1 /* 2 * vhost-net support 3 * 4 * Copyright Red Hat, Inc. 2010 5 * 6 * Authors: 7 * Michael S. Tsirkin <mst@redhat.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 * Contributions after 2012-01-13 are licensed under the terms of the 13 * GNU GPL, version 2 or (at your option) any later version. 14 */ 15 16 #include "net/net.h" 17 #include "net/tap.h" 18 19 #include "hw/virtio/virtio-net.h" 20 #include "net/vhost_net.h" 21 #include "qemu/error-report.h" 22 23 #include "config.h" 24 25 #ifdef CONFIG_VHOST_NET 26 #include <linux/vhost.h> 27 #include <sys/socket.h> 28 #include <linux/kvm.h> 29 #include <fcntl.h> 30 #include <sys/ioctl.h> 31 #include <linux/virtio_ring.h> 32 #include <netpacket/packet.h> 33 #include <net/ethernet.h> 34 #include <net/if.h> 35 #include <netinet/in.h> 36 37 #include <stdio.h> 38 39 #include "hw/virtio/vhost.h" 40 #include "hw/virtio/virtio-bus.h" 41 42 struct vhost_net { 43 struct vhost_dev dev; 44 struct vhost_virtqueue vqs[2]; 45 int backend; 46 NetClientState *nc; 47 }; 48 49 /* Features supported by host kernel. */ 50 static const int kernel_feature_bits[] = { 51 VIRTIO_F_NOTIFY_ON_EMPTY, 52 VIRTIO_RING_F_INDIRECT_DESC, 53 VIRTIO_RING_F_EVENT_IDX, 54 VIRTIO_NET_F_MRG_RXBUF, 55 VHOST_INVALID_FEATURE_BIT 56 }; 57 58 static const int *vhost_net_get_feature_bits(struct vhost_net *net) 59 { 60 const int *feature_bits = 0; 61 62 switch (net->nc->info->type) { 63 case NET_CLIENT_OPTIONS_KIND_TAP: 64 feature_bits = kernel_feature_bits; 65 break; 66 default: 67 error_report("Feature bits not defined for this type: %d", 68 net->nc->info->type); 69 break; 70 } 71 72 return feature_bits; 73 } 74 75 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features) 76 { 77 return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net), 78 features); 79 } 80 81 void vhost_net_ack_features(struct vhost_net *net, unsigned features) 82 { 83 vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features); 84 } 85 86 static int vhost_net_get_fd(NetClientState *backend) 87 { 88 switch (backend->info->type) { 89 case NET_CLIENT_OPTIONS_KIND_TAP: 90 return tap_get_fd(backend); 91 default: 92 fprintf(stderr, "vhost-net requires tap backend\n"); 93 return -EBADFD; 94 } 95 } 96 97 struct vhost_net *vhost_net_init(NetClientState *backend, int devfd, 98 bool force) 99 { 100 int r; 101 struct vhost_net *net = g_malloc(sizeof *net); 102 if (!backend) { 103 fprintf(stderr, "vhost-net requires backend to be setup\n"); 104 goto fail; 105 } 106 r = vhost_net_get_fd(backend); 107 if (r < 0) { 108 goto fail; 109 } 110 net->nc = backend; 111 net->dev.backend_features = qemu_has_vnet_hdr(backend) ? 0 : 112 (1 << VHOST_NET_F_VIRTIO_NET_HDR); 113 net->backend = r; 114 115 net->dev.nvqs = 2; 116 net->dev.vqs = net->vqs; 117 118 r = vhost_dev_init(&net->dev, devfd, "/dev/vhost-net", force); 119 if (r < 0) { 120 goto fail; 121 } 122 if (!qemu_has_vnet_hdr_len(backend, 123 sizeof(struct virtio_net_hdr_mrg_rxbuf))) { 124 net->dev.features &= ~(1 << VIRTIO_NET_F_MRG_RXBUF); 125 } 126 if (~net->dev.features & net->dev.backend_features) { 127 fprintf(stderr, "vhost lacks feature mask %" PRIu64 " for backend\n", 128 (uint64_t)(~net->dev.features & net->dev.backend_features)); 129 vhost_dev_cleanup(&net->dev); 130 goto fail; 131 } 132 133 /* Set sane init value. Override when guest acks. */ 134 vhost_net_ack_features(net, 0); 135 return net; 136 fail: 137 g_free(net); 138 return NULL; 139 } 140 141 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev) 142 { 143 return vhost_dev_query(&net->dev, dev); 144 } 145 146 static int vhost_net_start_one(struct vhost_net *net, 147 VirtIODevice *dev, 148 int vq_index) 149 { 150 struct vhost_vring_file file = { }; 151 int r; 152 153 if (net->dev.started) { 154 return 0; 155 } 156 157 net->dev.nvqs = 2; 158 net->dev.vqs = net->vqs; 159 net->dev.vq_index = vq_index; 160 161 r = vhost_dev_enable_notifiers(&net->dev, dev); 162 if (r < 0) { 163 goto fail_notifiers; 164 } 165 166 r = vhost_dev_start(&net->dev, dev); 167 if (r < 0) { 168 goto fail_start; 169 } 170 171 if (net->nc->info->poll) { 172 