1 #include "kvm/virtio-pci-dev.h" 2 #include "kvm/virtio-net.h" 3 #include "kvm/virtio.h" 4 #include "kvm/mutex.h" 5 #include "kvm/util.h" 6 #include "kvm/kvm.h" 7 #include "kvm/irq.h" 8 #include "kvm/uip.h" 9 #include "kvm/guest_compat.h" 10 #include "kvm/iovec.h" 11 12 #include <linux/vhost.h> 13 #include <linux/virtio_net.h> 14 #include <linux/if_tun.h> 15 #include <linux/types.h> 16 17 #include <arpa/inet.h> 18 #include <net/if.h> 19 20 #include <unistd.h> 21 #include <fcntl.h> 22 23 #include <sys/socket.h> 24 #include <sys/ioctl.h> 25 #include <sys/types.h> 26 #include <sys/wait.h> 27 #include <sys/eventfd.h> 28 29 #define VIRTIO_NET_QUEUE_SIZE 256 30 #define VIRTIO_NET_NUM_QUEUES 8 31 32 struct net_dev; 33 34 struct net_dev_operations { 35 int (*rx)(struct iovec *iov, u16 in, struct net_dev *ndev); 36 int (*tx)(struct iovec *iov, u16 in, struct net_dev *ndev); 37 }; 38 39 struct net_dev { 40 struct mutex mutex; 41 struct virtio_device vdev; 42 struct list_head list; 43 44 struct virt_queue vqs[VIRTIO_NET_NUM_QUEUES * 2 + 1]; 45 struct virtio_net_config config; 46 u32 features, rx_vqs, tx_vqs, queue_pairs; 47 48 pthread_t io_thread[VIRTIO_NET_NUM_QUEUES * 2 + 1]; 49 struct mutex io_lock[VIRTIO_NET_NUM_QUEUES * 2 + 1]; 50 pthread_cond_t io_cond[VIRTIO_NET_NUM_QUEUES * 2 + 1]; 51 52 int vhost_fd; 53 int tap_fd; 54 char tap_name[IFNAMSIZ]; 55 56 int mode; 57 58 struct uip_info info; 59 struct net_dev_operations *ops; 60 struct kvm *kvm; 61 62 struct virtio_net_params *params; 63 }; 64 65 static LIST_HEAD(ndevs); 66 static int compat_id = -1; 67 68 #define MAX_PACKET_SIZE 65550 69 70 static bool has_virtio_feature(struct net_dev *ndev, u32 feature) 71 { 72 return ndev->features & (1 << feature); 73 } 74 75 static void virtio_net_fix_tx_hdr(struct virtio_net_hdr *hdr, struct net_dev *ndev) 76 { 77 hdr->hdr_len = virtio_guest_to_host_u16(&ndev->vdev, hdr->hdr_len); 78 hdr->gso_size = virtio_guest_to_host_u16(&ndev->vdev, hdr->gso_size); 79 hdr->csum_start = virtio_guest_to_host_u16(&ndev->vdev, hdr->csum_start); 80 hdr->csum_offset = virtio_guest_to_host_u16(&ndev->vdev, hdr->csum_offset); 81 } 82 83 static void virtio_net_fix_rx_hdr(struct virtio_net_hdr_mrg_rxbuf *hdr, struct net_dev *ndev) 84 { 85 hdr->hdr.hdr_len = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr.hdr_len); 86 hdr->hdr.gso_size = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr.gso_size); 87 hdr->hdr.csum_start = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr.csum_start); 88 hdr->hdr.csum_offset = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr.csum_offset); 89 if (has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 90 hdr->num_buffers = virtio_host_to_guest_u16(&ndev->vdev, hdr->num_buffers); 91 } 92 93 static void *virtio_net_rx_thread(void *p) 94 { 95 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 96 struct virt_queue *vq; 97 struct kvm *kvm; 98 struct net_dev *ndev = p; 99 u16 out, in; 100 u16 head; 101 int len, copied; 102 u32 id; 103 104 mutex_lock(&ndev->mutex); 105 id = ndev->rx_vqs++ * 2; 106 mutex_unlock(&ndev->mutex); 107 108 kvm__set_thread_name("virtio-net-rx"); 109 110 kvm = ndev->kvm; 111 vq = &ndev->vqs[id]; 112 113 while (1) { 114 mutex_lock(&ndev->io_lock[id]); 115 if (!virt_queue__available(vq)) 116 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 117 mutex_unlock(&ndev->io_lock[id]); 118 119 while (virt_queue__available(vq)) { 120 unsigned char buffer[MAX_PACKET_SIZE + sizeof(struct virtio_net_hdr_mrg_rxbuf)]; 121 struct