1/* 2 * Copyright (C) 2013-2014 Universita` di Pisa. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 */ 25 26/* $FreeBSD: stable/11/sys/dev/netmap/netmap_freebsd.c 369514 2021-03-23 21:21:46Z git2svn $ */ 27#include "opt_inet.h" 28#include "opt_inet6.h" 29 30#include <sys/param.h> 31#include <sys/module.h> 32#include <sys/errno.h> 33#include <sys/jail.h> 34#include <sys/poll.h> /* POLLIN, POLLOUT */ 35#include <sys/kernel.h> /* types used in module initialization */ 36#include <sys/conf.h> /* DEV_MODULE_ORDERED */ 37#include <sys/endian.h> 38#include <sys/syscallsubr.h> /* kern_ioctl() */ 39 40#include <sys/rwlock.h> 41 42#include <vm/vm.h> /* vtophys */ 43#include <vm/pmap.h> /* vtophys */ 44#include <vm/vm_param.h> 45#include <vm/vm_object.h> 46#include <vm/vm_page.h> 47#include <vm/vm_pager.h> 48#include <vm/uma.h> 49 50 51#include <sys/malloc.h> 52#include <sys/socket.h> /* sockaddrs */ 53#include <sys/selinfo.h> 54#include <sys/kthread.h> /* kthread_add() */ 55#include <sys/proc.h> /* PROC_LOCK() */ 56#include <sys/unistd.h> /* RFNOWAIT */ 57#include <sys/sched.h> /* sched_bind() */ 58#include <sys/smp.h> /* mp_maxid */ 59#include <sys/taskqueue.h> /* taskqueue_enqueue(), taskqueue_create(), ... */ 60#include <net/if.h> 61#include <net/if_var.h> 62#include <net/if_types.h> /* IFT_ETHER */ 63#include <net/ethernet.h> /* ether_ifdetach */ 64#include <net/if_dl.h> /* LLADDR */ 65#include <machine/bus.h> /* bus_dmamap_* */ 66#include <netinet/in.h> /* in6_cksum_pseudo() */ 67#include <machine/in_cksum.h> /* in_pseudo(), in_cksum_hdr() */ 68 69#include <net/netmap.h> 70#include <dev/netmap/netmap_kern.h> 71#include <net/netmap_virt.h> 72#include <dev/netmap/netmap_mem2.h> 73 74 75/* ======================== FREEBSD-SPECIFIC ROUTINES ================== */ 76 77static void 78nm_kqueue_notify(void *opaque, int pending) 79{ 80 struct nm_selinfo *si = opaque; 81 82 /* We use a non-zero hint to distinguish this notification call 83 * from the call done in kqueue_scan(), which uses hint=0. 84 */ 85 KNOTE_UNLOCKED(&si->si.si_note, /*hint=*/0x100); 86} 87 88int nm_os_selinfo_init(NM_SELINFO_T *si, const char *name) { 89 int err; 90 91 TASK_INIT(&si->ntfytask, 0, nm_kqueue_notify, si); 92 si->ntfytq = taskqueue_create(name, M_NOWAIT, 93 taskqueue_thread_enqueue, &si->ntfytq); 94 if (si->ntfytq == NULL) 95 return -ENOMEM; 96 err = taskqueue_start_threads(&si->ntfytq, 1, PI_NET, "tq %s", name); 97 if (err) { 98 taskqueue_free(si->ntfytq); 99 si->ntfytq = NULL; 100 return err; 101 } 102 103 snprintf(si->mtxname, sizeof(si->mtxname), "nmkl%s", name); 104 mtx_init(&si->m, si->mtxname, NULL, MTX_DEF); 105 knlist_init_mtx(&si->si.si_note, &si->m); 106 si->kqueue_users = 0; 107 108 return (0); 109} 110 111void 112nm_os_selinfo_uninit(NM_SELINFO_T *si) 113{ 114 if (si->ntfytq == NULL) { 115 return; /* si was not initialized */ 116 } 117 taskqueue_drain(si->ntfytq, &si->ntfytask); 118 taskqueue_free(si->ntfytq); 119 si->ntfytq = NULL; 120 knlist_delete(&si->si.si_note, curthread, /*islocked=*/0); 121 knlist_destroy(&si->si.si_note); 122 /* now we don't need the mutex anymore */ 123 mtx_destroy(&si->m); 124} 125 126void * 127nm_os_malloc(size_t size) 128{ 129 return malloc(size, M_DEVBUF, M_NOWAIT | M_ZERO); 130} 131 132void * 133nm_os_realloc(void *addr, size_t new_size, size_t old_size __unused) 134{ 135 return realloc(addr, new_size, M_DEVBUF, M_NOWAIT | M_ZERO); 136} 137 138void 139nm_os_free(void *addr) 140{ 141 free(addr, M_DEVBUF); 142} 143 144void 145nm_os_ifnet_lock(void) 146{ 147 IFNET_RLOCK(); 148} 149 150void 151nm_os_ifnet_unlock(void) 152{ 153 IFNET_RUNLOCK(); 154} 155 156static int netmap_use_count = 0; 157 158void 159nm_os_get_module(void) 160{ 161 netmap_use_count++; 162} 163 164void 165nm_os_put_module(void) 166{ 167 netmap_use_count--; 168} 169 170static void 171netmap_ifnet_arrival_handler(void *arg __unused, struct ifnet *ifp) 172{ 173 netmap_undo_zombie(ifp); 174} 175 176static void 177netmap_ifnet_departure_handler(void *arg __unused, struct ifnet *ifp) 178{ 179 netmap_make_zombie(ifp); 180} 181 182static eventhandler_tag nm_ifnet_ah_tag; 183static eventhandler_tag nm_ifnet_dh_tag; 184 185int 186nm_os_ifnet_init(void) 187{ 188 nm_ifnet_ah_tag = 189 EVENTHANDLER_REGISTER(ifnet_arrival_event, 190 netmap_ifnet_arrival_handler, 191 NULL, EVENTHANDLER_PRI_ANY); 192 nm_ifnet_dh_tag = 193 EVENTHANDLER_REGISTER(ifnet_departure_event, 194 netmap_ifnet_departure_handler, 195 NULL, EVENTHANDLER_PRI_ANY); 196 return 0; 197} 198 199void 200nm_os_ifnet_fini(void) 201{ 202 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 203 nm_ifnet_ah_tag); 204 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 205 nm_ifnet_dh_tag); 206} 207 208unsigned 209nm_os_ifnet_mtu(struct ifnet *ifp) 210{ 211#if __FreeBSD_version < 1100030 212 return ifp->if_data.ifi_mtu; 213#else /* __FreeBSD_version >= 1100030 */ 214 return ifp->if_mtu; 215#endif 216} 217 218rawsum_t 219nm_os_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum) 220{ 221 /* TODO XXX please use the FreeBSD implementation for this. */ 222 uint16_t *words = (uint16_t *)data; 223 int nw = len / 2; 224 int i; 225 226 for (i = 0; i < nw; i++) 227 cur_sum += be16toh(words[i]); 228 229 if (len & 1) 230 cur_sum += (data[len-1] << 8); 231 232 return cur_sum; 233} 234 235/* Fold a raw checksum: 'cur_sum' is in host byte order, while the 236 * return value is in network byte order. 