sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c

   1 /*-
   2  * Copyright (c) 2010-2012 Citrix Inc.
   3  * Copyright (c) 2009-2012 Microsoft Corp.
   4  * Copyright (c) 2012 NetApp Inc.
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice unmodified, this list of conditions, and the following
  12  *    disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  27  */
  28
  29 /*-
  30  * Copyright (c) 2004-2006 Kip Macy
  31  * All rights reserved.
  32  *
  33  * Redistribution and use in source and binary forms, with or without
  34  * modification, are permitted provided that the following conditions
  35  * are met:
  36  * 1. Redistributions of source code must retain the above copyright
  37  *    notice, this list of conditions and the following disclaimer.
  38  * 2. Redistributions in binary form must reproduce the above copyright
  39  *    notice, this list of conditions and the following disclaimer in the
  40  *    documentation and/or other materials provided with the distribution.
  41  *
  42  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  43  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  44  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  45  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  46  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  47  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  48  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  49  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  50  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  51  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  52  * SUCH DAMAGE.
  53  */
  54
  55 #include <sys/cdefs.h>
  56 __FBSDID("$FreeBSD$");
  57
  58 #include "opt_inet6.h"
  59 #include "opt_inet.h"
  60
  61 #include <sys/param.h>
  62 #include <sys/systm.h>
  63 #include <sys/sockio.h>
  64 #include <sys/mbuf.h>
  65 #include <sys/malloc.h>
  66 #include <sys/module.h>
  67 #include <sys/kernel.h>
  68 #include <sys/socket.h>
  69 #include <sys/queue.h>
  70 #include <sys/lock.h>
  71 #include <sys/sx.h>
  72 #include <sys/sysctl.h>
  73
  74 #include <net/if.h>
  75 #include <net/if_arp.h>
  76 #include <net/ethernet.h>
  77 #include <net/if_dl.h>
  78 #include <net/if_media.h>
  79
  80 #include <net/bpf.h>
  81
  82 #include <net/if_var.h>
  83 #include <net/if_types.h>
  84 #include <net/if_vlan_var.h>
  85
  86 #include <netinet/in_systm.h>
  87 #include <netinet/in.h>
  88 #include <netinet/ip.h>
  89 #include <netinet/if_ether.h>
  90 #include <netinet/tcp.h>
  91 #include <netinet/udp.h>
  92 #include <netinet/ip6.h>
  93
  94 #include <vm/vm.h>
  95 #include <vm/vm_param.h>
  96 #include <vm/vm_kern.h>
  97 #include <vm/pmap.h>
  98
  99 #include <machine/bus.h>
 100 #include <machine/resource.h>
 101 #include <machine/frame.h>
 102 #include <machine/vmparam.h>
 103
 104 #include <sys/bus.h>
 105 #include <sys/rman.h>
 106 #include <sys/mutex.h>
 107 #include <sys/errno.h>
 108 #include <sys/types.h>
 109 #include <machine/atomic.h>
 110
 111 #include <machine/intr_machdep.h>
 112
 113 #include <machine/in_cksum.h>
 114
 115 #include <dev/hyperv/include/hyperv.h>
 116 #include "hv_net_vsc.h"
 117 #include "hv_rndis.h"
 118 #include "hv_rndis_filter.h"
 119
 120
 121 /* Short for Hyper-V network interface */
 122 #define NETVSC_DEVNAME    "hn"
 123
 124 /*
 125  * It looks like offset 0 of buf is reserved to hold the softc pointer.
 126  * The sc pointer evidently not needed, and is not presently populated.
 127  * The packet offset is where the netvsc_packet starts in the buffer.
 128  */
 129 #define HV_NV_SC_PTR_OFFSET_IN_BUF         0
 130 #define HV_NV_PACKET_OFFSET_IN_BUF         16
 131
 132 /*
 133  * A unified flag for all outbound check sum flags is useful,
 134  * and it helps avoiding unnecessary check sum calculation in
 135  * network forwarding scenario.
 136  */
 137 #define HV_CSUM_FOR_OUTBOUND                                            \
 138     (CSUM_IP|CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP|CSUM_IP_TSO|          \
 139     CSUM_IP_ISCSI|CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP|              \
 140     CSUM_IP6_TSO|CSUM_IP6_ISCSI)
 141
 142 /* XXX move to netinet/tcp_lro.h */
 143 #define HN_LRO_HIWAT_MAX                                65535
 144 #define HN_LRO_HIWAT_DEF                                HN_LRO_HIWAT_MAX
 145 /* YYY 2*MTU is a bit rough, but should be good enough. */
 146 #define HN_LRO_HIWAT_MTULIM(ifp)                        (2 * (ifp)->if_mtu)
 147 #define HN_LRO_HIWAT_ISVALID(sc, hiwat)                 \
 148     ((hiwat) >= HN_LRO_HIWAT_MTULIM((sc)->hn_ifp) ||    \
 149      (hiwat) <= HN_LRO_HIWAT_MAX)
 150
 151 /*
 152  * Be aware that this sleepable mutex will exhibit WITNESS errors when
 153  * certain TCP and ARP code paths are taken.  This appears to be a
 154  * well-known condition, as all other drivers checked use a sleeping
 155  * mutex to protect their transmit paths.
 156  * Also Be aware that mutexes do not play well with semaphores, and there
 157  * is a conflicting semaphore in a certain channel code path.