iovec dummy_iov = { 122 .iov_base = buffer, 123 .iov_len = sizeof(buffer), 124 }; 125 struct virtio_net_hdr_mrg_rxbuf *hdr; 126 int i; 127 128 len = ndev->ops->rx(&dummy_iov, 1, ndev); 129 if (len < 0) { 130 pr_warning("%s: rx on vq %u failed (%d), exiting thread\n", 131 __func__, id, len); 132 goto out_err; 133 } 134 135 copied = i = 0; 136 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 137 hdr = iov[0].iov_base; 138 while (copied < len) { 139 size_t iovsize = min_t(size_t, len - copied, iov_size(iov, in)); 140 141 memcpy_toiovec(iov, buffer + copied, iovsize); 142 copied += iovsize; 143 if (i++ == 0) 144 virtio_net_fix_rx_hdr(hdr, ndev); 145 if (has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) { 146 u16 num_buffers = virtio_guest_to_host_u16(vq, hdr->num_buffers); 147 hdr->num_buffers = virtio_host_to_guest_u16(vq, num_buffers + 1); 148 } 149 virt_queue__set_used_elem(vq, head, iovsize); 150 if (copied == len) 151 break; 152 while (!virt_queue__available(vq)) 153 sleep(0); 154 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 155 } 156 /* We should interrupt guest right now, otherwise latency is huge. */ 157 if (virtio_queue__should_signal(vq)) 158 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 159 } 160 } 161 162 out_err: 163 pthread_exit(NULL); 164 return NULL; 165 166 } 167 168 static void *virtio_net_tx_thread(void *p) 169 { 170 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 171 struct virt_queue *vq; 172 struct kvm *kvm; 173 struct net_dev *ndev = p; 174 u16 out, in; 175 u16 head; 176 int len; 177 u32 id; 178 179 mutex_lock(&ndev->mutex); 180 id = ndev->tx_vqs++ * 2 + 1; 181 mutex_unlock(&ndev->mutex); 182 183 kvm__set_thread_name("virtio-net-tx"); 184 185 kvm = ndev->kvm; 186 vq = &ndev->vqs[id]; 187 188 while (1) { 189 mutex_lock(&ndev->io_lock[id]); 190 if (!virt_queue__available(vq)) 191 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 192 mutex_unlock(&ndev->io_lock[id]); 193 194 while (virt_queue__available(vq)) { 195 struct virtio_net_hdr *hdr; 196 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 197 hdr = iov[0].iov_base; 198 virtio_net_fix_tx_hdr(hdr, ndev); 199 len = ndev->ops->tx(iov, out, ndev); 200 if (len < 0) { 201 pr_warning("%s: tx on vq %u failed (%d)\n", 202 __func__, id, errno); 203 goto out_err; 204 } 205 206 virt_queue__set_used_elem(vq, head, len); 207 } 208 209 if (virtio_queue__should_signal(vq)) 210 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 211 } 212 213 out_err: 214 pthread_exit(NULL); 215 return NULL; 216 } 217 218 static virtio_net_ctrl_ack virtio_net_handle_mq(struct kvm* kvm, struct net_dev *ndev, struct virtio_net_ctrl_hdr *ctrl) 219 { 220 /* Not much to do here */ 221 return VIRTIO_NET_OK; 222 } 223 224 static void *virtio_net_ctrl_thread(void *p) 225 { 226 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 227 u16 out, in, head; 228 struct net_dev *ndev = p; 229 struct kvm *kvm = ndev->kvm; 230 u32 id = ndev->queue_pairs * 2; 231 struct virt_queue *vq = &ndev->vqs[id]; 232 struct virtio_net_ctrl_hdr *ctrl; 233 virtio_net_ctrl_ack *ack; 234 235 kvm__set_thread_name("virtio-net-ctrl"); 236 237 while (1) { 238 mutex_lock(&ndev->io_lock[id]); 239 if (!virt_queue__available(vq)) 240 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 241 mutex_unlock(&ndev->io_lock[id]); 242 243 while (virt_queue__available(vq)) { 244 head = virt_queue__get_iov(&ndev->vqs[id], iov, &out, &in, kvm); 245 ctrl = iov[0].iov_base; 246 ack = iov[out].iov_base; 247 248 switch (ctrl->class) { 249 case VIRTIO_NET_CTRL_MQ: 250 *ack = virtio_net_handle_mq(kvm, ndev, ctrl); 251 break; 252 default: 253 *ack = VIRTIO_NET_ERR; 254 break; 255 } 256 virt_queue__set_used_elem(&ndev->vqs[id], head, iov[out].iov_len); 257 } 258 259 if (virtio_queue__should_signal(&ndev->vqs[id])) 260 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 261 } 262 263 pthread_exit(NULL); 264 265 return NULL; 266 } 267 268 static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue) 269 { 270 if ((u32)queue >= (ndev->queue_pairs * 2 + 1)) { 271 pr_warning("Unknown queue index %u", queue); 272 return; 273 } 274 275 mutex_lock(&ndev->io_lock[queue]); 276 pthread_cond_signal(&ndev->io_cond[queue]); 277 mutex_unlock(&ndev->io_lock[queue]); 278 } 279 280 static int virtio_net_request_tap(struct net_dev *ndev, struct ifreq *ifr, 281 const char *tapname) 282 { 283 int ret; 284 285 memset(ifr, 0, sizeof(*ifr)); 286 ifr->ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; 287 if (tapname) 288 strncpy(ifr->ifr_name, tapname, sizeof(ifr->ifr_name)); 289 290 ret = ioctl(ndev->tap_fd, TUNSETIFF, ifr); 291 292 if (ret >= 0) 293 strncpy(ndev->tap_name, ifr->ifr_name, sizeof(ndev->tap_name)); 294 return ret; 295 } 296 297 static bool virtio_net__tap_init(struct net_dev *ndev) 298 { 299 int sock = socket(AF_INET, SOCK_STREAM, 0); 300 int pid, status, offload, hdr_len; 301 struct sockaddr_in sin = {0}; 302 struct ifreq ifr; 303 const struct virtio_net_params *params = ndev->params; 304 bool skipconf = !!params->tapif; 305 bool macvtap = skipconf && (params->tapif[0] == '/'); 306 const char *tap_file = "/dev/net/tun"; 307 308 /* Did the user already gave us the FD? */ 309 if (params->fd) { 310 ndev->tap_fd = params->fd; 311 return 1; 312 } 313 314 if (macvtap) 315 tap_file = params->tapif; 316 317 ndev->tap_fd = open(tap_file, O_RDWR); 318 if (ndev->tap_fd < 0) { 319 pr_warning("Unable to open %s", tap_file); 320 goto fail; 321 } 322 323 if (!macvtap && 324 virtio_net_request_tap(ndev, &ifr, params->tapif) < 0) { 325 pr_warning("Config tap device error. Are you root?"); 326 goto fail; 327 } 328 329 hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 330 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 331 sizeof(struct virtio_net_hdr); 332 if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0) 333 pr_warning("Config tap device TUNSETVNETHDRSZ error"); 334 335 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO; 336 if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) { 337 pr_warning("Config tap device TUNSETOFFLOAD error"); 338 goto fail; 339 } 340 341 if (strcmp(params->script, "none")) { 342 pid = fork(); 343 if (pid == 0) { 344 execl(params->script, params->script, ndev->tap_name, NULL); 345 _exit(1); 346 } else { 347 waitpid(pid, &status, 0); 348 if (WIFEXITED(status) && WEXITSTATUS(status) != 0) { 349 pr_warning("Fail to setup tap by %s", params->script); 350 goto fail; 351 } 352 } 353 } else if (!skipconf) { 354 memset(&ifr, 0, sizeof(ifr)); 355 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name)); 356 sin.sin_addr.s_addr = inet_addr(params->host_ip); 357 memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr)); 358 ifr.ifr_addr.sa_family = AF_INET; 359 if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) { 360 pr_warning("Could not set ip address on tap device"); 361 goto fail; 362 } 363 } 364 365 if (!skipconf) { 366 memset(&ifr, 0, sizeof(ifr)); 367 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name)); 368 ioctl(sock, SIOCGIFFLAGS, &ifr); 