237 */ 238uint16_t 239nm_os_csum_fold(rawsum_t cur_sum) 240{ 241 /* TODO XXX please use the FreeBSD implementation for this. */ 242 while (cur_sum >> 16) 243 cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16); 244 245 return htobe16((~cur_sum) & 0xFFFF); 246} 247 248uint16_t nm_os_csum_ipv4(struct nm_iphdr *iph) 249{ 250#if 0 251 return in_cksum_hdr((void *)iph); 252#else 253 return nm_os_csum_fold(nm_os_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0)); 254#endif 255} 256 257void 258nm_os_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 259 size_t datalen, uint16_t *check) 260{ 261#ifdef INET 262 uint16_t pseudolen = datalen + iph->protocol; 263 264 /* Compute and insert the pseudo-header cheksum. */ 265 *check = in_pseudo(iph->saddr, iph->daddr, 266 htobe16(pseudolen)); 267 /* Compute the checksum on TCP/UDP header + payload 268 * (includes the pseudo-header). 269 */ 270 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 271#else 272 static int notsupported = 0; 273 if (!notsupported) { 274 notsupported = 1; 275 nm_prerr("inet4 segmentation not supported"); 276 } 277#endif 278} 279 280void 281nm_os_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 282 size_t datalen, uint16_t *check) 283{ 284#ifdef INET6 285 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0); 286 *check = nm_os_csum_fold(nm_os_csum_raw(data, datalen, 0)); 287#else 288 static int notsupported = 0; 289 if (!notsupported) { 290 notsupported = 1; 291 nm_prerr("inet6 segmentation not supported"); 292 } 293#endif 294} 295 296/* on FreeBSD we send up one packet at a time */ 297void * 298nm_os_send_up(struct ifnet *ifp, struct mbuf *m, struct mbuf *prev) 299{ 300 NA(ifp)->if_input(ifp, m); 301 return NULL; 302} 303 304int 305nm_os_mbuf_has_csum_offld(struct mbuf *m) 306{ 307 return m->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP | CSUM_SCTP | 308 CSUM_TCP_IPV6 | CSUM_UDP_IPV6 | 309 CSUM_SCTP_IPV6); 310} 311 312int 313nm_os_mbuf_has_seg_offld(struct mbuf *m) 314{ 315 return m->m_pkthdr.csum_flags & CSUM_TSO; 316} 317 318static void 319freebsd_generic_rx_handler(struct ifnet *ifp, struct mbuf *m) 320{ 321 int stolen; 322 323 if (unlikely(!NM_NA_VALID(ifp))) { 324 nm_prlim(1, "Warning: RX packet intercepted, but no" 325 " emulated adapter"); 326 return; 327 } 328 329 stolen = generic_rx_handler(ifp, m); 330 if (!stolen) { 331 struct netmap_generic_adapter *gna = 332 (struct netmap_generic_adapter *)NA(ifp); 333 gna->save_if_input(ifp, m); 334 } 335} 336 337/* 338 * Intercept the rx routine in the standard device driver. 339 * Second argument is non-zero to intercept, 0 to restore 340 */ 341int 342nm_os_catch_rx(struct netmap_generic_adapter *gna, int intercept) 343{ 344 struct netmap_adapter *na = &gna->up.up; 345 struct ifnet *ifp = na->ifp; 346 int ret = 0; 347 348 nm_os_ifnet_lock(); 349 if (intercept) { 350 if (gna->save_if_input) { 351 nm_prerr("RX on %s already intercepted", na->name); 352 ret = EBUSY; /* already set */ 353 goto out; 354 } 355 gna->save_if_input = ifp->if_input; 356 ifp->if_input = freebsd_generic_rx_handler; 357 } else { 358 if (!gna->save_if_input) { 359 nm_prerr("Failed to undo RX intercept on %s", 360 na->name); 361 ret = EINVAL; /* not saved */ 362 goto out; 363 } 364 ifp->if_input = gna->save_if_input; 365 gna->save_if_input = NULL; 366 } 367out: 368 nm_os_ifnet_unlock(); 369 370 return ret; 371} 372 373 374/* 375 * Intercept the packet steering routine in the tx path, 376 * so that we can decide which queue is used for an mbuf. 377 * Second argument is non-zero to intercept, 0 to restore. 378 * On freebsd we just intercept if_transmit. 379 */ 380int 381nm_os_catch_tx(struct netmap_generic_adapter *gna, int intercept) 382{ 383 struct netmap_adapter *na = &gna->up.up; 384 struct ifnet *ifp = netmap_generic_getifp(gna); 385 386 nm_os_ifnet_lock(); 387 if (intercept) { 388 na->if_transmit = ifp->if_transmit; 389 ifp->if_transmit = netmap_transmit; 390 } else { 391 ifp->if_transmit = na->if_transmit; 392 } 393 nm_os_ifnet_unlock(); 394 395 return 0; 396} 397 398 399/* 400 * Transmit routine used by generic_netmap_txsync(). Returns 0 on success 401 * and non-zero on error (which may be packet drops or other errors). 402 * addr and len identify the netmap buffer, m is the (preallocated) 403 * mbuf to use for transmissions. 404 * 405 * We should add a reference to the mbuf so the m_freem() at the end 406 * of the transmission does not consume resources. 407 * 408 * On FreeBSD, and on multiqueue cards, we can force the queue using 409 * if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 410 * i = m->m_pkthdr.flowid % adapter->num_queues; 411 * else 412 * i = curcpu % adapter->num_queues; 413 * 414 */ 415int 416nm_os_generic_xmit_frame(struct nm_os_gen_arg *a) 417{ 418 int ret; 419 u_int len = a->len; 420 struct ifnet *ifp = a->ifp; 421 struct mbuf *m = a->m; 422 423#if __FreeBSD_version < 1100000 424 /* 425 * Old FreeBSD versions. The mbuf has a cluster attached, 426 * we need to copy from the cluster to the netmap buffer. 