 158  */
 159 #define NV_LOCK_INIT(_sc, _name) \
 160             mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF)
 161 #define NV_LOCK(_sc)            mtx_lock(&(_sc)->hn_lock)
 162 #define NV_LOCK_ASSERT(_sc)     mtx_assert(&(_sc)->hn_lock, MA_OWNED)
 163 #define NV_UNLOCK(_sc)          mtx_unlock(&(_sc)->hn_lock)
 164 #define NV_LOCK_DESTROY(_sc)    mtx_destroy(&(_sc)->hn_lock)
 165
 166
 167 /*
 168  * Globals
 169  */
 170
 171 int hv_promisc_mode = 0;    /* normal mode by default */
 172
 173 /* Trust tcp segements verification on host side. */
 174 static int hn_trust_hosttcp = 0;
 175 TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp);
 176
 177 /*
 178  * Forward declarations
 179  */
 180 static void hn_stop(hn_softc_t *sc);
 181 static void hn_ifinit_locked(hn_softc_t *sc);
 182 static void hn_ifinit(void *xsc);
 183 static int  hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
 184 static int  hn_start_locked(struct ifnet *ifp);
 185 static void hn_start(struct ifnet *ifp);
 186 static int hn_ifmedia_upd(struct ifnet *ifp);
 187 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
 188 #ifdef HN_LRO_HIWAT
 189 static int hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS);
 190 #endif
 191 static int hn_check_iplen(const struct mbuf *, int);
 192
 193 static __inline void
 194 hn_set_lro_hiwat(struct hn_softc *sc, int hiwat)
 195 {
 196         sc->hn_lro_hiwat = hiwat;
 197 #ifdef HN_LRO_HIWAT
 198         sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
 199 #endif
 200 }
 201
 202 /*
 203  * NetVsc get message transport protocol type
 204  */
 205 static uint32_t get_transport_proto_type(struct mbuf *m_head)
 206 {
 207         uint32_t ret_val = TRANSPORT_TYPE_NOT_IP;
 208         uint16_t ether_type = 0;
 209         int ether_len = 0;
 210         struct ether_vlan_header *eh;
 211 #ifdef INET
 212         struct ip *iph;
 213 #endif
 214 #ifdef INET6
 215         struct ip6_hdr *ip6;
 216 #endif
 217
 218         eh = mtod(m_head, struct ether_vlan_header*);
 219         if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
 220                 ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
 221                 ether_type = eh->evl_proto;
 222         } else {
 223                 ether_len = ETHER_HDR_LEN;
 224                 ether_type = eh->evl_encap_proto;
 225         }
 226
 227         switch (ntohs(ether_type)) {
 228 #ifdef INET6
 229         case ETHERTYPE_IPV6:
 230                 ip6 = (struct ip6_hdr *)(m_head->m_data + ether_len);
 231
 232                 if (IPPROTO_TCP == ip6->ip6_nxt) {
 233                         ret_val = TRANSPORT_TYPE_IPV6_TCP;
 234                 } else if (IPPROTO_UDP == ip6->ip6_nxt) {
 235                         ret_val = TRANSPORT_TYPE_IPV6_UDP;
 236                 }
 237                 break;
 238 #endif
 239 #ifdef INET
 240         case ETHERTYPE_IP:
 241                 iph = (struct ip *)(m_head->m_data + ether_len);
 242
 243                 if (IPPROTO_TCP == iph->ip_p) {
 244                         ret_val = TRANSPORT_TYPE_IPV4_TCP;
 245                 } else if (IPPROTO_UDP == iph->ip_p) {
 246                         ret_val = TRANSPORT_TYPE_IPV4_UDP;
 247                 }
 248                 break;
 249 #endif
 250         default:
 251                 ret_val = TRANSPORT_TYPE_NOT_IP;
 252                 break;
 253         }
 254
 255         return (ret_val);
 256 }
 257
 258 static int
 259 hn_ifmedia_upd(struct ifnet *ifp __unused)
 260 {
 261
 262         return EOPNOTSUPP;
 263 }
 264
 265 static void
 266 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
 267 {
 268         struct hn_softc *sc = ifp->if_softc;
 269
 270         ifmr->ifm_status = IFM_AVALID;
 271         ifmr->ifm_active = IFM_ETHER;
 272
 273         if (!sc->hn_carrier) {
 274                 ifmr->ifm_active |= IFM_NONE;
 275                 return;
 276         }
 277         ifmr->ifm_status |= IFM_ACTIVE;
 278         ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
 279 }
 280
 281 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
 282 static const hv_guid g_net_vsc_device_type = {
 283         .data = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 284                 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
 285 };
 286
 287 /*
 288  * Standard probe entry point.
 289  *
 290  */
 291 static int
 292 netvsc_probe(device_t dev)
 293 {
 294         const char *p;
 295
 296         p = vmbus_get_type(dev);
 297         if (!memcmp(p, &g_net_vsc_device_type.data, sizeof(hv_guid))) {
 298                 device_set_desc(dev, "Synthetic Network Interface");
 299                 if (bootverbose)
 300                         printf("Netvsc probe... DONE \n");
 301
 302                 return (BUS_PROBE_DEFAULT);
 303         }
 304
 305         return (ENXIO);
 306 }
 307
 308 /*
 309  * Standard attach entry point.
 310  *
 311  * Called when the driver is loaded.  It allocates needed resources,
 312  * and initializes the "hardware" and software.
 313  */
 314 static int
 315 netvsc_attach(device_t dev)
 316 {
 317         struct hv_device *device_ctx = vmbus_get_devctx(dev);
 318         netvsc_device_info device_info;
 319         hn_softc_t *sc;
 320         int unit = device_get_unit(dev);
 321         struct ifnet *ifp;
 322         struct sysctl_oid_list *child;
 323         struct sysctl_ctx_list *ctx;
 324         int ret;
 325
 326         sc = device_get_softc(dev);
 327         if (sc == NULL) {
 328                 return (ENOMEM);
 329         }
 330
 331         bzero(sc, sizeof(hn_softc_t));
 332         sc->hn_unit = unit;
 333         sc->hn_dev = dev;
 334         sc->hn_lro_hiwat = HN_LRO_HIWAT_DEF;
 335         sc->hn_trust_hosttcp = hn_trust_hosttcp;
 336
 337         NV_LOCK_INIT(sc, "NetVSCLock");
 338
 339         sc->hn_dev_obj = device_ctx;
 340
 341         ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
 342         ifp->if_softc = sc;
 343
 344         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
 345         ifp->if_dunit = unit;
 346         ifp->if_dname = NETVSC_DEVNAME;
 347
 348         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
 349         ifp->if_ioctl = hn_ioctl;
 350         ifp->if_start = hn_start;
 351         ifp->if_init = hn_ifinit;
 352         /* needed by hv_rf_on_device_add() code */
 353         ifp->if_mtu = ETHERMTU;
 354         IFQ_SET_MAXLEN(&ifp->if_snd, 512);
 355         ifp->if_snd.ifq_drv_maxlen = 511;
 356         IFQ_SET_READY(&ifp->if_snd);
 357
 358         ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
 359         ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
 360         ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
 361         /* XXX ifmedia_set really should do this for us */
 362         sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
 363
 364         /*
 365          * Tell upper layers that we support full VLAN capability.
 366          */
 367         ifp->if_hdrlen = sizeof(struct ether_vlan_header);
 368         ifp->if_capabilities |=
 369             IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
 370             IFCAP_LRO;
 371         ifp->if_capenable |=
 372             IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO |
 373             IFCAP_LRO;
 374         /*
 375          * Only enable UDP checksum offloading when it is on 2012R2 or
 376          * later. UDP checksum offloading doesn't work on earlier
 377          * Windows releases.