net->nc->info->poll(net->nc, false); 173 } 174 175 qemu_set_fd_handler(net->backend, NULL, NULL, NULL); 176 file.fd = net->backend; 177 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) { 178 r = ioctl(net->dev.control, VHOST_NET_SET_BACKEND, &file); 179 if (r < 0) { 180 r = -errno; 181 goto fail; 182 } 183 } 184 return 0; 185 fail: 186 file.fd = -1; 187 while (file.index-- > 0) { 188 int r = ioctl(net->dev.control, VHOST_NET_SET_BACKEND, &file); 189 assert(r >= 0); 190 } 191 if (net->nc->info->poll) { 192 net->nc->info->poll(net->nc, true); 193 } 194 vhost_dev_stop(&net->dev, dev); 195 fail_start: 196 vhost_dev_disable_notifiers(&net->dev, dev); 197 fail_notifiers: 198 return r; 199 } 200 201 static void vhost_net_stop_one(struct vhost_net *net, 202 VirtIODevice *dev) 203 { 204 struct vhost_vring_file file = { .fd = -1 }; 205 206 if (!net->dev.started) { 207 return; 208 } 209 210 for (file.index = 0; file.index < net->dev.nvqs; ++file.index) { 211 int r = ioctl(net->dev.control, VHOST_NET_SET_BACKEND, &file); 212 assert(r >= 0); 213 } 214 if (net->nc->info->poll) { 215 net->nc->info->poll(net->nc, true); 216 } 217 vhost_dev_stop(&net->dev, dev); 218 vhost_dev_disable_notifiers(&net->dev, dev); 219 } 220 221 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs, 222 int total_queues) 223 { 224 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev))); 225 VirtioBusState *vbus = VIRTIO_BUS(qbus); 226 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus); 227 int r, i = 0; 228 229 if (!k->set_guest_notifiers) { 230 error_report("binding does not support guest notifiers"); 231 r = -ENOSYS; 232 goto err; 233 } 234 235 for (i = 0; i < total_queues; i++) { 236 r = vhost_net_start_one(tap_get_vhost_net(ncs[i].peer), dev, i * 2); 237 238 if (r < 0) { 239 goto err; 240 } 241 } 242 243 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true); 244 if (r < 0) { 245 error_report("Error binding guest notifier: %d", -r); 246 goto err; 247 } 248 249 return 0; 250 251 err: 252 while (--i >= 0) { 253 vhost_net_stop_one(tap_get_vhost_net(ncs[i].peer), dev); 254 } 255 return r; 256 } 257 258 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs, 259 int total_queues) 260 { 261 BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev))); 262 VirtioBusState *vbus = VIRTIO_BUS(qbus); 263 VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus); 264 int i, r; 265 266 r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false); 267 if (r < 0) { 268 fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r); 269 fflush(stderr); 270 } 271 assert(r >= 0); 272 273 for (i = 0; i < total_queues; i++) { 274 vhost_net_stop_one(tap_get_vhost_net(ncs[i].peer), dev); 275 } 276 } 277 278 void vhost_net_cleanup(struct vhost_net *net) 279 { 280 vhost_dev_cleanup(&net->dev); 281 g_free(net); 282 } 283 284 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx) 285 { 286 return vhost_virtqueue_pending(&net->dev, idx); 287 } 288 289 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev, 290 int idx, bool mask) 291 { 292 vhost_virtqueue_mask(&net->dev, dev, idx, mask); 293 } 294 #else 295 struct vhost_net *vhost_net_init(NetClientState *backend, int devfd, 296 bool force) 297 { 298 error_report("vhost-net support is not compiled in"); 299 return NULL; 300 } 301 302 bool vhost_net_query(VHostNetState *net, VirtIODevice *dev) 303 { 304 return false; 305 } 306 307 int vhost_net_start(VirtIODevice *dev, 308 NetClientState *ncs, 309 int total_queues) 310 { 311 return -ENOSYS; 312 } 313 void vhost_net_stop(VirtIODevice *dev, 314 NetClientState *ncs, 315 int total_queues) 316 { 317 } 318 319 void vhost_net_cleanup(struct vhost_net *net) 320 { 321 } 322 323 unsigned vhost_net_get_features(struct vhost_net *net, unsigned features) 324 { 325 return features; 326 } 327 void vhost_net_ack_features(struct vhost_net *net, unsigned features) 328 { 329 } 330 331 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx) 332 { 333 return false; 334 } 335 336 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev, 337 int idx, bool mask) 338 { 339 } 340 #endif 341