427 */ 428 if (MBUF_REFCNT(m) != 1) { 429 nm_prerr("invalid refcnt %d for %p", MBUF_REFCNT(m), m); 430 panic("in generic_xmit_frame"); 431 } 432 if (m->m_ext.ext_size < len) { 433 nm_prlim(2, "size %d < len %d", m->m_ext.ext_size, len); 434 len = m->m_ext.ext_size; 435 } 436 bcopy(a->addr, m->m_data, len); 437#else /* __FreeBSD_version >= 1100000 */ 438 /* New FreeBSD versions. Link the external storage to 439 * the netmap buffer, so that no copy is necessary. */ 440 m->m_ext.ext_buf = m->m_data = a->addr; 441 m->m_ext.ext_size = len; 442#endif /* __FreeBSD_version >= 1100000 */ 443 444 m->m_flags |= M_PKTHDR; 445 m->m_len = m->m_pkthdr.len = len; 446 447 /* mbuf refcnt is not contended, no need to use atomic 448 * (a memory barrier is enough). */ 449 SET_MBUF_REFCNT(m, 2); 450 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 451 m->m_pkthdr.flowid = a->ring_nr; 452 m->m_pkthdr.rcvif = ifp; /* used for tx notification */ 453 CURVNET_SET(ifp->if_vnet); 454 ret = NA(ifp)->if_transmit(ifp, m); 455 CURVNET_RESTORE(); 456 return ret ? -1 : 0; 457} 458 459 460#if __FreeBSD_version >= 1100005 461struct netmap_adapter * 462netmap_getna(if_t ifp) 463{ 464 return (NA((struct ifnet *)ifp)); 465} 466#endif /* __FreeBSD_version >= 1100005 */ 467 468/* 469 * The following two functions are empty until we have a generic 470 * way to extract the info from the ifp 471 */ 472int 473nm_os_generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx) 474{ 475 return 0; 476} 477 478 479void 480nm_os_generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq) 481{ 482 unsigned num_rings = netmap_generic_rings ? netmap_generic_rings : 1; 483 484 *txq = num_rings; 485 *rxq = num_rings; 486} 487 488void 489nm_os_generic_set_features(struct netmap_generic_adapter *gna) 490{ 491 492 gna->rxsg = 1; /* Supported through m_copydata. */ 493 gna->txqdisc = 0; /* Not supported. */ 494} 495 496void 497nm_os_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na) 498{ 499 mit->mit_pending = 0; 500 mit->mit_ring_idx = idx; 501 mit->mit_na = na; 502} 503 504 505void 506nm_os_mitigation_start(struct nm_generic_mit *mit) 507{ 508} 509 510 511void 512nm_os_mitigation_restart(struct nm_generic_mit *mit) 513{ 514} 515 516 517int 518nm_os_mitigation_active(struct nm_generic_mit *mit) 519{ 520 521 return 0; 522} 523 524 525void 526nm_os_mitigation_cleanup(struct nm_generic_mit *mit) 527{ 528} 529 530static int 531nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr) 532{ 533 534 return EINVAL; 535} 536 537static void 538nm_vi_start(struct ifnet *ifp) 539{ 540 panic("nm_vi_start() must not be called"); 541} 542 543/* 544 * Index manager of persistent virtual interfaces. 545 * It is used to decide the lowest byte of the MAC address. 546 * We use the same algorithm with management of bridge port index. 547 */ 548#define NM_VI_MAX 255 549static struct { 550 uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */ 551 uint8_t active; 552 struct mtx lock; 553} nm_vi_indices; 554 555void 556nm_os_vi_init_index(void) 557{ 558 int i; 559 for (i = 0; i < NM_VI_MAX; i++) 560 nm_vi_indices.index[i] = i; 561 nm_vi_indices.active = 0; 562 mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF); 563} 564 565/* return -1 if no index available */ 566static int 567nm_vi_get_index(void) 568{ 569 int ret; 570 571 mtx_lock(&nm_vi_indices.lock); 572 ret = nm_vi_indices.active == NM_VI_MAX ? -1 : 573 nm_vi_indices.index[nm_vi_indices.active++]; 574 mtx_unlock(&nm_vi_indices.lock); 575 return ret; 576} 577 578static void 579nm_vi_free_index(uint8_t val) 580{ 581 int i, lim; 582 583 mtx_lock(&nm_vi_indices.lock); 584 lim = nm_vi_indices.active; 585 for (i = 0; i < lim; i++) { 586 if (nm_vi_indices.index[i] == val) { 587 /* swap index[lim-1] and j */ 588 int tmp = nm_vi_indices.index[lim-1]; 589 nm_vi_indices.index[lim-1] = val; 590 nm_vi_indices.index[i] = tmp; 591 nm_vi_indices.active--; 592 break; 593 } 594 } 595 if (lim == nm_vi_indices.active) 596 nm_prerr("Index %u not found", val); 597 mtx_unlock(&nm_vi_indices.lock); 598} 599#undef NM_VI_MAX 600 601/* 602 * Implementation of a netmap-capable virtual interface that 603 * registered to the system. 604 * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9. 605 * 606 * Note: Linux sets refcount to 0 on allocation of net_device, 607 * then increments it on registration to the system. 608 * FreeBSD sets refcount to 1 on if_alloc(), and does not 609 * increment this refcount on if_attach(). 610 */ 611int 612nm_os_vi_persist(const char *name, struct ifnet **ret) 613{ 614 struct ifnet *ifp; 615 u_short macaddr_hi; 616 uint32_t macaddr_mid; 617 u_char eaddr[6]; 618 int unit = nm_vi_get_index(); /* just to decide MAC address */ 619 620 if (unit < 0) 621 return EBUSY; 622 /* 623 * We use the same MAC address generation method with tap 624 * except for the highest octet is 00:be instead of 00:bd 625 */ 626 macaddr_hi = htons(0x00be); /* XXX tap + 1 */ 627 macaddr_mid = (uint32_t) ticks; 628 bcopy(&macaddr_hi, eaddr, sizeof(short)); 629 bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t)); 630 eaddr[5] = (uint8_t)unit; 631 632 ifp = if_alloc(IFT_ETHER); 633 if (ifp == NULL) { 634 nm_prerr("if_alloc failed"); 635 return ENOMEM; 636 } 637 if_initname(ifp, name, IF_DUNIT_NONE); 638 ifp->if_mtu = 65536; 639 ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; 640 ifp->if_init = (void *)nm_vi_dummy; 641 ifp->if_ioctl = nm_vi_dummy; 642 