678{ 679 nm_prinf("freeing %zx bytes", (size_t)e->size); 680 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 681 nm_os_free(e); 682} 683 684char * 685nm_os_extmem_nextpage(struct nm_os_extmem *e) 686{ 687 char *rv = NULL; 688 if (e->scan < e->kva + e->size) { 689 rv = (char *)e->scan; 690 e->scan += PAGE_SIZE; 691 } 692 return rv; 693} 694 695int 696nm_os_extmem_isequal(struct nm_os_extmem *e1, struct nm_os_extmem *e2) 697{ 698 return (e1->obj == e2->obj); 699} 700 701int 702nm_os_extmem_nr_pages(struct nm_os_extmem *e) 703{ 704 return e->size >> PAGE_SHIFT; 705} 706 707struct nm_os_extmem * 708nm_os_extmem_create(unsigned long p, struct nmreq_pools_info *pi, int *perror) 709{ 710 vm_map_t map; 711 vm_map_entry_t entry; 712 vm_object_t obj; 713 vm_prot_t prot; 714 vm_pindex_t index; 715 boolean_t wired; 716 struct nm_os_extmem *e = NULL; 717 int rv, error = 0; 718 719 e = nm_os_malloc(sizeof(*e)); 720 if (e == NULL) { 721 error = ENOMEM; 722 goto out; 723 } 724 725 map = &curthread->td_proc->p_vmspace->vm_map; 726 rv = vm_map_lookup(&map, p, VM_PROT_RW, &entry, 727 &obj, &index, &prot, &wired); 728 if (rv != KERN_SUCCESS) { 729 nm_prerr("address %lx not found", p); 730 error = vm_mmap_to_errno(rv); 731 goto out_free; 732 } 733 vm_object_reference(obj); 734 735 /* check that we are given the whole vm_object ? */ 736 vm_map_lookup_done(map, entry); 737 738 e->obj = obj; 739 /* Wire the memory and add the vm_object to the kernel map, 740 * to make sure that it is not freed even if all the processes 741 * that are mmap()ing should munmap() it. 742 */ 743 e->kva = vm_map_min(kernel_map); 744 e->size = obj->size << PAGE_SHIFT; 745 rv = vm_map_find(kernel_map, obj, 0, &e->kva, e->size, 0, 746 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 747 VM_PROT_READ | VM_PROT_WRITE, 0); 748 if (rv != KERN_SUCCESS) { 749 nm_prerr("vm_map_find(%zx) failed", (size_t)e->size); 750 error = vm_mmap_to_errno(rv); 751 goto out_rel; 752 } 753 rv = vm_map_wire(kernel_map, e->kva, e->kva + e->size, 754 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 755 if (rv != KERN_SUCCESS) { 756 nm_prerr("vm_map_wire failed"); 757 error = vm_mmap_to_errno(rv); 758 goto out_rem; 759 } 760 761 e->scan = e->kva; 762 763 return e; 764 765out_rem: 766 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 767out_rel: 768 vm_object_deallocate(e->obj); 769 e->obj = NULL; 770out_free: 771 nm_os_free(e); 772out: 773 if (perror) 774 *perror = error; 775 return NULL; 776} 777#endif /* WITH_EXTMEM */ 778 779/* ================== PTNETMAP GUEST SUPPORT ==================== */ 780 781#ifdef WITH_PTNETMAP 782#include <sys/bus.h> 783#include <sys/rman.h> 784#include <machine/bus.h> /* bus_dmamap_* */ 785#include <machine/resource.h> 786#include <dev/pci/pcivar.h> 787#include <dev/pci/pcireg.h> 788/* 789 * ptnetmap memory device (memdev) for freebsd guest, 790 * ssed to expose host netmap memory to the guest through a PCI BAR. 791 */ 792 793/* 794 * ptnetmap memdev private data structure 795 */ 796struct ptnetmap_memdev { 797 device_t dev; 798 struct resource *pci_io; 799 struct resource *pci_mem; 800 struct netmap_mem_d *nm_mem; 801}; 802 803static int ptn_memdev_probe(device_t); 