369 ifr.ifr_flags |= IFF_UP | IFF_RUNNING; 370 if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0) 371 pr_warning("Could not bring tap device up"); 372 } 373 374 close(sock); 375 376 return 1; 377 378 fail: 379 if (sock >= 0) 380 close(sock); 381 if (ndev->tap_fd >= 0) 382 close(ndev->tap_fd); 383 384 return 0; 385 } 386 387 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 388 { 389 return writev(ndev->tap_fd, iov, out); 390 } 391 392 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 393 { 394 return readv(ndev->tap_fd, iov, in); 395 } 396 397 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 398 { 399 return uip_tx(iov, out, &ndev->info); 400 } 401 402 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 403 { 404 return uip_rx(iov, in, &ndev->info); 405 } 406 407 static struct net_dev_operations tap_ops = { 408 .rx = tap_ops_rx, 409 .tx = tap_ops_tx, 410 }; 411 412 static struct net_dev_operations uip_ops = { 413 .rx = uip_ops_rx, 414 .tx = uip_ops_tx, 415 }; 416 417 static u8 *get_config(struct kvm *kvm, void *dev) 418 { 419 struct net_dev *ndev = dev; 420 421 return ((u8 *)(&ndev->config)); 422 } 423 424 static u32 get_host_features(struct kvm *kvm, void *dev) 425 { 426 struct net_dev *ndev = dev; 427 428 return 1UL << VIRTIO_NET_F_MAC 429 | 1UL << VIRTIO_NET_F_CSUM 430 | 1UL << VIRTIO_NET_F_HOST_UFO 431 | 1UL << VIRTIO_NET_F_HOST_TSO4 432 | 1UL << VIRTIO_NET_F_HOST_TSO6 433 | 1UL << VIRTIO_NET_F_GUEST_UFO 434 | 1UL << VIRTIO_NET_F_GUEST_TSO4 435 | 1UL << VIRTIO_NET_F_GUEST_TSO6 436 | 1UL << VIRTIO_RING_F_EVENT_IDX 437 | 1UL << VIRTIO_RING_F_INDIRECT_DESC 438 | 1UL << VIRTIO_NET_F_CTRL_VQ 439 | 1UL << VIRTIO_NET_F_MRG_RXBUF 440 | 1UL << (ndev->queue_pairs > 1 ? VIRTIO_NET_F_MQ : 0); 441 } 442 443 static int virtio_net__vhost_set_features(struct net_dev *ndev) 444 { 445 u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX; 446 u64 vhost_features; 447 448 if (ioctl(ndev->vhost_fd, VHOST_GET_FEATURES, &vhost_features) != 0) 449 die_perror("VHOST_GET_FEATURES failed"); 450 451 /* make sure both side support mergable rx buffers */ 452 if (vhost_features & 1UL << VIRTIO_NET_F_MRG_RXBUF && 453 has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 454 features |= 1UL << VIRTIO_NET_F_MRG_RXBUF; 455 456 return ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features); 457 } 458 459 static void set_guest_features(struct kvm *kvm, void *dev, u32 features) 460 { 461 struct net_dev *ndev = dev; 462 struct virtio_net_config *conf = &ndev->config; 463 464 ndev->features = features; 465 466 conf->status = virtio_host_to_guest_u16(&ndev->vdev, conf->status); 467 conf->max_virtqueue_pairs = virtio_host_to_guest_u16(&ndev->vdev, 468 conf->max_virtqueue_pairs); 469 470 if (ndev->mode == NET_MODE_TAP) { 471 if (!virtio_net__tap_init(ndev)) 472 die_perror("You have requested a TAP device, but creation of one has failed because"); 473 if (ndev->vhost_fd && 474 virtio_net__vhost_set_features(ndev) != 0) 475 die_perror("VHOST_SET_FEATURES failed"); 476 } else { 477 ndev->info.vnet_hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 478 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 479 sizeof(struct virtio_net_hdr); 480 uip_init(&ndev->info); 481 } 482 } 483 484 static bool is_ctrl_vq(struct net_dev *ndev, u32 vq) 485 { 486 return vq == (u32)(ndev->queue_pairs * 2); 487 } 488 489 static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 page_size, u32 align, 490 u32 pfn) 491 { 492 struct vhost_vring_state state = { .index = vq }; 493 struct vhost_vring_addr addr; 494 struct net_dev *ndev = dev; 495 struct virt_queue *queue; 496 void *p; 497 int r; 498 499 compat__remove_message(compat_id); 500 501 queue = &ndev->vqs[vq]; 502 queue->pfn = pfn; 503 p = virtio_get_vq(kvm, queue->pfn, page_size); 504 505 vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, align); 506 virtio_init_device_vq(&ndev->vdev, queue); 507 508 mutex_init(&ndev->io_lock[vq]); 509 pthread_cond_init(&ndev->io_cond[vq], NULL); 510 if (is_ctrl_vq(ndev, vq)) { 511 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_ctrl_thread, ndev); 512 513 return 0; 514 } else if (ndev->vhost_fd == 0 ) { 515 if (vq & 1) 516 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_tx_thread, ndev); 517 else 518 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_rx_thread, ndev); 519 520 return 0; 521 } 522 523 if (queue->endian != VIRTIO_ENDIAN_HOST) 524 die_perror("VHOST requires VIRTIO_ENDIAN_HOST"); 525 526 state.num = queue->vring.num; 527 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state); 528 if (r < 0) 529 die_perror("VHOST_SET_VRING_NUM failed"); 530 state.num = 0; 531 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state); 532 if (r < 0) 533 die_perror("VHOST_SET_VRING_BASE failed"); 534 535 addr = (struct vhost_vring_addr) { 536 .index = vq, 537 .desc_user_addr = (u64)(unsigned long)queue->vring.desc, 538 .avail_user_addr = (u64)(unsigned long)queue->vring.avail, 539 .used_user_addr = (u64)(unsigned long)queue->vring.used, 540 }; 541 542 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr); 543 if (r < 0) 544 die_perror("VHOST_SET_VRING_ADDR failed"); 545 546 return 0; 547 } 548 549 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi) 550 { 551 struct net_dev *ndev = dev; 552 struct kvm_irqfd irq; 553 struct vhost_vring_file file; 554 int r; 555 556 if (ndev->vhost_fd == 0) 557 return; 558 559 irq = (struct kvm_irqfd) { 560 .gsi = gsi, 561 .fd = eventfd(0, 0), 562 }; 563 file = (struct vhost_vring_file) { 564 .index = vq, 565 .fd = irq.fd, 566 }; 567 568 r = ioctl(kvm->vm_fd, KVM_IRQFD, &irq); 569 if (r < 0) 570 die_perror("KVM_IRQFD failed"); 571 572 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file); 573 if (r < 0) 574 die_perror("VHOST_SET_VRING_CALL failed"); 575 file.fd = ndev->tap_fd; 576 r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file); 577 if (r != 0) 578 die("VHOST_NET_SET_BACKEND failed %d", errno); 579 580 } 581 582 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd) 583 { 584 struct net_dev *ndev = dev; 585 struct vhost_vring_file file = { 586 .index = vq, 587 .fd = efd, 588 }; 589 int r; 590 591 if (ndev->vhost_fd == 0 || is_ctrl_vq(ndev, vq)) 592 return; 593 594 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file); 595 if (r < 0) 596 die_perror("VHOST_SET_VRING_KICK failed"); 597 } 598 599 static int notify_vq(struct kvm *kvm, void *dev, u32 vq) 600 { 601 struct net_dev *ndev = dev; 602 603 virtio_net_handle_callback(kvm, ndev, vq); 604 605 return 0; 606 } 607 608 static int get_pfn_vq(struct kvm *kvm, void *dev, u32 vq) 609 { 610 struct net_dev *ndev = dev; 611 612 return ndev->vqs[vq].pfn; 613 } 614 615 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq) 616 { 617 /* FIXME: dynamic */ 618 return VIRTIO_NET_QUEUE_SIZE; 619 } 620 621 static int set_size_vq(struct kvm *kvm, void *dev, u32 vq, int size) 622 { 623 /* FIXME: dynamic */ 624 return size; 625 } 626 627 static struct virtio_ops net_dev_virtio_ops = (struct virtio_ops) { 