sys/dev/virtual/virtio/net/if_vtnet.c

   1 /*-
   2  * Copyright (c) 2011, Bryan Venteicher <bryanv@daemoninthecloset.org>
   3  * All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice unmodified, this list of conditions, and the following
  10  *    disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  *
  15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  25  */
  26
  27 /* Driver for VirtIO network devices. */
  28
  29 #include <sys/cdefs.h>
  30
  31 #include <sys/param.h>
  32 #include <sys/systm.h>
  33 #include <sys/kernel.h>
  34 #include <sys/sockio.h>
  35 #include <sys/mbuf.h>
  36 #include <sys/malloc.h>
  37 #include <sys/module.h>
  38 #include <sys/socket.h>
  39 #include <sys/sysctl.h>
  40 #include <sys/taskqueue.h>
  41 #include <sys/random.h>
  42 #include <sys/sglist.h>
  43 #include <sys/serialize.h>
  44 #include <sys/bus.h>
  45 #include <sys/rman.h>
  46
  47 #include <net/ethernet.h>
  48 #include <net/if.h>
  49 #include <net/if_arp.h>
  50 #include <net/if_dl.h>
  51 #include <net/if_types.h>
  52 #include <net/if_media.h>
  53 #include <net/vlan/if_vlan_var.h>
  54 #include <net/vlan/if_vlan_ether.h>
  55 #include <net/ifq_var.h>
  56
  57 #include <net/bpf.h>
  58
  59 #include <netinet/in_systm.h>
  60 #include <netinet/in.h>
  61 #include <netinet/ip.h>
  62 #include <netinet/ip6.h>
  63 #include <netinet/udp.h>
  64 #include <netinet/tcp.h>
  65
  66 #include <dev/virtual/virtio/virtio/virtio.h>
  67 #include <dev/virtual/virtio/virtio/virtqueue.h>
  68
  69 #include "virtio_net.h"
  70 #include "virtio_if.h"
  71
  72 struct vtnet_statistics {
  73         unsigned long           mbuf_alloc_failed;
  74
  75         unsigned long           rx_frame_too_large;
  76         unsigned long           rx_enq_replacement_failed;
  77         unsigned long           rx_mergeable_failed;
  78         unsigned long           rx_csum_bad_ethtype;
  79         unsigned long           rx_csum_bad_start;
  80         unsigned long           rx_csum_bad_ipproto;
  81         unsigned long           rx_csum_bad_offset;
  82         unsigned long           rx_csum_failed;
  83         unsigned long           rx_csum_offloaded;
  84         unsigned long           rx_task_rescheduled;
  85
  86         unsigned long           tx_csum_offloaded;
  87         unsigned long           tx_tso_offloaded;
  88         unsigned long           tx_csum_bad_ethtype;
  89         unsigned long           tx_tso_bad_ethtype;
  90         unsigned long           tx_task_rescheduled;
  91 };
  92
  93 struct vtnet_softc {
  94         device_t                vtnet_dev;
  95         struct ifnet            *vtnet_ifp;
  96         struct lwkt_serialize   vtnet_slz;
  97
  98         uint32_t                vtnet_flags;
  99 #define VTNET_FLAG_LINK         0x0001
 100 #define VTNET_FLAG_SUSPENDED    0x0002
 101 #define VTNET_FLAG_CTRL_VQ      0x0004
 102 #define VTNET_FLAG_CTRL_RX      0x0008
 103 #define VTNET_FLAG_VLAN_FILTER  0x0010
 104 #define VTNET_FLAG_TSO_ECN      0x0020
 105 #define VTNET_FLAG_MRG_RXBUFS   0x0040
 106 #define VTNET_FLAG_LRO_NOMRG    0x0080
 107
 108         struct virtqueue        *vtnet_rx_vq;
 109         struct virtqueue        *vtnet_tx_vq;
 110         struct virtqueue        *vtnet_ctrl_vq;
 111
 112         struct vtnet_tx_header  *vtnet_txhdrarea;
 113         uint32_t                vtnet_txhdridx;
 114         struct vtnet_mac_filter *vtnet_macfilter;
 115
 116         int                     vtnet_hdr_size;
 117         int                     vtnet_tx_size;
 118         int                     vtnet_rx_size;
 119         int                     vtnet_rx_process_limit;
 120         int                     vtnet_rx_mbuf_size;
 121         int                     vtnet_rx_mbuf_count;
 122         int                     vtnet_if_flags;
 123         int                     vtnet_watchdog_timer;
 124         uint64_t                vtnet_features;
 125
 126         struct task             vtnet_cfgchg_task;
 127
 128         struct vtnet_statistics vtnet_stats;
 129
 130         struct callout          vtnet_tick_ch;
 131
 132         eventhandler_tag        vtnet_vlan_attach;
 133         eventhandler_tag        vtnet_vlan_detach;
 134
 135         struct ifmedia          vtnet_media;
 136         /*
 137          * Fake media type; the host does not provide us with
 138          * any real media information.
 139          */
 140 #define VTNET_MEDIATYPE         (IFM_ETHER | IFM_1000_T | IFM_FDX)
 141         char                    vtnet_hwaddr[ETHER_ADDR_LEN];
 142
 143         /*
 144          * During reset, the host's VLAN filtering table is lost. The
 145          * array below is used to restore all the VLANs configured on
 146          * this interface after a reset.
 147          */
 148 #define VTNET_VLAN_SHADOW_SIZE  (4096 / 32)
 149         int                     vtnet_nvlans;
 150         uint32_t                vtnet_vlan_shadow[VTNET_VLAN_SHADOW_SIZE];
 151
 152         char                    vtnet_mtx_name[16];
 153 };
 154
 155 /*
 156  * When mergeable buffers are not negotiated, the vtnet_rx_header structure
 157  * below is placed at the beginning of the mbuf data. Use 4 bytes of pad to
 158  * both keep the VirtIO header and the data non-contiguous and to keep the
 159  * frame's payload 4 byte aligned.
 160  *
 161  * When mergeable buffers are negotiated, the host puts the VirtIO header in
 162  * the beginning of the first mbuf's data.
 163  */
 164 #define VTNET_RX_HEADER_PAD     4
 165 struct vtnet_rx_header {
 166         struct virtio_net_hdr   vrh_hdr;
 167         char                    vrh_pad[VTNET_RX_HEADER_PAD];
 168 } __packed;
 169
 170 /*
 171  * For each outgoing frame, the vtnet_tx_header below is allocated from
 172  * the vtnet_tx_header_zone.
 173  */
 174 struct vtnet_tx_header {
 175         union {
 176                 struct virtio_net_hdr           hdr;
 177                 struct virtio_net_hdr_mrg_rxbuf mhdr;
 178         } vth_uhdr;
 179
 180         struct mbuf             *vth_mbuf;
 181 };
 182
 183 MALLOC_DEFINE(M_VTNET, "VTNET_TX", "Outgoing VTNET TX frame header");
 184
 185 /*
 186  * The VirtIO specification does not place a limit on the number of MAC
 187  * addresses the guest driver may request to be filtered. In practice,
 188  * the host is constrained by available resources. To simplify this driver,
 189  * impose a reasonably high limit of MAC addresses we will filter before
 190  * falling back to promiscuous or all-multicast modes.
 191  */
 192 #define VTNET_MAX_MAC_ENTRIES   128
 193
 194 struct vtnet_mac_table {
 195         uint32_t                nentries;
 196         uint8_t                 macs[VTNET_MAX_MAC_ENTRIES][ETHER_ADDR_LEN];
 197 } __packed;
 198
 199 struct vtnet_mac_filter {
 200         struct vtnet_mac_table  vmf_unicast;
 201         uint32_t                vmf_pad; /* Make tables non-contiguous. */
 202         struct vtnet_mac_table  vmf_multicast;
 203 };
 204
 205 #define VTNET_WATCHDOG_TIMEOUT  5
 206 #define VTNET_CSUM_OFFLOAD      (CSUM_TCP | CSUM_UDP)
 207
 208 /* Features desired/implemented by this driver. */
 209 #define VTNET_FEATURES          \
 210     (VIRTIO_NET_F_MAC           | \
 211      VIRTIO_NET_F_STATUS        | \
 212      VIRTIO_NET_F_CTRL_VQ       | \
 213      VIRTIO_NET_F_CTRL_RX       | \
 214      VIRTIO_NET_F_CTRL_VLAN     | \
 215      VIRTIO_NET_F_CSUM          | \
 216      VIRTIO_NET_F_HOST_TSO4     | \
 217      VIRTIO_NET_F_HOST_TSO6     | \
 218      VIRTIO_NET_F_HOST_ECN      | \
 219      VIRTIO_NET_F_GUEST_CSUM    | \
 220      VIRTIO_NET_F_GUEST_TSO4    | \
 221      VIRTIO_NET_F_GUEST_TSO6    | \
 222      VIRTIO_NET_F_GUEST_ECN     | \
 223      VIRTIO_NET_F_MRG_RXBUF)
 224
 225 /*
 226  * The VIRTIO_NET_F_GUEST_TSO[46] features permit the host to send us
 227  * frames larger than 1514 bytes. We do not yet support software LRO
 228  * via tcp_lro_rx().
 229  */
 230 #define VTNET_LRO_FEATURES (VIRTIO_NET_F_GUEST_TSO4 | \
 231                             VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN)
 232
 233 #define VTNET_MAX_MTU           65536
 234 #define VTNET_MAX_RX_SIZE       65550
 235
 236 /*
 237  * Used to preallocate the Vq indirect descriptors. The first segment
 238  * is reserved for the header.
 239  */
 240 #define VTNET_MIN_RX_SEGS       2
 241 #define VTNET_MAX_RX_SEGS       34
 242 #define VTNET_MAX_TX_SEGS       34
 243
 244 #define IFCAP_TSO4              0x00100 /* can do TCP Segmentation Offload */
 245 #define IFCAP_TSO6              0x00200 /* can do TCP6 Segmentation Offload */
 246 #define IFCAP_LRO               0x00400 /* can do Large Receive Offload */
 247 #define IFCAP_VLAN_HWFILTER     0x10000 /* interface hw can filter vlan tag */
 248 #define IFCAP_VLAN_HWTSO        0x40000 /* can do IFCAP_TSO on VLANs */
 249
 250
 251 /*
 252  * Assert we can receive and transmit the maximum with regular
 253  * size clusters.
 254  */
 255 CTASSERT(((VTNET_MAX_RX_SEGS - 1) * MCLBYTES) >= VTNET_MAX_RX_SIZE);
 256 CTASSERT(((VTNET_MAX_TX_SEGS - 1) * MCLBYTES) >= VTNET_MAX_MTU);
 257
 258 /*
 259  * Determine how many mbufs are in each receive buffer. For LRO without
 260  * mergeable descriptors, we must allocate an mbuf chain large enough to
 261  * hold both the vtnet_rx_header and the maximum receivable data.