804static int ptn_memdev_attach(device_t); 805static int ptn_memdev_detach(device_t); 806static int ptn_memdev_shutdown(device_t); 807 808static device_method_t ptn_memdev_methods[] = { 809 DEVMETHOD(device_probe, ptn_memdev_probe), 810 DEVMETHOD(device_attach, ptn_memdev_attach), 811 DEVMETHOD(device_detach, ptn_memdev_detach), 812 DEVMETHOD(device_shutdown, ptn_memdev_shutdown), 813 DEVMETHOD_END 814}; 815 816static driver_t ptn_memdev_driver = { 817 PTNETMAP_MEMDEV_NAME, 818 ptn_memdev_methods, 819 sizeof(struct ptnetmap_memdev), 820}; 821 822/* We use (SI_ORDER_MIDDLE+1) here, see DEV_MODULE_ORDERED() invocation 823 * below. */ 824static devclass_t ptnetmap_devclass; 825DRIVER_MODULE_ORDERED(ptn_memdev, pci, ptn_memdev_driver, ptnetmap_devclass, 826 NULL, NULL, SI_ORDER_MIDDLE + 1); 827 828/* 829 * Map host netmap memory through PCI-BAR in the guest OS, 830 * returning physical (nm_paddr) and virtual (nm_addr) addresses 831 * of the netmap memory mapped in the guest. 832 */ 833int 834nm_os_pt_memdev_iomap(struct ptnetmap_memdev *ptn_dev, vm_paddr_t *nm_paddr, 835 void **nm_addr, uint64_t *mem_size) 836{ 837 int rid; 838 839 nm_prinf("ptn_memdev_driver iomap"); 840 841 rid = PCIR_BAR(PTNETMAP_MEM_PCI_BAR); 842 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_HI); 843 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_LO) | 844 (*mem_size << 32); 845 846 /* map memory allocator */ 847 ptn_dev->pci_mem = bus_alloc_resource(ptn_dev->dev, SYS_RES_MEMORY, 848 &rid, 0, ~0, *mem_size, RF_ACTIVE); 849 if (ptn_dev->pci_mem == NULL) { 850 *nm_paddr = 0; 851 *nm_addr = NULL; 852 return ENOMEM; 853 } 854 855 *nm_paddr = rman_get_start(ptn_dev->pci_mem); 856 *nm_addr = rman_get_virtual(ptn_dev->pci_mem); 857 858 nm_prinf("=== BAR %d start %lx len %lx mem_size %lx ===", 859 PTNETMAP_MEM_PCI_BAR, 860 (unsigned long)(*nm_paddr), 861 (unsigned long)rman_get_size(ptn_dev->pci_mem), 862 (unsigned long)*mem_size); 863 return (0); 864} 865 866uint32_t 867nm_os_pt_memdev_ioread(struct ptnetmap_memdev *ptn_dev, unsigned int reg) 868{ 869 return bus_read_4(ptn_dev->pci_io, reg); 870} 871 872/* Unmap host netmap memory. */ 873void 874nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *ptn_dev) 875{ 876 nm_prinf("ptn_memdev_driver iounmap"); 877 878 if (ptn_dev->pci_mem) { 879 bus_release_resource(ptn_dev->dev, SYS_RES_MEMORY, 880 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 881 ptn_dev->pci_mem = NULL; 882 } 883} 884 885/* Device identification routine, return BUS_PROBE_DEFAULT on success, 886 * positive on failure */ 887static int 888ptn_memdev_probe(device_t dev) 889{ 890 char desc[256]; 891 892 if (pci_get_vendor(dev) != PTNETMAP_PCI_VENDOR_ID) 893 return (ENXIO); 894 if (pci_get_device(dev) != PTNETMAP_PCI_DEVICE_ID) 895 return (ENXIO); 896 897 snprintf(desc, sizeof(desc), "%s PCI adapter", 898 PTNETMAP_MEMDEV_NAME); 899 device_set_desc_copy(dev, desc); 900 901 return (BUS_PROBE_DEFAULT); 902} 903 904/* Device initialization routine. */ 905static int 906ptn_memdev_attach(device_t dev) 907{ 908 struct ptnetmap_memdev *ptn_dev; 909 int rid; 910 uint16_t mem_id; 911 912 ptn_dev = device_get_softc(dev); 913 ptn_dev->dev = dev; 914 915 