hw/virtio-net.c

   1 /*
   2  * Virtio Network Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "virtio.h"
  15 #include "net.h"
  16 #include "qemu-timer.h"
  17 #include "virtio-net.h"
  18 #ifdef CONFIG_KVM
  19 #include "qemu-kvm.h"
  20 #endif
  21
  22 #define TAP_VNET_HDR
  23
  24 #define VIRTIO_NET_VM_VERSION    10
  25
  26 #define MAC_TABLE_ENTRIES    64
  27 #define MAX_VLAN    (1 << 12)   /* Per 802.1Q definition */
  28
  29 typedef struct VirtIONet
  30 {
  31     VirtIODevice vdev;
  32     uint8_t mac[ETH_ALEN];
  33     uint16_t status;
  34     VirtQueue *rx_vq;
  35     VirtQueue *tx_vq;
  36     VirtQueue *ctrl_vq;
  37     VLANClientState *vc;
  38     QEMUTimer *tx_timer;
  39     int tx_timer_active;
  40     struct {
  41         VirtQueueElement elem;
  42         ssize_t len;
  43     } async_tx;
  44     int mergeable_rx_bufs;
  45     uint8_t promisc;
  46     uint8_t allmulti;
  47     uint8_t alluni;
  48     uint8_t nomulti;
  49     uint8_t nouni;
  50     uint8_t nobcast;
  51     struct {
  52         int in_use;
  53         int first_multi;
  54         uint8_t multi_overflow;
  55         uint8_t uni_overflow;
  56         uint8_t *macs;
  57     } mac_table;
  58     uint32_t *vlans;
  59 } VirtIONet;
  60
  61 /* TODO
  62  * - we could suppress RX interrupt if we were so inclined.
  63  */
  64
  65 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  66 {
  67     return (VirtIONet *)vdev;
  68 }
  69
  70 static void virtio_net_get_config(VirtIODevice *vdev, uint8_t *config)
  71 {
  72     VirtIONet *n = to_virtio_net(vdev);
  73     struct virtio_net_config netcfg;
  74
  75     netcfg.status = n->status;
  76     memcpy(netcfg.mac, n->mac, ETH_ALEN);
  77     memcpy(config, &netcfg, sizeof(netcfg));
  78 }
  79
  80 static void virtio_net_set_config(VirtIODevice *vdev, const uint8_t *config)
  81 {
  82     VirtIONet *n = to_virtio_net(vdev);
  83     struct virtio_net_config netcfg;
  84
  85     memcpy(&netcfg, config, sizeof(netcfg));
  86
  87     if (memcmp(netcfg.mac, n->mac, ETH_ALEN)) {
  88         memcpy(n->mac, netcfg.mac, ETH_ALEN);
  89         qemu_format_nic_info_str(n->vc, n->mac);
  90     }
  91 }
  92
  93 static void virtio_net_set_link_status(VLANClientState *vc)
  94 {
  95     VirtIONet *n = vc->opaque;
  96     uint16_t old_status = n->status;
  97
  98     if (vc->link_down)
  99         n->status &= ~VIRTIO_NET_S_LINK_UP;
 100     else
 101         n->status |= VIRTIO_NET_S_LINK_UP;
 102
 103     if (n->status != old_status)
 104         virtio_notify_config(&n->vdev);
 105 }
 106
 107 static void virtio_net_reset(VirtIODevice *vdev)
 108 {
 109     VirtIONet *n = to_virtio_net(vdev);
 110
 111     /* Reset back to compatibility mode */
 112     n->promisc = 1;
 113     n->allmulti = 0;
 114     n->alluni = 0;
 115     n->nomulti = 0;
 116     n->nouni = 0;
 117     n->nobcast = 0;
 118
 119     /* Flush any MAC and VLAN filter table state */
 120     n->mac_table.in_use = 0;
 121     n->mac_table.first_multi = 0;
 122     n->mac_table.multi_overflow = 0;
 123     n->mac_table.uni_overflow = 0;
 124     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 125     memset(n->vlans, 0, MAX_VLAN >> 3);
 126 }
 127
 128 static uint32_t virtio_net_get_features(VirtIODevice *vdev)
 129 {
 130     uint32_t features = (1 << VIRTIO_NET_F_MAC) |
 131                         (1 << VIRTIO_NET_F_MRG_RXBUF) |
 132                         (1 << VIRTIO_NET_F_STATUS) |
 133                         (1 << VIRTIO_NET_F_CTRL_VQ) |
 134                         (1 << VIRTIO_NET_F_CTRL_RX) |
 135                         (1 << VIRTIO_NET_F_CTRL_VLAN) |
 136                         (1 << VIRTIO_NET_F_CTRL_RX_EXTRA);
 137
 138 #ifdef TAP_VNET_HDR