 262  */
 263 #define VTNET_NEEDED_RX_MBUFS(_sc)                                      \
 264         ((_sc)->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0 ? 1 :          \
 265         howmany(sizeof(struct vtnet_rx_header) + VTNET_MAX_RX_SIZE,     \
 266         (_sc)->vtnet_rx_mbuf_size)
 267
 268 static int      vtnet_modevent(module_t, int, void *);
 269
 270 static int      vtnet_probe(device_t);
 271 static int      vtnet_attach(device_t);
 272 static int      vtnet_detach(device_t);
 273 static int      vtnet_suspend(device_t);
 274 static int      vtnet_resume(device_t);
 275 static int      vtnet_shutdown(device_t);
 276 static int      vtnet_config_change(device_t);
 277
 278 static void     vtnet_negotiate_features(struct vtnet_softc *);
 279 static int      vtnet_alloc_virtqueues(struct vtnet_softc *);
 280 static void     vtnet_get_hwaddr(struct vtnet_softc *);
 281 static void     vtnet_set_hwaddr(struct vtnet_softc *);
 282 static int      vtnet_is_link_up(struct vtnet_softc *);
 283 static void     vtnet_update_link_status(struct vtnet_softc *);
 284 #if 0
 285 static void     vtnet_watchdog(struct vtnet_softc *);
 286 #endif
 287 static void     vtnet_config_change_task(void *, int);
 288 static int      vtnet_change_mtu(struct vtnet_softc *, int);
 289 static int      vtnet_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
 290
 291 static int      vtnet_init_rx_vq(struct vtnet_softc *);
 292 static void     vtnet_free_rx_mbufs(struct vtnet_softc *);
 293 static void     vtnet_free_tx_mbufs(struct vtnet_softc *);
 294 static void     vtnet_free_ctrl_vq(struct vtnet_softc *);
 295
 296 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int,
 297                     struct mbuf **);
 298 static int      vtnet_replace_rxbuf(struct vtnet_softc *,
 299                     struct mbuf *, int);
 300 static int      vtnet_newbuf(struct vtnet_softc *);
 301 static void     vtnet_discard_merged_rxbuf(struct vtnet_softc *, int);
 302 static void     vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *);
 303 static int      vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *);
 304 static void     vtnet_vlan_tag_remove(struct mbuf *);
 305 static int      vtnet_rx_csum(struct vtnet_softc *, struct mbuf *,
 306                     struct virtio_net_hdr *);
 307 static int      vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int);
 308 static int      vtnet_rxeof(struct vtnet_softc *, int, int *);
 309 static void     vtnet_rx_intr_task(void *);
 310 static int      vtnet_rx_vq_intr(void *);
 311
 312 static void     vtnet_txeof(struct vtnet_softc *);
 313 static struct mbuf * vtnet_tx_offload(struct vtnet_softc *, struct mbuf *,
 314                     struct virtio_net_hdr *);
 315 static int      vtnet_enqueue_txbuf(struct vtnet_softc *, struct mbuf **,
 316                     struct vtnet_tx_header *);
 317 static int      vtnet_encap(struct vtnet_softc *, struct mbuf **);
 318 static void     vtnet_start_locked(struct ifnet *, struct ifaltq_subque *);
 319 static void     vtnet_start(struct ifnet *, struct ifaltq_subque *);
 320 static void     vtnet_tick(void *);
 321 static void     vtnet_tx_intr_task(void *);
 322 static int      vtnet_tx_vq_intr(void *);
 323
 324 static void     vtnet_stop(struct vtnet_softc *);
 325 static int      vtnet_reinit(struct vtnet_softc *);
 326 static void     vtnet_init_locked(struct vtnet_softc *);
 327 static void     vtnet_init(void *);
 328
 329 static void     vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
 330                     struct sglist *, int, int);
 331
 332 static void     vtnet_rx_filter(struct vtnet_softc *sc);
 333 static int      vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
 334 static int      vtnet_set_promisc(struct vtnet_softc *, int);
 335 static int      vtnet_set_allmulti(struct vtnet_softc *, int);
 336 static void     vtnet_rx_filter_mac(struct vtnet_softc *);
 337
 338 static int      vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
 339 static void     vtnet_rx_filter_vlan(struct vtnet_softc *);
 340 static void     vtnet_set_vlan_filter(struct vtnet_softc *, int, uint16_t);
 341 static void     vtnet_register_vlan(void *, struct ifnet *, uint16_t);
 342 static void     vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
 343
 344 static int      vtnet_ifmedia_upd(struct ifnet *);
 345 static void     vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
 346
 347 static void     vtnet_add_statistics(struct vtnet_softc *);
 348
 349 static int      vtnet_enable_rx_intr(struct vtnet_softc *);
 350 static int      vtnet_enable_tx_intr(struct vtnet_softc *);
 351 static void     vtnet_disable_rx_intr(struct vtnet_softc *);
 352 static void     vtnet_disable_tx_intr(struct vtnet_softc *);
 353
 354 /* Tunables. */
 355 static int vtnet_csum_disable = 0;
 356 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
 357 static int vtnet_tso_disable = 1;
 358 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
 359 static int vtnet_lro_disable = 1;
 360 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
 361
 362 /*
 363  * Reducing the number of transmit completed interrupts can
 364  * improve performance. To do so, the define below keeps the
 365  * Tx vq interrupt disabled and adds calls to vtnet_txeof()
 366  * in the start and watchdog paths. The price to pay for this
 367  * is the m_free'ing of transmitted mbufs may be delayed until
 368  * the watchdog fires.
 369  */
 370 #define VTNET_TX_INTR_MODERATION
 371
 372 static struct virtio_feature_desc vtnet_feature_desc[] = {
 373         { VIRTIO_NET_F_CSUM,            "TxChecksum"    },
 374         { VIRTIO_NET_F_GUEST_CSUM,      "RxChecksum"    },
 375         { VIRTIO_NET_F_MAC,             "MacAddress"    },
 376         { VIRTIO_NET_F_GSO,             "TxAllGSO"      },
 377         { VIRTIO_NET_F_GUEST_TSO4,      "RxTSOv4"       },
 378         { VIRTIO_NET_F_GUEST_TSO6,      "RxTSOv6"       },
 379         { VIRTIO_NET_F_GUEST_ECN,       "RxECN"         },
 380         { VIRTIO_NET_F_GUEST_UFO,       "RxUFO"         },
 381         { VIRTIO_NET_F_HOST_TSO4,       "TxTSOv4"       },
 382         { VIRTIO_NET_F_HOST_TSO6,       "TxTSOv6"       },
 383         { VIRTIO_NET_F_HOST_ECN,        "TxTSOECN"      },
 384         { VIRTIO_NET_F_HOST_UFO,        "TxUFO"         },
 385         { VIRTIO_NET_F_MRG_RXBUF,       "MrgRxBuf"      },
 386         { VIRTIO_NET_F_STATUS,          "Status"        },
 387         { VIRTIO_NET_F_CTRL_VQ,         "ControlVq"     },
 388         { VIRTIO_NET_F_CTRL_RX,         "RxMode"        },
 389         { VIRTIO_NET_F_CTRL_VLAN,       "VLanFilter"    },
 390         { VIRTIO_NET_F_CTRL_RX_EXTRA,   "RxModeExtra"   },
 391         { VIRTIO_NET_F_MQ,              "RFS"           },
 392         { 0, NULL }
 393 };
 394
 395 static device_method_t vtnet_methods[] = {
 396         /* Device methods. */
 397         DEVMETHOD(device_probe,         vtnet_probe),
 398         DEVMETHOD(device_attach,        vtnet_attach),
 399         DEVMETHOD(device_detach,        vtnet_detach),
 400         DEVMETHOD(device_suspend,       vtnet_suspend),
 401         DEVMETHOD(device_resume,        vtnet_resume),
 402         DEVMETHOD(device_shutdown,      vtnet_shutdown),
 403
 404         /* VirtIO methods. */
 405         DEVMETHOD(virtio_config_change, vtnet_config_change),
 406
 407         { 0, 0 }
 408 };
 409
 410 static driver_t vtnet_driver = {
 411         "vtnet",
 412         vtnet_methods,
 413         sizeof(struct vtnet_softc)
 414 };
 415
 416 static devclass_t vtnet_devclass;
 417
 418 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
 419               vtnet_modevent, 0);
 420 MODULE_VERSION(vtnet, 1);
 421 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
 422
 423 static int
 424 vtnet_modevent(module_t mod, int type, void *unused)
 425 {
 426         int error;
 427
 428         error = 0;
 429
 430         switch (type) {
 431         case MOD_LOAD:
 432                 break;
 433         case MOD_UNLOAD:
 434                 break;
 435         case MOD_SHUTDOWN:
 436                 break;
 437         default:
 438                 error = EOPNOTSUPP;
 439                 break;
 440         }
 441
 442         return (error);
 443 }
 444
 445 static int
 446 vtnet_probe(device_t dev)
 447 {
 448         if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
 449                 return (ENXIO);
 450
 451         device_set_desc(dev, "VirtIO Networking Adapter");
 452
 453         return (BUS_PROBE_DEFAULT);
 454 }
 455
 456 static int
 457 vtnet_attach(device_t dev)
 458 {
 459         struct vtnet_softc *sc;
 460         struct ifnet *ifp;
 461         int tx_size, error;
 462
 463         sc = device_get_softc(dev);
 464         sc->vtnet_dev = dev;
 465
 466         lwkt_serialize_init(&sc->vtnet_slz);
 467         callout_init(&sc->vtnet_tick_ch);
 468
 469         ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
 470                      vtnet_ifmedia_sts);
 471         ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
 472         ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
 473
 474         vtnet_add_statistics(sc);
 475
 476         virtio_set_feature_desc(dev, vtnet_feature_desc);
 477         vtnet_negotiate_features(sc);
 478
 479         if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
 480                 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
 481                 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 482         } else {
 483                 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
 484         }
 485
 486         sc->vtnet_rx_mbuf_size = MCLBYTES;
 487         sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
 488
 489         if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
 490                 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
 491
 492                 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
 493                         sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
 494                 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
 495                         sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
 496         }
 497
 498         vtnet_get_hwaddr(sc);
 499
 500         error = vtnet_alloc_virtqueues(sc);
 501         if (error) {
 502                 device_printf(dev, "cannot allocate virtqueues\n");
 503                 goto fail;
 504         }
 505
 506         ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
 507         if (ifp == NULL) {
 508                 device_printf(dev, "cannot allocate ifnet structure\n");
 509                 error = ENOSPC;
 510                 goto fail;
 511         }
 512
 513         ifp->if_softc = sc;
 514         if_initname(ifp, device_get_name(dev), device_get_unit(dev));
 515         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
 516         ifp->if_init = vtnet_init;
 517         ifp->if_start = vtnet_start;
 518         ifp->if_ioctl = vtnet_ioctl;
 519
 520         sc->vtnet_rx_size = virtqueue_size(sc->vtnet_rx_vq);
 521         sc->vtnet_rx_process_limit = sc->vtnet_rx_size;
 522
 523         tx_size = virtqueue_size(sc->vtnet_tx_vq);
 524         sc->vtnet_tx_size = tx_size;
 525         sc->vtnet_txhdridx = 0;
 526         sc->vtnet_txhdrarea = contigmalloc(
 527             ((sc->vtnet_tx_size / 2) + 1) * sizeof(struct vtnet_tx_header),
 528             M_VTNET, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
 529         if (sc->vtnet_txhdrarea == NULL) {
 530                 device_printf(dev, "cannot contigmalloc the tx headers\n");
 531                 goto fail;
 532         }
 533         sc->vtnet_macfilter = contigmalloc(
 534             sizeof(struct vtnet_mac_filter),
 535             M_DEVBUF, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
 536         if (sc->vtnet_macfilter == NULL) {
 537                 device_printf(dev,
 538                     "cannot contigmalloc the mac filter table\n");
 539                 goto fail;
 540         }
 541         ifq_set_maxlen(&ifp->if_snd, tx_size - 1);
 542         ifq_set_ready(&ifp->if_snd);
 543
 544         ether_ifattach(ifp, sc->vtnet_hwaddr, NULL);
 545
 546         if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)){
 547                 //ifp->if_capabilities |= IFCAP_LINKSTATE;
 548                  kprintf("add dynamic link state\n");
 549         }
 550
 551         /* Tell the upper layer(s) we support long frames. */
 552         ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
 553         ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
 554
 555         if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
 556                 ifp->if_capabilities |= IFCAP_TXCSUM;
 557
 558                 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
 559                         ifp->if_capabilities |= IFCAP_TSO4;
 560                 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
 561                         ifp->if_capabilities |= IFCAP_TSO6;
 562                 if (ifp->if_capabilities & IFCAP_TSO)
 563                         ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
 564
 565                 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
 566                         sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
 567         }
 568
 569         if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
 570                 ifp->if_capabilities |= IFCAP_RXCSUM;
 571
 572                 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
 573                     virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
 574                         ifp->if_capabilities |= IFCAP_LRO;
 575         }
 576
 577         if (ifp->if_capabilities & IFCAP_HWCSUM) {
 578                 /*
 579                  * VirtIO does not support VLAN tagging, but we can fake
 580                  * it by inserting and removing the 802.1Q header during
 581                  * transmit and receive. We are then able to do checksum
 582                  * offloading of VLAN frames.
