netmap_freebsd.c revision 341275
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 341275 2018-11-30 02:06:30Z dab $ */ 27#include "opt_inet.h" 28#include "opt_inet6.h" 29 30#include <sys/types.h> 31#include <sys/module.h> 32#include <sys/errno.h> 33#include <sys/param.h> /* defines used in kernel.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 */ 37#include <sys/endian.h> 38 39#include <sys/rwlock.h> 40 41#include <vm/vm.h> /* vtophys */ 42#include <vm/pmap.h> /* vtophys */ 43#include <vm/vm_param.h> 44#include <vm/vm_object.h> 45#include <vm/vm_page.h> 46#include <vm/vm_pager.h> 47#include <vm/uma.h> 48 49 50#include <sys/malloc.h> 51#include <sys/socket.h> /* sockaddrs */ 52#include <sys/selinfo.h> 53#include <net/if.h> 54#include <net/if_var.h> 55#include <net/if_types.h> /* IFT_ETHER */ 56#include <net/ethernet.h> /* ether_ifdetach */ 57#include <net/if_dl.h> /* LLADDR */ 58#include <machine/bus.h> /* bus_dmamap_* */ 59#include <netinet/in.h> /* in6_cksum_pseudo() */ 60#include <machine/in_cksum.h> /* in_pseudo(), in_cksum_hdr() */ 61 62#include <net/netmap.h> 63#include <dev/netmap/netmap_kern.h> 64#include <dev/netmap/netmap_mem2.h> 65 66 67/* ======================== FREEBSD-SPECIFIC ROUTINES ================== */ 68 69rawsum_t 70nm_csum_raw(uint8_t *data, size_t len, rawsum_t cur_sum) 71{ 72 /* TODO XXX please use the FreeBSD implementation for this. */ 73 uint16_t *words = (uint16_t *)data; 74 int nw = len / 2; 75 int i; 76 77 for (i = 0; i < nw; i++) 78 cur_sum += be16toh(words[i]); 79 80 if (len & 1) 81 cur_sum += (data[len-1] << 8); 82 83 return cur_sum; 84} 85 86/* Fold a raw checksum: 'cur_sum' is in host byte order, while the 87 * return value is in network byte order. 88 */ 89uint16_t 90nm_csum_fold(rawsum_t cur_sum) 91{ 92 /* TODO XXX please use the FreeBSD implementation for this. */ 93 while (cur_sum >> 16) 94 cur_sum = (cur_sum & 0xFFFF) + (cur_sum >> 16); 95 96 return htobe16((~cur_sum) & 0xFFFF); 97} 98 99uint16_t nm_csum_ipv4(struct nm_iphdr *iph) 100{ 101#if 0 102 return in_cksum_hdr((void *)iph); 103#else 104 return nm_csum_fold(nm_csum_raw((uint8_t*)iph, sizeof(struct nm_iphdr), 0)); 105#endif 106} 107 108void 109nm_csum_tcpudp_ipv4(struct nm_iphdr *iph, void *data, 110 size_t datalen, uint16_t *check) 111{ 112#ifdef INET 113 uint16_t pseudolen = datalen + iph->protocol; 114 115 /* Compute and insert the pseudo-header cheksum. */ 116 *check = in_pseudo(iph->saddr, iph->daddr, 117 htobe16(pseudolen)); 118 /* Compute the checksum on TCP/UDP header + payload 119 * (includes the pseudo-header). 120 */ 121 *check = nm_csum_fold(nm_csum_raw(data, datalen, 0)); 122#else 123 static int notsupported = 0; 124 if (!notsupported) { 125 notsupported = 1; 126 D("inet4 segmentation not supported"); 127 } 128#endif 129} 130 131void 132nm_csum_tcpudp_ipv6(struct nm_ipv6hdr *ip6h, void *data, 133 size_t datalen, uint16_t *check) 134{ 135#ifdef INET6 136 *check = in6_cksum_pseudo((void*)ip6h, datalen, ip6h->nexthdr, 0); 137 *check = nm_csum_fold(nm_csum_raw(data, datalen, 0)); 138#else 139 static int notsupported = 0; 140 if (!notsupported) { 141 notsupported = 1; 142 D("inet6 segmentation not supported"); 143 } 144#endif 145} 146 147 148/* 149 * Intercept the rx routine in the standard device driver. 