 139     VirtIONet *n = to_virtio_net(vdev);
 140     VLANClientState *host = n->vc->vlan->first_client;
 141
 142     if (tap_has_vnet_hdr(host)) {
 143         tap_using_vnet_hdr(host, 1);
 144         features |= (1 << VIRTIO_NET_F_CSUM);
 145         features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 146         features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 147         features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 148         features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 149         features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 150         features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 151         features |= (1 << VIRTIO_NET_F_HOST_ECN);
 152         features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
 153         /* Kernel can't actually handle UFO in software currently. */
 154     }
 155 #endif
 156
 157     return features;
 158 }
 159
 160 static uint32_t virtio_net_bad_features(VirtIODevice *vdev)
 161 {
 162     uint32_t features = 0;
 163
 164     /* Linux kernel 2.6.25.  It understood MAC (as everyone must),
 165      * but also these: */
 166     features |= (1 << VIRTIO_NET_F_MAC);
 167     features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 168     features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 169     features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 170     features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 171
 172     return features & virtio_net_get_features(vdev);
 173 }
 174
 175 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 176 {
 177     VirtIONet *n = to_virtio_net(vdev);
 178 #ifdef TAP_VNET_HDR
 179     VLANClientState *host = n->vc->vlan->first_client;
 180 #endif
 181
 182     n->mergeable_rx_bufs = !!(features & (1 << VIRTIO_NET_F_MRG_RXBUF));
 183
 184 #ifdef TAP_VNET_HDR
 185     if (!tap_has_vnet_hdr(host) || !host->set_offload)
 186         return;
 187
 188     host->set_offload(host,
 189                       (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 190                       (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 191                       (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 192                       (features >> VIRTIO_NET_F_GUEST_ECN)  & 1);
 193 #endif
 194 }
 195
 196 static int virtio_net_handle_rx_mode(VirtIONet *n, uint8_t cmd,
 197                                      VirtQueueElement *elem)
 198 {
 199     uint8_t on;
 200
 201     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(on)) {
 202         fprintf(stderr, "virtio-net ctrl invalid rx mode command\n");
 203         exit(1);
 204     }
 205
 206     on = ldub_p(elem->out_sg[1].iov_base);
 207
 208     if (cmd == VIRTIO_NET_CTRL_RX_MODE_PROMISC)
 209         n->promisc = on;
 210     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLMULTI)
 211         n->allmulti = on;
 212     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_ALLUNI)
 213         n->alluni = on;
 214     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOMULTI)
 215         n->nomulti = on;
 216     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOUNI)
 217         n->nouni = on;
 218     else if (cmd == VIRTIO_NET_CTRL_RX_MODE_NOBCAST)
 219         n->nobcast = on;
 220     else
 221         return VIRTIO_NET_ERR;
 222
 223     return VIRTIO_NET_OK;
 224 }
 225
 226 static int virtio_net_handle_mac(VirtIONet *n, uint8_t cmd,
 227                                  VirtQueueElement *elem)
 228 {
 229     struct virtio_net_ctrl_mac mac_data;
 230
 231     if (cmd != VIRTIO_NET_CTRL_MAC_TABLE_SET || elem->out_num != 3 ||
 232         elem->out_sg[1].iov_len < sizeof(mac_data) ||
 233         elem->out_sg[2].iov_len < sizeof(mac_data))