ifp->if_start = nm_vi_start; 643 ifp->if_mtu = ETHERMTU; 644 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 645 ifp->if_capabilities |= IFCAP_LINKSTATE; 646 ifp->if_capenable |= IFCAP_LINKSTATE; 647 648 ether_ifattach(ifp, eaddr); 649 *ret = ifp; 650 return 0; 651} 652 653/* unregister from the system and drop the final refcount */ 654void 655nm_os_vi_detach(struct ifnet *ifp) 656{ 657 nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]); 658 ether_ifdetach(ifp); 659 if_free(ifp); 660} 661 662#ifdef WITH_EXTMEM 663#include <vm/vm_map.h> 664#include <vm/vm_extern.h> 665#include <vm/vm_kern.h> 666struct nm_os_extmem { 667 vm_object_t obj; 668 vm_offset_t kva; 669 vm_offset_t size; 670 uintptr_t scan; 671}; 672 673void 674nm_os_extmem_delete(struct nm_os_extmem *e) 675{ 676 nm_prinf("freeing %zx bytes", (size_t)e->size); 677 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 678 nm_os_free(e); 679} 680 681char * 682nm_os_extmem_nextpage(struct nm_os_extmem *e) 683{ 684 char *rv = NULL; 685 if (e->scan < e->kva + e->size) { 686 rv = (char *)e->scan; 687 e->scan += PAGE_SIZE; 688 } 689 return rv; 690} 691 692int 693nm_os_extmem_isequal(struct nm_os_extmem *e1, struct nm_os_extmem *e2) 694{ 695 return (e1->obj == e2->obj); 696} 697 698int 699nm_os_extmem_nr_pages(struct nm_os_extmem *e) 700{ 701 return e->size >> PAGE_SHIFT; 702} 703 704struct nm_os_extmem * 705nm_os_extmem_create(unsigned long p, struct nmreq_pools_info *pi, int *perror) 706{ 707 vm_map_t map; 708 vm_map_entry_t entry; 709 vm_object_t obj; 710 vm_prot_t prot; 711 vm_pindex_t index; 712 boolean_t wired; 713 struct nm_os_extmem *e = NULL; 714 int rv, error = 0; 715 716 e = nm_os_malloc(sizeof(*e)); 717 if (e == NULL) { 718 error = ENOMEM; 719 goto out; 720 } 721 722 map = &curthread->td_proc->p_vmspace->vm_map; 723 rv = vm_map_lookup(&map, p, VM_PROT_RW, &entry, 724 &obj, &index, &prot, &wired); 725 if (rv != KERN_SUCCESS) { 726 nm_prerr("address %lx not found", p); 727 error = vm_mmap_to_errno(rv); 728 goto out_free; 729 } 730 vm_object_reference(obj); 731 732 /* check that we are given the whole vm_object ? */ 733 vm_map_lookup_done(map, entry); 734 735 e->obj = obj; 736 /* Wire the memory and add the vm_object to the kernel map, 737 * to make sure that it is not freed even if all the processes 738 * that are mmap()ing should munmap() it. 739 */ 740 e->kva = vm_map_min(kernel_map); 741 e->size = obj->size << PAGE_SHIFT; 742 rv = vm_map_find(kernel_map, obj, 0, &e->kva, e->size, 0, 743 VMFS_OPTIMAL_SPACE, VM_PROT_READ | VM_PROT_WRITE, 744 VM_PROT_READ | VM_PROT_WRITE, 0); 745 if (rv != KERN_SUCCESS) { 746 nm_prerr("vm_map_find(%zx) failed", (size_t)e->size); 747 error = vm_mmap_to_errno(rv); 748 goto out_rel; 749 } 750 rv = vm_map_wire(kernel_map, e->kva, e->kva + e->size, 751 VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES); 752 if (rv != KERN_SUCCESS) { 753 nm_prerr("vm_map_wire failed"); 754 error = vm_mmap_to_errno(rv); 755 goto out_rem; 756 } 757 758 e->scan = e->kva; 759 760 return e; 761 762out_rem: 763 vm_map_remove(kernel_map, e->kva, e->kva + e->size); 764out_rel: 765 vm_object_deallocate(e->obj); 766 e->obj = NULL; 767out_free: 768 nm_os_free(e); 769out: 770 if (perror) 771 *perror = error; 772 return NULL; 773} 774#endif /* WITH_EXTMEM */ 775 776/* ================== PTNETMAP GUEST SUPPORT ==================== */ 777 778#ifdef WITH_PTNETMAP 779#include <sys/bus.h> 780#include <sys/rman.h> 781#include <machine/bus.h> /* bus_dmamap_* */ 782#include <machine/resource.h> 783#include <dev/pci/pcivar.h> 784#include <dev/pci/pcireg.h> 785/* 786 * ptnetmap memory device (memdev) for freebsd guest, 787 * ssed to expose host netmap memory to the guest through a PCI BAR. 788 */ 789 790/* 791 * ptnetmap memdev private data structure 792 */ 793struct ptnetmap_memdev { 794 device_t dev; 795 struct resource *pci_io; 796 struct resource *pci_mem; 797 struct netmap_mem_d *nm_mem; 798}; 799 800static int ptn_memdev_probe(device_t); 801static int ptn_memdev_attach(device_t); 802static int ptn_memdev_detach(device_t); 803static int ptn_memdev_shutdown(device_t); 804 805static device_method_t ptn_memdev_methods[] = { 806 DEVMETHOD(device_probe, ptn_memdev_probe), 807 DEVMETHOD(device_attach, ptn_memdev_attach), 808 DEVMETHOD(device_detach, ptn_memdev_detach), 809 DEVMETHOD(device_shutdown, ptn_memdev_shutdown), 810 DEVMETHOD_END 811}; 812 813static driver_t ptn_memdev_driver = { 814 PTNETMAP_MEMDEV_NAME, 815 ptn_memdev_methods, 816 sizeof(struct ptnetmap_memdev), 817}; 818 819/* We use (SI_ORDER_MIDDLE+1) here, see DEV_MODULE_ORDERED() invocation 820 * below. */ 821static devclass_t ptnetmap_devclass; 822DRIVER_MODULE_ORDERED(ptn_memdev, pci, ptn_memdev_driver, ptnetmap_devclass, 823 NULL, NULL, SI_ORDER_MIDDLE + 1); 824 825/* 826 * Map host netmap memory through PCI-BAR in the guest OS, 827 * returning physical (nm_paddr) and virtual (nm_addr) addresses 828 * of the netmap memory mapped in the guest. 