150 * Second argument is non-zero to intercept, 0 to restore 151 */ 152int 153netmap_catch_rx(struct netmap_generic_adapter *gna, int intercept) 154{ 155 struct netmap_adapter *na = &gna->up.up; 156 struct ifnet *ifp = na->ifp; 157 158 if (intercept) { 159 if (gna->save_if_input) { 160 D("cannot intercept again"); 161 return EINVAL; /* already set */ 162 } 163 gna->save_if_input = ifp->if_input; 164 ifp->if_input = generic_rx_handler; 165 } else { 166 if (!gna->save_if_input){ 167 D("cannot restore"); 168 return EINVAL; /* not saved */ 169 } 170 ifp->if_input = gna->save_if_input; 171 gna->save_if_input = NULL; 172 } 173 174 return 0; 175} 176 177 178/* 179 * Intercept the packet steering routine in the tx path, 180 * so that we can decide which queue is used for an mbuf. 181 * Second argument is non-zero to intercept, 0 to restore. 182 * On freebsd we just intercept if_transmit. 183 */ 184void 185netmap_catch_tx(struct netmap_generic_adapter *gna, int enable) 186{ 187 struct netmap_adapter *na = &gna->up.up; 188 struct ifnet *ifp = netmap_generic_getifp(gna); 189 190 if (enable) { 191 na->if_transmit = ifp->if_transmit; 192 ifp->if_transmit = netmap_transmit; 193 } else { 194 ifp->if_transmit = na->if_transmit; 195 } 196} 197 198 199/* 200 * Transmit routine used by generic_netmap_txsync(). Returns 0 on success 201 * and non-zero on error (which may be packet drops or other errors). 202 * addr and len identify the netmap buffer, m is the (preallocated) 203 * mbuf to use for transmissions. 204 * 205 * We should add a reference to the mbuf so the m_freem() at the end 206 * of the transmission does not consume resources. 207 * 208 * On FreeBSD, and on multiqueue cards, we can force the queue using 209 * if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 210 * i = m->m_pkthdr.flowid % adapter->num_queues; 211 * else 212 * i = curcpu % adapter->num_queues; 213 * 214 */ 215int 216generic_xmit_frame(struct ifnet *ifp, struct mbuf *m, 217 void *addr, u_int len, u_int ring_nr) 218{ 219 int ret; 220 221 /* Link the external storage to the netmap buffer, so that 222 * no copy is necessary. */ 223 m->m_ext.ext_buf = m->m_data = addr; 224 m->m_ext.ext_size = len; 225 226 m->m_len = m->m_pkthdr.len = len; 227 228 /* mbuf refcnt is not contended, no need to use atomic 229 * (a memory barrier is enough). */ 230 SET_MBUF_REFCNT(m, 2); 231 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 232 m->m_pkthdr.flowid = ring_nr; 233 m->m_pkthdr.rcvif = ifp; /* used for tx notification */ 234 ret = NA(ifp)->if_transmit(ifp, m); 235 return ret; 236} 237 238 239#if __FreeBSD_version >= 1100005 240struct netmap_adapter * 241netmap_getna(if_t ifp) 242{ 243 return (NA((struct ifnet *)ifp)); 244} 245#endif /* __FreeBSD_version >= 1100005 */ 246 247/* 248 * The following two functions are empty until we have a generic 249 * way to extract the info from