 234         return VIRTIO_NET_ERR;
 235
 236     n->mac_table.in_use = 0;
 237     n->mac_table.first_multi = 0;
 238     n->mac_table.uni_overflow = 0;
 239     n->mac_table.multi_overflow = 0;
 240     memset(n->mac_table.macs, 0, MAC_TABLE_ENTRIES * ETH_ALEN);
 241
 242     mac_data.entries = ldl_le_p(elem->out_sg[1].iov_base);
 243
 244     if (sizeof(mac_data.entries) +
 245         (mac_data.entries * ETH_ALEN) > elem->out_sg[1].iov_len)
 246         return VIRTIO_NET_ERR;
 247
 248     if (mac_data.entries <= MAC_TABLE_ENTRIES) {
 249         memcpy(n->mac_table.macs, elem->out_sg[1].iov_base + sizeof(mac_data),
 250                mac_data.entries * ETH_ALEN);
 251         n->mac_table.in_use += mac_data.entries;
 252     } else {
 253         n->mac_table.uni_overflow = 1;
 254     }
 255
 256     n->mac_table.first_multi = n->mac_table.in_use;
 257
 258     mac_data.entries = ldl_le_p(elem->out_sg[2].iov_base);
 259
 260     if (sizeof(mac_data.entries) +
 261         (mac_data.entries * ETH_ALEN) > elem->out_sg[2].iov_len)
 262         return VIRTIO_NET_ERR;
 263
 264     if (mac_data.entries) {
 265         if (n->mac_table.in_use + mac_data.entries <= MAC_TABLE_ENTRIES) {
 266             memcpy(n->mac_table.macs + (n->mac_table.in_use * ETH_ALEN),
 267                    elem->out_sg[2].iov_base + sizeof(mac_data),
 268                    mac_data.entries * ETH_ALEN);
 269             n->mac_table.in_use += mac_data.entries;
 270         } else {
 271             n->mac_table.multi_overflow = 1;
 272         }
 273     }
 274
 275     return VIRTIO_NET_OK;
 276 }
 277
 278 static int virtio_net_handle_vlan_table(VirtIONet *n, uint8_t cmd,
 279                                         VirtQueueElement *elem)
 280 {
 281     uint16_t vid;
 282
 283     if (elem->out_num != 2 || elem->out_sg[1].iov_len != sizeof(vid)) {
 284         fprintf(stderr, "virtio-net ctrl invalid vlan command\n");
 285         return VIRTIO_NET_ERR;
 286     }
 287
 288     vid = lduw_le_p(elem->out_sg[1].iov_base);
 289
 290     if (vid >= MAX_VLAN)
 291         return VIRTIO_NET_ERR;
 292
 293     if (cmd == VIRTIO_NET_CTRL_VLAN_ADD)
 294         n->vlans[vid >> 5] |= (1U << (vid & 0x1f));
 295     else if (cmd == VIRTIO_NET_CTRL_VLAN_DEL)
 296         n->vlans[vid >> 5] &= ~(1U << (vid & 0x1f));
 297     else
 298         return VIRTIO_NET_ERR;
 299
 300     return VIRTIO_NET_OK;
 301 }
 302
 303 static void virtio_net_handle_ctrl(VirtIODevice *vdev, VirtQueue *vq)
 304 {
 305     VirtIONet *n = to_virtio_net(vdev);
 306     struct virtio_net_ctrl_hdr ctrl;
 307     virtio_net_ctrl_ack status = VIRTIO_NET_ERR;
 308     VirtQueueElement elem;
 309
 310     while (virtqueue_pop(vq, &elem)) {
 311         if ((elem.in_num < 1) || (elem.out_num < 1)) {
 312             fprintf(stderr, "virtio-net ctrl missing headers\n");
 313             exit(1);
 314         }
 315
 316         if (elem.out_sg[0].iov_len < sizeof(ctrl) ||
 317             elem.in_sg[elem.in_num - 1].iov_len < sizeof(status)) {
 318             fprintf(stderr, "virtio-net ctrl header not in correct element\n");
 319             exit(1);
 320         }
 321
 322         ctrl.class = ldub_p(elem.out_sg[0].iov_base);
 323         ctrl.cmd = ldub_p(elem.out_sg[0].iov_base + sizeof(ctrl.class));
 324
 325         if (ctrl.class == VIRTIO_NET_CTRL_RX_MODE)
 326             status = virtio_net_handle_rx_mode(n, ctrl.cmd, &elem);
 327         else if (ctrl.class == VIRTIO_NET_CTRL_MAC)
 328             status = virtio_net_handle_mac(n, ctrl.cmd, &elem);
 329         else if (ctrl.class == VIRTIO_NET_CTRL_VLAN)
 330             status = virtio_net_handle_vlan_table(n, ctrl.cmd, &elem);
 331
 332         stb_p(elem.in_sg[elem.in_num - 1].iov_base, status);
 333