 583                  */
 584                 ifp->if_capabilities |=
 585                         IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
 586         }
 587
 588         ifp->if_capenable = ifp->if_capabilities;
 589
 590         /*
 591          * Capabilities after here are not enabled by default.
 592          */
 593
 594         if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
 595                 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
 596
 597                 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
 598                     vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
 599                 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
 600                     vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
 601         }
 602
 603         TASK_INIT(&sc->vtnet_cfgchg_task, 0, vtnet_config_change_task, sc);
 604
 605         error = virtio_setup_intr(dev, &sc->vtnet_slz);
 606         if (error) {
 607                 device_printf(dev, "cannot setup virtqueue interrupts\n");
 608                 ether_ifdetach(ifp);
 609                 goto fail;
 610         }
 611
 612         /*
 613          * Device defaults to promiscuous mode for backwards
 614          * compatibility. Turn it off if possible.
 615          */
 616         if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
 617                 lwkt_serialize_enter(&sc->vtnet_slz);
 618                 if (vtnet_set_promisc(sc, 0) != 0) {
 619                         ifp->if_flags |= IFF_PROMISC;
 620                         device_printf(dev,
 621                             "cannot disable promiscuous mode\n");
 622                 }
 623                 lwkt_serialize_exit(&sc->vtnet_slz);
 624         } else
 625                 ifp->if_flags |= IFF_PROMISC;
 626
 627 fail:
 628         if (error)
 629                 vtnet_detach(dev);
 630
 631         return (error);
 632 }
 633
 634 static int
 635 vtnet_detach(device_t dev)
 636 {
 637         struct vtnet_softc *sc;
 638         struct ifnet *ifp;
 639
 640         sc = device_get_softc(dev);
 641         ifp = sc->vtnet_ifp;
 642
 643         if (device_is_attached(dev)) {
 644                 lwkt_serialize_enter(&sc->vtnet_slz);
 645                 vtnet_stop(sc);
 646                 lwkt_serialize_exit(&sc->vtnet_slz);
 647
 648                 callout_stop(&sc->vtnet_tick_ch);
 649                 taskqueue_drain(taskqueue_swi, &sc->vtnet_cfgchg_task);
 650
 651                 ether_ifdetach(ifp);
 652         }
 653
 654         if (sc->vtnet_vlan_attach != NULL) {
 655                 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
 656                 sc->vtnet_vlan_attach = NULL;
 657         }
 658         if (sc->vtnet_vlan_detach != NULL) {
 659                 EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach);
 660                 sc->vtnet_vlan_detach = NULL;
 661         }
 662
 663         if (ifp) {
 664                 if_free(ifp);
 665                 sc->vtnet_ifp = NULL;
 666         }
 667
 668         if (sc->vtnet_rx_vq != NULL)
 669                 vtnet_free_rx_mbufs(sc);
 670         if (sc->vtnet_tx_vq != NULL)
 671                 vtnet_free_tx_mbufs(sc);
 672         if (sc->vtnet_ctrl_vq != NULL)
 673                 vtnet_free_ctrl_vq(sc);
 674
 675         if (sc->vtnet_txhdrarea != NULL) {
 676                 contigfree(sc->vtnet_txhdrarea,
 677                     ((sc->vtnet_tx_size / 2) + 1) *
 678                     sizeof(struct vtnet_tx_header), M_VTNET);
 679                 sc->vtnet_txhdrarea = NULL;
 680         }
 681         if (sc->vtnet_macfilter != NULL) {
 682                 contigfree(sc->vtnet_macfilter,
 683                     sizeof(struct vtnet_mac_filter), M_DEVBUF);
 684                 sc->vtnet_macfilter = NULL;
 685         }
 686
 687         ifmedia_removeall(&sc->vtnet_media);
 688
 689         return (0);
 690 }
 691
 692 static int
 693 vtnet_suspend(device_t dev)
 694 {
 695         struct vtnet_softc *sc;
 696
 697         sc = device_get_softc(dev);
 698
 699         lwkt_serialize_enter(&sc->vtnet_slz);
 700         vtnet_stop(sc);
 701         sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
 702         lwkt_serialize_exit(&sc->vtnet_slz);
 703
 704         return (0);
 705 }
 706
 707 static int
 708 vtnet_resume(device_t dev)
 709 {
 710         struct vtnet_softc *sc;
 711         struct ifnet *ifp;
 712
 713         sc = device_get_softc(dev);
 714         ifp = sc->vtnet_ifp;
 715
 716         lwkt_serialize_enter(&sc->vtnet_slz);
 717         if (ifp->if_flags & IFF_UP)
 718                 vtnet_init_locked(sc);
 719         sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
 720         lwkt_serialize_exit(&sc->vtnet_slz);
 721
 722         return (0);
 723 }
 724
 725 static int
 726 vtnet_shutdown(device_t dev)
 727 {
 728
 729         /*
 730          * Suspend already does all of what we need to
 731          * do here; we just never expect to be resumed.
 732          */
 733         return (vtnet_suspend(dev));
 734 }
 735
 736 static int
 737 vtnet_config_change(device_t dev)
 738 {
 739         struct vtnet_softc *sc;
 740
 741         sc = device_get_softc(dev);
 742
 743         taskqueue_enqueue(taskqueue_thread[mycpuid], &sc->vtnet_cfgchg_task);
 744
 745         return (1);
 746 }
 747
 748 static void
 749 vtnet_negotiate_features(struct vtnet_softc *sc)
 750 {
 751         device_t dev;
 752         uint64_t mask, features;
 753
 754         dev = sc->vtnet_dev;
 755         mask = 0;
 756
 757         if (vtnet_csum_disable)
 758                 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
 759
 760         /*
 761          * TSO and LRO are only available when their corresponding
 762          * checksum offload feature is also negotiated.
 763          */
 764
 765         if (vtnet_csum_disable || vtnet_tso_disable)
 766                 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
 767                     VIRTIO_NET_F_HOST_ECN;
 768
 769         if (vtnet_csum_disable || vtnet_lro_disable)
 770                 mask |= VTNET_LRO_FEATURES;
 771
 772         features = VTNET_FEATURES & ~mask;
 773         features |= VIRTIO_F_NOTIFY_ON_EMPTY;
 774         sc->vtnet_features = virtio_negotiate_features(dev, features);
 775 }
 776
 777 static int
 778 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
 779 {
 780         device_t dev;
 781         struct vq_alloc_info vq_info[3];
 782         int nvqs, rxsegs;
 783
 784         dev = sc->vtnet_dev;
 785         nvqs = 2;
 786
 787         /*
 788          * Indirect descriptors are not needed for the Rx
 789          * virtqueue when mergeable buffers are negotiated.
 790          * The header is placed inline with the data, not
 791          * in a separate descriptor, and mbuf clusters are
 792          * always physically contiguous.
 793          */
 794         if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
 795                 rxsegs = sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ?