the ifp 250 */ 251int 252generic_find_num_desc(struct ifnet *ifp, unsigned int *tx, unsigned int *rx) 253{ 254 D("called, in tx %d rx %d", *tx, *rx); 255 return 0; 256} 257 258 259void 260generic_find_num_queues(struct ifnet *ifp, u_int *txq, u_int *rxq) 261{ 262 D("called, in txq %d rxq %d", *txq, *rxq); 263 *txq = netmap_generic_rings; 264 *rxq = netmap_generic_rings; 265} 266 267 268void 269netmap_mitigation_init(struct nm_generic_mit *mit, int idx, struct netmap_adapter *na) 270{ 271 ND("called"); 272 mit->mit_pending = 0; 273 mit->mit_ring_idx = idx; 274 mit->mit_na = na; 275} 276 277 278void 279netmap_mitigation_start(struct nm_generic_mit *mit) 280{ 281 ND("called"); 282} 283 284 285void 286netmap_mitigation_restart(struct nm_generic_mit *mit) 287{ 288 ND("called"); 289} 290 291 292int 293netmap_mitigation_active(struct nm_generic_mit *mit) 294{ 295 ND("called"); 296 return 0; 297} 298 299 300void 301netmap_mitigation_cleanup(struct nm_generic_mit *mit) 302{ 303 ND("called"); 304} 305 306static int 307nm_vi_dummy(struct ifnet *ifp, u_long cmd, caddr_t addr) 308{ 309 return EINVAL; 310} 311 312static void 313nm_vi_start(struct ifnet *ifp) 314{ 315 panic("nm_vi_start() must not be called"); 316} 317 318/* 319 * Index manager of persistent virtual interfaces. 320 * It is used to decide the lowest byte of the MAC address. 321 * We use the same algorithm with management of bridge port index. 322 */ 323#define NM_VI_MAX 255 324static struct { 325 uint8_t index[NM_VI_MAX]; /* XXX just for a reasonable number */ 326 uint8_t active; 327 struct mtx lock; 328} nm_vi_indices; 329 330void 331nm_vi_init_index(void) 332{ 333 int i; 334 for (i = 0; i < NM_VI_MAX; i++) 335 nm_vi_indices.index[i] = i; 336 nm_vi_indices.active = 0; 337 mtx_init(&nm_vi_indices.lock, "nm_vi_indices_lock", NULL, MTX_DEF); 338} 339 340/* return -1 if no index available */ 341static int 342nm_vi_get_index(void) 343{ 344 int ret; 345 346 mtx_lock(&nm_vi_indices.lock); 347 ret = nm_vi_indices.active == NM_VI_MAX ? -1 : 348 nm_vi_indices.index[nm_vi_indices.active++]; 349 mtx_unlock(&nm_vi_indices.lock); 350 return ret; 351} 352 353static void 354nm_vi_free_index(uint8_t val) 355{ 356 int i, lim; 357 358 mtx_lock(&nm_vi_indices.lock); 359 lim = nm_vi_indices.active; 360 for (i = 0; i < lim; i++) { 361 if (nm_vi_indices.index[i] == val) { 362 /* swap index[lim-1] and j */ 363 int tmp = nm_vi_indices.index[lim-1]; 364 nm_vi_indices.index[lim-1] = val; 365 nm_vi_indices.index[i] = tmp; 366 nm_vi_indices.active--; 367 break; 368 } 369 } 370 if (lim == nm_vi_indices.active) 371 D("funny, index %u didn't found", val); 372 mtx_unlock(&nm_vi_indices.lock); 373} 374#undef NM_VI_MAX 375 376/* 377 * Implementation of a netmap-capable virtual interface that 378 * registered to the system. 379 * It is based on if_tap.c and ip_fw_log.c in FreeBSD 9. 380 * 381 * Note: Linux sets refcount to 0 on allocation of net_device, 382 * then increments it on registration to the system. 383 * FreeBSD sets refcount to 1 on if_alloc(), and does not 384 * increment this refcount on if_attach(). 