 378          */
 379         if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1)
 380                 ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_TSO;
 381         else
 382                 ifp->if_hwassist = CSUM_TCP | CSUM_TSO;
 383
 384         ret = hv_rf_on_device_add(device_ctx, &device_info);
 385         if (ret != 0) {
 386                 if_free(ifp);
 387
 388                 return (ret);
 389         }
 390         if (device_info.link_state == 0) {
 391                 sc->hn_carrier = 1;
 392         }
 393
 394         tcp_lro_init(&sc->hn_lro);
 395         /* Driver private LRO settings */
 396         sc->hn_lro.ifp = ifp;
 397 #ifdef HN_LRO_HIWAT
 398         sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
 399 #endif
 400
 401         ether_ifattach(ifp, device_info.mac_addr);
 402
 403         ctx = device_get_sysctl_ctx(dev);
 404         child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
 405
 406         SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_queued",
 407             CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
 408         SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_flushed",
 409             CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
 410         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
 411             CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
 412 #ifdef HN_LRO_HIWAT
 413         SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_hiwat",
 414             CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_hiwat_sysctl,
 415             "I", "LRO high watermark");
 416 #endif
 417         SYSCTL_ADD_INT(ctx, child, OID_AUTO, "trust_hosttcp",
 418             CTLFLAG_RW, &sc->hn_trust_hosttcp, 0,
 419             "Trust tcp segement verification on host side, "
 420             "when csum info is missing");
 421         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
 422             CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
 423         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
 424             CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
 425         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
 426             CTLFLAG_RW, &sc->hn_csum_trusted,
 427             "# of TCP segements that we trust host's csum verification");
 428         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
 429             CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
 430
 431         if (unit == 0) {
 432                 struct sysctl_ctx_list *dc_ctx;
 433                 struct sysctl_oid_list *dc_child;
 434                 devclass_t dc;
 435
 436                 /*
 437                  * Add sysctl nodes for devclass
 438                  */
 439                 dc = device_get_devclass(dev);
 440                 dc_ctx = devclass_get_sysctl_ctx(dc);
 441                 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
 442
 443                 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
 444                     CTLFLAG_RD, &hn_trust_hosttcp, 0,
 445                     "Trust tcp segement verification on host side, "
 446                     "when csum info is missing (global setting)");
 447         }
 448
 449         return (0);
 450 }
 451
 452 /*
 453  * Standard detach entry point
 454  */
 455 static int
 456 netvsc_detach(device_t dev)
 457 {
 458         struct hn_softc *sc = device_get_softc(dev);
 459         struct hv_device *hv_device = vmbus_get_devctx(dev);
 460
 461         if (bootverbose)
 462                 printf("netvsc_detach\n");
 463
 464         /*
 465          * XXXKYS:  Need to clean up all our
 466          * driver state; this is the driver
 467          * unloading.
 468          */
 469
 470         /*
 471          * XXXKYS:  Need to stop outgoing traffic and unregister
 472          * the netdevice.
 473          */
 474
 475         hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
 476
 477         ifmedia_removeall(&sc->hn_media);
 478         tcp_lro_free(&sc->hn_lro);
 479
 480         return (0);
 481 }
 482
 483 /*
 484  * Standard shutdown entry point
 485  */
 486 static int
 487 netvsc_shutdown(device_t dev)
 488 {
 489         return (0);
 490 }
 491
 492 /*
 493  * Send completion processing
 494  *
 495  * Note:  It looks like offset 0 of buf is reserved to hold the softc
 496  * pointer.  The sc pointer is not currently needed in this function, and
 497  * it is not presently populated by the TX function.
 498  */
 499 void
 500 netvsc_xmit_completion(void *context)
 501 {
 502         netvsc_packet *packet = (netvsc_packet *)context;
 503         struct mbuf *mb;
 504         uint8_t *buf;
 505
 506         mb = (struct mbuf *)(uintptr_t)packet->compl.send.send_completion_tid;
 507         buf = ((uint8_t *)packet) - HV_NV_PACKET_OFFSET_IN_BUF;
 508
 509         free(buf, M_NETVSC);
 510
 511         if (mb != NULL) {
 512                 m_freem(mb);
 513         }
 514 }
 515
 516 /*
 517  * Start a transmit of one or more packets
 518  */
 519 static int
 520 hn_start_locked(struct ifnet *ifp)
 521 {
 522         hn_softc_t *sc = ifp->if_softc;
 523         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
 524         netvsc_dev *net_dev = sc->net_dev;
 525         device_t dev = device_ctx->device;
 526         uint8_t *buf;
 527         netvsc_packet *packet;
 528         struct mbuf *m_head, *m;
 529         struct mbuf *mc_head = NULL;
 530         struct ether_vlan_header *eh;
 531         rndis_msg *rndis_mesg;
 532         rndis_packet *rndis_pkt;
 533         rndis_per_packet_info *rppi;
 534         ndis_8021q_info *rppi_vlan_info;
 535         rndis_tcp_ip_csum_info *csum_info;
 536         rndis_tcp_tso_info *tso_info;
 537         int ether_len;
 538         int i;
 539         int num_frags;
 540         int len;
 541         int retries = 0;
 542         int ret = 0;
 543         uint32_t rndis_msg_size = 0;
 544         uint32_t trans_proto_type;
 545         uint32_t send_buf_section_idx =
 546             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
 547
 548         while (!IFQ_DRV_IS_EMPTY(&sc->hn_ifp->if_snd)) {
 549                 IFQ_DRV_DEQUEUE(&sc->hn_ifp->if_snd, m_head);
 550                 if (m_head == NULL) {
 551                         break;
 552                 }
 553
 554                 len = 0;
 555                 num_frags = 0;
 556
 557                 /* Walk the mbuf list computing total length and num frags */
 558                 for (m = m_head; m != NULL; m = m->m_next) {
 559                         if (m->m_len != 0) {
 560                                 num_frags++;
 561                                 len += m->m_len;
 562                         }
 563                 }
 564
 565                 /*
 566                  * Reserve the number of pages requested.  Currently,
 567                  * one page is reserved for the message in the RNDIS
 568                  * filter packet
 569                  */
 570                 num_frags += HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
 571
 572                 /* If exceeds # page_buffers in netvsc_packet */
 573                 if (num_frags > NETVSC_PACKET_MAXPAGE) {
 574                         device_printf(dev, "exceed max page buffers,%d,%d\n",
 575                             num_frags, NETVSC_PACKET_MAXPAGE);
 576                         m_freem(m_head);
 577                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 578                         return (EINVAL);
 579                 }
 580
 581                 /*
 582                  * Allocate a buffer with space for a netvsc packet plus a
 583                  * number of reserved areas.  First comes a (currently 16
 584                  * bytes, currently unused) reserved data area.  Second is
 585                  * the netvsc_packet. Third is an area reserved for an
 586                  * rndis_filter_packet struct. Fourth (optional) is a
 587                  * rndis_per_packet_info struct.