 796                     VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
 797         } else
 798                 rxsegs = 0;
 799
 800         VQ_ALLOC_INFO_INIT(&vq_info[0], rxsegs,
 801             vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
 802             "%s receive", device_get_nameunit(dev));
 803
 804         VQ_ALLOC_INFO_INIT(&vq_info[1], VTNET_MAX_TX_SEGS,
 805             vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
 806             "%s transmit", device_get_nameunit(dev));
 807
 808         if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
 809                 nvqs++;
 810
 811                 VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
 812                     &sc->vtnet_ctrl_vq, "%s control",
 813                     device_get_nameunit(dev));
 814         }
 815
 816         return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
 817 }
 818
 819 static void
 820 vtnet_get_hwaddr(struct vtnet_softc *sc)
 821 {
 822         device_t dev;
 823
 824         dev = sc->vtnet_dev;
 825
 826         if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
 827                 virtio_read_device_config(dev,
 828                     offsetof(struct virtio_net_config, mac),
 829                     sc->vtnet_hwaddr, ETHER_ADDR_LEN);
 830         } else {
 831                 /* Generate random locally administered unicast address. */
 832                 sc->vtnet_hwaddr[0] = 0xB2;
 833                 karc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1);
 834
 835                 vtnet_set_hwaddr(sc);
 836         }
 837 }
 838
 839 static void
 840 vtnet_set_hwaddr(struct vtnet_softc *sc)
 841 {
 842         device_t dev;
 843
 844         dev = sc->vtnet_dev;
 845
 846         virtio_write_device_config(dev,
 847             offsetof(struct virtio_net_config, mac),
 848             sc->vtnet_hwaddr, ETHER_ADDR_LEN);
 849 }
 850
 851 static int
 852 vtnet_is_link_up(struct vtnet_softc *sc)
 853 {
 854         device_t dev;
 855         struct ifnet *ifp;
 856         uint16_t status;
 857
 858         dev = sc->vtnet_dev;
 859         ifp = sc->vtnet_ifp;
 860
 861         ASSERT_SERIALIZED(&sc->vtnet_slz);
 862
 863         status = virtio_read_dev_config_2(dev,
 864                         offsetof(struct virtio_net_config, status));
 865
 866         return ((status & VIRTIO_NET_S_LINK_UP) != 0);
 867 }
 868
 869 static void
 870 vtnet_update_link_status(struct vtnet_softc *sc)
 871 {
 872         device_t dev;
 873         struct ifnet *ifp;
 874         struct ifaltq_subque *ifsq;
 875         int link;
 876
 877         dev = sc->vtnet_dev;
 878         ifp = sc->vtnet_ifp;
 879         ifsq = ifq_get_subq_default(&ifp->if_snd);
 880
 881         link = vtnet_is_link_up(sc);
 882
 883         if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
 884                 sc->vtnet_flags |= VTNET_FLAG_LINK;
 885                 if (bootverbose)
 886                         device_printf(dev, "Link is up\n");
 887                 ifp->if_link_state = LINK_STATE_UP;
 888                 if_link_state_change(ifp);
 889                 if (!ifsq_is_empty(ifsq))
 890                         vtnet_start_locked(ifp, ifsq);
 891         } else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
 892                 sc->vtnet_flags &= ~VTNET_FLAG_LINK;
 893                 if (bootverbose)
 894                         device_printf(dev, "Link is down\n");
 895
 896                 ifp->if_link_state = LINK_STATE_DOWN;
 897                 if_link_state_change(ifp);
 898         }
 899 }
 900
 901 #if 0
 902 static void
 903 vtnet_watchdog(struct vtnet_softc *sc)
 904 {
 905         struct ifnet *ifp;
 906
 907         ifp = sc->vtnet_ifp;
 908
 909 #ifdef VTNET_TX_INTR_MODERATION
 910         vtnet_txeof(sc);
 911 #endif
 912
 913         if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
 914                 return;
 915
 916         if_printf(ifp, "watchdog timeout -- resetting\n");
 917 #ifdef VTNET_DEBUG
 918         virtqueue_dump(sc->vtnet_tx_vq);
 919 #endif
 920         ifp->if_oerrors++;
 921         ifp->if_flags &= ~IFF_RUNNING;
 922         vtnet_init_locked(sc);
 923 }
 924 #endif
 925
 926 static void
 927 vtnet_config_change_task(void *arg, int pending)
 928 {
 929         struct vtnet_softc *sc;
 930
 931         sc = arg;
 932
 933         lwkt_serialize_enter(&sc->vtnet_slz);
 934         vtnet_update_link_status(sc);
 935         lwkt_serialize_exit(&sc->vtnet_slz);
 936 }
 937
 938 static int
 939 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data,struct ucred *cr)
 940 {
 941         struct vtnet_softc *sc;
 942         struct ifreq *ifr;
 943         int reinit, mask, error;
 944
 945         sc = ifp->if_softc;
 946         ifr = (struct ifreq *) data;
 947         reinit = 0;
 948         error = 0;
 949
 950         switch (cmd) {
 951         case SIOCSIFMTU:
 952                 if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
 953                         error = EINVAL;
 954                 else if (ifp->if_mtu != ifr->ifr_mtu) {
 955                         lwkt_serialize_enter(&sc->vtnet_slz);
 956                         error = vtnet_change_mtu(sc, ifr->ifr_mtu);
 957                         lwkt_serialize_exit(&sc->vtnet_slz);
 958                 }
 959                 break;
 960
 961         case SIOCSIFFLAGS:
 962                 lwkt_serialize_enter(&sc->vtnet_slz);
 963                 if ((ifp->if_flags & IFF_UP) == 0) {
 964                         if (ifp->if_flags & IFF_RUNNING)
 965                                 vtnet_stop(sc);
 966                 } else if (ifp->if_flags & IFF_RUNNING) {
 967                         if ((ifp->if_flags ^ sc->vtnet_if_flags) &
 968                             (IFF_PROMISC | IFF_ALLMULTI)) {
 969                                 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
 970                                         vtnet_rx_filter(sc);
 971                                 else
 972                                         error = ENOTSUP;
 973                         }
 974                 } else
 975                         vtnet_init_locked(sc);
 976
 977                 if (error == 0)
 978                         sc->vtnet_if_flags = ifp->if_flags;
 979                 lwkt_serialize_exit(&sc->vtnet_slz);
 980                 break;
 981
 982         case SIOCADDMULTI:
 983         case SIOCDELMULTI:
 984                 lwkt_serialize_enter(&sc->vtnet_slz);
 985                 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
 986                     (ifp->if_flags & IFF_RUNNING))
 987                         vtnet_rx_filter_mac(sc);
 988                 lwkt_serialize_exit(&sc->vtnet_slz);
 989                 break;
 990
 991         case SIOCSIFMEDIA:
 992         case SIOCGIFMEDIA:
 993                 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
 994                 break;
 995
 996         case SIOCSIFCAP:
 997                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
 998
 999                 lwkt_serialize_enter(&sc->vtnet_slz);
1000
1001                 if (mask & IFCAP_TXCSUM) {
1002                         ifp->if_capenable ^= IFCAP_TXCSUM;
1003                         if (ifp->if_capenable & IFCAP_TXCSUM)
1004                                 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
1005                         else
1006                                 ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
1007                 }
1008
1009                 if (mask & IFCAP_TSO4) {
1010                         ifp->if_capenable ^= IFCAP_TSO4;
1011                         if (ifp->if_capenable & IFCAP_TSO4)
1012                                 ifp->if_hwassist |= CSUM_TSO;
1013                         else
1014                                 ifp->if_hwassist &= ~CSUM_TSO;
1015                 }
1016
1017                 if (mask & IFCAP_RXCSUM) {
1018                         ifp->if_capenable ^= IFCAP_RXCSUM;
1019                         reinit = 1;
1020                 }
1021
1022                 if (mask & IFCAP_LRO) {
1023                         ifp->if_capenable ^= IFCAP_LRO;
1024                         reinit = 1;
1025                 }
1026
1027                 if (mask & IFCAP_VLAN_HWFILTER) {
1028                         ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
1029                         reinit = 1;
1030                 }
1031
1032                 if (mask & IFCAP_VLAN_HWTSO)
1033                         ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1034
1035                 if (mask & IFCAP_VLAN_HWTAGGING)
1036                         ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1037
1038                 if (reinit && (ifp->if_flags & IFF_RUNNING)) {
1039                         ifp->if_flags &= ~IFF_RUNNING;
1040                         vtnet_init_locked(sc);
1041                 }
1042                 //VLAN_CAPABILITIES(ifp);
1043
1044                 lwkt_serialize_exit(&sc->vtnet_slz);
1045                 break;
1046
1047         default:
1048                 error = ether_ioctl(ifp, cmd, data);
1049                 break;
1050         }
1051
1052         return (error);
1053 }
1054
1055 static int
1056 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
1057 {
1058         struct ifnet *ifp;
1059         int new_frame_size, clsize;
1060
1061         ifp = sc->vtnet_ifp;
1062
1063         if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1064                 new_frame_size = sizeof(struct vtnet_rx_header) +
1065                     sizeof(struct ether_vlan_header) + new_mtu;
1066
1067                 if (new_frame_size > MJUM9BYTES)
1068                         return (EINVAL);
1069
1070                 if (new_frame_size <= MCLBYTES)
1071                         clsize = MCLBYTES;
1072                 else
1073                         clsize = MJUM9BYTES;
1074         } else {
1075                 new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
1076                     sizeof(struct ether_vlan_header) + new_mtu;
1077
1078                 if (new_frame_size <= MCLBYTES)
1079                         clsize = MCLBYTES;
1080                 else
1081                         clsize = MJUMPAGESIZE;
1082         }
1083
1084         sc->vtnet_rx_mbuf_size = clsize;
1085         sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
1086         KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
1087             ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
1088
1089         ifp->if_mtu = new_mtu;
1090
1091         if (ifp->if_flags & IFF_RUNNING) {
1092                 ifp->if_flags &= ~IFF_RUNNING;
1093                 vtnet_init_locked(sc);
1094         }
1095
1096         return (0);
1097 }
1098
1099 static int
1100 vtnet_init_rx_vq(struct vtnet_softc *sc)
1101 {
1102         struct virtqueue *vq;
1103         int nbufs, error;
1104
1105         vq = sc->vtnet_rx_vq;
1106         nbufs = 0;
1107         error = ENOSPC;
1108
1109         while (!virtqueue_full(vq)) {
1110                 if ((error = vtnet_newbuf(sc)) != 0)
1111                         break;
1112                 nbufs++;
1113         }
1114
1115         if (nbufs > 0) {
1116                 virtqueue_notify(vq, &sc->vtnet_slz);
1117
1118                 /*
1119                  * EMSGSIZE signifies the virtqueue did not have enough
1120                  * entries available to hold the last mbuf. This is not
1121                  * an error. We should not get ENOSPC since we check if
1122                  * the virtqueue is full before attempting to add a
1123                  * buffer.
1124                  */
1125                 if (error == EMSGSIZE)
1126                         error = 0;
1127         }
1128
1129         return (error);
1130 }
1131
1132 static void
1133 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
1134 {
1135         struct virtqueue *vq;
1136         struct mbuf *m;
1137         int last;
1138
1139         vq = sc->vtnet_rx_vq;
1140         last = 0;
1141
1142         while ((m = virtqueue_drain(vq, &last)) != NULL)
1143                 m_freem(m);
1144
1145         KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
1146 }
1147
1148 static void
1149 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1150 {
1151         struct virtqueue *vq;
1152         struct vtnet_tx_header *txhdr;
1153         int last;
1154
1155         vq = sc->vtnet_tx_vq;
1156         last = 0;
1157
1158         while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1159                 m_freem(txhdr->vth_mbuf);
1160         }
1161
1162         KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1163 }
1164
1165 static void
1166 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1167 {
1168         /*
1169          * The control virtqueue is only polled, therefore
1170          * it should already be empty.
1171          */
1172         KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1173                 ("Ctrl Vq not empty"));
1174 }
1175
1176 static struct mbuf *
1177 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1178 {
1179         struct mbuf *m_head, *m_tail, *m;
1180         int i, clsize;
1181
1182         clsize = sc->vtnet_rx_mbuf_size;
1183
1184         /*use getcl instead of getjcl. see  if_mxge.c comment line 2398*/
1185         //m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, clsize);
1186         m_head = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR );
1187         if (m_head == NULL)
1188                 goto fail;
1189
1190         m_head->m_len = clsize;
1191         m_tail = m_head;
1192
1193         if (nbufs > 1) {
1194                 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1195                         ("chained Rx mbuf requested without LRO_NOMRG"));
1196
1197                 for (i = 0; i < nbufs - 1; i++) {
1198                         //m = m_getjcl(M_DONTWAIT, MT_DATA, 0, clsize);
1199                         m = m_getcl(M_NOWAIT, MT_DATA, 0);
1200                         if (m == NULL)
1201                                 goto fail;
1202
1203                         m->m_len = clsize;
1204                         m_tail->m_next = m;
1205                         m_tail = m;
1206                 }
1207         }
1208
1209         if (m_tailp != NULL)
1210                 *m_tailp = m_tail;
1211
1212         return (m_head);
1213
1214 fail:
1215         sc->vtnet_stats.mbuf_alloc_failed++;
1216         m_freem(m_head);
1217
1218         return (NULL);
1219 }
1220
1221 static int
1222 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1223 {
1224         struct mbuf *m, *m_prev;
1225         struct mbuf *m_new, *m_tail;
1226         int len, clsize, nreplace, error;
1227
1228         m = m0;
1229         m_prev = NULL;
1230         len = len0;
1231
1232         m_tail = NULL;
1233         clsize = sc->vtnet_rx_mbuf_size;
1234         nreplace = 0;
1235
1236         if (m->m_next != NULL)
1237                 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1238                     ("chained Rx mbuf without LRO_NOMRG"));
1239
1240         /*
1241          * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1242          * allocating an entire chain for each received frame. When
1243          * the received frame's length is less than that of the chain,
1244          * the unused mbufs are reassigned to the new chain.
1245          */
1246         while (len > 0) {
1247                 /*
1248                  * Something is seriously wrong if we received
1249                  * a frame larger than the mbuf chain. Drop it.