385 */ 386int 387nm_vi_persist(const char *name, struct ifnet **ret) 388{ 389 struct ifnet *ifp; 390 u_short macaddr_hi; 391 uint32_t macaddr_mid; 392 u_char eaddr[6]; 393 int unit = nm_vi_get_index(); /* just to decide MAC address */ 394 395 if (unit < 0) 396 return EBUSY; 397 /* 398 * We use the same MAC address generation method with tap 399 * except for the highest octet is 00:be instead of 00:bd 400 */ 401 macaddr_hi = htons(0x00be); /* XXX tap + 1 */ 402 macaddr_mid = (uint32_t) ticks; 403 bcopy(&macaddr_hi, eaddr, sizeof(short)); 404 bcopy(&macaddr_mid, &eaddr[2], sizeof(uint32_t)); 405 eaddr[5] = (uint8_t)unit; 406 407 ifp = if_alloc(IFT_ETHER); 408 if (ifp == NULL) { 409 D("if_alloc failed"); 410 return ENOMEM; 411 } 412 if_initname(ifp, name, IF_DUNIT_NONE); 413 ifp->if_mtu = 65536; 414 ifp->if_flags = IFF_UP | IFF_SIMPLEX | IFF_MULTICAST; 415 ifp->if_init = (void *)nm_vi_dummy; 416 ifp->if_ioctl = nm_vi_dummy; 417 ifp->if_start = nm_vi_start; 418 ifp->if_mtu = ETHERMTU; 419 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 420 ifp->if_capabilities |= IFCAP_LINKSTATE; 421 ifp->if_capenable |= IFCAP_LINKSTATE; 422 423 ether_ifattach(ifp, eaddr); 424 *ret = ifp; 425 return 0; 426} 427/* unregister from the system and drop the final refcount */ 428void 429nm_vi_detach(struct ifnet *ifp) 430{ 431 nm_vi_free_index(((char *)IF_LLADDR(ifp))[5]); 432 ether_ifdetach(ifp); 433 if_free(ifp); 434} 435 436/* 437 * In order to track whether pages are still mapped, we hook into 438 * the standard cdev_pager and intercept the constructor and 439 * destructor. 440 */ 441 442struct netmap_vm_handle_t { 443 struct cdev *dev; 444 struct netmap_priv_d *priv; 445}; 446 447 448static int 449netmap_dev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 450 vm_ooffset_t foff, struct ucred *cred, u_short *color) 451{ 452 struct netmap_vm_handle_t *vmh = handle; 453 454 if (netmap_verbose) 455 D("handle %p size %jd prot %d foff %jd", 456 handle, (intmax_t)size, prot, (intmax_t)foff); 457 if (color) 458 *color = 0; 459 dev_ref(vmh->dev); 460 return 0; 461} 462 463 464static void 465netmap_dev_pager_dtor(void *handle) 466{ 467 struct netmap_vm_handle_t *vmh = handle; 468 struct cdev *dev = vmh->dev; 469 struct netmap_priv_d *priv = vmh->priv; 470 471 if (netmap_verbose) 472 D("handle %p", handle); 473 netmap_dtor(priv); 474 free(vmh, M_DEVBUF); 475 dev_rel(dev); 476} 477 478 479static int 480netmap_dev_pager_fault(vm_object_t object, vm_ooffset_t offset, 481 int prot, vm_page_t *mres) 482{ 483 struct netmap_vm_handle_t *vmh = object->handle; 484 struct netmap_priv_d *priv = vmh->priv; 485 struct netmap_adapter *na = priv->np_na; 486 vm_paddr_t paddr; 487 vm_page_t page; 488 vm_memattr_t memattr; 489 vm_pindex_t pidx; 490 491 ND("object %p offset %jd prot %d mres %p", 492 object, (intmax_t)offset, prot, mres); 493 memattr = object->memattr; 494 pidx = OFF_TO_IDX(offset); 495 paddr = netmap_mem_ofstophys(na->nm_mem, offset); 496 if (paddr == 0) 497 return VM_PAGER_FAIL; 498 499 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 500 /* 501 * If the passed in result page is a fake page, update it with 502 * the new physical address. 