829 */ 830int 831nm_os_pt_memdev_iomap(struct ptnetmap_memdev *ptn_dev, vm_paddr_t *nm_paddr, 832 void **nm_addr, uint64_t *mem_size) 833{ 834 int rid; 835 836 nm_prinf("ptn_memdev_driver iomap"); 837 838 rid = PCIR_BAR(PTNETMAP_MEM_PCI_BAR); 839 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_HI); 840 *mem_size = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMSIZE_LO) | 841 (*mem_size << 32); 842 843 /* map memory allocator */ 844 ptn_dev->pci_mem = bus_alloc_resource(ptn_dev->dev, SYS_RES_MEMORY, 845 &rid, 0, ~0, *mem_size, RF_ACTIVE); 846 if (ptn_dev->pci_mem == NULL) { 847 *nm_paddr = 0; 848 *nm_addr = NULL; 849 return ENOMEM; 850 } 851 852 *nm_paddr = rman_get_start(ptn_dev->pci_mem); 853 *nm_addr = rman_get_virtual(ptn_dev->pci_mem); 854 855 nm_prinf("=== BAR %d start %lx len %lx mem_size %lx ===", 856 PTNETMAP_MEM_PCI_BAR, 857 (unsigned long)(*nm_paddr), 858 (unsigned long)rman_get_size(ptn_dev->pci_mem), 859 (unsigned long)*mem_size); 860 return (0); 861} 862 863uint32_t 864nm_os_pt_memdev_ioread(struct ptnetmap_memdev *ptn_dev, unsigned int reg) 865{ 866 return bus_read_4(ptn_dev->pci_io, reg); 867} 868 869/* Unmap host netmap memory. */ 870void 871nm_os_pt_memdev_iounmap(struct ptnetmap_memdev *ptn_dev) 872{ 873 nm_prinf("ptn_memdev_driver iounmap"); 874 875 if (ptn_dev->pci_mem) { 876 bus_release_resource(ptn_dev->dev, SYS_RES_MEMORY, 877 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 878 ptn_dev->pci_mem = NULL; 879 } 880} 881 882/* Device identification routine, return BUS_PROBE_DEFAULT on success, 883 * positive on failure */ 884static int 885ptn_memdev_probe(device_t dev) 886{ 887 char desc[256]; 888 889 if (pci_get_vendor(dev) != PTNETMAP_PCI_VENDOR_ID) 890 return (ENXIO); 891 if (pci_get_device(dev) != PTNETMAP_PCI_DEVICE_ID) 892 return (ENXIO); 893 894 snprintf(desc, sizeof(desc), "%s PCI adapter", 895 PTNETMAP_MEMDEV_NAME); 896 device_set_desc_copy(dev, desc); 897 898 return (BUS_PROBE_DEFAULT); 899} 900 901/* Device initialization routine. */ 902static int 903ptn_memdev_attach(device_t dev) 904{ 905 struct ptnetmap_memdev *ptn_dev; 906 int rid; 907 uint16_t mem_id; 908 909 ptn_dev = device_get_softc(dev); 910 ptn_dev->dev = dev; 911 912 pci_enable_busmaster(dev); 913 914 rid = PCIR_BAR(PTNETMAP_IO_PCI_BAR); 915 ptn_dev->pci_io = bus_alloc_resource_any(dev, SYS_RES_IOPORT, &rid, 916 RF_ACTIVE); 917 if (ptn_dev->pci_io == NULL) { 918 device_printf(dev, "cannot map I/O space\n"); 919 return (ENXIO); 920 } 921 922 mem_id = bus_read_4(ptn_dev->pci_io, PTNET_MDEV_IO_MEMID); 923 924 /* create guest allocator */ 925 ptn_dev->nm_mem = netmap_mem_pt_guest_attach(ptn_dev, mem_id); 926 if (ptn_dev->nm_mem == NULL) { 927 ptn_memdev_detach(dev); 928 return (ENOMEM); 929 } 930 netmap_mem_get(ptn_dev->nm_mem); 931 932 nm_prinf("ptnetmap memdev attached, host memid: %u", mem_id); 933 934 return (0); 935} 936 937/* Device removal routine. */ 938static int 939ptn_memdev_detach(device_t dev) 940{ 941 struct ptnetmap_memdev *ptn_dev; 942 943 ptn_dev = device_get_softc(dev); 944 945 if (ptn_dev->nm_mem) { 946 nm_prinf("ptnetmap memdev detached, host memid %u", 947 netmap_mem_get_id(ptn_dev->nm_mem)); 948 netmap_mem_put(ptn_dev->nm_mem); 949 ptn_dev->nm_mem = NULL; 950 } 951 if (ptn_dev->pci_mem) { 952 bus_release_resource(dev, SYS_RES_MEMORY, 953 PCIR_BAR(PTNETMAP_MEM_PCI_BAR), ptn_dev->pci_mem); 954 ptn_dev->pci_mem = NULL; 955 } 956 if (ptn_dev->pci_io) { 957 bus_release_resource(dev, SYS_RES_IOPORT, 958 PCIR_BAR(PTNETMAP_IO_PCI_BAR), ptn_dev->pci_io); 959 ptn_dev->pci_io = NULL; 960 } 961 962 return (0); 963} 964 965static int 966ptn_memdev_shutdown(device_t dev) 967{ 968 return bus_generic_shutdown(dev); 969} 970 971#endif /* WITH_PTNETMAP */ 972 973/* 974 * In order to track whether pages are still mapped, we hook into 975 * the standard cdev_pager and intercept the constructor and 976 * destructor. 977 */ 978 979struct netmap_vm_handle_t { 980 struct cdev *dev; 981 struct netmap_priv_d *priv; 982}; 983 984 985static int 986netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 987 vm_ooffset_t foff, struct ucred *cred, u_short *color) 988{ 989 struct netmap_vm_handle_t *vmh = handle; 990 991 if (netmap_verbose) 992 nm_prinf("handle %p size %jd prot %d foff %jd", 993 handle, (intmax_t)size, prot, (intmax_t)foff); 994 if (color) 995 *color = 0; 996 dev_ref(vmh->dev); 997 return 0; 998} 999 1000 1001static void 1002netmap_dev_pager_dtor(void *handle) 1003{ 1004 struct netmap_vm_handle_t *vmh = handle; 1005 struct cdev *dev = vmh->dev; 1006 struct netmap_priv_d *priv = vmh->priv; 1007 1008 if (netmap_verbose) 1009 nm_prinf("handle %p", handle); 1010 netmap_dtor(priv); 1011 free(vmh, M_DEVBUF); 1012 dev_rel(dev); 1013} 1014 1015 1016static int 1017netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, 1018 int prot, vm_page_t *mres) 1019{ 1020 struct netmap_vm_handle_t *vmh = object->handle; 1021 struct netmap_priv_d *priv = vmh->priv; 1022 struct netmap_adapter *na = priv->np_na; 1023 vm_paddr_t paddr; 1024 vm_page_t page; 1025 vm_memattr_t memattr; 1026 vm_pindex_t pidx; 1027 1028 nm_prdis("object %p offset %jd prot %d mres %p", 1029 object, (intmax_t)offset, prot, mres); 1030 memattr = object->memattr; 1031 pidx = OFF_TO_IDX(offset); 1032 paddr = netmap_mem_ofstophys(na->nm_mem, offset); 1033 if (paddr == 0) 1034 return VM_PAGER_FAIL; 1035 1036 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 1037 /* 1038 * If the passed in result page is a fake page, update it with 1039 * the new physical address. 