1250                  */
1251                 if (m == NULL) {
1252                         sc->vtnet_stats.rx_frame_too_large++;
1253                         return (EMSGSIZE);
1254                 }
1255
1256                 KASSERT(m->m_len == clsize,
1257                     ("mbuf length not expected cluster size: %d",
1258                     m->m_len));
1259
1260                 m->m_len = MIN(m->m_len, len);
1261                 len -= m->m_len;
1262
1263                 m_prev = m;
1264                 m = m->m_next;
1265                 nreplace++;
1266         }
1267
1268         KASSERT(m_prev != NULL, ("m_prev == NULL"));
1269         KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1270                 ("too many replacement mbufs: %d/%d", nreplace,
1271                 sc->vtnet_rx_mbuf_count));
1272
1273         m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1274         if (m_new == NULL) {
1275                 m_prev->m_len = clsize;
1276                 return (ENOBUFS);
1277         }
1278
1279         /*
1280          * Move unused mbufs, if any, from the original chain
1281          * onto the end of the new chain.
1282          */
1283         if (m_prev->m_next != NULL) {
1284                 m_tail->m_next = m_prev->m_next;
1285                 m_prev->m_next = NULL;
1286         }
1287
1288         error = vtnet_enqueue_rxbuf(sc, m_new);
1289         if (error) {
1290                 /*
1291                  * BAD! We could not enqueue the replacement mbuf chain. We
1292                  * must restore the m0 chain to the original state if it was
1293                  * modified so we can subsequently discard it.
1294                  *
1295                  * NOTE: The replacement is suppose to be an identical copy
1296                  * to the one just dequeued so this is an unexpected error.
1297                  */
1298                 sc->vtnet_stats.rx_enq_replacement_failed++;
1299
1300                 if (m_tail->m_next != NULL) {
1301                         m_prev->m_next = m_tail->m_next;
1302                         m_tail->m_next = NULL;
1303                 }
1304
1305                 m_prev->m_len = clsize;
1306                 m_freem(m_new);
1307         }
1308
1309         return (error);
1310 }
1311
1312 static int
1313 vtnet_newbuf(struct vtnet_softc *sc)
1314 {
1315         struct mbuf *m;
1316         int error;
1317
1318         m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1319         if (m == NULL)
1320                 return (ENOBUFS);
1321
1322         error = vtnet_enqueue_rxbuf(sc, m);
1323         if (error)
1324                 m_freem(m);
1325
1326         return (error);
1327 }
1328
1329 static void
1330 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1331 {
1332         struct virtqueue *vq;
1333         struct mbuf *m;
1334
1335         vq = sc->vtnet_rx_vq;
1336
1337         while (--nbufs > 0) {
1338                 if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1339                         break;
1340                 vtnet_discard_rxbuf(sc, m);
1341         }
1342 }
1343
1344 static void
1345 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1346 {
1347         int error;
1348
1349         /*
1350          * Requeue the discarded mbuf. This should always be
1351          * successful since it was just dequeued.
1352          */
1353         error = vtnet_enqueue_rxbuf(sc, m);
1354         KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1355 }
1356
1357 static int
1358 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1359 {
1360         struct sglist sg;
1361         struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1362         struct vtnet_rx_header *rxhdr;
1363         struct virtio_net_hdr *hdr;
1364         uint8_t *mdata;
1365         int offset, error;
1366
1367         ASSERT_SERIALIZED(&sc->vtnet_slz);
1368         if ((sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0)
1369                 KASSERT(m->m_next == NULL, ("chained Rx mbuf"));
1370
1371         sglist_init(&sg, VTNET_MAX_RX_SEGS, segs);
1372
1373         mdata = mtod(m, uint8_t *);
1374         offset = 0;
1375
1376         if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1377                 rxhdr = (struct vtnet_rx_header *) mdata;
1378                 hdr = &rxhdr->vrh_hdr;
1379                 offset += sizeof(struct vtnet_rx_header);
1380
1381                 error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1382                 KASSERT(error == 0, ("cannot add header to sglist"));
1383         }
1384
1385         error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1386         if (error)
1387                 return (error);
1388
1389         if (m->m_next != NULL) {
1390                 error = sglist_append_mbuf(&sg, m->m_next);
1391                 if (error)
1392                         return (error);
1393         }
1394
1395         return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1396 }
1397
1398 static void
1399 vtnet_vlan_tag_remove(struct mbuf *m)
1400 {
1401         struct ether_vlan_header *evl;
1402
1403         evl = mtod(m, struct ether_vlan_header *);
1404
1405         m->m_pkthdr.ether_vlantag = ntohs(evl->evl_tag);
1406         m->m_flags |= M_VLANTAG;
1407
1408         /* Strip the 802.1Q header. */
1409         bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1410             ETHER_HDR_LEN - ETHER_TYPE_LEN);
1411         m_adj(m, ETHER_VLAN_ENCAP_LEN);
1412 }
1413
1414 /*
1415  * Alternative method of doing receive checksum offloading. Rather
1416  * than parsing the received frame down to the IP header, use the
1417  * csum_offset to determine which CSUM_* flags are appropriate. We
1418  * can get by with doing this only because the checksum offsets are
1419  * unique for the things we care about.
1420  */
1421 static int
1422 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1423     struct virtio_net_hdr *hdr)
1424 {
1425         struct ether_header *eh;
1426         struct ether_vlan_header *evh;
1427         struct udphdr *udp;
1428         int csum_len;
1429         uint16_t eth_type;
1430
1431         csum_len = hdr->csum_start + hdr->csum_offset;
1432
1433         if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1434                 return (1);
1435         if (m->m_len < csum_len)
1436                 return (1);
1437
1438         eh = mtod(m, struct ether_header *);
1439         eth_type = ntohs(eh->ether_type);
1440         if (eth_type == ETHERTYPE_VLAN) {
1441                 evh = mtod(m, struct ether_vlan_header *);
1442                 eth_type = ntohs(evh->evl_proto);
1443         }
1444
1445         if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1446                 sc->vtnet_stats.rx_csum_bad_ethtype++;
1447                 return (1);
1448         }
1449
1450         /* Use the offset to determine the appropriate CSUM_* flags. */
1451         switch (hdr->csum_offset) {
1452         case offsetof(struct udphdr, uh_sum):
1453                 if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1454                         return (1);
1455                 udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1456                 if (udp->uh_sum == 0)
1457                         return (0);
1458
1459                 /* FALLTHROUGH */
1460
1461         case offsetof(struct tcphdr, th_sum):
1462                 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1463                 m->m_pkthdr.csum_data = 0xFFFF;
1464                 break;
1465
1466         default:
1467                 sc->vtnet_stats.rx_csum_bad_offset++;
1468                 return (1);
1469         }
1470
1471         sc->vtnet_stats.rx_csum_offloaded++;
1472
1473         return (0);
1474 }
1475
1476 static int
1477 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1478 {
1479         struct ifnet *ifp;
1480         struct virtqueue *vq;
1481         struct mbuf *m, *m_tail;
1482         int len;
1483
1484         ifp = sc->vtnet_ifp;
1485         vq = sc->vtnet_rx_vq;
1486         m_tail = m_head;
1487
1488         while (--nbufs > 0) {
1489                 m = virtqueue_dequeue(vq, &len);
1490                 if (m == NULL) {
1491                         ifp->if_ierrors++;
1492                         goto fail;
1493                 }
1494
1495                 if (vtnet_newbuf(sc) != 0) {
1496                         ifp->if_iqdrops++;
1497                         vtnet_discard_rxbuf(sc, m);
1498                         if (nbufs > 1)
1499                                 vtnet_discard_merged_rxbuf(sc, nbufs);
1500                         goto fail;
1501                 }
1502
1503                 if (m->m_len < len)
1504                         len = m->m_len;
1505
1506                 m->m_len = len;
1507                 m->m_flags &= ~M_PKTHDR;
1508
1509                 m_head->m_pkthdr.len += len;
1510                 m_tail->m_next = m;
1511                 m_tail = m;
1512         }
1513
1514         return (0);
1515
1516 fail:
1517         sc->vtnet_stats.rx_mergeable_failed++;
1518         m_freem(m_head);
1519
1520         return (1);
1521 }
1522
1523 static int
1524 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1525 {
1526         struct virtio_net_hdr lhdr;
1527         struct ifnet *ifp;
1528         struct virtqueue *vq;
1529         struct mbuf *m;
1530         struct ether_header *eh;
1531         struct virtio_net_hdr *hdr;
1532         struct virtio_net_hdr_mrg_rxbuf *mhdr;
1533         int len, deq, nbufs, adjsz, rx_npkts;
1534
1535         ifp = sc->vtnet_ifp;
1536         vq = sc->vtnet_rx_vq;
1537         hdr = &lhdr;
1538         deq = 0;
1539         rx_npkts = 0;
1540
1541         ASSERT_SERIALIZED(&sc->vtnet_slz);
1542
1543         while (--count >= 0) {
1544                 m = virtqueue_dequeue(vq, &len);
1545                 if (m == NULL)
1546                         break;
1547                 deq++;
1548
1549                 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1550                         ifp->if_ierrors++;
1551                         vtnet_discard_rxbuf(sc, m);
1552                         continue;
1553                 }
1554
1555                 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1556                         nbufs = 1;
1557                         adjsz = sizeof(struct vtnet_rx_header);
1558                         /*
1559                          * Account for our pad between the header and
1560                          * the actual start of the frame.
1561                          */
1562                         len += VTNET_RX_HEADER_PAD;
1563                 } else {
1564                         mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1565                         nbufs = mhdr->num_buffers;
1566                         adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1567                 }
1568
1569                 if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1570                         ifp->if_iqdrops++;
1571                         vtnet_discard_rxbuf(sc, m);
1572                         if (nbufs > 1)
1573                                 vtnet_discard_merged_rxbuf(sc, nbufs);
1574                         continue;
1575                 }
1576
1577                 m->m_pkthdr.len = len;
1578                 m->m_pkthdr.rcvif = ifp;
1579                 m->m_pkthdr.csum_flags = 0;
1580
1581                 if (nbufs > 1) {
1582                         if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1583                                 continue;
1584                 }
1585
1586                 ifp->if_ipackets++;
1587
1588                 /*
1589                  * Save copy of header before we strip it. For both mergeable
1590                  * and non-mergeable, the VirtIO header is placed first in the
1591                  * mbuf's data. We no longer need num_buffers, so always use a
1592                  * virtio_net_hdr.
1593                  */
1594                 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1595                 m_adj(m, adjsz);
1596
1597                 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1598                         eh = mtod(m, struct ether_header *);
1599                         if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1600                                 vtnet_vlan_tag_remove(m);
1601
1602                                 /*
1603                                  * With the 802.1Q header removed, update the
1604                                  * checksum starting location accordingly.
1605                                  */
1606                                 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1607                                         hdr->csum_start -=
1608                                             ETHER_VLAN_ENCAP_LEN;
1609                         }
1610                 }
1611
1612                 if (ifp->if_capenable & IFCAP_RXCSUM &&
1613                     hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1614                         if (vtnet_rx_csum(sc, m, hdr) != 0)
1615                                 sc->vtnet_stats.rx_csum_failed++;
1616                 }
1617
1618                 lwkt_serialize_exit(&sc->vtnet_slz);
1619                 rx_npkts++;
1620                 ifp->if_input(ifp, m, NULL, -1);
1621                 lwkt_serialize_enter(&sc->vtnet_slz);
1622
1623                 /*
1624                  * The interface may have been stopped while we were
1625                  * passing the packet up the network stack.