 588                  * Changed malloc to M_NOWAIT to avoid sleep under spin lock.
 589                  * No longer reserving extra space for page buffers, as they
 590                  * are already part of the netvsc_packet.
 591                  */
 592                 buf = malloc(HV_NV_PACKET_OFFSET_IN_BUF +
 593                         sizeof(netvsc_packet) +
 594                         sizeof(rndis_msg) +
 595                         RNDIS_VLAN_PPI_SIZE +
 596                         RNDIS_TSO_PPI_SIZE +
 597                         RNDIS_CSUM_PPI_SIZE,
 598                         M_NETVSC, M_ZERO | M_NOWAIT);
 599                 if (buf == NULL) {
 600                         device_printf(dev, "hn:malloc packet failed\n");
 601                         m_freem(m_head);
 602                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 603                         return (ENOMEM);
 604                 }
 605
 606                 packet = (netvsc_packet *)(buf + HV_NV_PACKET_OFFSET_IN_BUF);
 607                 *(vm_offset_t *)buf = HV_NV_SC_PTR_OFFSET_IN_BUF;
 608
 609                 packet->is_data_pkt = TRUE;
 610
 611                 /* Set up the rndis header */
 612                 packet->page_buf_count = num_frags;
 613
 614                 /* Initialize it from the mbuf */
 615                 packet->tot_data_buf_len = len;
 616
 617                 /*
 618                  * extension points to the area reserved for the
 619                  * rndis_filter_packet, which is placed just after
 620                  * the netvsc_packet (and rppi struct, if present;
 621                  * length is updated later).
 622                  */
 623                 packet->rndis_mesg = packet + 1;
 624                 rndis_mesg = (rndis_msg *)packet->rndis_mesg;
 625                 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
 626
 627                 rndis_pkt = &rndis_mesg->msg.packet;
 628                 rndis_pkt->data_offset = sizeof(rndis_packet);
 629                 rndis_pkt->data_length = packet->tot_data_buf_len;
 630                 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
 631
 632                 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
 633
 634                 /*
 635                  * If the Hyper-V infrastructure needs to embed a VLAN tag,
 636                  * initialize netvsc_packet and rppi struct values as needed.
 637                  */
 638                 if (m_head->m_flags & M_VLANTAG) {
 639                         /*
 640                          * set up some additional fields so the Hyper-V infrastructure will stuff the VLAN tag
 641                          * into the frame.
 642                          */
 643                         packet->vlan_tci = m_head->m_pkthdr.ether_vtag;
 644
 645                         rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
 646
 647                         rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
 648                             ieee_8021q_info);
 649
 650                         /* VLAN info immediately follows rppi struct */
 651                         rppi_vlan_info = (ndis_8021q_info *)((char*)rppi +
 652                             rppi->per_packet_info_offset);
 653                         /* FreeBSD does not support CFI or priority */
 654                         rppi_vlan_info->u1.s1.vlan_id =
 655                             packet->vlan_tci & 0xfff;
 656                 }
 657
 658                 /* Only check the flags for outbound and ignore the ones for inbound */
 659                 if (0 == (m_head->m_pkthdr.csum_flags & HV_CSUM_FOR_OUTBOUND)) {
 660                         goto pre_send;
 661                 }
 662
 663                 eh = mtod(m_head, struct ether_vlan_header*);
 664                 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
 665                         ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
 666                 } else {
 667                         ether_len = ETHER_HDR_LEN;
 668                 }
 669
 670                 trans_proto_type = get_transport_proto_type(m_head);
 671                 if (TRANSPORT_TYPE_NOT_IP == trans_proto_type) {
 672                         goto pre_send;
 673                 }
 674
 675                 /*
 676                  * TSO packet needless to setup the send side checksum
 677                  * offload.
 678                  */
 679                 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
 680                         goto do_tso;
 681                 }
 682
 683                 /* setup checksum offload */
 684                 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
 685                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
 686                     tcpip_chksum_info);
 687                 csum_info = (rndis_tcp_ip_csum_info *)((char*)rppi +
 688                     rppi->per_packet_info_offset);
 689
 690                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
 691                         csum_info->xmit.is_ipv4 = 1;
 692                 } else {
 693                         csum_info->xmit.is_ipv6 = 1;
 694                 }
 695
 696                 if (trans_proto_type & TYPE_TCP) {
 697                         csum_info->xmit.tcp_csum = 1;
 698                         csum_info->xmit.tcp_header_offset = 0;
 699                 } else if (trans_proto_type & TYPE_UDP) {
 700                         csum_info->xmit.udp_csum = 1;
 701                 }
 702
 703                 goto pre_send;
 704
 705 do_tso:
 706                 /* setup TCP segmentation offload */
 707                 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
 708                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
 709                     tcp_large_send_info);
 710
 711                 tso_info = (rndis_tcp_tso_info *)((char *)rppi +
 712                     rppi->per_packet_info_offset);
 713                 tso_info->lso_v2_xmit.type =
 714                     RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 715
 716 #ifdef INET
 717                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
 718                         struct ip *ip =
 719                             (struct ip *)(m_head->m_data + ether_len);
 720                         unsigned long iph_len = ip->ip_hl << 2;
 721                         struct tcphdr *th =
 722                             (struct tcphdr *)((caddr_t)ip + iph_len);
 723
 724                         tso_info->lso_v2_xmit.ip_version =
 725                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 726                         ip->ip_len = 0;
 727                         ip->ip_sum = 0;
 728
 729                         th->th_sum = in_pseudo(ip->ip_src.s_addr,
 730                             ip->ip_dst.s_addr,
 731                             htons(IPPROTO_TCP));
 732                 }
 733 #endif
 734 #if defined(INET6) && defined(INET)
 735                 else
 736 #endif
 737 #ifdef INET6
 738                 {
 739                         struct ip6_hdr *ip6 =
 740                             (struct ip6_hdr *)(m_head->m_data + ether_len);
 741                         struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
 742
 743                         tso_info->lso_v2_xmit.ip_version =
 744                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 745                         ip6->ip6_plen = 0;
 746                         th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
 747                 }
 748 #endif
 749                 tso_info->lso_v2_xmit.tcp_header_offset = 0;
 750                 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
 751
 752 pre_send:
 753                 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
 754                 packet->tot_data_buf_len = rndis_mesg->msg_len;
 755
 756                 /* send packet with send buffer */
 757                 if (packet->tot_data_buf_len < net_dev->send_section_size) {
 758                         send_buf_section_idx =
 759                             hv_nv_get_next_send_section(net_dev);
 760                         if (send_buf_section_idx !=
 761                             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
 762                                 char *dest = ((char *)net_dev->send_buf +
 763                                     send_buf_section_idx *
 764                                     net_dev->send_section_size);
 765
 766                                 memcpy(dest, rndis_mesg, rndis_msg_size);
 767                                 dest += rndis_msg_size;
