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 #if defined(INET) || defined(INET6)
 395         tcp_lro_init(&sc->hn_lro);
 396         /* Driver private LRO settings */
 397         sc->hn_lro.ifp = ifp;
 398 #ifdef HN_LRO_HIWAT
 399         sc->hn_lro.lro_hiwat = sc->hn_lro_hiwat;
 400 #endif
 401 #endif  /* INET || INET6 */
 402
 403         ether_ifattach(ifp, device_info.mac_addr);
 404
 405         ctx = device_get_sysctl_ctx(dev);
 406         child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
 407
 408         SYSCTL_ADD_U64(ctx, child, OID_AUTO, "lro_queued",
 409             CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued");
 410         SYSCTL_ADD_U64(ctx, child, OID_AUTO, "lro_flushed",
 411             CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed");
 412         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried",
 413             CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries");
 414 #ifdef HN_LRO_HIWAT
 415         SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_hiwat",
 416             CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_hiwat_sysctl,
 417             "I", "LRO high watermark");
 418 #endif
 419         SYSCTL_ADD_INT(ctx, child, OID_AUTO, "trust_hosttcp",
 420             CTLFLAG_RW, &sc->hn_trust_hosttcp, 0,
 421             "Trust tcp segement verification on host side, "
 422             "when csum info is missing");
 423         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip",
 424             CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP");
 425         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp",
 426             CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP");
 427         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted",
 428             CTLFLAG_RW, &sc->hn_csum_trusted,
 429             "# of TCP segements that we trust host's csum verification");
 430         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts",
 431             CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received");
 432
 433         if (unit == 0) {
 434                 struct sysctl_ctx_list *dc_ctx;
 435                 struct sysctl_oid_list *dc_child;
 436                 devclass_t dc;
 437
 438                 /*
 439                  * Add sysctl nodes for devclass
 440                  */
 441                 dc = device_get_devclass(dev);
 442                 dc_ctx = devclass_get_sysctl_ctx(dc);
 443                 dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc));
 444
 445                 SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp",
 446                     CTLFLAG_RD, &hn_trust_hosttcp, 0,
 447                     "Trust tcp segement verification on host side, "
 448                     "when csum info is missing (global setting)");
 449         }
 450
 451         return (0);
 452 }
 453
 454 /*
 455  * Standard detach entry point
 456  */
 457 static int
 458 netvsc_detach(device_t dev)
 459 {
 460         struct hn_softc *sc = device_get_softc(dev);
 461         struct hv_device *hv_device = vmbus_get_devctx(dev);
 462
 463         if (bootverbose)
 464                 printf("netvsc_detach\n");
 465
 466         /*
 467          * XXXKYS:  Need to clean up all our
 468          * driver state; this is the driver
 469          * unloading.
 470          */
 471
 472         /*
 473          * XXXKYS:  Need to stop outgoing traffic and unregister
 474          * the netdevice.
 475          */
 476
 477         hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL);
 478
 479         ifmedia_removeall(&sc->hn_media);
 480 #if defined(INET) || defined(INET6)
 481         tcp_lro_free(&sc->hn_lro);
 482 #endif
 483
 484         return (0);
 485 }
 486
 487 /*
 488  * Standard shutdown entry point
 489  */
 490 static int
 491 netvsc_shutdown(device_t dev)
 492 {
 493         return (0);
 494 }
 495
 496 /*
 497  * Send completion processing
 498  *
 499  * Note:  It looks like offset 0 of buf is reserved to hold the softc
 500  * pointer.  The sc pointer is not currently needed in this function, and
 501  * it is not presently populated by the TX function.
 502  */
 503 void
 504 netvsc_xmit_completion(void *context)
 505 {
 506         netvsc_packet *packet = (netvsc_packet *)context;
 507         struct mbuf *mb;
 508         uint8_t *buf;
 509
 510         mb = (struct mbuf *)(uintptr_t)packet->compl.send.send_completion_tid;
 511         buf = ((uint8_t *)packet) - HV_NV_PACKET_OFFSET_IN_BUF;
 512
 513         free(buf, M_NETVSC);
 514
 515         if (mb != NULL) {
 516                 m_freem(mb);
 517         }
 518 }
 519
 520 /*
 521  * Start a transmit of one or more packets
 522  */
 523 static int
 524 hn_start_locked(struct ifnet *ifp)
 525 {
 526         hn_softc_t *sc = ifp->if_softc;
 527         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
 528         netvsc_dev *net_dev = sc->net_dev;
 529         device_t dev = device_ctx->device;
 530         uint8_t *buf;
 531         netvsc_packet *packet;
 532         struct mbuf *m_head, *m;
 533         struct mbuf *mc_head = NULL;
 534         struct ether_vlan_header *eh;
 535         rndis_msg *rndis_mesg;
 536         rndis_packet *rndis_pkt;
 537         rndis_per_packet_info *rppi;
 538         ndis_8021q_info *rppi_vlan_info;
 539         rndis_tcp_ip_csum_info *csum_info;
 540         rndis_tcp_tso_info *tso_info;
 541         int ether_len;
 542         int i;
 543         int num_frags;
 544         int len;
 545         int retries = 0;
 546         int ret = 0;
 547         uint32_t rndis_msg_size = 0;
 548         uint32_t trans_proto_type;
 549         uint32_t send_buf_section_idx =
 550             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
 551
 552         while (!IFQ_DRV_IS_EMPTY(&sc->hn_ifp->if_snd)) {
 553                 IFQ_DRV_DEQUEUE(&sc->hn_ifp->if_snd, m_head);
 554                 if (m_head == NULL) {
 555                         break;
 556                 }
 557
 558                 len = 0;