 334         virtqueue_push(vq, &elem, sizeof(status));
 335         virtio_notify(vdev, vq);
 336     }
 337 }
 338
 339 /* RX */
 340
 341 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 342 {
 343     VirtIONet *n = to_virtio_net(vdev);
 344
 345     qemu_flush_queued_packets(n->vc);
 346
 347     /* We now have RX buffers, signal to the IO thread to break out of the
 348      * select to re-poll the tap file descriptor */
 349     qemu_notify_event();
 350 }
 351
 352 static int do_virtio_net_can_receive(VirtIONet *n, int bufsize)
 353 {
 354     if (!virtio_queue_ready(n->rx_vq) ||
 355         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 356         return 0;
 357
 358     if (virtio_queue_empty(n->rx_vq) ||
 359         (n->mergeable_rx_bufs &&
 360          !virtqueue_avail_bytes(n->rx_vq, bufsize, 0))) {
 361         virtio_queue_set_notification(n->rx_vq, 1);
 362         return 0;
 363     }
 364
 365     virtio_queue_set_notification(n->rx_vq, 0);
 366     return 1;
 367 }
 368
 369 static int virtio_net_can_receive(VLANClientState *vc)
 370 {
 371     VirtIONet *n = vc->opaque;
 372
 373     return do_virtio_net_can_receive(n, VIRTIO_NET_MAX_BUFSIZE);
 374 }
 375
 376 #ifdef TAP_VNET_HDR
 377 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 378  * it never finds out that the packets don't have valid checksums.  This
 379  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 380  * fix this with Xen but it hasn't appeared in an upstream release of
 381  * dhclient yet.
 382  *
 383  * To avoid breaking existing guests, we catch udp packets and add
 384  * checksums.  This is terrible but it's better than hacking the guest
 385  * kernels.
 386  *
 387  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 388  * we should provide a mechanism to disable it to avoid polluting the host
 389  * cache.
 390  */
 391 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 392                                         const uint8_t *buf, size_t size)
 393 {
 394     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 395         (size > 27 && size < 1500) && /* normal sized MTU */
 396         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 397         (buf[23] == 17) && /* ip.protocol == UDP */
 398         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 399         /* FIXME this cast is evil */
 400         net_checksum_calculate((uint8_t *)buf, size);
 401         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 402     }
 403 }
 404 #endif
 405
 406 static int iov_fill(struct iovec *iov, int iovcnt, const void *buf, int count)
 407 {
 408     int offset, i;
 409
 410     offset = i = 0;
 411     while (offset < count && i < iovcnt) {
 412         int len = MIN(iov[i].iov_len, count - offset);
 413         memcpy(iov[i].iov_base, buf + offset, len);
 414         offset += len;
 415         i++;
 416     }
 417
 418     return offset;
 419 }
 420
 421 static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
 422                           const void *buf, size_t size, size_t hdr_len, int raw)
 423 {
 424     struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
 425     int offset = 0;
 426
 427     hdr->flags = 0;
 428     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 429
 430 #ifdef TAP_VNET_HDR
 431     if (tap_has_vnet_hdr(n->vc->vlan->first_client)) {
 432         if (!raw) {
 433             memcpy(hdr, buf, sizeof(*hdr));
 434         } else {
 435             memset(hdr, 0, sizeof(*hdr));
 436         }
 437         offset = sizeof(*hdr);
 438         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 439     }
 440 #endif
 441
 442     /* We only ever receive a struct virtio_net_hdr from the tapfd,
 443      * but we may be passing along a larger header to the guest.