pci_enable_busmaster(dev); 916 917 rid = PCIR_BAR(PTNETMAP_IO_PCI_BAR); 918 ptn_dev->pci_io = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid, 919 RF_ACTIVE); 920 if (ptn_dev->pci_io == NULL) { 921 device_printf(dev, "cannot map I/O space\n"); 922 return (ENXIO); 923 } 924 925 mem_id = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMID); 926 927 /* create guest allocator */ 928 ptn_dev->nm_mem = netmap_mem_pt_guest_attach(ptn_dev, mem_id); 929 if (ptn_dev->nm_mem == NULL) { 930 ptn_memdev_detach(dev); 931 return (ENOMEM); 932 } 933 netmap_mem_get(ptn_dev->nm_mem); 934 935 nm_prinf("ptnetmap memdev attached, host memid: %u", mem_id); 936 937 return (0); 938} 939 940/* Device removal routine. */ 941static int 942ptn_memdev_detach(device_t dev) 943{ 944 struct ptnetmap_memdev *ptn_dev; 945 946 ptn_dev = device_get_softc(dev); 947 948 if (ptn_dev->nm_mem) { 949 nm_prinf("ptnetmap memdev detached, host memid %u", 950 netmap_mem_get_id(ptn_dev->nm_mem)); 951 netmap_mem_put(ptn_dev->nm_mem); 952 ptn_dev->nm_mem = NULL; 953 } 954 if (ptn_dev->pci_mem) { 955 bus_release_resource(dev, SYS_RES_MEMORY, 956 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 957 ptn_dev->pci_mem = NULL; 958 } 959 if (ptn_dev->pci_io) { 960 bus_release_resource(dev, SYS_RES_IOPORT, 961 PCIR_BAR(PTNETMAP_IO_PCI_BAR), ptn_dev->pci_io); 962 ptn_dev->pci_io = NULL; 963 } 964 965 return (0); 966} 967 968static int 969ptn_memdev_shutdown(device_t dev) 970{ 971 return bus_generic_shutdown(dev); 972} 973 974#endif /* WITH_PTNETMAP */ 975 976/* 977 * In order to track whether pages are still mapped, we hook into 978 * the standard cdev_pager and intercept the constructor and 979 * destructor. 980 */ 981 982struct netmap_vm_handle_t { 983 struct cdev *dev; 984 struct netmap_priv_d *priv; 985}; 986 987 988static int 989netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 990 vm_ooffset_t foff, struct ucred *cred, u_short *color) 991{ 992 struct netmap_vm_handle_t *vmh = handle; 993 994 if (netmap_verbose) 995 nm_prinf("handle %p size %jd prot %d foff %jd", 996 handle, (intmax_t)size, prot, (intmax_t)foff); 997 if (color) 998 *color = 0; 999 dev_ref(vmh->dev); 1000 return 0; 1001} 1002 1003 1004static void 1005netmap_dev_pager_dtor(void *handle) 1006{ 1007 struct netmap_vm_handle_t *vmh = handle; 1008 struct cdev *dev = vmh->dev; 1009 struct netmap_priv_d *priv = vmh->priv; 1010 1011 if (netmap_verbose) 1012 nm_prinf("handle %p", handle); 1013 netmap_dtor(priv); 1014 free(vmh, M_DEVBUF); 1015 dev_rel(dev); 1016} 1017 1018 1019static int 1020netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, 1021 int prot, vm_page_t *mres) 1022{ 1023 struct netmap_vm_handle_t *vmh = object->handle; 1024 struct netmap_priv_d *priv = vmh->priv; 1025 struct netmap_adapter *na = priv->np_na; 1026 vm_paddr_t paddr; 1027 vm_page_t page; 1028 vm_memattr_t memattr; 1029 1030 nm_prdis("object %p offset %jd prot %d mres %p", 1031 object, (intmax_t)offset, prot, mres); 1032 memattr = object->memattr; 1033 paddr = netmap_mem_ofstophys(na->nm_mem, offset); 1034 if (paddr == 0) 1035 return VM_PAGER_FAIL; 1036 1037 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 1038 /* 1039 * If the passed in result page is a fake page, update it with 1040 * the new physical address. 