1040 */ 1041 page = *mres; 1042 vm_page_updatefake(page, paddr, memattr); 1043 } else { 1044 /* 1045 * Replace the passed in reqpage page with our own fake page and 1046 * free up the all of the original pages. 1047 */ 1048#ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */ 1049#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK 1050#define VM_OBJECT_WLOCK VM_OBJECT_LOCK 1051#endif /* VM_OBJECT_WUNLOCK */ 1052 1053 VM_OBJECT_WUNLOCK(object); 1054 page = vm_page_getfake(paddr, memattr); 1055 VM_OBJECT_WLOCK(object); 1056 vm_page_lock(*mres); 1057 vm_page_free(*mres); 1058 vm_page_unlock(*mres); 1059 *mres = page; 1060 vm_page_insert(page, object, pidx); 1061 } 1062 page->valid = VM_PAGE_BITS_ALL; 1063 return (VM_PAGER_OK); 1064} 1065 1066 1067static struct cdev_pager_ops netmap_cdev_pager_ops = { 1068 .cdev_pg_ctor = netmap_dev_pager_ctor, 1069 .cdev_pg_dtor = netmap_dev_pager_dtor, 1070 .cdev_pg_fault = netmap_dev_pager_fault, 1071}; 1072 1073 1074static int 1075netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff, 1076 vm_size_t objsize, vm_object_t *objp, int prot) 1077{ 1078 int error; 1079 struct netmap_vm_handle_t *vmh; 1080 struct netmap_priv_d *priv; 1081 vm_object_t obj; 1082 1083 if (netmap_verbose) 1084 nm_prinf("cdev %p foff %jd size %jd objp %p prot %d", cdev, 1085 (intmax_t )*foff, (intmax_t )objsize, objp, prot); 1086 1087 vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF, 1088 M_NOWAIT | M_ZERO); 1089 if (vmh == NULL) 1090 return ENOMEM; 1091 vmh->dev = cdev; 1092 1093 NMG_LOCK(); 1094 error = devfs_get_cdevpriv((void**)&priv); 1095 if (error) 1096 goto err_unlock; 1097 if (priv->np_nifp == NULL) { 1098 error = EINVAL; 1099 goto err_unlock; 1100 } 1101 vmh->priv = priv; 1102 priv->np_refs++; 1103 NMG_UNLOCK(); 1104 1105 obj = cdev_pager_allocate(vmh, OBJT_DEVICE, 1106 &netmap_cdev_pager_ops, objsize, prot, 1107 *foff, NULL); 1108 if (obj == NULL) { 1109 nm_prerr("cdev_pager_allocate failed"); 1110 error = EINVAL; 1111 goto err_deref; 1112 } 1113 1114 *objp = obj; 1115 return 0; 1116 1117err_deref: 1118 NMG_LOCK(); 1119 priv->np_refs--; 1120err_unlock: 1121 NMG_UNLOCK(); 1122// err: 1123 free(vmh, M_DEVBUF); 1124 return error; 1125} 1126 1127/* 1128 * On FreeBSD the close routine is only called on the last close on 1129 * the device (/dev/netmap) so we cannot do anything useful. 1130 * To track close() on individual file descriptors we pass netmap_dtor() to 1131 * devfs_set_cdevpriv() on open(). The FreeBSD kernel will call the destructor 1132 * when the last fd pointing to the device is closed. 1133 * 1134 * Note that FreeBSD does not even munmap() on close() so we also have 1135 * to track mmap() ourselves, and postpone the call to 1136 * netmap_dtor() is called when the process has no open fds and no active 1137 * memory maps on /dev/netmap, as in linux. 1138 */ 1139static int 1140netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 1141{ 1142 if (netmap_verbose) 1143 nm_prinf("dev %p fflag 0x%x devtype %d td %p", 1144 dev, fflag, devtype, td); 1145 return 0; 1146} 1147 1148 1149static int 1150netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 1151{ 1152 struct netmap_priv_d *priv; 1153 int error; 1154 1155 (void)dev; 1156 (void)oflags; 1157 (void)devtype; 1158 (void)td; 1159 1160 NMG_LOCK(); 1161 priv = netmap_priv_new(); 1162 if (priv == NULL) { 1163 error = ENOMEM; 1164 goto out; 1165 } 1166 error = devfs_set_cdevpriv(priv, netmap_dtor); 1167 if (error) { 1168 netmap_priv_delete(priv); 1169 } 1170out: 1171 NMG_UNLOCK(); 1172 return error; 1173} 1174 1175/******************** kthread wrapper ****************/ 1176#include <sys/sysproto.h> 1177u_int 1178nm_os_ncpus(void) 1179{ 1180 return mp_maxid + 1; 1181} 1182 1183struct nm_kctx_ctx { 1184 /* Userspace thread (kthread creator). */ 1185 struct thread *user_td; 1186 1187 /* worker function and parameter */ 1188 nm_kctx_worker_fn_t worker_fn; 1189 void *worker_private; 1190 1191 struct nm_kctx *nmk; 1192 1193 /* integer to manage multiple worker contexts (e.g., RX or TX on ptnetmap) */ 1194 long type; 1195}; 1196 1197struct nm_kctx { 1198 struct thread *worker; 1199 struct mtx worker_lock; 1200 struct nm_kctx_ctx worker_ctx; 1201 int run; /* used to stop kthread */ 1202 int attach_user; /* kthread attached to user_process */ 1203 int affinity; 1204}; 1205 1206static void 1207nm_kctx_worker(void *data) 1208{ 1209 struct nm_kctx *nmk = data; 1210 struct nm_kctx_ctx *ctx = &nmk->worker_ctx; 1211 1212 if (nmk->affinity >= 0) { 1213 thread_lock(curthread); 1214 sched_bind(curthread, nmk->affinity); 1215 thread_unlock(curthread); 1216 } 1217 1218 while (nmk->run) { 1219 /* 1220 * check if the parent process dies 1221 * (when kthread is attached to user process) 1222 */ 1223 if (ctx->user_td) { 1224 PROC_LOCK(curproc); 1225 thread_suspend_check(0); 1226 PROC_UNLOCK(curproc); 1227 } else { 1228 kthread_suspend_check(); 1229 } 1230 1231 /* Continuously execute worker process. */ 1232 ctx->worker_fn(ctx->worker_private); /* worker body */ 1233 } 1234 1235 kthread_exit(); 1236} 1237 1238void 1239nm_os_kctx_worker_setaff(struct nm_kctx *nmk, int affinity) 1240{ 1241 nmk->affinity = affinity; 1242} 1243 1244struct nm_kctx * 1245nm_os_kctx_create(struct nm_kctx_cfg *cfg, void *opaque) 1246{ 1247 struct nm_kctx *nmk = NULL; 1248 1249 nmk = malloc(sizeof(*nmk), M_DEVBUF, M_NOWAIT | M_ZERO); 1250 if (!nmk) 