 559                 num_frags = 0;
 560
 561                 /* Walk the mbuf list computing total length and num frags */
 562                 for (m = m_head; m != NULL; m = m->m_next) {
 563                         if (m->m_len != 0) {
 564                                 num_frags++;
 565                                 len += m->m_len;
 566                         }
 567                 }
 568
 569                 /*
 570                  * Reserve the number of pages requested.  Currently,
 571                  * one page is reserved for the message in the RNDIS
 572                  * filter packet
 573                  */
 574                 num_frags += HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
 575
 576                 /* If exceeds # page_buffers in netvsc_packet */
 577                 if (num_frags > NETVSC_PACKET_MAXPAGE) {
 578                         device_printf(dev, "exceed max page buffers,%d,%d\n",
 579                             num_frags, NETVSC_PACKET_MAXPAGE);
 580                         m_freem(m_head);
 581                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 582                         return (EINVAL);
 583                 }
 584
 585                 /*
 586                  * Allocate a buffer with space for a netvsc packet plus a
 587                  * number of reserved areas.  First comes a (currently 16
 588                  * bytes, currently unused) reserved data area.  Second is
 589                  * the netvsc_packet. Third is an area reserved for an
 590                  * rndis_filter_packet struct. Fourth (optional) is a
 591                  * rndis_per_packet_info struct.
 592                  * Changed malloc to M_NOWAIT to avoid sleep under spin lock.
 593                  * No longer reserving extra space for page buffers, as they
 594                  * are already part of the netvsc_packet.
 595                  */
 596                 buf = malloc(HV_NV_PACKET_OFFSET_IN_BUF +
 597                         sizeof(netvsc_packet) +
 598                         sizeof(rndis_msg) +
 599                         RNDIS_VLAN_PPI_SIZE +
 600                         RNDIS_TSO_PPI_SIZE +
 601                         RNDIS_CSUM_PPI_SIZE,
 602                         M_NETVSC, M_ZERO | M_NOWAIT);
 603                 if (buf == NULL) {
 604                         device_printf(dev, "hn:malloc packet failed\n");
 605                         m_freem(m_head);
 606                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 607                         return (ENOMEM);
 608                 }
 609
 610                 packet = (netvsc_packet *)(buf + HV_NV_PACKET_OFFSET_IN_BUF);
 611                 *(vm_offset_t *)buf = HV_NV_SC_PTR_OFFSET_IN_BUF;
 612
 613                 packet->is_data_pkt = TRUE;
 614
 615                 /* Set up the rndis header */
 616                 packet->page_buf_count = num_frags;
 617
 618                 /* Initialize it from the mbuf */
 619                 packet->tot_data_buf_len = len;
 620
 621                 /*
 622                  * extension points to the area reserved for the
 623                  * rndis_filter_packet, which is placed just after
 624                  * the netvsc_packet (and rppi struct, if present;
 625                  * length is updated later).
 626                  */
 627                 packet->rndis_mesg = packet + 1;
 628                 rndis_mesg = (rndis_msg *)packet->rndis_mesg;
 629                 rndis_mesg->ndis_msg_type = REMOTE_NDIS_PACKET_MSG;
 630
 631                 rndis_pkt = &rndis_mesg->msg.packet;
 632                 rndis_pkt->data_offset = sizeof(rndis_packet);
 633                 rndis_pkt->data_length = packet->tot_data_buf_len;
 634                 rndis_pkt->per_pkt_info_offset = sizeof(rndis_packet);
 635
 636                 rndis_msg_size = RNDIS_MESSAGE_SIZE(rndis_packet);
 637
 638                 /*
 639                  * If the Hyper-V infrastructure needs to embed a VLAN tag,
 640                  * initialize netvsc_packet and rppi struct values as needed.
 641                  */
 642                 if (m_head->m_flags & M_VLANTAG) {
 643                         /*
 644                          * set up some additional fields so the Hyper-V infrastructure will stuff the VLAN tag
 645                          * into the frame.
 646                          */
 647                         packet->vlan_tci = m_head->m_pkthdr.ether_vtag;
 648
 649                         rndis_msg_size += RNDIS_VLAN_PPI_SIZE;
 650
 651                         rppi = hv_set_rppi_data(rndis_mesg, RNDIS_VLAN_PPI_SIZE,
 652                             ieee_8021q_info);
 653
 654                         /* VLAN info immediately follows rppi struct */
 655                         rppi_vlan_info = (ndis_8021q_info *)((char*)rppi +
 656                             rppi->per_packet_info_offset);
 657                         /* FreeBSD does not support CFI or priority */
 658                         rppi_vlan_info->u1.s1.vlan_id =
 659                             packet->vlan_tci & 0xfff;
 660                 }
 661
 662                 /* Only check the flags for outbound and ignore the ones for inbound */
 663                 if (0 == (m_head->m_pkthdr.csum_flags & HV_CSUM_FOR_OUTBOUND)) {
 664                         goto pre_send;
 665                 }
 666
 667                 eh = mtod(m_head, struct ether_vlan_header*);
 668                 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
 669                         ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
 670                 } else {
 671                         ether_len = ETHER_HDR_LEN;
 672                 }
 673
 674                 trans_proto_type = get_transport_proto_type(m_head);
 675                 if (TRANSPORT_TYPE_NOT_IP == trans_proto_type) {
 676                         goto pre_send;
 677                 }
 678
 679                 /*
 680                  * TSO packet needless to setup the send side checksum
 681                  * offload.
