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 "qemu-kvm.h"
  18
  19 /* from Linux's virtio_net.h */
  20
  21 /* The ID for virtio_net */
  22 #define VIRTIO_ID_NET   1
  23
  24 /* The feature bitmap for virtio net */
  25 #define VIRTIO_NET_F_CSUM       0       /* Host handles pkts w/ partial csum */
  26 #define VIRTIO_NET_F_GUEST_CSUM 1       /* Guest handles pkts w/ partial csum */
  27 #define VIRTIO_NET_F_MAC        5       /* Host has given MAC address. */
  28 #define VIRTIO_NET_F_GSO        6       /* Host handles pkts w/ any GSO type */
  29 #define VIRTIO_NET_F_GUEST_TSO4 7       /* Guest can handle TSOv4 in. */
  30 #define VIRTIO_NET_F_GUEST_TSO6 8       /* Guest can handle TSOv6 in. */
  31 #define VIRTIO_NET_F_GUEST_ECN  9       /* Guest can handle TSO[6] w/ ECN in. */
  32 #define VIRTIO_NET_F_GUEST_UFO  10      /* Guest can handle UFO in. */
  33 #define VIRTIO_NET_F_HOST_TSO4  11      /* Host can handle TSOv4 in. */
  34 #define VIRTIO_NET_F_HOST_TSO6  12      /* Host can handle TSOv6 in. */
  35 #define VIRTIO_NET_F_HOST_ECN   13      /* Host can handle TSO[6] w/ ECN in. */
  36 #define VIRTIO_NET_F_HOST_UFO   14      /* Host can handle UFO in. */
  37
  38 #define TX_TIMER_INTERVAL 150000 /* 150 us */
  39
  40 /* The config defining mac address (6 bytes) */
  41 struct virtio_net_config
  42 {
  43     uint8_t mac[6];
  44 } __attribute__((packed));
  45
  46 /* This is the first element of the scatter-gather list.  If you don't
  47  * specify GSO or CSUM features, you can simply ignore the header. */
  48 struct virtio_net_hdr
  49 {
  50 #define VIRTIO_NET_HDR_F_NEEDS_CSUM     1       // Use csum_start, csum_offset
  51     uint8_t flags;
  52 #define VIRTIO_NET_HDR_GSO_NONE         0       // Not a GSO frame
  53 #define VIRTIO_NET_HDR_GSO_TCPV4        1       // GSO frame, IPv4 TCP (TSO)
  54 #define VIRTIO_NET_HDR_GSO_UDP          3       // GSO frame, IPv4 UDP (UFO)
  55 #define VIRTIO_NET_HDR_GSO_TCPV6        4       // GSO frame, IPv6 TCP
  56 #define VIRTIO_NET_HDR_GSO_ECN          0x80    // TCP has ECN set
  57     uint8_t gso_type;
  58     uint16_t hdr_len;
  59     uint16_t gso_size;
  60     uint16_t csum_start;
  61     uint16_t csum_offset;
  62 };
  63
  64 typedef struct VirtIONet
  65 {
  66     VirtIODevice vdev;
  67     uint8_t mac[6];
  68     VirtQueue *rx_vq;
  69     VirtQueue *tx_vq;
  70     VLANClientState *vc;
  71     QEMUTimer *tx_timer;
  72     int tx_timer_active;
  73 } VirtIONet;
  74
  75 /* TODO
  76  * - we could suppress RX interrupt if we were so inclined.