 444      */
 445     iov[0].iov_base += hdr_len;
 446     iov[0].iov_len  -= hdr_len;
 447
 448     return offset;
 449 }
 450
 451 static int receive_filter(VirtIONet *n, const uint8_t *buf, int size)
 452 {
 453     static const uint8_t bcast[] = {0xff, 0xff, 0xff, 0xff, 0xff, 0xff};
 454     static const uint8_t vlan[] = {0x81, 0x00};
 455     uint8_t *ptr = (uint8_t *)buf;
 456     int i;
 457
 458     if (n->promisc)
 459         return 1;
 460
 461 #ifdef TAP_VNET_HDR
 462     if (tap_has_vnet_hdr(n->vc->vlan->first_client))
 463         ptr += sizeof(struct virtio_net_hdr);
 464 #endif
 465
 466     if (!memcmp(&ptr[12], vlan, sizeof(vlan))) {
 467         int vid = be16_to_cpup((uint16_t *)(ptr + 14)) & 0xfff;
 468         if (!(n->vlans[vid >> 5] & (1U << (vid & 0x1f))))
 469             return 0;
 470     }
 471
 472     if (ptr[0] & 1) { // multicast
 473         if (!memcmp(ptr, bcast, sizeof(bcast))) {
 474             return !n->nobcast;
 475         } else if (n->nomulti) {
 476             return 0;
 477         } else if (n->allmulti || n->mac_table.multi_overflow) {
 478             return 1;
 479         }
 480
 481         for (i = n->mac_table.first_multi; i < n->mac_table.in_use; i++) {
 482             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 483                 return 1;
 484             }
 485         }
 486     } else { // unicast
 487         if (n->nouni) {
 488             return 0;
 489         } else if (n->alluni || n->mac_table.uni_overflow) {
 490             return 1;
 491         } else if (!memcmp(ptr, n->mac, ETH_ALEN)) {
 492             return 1;
 493         }
 494
 495         for (i = 0; i < n->mac_table.first_multi; i++) {
 496             if (!memcmp(ptr, &n->mac_table.macs[i * ETH_ALEN], ETH_ALEN)) {
 497                 return 1;
 498             }
 499         }
 500     }
 501
 502     return 0;
 503 }
 504
 505 static ssize_t virtio_net_receive2(VLANClientState *vc, const uint8_t *buf, size_t size, int raw)
 506 {
 507     VirtIONet *n = vc->opaque;
 508     struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
 509     size_t hdr_len, offset, i;
 510
 511     if (!do_virtio_net_can_receive(n, size))
 512         return 0;
 513
 514     if (!receive_filter(n, buf, size))
 515         return size;
 516
 517     /* hdr_len refers to the header we supply to the guest */
 518     hdr_len = n->mergeable_rx_bufs ?
