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 #include "kvm/strbuf.h" 12 13 #include <linux/vhost.h> 14 #include <linux/virtio_net.h> 15 #include <linux/if_tun.h> 16 #include <linux/types.h> 17 18 #include <arpa/inet.h> 19 #include <net/if.h> 20 21 #include <unistd.h> 22 #include <fcntl.h> 23 24 #include <sys/socket.h> 25 #include <sys/ioctl.h> 26 #include <sys/types.h> 27 #include <sys/wait.h> 28 #include <sys/eventfd.h> 29 30 #define VIRTIO_NET_QUEUE_SIZE 256 31 #define VIRTIO_NET_NUM_QUEUES 8 32 33 struct net_dev; 34 35 struct net_dev_operations { 36 int (*rx)(struct iovec *iov, u16 in, struct net_dev *ndev); 37 int (*tx)(struct iovec *iov, u16 in, struct net_dev *ndev); 38 }; 39 40 struct net_dev_queue { 41 int id; 42 struct net_dev *ndev; 43 struct virt_queue vq; 44 pthread_t thread; 45 struct mutex lock; 46 pthread_cond_t cond; 47 int gsi; 48 int irqfd; 49 }; 50 51 struct net_dev { 52 struct mutex mutex; 53 struct virtio_device vdev; 54 struct list_head list; 55 56 struct net_dev_queue queues[VIRTIO_NET_NUM_QUEUES * 2 + 1]; 57 struct virtio_net_config config; 58 u32 features, queue_pairs; 59 60 int vhost_fd; 61 int tap_fd; 62 char tap_name[IFNAMSIZ]; 63 bool tap_ufo; 64 65 int mode; 66 67 struct uip_info info; 68 struct net_dev_operations *ops; 69 struct kvm *kvm; 70 71 struct virtio_net_params *params; 72 }; 73 74 static LIST_HEAD(ndevs); 75 static int compat_id = -1; 76 77 #define MAX_PACKET_SIZE 65550 78 79 static bool has_virtio_feature(struct net_dev *ndev, u32 feature) 80 { 81 return ndev->features & (1 << feature); 82 } 83 84 static void virtio_net_fix_tx_hdr(struct virtio_net_hdr *hdr, struct net_dev *ndev) 85 { 86 hdr->hdr_len = virtio_guest_to_host_u16(&ndev->vdev, hdr->hdr_len); 87 hdr->gso_size = virtio_guest_to_host_u16(&ndev->vdev, hdr->gso_size); 88 hdr->csum_start = virtio_guest_to_host_u16(&ndev->vdev, hdr->csum_start); 89 hdr->csum_offset = virtio_guest_to_host_u16(&ndev->vdev, hdr->csum_offset); 90 } 91 92 static void virtio_net_fix_rx_hdr(struct virtio_net_hdr *hdr, struct net_dev *ndev) 93 { 94 hdr->hdr_len = virtio_host_to_guest_u16(&ndev->vdev, hdr->hdr_len); 95 hdr->gso_size = virtio_host_to_guest_u16(&ndev->vdev, hdr->gso_size); 96 hdr->csum_start = virtio_host_to_guest_u16(&ndev->vdev, hdr->csum_start); 97 hdr->csum_offset = virtio_host_to_guest_u16(&ndev->vdev, hdr->csum_offset); 98 } 99 100 static void *virtio_net_rx_thread(void *p) 101 { 102 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 103 struct net_dev_queue *queue = p; 104 struct virt_queue *vq = &queue->vq; 105 struct net_dev *ndev = queue->ndev; 106 struct kvm *kvm; 107 u16 out, in; 108 u16 head; 109 int len, copied; 110 111 kvm__set_thread_name("virtio-net-rx"); 112 113 kvm = ndev->kvm; 114 while (1) { 115 mutex_lock(&queue->lock); 116 if (!virt_queue__available(vq)) 117 pthread_cond_wait(&queue->cond, &queue->lock.mutex); 118 mutex_unlock(&queue->lock); 119 120 while (virt_queue__available(vq)) { 121 unsigned char buffer[MAX_PACKET_SIZE + sizeof(struct virtio_net_hdr_mrg_rxbuf)]; 122 struct iovec dummy_iov = { 123 .iov_base = buffer, 124 .iov_len = sizeof(buffer), 125 }; 126 struct virtio_net_hdr_mrg_rxbuf *hdr; 127 u16 num_buffers; 128 129 len = ndev->ops->rx(&dummy_iov, 