 768                                 for (m = m_head; m != NULL; m = m->m_next) {
 769                                         if (m->m_len) {
 770                                                 memcpy(dest,
 771                                                     (void *)mtod(m, vm_offset_t),
 772                                                     m->m_len);
 773                                                 dest += m->m_len;
 774                                         }
 775                                 }
 776
 777                                 packet->send_buf_section_idx =
 778                                     send_buf_section_idx;
 779                                 packet->send_buf_section_size =
 780                                     packet->tot_data_buf_len;
 781                                 packet->page_buf_count = 0;
 782                                 goto do_send;
 783                         }
 784                 }
 785
 786                 /* send packet with page buffer */
 787                 packet->page_buffers[0].pfn =
 788                     atop(hv_get_phys_addr(rndis_mesg));
 789                 packet->page_buffers[0].offset =
 790                     (unsigned long)rndis_mesg & PAGE_MASK;
 791                 packet->page_buffers[0].length = rndis_msg_size;
 792
 793                 /*
 794                  * Fill the page buffers with mbuf info starting at index
 795                  * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
 796                  */
 797                 i = HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
 798                 for (m = m_head; m != NULL; m = m->m_next) {
 799                         if (m->m_len) {
 800                                 vm_offset_t paddr =
 801                                     vtophys(mtod(m, vm_offset_t));
 802                                 packet->page_buffers[i].pfn =
 803                                     paddr >> PAGE_SHIFT;
 804                                 packet->page_buffers[i].offset =
 805                                     paddr & (PAGE_SIZE - 1);
 806                                 packet->page_buffers[i].length = m->m_len;
 807                                 i++;
 808                         }
 809                 }
 810
 811                 packet->send_buf_section_idx =
 812                     NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
 813                 packet->send_buf_section_size = 0;
 814
 815 do_send:
 816
 817                 /*
 818                  * If bpf, copy the mbuf chain.  This is less expensive than
 819                  * it appears; the mbuf clusters are not copied, only their
 820                  * reference counts are incremented.
 821                  * Needed to avoid a race condition where the completion
 822                  * callback is invoked, freeing the mbuf chain, before the
 823                  * bpf_mtap code has a chance to run.
 824                  */
 825                 if (ifp->if_bpf) {
 826                         mc_head = m_copypacket(m_head, M_NOWAIT);
 827                 }
 828 retry_send:
 829                 /* Set the completion routine */
 830                 packet->compl.send.on_send_completion = netvsc_xmit_completion;
 831                 packet->compl.send.send_completion_context = packet;
 832                 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)m_head;
 833
 834                 /* Removed critical_enter(), does not appear necessary */
 835                 ret = hv_nv_on_send(device_ctx, packet);
 836                 if (ret == 0) {
 837                         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
 838                         /* if bpf && mc_head, call bpf_mtap code */
 839                         if (mc_head) {
 840                                 ETHER_BPF_MTAP(ifp, mc_head);
 841                         }
 842                 } else {
 843                         retries++;
 844                         if (retries < 4) {
 845                                 goto retry_send;
 846                         }
 847
 848                         IF_PREPEND(&ifp->if_snd, m_head);
 849                         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 850
 851                         /*
 852                          * Null the mbuf pointer so the completion function
 853                          * does not free the mbuf chain.  We just pushed the
 854                          * mbuf chain back on the if_snd queue.
 855                          */
 856                         packet->compl.send.send_completion_tid = 0;
 857
 858                         /*
 859                          * Release the resources since we will not get any
 860                          * send completion
 861                          */
 862                         netvsc_xmit_completion(packet);
 863                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 864                 }
 865
 866                 /* if bpf && mc_head, free the mbuf chain copy */
 867                 if (mc_head) {
 868                         m_freem(mc_head);
 869                 }
 870         }
 871
 872         return (ret);
 873 }
 874
 875 /*
 876  * Link up/down notification
 877  */
 878 void
 879 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
 880 {
 881         hn_softc_t *sc = device_get_softc(device_obj->device);
 882
 883         if (sc == NULL) {
 884                 return;
 885         }
 886
 887         if (status == 1) {
 888                 sc->hn_carrier = 1;
 889         } else {
 890                 sc->hn_carrier = 0;
 891         }
 892 }
 893
 894 /*
 895  * Append the specified data to the indicated mbuf chain,
 896  * Extend the mbuf chain if the new data does not fit in
 897  * existing space.
 898  *
 899  * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
 900  * There should be an equivalent in the kernel mbuf code,
 901  * but there does not appear to be one yet.
 902  *
 903  * Differs from m_append() in that additional mbufs are
 904  * allocated with cluster size MJUMPAGESIZE, and filled
 905  * accordingly.
 906  *
 907  * Return 1 if able to complete the job; otherwise 0.
 908  */
 909 static int
 910 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
 911 {
 912         struct mbuf *m, *n;
 913         int remainder, space;
 914
 915         for (m = m0; m->m_next != NULL; m = m->m_next)
 916                 ;
 917         remainder = len;
 918         space = M_TRAILINGSPACE(m);
 919         if (space > 0) {
 920                 /*
 921                  * Copy into available space.
 922                  */
 923                 if (space > remainder)
 924                         space = remainder;
 925                 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
 926                 m->m_len += space;
 927                 cp += space;
 928                 remainder -= space;
 929         }
 930         while (remainder > 0) {
 931                 /*
 932                  * Allocate a new mbuf; could check space
 933                  * and allocate a cluster instead.
 934                  */
 935                 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
 936                 if (n == NULL)
 937                         break;
 938                 n->m_len = min(MJUMPAGESIZE, remainder);
 939                 bcopy(cp, mtod(n, caddr_t), n->m_len);
 940                 cp += n->m_len;
 941                 remainder -= n->m_len;
 942                 m->m_next = n;
 943                 m = n;
 944         }
 945         if (m0->m_flags & M_PKTHDR)
 946                 m0->m_pkthdr.len += len - remainder;
 947
 948         return (remainder == 0);
 949 }
 950
 951
 952 /*
 953  * Called when we receive a data packet from the "wire" on the
 954  * specified device
 955  *
 956  * Note:  This is no longer used as a callback
 957  */
 958 int
 959 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
 960     rndis_tcp_ip_csum_info *csum_info)
 961 {
 962         hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
 963         struct mbuf *m_new;
 964         struct ifnet *ifp;
 965         device_t dev = device_ctx->device;
 966         int size, do_lro = 0;
 967
 968         if (sc == NULL) {
 969                 return (0); /* TODO: KYS how can this be! */
 970         }
 971
 972         ifp = sc->hn_ifp;
 973
 974         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
 975                 return (0);
 976         }
 977
 978         /*
 979          * Bail out if packet contains more data than configured MTU.