 682                  */
 683                 if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
 684                         goto do_tso;
 685                 }
 686
 687                 /* setup checksum offload */
 688                 rndis_msg_size += RNDIS_CSUM_PPI_SIZE;
 689                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_CSUM_PPI_SIZE,
 690                     tcpip_chksum_info);
 691                 csum_info = (rndis_tcp_ip_csum_info *)((char*)rppi +
 692                     rppi->per_packet_info_offset);
 693
 694                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
 695                         csum_info->xmit.is_ipv4 = 1;
 696                 } else {
 697                         csum_info->xmit.is_ipv6 = 1;
 698                 }
 699
 700                 if (trans_proto_type & TYPE_TCP) {
 701                         csum_info->xmit.tcp_csum = 1;
 702                         csum_info->xmit.tcp_header_offset = 0;
 703                 } else if (trans_proto_type & TYPE_UDP) {
 704                         csum_info->xmit.udp_csum = 1;
 705                 }
 706
 707                 goto pre_send;
 708
 709 do_tso:
 710                 /* setup TCP segmentation offload */
 711                 rndis_msg_size += RNDIS_TSO_PPI_SIZE;
 712                 rppi = hv_set_rppi_data(rndis_mesg, RNDIS_TSO_PPI_SIZE,
 713                     tcp_large_send_info);
 714
 715                 tso_info = (rndis_tcp_tso_info *)((char *)rppi +
 716                     rppi->per_packet_info_offset);
 717                 tso_info->lso_v2_xmit.type =
 718                     RNDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 719
 720 #ifdef INET
 721                 if (trans_proto_type & (TYPE_IPV4 << 16)) {
 722                         struct ip *ip =
 723                             (struct ip *)(m_head->m_data + ether_len);
 724                         unsigned long iph_len = ip->ip_hl << 2;
 725                         struct tcphdr *th =
 726                             (struct tcphdr *)((caddr_t)ip + iph_len);
 727
 728                         tso_info->lso_v2_xmit.ip_version =
 729                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 730                         ip->ip_len = 0;
 731                         ip->ip_sum = 0;
 732
 733                         th->th_sum = in_pseudo(ip->ip_src.s_addr,
 734                             ip->ip_dst.s_addr,
 735                             htons(IPPROTO_TCP));
 736                 }
 737 #endif
 738 #if defined(INET6) && defined(INET)
 739                 else
 740 #endif
 741 #ifdef INET6
 742                 {
 743                         struct ip6_hdr *ip6 =
 744                             (struct ip6_hdr *)(m_head->m_data + ether_len);
 745                         struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
 746
 747                         tso_info->lso_v2_xmit.ip_version =
 748                             RNDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 749                         ip6->ip6_plen = 0;
 750                         th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
 751                 }
 752 #endif
 753                 tso_info->lso_v2_xmit.tcp_header_offset = 0;
 754                 tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz;
 755
 756 pre_send:
 757                 rndis_mesg->msg_len = packet->tot_data_buf_len + rndis_msg_size;
 758                 packet->tot_data_buf_len = rndis_mesg->msg_len;
 759
 760                 /* send packet with send buffer */
 761                 if (packet->tot_data_buf_len < net_dev->send_section_size) {
 762                         send_buf_section_idx =
 763                             hv_nv_get_next_send_section(net_dev);
 764                         if (send_buf_section_idx !=
 765                             NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX) {
 766                                 char *dest = ((char *)net_dev->send_buf +
 767                                     send_buf_section_idx *
 768                                     net_dev->send_section_size);
 769
 770                                 memcpy(dest, rndis_mesg, rndis_msg_size);
 771                                 dest += rndis_msg_size;
 772                                 for (m = m_head; m != NULL; m = m->m_next) {
 773                                         if (m->m_len) {
 774                                                 memcpy(dest,
 775                                                     (void *)mtod(m, vm_offset_t),
 776                                                     m->m_len);
 777                                                 dest += m->m_len;
 778                                         }
 779                                 }
 780
 781                                 packet->send_buf_section_idx =
 782                                     send_buf_section_idx;
 783                                 packet->send_buf_section_size =
 784                                     packet->tot_data_buf_len;
 785                                 packet->page_buf_count = 0;
 786                                 goto do_send;
 787                         }
 788                 }
 789
 790                 /* send packet with page buffer */
 791                 packet->page_buffers[0].pfn =
 792                     atop(hv_get_phys_addr(rndis_mesg));
 793                 packet->page_buffers[0].offset =
 794                     (unsigned long)rndis_mesg & PAGE_MASK;
 795                 packet->page_buffers[0].length = rndis_msg_size;
 796
 797                 /*
 798                  * Fill the page buffers with mbuf info starting at index
 799                  * HV_RF_NUM_TX_RESERVED_PAGE_BUFS.