503 */ 504 page = *mres; 505 vm_page_updatefake(page, paddr, memattr); 506 } else { 507 /* 508 * Replace the passed in reqpage page with our own fake page and 509 * free up the all of the original pages. 510 */ 511#ifndef VM_OBJECT_WUNLOCK /* FreeBSD < 10.x */ 512#define VM_OBJECT_WUNLOCK VM_OBJECT_UNLOCK 513#define VM_OBJECT_WLOCK VM_OBJECT_LOCK 514#endif /* VM_OBJECT_WUNLOCK */ 515 516 VM_OBJECT_WUNLOCK(object); 517 page = vm_page_getfake(paddr, memattr); 518 VM_OBJECT_WLOCK(object); 519 vm_page_lock(*mres); 520 vm_page_free(*mres); 521 vm_page_unlock(*mres); 522 *mres = page; 523 vm_page_insert(page, object, pidx); 524 } 525 page->valid = VM_PAGE_BITS_ALL; 526 return (VM_PAGER_OK); 527} 528 529 530static struct cdev_pager_ops netmap_cdev_pager_ops = { 531 .cdev_pg_ctor = netmap_dev_pager_ctor, 532 .cdev_pg_dtor = netmap_dev_pager_dtor, 533 .cdev_pg_fault = netmap_dev_pager_fault, 534}; 535 536 537static int 538netmap_mmap_single(struct cdev *cdev, vm_ooffset_t *foff, 539 vm_size_t objsize, vm_object_t *objp, int prot) 540{ 541 int error; 542 struct netmap_vm_handle_t *vmh; 543 struct netmap_priv_d *priv; 544 vm_object_t obj; 545 546 if (netmap_verbose) 547 D("cdev %p foff %jd size %jd objp %p prot %d", cdev, 548 (intmax_t )*foff, (intmax_t )objsize, objp, prot); 549 550 vmh = malloc(sizeof(struct netmap_vm_handle_t), M_DEVBUF, 551 M_NOWAIT | M_ZERO); 552 if (vmh == NULL) 553 return ENOMEM; 554 vmh->dev = cdev; 555 556 NMG_LOCK(); 557 error = devfs_get_cdevpriv((void**)&priv); 558 if (error) 559 goto err_unlock; 560 if (priv->np_nifp == NULL) { 561 error = EINVAL; 562 goto err_unlock; 563 } 564 vmh->priv = priv; 565 priv->np_refs++; 566 NMG_UNLOCK(); 567 568 obj = cdev_pager_allocate(vmh, OBJT_DEVICE, 569 &netmap_cdev_pager_ops, objsize, prot, 570 *foff, NULL); 571 if (obj == NULL) { 572 D("cdev_pager_allocate failed"); 573 error = EINVAL; 574 goto err_deref; 575 } 576 577 *objp = obj; 578 return 0; 579 580err_deref: 581 NMG_LOCK(); 582 priv->np_refs--; 583err_unlock: 584 NMG_UNLOCK(); 585// err: 586 free(vmh, M_DEVBUF); 587 return error; 588} 589 590/* 591 * On FreeBSD the close routine is only called on the last close on 592 * the device (/dev/netmap) so we cannot do anything useful. 593 * To track close() on individual file descriptors we pass netmap_dtor() to 594 * devfs_set_cdevpriv() on open(). The FreeBSD kernel will call the destructor 595 * when the last fd pointing to the device is closed. 596 * 597 * Note that FreeBSD does not even munmap() on close() so we also have 598 * to track mmap() ourselves, and postpone the call to 599 * netmap_dtor() is called when the process has no open fds and no active 600 * memory maps on /dev/netmap, as in linux. 