1041 */ 1042 page = *mres; 1043 vm_page_updatefake(page, paddr, memattr); 1044 } else { 1045 /* 1046 * Replace the passed in reqpage page with our own fake page and 1047 * free up the all of the original pages. 1048 */ 1049#ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */ 1050#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK 1051#define VM_OBJECT_WLOCK VM_OBJECT_LOCK 1052#endif /* VM_OBJECT_WUNLOCK */ 1053 1054 VM_OBJECT_WUNLOCK(object); 1055 page = vm_page_getfake(paddr, memattr); 1056 VM_OBJECT_WLOCK(object); 1057 vm_page_replace(page, object, (*mres)->pindex, *mres); 1058 *mres = page; 1059 } 1060 vm_page_valid(page); 1061 return (VM_PAGER_OK); 1062} 1063 1064 1065static struct cdev_pager_ops netmap_cdev_pager_ops = { 1066 .cdev_pg_ctor = netmap_dev_pager_ctor, 1067 .cdev_pg_dtor = netmap_dev_pager_dtor, 1068 .cdev_pg_fault = netmap_dev_pager_fault, 1069}; 1070 1071 1072static int 1073netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff, 1074 vm_size_t objsize, vm_object_t *objp, int prot) 1075{ 1076 int error; 1077 struct netmap_vm_handle_t *vmh; 1078 struct netmap_priv_d *priv; 1079 vm_object_t obj; 1080 1081 if (netmap_verbose) 1082 nm_prinf("cdev %p foff %jd size %jd objp %p prot %d", cdev, 1083 (intmax_t )*foff, (intmax_t )objsize, objp, prot); 1084 1085 vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF, 1086 M_NOWAIT | M_ZERO); 1087 if (vmh == NULL) 1088 return ENOMEM; 1089 vmh->dev = cdev; 1090 1091 NMG_LOCK(); 1092 error = devfs_get_cdevpriv((void**)&priv); 1093 if (error) 1094 goto err_unlock; 1095 if (priv->np_nifp == NULL) { 1096 error = EINVAL; 1097 goto err_unlock; 1098 } 1099 vmh->priv = priv; 1100 priv->np_refs++; 1101 NMG_UNLOCK(); 1102 1103 obj = cdev_pager_allocate(vmh, OBJT_DEVICE, 1104 &netmap_cdev_pager_ops, objsize, prot, 1105 *foff, NULL); 1106 if (obj == NULL) { 1107 nm_prerr("cdev_pager_allocate failed"); 1108 error = EINVAL; 1109 goto err_deref; 1110 } 1111 1112 *objp = obj; 1113 return 0; 1114 1115err_deref: 1116 NMG_LOCK(); 1117 priv->np_refs--; 1118err_unlock: 1119 NMG_UNLOCK(); 1120// err: 1121 free(vmh, M_DEVBUF); 1122 return error; 1123} 1124 1125/* 1126 * On FreeBSD the close routine is only called on the last close on 1127 * the device (/dev/netmap) so we cannot do anything useful. 1128 * To track close() on individual file descriptors we pass netmap_dtor() to 1129 * devfs_set_cdevpriv() on open(). The FreeBSD kernel will call the destructor 1130 * when the last fd pointing to the device is closed. 1131 * 1132 * Note that FreeBSD does not even munmap() on close() so we also have 1133 * to track mmap() ourselves, and postpone the call to 1134 * netmap_dtor() is called when the process has no open fds and no active 1135 * memory maps on /dev/netmap, as in linux. 