 800                  */
 801                 i = HV_RF_NUM_TX_RESERVED_PAGE_BUFS;
 802                 for (m = m_head; m != NULL; m = m->m_next) {
 803                         if (m->m_len) {
 804                                 vm_offset_t paddr =
 805                                     vtophys(mtod(m, vm_offset_t));
 806                                 packet->page_buffers[i].pfn =
 807                                     paddr >> PAGE_SHIFT;
 808                                 packet->page_buffers[i].offset =
 809                                     paddr & (PAGE_SIZE - 1);
 810                                 packet->page_buffers[i].length = m->m_len;
 811                                 i++;
 812                         }
 813                 }
 814
 815                 packet->send_buf_section_idx =
 816                     NVSP_1_CHIMNEY_SEND_INVALID_SECTION_INDEX;
 817                 packet->send_buf_section_size = 0;
 818
 819 do_send:
 820
 821                 /*
 822                  * If bpf, copy the mbuf chain.  This is less expensive than
 823                  * it appears; the mbuf clusters are not copied, only their
 824                  * reference counts are incremented.
 825                  * Needed to avoid a race condition where the completion
 826                  * callback is invoked, freeing the mbuf chain, before the
 827                  * bpf_mtap code has a chance to run.
 828                  */
 829                 if (ifp->if_bpf) {
 830                         mc_head = m_copypacket(m_head, M_NOWAIT);
 831                 }
 832 retry_send:
 833                 /* Set the completion routine */
 834                 packet->compl.send.on_send_completion = netvsc_xmit_completion;
 835                 packet->compl.send.send_completion_context = packet;
 836                 packet->compl.send.send_completion_tid = (uint64_t)(uintptr_t)m_head;
 837
 838                 /* Removed critical_enter(), does not appear necessary */
 839                 ret = hv_nv_on_send(device_ctx, packet);
 840                 if (ret == 0) {
 841                         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
 842                         /* if bpf && mc_head, call bpf_mtap code */
 843                         if (mc_head) {
 844                                 ETHER_BPF_MTAP(ifp, mc_head);
 845                         }
 846                 } else {
 847                         retries++;
 848                         if (retries < 4) {
 849                                 goto retry_send;
 850                         }
 851
 852                         IF_PREPEND(&ifp->if_snd, m_head);
 853                         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 854
 855                         /*
 856                          * Null the mbuf pointer so the completion function
 857                          * does not free the mbuf chain.  We just pushed the
 858                          * mbuf chain back on the if_snd queue.
 859                          */
 860                         packet->compl.send.send_completion_tid = 0;
 861
 862                         /*
 863                          * Release the resources since we will not get any
 864                          * send completion
 865                          */
 866                         netvsc_xmit_completion(packet);
 867                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 868                 }
 869
 870                 /* if bpf && mc_head, free the mbuf chain copy */
 871                 if (mc_head) {
 872                         m_freem(mc_head);
 873                 }
 874         }
 875
 876         return (ret);
 877 }
 878
 879 /*
 880  * Link up/down notification
 881  */
 882 void
 883 netvsc_linkstatus_callback(struct hv_device *device_obj, uint32_t status)
 884 {
 885         hn_softc_t *sc = device_get_softc(device_obj->device);
 886
 887         if (sc == NULL) {
 888                 return;
 889         }
 890
 891         if (status == 1) {
 892                 sc->hn_carrier = 1;
 893         } else {
 894                 sc->hn_carrier = 0;
 895         }
 896 }
 897
 898 /*
 899  * Append the specified data to the indicated mbuf chain,
 900  * Extend the mbuf chain if the new data does not fit in
 901  * existing space.
 902  *
 903  * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
 904  * There should be an equivalent in the kernel mbuf code,
 905  * but there does not appear to be one yet.
 906  *
 907  * Differs from m_append() in that additional mbufs are
 908  * allocated with cluster size MJUMPAGESIZE, and filled
 909  * accordingly.
 910  *
 911  * Return 1 if able to complete the job; otherwise 0.
 912  */
 913 static int
 914 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
 915 {
 916         struct mbuf *m, *n;
 917         int remainder, space;
 918
 919         for (m = m0; m->m_next != NULL; m = m->m_next)
 920                 ;
 921         remainder = len;
 922         space = M_TRAILINGSPACE(m);
 923         if (space > 0) {
 924                 /*
 925                  * Copy into available space.
 926                  */
 927                 if (space > remainder)
 928                         space = remainder;
 929                 bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
 930                 m->m_len += space;
 931                 cp += space;
 932                 remainder -= space;
 933         }
 934         while (remainder > 0) {
 935                 /*
 936                  * Allocate a new mbuf; could check space
 937                  * and allocate a cluster instead.
 938                  */
 939                 n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
 940                 if (n == NULL)
 941                         break;
 942                 n->m_len = min(MJUMPAGESIZE, remainder);
 943                 bcopy(cp, mtod(n, caddr_t), n->m_len);
 944                 cp += n->m_len;
 945                 remainder -= n->m_len;
 946                 m->m_next = n;
 947                 m = n;
 948         }
 949         if (m0->m_flags & M_PKTHDR)
 950                 m0->m_pkthdr.len += len - remainder;
 951
 952         return (remainder == 0);
 953 }
 954
 955
 956 /*
 957  * Called when we receive a data packet from the "wire" on the
 958  * specified device
 959  *
 960  * Note:  This is no longer used as a callback
 961  */
 962 int
 963 netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet,
 964     rndis_tcp_ip_csum_info *csum_info)
 965 {
 966         hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device);
 967         struct mbuf *m_new;
 968         struct ifnet *ifp;
 969         device_t dev = device_ctx->device;
 970         int size, do_lro = 0;
 971
 972         if (sc == NULL) {
 973                 return (0); /* TODO: KYS how can this be! */
 974         }
 975
 976         ifp = sc->hn_ifp;
 977
 978         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
 979                 return (0);
 980         }
 981
 982         /*
 983          * Bail out if packet contains more data than configured MTU.