 519         sizeof(struct virtio_net_hdr_mrg_rxbuf) : sizeof(struct virtio_net_hdr);
 520
 521     offset = i = 0;
 522
 523     while (offset < size) {
 524         VirtQueueElement elem;
 525         int len, total;
 526         struct iovec sg[VIRTQUEUE_MAX_SIZE];
 527
 528         len = total = 0;
 529
 530         if ((i != 0 && !n->mergeable_rx_bufs) ||
 531             virtqueue_pop(n->rx_vq, &elem) == 0) {
 532             if (i == 0)
 533                 return -1;
 534             fprintf(stderr, "virtio-net truncating packet\n");
 535             exit(1);
 536         }
 537
 538         if (elem.in_num < 1) {
 539             fprintf(stderr, "virtio-net receive queue contains no in buffers\n");
 540             exit(1);
 541         }
 542
 543         if (!n->mergeable_rx_bufs && elem.in_sg[0].iov_len != hdr_len) {
 544             fprintf(stderr, "virtio-net header not in first element\n");
 545             exit(1);
 546         }
 547
 548         memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
 549
 550         if (i == 0) {
 551             if (n->mergeable_rx_bufs)
 552                 mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
 553
 554             offset += receive_header(n, sg, elem.in_num,
 555                                      buf + offset, size - offset, hdr_len, raw);
 556             total += hdr_len;
 557         }
 558
 559         /* copy in packet.  ugh */
 560         len = iov_fill(sg, elem.in_num,
 561                        buf + offset, size - offset);
 562         total += len;
 563
 564         /* signal other side */
 565         virtqueue_fill(n->rx_vq, &elem, total, i++);
 566
 567         offset += len;
 568     }
 569
 570     if (mhdr)
 571         mhdr->num_buffers = i;
 572
 573     virtqueue_flush(n->rx_vq, i);
 574     virtio_notify(&n->vdev, n->rx_vq);
 575
 576     return size;
 577 }
 578
 579 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq);
 580
 581 static void virtio_net_tx_complete(VLANClientState *vc, ssize_t len)
 582 {
 583     VirtIONet *n = vc->opaque;
 584
 585     virtqueue_push(n->tx_vq, &n->async_tx.elem, n->async_tx.len);
 586     virtio_notify(&n->vdev, n->tx_vq);
 587
 588     n->async_tx.elem.out_num = n->async_tx.len = 0;
 589
 590     virtio_queue_set_notification(n->tx_vq, 1);
 591     virtio_net_flush_tx(n, n->tx_vq);
 592 }
 593
 594 static ssize_t virtio_net_receive(VLANClientState *vc, const uint8_t *buf, size_t size)
 595 {
 596     return virtio_net_receive2(vc, buf, size, 0);
 597 }
 598
 599 static ssize_t virtio_net_receive_raw(VLANClientState *vc, const uint8_t *buf, size_t size)
 600 {
 601     return virtio_net_receive2(vc, buf, size, 1);
 602 }
 603
 604 /* TX */
 605 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 606 {
 607     VirtQueueElement elem;
 608 #ifdef TAP_VNET_HDR
 609     int has_vnet_hdr = tap_has_vnet_hdr(n->vc->vlan->first_client);
 610 #else
 611     int has_vnet_hdr = 0;
 612 #endif
 613
 614     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 615         return;
 616
 617     if (n->async_tx.elem.out_num) {
 618         virtio_queue_set_notification(n->tx_vq, 0);
 619         return;
 620     }
 621
 622     while (virtqueue_pop(vq, &elem)) {
 623         ssize_t ret, len = 0;
 624         unsigned int out_num = elem.out_num;
 625         struct iovec *out_sg = &elem.out_sg[0];
 626         unsigned hdr_len;
 627
 628         /* hdr_len refers to the header received from the guest */
 629         hdr_len = n->mergeable_rx_bufs ?
 630             sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 631             sizeof(struct virtio_net_hdr);
 632
 633         if (out_num < 1 || out_sg->iov_len != hdr_len) {
 634             fprintf(stderr, "virtio-net header not in first element\n");
 635             exit(1);
 636         }
 637
 638         /* ignore the header if GSO is not supported */
 639         if (!has_vnet_hdr) {
 640             out_num--;