 980          */
 981         if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
 982                 return (0);
 983         } else if (packet->tot_data_buf_len <= MHLEN) {
 984                 m_new = m_gethdr(M_NOWAIT, MT_DATA);
 985                 if (m_new == NULL)
 986                         return (0);
 987                 memcpy(mtod(m_new, void *), packet->data,
 988                     packet->tot_data_buf_len);
 989                 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
 990                 sc->hn_small_pkts++;
 991         } else {
 992                 /*
 993                  * Get an mbuf with a cluster.  For packets 2K or less,
 994                  * get a standard 2K cluster.  For anything larger, get a
 995                  * 4K cluster.  Any buffers larger than 4K can cause problems
 996                  * if looped around to the Hyper-V TX channel, so avoid them.
 997                  */
 998                 size = MCLBYTES;
 999                 if (packet->tot_data_buf_len > MCLBYTES) {
1000                         /* 4096 */
1001                         size = MJUMPAGESIZE;
1002                 }
1003
1004                 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1005                 if (m_new == NULL) {
1006                         device_printf(dev, "alloc mbuf failed.\n");
1007                         return (0);
1008                 }
1009
1010                 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1011         }
1012         m_new->m_pkthdr.rcvif = ifp;
1013
1014         /* receive side checksum offload */
1015         if (NULL != csum_info) {
1016                 /* IP csum offload */
1017                 if (csum_info->receive.ip_csum_succeeded) {
1018                         m_new->m_pkthdr.csum_flags |=
1019                             (CSUM_IP_CHECKED | CSUM_IP_VALID);
1020                         sc->hn_csum_ip++;
1021                 }
1022
1023                 /* TCP csum offload */
1024                 if (csum_info->receive.tcp_csum_succeeded) {
1025                         m_new->m_pkthdr.csum_flags |=
1026                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1027                         m_new->m_pkthdr.csum_data = 0xffff;
1028                         sc->hn_csum_tcp++;
1029                 }
1030
1031                 if (csum_info->receive.ip_csum_succeeded &&
1032                     csum_info->receive.tcp_csum_succeeded)
1033                         do_lro = 1;
1034         } else {
1035                 const struct ether_header *eh;
1036                 uint16_t etype;
1037                 int hoff;
1038
1039                 hoff = sizeof(*eh);
1040                 if (m_new->m_len < hoff)
1041                         goto skip;
1042                 eh = mtod(m_new, struct ether_header *);
1043                 etype = ntohs(eh->ether_type);
1044                 if (etype == ETHERTYPE_VLAN) {
1045                         const struct ether_vlan_header *evl;
1046
1047                         hoff = sizeof(*evl);
1048                         if (m_new->m_len < hoff)
1049                                 goto skip;
1050                         evl = mtod(m_new, struct ether_vlan_header *);
1051                         etype = ntohs(evl->evl_proto);
1052                 }
1053
1054                 if (etype == ETHERTYPE_IP) {
1055                         int pr;
1056
1057                         pr = hn_check_iplen(m_new, hoff);
1058                         if (pr == IPPROTO_TCP) {
1059                                 if (sc->hn_trust_hosttcp) {
1060                                         sc->hn_csum_trusted++;
1061                                         m_new->m_pkthdr.csum_flags |=
1062                                            (CSUM_IP_CHECKED | CSUM_IP_VALID |
1063                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1064                                         m_new->m_pkthdr.csum_data = 0xffff;
1065                                 }
1066                                 /* Rely on SW csum verification though... */
1067                                 do_lro = 1;
1068                         }
1069                 }
1070         }
1071 skip:
1072         if ((packet->vlan_tci != 0) &&
1073             (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1074                 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1075                 m_new->m_flags |= M_VLANTAG;
1076         }
1077
1078         /*
1079          * Note:  Moved RX completion back to hv_nv_on_receive() so all
1080          * messages (not just data messages) will trigger a response.
1081          */
1082
1083         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1084
1085         if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1086                 struct lro_ctrl *lro = &sc->hn_lro;
1087
1088                 if (lro->lro_cnt) {
1089                         sc->hn_lro_tried++;
1090                         if (tcp_lro_rx(lro, m_new, 0) == 0) {
1091                                 /* DONE! */
1092                                 return 0;
1093                         }
1094                 }
1095         }
1096
1097         /* We're not holding the lock here, so don't release it */
1098         (*ifp->if_input)(ifp, m_new);
1099
1100         return (0);
1101 }
1102
1103 void
1104 netvsc_recv_rollup(struct hv_device *device_ctx)
1105 {
1106         hn_softc_t *sc = device_get_softc(device_ctx->device);
1107         struct lro_ctrl *lro = &sc->hn_lro;
1108         struct lro_entry *queued;
1109
1110         while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
1111                 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1112                 tcp_lro_flush(lro, queued);
1113         }
1114 }
1115
1116 /*
1117  * Rules for using sc->temp_unusable:
1118  * 1.  sc->temp_unusable can only be read or written while holding NV_LOCK()
1119  * 2.  code reading sc->temp_unusable under NV_LOCK(), and finding
1120  *     sc->temp_unusable set, must release NV_LOCK() and exit
1121  * 3.  to retain exclusive control of the interface,
1122  *     sc->temp_unusable must be set by code before releasing NV_LOCK()
1123  * 4.  only code setting sc->temp_unusable can clear sc->temp_unusable
1124  * 5.  code setting sc->temp_unusable must eventually clear sc->temp_unusable
1125  */
1126
1127 /*
1128  * Standard ioctl entry point.  Called when the user wants to configure
1129  * the interface.
1130  */
1131 static int
1132 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1133 {
1134         hn_softc_t *sc = ifp->if_softc;
1135         struct ifreq *ifr = (struct ifreq *)data;
1136 #ifdef INET
1137         struct ifaddr *ifa = (struct ifaddr *)data;
1138 #endif
1139         netvsc_device_info device_info;
1140         struct hv_device *hn_dev;
1141         int mask, error = 0;
1142         int retry_cnt = 500;
1143
1144         switch(cmd) {
1145
1146         case SIOCSIFADDR:
1147 #ifdef INET
1148                 if (ifa->ifa_addr->sa_family == AF_INET) {
1149                         ifp->if_flags |= IFF_UP;
1150                         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1151                                 hn_ifinit(sc);
1152                         arp_ifinit(ifp, ifa);
1153                 } else
1154 #endif
1155                 error = ether_ioctl(ifp, cmd, data);
1156                 break;
1157         case SIOCSIFMTU:
1158                 hn_dev = vmbus_get_devctx(sc->hn_dev);
1159
1160                 /* Check MTU value change */
1161                 if (ifp->if_mtu == ifr->ifr_mtu)
1162                         break;
1163
1164                 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1165                         error = EINVAL;
1166                         break;
1167                 }
1168
1169                 /* Obtain and record requested MTU */
1170                 ifp->if_mtu = ifr->ifr_mtu;
1171                 /*
1172                  * Make sure that LRO high watermark is still valid,
1173                  * after MTU change (the 2*MTU limit).
