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 *hdr, struct net_dev *ndev) 84 { 85 hdr->hdr_len = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr_len); 86 hdr->gso_size = virtio_host_to_guest_u16(&ndev->vdev, hdr->gso_size); 87 hdr->csum_start = virtio_host_to_guest_u16(&ndev->vdev, hdr->csum_start); 88 hdr->csum_offset = virtio_host_to_guest_u16(&ndev->vdev, hdr->csum_offset); 89 } 90 91 static void *virtio_net_rx_thread(void *p) 92 { 93 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 94 struct virt_queue *vq; 95 struct kvm *kvm; 96 struct net_dev *ndev = p; 97 u16 out, in; 98 u16 head; 99 int len, copied; 100 u32 id; 101 102 mutex_lock(&ndev->mutex); 103 id = ndev->rx_vqs++ * 2; 104 mutex_unlock(&ndev->mutex); 105 106 kvm__set_thread_name("virtio-net-rx"); 107 108 kvm = ndev->kvm; 109 vq = &ndev->vqs[id]; 110 111 while (1) { 112 mutex_lock(&ndev->io_lock[id]); 113 if (!virt_queue__available(vq)) 114 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 115 mutex_unlock(&ndev->io_lock[id]); 116 117 while (virt_queue__available(vq)) { 118 unsigned char buffer[MAX_PACKET_SIZE + sizeof(struct virtio_net_hdr_mrg_rxbuf)]; 119 struct iovec dummy_iov = { 120 .iov_base = buffer, 121 .iov_len = sizeof(buffer), 122 }; 123 struct virtio_net_hdr_mrg_rxbuf *hdr; 124 u16 num_buffers; 125 126 len = ndev->ops->rx(&dummy_iov, 1, ndev); 127 if (len < 0) { 128 pr_warning("%s: rx on vq %u failed (%d), exiting thread\n", 129 __func__, id, len); 130 goto out_err; 131 } 132 133 copied = num_buffers = 0; 134 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 135 hdr = iov[0].iov_base; 136 while (copied < len) { 137 size_t iovsize = min_t(size_t, len - copied, iov_size(iov, in)); 138 139 memcpy_toiovec(iov, buffer + copied, iovsize); 140 copied += iovsize; 141 virt_queue__set_used_elem_no_update(vq, head, iovsize, num_buffers++); 142 if (copied == len) 143 break; 144 while (!virt_queue__available(vq)) 145 sleep(0); 146 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 147 } 148 149 virtio_net_fix_rx_hdr(&hdr->hdr, ndev); 150 if (has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 151 hdr->num_buffers = virtio_host_to_guest_u16(vq, num_buffers); 152 153 virt_queue__used_idx_advance(vq, num_buffers); 154 155 /* We should interrupt guest right now, otherwise latency is huge. */ 156 if (virtio_queue__should_signal(vq)) 157 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 158 } 159 } 160 161 out_err: 162 pthread_exit(NULL); 163 return NULL; 164 165 } 166 167 static void *virtio_net_tx_thread(void *p) 168 { 169 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 170 struct virt_queue *vq; 171 struct kvm *kvm; 172 struct net_dev *ndev = p; 173 u16 out, in; 174 u16 head; 175 int len; 176 u32 id; 177 178 mutex_lock(&ndev->mutex); 179 id = ndev->tx_vqs++ * 2 + 1; 180 mutex_unlock(&ndev->mutex); 181 182 kvm__set_thread_name("virtio-net-tx"); 183 184 kvm = ndev->kvm; 185 vq = &ndev->vqs[id]; 186 187 while (1) { 188 mutex_lock(&ndev->io_lock[id]); 189 if (!virt_queue__available(vq)) 190 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 191 mutex_unlock(&ndev->io_lock[id]); 192 193 while (virt_queue__available(vq)) { 194 struct virtio_net_hdr *hdr; 195 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 196 hdr = iov[0].iov_base; 197 