1626                  */
1627                 if ((ifp->if_flags & IFF_RUNNING) == 0)
1628                         break;
1629         }
1630
1631         virtqueue_notify(vq, &sc->vtnet_slz);
1632
1633         if (rx_npktsp != NULL)
1634                 *rx_npktsp = rx_npkts;
1635
1636         return (count > 0 ? 0 : EAGAIN);
1637 }
1638
1639 static void
1640 vtnet_rx_intr_task(void *arg)
1641 {
1642         struct vtnet_softc *sc;
1643         struct ifnet *ifp;
1644         int more;
1645
1646         sc = arg;
1647         ifp = sc->vtnet_ifp;
1648
1649 next:
1650 //      lwkt_serialize_enter(&sc->vtnet_slz);
1651
1652         if ((ifp->if_flags & IFF_RUNNING) == 0) {
1653                 vtnet_enable_rx_intr(sc);
1654 //              lwkt_serialize_exit(&sc->vtnet_slz);
1655                 return;
1656         }
1657
1658         more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1659         if (!more && vtnet_enable_rx_intr(sc) != 0) {
1660                 vtnet_disable_rx_intr(sc);
1661                 more = 1;
1662         }
1663
1664 //      lwkt_serialize_exit(&sc->vtnet_slz);
1665
1666         if (more) {
1667                 sc->vtnet_stats.rx_task_rescheduled++;
1668                 goto next;
1669         }
1670 }
1671
1672 static int
1673 vtnet_rx_vq_intr(void *xsc)
1674 {
1675         struct vtnet_softc *sc;
1676
1677         sc = xsc;
1678
1679         vtnet_disable_rx_intr(sc);
1680         vtnet_rx_intr_task(sc);
1681
1682         return (1);
1683 }
1684
1685 static void
1686 vtnet_txeof(struct vtnet_softc *sc)
1687 {
1688         struct virtqueue *vq;
1689         struct ifnet *ifp;
1690         struct vtnet_tx_header *txhdr;
1691         int deq;
1692
1693         vq = sc->vtnet_tx_vq;
1694         ifp = sc->vtnet_ifp;
1695         deq = 0;
1696
1697         ASSERT_SERIALIZED(&sc->vtnet_slz);
1698
1699         while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1700                 deq++;
1701                 ifp->if_opackets++;
1702                 m_freem(txhdr->vth_mbuf);
1703         }
1704
1705         if (deq > 0) {
1706                 ifq_clr_oactive(&ifp->if_snd);
1707                 if (virtqueue_empty(vq))
1708                         sc->vtnet_watchdog_timer = 0;
1709         }
1710 }
1711
1712 static struct mbuf *
1713 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1714     struct virtio_net_hdr *hdr)
1715 {
1716         struct ifnet *ifp;
1717         struct ether_header *eh;
1718         struct ether_vlan_header *evh;
1719         struct ip *ip;
1720         struct ip6_hdr *ip6;
1721         struct tcphdr *tcp;
1722         int ip_offset;
1723         uint16_t eth_type, csum_start;
1724         uint8_t ip_proto, gso_type;
1725
1726         ifp = sc->vtnet_ifp;
1727         M_ASSERTPKTHDR(m);
1728
1729         ip_offset = sizeof(struct ether_header);
1730         if (m->m_len < ip_offset) {
1731                 if ((m = m_pullup(m, ip_offset)) == NULL)
1732                         return (NULL);
1733         }
1734
1735         eh = mtod(m, struct ether_header *);
1736         eth_type = ntohs(eh->ether_type);
1737         if (eth_type == ETHERTYPE_VLAN) {
1738                 ip_offset = sizeof(struct ether_vlan_header);
1739                 if (m->m_len < ip_offset) {
1740                         if ((m = m_pullup(m, ip_offset)) == NULL)
1741                                 return (NULL);
1742                 }
1743                 evh = mtod(m, struct ether_vlan_header *);
1744                 eth_type = ntohs(evh->evl_proto);
1745         }
1746
1747         switch (eth_type) {
1748         case ETHERTYPE_IP:
1749                 if (m->m_len < ip_offset + sizeof(struct ip)) {
1750                         m = m_pullup(m, ip_offset + sizeof(struct ip));
1751                         if (m == NULL)
1752                                 return (NULL);
1753                 }
1754
1755                 ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1756                 ip_proto = ip->ip_p;
1757                 csum_start = ip_offset + (ip->ip_hl << 2);
1758                 gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1759                 break;
1760
1761         case ETHERTYPE_IPV6:
1762                 if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1763                         m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1764                         if (m == NULL)
1765                                 return (NULL);
1766                 }
1767
1768                 ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1769                 /*
1770                  * XXX Assume no extension headers are present. Presently,
1771                  * this will always be true in the case of TSO, and FreeBSD
1772                  * does not perform checksum offloading of IPv6 yet.
1773                  */
1774                 ip_proto = ip6->ip6_nxt;
1775                 csum_start = ip_offset + sizeof(struct ip6_hdr);
1776                 gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1777                 break;
1778
1779         default:
1780                 return (m);
1781         }
1782
1783         if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1784                 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1785                 hdr->csum_start = csum_start;
1786                 hdr->csum_offset = m->m_pkthdr.csum_data;
1787
1788                 sc->vtnet_stats.tx_csum_offloaded++;
1789         }
1790
1791         if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1792                 if (ip_proto != IPPROTO_TCP)
1793                         return (m);
1794
1795                 if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1796                         m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1797                         if (m == NULL)
1798                                 return (NULL);
1799                 }
1800
1801                 tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1802                 hdr->gso_type = gso_type;
1803                 hdr->hdr_len = csum_start + (tcp->th_off << 2);
1804                 hdr->gso_size = m->m_pkthdr.tso_segsz;
1805
1806                 if (tcp->th_flags & TH_CWR) {
1807                         /*
1808                          * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1809                          * ECN support is only configurable globally with the
1810                          * net.inet.tcp.ecn.enable sysctl knob.
1811                          */
1812                         if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1813                                 if_printf(ifp, "TSO with ECN not supported "
1814                                     "by host\n");
1815                                 m_freem(m);
1816                                 return (NULL);
1817                         }
1818
1819                         hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1820                 }
1821
1822                 sc->vtnet_stats.tx_tso_offloaded++;
1823         }
1824
1825         return (m);
1826 }
1827
1828 static int
1829 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1830     struct vtnet_tx_header *txhdr)
1831 {
1832         struct sglist sg;
1833         struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1834         struct virtqueue *vq;
1835         struct mbuf *m;
1836         int collapsed, error;
1837
1838         vq = sc->vtnet_tx_vq;
1839         m = *m_head;
1840         collapsed = 0;
1841
1842         sglist_init(&sg, VTNET_MAX_TX_SEGS, segs);
1843         error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1844         KASSERT(error == 0 && sg.sg_nseg == 1,
1845             ("cannot add header to sglist"));
1846
1847 again:
1848         error = sglist_append_mbuf(&sg, m);
1849         if (error) {
1850                 if (collapsed)
1851                         goto fail;
1852
1853                 //m = m_collapse(m, M_NOWAIT, VTNET_MAX_TX_SEGS - 1);
1854                 m = m_defrag(m, M_NOWAIT);
1855                 if (m == NULL)
1856                         goto fail;
1857
1858                 *m_head = m;
1859                 collapsed = 1;
1860                 goto again;
1861         }
1862
1863         txhdr->vth_mbuf = m;
1864
1865         return (virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0));
1866
1867 fail:
1868         m_freem(*m_head);
1869         *m_head = NULL;
1870
1871         return (ENOBUFS);
1872 }
1873
1874 static struct mbuf *
1875 vtnet_vlan_tag_insert(struct mbuf *m)
1876 {
1877         struct mbuf *n;
1878         struct ether_vlan_header *evl;
1879
1880         if (M_WRITABLE(m) == 0) {
1881                 n = m_dup(m, M_NOWAIT);
1882                 m_freem(m);
1883                 if ((m = n) == NULL)
1884                         return (NULL);
1885         }
1886
1887         M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT);
1888         if (m == NULL)
1889                 return (NULL);
1890         if (m->m_len < sizeof(struct ether_vlan_header)) {
1891                 m = m_pullup(m, sizeof(struct ether_vlan_header));
1892                 if (m == NULL)
1893                         return (NULL);
1894         }
1895
1896         /* Insert 802.1Q header into the existing Ethernet header. */
1897         evl = mtod(m, struct ether_vlan_header *);
1898         bcopy((char *) evl + ETHER_VLAN_ENCAP_LEN,
1899               (char *) evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
1900         evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1901         evl->evl_tag = htons(m->m_pkthdr.ether_vlantag);
1902         m->m_flags &= ~M_VLANTAG;
1903
1904         return (m);
1905 }
1906
1907 static int
1908 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1909 {
1910         struct vtnet_tx_header *txhdr;
1911         struct virtio_net_hdr *hdr;
1912         struct mbuf *m;
1913         int error;
1914
1915         txhdr = &sc->vtnet_txhdrarea[sc->vtnet_txhdridx];
1916         memset(txhdr, 0, sizeof(struct vtnet_tx_header));
1917
1918         /*
1919          * Always use the non-mergeable header to simplify things. When
1920          * the mergeable feature is negotiated, the num_buffers field
1921          * must be set to zero. We use vtnet_hdr_size later to enqueue
1922          * the correct header size to the host.