1, ndev); 130 if (len < 0) { 131 pr_warning("%s: rx on vq %u failed (%d), exiting thread\n", 132 __func__, queue->id, len); 133 goto out_err; 134 } 135 136 copied = num_buffers = 0; 137 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 138 hdr = iov[0].iov_base; 139 while (copied < len) { 140 size_t iovsize = min_t(size_t, len - copied, iov_size(iov, in)); 141 142 memcpy_toiovec(iov, buffer + copied, iovsize); 143 copied += iovsize; 144 virt_queue__set_used_elem_no_update(vq, head, iovsize, num_buffers++); 145 if (copied == len) 146 break; 147 while (!virt_queue__available(vq)) 148 sleep(0); 149 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 150 } 151 152 virtio_net_fix_rx_hdr(&hdr->hdr, ndev); 153 if (has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 154 hdr->num_buffers = virtio_host_to_guest_u16(vq, num_buffers); 155 156 virt_queue__used_idx_advance(vq, num_buffers); 157 158 /* We should interrupt guest right now, otherwise latency is huge. */ 159 if (virtio_queue__should_signal(vq)) 160 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, queue->id); 161 } 162 } 163 164 out_err: 165 pthread_exit(NULL); 166 return NULL; 167 168 } 169 170 static void *virtio_net_tx_thread(void *p) 171 { 172 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 173 struct net_dev_queue *queue = p; 174 struct virt_queue *vq = &queue->vq; 175 struct net_dev *ndev = queue->ndev; 176 struct kvm *kvm; 177 u16 out, in; 178 u16 head; 179 int len; 180 181 kvm__set_thread_name("virtio-net-tx"); 182 183 kvm = ndev->kvm; 184 185 while (1) { 186 mutex_lock(&queue->lock); 187 if (!virt_queue__available(vq)) 188 pthread_cond_wait(&queue->cond, &queue->lock.mutex); 189 mutex_unlock(&queue->lock); 190 191 while (virt_queue__available(vq)) { 192 struct virtio_net_hdr *hdr; 193 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 194 hdr = iov[0].iov_base; 195 virtio_net_fix_tx_hdr(hdr, ndev); 196 len = ndev->ops->tx(iov, out, ndev); 197 if (len < 0) { 198 pr_warning("%s: tx on vq %u failed (%d)\n", 199 __func__, queue->id, errno); 200 goto out_err; 201 } 202 203 virt_queue__set_used_elem(vq, head, len); 204 } 205 206 if (virtio_queue__should_signal(vq)) 207 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, queue->id); 208 } 209 210 out_err: 211 pthread_exit(NULL); 212 return NULL; 213 } 214 215 static virtio_net_ctrl_ack virtio_net_handle_mq(struct kvm* kvm, struct net_dev *ndev, struct virtio_net_ctrl_hdr *ctrl) 216 { 217 /* Not much to do here */ 218 return VIRTIO_NET_OK; 219 } 220 221 static void *virtio_net_ctrl_thread(void *p) 222 { 223 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 224 struct net_dev_queue *queue = p; 225 struct virt_queue *vq = &queue->vq; 226 struct net_dev *ndev = queue->ndev; 227 u16 out, in, head; 228 struct kvm *kvm = ndev->kvm; 229 struct virtio_net_ctrl_hdr *ctrl; 230 virtio_net_ctrl_ack *ack; 231 232 kvm__set_thread_name("virtio-net-ctrl"); 233 234 while (1) { 235 mutex_lock(&queue->lock); 236 if (!virt_queue__available(vq)) 237 pthread_cond_wait(&queue->cond, &queue->lock.mutex); 238 mutex_unlock(&queue->lock); 239 240 while (virt_queue__available(vq)) { 241 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 242 ctrl = iov[0].iov_base; 