1136 */ 1137static int 1138netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 1139{ 1140 if (netmap_verbose) 1141 nm_prinf("dev %p fflag 0x%x devtype %d td %p", 1142 dev, fflag, devtype, td); 1143 return 0; 1144} 1145 1146 1147static int 1148netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 1149{ 1150 struct netmap_priv_d *priv; 1151 int error; 1152 1153 (void)dev; 1154 (void)oflags; 1155 (void)devtype; 1156 (void)td; 1157 1158 NMG_LOCK(); 1159 priv = netmap_priv_new(); 1160 if (priv == NULL) { 1161 error = ENOMEM; 1162 goto out; 1163 } 1164 error = devfs_set_cdevpriv(priv, netmap_dtor); 1165 if (error) { 1166 netmap_priv_delete(priv); 1167 } 1168out: 1169 NMG_UNLOCK(); 1170 return error; 1171} 1172 1173/******************** kthread wrapper ****************/ 1174#include <sys/sysproto.h> 1175u_int 1176nm_os_ncpus(void) 1177{ 1178 return mp_maxid + 1; 1179} 1180 1181struct nm_kctx_ctx { 1182 /* Userspace thread (kthread creator). */ 1183 struct thread *user_td; 1184 1185 /* worker function and parameter */ 1186 nm_kctx_worker_fn_t worker_fn; 1187 void *worker_private; 1188 1189 struct nm_kctx *nmk; 1190 1191 /* integer to manage multiple worker contexts (e.g., RX or TX on ptnetmap) */ 1192 long type; 1193}; 1194 1195struct nm_kctx { 1196 struct thread *worker; 1197 struct mtx worker_lock; 1198 struct nm_kctx_ctx worker_ctx; 1199 int run; /* used to stop kthread */ 1200 int attach_user; /* kthread attached to user_process */ 1201 int affinity; 1202}; 1203 1204static void 1205nm_kctx_worker(void *data) 1206{ 1207 struct nm_kctx *nmk = data; 1208 struct nm_kctx_ctx *ctx = &nmk->worker_ctx; 1209 1210 if (nmk->affinity >= 0) { 1211 thread_lock(curthread); 1212 sched_bind(curthread, nmk->affinity); 1213 thread_unlock(curthread); 1214 } 1215 1216 while (nmk->run) { 1217 /* 1218 * check if the parent process dies 1219 * (when kthread is attached to user process) 1220 */ 1221 if (ctx->user_td) { 1222 PROC_LOCK(curproc); 1223 thread_suspend_check(0); 1224 PROC_UNLOCK(curproc); 1225 } else { 1226 kthread_suspend_check(); 1227 } 1228 1229 /* Continuously execute worker process. */ 1230 ctx->worker_fn(ctx->worker_private); /* worker body */ 1231 } 1232 1233 kthread_exit(); 1234} 1235 1236void 1237nm_os_kctx_worker_setaff(struct nm_kctx *nmk, int affinity) 1238{ 1239 nmk->affinity = affinity; 1240} 1241 1242struct nm_kctx * 1243nm_os_kctx_create(struct nm_kctx_cfg *cfg, void *opaque) 1244{ 1245 struct nm_kctx *nmk = NULL; 1246 1247 nmk = malloc(sizeof(*nmk), M_DEVBUF, M_NOWAIT | M_ZERO); 1248 if (!nmk) 1249 return NULL; 1250 1251 mtx_init(&nmk->worker_lock, "nm_kthread lock", NULL, MTX_DEF); 1252 nmk->worker_ctx.worker_fn = cfg->worker_fn; 1253 nmk->worker_ctx.worker_private = cfg->worker_private; 1254 nmk->worker_ctx.type = cfg->type; 1255 nmk->affinity = -1; 1256 1257 /* attach kthread to user process (ptnetmap) */ 1258 nmk->attach_user = cfg->attach_user; 1259 1260 return nmk; 1261} 1262 1263int 1264nm_os_kctx_worker_start(struct nm_kctx *nmk) 1265{ 1266 struct proc *p = NULL; 1267 int error = 0; 1268 1269 /* Temporarily disable this function as it is currently broken 1270 * and causes kernel crashes. The failure can be triggered by 1271 * the "vale_polling_enable_disable" test in ctrl-api-test.c. */ 1272 return EOPNOTSUPP; 1273 1274 if (nmk->worker) 1275 return EBUSY; 1276 1277 /* check if we want to attach kthread to user process */ 1278 if (nmk->attach_user) { 1279 nmk->worker_ctx.user_td = curthread; 1280 p = curthread->td_proc; 1281 } 1282 1283 /* enable kthread main