 984          */
 985         if (packet->tot_data_buf_len > (ifp->if_mtu + ETHER_HDR_LEN)) {
 986                 return (0);
 987         } else if (packet->tot_data_buf_len <= MHLEN) {
 988                 m_new = m_gethdr(M_NOWAIT, MT_DATA);
 989                 if (m_new == NULL)
 990                         return (0);
 991                 memcpy(mtod(m_new, void *), packet->data,
 992                     packet->tot_data_buf_len);
 993                 m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len;
 994                 sc->hn_small_pkts++;
 995         } else {
 996                 /*
 997                  * Get an mbuf with a cluster.  For packets 2K or less,
 998                  * get a standard 2K cluster.  For anything larger, get a
 999                  * 4K cluster.  Any buffers larger than 4K can cause problems
1000                  * if looped around to the Hyper-V TX channel, so avoid them.
1001                  */
1002                 size = MCLBYTES;
1003                 if (packet->tot_data_buf_len > MCLBYTES) {
1004                         /* 4096 */
1005                         size = MJUMPAGESIZE;
1006                 }
1007
1008                 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
1009                 if (m_new == NULL) {
1010                         device_printf(dev, "alloc mbuf failed.\n");
1011                         return (0);
1012                 }
1013
1014                 hv_m_append(m_new, packet->tot_data_buf_len, packet->data);
1015         }
1016         m_new->m_pkthdr.rcvif = ifp;
1017
1018         /* receive side checksum offload */
1019         if (NULL != csum_info) {
1020                 /* IP csum offload */
1021                 if (csum_info->receive.ip_csum_succeeded) {
1022                         m_new->m_pkthdr.csum_flags |=
1023                             (CSUM_IP_CHECKED | CSUM_IP_VALID);
1024                         sc->hn_csum_ip++;
1025                 }
1026
1027                 /* TCP csum offload */
1028                 if (csum_info->receive.tcp_csum_succeeded) {
1029                         m_new->m_pkthdr.csum_flags |=
1030                             (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1031                         m_new->m_pkthdr.csum_data = 0xffff;
1032                         sc->hn_csum_tcp++;
1033                 }
1034
1035                 if (csum_info->receive.ip_csum_succeeded &&
1036                     csum_info->receive.tcp_csum_succeeded)
1037                         do_lro = 1;
1038         } else {
1039                 const struct ether_header *eh;
1040                 uint16_t etype;
1041                 int hoff;
1042
1043                 hoff = sizeof(*eh);
1044                 if (m_new->m_len < hoff)
1045                         goto skip;
1046                 eh = mtod(m_new, struct ether_header *);
1047                 etype = ntohs(eh->ether_type);
1048                 if (etype == ETHERTYPE_VLAN) {
1049                         const struct ether_vlan_header *evl;
1050
1051                         hoff = sizeof(*evl);
1052                         if (m_new->m_len < hoff)
1053                                 goto skip;
1054                         evl = mtod(m_new, struct ether_vlan_header *);
1055                         etype = ntohs(evl->evl_proto);
1056                 }
1057
1058                 if (etype == ETHERTYPE_IP) {
1059                         int pr;
1060
1061                         pr = hn_check_iplen(m_new, hoff);
1062                         if (pr == IPPROTO_TCP) {
1063                                 if (sc->hn_trust_hosttcp) {
1064                                         sc->hn_csum_trusted++;
1065                                         m_new->m_pkthdr.csum_flags |=
1066                                            (CSUM_IP_CHECKED | CSUM_IP_VALID |
1067                                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1068                                         m_new->m_pkthdr.csum_data = 0xffff;
1069                                 }
1070                                 /* Rely on SW csum verification though... */
1071                                 do_lro = 1;
1072                         }
1073                 }
1074         }
1075 skip:
1076         if ((packet->vlan_tci != 0) &&
1077             (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0) {
1078                 m_new->m_pkthdr.ether_vtag = packet->vlan_tci;
1079                 m_new->m_flags |= M_VLANTAG;
1080         }
1081
1082         /*
1083          * Note:  Moved RX completion back to hv_nv_on_receive() so all
1084          * messages (not just data messages) will trigger a response.