601 */ 602static int 603netmap_close(struct cdev *dev, int fflag, int devtype, struct thread *td) 604{ 605 if (netmap_verbose) 606 D("dev %p fflag 0x%x devtype %d td %p", 607 dev, fflag, devtype, td); 608 return 0; 609} 610 611 612static int 613netmap_open(struct cdev *dev, int oflags, int devtype, struct thread *td) 614{ 615 struct netmap_priv_d *priv; 616 int error; 617 618 (void)dev; 619 (void)oflags; 620 (void)devtype; 621 (void)td; 622 623 priv = malloc(sizeof(struct netmap_priv_d), M_DEVBUF, 624 M_NOWAIT | M_ZERO); 625 if (priv == NULL) 626 return ENOMEM; 627 priv->np_refs = 1; 628 error = devfs_set_cdevpriv(priv, netmap_dtor); 629 if (error) { 630 free(priv, M_DEVBUF); 631 } else { 632 NMG_LOCK(); 633 netmap_use_count++; 634 NMG_UNLOCK(); 635 } 636 return error; 637} 638 639/******************** kqueue support ****************/ 640 641/* 642 * The OS_selwakeup also needs to issue a KNOTE_UNLOCKED. 643 * We use a non-zero argument to distinguish the call from the one 644 * in kevent_scan() which instead also needs to run netmap_poll(). 645 * The knote uses a global mutex for the time being. We might 646 * try to reuse the one in the si, but it is not allocated 647 * permanently so it might be a bit tricky. 648 * 649 * The *kqfilter function registers one or another f_event 650 * depending on read or write mode. 651 * In the call to f_event() td_fpop is NULL so any child function 652 * calling devfs_get_cdevpriv() would fail - and we need it in 653 * netmap_poll(). As a workaround we store priv into kn->kn_hook 654 * and pass it as first argument to netmap_poll(), which then 655 * uses the failure to tell that we are called from f_event() 656 * and do not need the selrecord(). 657 */ 658 659 660void 661freebsd_selwakeup(struct nm_selinfo *si, int pri) 662{ 663 if (netmap_verbose) 664 D("on knote %p", &si->si.si_note); 665 selwakeuppri(&si->si, pri); 666 /* use a non-zero hint to tell the notification from the 667 * call done in kqueue_scan() which uses 0 668 */ 669 KNOTE_UNLOCKED(&si->si.si_note, 0x100 /* notification */); 670} 671 672static void 673netmap_knrdetach(struct knote *kn) 674{ 675 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 676 struct selinfo *si = &priv->np_si[NR_RX]->si; 677 678 D("remove selinfo %p", si); 679 knlist_remove(&si->si_note, kn, 0); 680} 681 682static void 683netmap_knwdetach(struct knote *kn) 684{ 685 struct netmap_priv_d *priv = (struct netmap_priv_d *)kn->kn_hook; 686 struct selinfo *si = &priv->np_si[NR_TX]->si; 687 688 D("remove selinfo %p", si); 689 knlist_remove(&si->si_note, kn, 0); 690} 691 692/* 693 * callback from notifies (generated externally) and our 694 * calls to kevent(). The former we just return 1 (ready) 695 * since we do not know better. 696 * In the latter we call netmap_poll and return 0/1 accordingly. 697 */ 698static int 699netmap_knrw(struct knote *kn, long hint, int events) 700{ 701 struct netmap_priv_d *priv; 702 int revents; 703 704 if (hint != 0) { 705 ND(5, "call from notify"); 706 return 1; /* assume we are ready */ 707 } 708 priv = kn->kn_hook; 709 /* the notification may come from an external thread, 710 * in which case we do not want to run the netmap_poll 711 * This should be filtered above, but check just in case. 712 */ 713 if (curthread != priv->np_td) { /* should not happen */ 714 RD(5, "curthread changed %p %p", curthread, priv->np_td); 715 return 1; 716 } else { 717 revents = netmap_poll((void *)priv, events, curthread); 718 return (events & revents) ? 