243 ack = iov[out].iov_base; 244 245 switch (ctrl->class) { 246 case VIRTIO_NET_CTRL_MQ: 247 *ack = virtio_net_handle_mq(kvm, ndev, ctrl); 248 break; 249 default: 250 *ack = VIRTIO_NET_ERR; 251 break; 252 } 253 virt_queue__set_used_elem(vq, head, iov[out].iov_len); 254 } 255 256 if (virtio_queue__should_signal(vq)) 257 ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, queue->id); 258 } 259 260 pthread_exit(NULL); 261 262 return NULL; 263 } 264 265 static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue) 266 { 267 struct net_dev_queue *net_queue = &ndev->queues[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(&net_queue->lock); 275 pthread_cond_signal(&net_queue->cond); 276 mutex_unlock(&net_queue->lock); 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 strlcpy(ifr->ifr_name, tapname, sizeof(ifr->ifr_name)); 288 289 ret = ioctl(ndev->tap_fd, TUNSETIFF, ifr); 290 291 if (ret >= 0) 292 strlcpy(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 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; 323 324 hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 325 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 326 sizeof(struct virtio_net_hdr); 327 if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0) 328 pr_warning("Config tap device TUNSETVNETHDRSZ error"); 329 330 if (strcmp(params->script, "none")) { 331 if (virtio_net_exec_script(params->script, ndev->tap_name) < 0) 332 goto fail; 333 } else if (!skipconf) { 334 memset(&ifr, 0, sizeof(ifr)); 335 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ifr.ifr_name)); 336 sin.sin_addr.s_addr = inet_addr(params->host_ip); 337 memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr)); 338 ifr.ifr_addr.sa_family = AF_INET; 339 if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) { 340 pr_warning("Could not set ip address on tap device"); 341 goto fail; 342 } 343 } 344 345 if (!skipconf) { 346 memset(&ifr, 0, sizeof(ifr)); 347 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ifr.ifr_name)); 348 ioctl(sock, SIOCGIFFLAGS, &ifr); 349 ifr.ifr_flags |= IFF_UP | IFF_RUNNING; 350 if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0) 351 pr_warning("Could not bring tap device up"); 352 } 353 354 close(sock); 355 356 return 1; 357 358 fail: 359 if (sock >= 0) 360 close(sock); 361 if (ndev->tap_fd >= 0) 362 close(ndev->tap_fd); 363 364 return 0; 365 } 366 367 static void virtio_net__tap_exit(struct net_dev *ndev) 368 { 369 int sock; 370 struct ifreq ifr; 371 372 if (ndev->params->tapif) 373 return; 374 375 sock = socket(AF_INET, SOCK_STREAM, 0); 376 strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ifr.ifr_name)); 377 ioctl(sock, SIOCGIFFLAGS, &ifr); 378 ifr.ifr_flags &= ~(IFF_UP | IFF_RUNNING); 379 if (ioctl(sock, SIOCGIFFLAGS, &ifr) < 0) 380 pr_warning("Count not bring tap device down"); 381 close(sock); 382 } 383 384 static bool virtio_net__tap_create(struct net_dev *ndev) 385 { 386 int offload; 387 struct ifreq ifr; 388 const struct virtio_net_params *params = ndev->params; 389 bool macvtap = (!!params->tapif) && (params->tapif[0] == '/'); 390 391 /* Did the user already gave us the FD? */ 392 if (params->fd) 393 ndev->tap_fd = params->fd; 394 else { 395 const char *tap_file = "/dev/net/tun"; 396 397 /* Did the user ask us to use macvtap? */ 398 if (macvtap) 399 tap_file = params->tapif; 400 401 ndev->tap_fd = open(tap_file, O_RDWR); 402 if (ndev->tap_fd < 0) { 403 pr_warning("Unable to open %s", tap_file); 404 return 0; 405 } 406 } 407 408 if (!macvtap && 409 virtio_net_request_tap(ndev, &ifr, params->tapif) < 0) { 410 pr_warning("Config tap device error. Are you root?"); 411 goto fail; 412 } 413 414 /* 415 * The UFO support had been removed from kernel in commit: 416 * ID: fb652fdfe83710da0ca13448a41b7ed027d0a984 417 * https://www.spinics.net/lists/netdev/msg443562.html 418 * In oder to support the older kernels without this commit, 419 * we set the TUN_F_UFO to offload by default to test the status of 420 * UFO kernel support. 421 */ 422 ndev->tap_ufo = true; 423 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO; 424 if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) { 425 /* 426 * Is this failure caused by kernel remove the UFO support? 427 * Try TUNSETOFFLOAD without TUN_F_UFO. 428 */ 429 offload &= ~TUN_F_UFO; 430 if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) { 431 pr_warning("Config tap device TUNSETOFFLOAD error"); 432 goto fail; 433 } 434 ndev->tap_ufo = false; 435 } 436 437 return 1; 438 439 fail: 440 if ((ndev->tap_fd >= 0) || (!params->fd) ) 441 close(ndev->tap_fd); 442 443 return 0; 444 } 445 446 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 447 { 448 return writev(ndev->tap_fd, iov, out); 449 } 450 451 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 452 { 453 return readv(ndev->tap_fd, iov, in); 454 } 455 456 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 457 { 458 return uip_tx(iov, out, &ndev->info); 459 } 460 461 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 462 { 463 return uip_rx(iov, in, &ndev->info); 464 } 465 466 static struct net_dev_operations tap_ops = { 467 .rx = tap_ops_rx, 468 .tx = tap_ops_tx, 469 }; 470 471 static struct net_dev_operations uip_ops = { 472 .rx = uip_ops_rx, 473 .tx = uip_ops_tx, 474 }; 475 476 static u8 *get_config(struct kvm *kvm, void *dev) 477 { 478 struct net_dev *ndev = dev; 479 480 return ((u8 *)(&ndev->config)); 481 } 482 483 static size_t get_config_size(struct kvm *kvm, void *dev) 484 { 485 struct net_dev *ndev = dev; 486 487 return sizeof(ndev->config); 488 } 489 490 static u32 get_host_features(struct kvm *kvm, void *dev) 491 { 492 u32 features; 493 struct net_dev *ndev = dev; 494 495 features = 1UL << VIRTIO_NET_F_MAC 496 | 1UL << VIRTIO_NET_F_CSUM 497 | 1UL << VIRTIO_NET_F_HOST_TSO4 498 | 1UL << VIRTIO_NET_F_HOST_TSO6 499 | 1UL << VIRTIO_NET_F_GUEST_TSO4 500 | 1UL << VIRTIO_NET_F_GUEST_TSO6 501 | 1UL << VIRTIO_RING_F_EVENT_IDX 502 | 1UL << VIRTIO_RING_F_INDIRECT_DESC 503 | 1UL << VIRTIO_NET_F_CTRL_VQ 504 | 1UL << VIRTIO_NET_F_MRG_RXBUF 505 | 1UL << (ndev->queue_pairs > 1 ? VIRTIO_NET_F_MQ : 0); 506 507 /* 508 * The UFO feature for host and guest only can be enabled when the 509 * kernel has TAP UFO support. 