virtio_net_fix_tx_hdr(hdr, ndev); 198 len = ndev->ops->tx(iov, out, ndev); 199 if (len < 0) { 200 pr_warning("%s: tx on vq %u failed (%d)\n", 201 __func__, id, errno); 202 goto out_err; 203 } 204 205 virt_queue__set_used_elem(vq, head, len); 206 } 207 208 if (virtio_queue__should_signal(vq)) 209 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 210 } 211 212 out_err: 213 pthread_exit(NULL); 214 return NULL; 215 } 216 217 static virtio_net_ctrl_ack virtio_net_handle_mq(struct kvm* kvm, struct net_dev *ndev, struct virtio_net_ctrl_hdr *ctrl) 218 { 219 /* Not much to do here */ 220 return VIRTIO_NET_OK; 221 } 222 223 static void *virtio_net_ctrl_thread(void *p) 224 { 225 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 226 u16 out, in, head; 227 struct net_dev *ndev = p; 228 struct kvm *kvm = ndev->kvm; 229 u32 id = ndev->queue_pairs * 2; 230 struct virt_queue *vq = &ndev->vqs[id]; 231 struct virtio_net_ctrl_hdr *ctrl; 232 virtio_net_ctrl_ack *ack; 233 234 kvm__set_thread_name("virtio-net-ctrl"); 235 236 while (1) { 237 mutex_lock(&ndev->io_lock[id]); 238 if (!virt_queue__available(vq)) 239 pthread_cond_wait(&ndev->io_cond[id], &ndev->io_lock[id].mutex); 240 mutex_unlock(&ndev->io_lock[id]); 241 242 while (virt_queue__available(vq)) { 243 head = virt_queue__get_iov(&ndev->vqs[id], iov, &out, &in, kvm); 244 ctrl = iov[0].iov_base; 245 ack = iov[out].iov_base; 246 247 switch (ctrl->class) { 248 case VIRTIO_NET_CTRL_MQ: 249 *ack = virtio_net_handle_mq(kvm, ndev, ctrl); 250 break; 251 default: 252 *ack = VIRTIO_NET_ERR; 253 break; 254 } 255 virt_queue__set_used_elem(&ndev->vqs[id], head, iov[out].iov_len); 256 } 257 258 if (virtio_queue__should_signal(&ndev->vqs[id])) 259 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, id); 260 } 261 262 pthread_exit(NULL); 263 264 return NULL; 265 } 266 267 static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue) 268 { 269 if ((u32)queue >= (ndev->queue_pairs * 2 + 1)) { 270 pr_warning("Unknown queue index %u", queue); 271 return; 272 } 273 274 mutex_lock(&ndev->io_lock[queue]); 275 pthread_cond_signal(&ndev->io_cond[queue]); 276 mutex_unlock(&ndev->io_lock[queue]); 277 } 278 279 static int virtio_net_request_tap(struct net_dev *ndev, struct ifreq *ifr, 280 const char *tapname) 281 { 282 int ret; 283 284 memset(ifr, 0, sizeof(*ifr)); 285 ifr->ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; 286 if (tapname) 287 strncpy(ifr->ifr_name, tapname, sizeof(ifr->ifr_name)); 288 289 ret = ioctl(ndev->tap_fd, TUNSETIFF, ifr); 290 291 if (ret >= 0) 292 strncpy(ndev->tap_name, ifr->ifr_name, sizeof(ndev->tap_name)); 293 return ret; 294 } 295 296 static int virtio_net_exec_script(const char* script, const char *tap_name) 297 { 298 pid_t pid; 299 int status; 300 301 pid = fork(); 302 if (pid == 0) { 303 execl(script, script, tap_name, NULL); 304 _exit(1); 305 } else { 306 waitpid(pid, &status, 0); 307 if (WIFEXITED(status) && WEXITSTATUS(status) != 0) { 308 pr_warning("Fail to setup tap by %s", script); 309 return -1; 310 } 311 } 312 return 0; 313 } 314 315 static bool virtio_net__tap_init(struct net_dev *ndev) 316 { 317 int sock = socket(AF_INET, SOCK_STREAM, 0); 318 int offload, hdr_len; 319 struct sockaddr_in sin = {0}; 320 struct ifreq ifr; 321 const struct virtio_net_params *params = ndev->params; 322 bool