1923          */
1924         hdr = &txhdr->vth_uhdr.hdr;
1925         m = *m_head;
1926
1927         error = ENOBUFS;
1928
1929         if (m->m_flags & M_VLANTAG) {
1930                 //m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1931                 m = vtnet_vlan_tag_insert(m);
1932                 if ((*m_head = m) == NULL)
1933                         goto fail;
1934                 m->m_flags &= ~M_VLANTAG;
1935         }
1936
1937         if (m->m_pkthdr.csum_flags != 0) {
1938                 m = vtnet_tx_offload(sc, m, hdr);
1939                 if ((*m_head = m) == NULL)
1940                         goto fail;
1941         }
1942
1943         error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1944         if (error == 0)
1945                 sc->vtnet_txhdridx =
1946                     (sc->vtnet_txhdridx + 1) % ((sc->vtnet_tx_size / 2) + 1);
1947 fail:
1948         return (error);
1949 }
1950
1951 static void
1952 vtnet_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1953 {
1954         struct vtnet_softc *sc;
1955
1956         sc = ifp->if_softc;
1957
1958         ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1959         lwkt_serialize_enter(&sc->vtnet_slz);
1960         vtnet_start_locked(ifp, ifsq);
1961         lwkt_serialize_exit(&sc->vtnet_slz);
1962 }
1963
1964 static void
1965 vtnet_start_locked(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1966 {
1967         struct vtnet_softc *sc;
1968         struct virtqueue *vq;
1969         struct mbuf *m0;
1970         int enq;
1971
1972         sc = ifp->if_softc;
1973         vq = sc->vtnet_tx_vq;
1974         enq = 0;
1975
1976         ASSERT_SERIALIZED(&sc->vtnet_slz);
1977
1978         if ((ifp->if_flags & (IFF_RUNNING)) !=
1979             IFF_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
1980                 return;
1981
1982 #ifdef VTNET_TX_INTR_MODERATION
1983         if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
1984                 vtnet_txeof(sc);
1985 #endif
1986
1987         while (!ifsq_is_empty(ifsq)) {
1988                 if (virtqueue_full(vq)) {
1989                         ifq_set_oactive(&ifp->if_snd);
1990                         break;
1991                 }
1992
1993                 m0 = ifq_dequeue(&ifp->if_snd);
1994                 if (m0 == NULL)
1995                         break;
1996
1997                 if (vtnet_encap(sc, &m0) != 0) {
1998                         if (m0 == NULL)
1999                                 break;
2000                         ifq_prepend(&ifp->if_snd, m0);
2001                         ifq_set_oactive(&ifp->if_snd);
2002                         break;
2003                 }
2004
2005                 enq++;
2006                 ETHER_BPF_MTAP(ifp, m0);
2007         }
2008
2009         if (enq > 0) {
2010                 virtqueue_notify(vq, &sc->vtnet_slz);
2011                 sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
2012         }
2013 }
2014
2015 static void
2016 vtnet_tick(void *xsc)
2017 {
2018         struct vtnet_softc *sc;
2019
2020         sc = xsc;
2021
2022 #if 0
2023         ASSERT_SERIALIZED(&sc->vtnet_slz);
2024 #ifdef VTNET_DEBUG
2025         virtqueue_dump(sc->vtnet_rx_vq);
2026         virtqueue_dump(sc->vtnet_tx_vq);
2027 #endif
2028
2029         vtnet_watchdog(sc);
2030         callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2031 #endif
2032 }
2033
2034 static void
2035 vtnet_tx_intr_task(void *arg)
2036 {
2037         struct vtnet_softc *sc;
2038         struct ifnet *ifp;
2039         struct ifaltq_subque *ifsq;
2040
2041         sc = arg;
2042         ifp = sc->vtnet_ifp;
2043         ifsq = ifq_get_subq_default(&ifp->if_snd);
2044
2045 next:
2046 //      lwkt_serialize_enter(&sc->vtnet_slz);
2047
2048         if ((ifp->if_flags & IFF_RUNNING) == 0) {
2049                 vtnet_enable_tx_intr(sc);
2050 //              lwkt_serialize_exit(&sc->vtnet_slz);
2051                 return;
2052         }
2053
2054         vtnet_txeof(sc);
2055
2056         if (!ifsq_is_empty(ifsq))
2057                 vtnet_start_locked(ifp, ifsq);
2058
2059         if (vtnet_enable_tx_intr(sc) != 0) {
2060                 vtnet_disable_tx_intr(sc);
2061                 sc->vtnet_stats.tx_task_rescheduled++;
2062 //              lwkt_serialize_exit(&sc->vtnet_slz);
2063                 goto next;
2064         }
2065
2066 //      lwkt_serialize_exit(&sc->vtnet_slz);
2067 }
2068
2069 static int
2070 vtnet_tx_vq_intr(void *xsc)
2071 {
2072         struct vtnet_softc *sc;
2073
2074         sc = xsc;
2075
2076         vtnet_disable_tx_intr(sc);
2077         vtnet_tx_intr_task(sc);
2078
2079         return (1);
2080 }
2081
2082 static void
2083 vtnet_stop(struct vtnet_softc *sc)
2084 {
2085         device_t dev;
2086         struct ifnet *ifp;
2087
2088         dev = sc->vtnet_dev;
2089         ifp = sc->vtnet_ifp;
2090
2091         ASSERT_SERIALIZED(&sc->vtnet_slz);
2092
2093         sc->vtnet_watchdog_timer = 0;
2094         callout_stop(&sc->vtnet_tick_ch);
2095         ifq_clr_oactive(&ifp->if_snd);
2096         ifp->if_flags &= ~(IFF_RUNNING);
2097
2098         vtnet_disable_rx_intr(sc);
2099         vtnet_disable_tx_intr(sc);
2100
2101         /*
2102          * Stop the host VirtIO adapter. Note this will reset the host
2103          * adapter's state back to the pre-initialized state, so in
2104          * order to make the device usable again, we must drive it
2105          * through virtio_reinit() and virtio_reinit_complete().
2106          */
2107         virtio_stop(dev);
2108
2109         sc->vtnet_flags &= ~VTNET_FLAG_LINK;
2110
2111         vtnet_free_rx_mbufs(sc);
2112         vtnet_free_tx_mbufs(sc);
2113 }
2114
2115 static int
2116 vtnet_reinit(struct vtnet_softc *sc)
2117 {
2118         struct ifnet *ifp;
2119         uint64_t features;
2120
2121         ifp = sc->vtnet_ifp;
2122         features = sc->vtnet_features;
2123
2124         /*
2125          * Re-negotiate with the host, removing any disabled receive
2126          * features. Transmit features are disabled only on our side
2127          * via if_capenable and if_hwassist.
2128          */
2129
2130         if (ifp->if_capabilities & IFCAP_RXCSUM) {
2131                 if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2132                         features &= ~VIRTIO_NET_F_GUEST_CSUM;
2133         }
2134
2135         if (ifp->if_capabilities & IFCAP_LRO) {
2136                 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2137                         features &= ~VTNET_LRO_FEATURES;
2138         }
2139
2140         if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2141                 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2142                         features &= ~VIRTIO_NET_F_CTRL_VLAN;
2143         }
2144
2145         return (virtio_reinit(sc->vtnet_dev, features));
2146 }
2147
2148 static void
2149 vtnet_init_locked(struct vtnet_softc *sc)
2150 {
2151         device_t dev;
2152         struct ifnet *ifp;
2153         int error;
2154
2155         dev = sc->vtnet_dev;
2156         ifp = sc->vtnet_ifp;
2157
2158         ASSERT_SERIALIZED(&sc->vtnet_slz);
2159
2160         if (ifp->if_flags & IFF_RUNNING)
2161                 return;
2162
2163         /* Stop host's adapter, cancel any pending I/O. */
2164         vtnet_stop(sc);
2165
2166         /* Reinitialize the host device. */
2167         error = vtnet_reinit(sc);
2168         if (error) {
2169                 device_printf(dev,
2170                     "reinitialization failed, stopping device...\n");
2171                 vtnet_stop(sc);
2172                 return;
2173         }
2174
2175         /* Update host with assigned MAC address. */
2176         bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2177         vtnet_set_hwaddr(sc);
2178
2179         ifp->if_hwassist = 0;
2180         if (ifp->if_capenable & IFCAP_TXCSUM)
2181                 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2182         if (ifp->if_capenable & IFCAP_TSO4)
2183                 ifp->if_hwassist |= CSUM_TSO;
2184
2185         error = vtnet_init_rx_vq(sc);
2186         if (error) {
2187                 device_printf(dev,
2188                     "cannot allocate mbufs for Rx virtqueue\n");
2189                 vtnet_stop(sc);
2190                 return;
2191         }
2192
2193         if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2194                 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2195                         /* Restore promiscuous and all-multicast modes. */
2196                         vtnet_rx_filter(sc);
2197
2198                         /* Restore filtered MAC addresses. */
2199                         vtnet_rx_filter_mac(sc);
2200                 }
2201
2202                 /* Restore VLAN filters. */
2203                 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2204                         vtnet_rx_filter_vlan(sc);
2205         }
2206
2207         {
2208                 vtnet_enable_rx_intr(sc);
2209                 vtnet_enable_tx_intr(sc);
2210         }
2211
2212         ifp->if_flags |= IFF_RUNNING;
2213         ifq_clr_oactive(&ifp->if_snd);
2214
2215         virtio_reinit_complete(dev);
2216
2217         vtnet_update_link_status(sc);
2218         callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2219 }
2220
2221 static void
2222 vtnet_init(void *xsc)
2223 {
2224         struct vtnet_softc *sc;
2225
2226         sc = xsc;
2227
2228         lwkt_serialize_enter(&sc->vtnet_slz);
2229         vtnet_init_locked(sc);
2230         lwkt_serialize_exit(&sc->vtnet_slz);
2231 }
2232
2233 static void
2234 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2235     struct sglist *sg, int readable, int writable)
2236 {
2237         struct virtqueue *vq;
2238         void *c;
2239
2240         vq = sc->vtnet_ctrl_vq;
2241
2242         ASSERT_SERIALIZED(&sc->vtnet_slz);
2243         KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2244             ("no control virtqueue"));
2245         KASSERT(virtqueue_empty(vq),
2246             ("control command already enqueued"));
2247
2248         if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2249                 return;
2250
2251         virtqueue_notify(vq, &sc->vtnet_slz);
2252
2253         /*
2254          * Poll until the command is complete. Previously, we would
2255          * sleep until the control virtqueue interrupt handler woke
2256          * us up, but dropping the VTNET_MTX leads to serialization
2257          * difficulties.
2258          *
2259          * Furthermore, it appears QEMU/KVM only allocates three MSIX
2260          * vectors. Two of those vectors are needed for the Rx and Tx
2261          * virtqueues. We do not support sharing both a Vq and config
2262          * changed notification on the same MSIX vector.