510 */ 511 if (ndev->tap_ufo) 512 features |= (1UL << VIRTIO_NET_F_HOST_UFO 513 | 1UL << VIRTIO_NET_F_GUEST_UFO); 514 515 return features; 516 } 517 518 static int virtio_net__vhost_set_features(struct net_dev *ndev) 519 { 520 u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX; 521 u64 vhost_features; 522 523 if (ioctl(ndev->vhost_fd, VHOST_GET_FEATURES, &vhost_features) != 0) 524 die_perror("VHOST_GET_FEATURES failed"); 525 526 /* make sure both side support mergable rx buffers */ 527 if (vhost_features & 1UL << VIRTIO_NET_F_MRG_RXBUF && 528 has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF)) 529 features |= 1UL << VIRTIO_NET_F_MRG_RXBUF; 530 531 return ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features); 532 } 533 534 static void set_guest_features(struct kvm *kvm, void *dev, u32 features) 535 { 536 struct net_dev *ndev = dev; 537 struct virtio_net_config *conf = &ndev->config; 538 539 ndev->features = features; 540 541 conf->status = virtio_host_to_guest_u16(&ndev->vdev, conf->status); 542 conf->max_virtqueue_pairs = virtio_host_to_guest_u16(&ndev->vdev, 543 conf->max_virtqueue_pairs); 544 } 545 546 static void virtio_net_start(struct net_dev *ndev) 547 { 548 if (ndev->mode == NET_MODE_TAP) { 549 if (!virtio_net__tap_init(ndev)) 550 die_perror("TAP device initialized failed because"); 551 552 if (ndev->vhost_fd && 553 virtio_net__vhost_set_features(ndev) != 0) 554 die_perror("VHOST_SET_FEATURES failed"); 555 } else { 556 ndev->info.vnet_hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ? 557 sizeof(struct virtio_net_hdr_mrg_rxbuf) : 558 sizeof(struct virtio_net_hdr); 559 uip_init(&ndev->info); 560 } 561 } 562 563 static void virtio_net_stop(struct net_dev *ndev) 564 { 565 /* Undo whatever start() did */ 566 if (ndev->mode == NET_MODE_TAP) 567 virtio_net__tap_exit(ndev); 568 else 569 uip_exit(&ndev->info); 570 } 571 572 static void notify_status(struct kvm *kvm, void *dev, u32 status) 573 { 574 if (status & VIRTIO__STATUS_START) 575 virtio_net_start(dev); 576 else if (status & VIRTIO__STATUS_STOP) 577 virtio_net_stop(dev); 578 } 579 580 static bool is_ctrl_vq(struct net_dev *ndev, u32 vq) 581 { 582 return vq == (u32)(ndev->queue_pairs * 2); 583 } 584 585 static int init_vq(struct kvm *kvm, void *dev, u32 vq) 586 { 587 struct vhost_vring_state state = { .index = vq }; 588 struct net_dev_queue *net_queue; 589 struct vhost_vring_addr addr; 590 struct net_dev *ndev = dev; 591 struct virt_queue *queue; 592 int r; 593 594 compat__remove_message(compat_id); 595 596 net_queue = &ndev->queues[vq]; 597 net_queue->id = vq; 598 net_queue->ndev = ndev; 599 queue = &net_queue->vq; 600 virtio_init_device_vq(kvm, &ndev->vdev, queue, VIRTIO_NET_QUEUE_SIZE); 601 602 mutex_init(&net_queue->lock); 603 pthread_cond_init(&net_queue->cond, NULL); 604 if (is_ctrl_vq(ndev, vq)) { 605 pthread_create(&net_queue->thread, NULL, virtio_net_ctrl_thread, 606 net_queue); 607 608 return 0; 609 } else if (ndev->vhost_fd == 0 ) { 610 if (vq & 1) 611 pthread_create(&net_queue->thread, NULL, 612 virtio_net_tx_thread, net_queue); 613 else 614 pthread_create(&net_queue->thread, NULL, 615 virtio_net_rx_thread, net_queue); 616 617 return 0; 618 } 619 620 if (queue->endian != VIRTIO_ENDIAN_HOST) 621 die_perror("VHOST requires the same endianness in guest and host"); 622 623 state.num = queue->vring.num; 624 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state); 625 if (r < 0) 626 die_perror("VHOST_SET_VRING_NUM