skipconf = !!params->tapif; 323 bool macvtap = skipconf && (params->tapif[0] == '/'); 324 const char *tap_file = "/dev/net/tun"; 325 326 /* Did the user already gave us the FD? */ 327 if (params->fd) { 328 ndev->tap_fd = params->fd; 329 return 1; 330 } 331 332 if (macvtap) 333 tap_file = params->tapif; 334 335 ndev->tap_fd = open(tap_file, O_RDWR); 336 if (ndev->tap_fd < 0) { 337 pr_warning("Unable to open %s", tap_file); 338 goto fail; 339 } 340 341 if (!macvtap && 342 virtio_net_request_tap(ndev, &ifr, params->tapif) < 0) { 343 pr_warning("Config tap device error. Are you root?"); 344 goto fail; 345 } 346 347 hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 348 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 349 sizeof(struct virtio_net_hdr); 350 if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0) 351 pr_warning("Config tap device TUNSETVNETHDRSZ error"); 352 353 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO; 354 if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) { 355 pr_warning("Config tap device TUNSETOFFLOAD error"); 356 goto fail; 357 } 358 359 if (strcmp(params->script, "none")) { 360 if (virtio_net_exec_script(params->script, ndev->tap_name) < 0) 361 goto fail; 362 } else if (!skipconf) { 363 memset(&ifr, 0, sizeof(ifr)); 364 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name)); 365 sin.sin_addr.s_addr = inet_addr(params->host_ip); 366 memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr)); 367 ifr.ifr_addr.sa_family = AF_INET; 368 if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) { 369 pr_warning("Could not set ip address on tap device"); 370 goto fail; 371 } 372 } 373 374 if (!skipconf) { 375 memset(&ifr, 0, sizeof(ifr)); 376 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name)); 377 ioctl(sock, SIOCGIFFLAGS, &ifr); 378 ifr.ifr_flags |= IFF_UP | IFF_RUNNING; 379 if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0) 380 pr_warning("Could not bring tap device up"); 381 } 382 383 close(sock); 384 385 return 1; 386 387 fail: 388 if (sock >= 0) 389 close(sock); 390 if (ndev->tap_fd >= 0) 391 close(ndev->tap_fd); 392 393 return 0; 394 } 395 396 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 397 { 398 return writev(ndev->tap_fd, iov, out); 399 } 400 401 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 402 { 403 return readv(ndev->tap_fd, iov, in); 404 } 405 406 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 407 { 408 return uip_tx(iov, out, &ndev->info); 409 } 410 411 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 412 { 413 return uip_rx(iov, in, &ndev->info); 414 } 415 416 static struct net_dev_operations tap_ops = { 417 .rx = tap_ops_rx, 418 .tx = tap_ops_tx, 419 }; 420 421 static struct net_dev_operations uip_ops = { 422 .rx = uip_ops_rx, 423 .tx = uip_ops_tx, 424 }; 425 426 static u8 *get_config(struct kvm *kvm, void *dev) 427 { 428 struct net_dev *ndev = dev; 429 430 return ((u8 *)(&ndev->config)); 431 } 432 433 static u32 get_host_features(struct kvm *kvm, void *dev) 434 { 435 struct net_dev *ndev = dev; 436 437 return 1UL << VIRTIO_NET_F_MAC 438 | 1UL << VIRTIO_NET_F_CSUM 439 | 1UL << VIRTIO_NET_F_HOST_UFO 440 | 1UL << VIRTIO_NET_F_HOST_TSO4 441 | 1UL << VIRTIO_NET_F_HOST_TSO6 442 | 1UL << VIRTIO_NET_F_GUEST_UFO 443 | 1UL << VIRTIO_NET_F_GUEST_TSO4 444 | 1UL << VIRTIO_NET_F_GUEST_TSO6 445 | 1UL << VIRTIO_RING_F_EVENT_IDX 446 | 1UL << VIRTIO_RING_F_INDIRECT_DESC 447 | 1UL << VIRTIO_NET_F_CTRL_VQ 448 | 1UL << VIRTIO_NET_F_MRG_RXBUF 449 | 1UL << (ndev->queue_pairs > 1 ? VIRTIO_NET_F_MQ : 0); 450 } 451 452 static int virtio_net__vhost_set_features(struct net_dev *ndev) 453 { 454 u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX; 455 u64 vhost_features; 456 457 if (ioctl(ndev->vhost_fd, VHOST_GET_FEATURES, &vhost_features) != 0) 458 die_perror("VHOST_GET_FEATURES failed"); 459 460 /* make sure both side support mergable rx buffers */ 461 if (vhost_features & 1UL << VIRTIO_NET_F_MRG_RXBUF && 462 has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 463 features |= 1UL << VIRTIO_NET_F_MRG_RXBUF; 464 465 return ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features); 466 } 467 468 static void set_guest_features(struct kvm *kvm, void *dev, u32 features) 469 { 470 struct net_dev *ndev = dev; 471 struct virtio_net_config *conf = &ndev->config; 472 473 ndev->features = features; 474 475 conf->status = virtio_host_to_guest_u16(&ndev->vdev, conf->status); 476 conf->max_virtqueue_pairs = virtio_host_to_guest_u16(&ndev->vdev, 477 conf->max_virtqueue_pairs); 478 479 if (ndev->mode == NET_MODE_TAP) { 480 if (!virtio_net__tap_init(ndev)) 481 die_perror("You have requested a TAP device, but creation of one has failed because"); 482 if (ndev->vhost_fd && 483 virtio_net__vhost_set_features(ndev) != 0) 484 die_perror("VHOST_SET_FEATURES failed"); 485 } else { 486 ndev->info.vnet_hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 487 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 488 sizeof(struct virtio_net_hdr); 489 uip_init(&ndev->info); 490 } 491 } 492 493 static bool is_ctrl_vq(struct net_dev *ndev, u32 vq) 494 { 495 return vq == (u32)(ndev->queue_pairs * 2); 496 } 497 498 static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 page_size, u32 align, 499 u32 pfn) 500 { 501 struct vhost_vring_state state = { .index = vq }; 502 struct vhost_vring_addr addr; 503 struct net_dev *ndev = dev; 504 struct virt_queue *queue; 505 void *p; 506 int r; 507 508 compat__remove_message(compat_id); 509 510 queue = &ndev->vqs[vq]; 511 queue->pfn = pfn; 512 p = virtio_get_vq(kvm, queue->pfn, page_size); 513 514 vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, align); 515 virtio_init_device_vq(&ndev->vdev, queue); 516 517 mutex_init(&ndev->io_lock[vq]); 518 pthread_cond_init(&ndev->io_cond[vq], NULL); 519 if (is_ctrl_vq(ndev, vq)) { 520 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_ctrl_thread, ndev); 521 522 return 0; 523 } else if (ndev->vhost_fd == 0 ) { 524 if (vq & 1) 525 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_tx_thread, ndev); 526 else 527 pthread_create(&ndev->io_thread[vq], NULL, virtio_net_rx_thread, ndev); 528 529 return 0; 530 } 531 532 if (queue->endian != VIRTIO_ENDIAN_HOST) 533 die_perror("VHOST requires the same endianness in guest and host"); 534 535 state.num = queue->vring.num; 536 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state); 537 if (r < 0) 538 die_perror("VHOST_SET_VRING_NUM failed"); 539 state.num = 0; 540 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state); 541 if (r < 0) 542 die_perror("VHOST_SET_VRING_BASE failed"); 543 544 addr = (struct vhost_vring_addr) { 545 .index = vq, 546 .desc_user_addr = (u64)(unsigned long)queue->vring.desc, 547 .avail_user_addr = (u64)(unsigned long)queue->vring.avail, 548 .used_user_addr = (u64)(unsigned long)queue->vring.used, 549 }; 550 551 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr); 552 if (r < 0) 553 die_perror("VHOST_SET_VRING_ADDR failed"); 554 555 return 0; 556 } 557 558 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi) 559 { 560 struct net_dev *ndev = dev; 561 struct kvm_irqfd irq; 562 struct vhost_vring_file file; 563 int r; 564 565 if (ndev->vhost_fd == 0) 566 return; 567 568 irq = (struct kvm_irqfd) { 569 .gsi = gsi, 570 .fd = eventfd(0, 0), 571 }; 572 file = (struct vhost_vring_file) { 573 .index = vq, 574 .fd = irq.fd, 575 }; 576 577 r = ioctl(kvm->vm_fd, KVM_IRQFD, &irq); 578 if (r < 0) 579 die_perror("KVM_IRQFD failed"); 580 581 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file); 582 if (r < 0) 583 die_perror("VHOST_SET_VRING_CALL failed"); 584 file.fd = ndev->tap_fd; 585 r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file); 586 if (r != 0) 587 die("VHOST_NET_SET_BACKEND failed %d", errno); 588 589 } 590 591 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd) 592 { 593 struct net_dev *ndev = dev; 594 struct vhost_vring_file file = { 595 .index = vq, 596 .fd = efd, 597 }; 598 int r; 599 600 if (ndev->vhost_fd == 0 || is_ctrl_vq(ndev, vq)) 601 return; 602 603 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file); 604 if (r < 0) 605 die_perror("VHOST_SET_VRING_KICK failed"); 606 } 607 608 static int notify_vq(struct kvm *kvm, void *dev, u32 vq) 609 { 610 struct net_dev *ndev = dev; 611 612 virtio_net_handle_callback(kvm, ndev, vq); 613 614 return 0; 615 } 616 617 static int get_pfn_vq(struct kvm *kvm, void *dev, u32 vq) 618 { 619 struct net_dev *ndev = dev; 620 621 return ndev->vqs[vq].pfn; 622 } 623 624 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq) 625 { 626 /* FIXME: dynamic */ 627 return VIRTIO_NET_QUEUE_SIZE; 628 } 629 630 static int set_size_vq(struct kvm *kvm, void *dev, u32 vq, int size) 631 { 632 /* FIXME: dynamic */ 633 return size; 634 } 635 636 static struct virtio_ops net_dev_virtio_ops = { 637 .get_config = get_config, 638 .get_host_features = get_host_features, 639 .set_guest_features = set_guest_features, 640 .init_vq = init_vq, 641 .get_pfn_vq = get_pfn_vq, 642 .get_size_vq = get_size_vq, 643 .set_size_vq = set_size_vq, 644 .notify_vq = notify_vq, 645 .notify_vq_gsi = notify_vq_gsi, 646 .notify_vq_eventfd = notify_vq_eventfd, 647 }; 648 649 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev) 650 { 651 struct kvm_mem_bank *bank; 652 struct vhost_memory *mem; 653 int r, i; 654 655 ndev->vhost_fd = open("/dev/vhost-net", O_RDWR); 656 if (ndev->vhost_fd < 0) 657 die_perror("Failed openning vhost-net device"); 658 659 mem = calloc(1, sizeof(*mem) + kvm->mem_slots * sizeof(struct vhost_memory_region)); 660 if (mem == NULL) 661 die("Failed allocating memory for vhost memory map"); 662 663 i = 0; 664 list_for_each_entry(bank, &kvm->mem_banks, list) { 665 mem->regions[i] = (struct vhost_memory_region) { 666 .guest_phys_addr = bank->guest_phys_addr, 667 .memory_size = bank->size, 668 .userspace_addr = (unsigned long)bank->host_addr, 669 }; 670 i++; 671 } 672 mem->nregions = i; 673 674 r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER); 675 if (r != 0) 676 die_perror("VHOST_SET_OWNER failed"); 677 678 r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem); 679 if (r != 0) 680 die_perror("VHOST_SET_MEM_TABLE failed"); 681 682 ndev->vdev.use_vhost = true; 683 684 free(mem); 685 } 686 687 static inline void str_to_mac(const char *str, char *mac) 688 { 689 sscanf(str, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx", 690 mac, mac+1, mac+2, mac+3, mac+4, mac+5); 691 } 692 static int set_net_param(struct kvm *kvm, struct virtio_net_params *p, 693 const char *param, const char *val) 694 { 695 if (strcmp(param, "guest_mac") == 0) { 696 str_to_mac(val, p->guest_mac); 697 } else if (strcmp(param, "mode") == 0) { 698 if (!strncmp(val, "user", 4)) { 699 int i; 700 701 for (i = 0; i < kvm->cfg.num_net_devices; i++) 702 if (kvm->cfg.net_params[i].mode == NET_MODE_USER) 703 die("Only one usermode network device allowed at a time"); 704 p->mode = NET_MODE_USER; 705 } else if (!strncmp(val, "tap", 3)) { 706 p->mode = NET_MODE_TAP; 707 } else if (!strncmp(val, "none", 4)) { 708 kvm->cfg.no_net = 1; 709 return -1; 710 } else 711 die("Unknown network mode %s, please use user, tap or none", kvm->cfg.network); 712 } else if (strcmp(param, "script") == 0) { 713 p->script = strdup(val); 714 } else if (strcmp(param, "downscript") == 0) { 715 p->downscript = strdup(val); 716 } else if (strcmp(param, "guest_ip") == 0) { 717 p->guest_ip = strdup(val); 718 } else if (strcmp(param, "host_ip") == 0) { 719 p->host_ip = strdup(val); 720 } else if (strcmp(param, "trans") == 0) { 721 p->trans = strdup(val); 722 } else if (strcmp(param, "tapif") == 0) { 723 p->tapif = strdup(val); 724 } else if (strcmp(param, "vhost") == 0) { 725 p->vhost = atoi(val); 726 } else if (strcmp(param, "fd") == 0) { 727 p->fd = atoi(val); 728 } else if (strcmp(param, "mq") == 0) { 729 p->mq = atoi(val); 730 } else 731 die("Unknown network parameter %s", param); 732 733 return 0; 734 } 735 736 int netdev_parser(const struct option *opt, const char *arg, int unset) 737 { 738 struct virtio_net_params p; 739 char *buf = NULL, *cmd = NULL, *cur = NULL; 740 bool on_cmd = true; 741 struct kvm *kvm = opt->ptr; 742 743 if (arg) { 744 buf = strdup(arg); 745 if (buf == NULL) 746 die("Failed allocating new net buffer"); 747 cur = strtok(buf, ",="); 748 } 749 750 p = (struct virtio_net_params) { 751 .guest_ip = DEFAULT_GUEST_ADDR, 752 .host_ip = DEFAULT_HOST_ADDR, 753 .script = DEFAULT_SCRIPT, 754 .downscript = DEFAULT_SCRIPT, 755 .mode = NET_MODE_TAP, 756 }; 757 758 str_to_mac(DEFAULT_GUEST_MAC, p.guest_mac); 759 p.guest_mac[5] += kvm->cfg.num_net_devices; 760 761 while (cur) { 762 if (on_cmd) { 763 cmd = cur; 764 } else { 765 if (set_net_param(kvm, &p, cmd, cur) < 0) 766 goto done; 767 } 768 on_cmd = !on_cmd; 769 770 cur = strtok(NULL, ",="); 771 }; 772 773 kvm->cfg.num_net_devices++; 774 775 kvm->cfg.net_params = realloc(kvm->cfg.net_params, kvm->cfg.num_net_devices * sizeof(*kvm->cfg.net_params)); 776 if (kvm->cfg.net_params == NULL) 777 die("Failed adding new network device"); 778 779 kvm->cfg.net_params[kvm->cfg.num_net_devices - 1] = p; 780 781 done: 782 free(buf); 783 return 0; 784 } 785 786 static int virtio_net__init_one(struct virtio_net_params *params) 