loop */ 1284 nmk->run = 1; 1285 /* create kthread */ 1286 if((error = kthread_add(nm_kctx_worker, nmk, p, 1287 &nmk->worker, RFNOWAIT /* to be checked */, 0, "nm-kthread-%ld", 1288 nmk->worker_ctx.type))) { 1289 goto err; 1290 } 1291 1292 nm_prinf("nm_kthread started td %p", nmk->worker); 1293 1294 return 0; 1295err: 1296 nm_prerr("nm_kthread start failed err %d", error); 1297 nmk->worker = NULL; 1298 return error; 1299} 1300 1301void 1302nm_os_kctx_worker_stop(struct nm_kctx *nmk) 1303{ 1304 if (!nmk->worker) 1305 return; 1306 1307 /* tell to kthread to exit from main loop */ 1308 nmk->run = 0; 1309 1310 /* wake up kthread if it sleeps */ 1311 kthread_resume(nmk->worker); 1312 1313 nmk->worker = NULL; 1314} 1315 1316void 1317nm_os_kctx_destroy(struct nm_kctx *nmk) 1318{ 1319 if (!nmk) 1320 return; 1321 1322 if (nmk->worker) 1323 nm_os_kctx_worker_stop(nmk); 1324 1325 free(nmk, M_DEVBUF); 1326} 1327 1328/******************** kqueue support ****************/ 1329 1330/* 1331 * In addition to calling selwakeuppri(), nm_os_selwakeup() also 1332 * needs to call knote() to wake up kqueue listeners. 1333 * This operation is deferred to a taskqueue in order to avoid possible 1334 * lock order reversals; these may happen because knote() grabs a 1335 * private lock associated to the 'si' (see struct selinfo, 1336 * struct nm_selinfo, and nm_os_selinfo_init), and nm_os_selwakeup() 1337 * can be called while holding the lock associated to a different 1338 * 'si'. 1339 * When calling knote() we use a non-zero 'hint' argument to inform 1340 * the netmap_knrw() function that it is being called from 1341 * 'nm_os_selwakeup'; this is necessary because when netmap_knrw() is 1342 * called by the kevent subsystem (i.e. kevent_scan()) we also need to 1343 * call netmap_poll(). 1344 * 1345 * The netmap_kqfilter() function registers one or another f_event 1346 * depending on read or write mode. A pointer to the struct 1347 * 'netmap_priv_d' is stored into kn->kn_hook, so that it can later 1348 * be passed to netmap_poll(). We pass NULL as a third argument to 1349 * netmap_poll(), so that the latter only runs the txsync/rxsync 1350 * (if necessary), and skips the nm_os_selrecord() calls. 1351 */ 1352 1353 1354void 1355nm_os_selwakeup(struct nm_selinfo *si) 1356{ 1357 selwakeuppri(&si->si, PI_NET); 1358 if (si->kqueue_users > 0) { 1359 taskqueue_enqueue(si->ntfytq, &si->ntfytask); 1360 } 1361} 1362 1363void 1364nm_os_selrecord(struct thread *td, struct nm_selinfo *si) 1365{ 1366 selrecord(td, &si->si); 1367} 1368 1369static void 1370netmap_knrdetach(struct knote *kn) 1371{ 1372 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1373 struct nm_selinfo *si = priv->np_si[NR_RX]; 1374 1375 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1376 NMG_LOCK(); 1377 KASSERT(si->kqueue_users > 0, ("kqueue_user underflow on %s", 1378 si->mtxname)); 1379 si->kqueue_users--; 1380 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1381 NMG_UNLOCK(); 1382} 1383 1384static void 1385netmap_knwdetach(struct knote *kn) 1386{ 1387 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1388 struct nm_selinfo *si = priv->np_si[NR_TX]; 1389 1390 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1391 NMG_LOCK(); 1392 si->kqueue_users--; 1393 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1394 NMG_UNLOCK(); 1395} 1396 1397/* 1398 * Callback triggered by netmap notifications (see netmap_notify()), 1399 * and by the application calling kevent(). In the former case we 1400 * just return 1 (events ready), since we are not able to do better. 