1251 return NULL; 1252 1253 mtx_init(&nmk->worker_lock, "nm_kthread lock", NULL, MTX_DEF); 1254 nmk->worker_ctx.worker_fn = cfg->worker_fn; 1255 nmk->worker_ctx.worker_private = cfg->worker_private; 1256 nmk->worker_ctx.type = cfg->type; 1257 nmk->affinity = -1; 1258 1259 /* attach kthread to user process (ptnetmap) */ 1260 nmk->attach_user = cfg->attach_user; 1261 1262 return nmk; 1263} 1264 1265int 1266nm_os_kctx_worker_start(struct nm_kctx *nmk) 1267{ 1268 struct proc *p = NULL; 1269 int error = 0; 1270 1271 /* Temporarily disable this function as it is currently broken 1272 * and causes kernel crashes. The failure can be triggered by 1273 * the "vale_polling_enable_disable" test in ctrl-api-test.c. */ 1274 return EOPNOTSUPP; 1275 1276 if (nmk->worker) 1277 return EBUSY; 1278 1279 /* check if we want to attach kthread to user process */ 1280 if (nmk->attach_user) { 1281 nmk->worker_ctx.user_td = curthread; 1282 p = curthread->td_proc; 1283 } 1284 1285 /* enable kthread main loop */ 1286 nmk->run = 1; 1287 /* create kthread */ 1288 if((error = kthread_add(nm_kctx_worker, nmk, p, 1289 &nmk->worker, RFNOWAIT /* to be checked */, 0, "nm-kthread-%ld", 1290 nmk->worker_ctx.type))) { 1291 goto err; 1292 } 1293 1294 nm_prinf("nm_kthread started td %p", nmk->worker); 1295 1296 return 0; 1297err: 1298 nm_prerr("nm_kthread start failed err %d", error); 1299 nmk->worker = NULL; 1300 return error; 1301} 1302 1303void 1304nm_os_kctx_worker_stop(struct nm_kctx *nmk) 1305{ 1306 if (!nmk->worker) 1307 return; 1308 1309 /* tell to kthread to exit from main loop */ 1310 nmk->run = 0; 1311 1312 /* wake up kthread if it sleeps */ 1313 kthread_resume(nmk->worker); 1314 1315 nmk->worker = NULL; 1316} 1317 1318void 1319nm_os_kctx_destroy(struct nm_kctx *nmk) 1320{ 1321 if (!nmk) 1322 return; 1323 1324 if (nmk->worker) 1325 nm_os_kctx_worker_stop(nmk); 1326 1327 free(nmk, M_DEVBUF); 1328} 1329 1330/******************** kqueue support ****************/ 1331 1332/* 1333 * In addition to calling selwakeuppri(), nm_os_selwakeup() also 1334 * needs to call knote() to wake up kqueue listeners. 1335 * This operation is deferred to a taskqueue in order to avoid possible 1336 * lock order reversals; these may happen because knote() grabs a 1337 * private lock associated to the 'si' (see struct selinfo, 1338 * struct nm_selinfo, and nm_os_selinfo_init), and nm_os_selwakeup() 1339 * can be called while holding the lock associated to a different 1340 * 'si'. 1341 * When calling knote() we use a non-zero 'hint' argument to inform 1342 * the netmap_knrw() function that it is being called from 1343 * 'nm_os_selwakeup'; this is necessary because when netmap_knrw() is 1344 * called by the kevent subsystem (i.e. kevent_scan()) we also need to 1345 * call netmap_poll(). 1346 * 1347 * The netmap_kqfilter() function registers one or another f_event 1348 * depending on read or write mode. A pointer to the struct 1349 * 'netmap_priv_d' is stored into kn->kn_hook, so that it can later 1350 * be passed to netmap_poll(). We pass NULL as a third argument to 1351 * netmap_poll(), so that the latter only runs the txsync/rxsync 1352 * (if necessary), and skips the nm_os_selrecord() calls. 1353 */ 1354 1355 1356void 1357nm_os_selwakeup(struct nm_selinfo *si) 1358{ 1359 selwakeuppri(&si->si, PI_NET); 1360 if (si->kqueue_users > 0) { 1361 taskqueue_enqueue(si->ntfytq, &si->ntfytask); 1362 } 1363} 1364 1365void 1366nm_os_selrecord(struct thread *td, struct nm_selinfo *si) 1367{ 1368 selrecord(td, &si->si); 1369} 1370 1371static void 1372netmap_knrdetach(struct knote *kn) 1373{ 1374 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1375 struct nm_selinfo *si = priv->np_si[NR_RX]; 1376 1377 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1378 NMG_LOCK(); 1379 KASSERT(si->kqueue_users > 0, ("kqueue_user underflow on %s", 1380 si->mtxname)); 1381 si->kqueue_users--; 1382 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1383 NMG_UNLOCK(); 1384} 1385 1386static void 1387netmap_knwdetach(struct knote *kn) 1388{ 1389 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 1390 struct nm_selinfo *si = priv->np_si[NR_TX]; 1391 1392 knlist_remove(&si->si.si_note, kn, /*islocked=*/0); 1393 NMG_LOCK(); 1394 si->kqueue_users--; 1395 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1396 NMG_UNLOCK(); 1397} 1398 1399/* 1400 * Callback triggered by netmap notifications (see netmap_notify()), 1401 * and by the application calling kevent(). In the former case we 1402 * just return 1 (events ready), since we are not able to do better. 1403 * In the latter case we use netmap_poll() to see which events are 1404 * ready. 1405 */ 1406static int 1407netmap_knrw(struct knote *kn, long hint, int events) 1408{ 1409 struct netmap_priv_d *priv; 1410 int revents; 1411 1412 if (hint != 0) { 1413 /* Called from netmap_notify(), typically from a 1414 * thread different from the one issuing kevent(). 1415 * Assume we are ready. */ 1416 return 1; 1417 } 1418 1419 /* Called from kevent(). */ 1420 priv = kn->kn_hook; 1421 revents = netmap_poll(priv, events, /*thread=*/NULL); 1422 1423 return (events & revents) ? 