787 { 788 int i, err; 789 struct net_dev *ndev; 790 struct virtio_ops *ops; 791 enum virtio_trans trans = VIRTIO_DEFAULT_TRANS(params->kvm); 792 793 ndev = calloc(1, sizeof(struct net_dev)); 794 if (ndev == NULL) 795 return -ENOMEM; 796 797 ops = malloc(sizeof(*ops)); 798 if (ops == NULL) { 799 err = -ENOMEM; 800 goto err_free_ndev; 801 } 802 803 list_add_tail(&ndev->list, &ndevs); 804 805 ndev->kvm = params->kvm; 806 ndev->params = params; 807 808 mutex_init(&ndev->mutex); 809 ndev->queue_pairs = max(1, min(VIRTIO_NET_NUM_QUEUES, params->mq)); 810 ndev->config.status = VIRTIO_NET_S_LINK_UP; 811 if (ndev->queue_pairs > 1) 812 ndev->config.max_virtqueue_pairs = ndev->queue_pairs; 813 814 for (i = 0 ; i < 6 ; i++) { 815 ndev->config.mac[i] = params->guest_mac[i]; 816 ndev->info.guest_mac.addr[i] = params->guest_mac[i]; 817 ndev->info.host_mac.addr[i] = params->host_mac[i]; 818 } 819 820 ndev->mode = params->mode; 821 if (ndev->mode == NET_MODE_TAP) { 822 ndev->ops = &tap_ops; 823 } else { 824 ndev->info.host_ip = ntohl(inet_addr(params->host_ip)); 825 ndev->info.guest_ip = ntohl(inet_addr(params->guest_ip)); 826 ndev->info.guest_netmask = ntohl(inet_addr("255.255.255.0")); 827 ndev->info.buf_nr = 20, 828 ndev->ops = &uip_ops; 829 uip_static_init(&ndev->info); 830 } 831 832 *ops = net_dev_virtio_ops; 833 834 if (params->trans) { 835 if (strcmp(params->trans, "mmio") == 0) 836 trans = VIRTIO_MMIO; 837 else if (strcmp(params->trans, "pci") == 0) 838 trans = VIRTIO_PCI; 839 else 840 pr_warning("virtio-net: Unknown transport method : %s, " 841 "falling back to %s.", params->trans, 842 virtio_trans_name(trans)); 843 } 844 845 virtio_init(params->kvm, ndev, &ndev->vdev, ops, trans, 846 PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET); 847 848 if (params->vhost) 849 virtio_net__vhost_init(params->kvm, ndev); 850 851 if (compat_id == -1) 852 compat_id = virtio_compat_add_message("virtio-net", "CONFIG_VIRTIO_NET"); 853 854 return 0; 855 856 err_free_ndev: 857 free(ndev); 858 return err; 859 } 860 861 int virtio_net__init(struct kvm *kvm) 862 { 863 int i; 864 865 for (i = 0; i < kvm->cfg.num_net_devices; i++) { 866 kvm->cfg.net_params[i].kvm = kvm; 867 virtio_net__init_one(&kvm->cfg.net_params[i]); 868 } 869 870 if (kvm->cfg.num_net_devices == 0 && kvm->cfg.no_net == 0) { 871 static struct virtio_net_params net_params; 872 873 net_params = (struct virtio_net_params) { 874 .guest_ip = kvm->cfg.guest_ip, 875 .host_ip = kvm->cfg.host_ip, 876 .kvm = kvm, 877 .script = kvm->cfg.script, 878 .mode = NET_MODE_USER, 879 }; 880 str_to_mac(kvm->cfg.guest_mac, net_params.guest_mac); 881 str_to_mac(kvm->cfg.host_mac, net_params.host_mac); 882 883 virtio_net__init_one(&net_params); 884 } 885 886 return 0; 887 } 888 virtio_dev_init(virtio_net__init); 889 890 int virtio_net__exit(struct kvm *kvm) 891 { 892 struct virtio_net_params *params; 893 struct net_dev *ndev; 894 struct list_head *ptr; 895 896 list_for_each(ptr, &ndevs) { 897 ndev = list_entry(ptr, struct net_dev, list); 898 params = ndev->params; 899 /* Cleanup any tap device which attached to bridge */ 900 if (ndev->mode == NET_MODE_TAP && 901 strcmp(params->downscript, "none")) 902 virtio_net_exec_script(params->downscript, ndev->tap_name); 903 } 904 return 0; 905 } 906 virtio_dev_exit(virtio_net__exit); 907