628 .get_config = get_config, 629 .get_host_features = get_host_features, 630 .set_guest_features = set_guest_features, 631 .init_vq = init_vq, 632 .get_pfn_vq = get_pfn_vq, 633 .get_size_vq = get_size_vq, 634 .set_size_vq = set_size_vq, 635 .notify_vq = notify_vq, 636 .notify_vq_gsi = notify_vq_gsi, 637 .notify_vq_eventfd = notify_vq_eventfd, 638 }; 639 640 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev) 641 { 642 struct vhost_memory *mem; 643 int r; 644 645 ndev->vhost_fd = open("/dev/vhost-net", O_RDWR); 646 if (ndev->vhost_fd < 0) 647 die_perror("Failed openning vhost-net device"); 648 649 mem = calloc(1, sizeof(*mem) + sizeof(struct vhost_memory_region)); 650 if (mem == NULL) 651 die("Failed allocating memory for vhost memory map"); 652 653 mem->nregions = 1; 654 mem->regions[0] = (struct vhost_memory_region) { 655 .guest_phys_addr = 0, 656 .memory_size = kvm->ram_size, 657 .userspace_addr = (unsigned long)kvm->ram_start, 658 }; 659 660 r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER); 661 if (r != 0) 662 die_perror("VHOST_SET_OWNER failed"); 663 664 r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem); 665 if (r != 0) 666 die_perror("VHOST_SET_MEM_TABLE failed"); 667 668 ndev->vdev.use_vhost = true; 669 670 free(mem); 671 } 672 673 static inline void str_to_mac(const char *str, char *mac) 674 { 675 sscanf(str, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx", 676 mac, mac+1, mac+2, mac+3, mac+4, mac+5); 677 } 678 static int set_net_param(struct kvm *kvm, struct virtio_net_params *p, 679 const char *param, const char *val) 680 { 681 if (strcmp(param, "guest_mac") == 0) { 682 str_to_mac(val, p->guest_mac); 683 } else if (strcmp(param, "mode") == 0) { 684 if (!strncmp(val, "user", 4)) { 685 int i; 686 687 for (i = 0; i < kvm->cfg.num_net_devices; i++) 688 if (kvm->cfg.net_params[i].mode == NET_MODE_USER) 689 die("Only one usermode network device allowed at a time"); 690 p->mode = NET_MODE_USER; 691 } else if (!strncmp(val, "tap", 3)) { 692 p->mode = NET_MODE_TAP; 693 } else if (!strncmp(val, "none", 4)) { 694 kvm->cfg.no_net = 1; 695 return -1; 696 } else 697 die("Unknown network mode %s, please use user, tap or none", kvm->cfg.network); 698 } else if (strcmp(param, "script") == 0) { 699 p->script = strdup(val); 700 } else if (strcmp(param, "guest_ip") == 0) { 701 p->guest_ip = strdup(val); 702 } else if (strcmp(param, "host_ip") == 0) { 703 p->host_ip = strdup(val); 704 } else if (strcmp(param, "trans") == 0) { 705 p->trans = strdup(val); 706 } else if (strcmp(param, "tapif") == 0) { 707 p->tapif = strdup(val); 708 } else if (strcmp(param, "vhost") == 0) { 709 p->vhost = atoi(val); 710 } else if (strcmp(param, "fd") == 0) { 711 p->fd = atoi(val); 712 } else if (strcmp(param, "mq") == 0) { 713 p->mq = atoi(val); 714 } else 715 die("Unknown network parameter %s", param); 716 717 return 0; 718 } 719 720 int netdev_parser(const struct option *opt, const char *arg, int unset) 721 { 722 struct virtio_net_params p; 723 char *buf = NULL, *cmd = NULL, *cur = NULL; 724 bool on_cmd = true; 725 struct kvm *kvm = opt->ptr; 726 727 if (arg) { 728 buf = strdup(arg); 729 if (buf == NULL) 730 die("Failed allocating new net buffer"); 731 cur = strtok(buf, ",="); 732 } 733 734 p = (struct virtio_net_params) { 735 .guest_ip = DEFAULT_GUEST_ADDR, 736 .host_ip = DEFAULT_HOST_ADDR, 737 .script = DEFAULT_SCRIPT, 738 .mode = NET_MODE_TAP, 739 }; 740 741 str_to_mac(DEFAULT_GUEST_MAC, p.guest_mac); 742 p.guest_mac[5] += kvm->cfg.num_net_devices; 