1 : 0; 719 } 720} 721 722static int 723netmap_knread(struct knote *kn, long hint) 724{ 725 return netmap_knrw(kn, hint, POLLIN); 726} 727 728static int 729netmap_knwrite(struct knote *kn, long hint) 730{ 731 return netmap_knrw(kn, hint, POLLOUT); 732} 733 734static struct filterops netmap_rfiltops = { 735 .f_isfd = 1, 736 .f_detach = netmap_knrdetach, 737 .f_event = netmap_knread, 738}; 739 740static struct filterops netmap_wfiltops = { 741 .f_isfd = 1, 742 .f_detach = netmap_knwdetach, 743 .f_event = netmap_knwrite, 744}; 745 746 747/* 748 * This is called when a thread invokes kevent() to record 749 * a change in the configuration of the kqueue(). 750 * The 'priv' should be the same as in the netmap device. 751 */ 752static int 753netmap_kqfilter(struct cdev *dev, struct knote *kn) 754{ 755 struct netmap_priv_d *priv; 756 int error; 757 struct netmap_adapter *na; 758 struct nm_selinfo *si; 759 int ev = kn->kn_filter; 760 761 if (ev != EVFILT_READ && ev != EVFILT_WRITE) { 762 D("bad filter request %d", ev); 763 return 1; 764 } 765 error = devfs_get_cdevpriv((void**)&priv); 766 if (error) { 767 D("device not yet setup"); 768 return 1; 769 } 770 na = priv->np_na; 771 if (na == NULL) { 772 D("no netmap adapter for this file descriptor"); 773 return 1; 774 } 775 /* the si is indicated in the priv */ 776 si = priv->np_si[(ev == EVFILT_WRITE) ? NR_TX : NR_RX]; 777 // XXX lock(priv) ? 778 kn->kn_fop = (ev == EVFILT_WRITE) ? 779 &netmap_wfiltops : &netmap_rfiltops; 780 kn->kn_hook = priv; 781 knlist_add(&si->si.si_note, kn, 0); 782 // XXX unlock(priv) 783 ND("register %p %s td %p priv %p kn %p np_nifp %p kn_fp/fpop %s", 784 na, na->ifp->if_xname, curthread, priv, kn, 785 priv->np_nifp, 786 kn->kn_fp == curthread->td_fpop ? "match" : "MISMATCH"); 787 return 0; 788} 789 790struct cdevsw netmap_cdevsw = { 791 .d_version = D_VERSION, 792 .d_name = "netmap", 793 .d_open = netmap_open, 794 .d_mmap_single = netmap_mmap_single, 795 .d_ioctl = netmap_ioctl, 796 .d_poll = netmap_poll, 797 .d_kqfilter = netmap_kqfilter, 798 .d_close = netmap_close, 799}; 800/*--- end of kqueue support ----*/ 801 802/* 803 * Kernel entry point. 804 * 805 * Initialize/finalize the module and return. 806 * 807 * Return 0 on success, errno on failure. 808 */ 809static int 810netmap_loader(__unused struct module *module, int event, __unused void *arg) 811{ 812 int error = 0; 813 814 switch (event) { 815 case MOD_LOAD: 816 error = netmap_init(); 817 break; 818 819 case MOD_UNLOAD: 820 /* 821 * if some one is still using netmap, 822 * then the module can not be unloaded. 823 */ 824 if (netmap_use_count) { 825 D("netmap module can not be unloaded - netmap_use_count: %d", 826 netmap_use_count); 827 error = EBUSY; 828 break; 829 } 830 netmap_fini(); 831 break; 832 833 default: 834 error = EOPNOTSUPP; 835 break; 836 } 837 838 return (error); 839} 840 841 842DEV_MODULE(netmap, netmap_loader, NULL); 843MODULE_VERSION(netmap, 1); 844