  77  */
  78
  79 static VirtIONet *to_virtio_net(VirtIODevice *vdev)
  80 {
  81     return (VirtIONet *)vdev;
  82 }
  83
  84 static void virtio_net_update_config(VirtIODevice *vdev, uint8_t *config)
  85 {
  86     VirtIONet *n = to_virtio_net(vdev);
  87     struct virtio_net_config netcfg;
  88
  89     memcpy(netcfg.mac, n->mac, 6);
  90     memcpy(config, &netcfg, sizeof(netcfg));
  91 }
  92
  93 static uint32_t virtio_net_get_features(VirtIODevice *vdev)
  94 {
  95     VirtIONet *n = to_virtio_net(vdev);
  96     VLANClientState *host = n->vc->vlan->first_client;
  97     uint32_t features = (1 << VIRTIO_NET_F_MAC);
  98
  99     if (tap_has_vnet_hdr(host)) {
 100         tap_using_vnet_hdr(host, 1);
 101         features |= (1 << VIRTIO_NET_F_CSUM);
 102         features |= (1 << VIRTIO_NET_F_GUEST_CSUM);
 103         features |= (1 << VIRTIO_NET_F_GUEST_TSO4);
 104         features |= (1 << VIRTIO_NET_F_GUEST_TSO6);
 105         features |= (1 << VIRTIO_NET_F_GUEST_ECN);
 106         features |= (1 << VIRTIO_NET_F_HOST_TSO4);
 107         features |= (1 << VIRTIO_NET_F_HOST_TSO6);
 108         features |= (1 << VIRTIO_NET_F_HOST_ECN);
 109         /* Kernel can't actually handle UFO in software currently. */
 110     }
 111
 112     return features;
 113 }
 114
 115 static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
 116 {
 117     VirtIONet *n = to_virtio_net(vdev);
 118     VLANClientState *host = n->vc->vlan->first_client;
 119
 120     if (!tap_has_vnet_hdr(host) || !host->set_offload)
 121         return;
 122
 123     host->set_offload(host,
 124                       (features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
 125                       (features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
 126                       (features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
 127                       (features >> VIRTIO_NET_F_GUEST_ECN)  & 1);
 128 }
 129
 130 /* RX */
 131
 132 static void virtio_net_handle_rx(VirtIODevice *vdev, VirtQueue *vq)
 133 {
 134     /* We now have RX buffers, signal to the IO thread to break out of the
 135        select to re-poll the tap file descriptor */
 136     if (kvm_enabled())
 137         qemu_kvm_notify_work();
 138 }
 139
 140 static int virtio_net_can_receive(void *opaque)
 141 {
 142     VirtIONet *n = opaque;
 143
 144     if (n->rx_vq->vring.avail == NULL ||
 145         !(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 146         return 0;
 147
 148     if (n->rx_vq->vring.avail->idx == n->rx_vq->last_avail_idx) {
 149         n->rx_vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
 150         return 0;
 151     }
 152
 153     n->rx_vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
 154     return 1;
 155 }
 156
 157 /* dhclient uses AF_PACKET but doesn't pass auxdata to the kernel so
 158  * it never finds out that the packets don't have valid checksums.  This
 159  * causes dhclient to get upset.  Fedora's carried a patch for ages to
 160  * fix this with Xen but it hasn't appeared in an upstream release of
 161  * dhclient yet.
 162  *
 163  * To avoid breaking existing guests, we catch udp packets and add
 164  * checksums.  This is terrible but it's better than hacking the guest
 165  * kernels.
 166  *
 167  * N.B. if we introduce a zero-copy API, this operation is no longer free so
 168  * we should provide a mechanism to disable it to avoid polluting the host
 169  * cache.
 170  */
 171 static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
 172                                         const uint8_t *buf, size_t size)
 173 {
 174     if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
 175         (size > 27 && size < 1500) && /* normal sized MTU */
 176         (buf[12] == 0x08 && buf[13] == 0x00) && /* ethertype == IPv4 */
 177         (buf[23] == 17) && /* ip.protocol == UDP */
 178         (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
 179         /* FIXME this cast is evil */
 180         net_checksum_calculate((uint8_t *)buf, size);
 181         hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
 182     }
 183 }
 184
 185 static void virtio_net_receive(void *opaque, const uint8_t *buf, int size)
 186 {
 187     VirtIONet *n = opaque;
 188     VirtQueueElement elem;
 189     struct virtio_net_hdr *hdr;
 190     int offset, i;
 191     int total;
 192
 193     if (virtqueue_pop(n->rx_vq, &elem) == 0)
 194         return;
 195
 196     if (elem.in_num < 1 || elem.in_sg[0].iov_len != sizeof(*hdr)) {
 197         fprintf(stderr, "virtio-net header not in first element\n");
 198         exit(1);
 199     }
 200
 201     hdr = (void *)elem.in_sg[0].iov_base;
 202     hdr->flags = 0;
 203     hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
 204