1 : 0; 1424} 1425 1426static int 1427netmap_knread(struct knote *kn, long hint) 1428{ 1429 return netmap_knrw(kn, hint, POLLIN); 1430} 1431 1432static int 1433netmap_knwrite(struct knote *kn, long hint) 1434{ 1435 return netmap_knrw(kn, hint, POLLOUT); 1436} 1437 1438static struct filterops netmap_rfiltops = { 1439 .f_isfd = 1, 1440 .f_detach = netmap_knrdetach, 1441 .f_event = netmap_knread, 1442}; 1443 1444static struct filterops netmap_wfiltops = { 1445 .f_isfd = 1, 1446 .f_detach = netmap_knwdetach, 1447 .f_event = netmap_knwrite, 1448}; 1449 1450 1451/* 1452 * This is called when a thread invokes kevent() to record 1453 * a change in the configuration of the kqueue(). 1454 * The 'priv' is the one associated to the open netmap device. 1455 */ 1456static int 1457netmap_kqfilter(struct cdev *dev, struct knote *kn) 1458{ 1459 struct netmap_priv_d *priv; 1460 int error; 1461 struct netmap_adapter *na; 1462 struct nm_selinfo *si; 1463 int ev = kn->kn_filter; 1464 1465 if (ev != EVFILT_READ && ev != EVFILT_WRITE) { 1466 nm_prerr("bad filter request %d", ev); 1467 return 1; 1468 } 1469 error = devfs_get_cdevpriv((void**)&priv); 1470 if (error) { 1471 nm_prerr("device not yet setup"); 1472 return 1; 1473 } 1474 na = priv->np_na; 1475 if (na == NULL) { 1476 nm_prerr("no netmap adapter for this file descriptor"); 1477 return 1; 1478 } 1479 /* the si is indicated in the priv */ 1480 si = priv->np_si[(ev == EVFILT_WRITE) ? NR_TX : NR_RX]; 1481 kn->kn_fop = (ev == EVFILT_WRITE) ? 1482 &netmap_wfiltops : &netmap_rfiltops; 1483 kn->kn_hook = priv; 1484 NMG_LOCK(); 1485 si->kqueue_users++; 1486 nm_prinf("kqueue users for %s: %d", si->mtxname, si->kqueue_users); 1487 NMG_UNLOCK(); 1488 knlist_add(&si->si.si_note, kn, /*islocked=*/0); 1489 1490 return 0; 1491} 1492 1493static int 1494freebsd_netmap_poll(struct cdev *cdevi __unused, int events, struct thread *td) 1495{ 1496 struct netmap_priv_d *priv; 1497 if (devfs_get_cdevpriv((void **)&priv)) { 1498 return POLLERR; 1499 } 1500 return netmap_poll(priv, events, td); 1501} 1502 1503static int 1504freebsd_netmap_ioctl(struct cdev *dev __unused, u_long cmd, caddr_t data, 1505 int ffla __unused, struct thread *td) 1506{ 1507 int error; 1508 struct netmap_priv_d *priv; 1509 1510 CURVNET_SET(TD_TO_VNET(td)); 1511 error = devfs_get_cdevpriv((void **)&priv); 1512 if (error) { 1513 /* XXX ENOENT should be impossible, since the priv 1514 * is now created in the open */ 1515 if (error == ENOENT) 1516 error = ENXIO; 1517 goto out; 1518 } 1519 error = netmap_ioctl(priv, cmd, data, td, /*nr_body_is_user=*/1); 1520out: 1521 CURVNET_RESTORE(); 1522 1523 return error; 1524} 1525 1526void 1527nm_os_onattach(struct ifnet *ifp) 1528{ 1529 ifp->if_capabilities |= IFCAP_NETMAP; 1530} 1531 1532void 1533nm_os_onenter(struct ifnet *ifp) 1534{ 1535 struct netmap_adapter *na = NA(ifp); 1536 1537 na->if_transmit = ifp->if_transmit; 1538 ifp->if_transmit = netmap_transmit; 1539 ifp->if_capenable |= IFCAP_NETMAP; 1540} 1541 1542void 1543nm_os_onexit(struct ifnet *ifp) 1544{ 1545 struct netmap_adapter *na = NA(ifp); 1546 1547 ifp->if_transmit = na->if_transmit; 1548 ifp->if_capenable &= ~IFCAP_NETMAP; 1549} 1550 1551extern struct cdevsw netmap_cdevsw; /* XXX used in netmap.c, should go elsewhere */ 1552struct cdevsw netmap_cdevsw = { 1553 .d_version = D_VERSION, 1554 .d_name = "netmap", 1555 .d_open = netmap_open, 1556 .d_mmap_single = netmap_mmap_single, 1557 .d_ioctl = freebsd_netmap_ioctl, 1558 .d_poll = freebsd_netmap_poll, 1559 .d_kqfilter = netmap_kqfilter, 1560 .d_close = netmap_close, 1561}; 1562/*--- end of kqueue support ----*/ 1563 1564/* 1565 * Kernel entry point. 1566 * 1567 * Initialize/finalize the module and return. 1568 * 1569 * Return 0 on success, errno on failure. 1570 */ 1571static int 1572netmap_loader(__unused struct module *module, int event, __unused void *arg) 1573{ 1574 int error = 0; 1575 1576 switch (event) { 1577 case MOD_LOAD: 1578 error = netmap_init(); 1579 break; 1580 1581 case MOD_UNLOAD: 1582 /* 1583 * if some one is still using netmap, 1584 * then the module can not be unloaded. 1585 */ 1586 if (netmap_use_count) { 1587 nm_prerr("netmap module can not be unloaded - netmap_use_count: %d", 1588 netmap_use_count); 1589 error = EBUSY; 1590 break; 1591 } 1592 netmap_fini(); 1593 break; 1594 1595 default: 1596 error = EOPNOTSUPP; 1597 break; 1598 } 1599 1600 return (error); 1601} 1602 1603#ifdef DEV_MODULE_ORDERED 1604/* 1605 * The netmap module contains three drivers: (i) the netmap character device 1606 * driver; (ii) the ptnetmap memdev PCI device driver, (iii) the ptnet PCI 1607 * device driver. The attach() routines of both (ii) and (iii) need the 1608 * lock of the global allocator, and such lock is initialized in netmap_init(), 1609 * which is part of (i). 1610 * Therefore, we make sure that (i) is loaded before (ii) and (iii), using 1611 * the 'order' parameter of driver declaration macros. For (i), we specify 1612 * SI_ORDER_MIDDLE, while higher orders are used with the DRIVER_MODULE_ORDERED 1613 * macros for (ii) and (iii). 1614 */ 1615DEV_MODULE_ORDERED(netmap, netmap_loader, NULL, SI_ORDER_MIDDLE); 1616#else /* !DEV_MODULE_ORDERED */ 1617DEV_MODULE(netmap, netmap_loader, NULL); 1618#endif /* DEV_MODULE_ORDERED */ 1619MODULE_DEPEND(netmap, pci, 1, 1, 1); 1620MODULE_VERSION(netmap, 1); 1621/* reduce conditional code */ 1622// linux API, use for the knlist in FreeBSD 1623/* use a private mutex for the knlist */ 1624