1085          */
1086
1087         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1088
1089         if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
1090 #if defined(INET) || defined(INET6)
1091                 struct lro_ctrl *lro = &sc->hn_lro;
1092
1093                 if (lro->lro_cnt) {
1094                         sc->hn_lro_tried++;
1095                         if (tcp_lro_rx(lro, m_new, 0) == 0) {
1096                                 /* DONE! */
1097                                 return 0;
1098                         }
1099                 }
1100 #endif
1101         }
1102
1103         /* We're not holding the lock here, so don't release it */
1104         (*ifp->if_input)(ifp, m_new);
1105
1106         return (0);
1107 }
1108
1109 void
1110 netvsc_recv_rollup(struct hv_device *device_ctx)
1111 {
1112 #if defined(INET) || defined(INET6)
1113         hn_softc_t *sc = device_get_softc(device_ctx->device);
1114         struct lro_ctrl *lro = &sc->hn_lro;
1115         struct lro_entry *queued;
1116
1117         while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) {
1118                 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1119                 tcp_lro_flush(lro, queued);
1120         }
1121 #endif
1122 }
1123
1124 /*
1125  * Rules for using sc->temp_unusable:
1126  * 1.  sc->temp_unusable can only be read or written while holding NV_LOCK()
1127  * 2.  code reading sc->temp_unusable under NV_LOCK(), and finding
1128  *     sc->temp_unusable set, must release NV_LOCK() and exit
1129  * 3.  to retain exclusive control of the interface,
1130  *     sc->temp_unusable must be set by code before releasing NV_LOCK()
1131  * 4.  only code setting sc->temp_unusable can clear sc->temp_unusable
1132  * 5.  code setting sc->temp_unusable must eventually clear sc->temp_unusable
1133  */
1134
1135 /*
1136  * Standard ioctl entry point.  Called when the user wants to configure
1137  * the interface.
1138  */
1139 static int
1140 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1141 {
1142         hn_softc_t *sc = ifp->if_softc;
1143         struct ifreq *ifr = (struct ifreq *)data;
1144 #ifdef INET
1145         struct ifaddr *ifa = (struct ifaddr *)data;
1146 #endif
1147         netvsc_device_info device_info;
1148         struct hv_device *hn_dev;
1149         int mask, error = 0;
1150         int retry_cnt = 500;
1151
1152         switch(cmd) {
1153
1154         case SIOCSIFADDR:
1155 #ifdef INET
1156                 if (ifa->ifa_addr->sa_family == AF_INET) {
1157                         ifp->if_flags |= IFF_UP;
1158                         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1159                                 hn_ifinit(sc);
1160                         arp_ifinit(ifp, ifa);
1161                 } else
1162 #endif
1163                 error = ether_ioctl(ifp, cmd, data);
1164                 break;
1165         case SIOCSIFMTU:
1166                 hn_dev = vmbus_get_devctx(sc->hn_dev);
1167
1168                 /* Check MTU value change */
1169                 if (ifp->if_mtu == ifr->ifr_mtu)
1170                         break;
1171
1172                 if (ifr->ifr_mtu > NETVSC_MAX_CONFIGURABLE_MTU) {
1173                         error = EINVAL;
1174                         break;
1175                 }
1176
1177                 /* Obtain and record requested MTU */
1178                 ifp->if_mtu = ifr->ifr_mtu;
1179                 /*
1180                  * Make sure that LRO high watermark is still valid,
1181                  * after MTU change (the 2*MTU limit).
1182                  */
1183                 if (!HN_LRO_HIWAT_ISVALID(sc, sc->hn_lro_hiwat))
1184                         hn_set_lro_hiwat(sc, HN_LRO_HIWAT_MTULIM(ifp));
1185
1186                 do {
1187                         NV_LOCK(sc);
1188                         if (!sc->temp_unusable) {
1189                                 sc->temp_unusable = TRUE;
1190                                 retry_cnt = -1;
1191                         }
1192                         NV_UNLOCK(sc);
1193                         if (retry_cnt > 0) {
1194                                 retry_cnt--;
1195                                 DELAY(5 * 1000);
1196                         }
1197                 } while (retry_cnt > 0);
1198
1199                 if (retry_cnt == 0) {
1200                         error = EINVAL;
1201                         break;
1202                 }
1203
1204                 /* We must remove and add back the device to cause the new
1205                  * MTU to take effect.  This includes tearing down, but not
1206                  * deleting the channel, then bringing it back up.
1207                  */
1208                 error = hv_rf_on_device_remove(hn_dev, HV_RF_NV_RETAIN_CHANNEL);
1209                 if (error) {
1210                         NV_LOCK(sc);
1211                         sc->temp_unusable = FALSE;
1212                         NV_UNLOCK(sc);
1213                         break;
1214                 }
1215                 error = hv_rf_on_device_add(hn_dev, &device_info);
1216                 if (error) {
1217                         NV_LOCK(sc);
1218                         sc->temp_unusable = FALSE;
1219                         NV_UNLOCK(sc);
1220                         break;
1221                 }
1222
1223                 hn_ifinit_locked(sc);
1224
1225                 NV_LOCK(sc);
1226                 sc->temp_unusable = FALSE;
1227                 NV_UNLOCK(sc);
1228                 break;
1229         case SIOCSIFFLAGS:
1230                 do {
1231                        NV_LOCK(sc);
1232                        if (!sc->temp_unusable) {
1233                                sc->temp_unusable = TRUE;
1234                                retry_cnt = -1;
1235                        }
1236                        NV_UNLOCK(sc);
1237                        if (retry_cnt > 0) {
1238                                 retry_cnt--;
1239                                 DELAY(5 * 1000);
1240                        }
1241                 } while (retry_cnt > 0);
1242
1243                 if (retry_cnt == 0) {
1244                        error = EINVAL;
1245                        break;
1246                 }
1247
1248                 if (ifp->if_flags & IFF_UP) {
1249                         /*
1250                          * If only the state of the PROMISC flag changed,
1251                          * then just use the 'set promisc mode' command
1252                          * instead of reinitializing the entire NIC. Doing
1253                          * a full re-init means reloading the firmware and
1254                          * waiting for it to start up, which may take a
1255                          * second or two.