 205     offset = 0;
 206     total = sizeof(*hdr);
 207
 208     if (tap_has_vnet_hdr(n->vc->vlan->first_client)) {
 209         memcpy(hdr, buf, sizeof(*hdr));
 210         offset += total;
 211         work_around_broken_dhclient(hdr, buf + offset, size - offset);
 212     }
 213
 214     /* copy in packet.  ugh */
 215     i = 1;
 216     while (offset < size && i < elem.in_num) {
 217         int len = MIN(elem.in_sg[i].iov_len, size - offset);
 218         memcpy(elem.in_sg[i].iov_base, buf + offset, len);
 219         offset += len;
 220         total += len;
 221         i++;
 222     }
 223
 224     /* signal other side */
 225     virtqueue_push(n->rx_vq, &elem, total);
 226     virtio_notify(&n->vdev, n->rx_vq);
 227 }
 228
 229 /* TX */
 230 static void virtio_net_flush_tx(VirtIONet *n, VirtQueue *vq)
 231 {
 232     VirtQueueElement elem;
 233     int has_vnet_hdr = tap_has_vnet_hdr(n->vc->vlan->first_client);
 234
 235     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 236         return;
 237
 238     while (virtqueue_pop(vq, &elem)) {
 239         ssize_t len = 0;
 240         unsigned int out_num = elem.out_num;
 241         struct iovec *out_sg = &elem.out_sg[0];
 242
 243         if (out_num < 1 || out_sg->iov_len != sizeof(struct virtio_net_hdr)) {
 244             fprintf(stderr, "virtio-net header not in first element\n");
 245             exit(1);
 246         }
 247
 248         /* ignore the header if GSO is not supported */
 249         if (!has_vnet_hdr) {
 250             out_num--;
 251             out_sg++;
 252             len += sizeof(struct virtio_net_hdr);
 253         }
 254
 255         len += qemu_sendv_packet(n->vc, out_sg, out_num);
 256
 257         virtqueue_push(vq, &elem, len);
 258         virtio_notify(&n->vdev, vq);
 259     }
 260 }
 261
 262 static void virtio_net_handle_tx(VirtIODevice *vdev, VirtQueue *vq)
 263 {
 264     VirtIONet *n = to_virtio_net(vdev);
 265
 266     if (n->tx_timer_active) {
 267         vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
 268         qemu_del_timer(n->tx_timer);
 269         n->tx_timer_active = 0;
 270         virtio_net_flush_tx(n, vq);
 271     } else {
 272         qemu_mod_timer(n->tx_timer,
 273                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 274         n->tx_timer_active = 1;
 275         vq->vring.used->flags |= VRING_USED_F_NO_NOTIFY;
 276     }
 277 }
 278
 279 static void virtio_net_tx_timer(void *opaque)
 280 {
 281     VirtIONet *n = opaque;
 282
 283     n->tx_timer_active = 0;
 284
 285     /* Just in case the driver is not ready on more */
 286     if (!(n->vdev.status & VIRTIO_CONFIG_S_DRIVER_OK))
 287         return;
 288
 289     n->tx_vq->vring.used->flags &= ~VRING_USED_F_NO_NOTIFY;
 290     virtio_net_flush_tx(n, n->tx_vq);
 291 }
 292
 293 static void virtio_net_save(QEMUFile *f, void *opaque)
 294 {
 295     VirtIONet *n = opaque;
 296
 297     virtio_save(&n->vdev, f);
 298
 299     qemu_put_buffer(f, n->mac, 6);
 300     qemu_put_be32(f, n->tx_timer_active);
 301 }
 302
 303 static int virtio_net_load(QEMUFile *f, void *opaque, int version_id)
 304 {
 305     VirtIONet *n = opaque;
 306
 307     if (version_id != 1)
 308         return -EINVAL;
 309
 310     virtio_load(&n->vdev, f);
 311
 312     qemu_get_buffer(f, n->mac, 6);
 313     n->tx_timer_active = qemu_get_be32(f);
 314
 315     if (n->tx_timer_active) {
 316         qemu_mod_timer(n->tx_timer,
 317                        qemu_get_clock(vm_clock) + TX_TIMER_INTERVAL);
 318     }
 319
 320     return 0;
 321 }
 322
 323 PCIDevice *virtio_net_init(PCIBus *bus, NICInfo *nd, int devfn)
 324 {
 325     VirtIONet *n;
 326     static int virtio_net_id;
 327
 328     n = (VirtIONet *)virtio_init_pci(bus, "virtio-net", 6900, 0x1000,
 329                                      0, VIRTIO_ID_NET,
 330                                      0x02, 0x00, 0x00,
 331                                      6, sizeof(VirtIONet));
 332     if (!n)
 333         return NULL;
 334
 335     n->vdev.get_config = virtio_net_update_config;
 336     n->vdev.get_features = virtio_net_get_features;
 337     n->vdev.set_features = virtio_net_set_features;
 338     n->rx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_rx);
 339     n->tx_vq = virtio_add_queue(&n->vdev, 256, virtio_net_handle_tx);
 340     memcpy(n->mac, nd->macaddr, 6);
 341     n->vc = qemu_new_vlan_client(nd->vlan, virtio_net_receive,
 342                                  virtio_net_can_receive, n);
 343
 344     n->tx_timer = qemu_new_timer(vm_clock, virtio_net_tx_timer, n);
 345     n->tx_timer_active = 0;
 346
 347     register_savevm("virtio-net", virtio_net_id++, 1,
 348                     virtio_net_save, virtio_net_load, n);
 349
 350     return (PCIDevice *)n;
 351 }