1401 * In the latter case we use netmap_poll() to see which events are 1402 * ready. 1403 */ 1404static int 1405netmap_knrw(struct knote *kn, long hint, int events) 1406{ 1407 struct netmap_priv_d *priv; 1408 int revents; 1409 1410 if (hint != 0) { 1411 /* Called from netmap_notify(), typically from a 1412 * thread different from the one issuing kevent(). 1413 * Assume we are ready. */ 1414 return 1; 1415 } 1416 1417 /* Called from kevent(). */ 1418 priv = kn->kn_hook; 1419 revents = netmap_poll(priv, events, /*thread=*/NULL); 1420 1421 return (events & revents) ? 1 : 0; 1422} 1423 1424static int 1425netmap_knread(struct knote *kn, long hint) 1426{ 1427 return netmap_knrw(kn, hint, POLLIN); 1428} 1429 1430static int 1431netmap_knwrite(struct knote *kn, long hint) 1432{ 1433 return netmap_knrw(kn, hint, POLLOUT); 1434} 1435 1436static struct filterops netmap_rfiltops = { 1437 .f_isfd = 1, 1438 .f_detach = netmap_knrdetach, 1439 .f_event = netmap_knread, 1440}; 1441 1442static struct filterops netmap_wfiltops = { 1443 .f_isfd = 1, 1444 .f_detach = netmap_knwdetach, 1445 .f_event = netmap_knwrite, 1446}; 1447 1448 1449/* 1450 * This is called when a thread invokes kevent() to record 1451 * a change in the configuration of the kqueue(). 1452 * The 'priv' is the one associated to the open netmap device. 1453 */ 1454static int 1455netmap_kqfilter(struct cdev *dev, struct knote *kn) 1456{ 1457 struct netmap_priv_d *priv; 1458 int error; 1459 struct netmap_adapter *na; 1460 struct nm_selinfo *si; 1461 int ev = kn->kn_filter; 1462 1463 if (ev != EVFILT_READ && ev != EVFILT_WRITE) { 1464 nm_prerr("bad filter request %d", ev); 1465 return 1; 1466 } 1467 error = devfs_get_cdevpriv((void**)&priv); 1468 if (error) { 1469 nm_prerr("device not yet setup"); 1470 return 1; 1471 } 1472 na = priv->np_na; 1473 if (na == NULL) { 1474 nm_prerr("no netmap adapter for this file descriptor"); 1475 return 1; 1476 } 1477 /* the si is indicated in the priv */ 1478 si = priv->np_si[(ev == EVFILT_WRITE) ? NR_TX : NR_RX]; 1479 kn->kn_fop = (ev == EVFILT_WRITE) ? 1480 &netmap_wfiltops : &netmap_rfiltops; 1481 kn->kn_hook = priv; 1482 NMG_LOCK(); 1483 si->kqueue_users++; 1484 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1485 NMG_UNLOCK(); 1486 knlist_add(&si->si.si_note, kn, /*islocked=*/0); 1487 1488 return 0; 1489} 1490 1491static int 1492freebsd_netmap_poll(struct cdev *cdevi __unused, int events, struct thread *td) 1493{ 1494 struct netmap_priv_d *priv; 1495 if (devfs_get_cdevpriv((void **)&priv)) { 1496 return POLLERR; 1497 } 1498 return netmap_poll(priv, events, td); 1499} 1500 1501static int 1502freebsd_netmap_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t data, 1503 int ffla __unused, struct thread *td) 1504{ 1505 int error; 1506 struct netmap_priv_d *priv; 1507 1508 CURVNET_SET(TD_TO_VNET(td)); 