 641             out_sg++;
 642             len += hdr_len;
 643         } else if (n->mergeable_rx_bufs) {
 644             /* tapfd expects a struct virtio_net_hdr */
 645             hdr_len -= sizeof(struct virtio_net_hdr);
 646             out_sg->iov_len -= hdr_len;
 647             len += hdr_len;
 648         }
 649
 650         ret = qemu_sendv_packet_async(n->vc, out_sg, out_num,
 651                                       virtio_net_tx_complete);
 652         if (ret == 0) {
 653             virtio_queue_set_notification(n->tx_vq, 0);
 654             n->async_tx.elem = elem;
 655             n->async_tx.len  = len;
 656             return;
 657         }
 658
 659         len += ret;
 660
 661         virtqueue_push(vq, &elem, len);
 662         virtio_notify(&n->vdev, vq);
 663     }
 664 }
 665
 666 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 667 {
 668     VirtIONet *n = to_virtio_net(vdev);
 669
 670     if (n->tx_timer_active) {
 671         virtio_queue_set_notification(vq, 1);
 672         qemu_del_timer(n->tx_timer);
 673         n->tx_timer_active = 0;
 674         virtio_net_flush_tx(n, vq);
 675     } else {
 676         qemu_mod_timer(n->tx_timer,
 677                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 678         n->tx_timer_active = 1;
 679         virtio_queue_set_notification(vq, 0);
 680     }
 681 }
 682
 683 static void virtio_net_tx_timer(void *opaque)
 684 {
 685     VirtIONet *n = opaque;
 686
 687     n->tx_timer_active = 0;
 688
 689     /* Just in case the driver is not ready on more */
 690     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 691         return;
 692
 693     virtio_queue_set_notification(n->tx_vq, 1);
 694     virtio_net_flush_tx(n, n->tx_vq);
 695 }
 696
 697 static void virtio_net_save(QEMUFile *f, void *opaque)
 698 {
 699     VirtIONet *n = opaque;
 700
 701     virtio_save(&n->vdev, f);
 702
 703     qemu_put_buffer(f, n->mac, ETH_ALEN);
 704     qemu_put_be32(f, n->tx_timer_active);
 705     qemu_put_be32(f, n->mergeable_rx_bufs);
 706     qemu_put_be16(f, n->status);
 707     qemu_put_byte(f, n->promisc);
 708     qemu_put_byte(f, n->allmulti);
 709     qemu_put_be32(f, n->mac_table.in_use);
 710     qemu_put_buffer(f, n->mac_table.macs, n->mac_table.in_use * ETH_ALEN);
 711     qemu_put_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 712
 713 #ifdef TAP_VNET_HDR
 714     qemu_put_be32(f, tap_has_vnet_hdr(n->vc->vlan->first_client));
 715 #else
 716     qemu_put_be32(f, 0);
 717 #endif
 718
 719     qemu_put_byte(f, n->mac_table.multi_overflow);
 720     qemu_put_byte(f, n->mac_table.uni_overflow);
 721     qemu_put_byte(f, n->alluni);
 722     qemu_put_byte(f, n->nomulti);
 723     qemu_put_byte(f, n->nouni);
 724     qemu_put_byte(f, n->nobcast);
 725 }
 726
 727 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 728 {
 729     VirtIONet *n = opaque;
 730     int i;
 731
 732     if (version_id < 2 || version_id > VIRTIO_NET_VM_VERSION)
 733         return -EINVAL;
 734
 735     virtio_load(&n->vdev, f);
 736
 737     qemu_get_buffer(f, n->mac, ETH_ALEN);
 738     n->tx_timer_active = qemu_get_be32(f);
 739     n->mergeable_rx_bufs = qemu_get_be32(f);
 740
 741     if (version_id >= 3)
 742         n->status = qemu_get_be16(f);
 743
 744     if (version_id >= 4) {
 745         if (version_id < 8) {
 746             n->promisc = qemu_get_be32(f);
 747             n->allmulti = qemu_get_be32(f);
 748         } else {
 749             n->promisc = qemu_get_byte(f);
 750             n->allmulti = qemu_get_byte(f);
 751         }
 752     }
 753
 754     if (version_id >= 5) {
 755         n->mac_table.in_use = qemu_get_be32(f);
 756         /* MAC_TABLE_ENTRIES may be different from the saved image */
 757         if (n->mac_table.in_use <= MAC_TABLE_ENTRIES) {
 758             qemu_get_buffer(f, n->mac_table.macs,
 759                             n->mac_table.in_use * ETH_ALEN);