1256                          */
1257 #ifdef notyet
1258                         /* Fixme:  Promiscuous mode? */
1259                         if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1260                             ifp->if_flags & IFF_PROMISC &&
1261                             !(sc->hn_if_flags & IFF_PROMISC)) {
1262                                 /* do something here for Hyper-V */
1263                         } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1264                             !(ifp->if_flags & IFF_PROMISC) &&
1265                             sc->hn_if_flags & IFF_PROMISC) {
1266                                 /* do something here for Hyper-V */
1267                         } else
1268 #endif
1269                                 hn_ifinit_locked(sc);
1270                 } else {
1271                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1272                                 hn_stop(sc);
1273                         }
1274                 }
1275                 NV_LOCK(sc);
1276                 sc->temp_unusable = FALSE;
1277                 NV_UNLOCK(sc);
1278                 sc->hn_if_flags = ifp->if_flags;
1279                 error = 0;
1280                 break;
1281         case SIOCSIFCAP:
1282                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1283                 if (mask & IFCAP_TXCSUM) {
1284                         if (IFCAP_TXCSUM & ifp->if_capenable) {
1285                                 ifp->if_capenable &= ~IFCAP_TXCSUM;
1286                                 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP);
1287                         } else {
1288                                 ifp->if_capenable |= IFCAP_TXCSUM;
1289                                 /*
1290                                  * Only enable UDP checksum offloading on
1291                                  * Windows Server 2012R2 or later releases.
1292                                  */
1293                                 if (hv_vmbus_protocal_version >=
1294                                     HV_VMBUS_VERSION_WIN8_1) {
1295                                         ifp->if_hwassist |=
1296                                             (CSUM_TCP | CSUM_UDP);
1297                                 } else {
1298                                         ifp->if_hwassist |= CSUM_TCP;
1299                                 }
1300                         }
1301                 }
1302
1303                 if (mask & IFCAP_RXCSUM) {
1304                         if (IFCAP_RXCSUM & ifp->if_capenable) {
1305                                 ifp->if_capenable &= ~IFCAP_RXCSUM;
1306                         } else {
1307                                 ifp->if_capenable |= IFCAP_RXCSUM;
1308                         }
1309                 }
1310                 if (mask & IFCAP_LRO)
1311                         ifp->if_capenable ^= IFCAP_LRO;
1312
1313                 if (mask & IFCAP_TSO4) {
1314                         ifp->if_capenable ^= IFCAP_TSO4;
1315                         ifp->if_hwassist ^= CSUM_IP_TSO;
1316                 }
1317
1318                 if (mask & IFCAP_TSO6) {
1319                         ifp->if_capenable ^= IFCAP_TSO6;
1320                         ifp->if_hwassist ^= CSUM_IP6_TSO;
1321                 }
1322
1323                 error = 0;
1324                 break;
1325         case SIOCADDMULTI:
1326         case SIOCDELMULTI:
1327 #ifdef notyet
1328                 /* Fixme:  Multicast mode? */
1329                 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1330                         NV_LOCK(sc);
1331                         netvsc_setmulti(sc);
1332                         NV_UNLOCK(sc);
1333                         error = 0;
1334                 }
1335 #endif
1336                 error = EINVAL;
1337                 break;
1338         case SIOCSIFMEDIA:
1339         case SIOCGIFMEDIA:
1340                 error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
1341                 break;
1342         default:
1343                 error = ether_ioctl(ifp, cmd, data);
1344                 break;
1345         }
1346
1347         return (error);
1348 }
1349
1350 /*
1351  *
1352  */
1353 static void
1354 hn_stop(hn_softc_t *sc)
1355 {
1356         struct ifnet *ifp;
1357         int ret;
1358         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1359
1360         ifp = sc->hn_ifp;
1361
1362         if (bootverbose)
1363                 printf(" Closing Device ...\n");
1364
1365         ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1366         if_link_state_change(ifp, LINK_STATE_DOWN);
1367         sc->hn_initdone = 0;
1368
1369         ret = hv_rf_on_close(device_ctx);
1370 }
1371
1372 /*
1373  * FreeBSD transmit entry point
1374  */
1375 static void
1376 hn_start(struct ifnet *ifp)
1377 {
1378         hn_softc_t *sc;
1379
1380         sc = ifp->if_softc;
1381         NV_LOCK(sc);
1382         if (sc->temp_unusable) {
1383                 NV_UNLOCK(sc);
1384                 return;
1385         }
1386         hn_start_locked(ifp);
1387         NV_UNLOCK(sc);
1388 }
1389
1390 /*
1391  *
1392  */
1393 static void
1394 hn_ifinit_locked(hn_softc_t *sc)
1395 {
1396         struct ifnet *ifp;
1397         struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev);
1398         int ret;
1399
1400         ifp = sc->hn_ifp;
1401
1402         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1403                 return;