743 744 while (cur) { 745 if (on_cmd) { 746 cmd = cur; 747 } else { 748 if (set_net_param(kvm, &p, cmd, cur) < 0) 749 goto done; 750 } 751 on_cmd = !on_cmd; 752 753 cur = strtok(NULL, ",="); 754 }; 755 756 kvm->cfg.num_net_devices++; 757 758 kvm->cfg.net_params = realloc(kvm->cfg.net_params, kvm->cfg.num_net_devices * sizeof(*kvm->cfg.net_params)); 759 if (kvm->cfg.net_params == NULL) 760 die("Failed adding new network device"); 761 762 kvm->cfg.net_params[kvm->cfg.num_net_devices - 1] = p; 763 764 done: 765 free(buf); 766 return 0; 767 } 768 769 static int virtio_net__init_one(struct virtio_net_params *params) 770 { 771 int i, err; 772 struct net_dev *ndev; 773 struct virtio_ops *ops; 774 enum virtio_trans trans = VIRTIO_DEFAULT_TRANS(params->kvm); 775 776 ndev = calloc(1, sizeof(struct net_dev)); 777 if (ndev == NULL) 778 return -ENOMEM; 779 780 ops = malloc(sizeof(*ops)); 781 if (ops == NULL) { 782 err = -ENOMEM; 783 goto err_free_ndev; 784 } 785 786 list_add_tail(&ndev->list, &ndevs); 787 788 ndev->kvm = params->kvm; 789 ndev->params = params; 790 791 mutex_init(&ndev->mutex); 792 ndev->queue_pairs = max(1, min(VIRTIO_NET_NUM_QUEUES, params->mq)); 793 ndev->config.status = VIRTIO_NET_S_LINK_UP; 794 if (ndev->queue_pairs > 1) 795 ndev->config.max_virtqueue_pairs = ndev->queue_pairs; 796 797 for (i = 0 ; i < 6 ; i++) { 798 ndev->config.mac[i] = params->guest_mac[i]; 799 ndev->info.guest_mac.addr[i] = params->guest_mac[i]; 800 ndev->info.host_mac.addr[i] = params->host_mac[i]; 801 } 802 803 ndev->mode = params->mode; 804 if (ndev->mode == NET_MODE_TAP) { 805 ndev->ops = &tap_ops; 806 } else { 807 ndev->info.host_ip = ntohl(inet_addr(params->host_ip)); 808 ndev->info.guest_ip = ntohl(inet_addr(params->guest_ip)); 809 ndev->info.guest_netmask = ntohl(inet_addr("255.255.255.0")); 810 ndev->info.buf_nr = 20, 811 ndev->ops = &uip_ops; 812 uip_static_init(&ndev->info); 813 } 814 815 *ops = net_dev_virtio_ops; 816 817 if (params->trans) { 818 if (strcmp(params->trans, "mmio") == 0) 819 trans = VIRTIO_MMIO; 820 else if (strcmp(params->trans, "pci") == 0) 821 trans = VIRTIO_PCI; 822 else 823 pr_warning("virtio-net: Unknown transport method : %s, " 824 "falling back to %s.", params->trans, 825 virtio_trans_name(trans)); 826 } 827 828 virtio_init(params->kvm, ndev, &ndev->vdev, ops, trans, 829 PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET); 830 831 if (params->vhost) 832 virtio_net__vhost_init(params->kvm, ndev); 833 834 if (compat_id == -1) 835 compat_id = virtio_compat_add_message("virtio-net", "CONFIG_VIRTIO_NET"); 836 837 return 0; 838 839 err_free_ndev: 840 free(ndev); 841 return err; 842 } 843 844 int virtio_net__init(struct kvm *kvm) 845 { 846 int i; 847 848 for (i = 0; i < kvm->cfg.num_net_devices; i++) { 849 kvm->cfg.net_params[i].kvm = kvm; 850 virtio_net__init_one(&kvm->cfg.net_params[i]); 851 } 852 853 if (kvm->cfg.num_net_devices == 0 && kvm->cfg.no_net == 0) { 854 static struct virtio_net_params net_params; 855 856 net_params = (struct virtio_net_params) { 857 .guest_ip = kvm->cfg.guest_ip, 858 .host_ip = kvm->cfg.host_ip, 859 .kvm = kvm, 860 .script = kvm->cfg.script, 861 .mode = NET_MODE_USER, 862 }; 863 str_to_mac(kvm->cfg.guest_mac, net_params.guest_mac); 864 str_to_mac(kvm->cfg.host_mac, net_params.host_mac); 865 866 virtio_net__init_one(&net_params); 867 } 868 869 return 0; 870 } 871 virtio_dev_init(virtio_net__init); 872 873 int virtio_net__exit(struct kvm *kvm) 874 { 875 return 0; 876 } 877 virtio_dev_exit(virtio_net__exit); 878