 760         } else if (n->mac_table.in_use) {
 761             qemu_fseek(f, n->mac_table.in_use * ETH_ALEN, SEEK_CUR);
 762             n->mac_table.multi_overflow = n->mac_table.uni_overflow = 1;
 763             n->mac_table.in_use = 0;
 764         }
 765     }
 766
 767     if (version_id >= 6)
 768         qemu_get_buffer(f, (uint8_t *)n->vlans, MAX_VLAN >> 3);
 769
 770     if (version_id >= 7 && qemu_get_be32(f)) {
 771 #ifdef TAP_VNET_HDR
 772         tap_using_vnet_hdr(n->vc->vlan->first_client, 1);
 773 #else
 774         fprintf(stderr,
 775                 "virtio-net: saved image requires vnet header support\n");
 776         exit(1);
 777 #endif
 778     }
 779
 780     if (version_id >= 9) {
 781         n->mac_table.multi_overflow = qemu_get_byte(f);
 782         n->mac_table.uni_overflow = qemu_get_byte(f);
 783     }
 784
 785     if (version_id >= 10) {
 786         n->alluni = qemu_get_byte(f);
 787         n->nomulti = qemu_get_byte(f);
 788         n->nouni = qemu_get_byte(f);
 789         n->nobcast = qemu_get_byte(f);
 790     }
 791
 792     /* Find the first multicast entry in the saved MAC filter */
 793     for (i = 0; i < n->mac_table.in_use; i++) {
 794         if (n->mac_table.macs[i * ETH_ALEN] & 1) {
 795             break;
 796         }
 797     }
 798     n->mac_table.first_multi = i;
 799
 800     if (n->tx_timer_active) {
 801         qemu_mod_timer(n->tx_timer,
 802                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 803     }
 804
 805     return 0;
 806 }
 807
 808 static void virtio_net_cleanup(VLANClientState *vc)
 809 {
 810     VirtIONet *n = vc->opaque;
 811
 812     qemu_purge_queued_packets(vc);
 813
 814     unregister_savevm("virtio-net", n);
 815
 816     qemu_free(n->mac_table.macs);
 817     qemu_free(n->vlans);
 818
 819     qemu_del_timer(n->tx_timer);
 820     qemu_free_timer(n->tx_timer);
 821
 822     virtio_cleanup(&n->vdev);
 823 }
 824
 825 VirtIODevice *virtio_net_init(DeviceState *dev)
 826 {
 827     VirtIONet *n;
 828     static int virtio_net_id;
 829
 830     n = (VirtIONet *)virtio_common_init("virtio-net", VIRTIO_ID_NET,
 831                                         sizeof(struct virtio_net_config),
 832                                         sizeof(VirtIONet));
 833
 834     n->vdev.get_config = virtio_net_get_config;
 835     n->vdev.set_config = virtio_net_set_config;
 836     n->vdev.get_features = virtio_net_get_features;
 837     n->vdev.set_features = virtio_net_set_features;
 838     n->vdev.bad_features = virtio_net_bad_features;
 839     n->vdev.reset = virtio_net_reset;
 840     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 841     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 842     n->ctrl_vq = virtio_add_queue(&n->vdev, 64, virtio_net_handle_ctrl);
 843     qdev_get_macaddr(dev, n->mac);
 844     n->status = VIRTIO_NET_S_LINK_UP;
 845     n->vc = qdev_get_vlan_client(dev,
 846                                  virtio_net_can_receive,
 847                                  virtio_net_receive, NULL,
 848                                  virtio_net_cleanup, n);
 849     n->vc->link_status_changed = virtio_net_set_link_status;
 850     n->vc->receive_raw = virtio_net_receive_raw;
 851
 852     qemu_format_nic_info_str(n->vc, n->mac);
 853
 854     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 855     n->tx_timer_active = 0;
 856     n->mergeable_rx_bufs = 0;
 857     n->promisc = 1; /* for compatibility */
 858
 859     n->mac_table.macs = qemu_mallocz(MAC_TABLE_ENTRIES * ETH_ALEN);
 860
 861     n->vlans = qemu_mallocz(MAX_VLAN >> 3);
 862     if (dev->nd->nvectors == NIC_NVECTORS_UNSPECIFIED)
 863         n->vdev.nvectors = 3;
 864     else
 865         n->vdev.nvectors = dev->nd->nvectors;
 866
 867     register_savevm("virtio-net", virtio_net_id++, VIRTIO_NET_VM_VERSION,
 868                     virtio_net_save, virtio_net_load, n);
 869
 870     return &n->vdev;
 871 }