1404         }
1405
1406         hv_promisc_mode = 1;
1407
1408         ret = hv_rf_on_open(device_ctx);
1409         if (ret != 0) {
1410                 return;
1411         } else {
1412                 sc->hn_initdone = 1;
1413         }
1414         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1415         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1416         if_link_state_change(ifp, LINK_STATE_UP);
1417 }
1418
1419 /*
1420  *
1421  */
1422 static void
1423 hn_ifinit(void *xsc)
1424 {
1425         hn_softc_t *sc = xsc;
1426
1427         NV_LOCK(sc);
1428         if (sc->temp_unusable) {
1429                 NV_UNLOCK(sc);
1430                 return;
1431         }
1432         sc->temp_unusable = TRUE;
1433         NV_UNLOCK(sc);
1434
1435         hn_ifinit_locked(sc);
1436
1437         NV_LOCK(sc);
1438         sc->temp_unusable = FALSE;
1439         NV_UNLOCK(sc);
1440 }
1441
1442 #ifdef LATER
1443 /*
1444  *
1445  */
1446 static void
1447 hn_watchdog(struct ifnet *ifp)
1448 {
1449         hn_softc_t *sc;
1450         sc = ifp->if_softc;
1451
1452         printf("hn%d: watchdog timeout -- resetting\n", sc->hn_unit);
1453         hn_ifinit(sc);    /*???*/
1454         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1455 }
1456 #endif
1457
1458 #ifdef HN_LRO_HIWAT
1459 static int
1460 hn_lro_hiwat_sysctl(SYSCTL_HANDLER_ARGS)
1461 {
1462         struct hn_softc *sc = arg1;
1463         int hiwat, error;
1464
1465         hiwat = sc->hn_lro_hiwat;
1466         error = sysctl_handle_int(oidp, &hiwat, 0, req);
1467         if (error || req->newptr == NULL)
1468                 return error;
1469
1470         if (!HN_LRO_HIWAT_ISVALID(sc, hiwat))
1471                 return EINVAL;
1472
1473         if (sc->hn_lro_hiwat != hiwat)
1474                 hn_set_lro_hiwat(sc, hiwat);
1475         return 0;
1476 }
1477 #endif  /* HN_LRO_HIWAT */
1478
1479 static int
1480 hn_check_iplen(const struct mbuf *m, int hoff)
1481 {
1482         const struct ip *ip;
1483         int len, iphlen, iplen;
1484         const struct tcphdr *th;
1485         int thoff;                              /* TCP data offset */
1486
1487         len = hoff + sizeof(struct ip);
1488
1489         /* The packet must be at least the size of an IP header. */
1490         if (m->m_pkthdr.len < len)
1491                 return IPPROTO_DONE;
1492
1493         /* The fixed IP header must reside completely in the first mbuf. */
1494         if (m->m_len < len)
1495                 return IPPROTO_DONE;
1496
1497         ip = mtodo(m, hoff);
1498
1499         /* Bound check the packet's stated IP header length. */
1500         iphlen = ip->ip_hl << 2;
1501         if (iphlen < sizeof(struct ip))         /* minimum header length */
1502                 return IPPROTO_DONE;
1503
1504         /* The full IP header must reside completely in the one mbuf. */
1505         if (m->m_len < hoff + iphlen)
1506                 return IPPROTO_DONE;
1507
1508         iplen = ntohs(ip->ip_len);
1509
1510         /*
1511          * Check that the amount of data in the buffers is as
1512          * at least much as the IP header would have us expect.
1513          */
1514         if (m->m_pkthdr.len < hoff + iplen)
1515                 return IPPROTO_DONE;
1516
1517         /*
1518          * Ignore IP fragments.
1519          */
1520         if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
1521                 return IPPROTO_DONE;
1522
1523         /*
1524          * The TCP/IP or UDP/IP header must be entirely contained within
1525          * the first fragment of a packet.
1526          */
1527         switch (ip->ip_p) {
1528         case IPPROTO_TCP:
1529                 if (iplen < iphlen + sizeof(struct tcphdr))
1530                         return IPPROTO_DONE;
1531                 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
1532                         return IPPROTO_DONE;
1533                 th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
1534                 thoff = th->th_off << 2;
1535                 if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
1536                         return IPPROTO_DONE;
1537                 if (m->m_len < hoff + iphlen + thoff)
1538                         return IPPROTO_DONE;
1539                 break;
1540         case IPPROTO_UDP:
1541                 if (iplen < iphlen + sizeof(struct udphdr))
1542                         return IPPROTO_DONE;
1543                 if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
1544                         return IPPROTO_DONE;
1545                 break;
1546         default:
1547                 if (iplen < iphlen)
1548                         return IPPROTO_DONE;
1549                 break;
1550         }
1551         return ip->ip_p;
1552 }
1553
1554 static device_method_t netvsc_methods[] = {
1555         /* Device interface */
1556         DEVMETHOD(device_probe,         netvsc_probe),
1557         DEVMETHOD(device_attach,        netvsc_attach),
1558         DEVMETHOD(device_detach,        netvsc_detach),
1559         DEVMETHOD(device_shutdown,      netvsc_shutdown),
1560
1561         { 0, 0 }
1562 };
1563
1564 static driver_t netvsc_driver = {
1565         NETVSC_DEVNAME,
1566         netvsc_methods,
1567         sizeof(hn_softc_t)
1568 };
1569
1570 static devclass_t netvsc_devclass;
1571
1572 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
1573 MODULE_VERSION(hn, 1);
1574 MODULE_DEPEND(hn, vmbus, 1, 1, 1);