1174                  */
1175                 if (!HN_LRO_HIWAT_ISVALID(sc, sc->hn_lro_hiwat))
1176                         hn_set_lro_hiwat(sc, HN_LRO_HIWAT_MTULIM(ifp));
1177
1178                 do {
1179                         NV_LOCK(sc);
1180                         if (!sc->temp_unusable) {
1181                                 sc->temp_unusable = TRUE;
1182                                 retry_cnt = -1;
1183                         }
1184                         NV_UNLOCK(sc);
1185                         if (retry_cnt > 0) {
1186                                 retry_cnt--;
1187                                 DELAY(5 * 1000);
1188                         }
1189                 } while (retry_cnt > 0);
1190
1191                 if (retry_cnt == 0) {
1192                         error = EINVAL;
1193                         break;
1194                 }
1195
1196                 /* We must remove and add back the device to cause the new
1197                  * MTU to take effect.  This includes tearing down, but not
1198                  * deleting the channel, then bringing it back up.
1199                  */
1200                 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1201                 if (error) {
1202                         NV_LOCK(sc);
1203                         sc->temp_unusable = FALSE;
1204                         NV_UNLOCK(sc);
1205                         break;
1206                 }
1207                 error = hv_rf_on_device_add(hn_dev, &device_info);
1208                 if (error) {
1209                         NV_LOCK(sc);
1210                         sc->temp_unusable = FALSE;
1211                         NV_UNLOCK(sc);
1212                         break;
1213                 }
1214
1215                 hn_ifinit_locked(sc);
1216
1217                 NV_LOCK(sc);
1218                 sc->temp_unusable = FALSE;
1219                 NV_UNLOCK(sc);
1220                 break;
1221         case SIOCSIFFLAGS:
1222                 do {
1223                        NV_LOCK(sc);
1224                        if (!sc->temp_unusable) {
1225                                sc->temp_unusable = TRUE;
1226                                retry_cnt = -1;
1227                        }
1228                        NV_UNLOCK(sc);
1229                        if (retry_cnt > 0) {
1230                                 retry_cnt--;
1231                                 DELAY(5 * 1000);
1232                        }
1233                 } while (retry_cnt > 0);
1234
1235                 if (retry_cnt == 0) {
1236                        error = EINVAL;
1237                        break;
1238                 }
1239
1240                 if (ifp->if_flags & IFF_UP) {
1241                         /*
1242                          * If only the state of the PROMISC flag changed,
1243                          * then just use the 'set promisc mode' command
1244                          * instead of reinitializing the entire NIC. Doing
1245                          * a full re-init means reloading the firmware and
1246                          * waiting for it to start up, which may take a
1247                          * second or two.
1248                          */
1249 #ifdef notyet
1250                         /* Fixme:  Promiscuous mode? */
1251                         if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1252                             ifp->if_flags & IFF_PROMISC &&
1253                             !(sc->hn_if_flags & IFF_PROMISC)) {
1254                                 /* do something here for Hyper-V */
1255                         } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1256                             !(ifp->if_flags & IFF_PROMISC) &&
1257                             sc->hn_if_flags & IFF_PROMISC) {
1258                                 /* do something here for Hyper-V */
1259                         } else
1260 #endif
1261                                 hn_ifinit_locked(sc);
1262                 } else {
1263                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1264                                 hn_stop(sc);
1265                         }
1266                 }
1267                 NV_LOCK(sc);
1268                 sc->temp_unusable = FALSE;
1269                 NV_UNLOCK(sc);
1270                 sc->hn_if_flags = ifp->if_flags;
1271                 error = 0;
1272                 break;
1273         case SIOCSIFCAP:
1274                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1275                 if (mask & IFCAP_TXCSUM) {
1276                         if (IFCAP_TXCSUM & ifp->if_capenable) {
1277                                 ifp->if_capenable &= ~IFCAP_TXCSUM;
1278                                 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
1279                         } else {
1280                                 ifp->if_capenable |= IFCAP_TXCSUM;
1281                                 /*
1282                                  * Only enable UDP checksum offloading on
1283                                  * Windows Server 2012R2 or later releases.
1284                                  */
1285                                 if (hv_vmbus_protocal_version >=
1286                                     HV_VMBUS_VERSION_WIN8_1) {
1287                                         ifp->if_hwassist |=
1288                                             (CSUM_TCP | CSUM_UDP);
1289                                 } else {
1290                                         ifp->if_hwassist |= CSUM_TCP;
1291                                 }
1292                         }
1293                 }
1294
1295                 if (mask & IFCAP_RXCSUM) {
1296                         if (IFCAP_RXCSUM & ifp->if_capenable) {
1297                                 ifp->if_capenable &= ~IFCAP_RXCSUM;
1298                         } else {
1299                                 ifp->if_capenable |= IFCAP_RXCSUM;
1300                         }
1301                 }
1302                 if (mask & IFCAP_LRO)
1303                         ifp->if_capenable ^= IFCAP_LRO;
1304
1305                 if (mask & IFCAP_TSO4) {
1306                         ifp->if_capenable ^= IFCAP_TSO4;
1307                         ifp->if_hwassist ^= CSUM_IP_TSO;
1308                 }
1309
1310                 if (mask & IFCAP_TSO6) {
1311                         ifp->if_capenable ^= IFCAP_TSO6;
1312                         ifp->if_hwassist ^= CSUM_IP6_TSO;
1313                 }
1314
1315                 error = 0;
1316                 break;
1317         case SIOCADDMULTI:
1318         case SIOCDELMULTI:
1319 #ifdef notyet
1320                 /* Fixme:  Multicast mode? */
1321                 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1322                         NV_LOCK(sc);
1323                         netvsc_setmulti(sc);
1324                         NV_UNLOCK(sc);
1325                         error = 0;
1326                 }
1327 #endif
1328                 error = EINVAL;
1329                 break;
1330         case SIOCSIFMEDIA:
1331         case SIOCGIFMEDIA:
1332                 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1333                 break;
1334         default:
1335                 error = ether_ioctl(ifp, cmd, data);
1336                 break;
1337         }
1338
1339         return (error);
1340 }
1341
1342 /*
1343  *
1344  */
1345 static void
1346 hn_stop(hn_softc_t *sc)
1347 {
1348         struct ifnet *ifp;
1349         int ret;
1350         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1351
1352         ifp = sc->hn_ifp;
1353
1354         if (bootverbose)
1355                 printf(" Closing Device ...\n");
1356
1357         ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1358         if_link_state_change(ifp, LINK_STATE_DOWN);
1359         sc->hn_initdone = 0;
1360
1361         ret = hv_rf_on_close(device_ctx);
1362 }
1363
1364 /*
1365  * FreeBSD transmit entry point
1366  */
1367 static void
1368 hn_start(struct ifnet *ifp)
1369 {
1370         hn_softc_t *sc;
1371
1372         sc = ifp->if_softc;
1373         NV_LOCK(sc);
1374         if (sc->temp_unusable) {
1375                 NV_UNLOCK(sc);
1376                 return;
1377         }
1378         hn_start_locked(ifp);
1379         NV_UNLOCK(sc);
1380 }
1381
1382 /*
1383  *
1384  */
1385 static void
1386 hn_ifinit_locked(hn_softc_t *sc)
1387 {
1388         struct ifnet *ifp;
1389         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1390         int ret;
1391
1392         ifp = sc->hn_ifp;
1393
1394         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1395                 return;
1396         }
1397
1398         hv_promisc_mode = 1;
1399
1400         ret = hv_rf_on_open(device_ctx);
1401         if (ret != 0) {
1402                 return;
1403         } else {
1404                 sc->hn_initdone = 1;
1405         }
1406         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1407         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1408         if_link_state_change(ifp, LINK_STATE_UP);
1409 }
1410
1411 /*
1412  *
1413  */
1414 static void
1415 hn_ifinit(void *xsc)
1416 {
1417         hn_softc_t *sc = xsc;
1418
1419         NV_LOCK(sc);
1420         if (sc->temp_unusable) {
1421                 NV_UNLOCK(sc);
1422                 return;
1423         }
1424         sc->temp_unusable = TRUE;
1425         NV_UNLOCK(sc);
1426
1427         hn_ifinit_locked(sc);
1428
1429         NV_LOCK(sc);
1430         sc->temp_unusable = FALSE;
1431         NV_UNLOCK(sc);
1432 }
1433
1434 #ifdef LATER
1435 /*
1436  *
1437  */
1438 static void
1439 hn_watchdog(struct ifnet *ifp)
1440 {
1441         hn_softc_t *sc;
1442         sc = ifp->if_softc;
1443
1444         printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1445         hn_ifinit(sc);    /*???*/
1446         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1447 }
1448 #endif
1449
1450 #ifdef HN_LRO_HIWAT
1451 static int
1452 hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS)
1453 {
1454         struct hn_softc *sc = arg1;
1455         int hiwat, error;
1456
1457         hiwat = sc->hn_lro_hiwat;
1458         error = sysctl_handle_int(oidp, &hiwat, 0, req);
1459         if (error || req->newptr == NULL)
1460                 return error;
1461
1462         if (!HN_LRO_HIWAT_ISVALID(sc, hiwat))
1463                 return EINVAL;
1464
1465         if (sc->hn_lro_hiwat != hiwat)
1466                 hn_set_lro_hiwat(sc, hiwat);
1467         return 0;
1468 }
1469 #endif  /* HN_LRO_HIWAT */
1470
1471 static int
1472 hn_check_iplen(const struct mbuf *m, int hoff)
1473 {
1474         const struct ip *ip;
1475         int len, iphlen, iplen;
1476         const struct tcphdr *th;
1477         int thoff;                              /* TCP data offset */
1478
1479         len = hoff + sizeof(struct ip);
1480
1481         /* The packet must be at least the size of an IP header. */
1482         if (m->m_pkthdr.len < len)
1483                 return IPPROTO_DONE;
1484
1485         /* The fixed IP header must reside completely in the first mbuf. */
1486         if (m->m_len < len)
1487                 return IPPROTO_DONE;
1488
1489         ip = mtodo(m, hoff);
1490
1491         /* Bound check the packet's stated IP header length. */
1492         iphlen = ip->ip_hl << 2;
1493         if (iphlen < sizeof(struct ip))         /* minimum header length */
1494                 return IPPROTO_DONE;
1495
1496         /* The full IP header must reside completely in the one mbuf. */
1497         if (m->m_len < hoff + iphlen)
1498                 return IPPROTO_DONE;
1499
1500         iplen = ntohs(ip->ip_len);
1501
1502         /*
1503          * Check that the amount of data in the buffers is as
1504          * at least much as the IP header would have us expect.
1505          */
1506         if (m->m_pkthdr.len < hoff + iplen)
1507                 return IPPROTO_DONE;
1508
1509         /*
1510          * Ignore IP fragments.
1511          */
1512         if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1513                 return IPPROTO_DONE;
1514
1515         /*
1516          * The TCP/IP or UDP/IP header must be entirely contained within
1517          * the first fragment of a packet.
1518          */
1519         switch (ip->ip_p) {
1520         case IPPROTO_TCP:
1521                 if (iplen < iphlen + sizeof(struct tcphdr))
1522                         return IPPROTO_DONE;
1523                 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1524                         return IPPROTO_DONE;
1525                 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1526                 thoff = th->th_off << 2;
1527                 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1528                         return IPPROTO_DONE;
1529                 if (m->m_len < hoff + iphlen + thoff)
1530                         return IPPROTO_DONE;
1531                 break;
1532         case IPPROTO_UDP:
1533                 if (iplen < iphlen + sizeof(struct udphdr))
1534                         return IPPROTO_DONE;
1535                 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1536                         return IPPROTO_DONE;
1537                 break;
1538         default:
1539                 if (iplen < iphlen)
1540                         return IPPROTO_DONE;
1541                 break;
1542         }
1543         return ip->ip_p;
1544 }
1545
1546 static device_method_t netvsc_methods[] = {
1547         /* Device interface */
1548         DEVMETHOD(device_probe,         netvsc_probe),
1549         DEVMETHOD(device_attach,        netvsc_attach),
1550         DEVMETHOD(device_detach,        netvsc_detach),
1551         DEVMETHOD(device_shutdown,      netvsc_shutdown),
1552
1553         { 0, 0 }
1554 };
1555
1556 static driver_t netvsc_driver = {
1557         NETVSC_DEVNAME,
1558         netvsc_methods,
1559         sizeof(hn_softc_t)
1560 };
1561
1562 static devclass_t netvsc_devclass;
1563
1564 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
1565 MODULE_VERSION(hn, 1);
1566 MODULE_DEPEND(hn, vmbus, 1, 1, 1);