2263          */
2264         c = virtqueue_poll(vq, NULL);
2265         KASSERT(c == cookie, ("unexpected control command response"));
2266 }
2267
2268 static void
2269 vtnet_rx_filter(struct vtnet_softc *sc)
2270 {
2271         device_t dev;
2272         struct ifnet *ifp;
2273
2274         dev = sc->vtnet_dev;
2275         ifp = sc->vtnet_ifp;
2276
2277         ASSERT_SERIALIZED(&sc->vtnet_slz);
2278         KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2279             ("CTRL_RX feature not negotiated"));
2280
2281         if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2282                 device_printf(dev, "cannot %s promiscuous mode\n",
2283                     ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
2284
2285         if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2286                 device_printf(dev, "cannot %s all-multicast mode\n",
2287                     ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
2288 }
2289
2290 static int
2291 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2292 {
2293         struct virtio_net_ctrl_hdr hdr __aligned(2);
2294         struct sglist_seg segs[3];
2295         struct sglist sg;
2296         uint8_t onoff, ack;
2297         int error;
2298
2299         if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0)
2300                 return (ENOTSUP);
2301
2302         error = 0;
2303
2304         hdr.class = VIRTIO_NET_CTRL_RX;
2305         hdr.cmd = cmd;
2306         onoff = !!on;
2307         ack = VIRTIO_NET_ERR;
2308
2309         sglist_init(&sg, 3, segs);
2310         error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2311         error |= sglist_append(&sg, &onoff, sizeof(uint8_t));
2312         error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2313         KASSERT(error == 0 && sg.sg_nseg == 3,
2314             ("error adding Rx filter message to sglist"));
2315
2316         vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2317
2318         return (ack == VIRTIO_NET_OK ? 0 : EIO);
2319 }
2320
2321 static int
2322 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2323 {
2324
2325         return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2326 }
2327
2328 static int
2329 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2330 {
2331
2332         return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2333 }
2334
2335 static void
2336 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2337 {
2338         struct virtio_net_ctrl_hdr hdr __aligned(2);
2339         struct vtnet_mac_filter *filter;
2340         struct sglist_seg segs[4];
2341         struct sglist sg;
2342         struct ifnet *ifp;
2343         struct ifaddr *ifa;
2344         struct ifaddr_container *ifac;
2345         struct ifmultiaddr *ifma;
2346         int ucnt, mcnt, promisc, allmulti, error;
2347         uint8_t ack;
2348
2349         ifp = sc->vtnet_ifp;
2350         ucnt = 0;
2351         mcnt = 0;
2352         promisc = 0;
2353         allmulti = 0;
2354         error = 0;
2355
2356         ASSERT_SERIALIZED(&sc->vtnet_slz);
2357         KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2358             ("CTRL_RX feature not negotiated"));
2359
2360         /* Use the MAC filtering table allocated in vtnet_attach. */
2361         filter = sc->vtnet_macfilter;
2362         memset(filter, 0, sizeof(struct vtnet_mac_filter));
2363
2364         /* Unicast MAC addresses: */
2365         //if_addr_rlock(ifp);
2366         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2367                 ifa = ifac->ifa;
2368                 if (ifa->ifa_addr->sa_family != AF_LINK)
2369                         continue;
2370                 else if (ucnt == VTNET_MAX_MAC_ENTRIES)
2371                         break;
2372
2373                 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2374                     &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2375                 ucnt++;
2376         }
2377         //if_addr_runlock(ifp);
2378
2379         if (ucnt >= VTNET_MAX_MAC_ENTRIES) {
2380                 promisc = 1;
2381                 filter->vmf_unicast.nentries = 0;
2382
2383                 if_printf(ifp, "more than %d MAC addresses assigned, "
2384                     "falling back to promiscuous mode\n",
2385                     VTNET_MAX_MAC_ENTRIES);
2386         } else
2387                 filter->vmf_unicast.nentries = ucnt;
2388
2389         /* Multicast MAC addresses: */
2390         //if_maddr_rlock(ifp);
2391         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2392                 if (ifma->ifma_addr->sa_family != AF_LINK)
2393                         continue;
2394                 else if (mcnt == VTNET_MAX_MAC_ENTRIES)
2395                         break;
2396
2397                 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2398                     &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2399                 mcnt++;
2400         }
2401         //if_maddr_runlock(ifp);
2402
2403         if (mcnt >= VTNET_MAX_MAC_ENTRIES) {
2404                 allmulti = 1;
2405                 filter->vmf_multicast.nentries = 0;
2406
2407                 if_printf(ifp, "more than %d multicast MAC addresses "
2408                     "assigned, falling back to all-multicast mode\n",
2409                     VTNET_MAX_MAC_ENTRIES);
2410         } else
2411                 filter->vmf_multicast.nentries = mcnt;
2412
2413         if (promisc && allmulti)
2414                 goto out;
2415
2416         hdr.class = VIRTIO_NET_CTRL_MAC;
2417         hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2418         ack = VIRTIO_NET_ERR;
2419
2420         sglist_init(&sg, 4, segs);
2421         error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2422         error |= sglist_append(&sg, &filter->vmf_unicast,
2423             sizeof(struct vtnet_mac_table));
2424         error |= sglist_append(&sg, &filter->vmf_multicast,
2425             sizeof(struct vtnet_mac_table));
2426         error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2427         KASSERT(error == 0 && sg.sg_nseg == 4,
2428             ("error adding MAC filtering message to sglist"));
2429
2430         vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2431
2432         if (ack != VIRTIO_NET_OK)
2433                 if_printf(ifp, "error setting host MAC filter table\n");
2434
2435 out:
2436         if (promisc)
2437                 if (vtnet_set_promisc(sc, 1) != 0)
2438                         if_printf(ifp, "cannot enable promiscuous mode\n");
2439         if (allmulti)
2440                 if (vtnet_set_allmulti(sc, 1) != 0)
2441                         if_printf(ifp, "cannot enable all-multicast mode\n");
2442 }
2443
2444 static int
2445 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2446 {
2447         struct virtio_net_ctrl_hdr hdr __aligned(2);
2448         struct sglist_seg segs[3];
2449         struct sglist sg;
2450         uint8_t ack;
2451         int error;
2452
2453         hdr.class = VIRTIO_NET_CTRL_VLAN;
2454         hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2455         ack = VIRTIO_NET_ERR;
2456         error = 0;
2457
2458         sglist_init(&sg, 3, segs);
2459         error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2460         error |= sglist_append(&sg, &tag, sizeof(uint16_t));
2461         error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2462         KASSERT(error == 0 && sg.sg_nseg == 3,
2463             ("error adding VLAN control message to sglist"));
2464
2465         vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2466
2467         return (ack == VIRTIO_NET_OK ? 0 : EIO);
2468 }
2469
2470 static void
2471 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2472 {
2473         device_t dev;
2474         uint32_t w, mask;
2475         uint16_t tag;
2476         int i, nvlans, error;
2477
2478         ASSERT_SERIALIZED(&sc->vtnet_slz);
2479         KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2480             ("VLAN_FILTER feature not negotiated"));
2481
2482         dev = sc->vtnet_dev;
2483         nvlans = sc->vtnet_nvlans;
2484         error = 0;
2485
2486         /* Enable filtering for each configured VLAN. */
2487         for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2488                 w = sc->vtnet_vlan_shadow[i];
2489                 for (mask = 1, tag = i * 32; w != 0; mask <<= 1, tag++) {
2490                         if ((w & mask) != 0) {
2491                                 w &= ~mask;
2492                                 nvlans--;
2493                                 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0)
2494                                         error++;
2495                         }
2496                 }
2497         }
2498
2499         KASSERT(nvlans == 0, ("VLAN count incorrect"));
2500         if (error)
2501                 device_printf(dev, "cannot restore VLAN filter table\n");
2502 }
2503
2504 static void
2505 vtnet_set_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2506 {
2507         struct ifnet *ifp;
2508         int idx, bit;
2509
2510         KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2511             ("VLAN_FILTER feature not negotiated"));
2512
2513         if ((tag == 0) || (tag > 4095))
2514                 return;
2515
2516         ifp = sc->vtnet_ifp;
2517         idx = (tag >> 5) & 0x7F;
2518         bit = tag & 0x1F;
2519
2520         lwkt_serialize_enter(&sc->vtnet_slz);
2521
2522         /* Update shadow VLAN table. */
2523         if (add) {
2524                 sc->vtnet_nvlans++;
2525                 sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2526         } else {
2527                 sc->vtnet_nvlans--;
2528                 sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2529         }
2530
2531         if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) {
2532                 if (vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2533                         device_printf(sc->vtnet_dev,
2534                             "cannot %s VLAN %d %s the host filter table\n",
2535                             add ? "add" : "remove", tag,
2536                             add ? "to" : "from");
2537                 }
2538         }
2539
2540         lwkt_serialize_exit(&sc->vtnet_slz);
2541 }
2542
2543 static void
2544 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2545 {
2546
2547         if (ifp->if_softc != arg)
2548                 return;
2549
2550         vtnet_set_vlan_filter(arg, 1, tag);
2551 }
2552
2553 static void
2554 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2555 {
2556
2557         if (ifp->if_softc != arg)
2558                 return;
2559
2560         vtnet_set_vlan_filter(arg, 0, tag);
2561 }
2562
2563 static int
2564 vtnet_ifmedia_upd(struct ifnet *ifp)
2565 {
2566         struct vtnet_softc *sc;
2567         struct ifmedia *ifm;
2568
2569         sc = ifp->if_softc;
2570         ifm = &sc->vtnet_media;
2571
2572         if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2573                 return (EINVAL);
2574
2575         return (0);
2576 }
2577
2578 static void
2579 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2580 {
2581         struct vtnet_softc *sc;
2582
2583         sc = ifp->if_softc;
2584
2585         ifmr->ifm_status = IFM_AVALID;
2586         ifmr->ifm_active = IFM_ETHER;
2587
2588         lwkt_serialize_enter(&sc->vtnet_slz);
2589         if (vtnet_is_link_up(sc) != 0) {
2590                 ifmr->ifm_status |= IFM_ACTIVE;
2591                 ifmr->ifm_active |= VTNET_MEDIATYPE;
2592         } else
2593                 ifmr->ifm_active |= IFM_NONE;
2594         lwkt_serialize_exit(&sc->vtnet_slz);
2595 }
2596
2597 static void
2598 vtnet_add_statistics(struct vtnet_softc *sc)
2599 {
2600         device_t dev;
2601         struct vtnet_statistics *stats;
2602         struct sysctl_ctx_list *ctx;
2603         struct sysctl_oid *tree;
2604         struct sysctl_oid_list *child;
2605
2606         dev = sc->vtnet_dev;
2607         stats = &sc->vtnet_stats;
2608         ctx = device_get_sysctl_ctx(dev);
2609         tree = device_get_sysctl_tree(dev);
2610         child = SYSCTL_CHILDREN(tree);
2611
2612         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2613             CTLFLAG_RD, &stats->mbuf_alloc_failed,
2614             "Mbuf cluster allocation failures");
2615         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_frame_too_large",
2616             CTLFLAG_RD, &stats->rx_frame_too_large,
2617             "Received frame larger than the mbuf chain");
2618         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2619             CTLFLAG_RD, &stats->rx_enq_replacement_failed,
2620             "Enqueuing the replacement receive mbuf failed");
2621         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_mergeable_failed",
2622             CTLFLAG_RD, &stats->rx_mergeable_failed,
2623             "Mergeable buffers receive failures");
2624         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2625             CTLFLAG_RD, &stats->rx_csum_bad_ethtype,
2626             "Received checksum offloaded buffer with unsupported "
2627             "Ethernet type");
2628         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_start",
2629             CTLFLAG_RD, &stats->rx_csum_bad_start,
2630             "Received checksum offloaded buffer with incorrect start offset");
2631         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2632             CTLFLAG_RD, &stats->rx_csum_bad_ipproto,
2633             "Received checksum offloaded buffer with incorrect IP protocol");
2634         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2635             CTLFLAG_RD, &stats->rx_csum_bad_offset,
2636             "Received checksum offloaded buffer with incorrect offset");
2637         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_failed",
2638             CTLFLAG_RD, &stats->rx_csum_failed,
2639             "Received buffer checksum offload failed");
2640         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_csum_offloaded",
2641             CTLFLAG_RD, &stats->rx_csum_offloaded,
2642             "Received buffer checksum offload succeeded");
2643         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "rx_task_rescheduled",
2644             CTLFLAG_RD, &stats->rx_task_rescheduled,
2645             "Times the receive interrupt task rescheduled itself");
2646
2647         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_offloaded",
2648             CTLFLAG_RD, &stats->tx_csum_offloaded,
2649             "Offloaded checksum of transmitted buffer");
2650         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_offloaded",
2651             CTLFLAG_RD, &stats->tx_tso_offloaded,
2652             "Segmentation offload of transmitted buffer");
2653         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2654             CTLFLAG_RD, &stats->tx_csum_bad_ethtype,
2655             "Aborted transmit of checksum offloaded buffer with unknown "
2656             "Ethernet type");
2657         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2658             CTLFLAG_RD, &stats->tx_tso_bad_ethtype,
2659             "Aborted transmit of TSO buffer with unknown Ethernet type");
2660         SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_task_rescheduled",
2661             CTLFLAG_RD, &stats->tx_task_rescheduled,
2662             "Times the transmit interrupt task rescheduled itself");
2663 }
2664
2665 static int
2666 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2667 {
2668
2669         return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2670 }
2671
2672 static void
2673 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2674 {
2675
2676         virtqueue_disable_intr(sc->vtnet_rx_vq);
2677 }
2678
2679 static int
2680 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2681 {
2682
2683 #ifdef VTNET_TX_INTR_MODERATION
2684         return (0);
2685 #else
2686         return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2687 #endif
2688 }
2689
2690 static void
2691 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2692 {
2693
2694         virtqueue_disable_intr(sc->vtnet_tx_vq);
2695 }