1509 error = devfs_get_cdevpriv((void **)&priv); 1510 if (error) { 1511 /* XXX ENOENT should be impossible, since the priv 1512 * is now created in the open */ 1513 if (error == ENOENT) 1514 error = ENXIO; 1515 goto out; 1516 } 1517 error = netmap_ioctl(priv, cmd, data, td, /*nr_body_is_user=*/1); 1518out: 1519 CURVNET_RESTORE(); 1520 1521 return error; 1522} 1523 1524void 1525nm_os_onattach(struct ifnet *ifp) 1526{ 1527 ifp->if_capabilities |= IFCAP_NETMAP; 1528} 1529 1530void 1531nm_os_onenter(struct ifnet *ifp) 1532{ 1533 struct netmap_adapter *na = NA(ifp); 1534 1535 na->if_transmit = ifp->if_transmit; 1536 ifp->if_transmit = netmap_transmit; 1537 ifp->if_capenable |= IFCAP_NETMAP; 1538} 1539 1540void 1541nm_os_onexit(struct ifnet *ifp) 1542{ 1543 struct netmap_adapter *na = NA(ifp); 1544 1545 ifp->if_transmit = na->if_transmit; 1546 ifp->if_capenable &= ~IFCAP_NETMAP; 1547} 1548 1549extern struct cdevsw netmap_cdevsw; /* XXX used in netmap.c, should go elsewhere */ 1550struct cdevsw netmap_cdevsw = { 1551 .d_version = D_VERSION, 1552 .d_name = "netmap", 1553 .d_open = netmap_open, 1554 .d_mmap_single = netmap_mmap_single, 1555 .d_ioctl = freebsd_netmap_ioctl, 1556 .d_poll = freebsd_netmap_poll, 1557 .d_kqfilter = netmap_kqfilter, 1558 .d_close = netmap_close, 1559}; 1560/*--- end of kqueue support ----*/ 1561 1562/* 1563 * Kernel entry point. 1564 * 1565 * Initialize/finalize the module and return. 1566 * 1567 * Return 0 on success, errno on failure. 1568 */ 1569static int 1570netmap_loader(__unused struct module *module, int event, __unused void *arg) 1571{ 1572 int error = 0; 1573 1574 switch (event) { 1575 case MOD_LOAD: 1576 error = netmap_init(); 1577 break; 1578 1579 case MOD_UNLOAD: 1580 /* 1581 * if some one is still using netmap, 1582 * then the module can not be unloaded. 1583 */ 1584 if (netmap_use_count) { 1585 nm_prerr("netmap module can not be unloaded - netmap_use_count: %d", 1586 netmap_use_count); 1587 error = EBUSY; 1588 break; 1589 } 1590 netmap_fini(); 1591 break; 1592 1593 default: 1594 error = EOPNOTSUPP; 1595 break; 1596 } 1597 1598 return (error); 1599} 1600 1601#ifdef DEV_MODULE_ORDERED 1602/* 1603 * The netmap module contains three drivers: (i) the netmap character device 1604 * driver; (ii) the ptnetmap memdev PCI device driver, (iii) the ptnet PCI 1605 * device driver. The attach() routines of both (ii) and (iii) need the 1606 * lock of the global allocator, and such lock is initialized in netmap_init(), 1607 * which is part of (i). 1608 * Therefore, we make sure that (i) is loaded before (ii) and (iii), using 1609 * the 'order' parameter of driver declaration macros. For (i), we specify 1610 * SI_ORDER_MIDDLE, while higher orders are used with the DRIVER_MODULE_ORDERED 1611 * macros for (ii) and (iii). 1612 */ 1613DEV_MODULE_ORDERED(netmap, netmap_loader, NULL, SI_ORDER_MIDDLE); 1614#else /* !DEV_MODULE_ORDERED */ 1615DEV_MODULE(netmap, netmap_loader, NULL); 1616#endif /* DEV_MODULE_ORDERED */ 1617MODULE_DEPEND(netmap, pci, 1, 1, 1); 1618MODULE_VERSION(netmap, 1); 1619/* reduce conditional code */ 1620// linux API, use for the knlist in FreeBSD 1621/* use a private mutex for the knlist */ 1622