failed"); 627 state.num = 0; 628 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state); 629 if (r < 0) 630 die_perror("VHOST_SET_VRING_BASE failed"); 631 632 addr = (struct vhost_vring_addr) { 633 .index = vq, 634 .desc_user_addr = (u64)(unsigned long)queue->vring.desc, 635 .avail_user_addr = (u64)(unsigned long)queue->vring.avail, 636 .used_user_addr = (u64)(unsigned long)queue->vring.used, 637 }; 638 639 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr); 640 if (r < 0) 641 die_perror("VHOST_SET_VRING_ADDR failed"); 642 643 return 0; 644 } 645 646 static void exit_vq(struct kvm *kvm, void *dev, u32 vq) 647 { 648 struct net_dev *ndev = dev; 649 struct net_dev_queue *queue = &ndev->queues[vq]; 650 651 if (!is_ctrl_vq(ndev, vq) && queue->gsi) { 652 irq__del_irqfd(kvm, queue->gsi, queue->irqfd); 653 close(queue->irqfd); 654 queue->gsi = queue->irqfd = 0; 655 } 656 657 /* 658 * TODO: vhost reset owner. It's the only way to cleanly stop vhost, but 659 * we can't restart it at the moment. 660 */ 661 if (ndev->vhost_fd && !is_ctrl_vq(ndev, vq)) { 662 pr_warning("Cannot reset VHOST queue"); 663 ioctl(ndev->vhost_fd, VHOST_RESET_OWNER); 664 return; 665 } 666 667 /* 668 * Threads are waiting on cancellation points (readv or 669 * pthread_cond_wait) and should stop gracefully. 670 */ 671 pthread_cancel(queue->thread); 672 pthread_join(queue->thread, NULL); 673 } 674 675 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi) 676 { 677 struct net_dev *ndev = dev; 678 struct net_dev_queue *queue = &ndev->queues[vq]; 679 struct vhost_vring_file file; 680 int r; 681 682 if (ndev->vhost_fd == 0) 683 return; 684 685 file = (struct vhost_vring_file) { 686 .index = vq, 687 .fd = eventfd(0, 0), 688 }; 689 690 r = irq__add_irqfd(kvm, gsi, file.fd, -1); 691 if (r < 0) 692 die_perror("KVM_IRQFD failed"); 693 694 queue->irqfd = file.fd; 695 queue->gsi = gsi; 696 697 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file); 698 if (r < 0) 699 die_perror("VHOST_SET_VRING_CALL failed"); 700 file.fd = ndev->tap_fd; 701 r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file); 702 if (r != 0) 703 die("VHOST_NET_SET_BACKEND failed %d", errno); 704 705 } 706 707 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd) 708 { 709 struct net_dev *ndev = dev; 710 struct vhost_vring_file file = { 711 .index = vq, 712 .fd = efd, 713 }; 714 int r; 715 716 if (ndev->vhost_fd == 0 || is_ctrl_vq(ndev, vq)) 717 return; 718 719 r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file); 720 if (r < 0) 721 die_perror("VHOST_SET_VRING_KICK failed"); 722 } 723 724 static int notify_vq(struct kvm *kvm, void *dev, u32 vq) 725 { 726 struct net_dev *ndev = dev; 727 728 virtio_net_handle_callback(kvm, ndev, vq); 729 730 return 0; 731 } 732 733 static struct virt_queue *get_vq(struct kvm *kvm, void *dev, u32 vq) 734 { 735 struct net_dev *ndev = dev; 736 737 return &ndev->queues[vq].vq; 738 } 739 740 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq) 741 { 742 /* FIXME: dynamic */ 743 return VIRTIO_NET_QUEUE_SIZE; 744 } 745 746 static int set_size_vq(struct kvm *kvm, void *dev, u32 vq, int size) 747 { 748 /* FIXME: dynamic */ 749 return size; 750 } 751 752 static unsigned int get_vq_count(struct kvm *kvm, void *dev) 753 { 754 struct net_dev *ndev = dev; 755 756 return ndev->queue_pairs * 2 + 1; 757 } 758 759 static struct virtio_ops net_dev_virtio_ops = { 760 .get_config = get_config, 761 .get_config_size = get_config_size, 762 .get_host_features = get_host_features, 763 .set_guest_features = set_guest_features, 764 .get_vq_count = get_vq_count, 765 .init_vq = init_vq, 766 .exit_vq = exit_vq, 767 .get_vq = get_vq, 768 .get_size_vq = get_size_vq, 769 .set_size_vq = set_size_vq, 770 .notify_vq = notify_vq, 771 .notify_vq_gsi = notify_vq_gsi, 772 .notify_vq_eventfd = notify_vq_eventfd, 773 .notify_status = notify_status, 774 }; 775 776 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev) 777 { 778 struct kvm_mem_bank *bank; 779 struct vhost_memory *mem; 780 int r, i; 781 782 ndev->vhost_fd = open("/dev/vhost-net", O_RDWR); 783 if (ndev->vhost_fd < 0) 784 die_perror("Failed openning vhost-net device"); 785 786 mem = calloc(1, sizeof(*mem) + kvm->mem_slots * sizeof(struct vhost_memory_region)); 787 if (mem == NULL) 788 die("Failed allocating memory for vhost memory map"); 789 790 i = 0; 791 list_for_each_entry(bank, &kvm->mem_banks, list) { 792 mem->regions[i] = (struct vhost_memory_region) { 793 .guest_phys_addr = bank->guest_phys_addr, 794 .memory_size = bank->size, 795 .userspace_addr = (unsigned long)bank->host_addr, 796 }; 797 i++; 798 } 799 mem->nregions = i; 800 801 r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER); 802 if (r != 0) 803 die_perror("VHOST_SET_OWNER failed"); 804 805 r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem); 806 if (r != 0) 807 die_perror("VHOST_SET_MEM_TABLE failed"); 808 809 ndev->vdev.use_vhost = true; 810 811 free(mem); 812 } 813 814 static inline void str_to_mac(const char *str, char *mac) 815 { 816 sscanf(str, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx", 817 mac, mac+1, mac+2, mac+3, mac+4, mac+5); 818 } 819 static int set_net_param(struct kvm *kvm, struct virtio_net_params *p, 820 const char *param, const char *val) 821 { 822 if (strcmp(param, "guest_mac") == 0) { 823 str_to_mac(val, p->guest_mac); 824 } else if (strcmp(param, "mode") == 0) { 825 if (!strncmp(val, "user", 4)) { 826 int i; 827 828 for (i = 0; i < kvm->cfg.num_net_devices; i++) 829 if (kvm->cfg.net_params[i].mode == NET_MODE_USER) 830 die("Only one usermode network device allowed at a time"); 831 p->mode = NET_MODE_USER; 832 } else if (!strncmp(val, "tap", 3)) { 833 p->mode = NET_MODE_TAP; 834 } else if (!strncmp(val, "none", 4)) { 835 kvm->cfg.no_net = 1; 836 return -1; 837 } else 838 die("Unknown network mode %s, please use user, tap or none", kvm->cfg.network); 839 } else if (strcmp(param, "script") == 0) { 840 p->script = strdup(val); 841 } else if (strcmp(param, "downscript") == 0) { 842 p->downscript = strdup(val); 843 } else if (strcmp(param, "guest_ip") == 0) { 844 p->guest_ip = strdup(val); 845 } else if (strcmp(param, "host_ip") == 0) { 846 p->host_ip = strdup(val); 847 } else if (strcmp(param, "trans") == 0) { 848 p->trans = strdup(val); 849 } else if (strcmp(param, "tapif") == 0) { 850 p->tapif = strdup(val); 851 } else if (strcmp(param, "vhost") == 0) { 852 p->vhost = atoi(val); 853 } else if (strcmp(param, "fd") == 0) { 854 p->fd = atoi(val); 855 } else if (strcmp(param, "mq") == 0) { 856 p->mq = atoi(val); 857 } else 858 die("Unknown network parameter %s", param); 859 860 return 0; 861 } 862 863 int netdev_parser(const struct option *opt, const char *arg, int unset) 864 { 865 struct virtio_net_params p; 866 char *buf = NULL, *cmd = NULL, *cur = NULL; 867 bool on_cmd = true; 868 struct kvm *kvm = opt->ptr; 869 870 if (arg) { 871 buf = strdup(arg); 872 if (buf == NULL) 873 die("Failed allocating new net buffer"); 874 cur = strtok(buf, ",="); 875 } 876 877 p = (struct virtio_net_params) { 878 .guest_ip = DEFAULT_GUEST_ADDR, 879 .host_ip = DEFAULT_HOST_ADDR, 880 .script = DEFAULT_SCRIPT, 881 .downscript = DEFAULT_SCRIPT, 882 .mode = NET_MODE_TAP, 883 }; 884 885 str_to_mac(DEFAULT_GUEST_MAC, p.guest_mac); 886 p.guest_mac[5] += kvm->cfg.num_net_devices; 887 888 while (cur) { 889 if (on_cmd) { 890 cmd = cur; 891 } else { 892 if (set_net_param(kvm, &p, cmd, cur) < 0) 893 goto done; 894 } 895 on_cmd = !on_cmd; 896 897 cur = strtok(NULL, ",="); 898 }; 899 900 kvm->cfg.num_net_devices++; 901 902 kvm->cfg.net_params = realloc(kvm->cfg.net_params, kvm->cfg.num_net_devices * sizeof(*kvm->cfg.net_params)); 903 if (kvm->cfg.net_params == NULL) 904 die("Failed adding new network device"); 905 906 kvm->cfg.net_params[kvm->cfg.num_net_devices - 1] = p; 907 908 done: 909 free(buf); 910 return 0; 911 } 912 913 static int virtio_net__init_one(struct virtio_net_params *params) 914 { 915 int i, r; 916 struct net_dev *ndev; 917 struct virtio_ops *ops; 918 enum virtio_trans trans = VIRTIO_DEFAULT_TRANS(params->kvm); 919 920 ndev = calloc(1, sizeof(struct net_dev)); 921 if (ndev == NULL) 922 return -ENOMEM; 923 924 list_add_tail(&ndev->list, &ndevs); 925 926 ops = malloc(sizeof(*ops)); 927 if (ops == NULL) 928 return -ENOMEM; 929 930 ndev->kvm = params->kvm; 931 ndev->params = params; 932 933 mutex_init(&ndev->mutex); 934 ndev->queue_pairs = max(1, min(VIRTIO_NET_NUM_QUEUES, params->mq)); 935 ndev->config.status = VIRTIO_NET_S_LINK_UP; 936 if (ndev->queue_pairs > 1) 937 ndev->config.max_virtqueue_pairs = ndev->queue_pairs; 938 939 for (i = 0 ; i < 6 ; i++) { 940 ndev->config.mac[i] = params->guest_mac[i]; 941 ndev->info.guest_mac.addr[i] = params->guest_mac[i]; 942 ndev->info.host_mac.addr[i] = params->host_mac[i]; 943 } 944 945 ndev->mode = params->mode; 946 if (ndev->mode == NET_MODE_TAP) { 947 ndev->ops = &tap_ops; 948 if (!virtio_net__tap_create(ndev)) 949 die_perror("You have requested a TAP device, but creation of one has failed because"); 950 } else { 951 ndev->info.host_ip = ntohl(inet_addr(params->host_ip)); 952 ndev->info.guest_ip = ntohl(inet_addr(params->guest_ip)); 953 ndev->info.guest_netmask = ntohl(inet_addr("255.255.255.0")); 954 ndev->info.buf_nr = 20, 955 ndev->ops = &uip_ops; 956 uip_static_init(&ndev->info); 957 } 958 959 *ops = net_dev_virtio_ops; 960 961 if (params->trans) { 962 if (strcmp(params->trans, "mmio") == 0) 963 trans = VIRTIO_MMIO; 964 else if (strcmp(params->trans, "pci") == 0) 965 trans = VIRTIO_PCI; 966 else 967 pr_warning("virtio-net: Unknown transport method : %s, " 968 "falling back to %s.", params->trans, 969 virtio_trans_name(trans)); 970 } 971 972 r = virtio_init(params->kvm, ndev, &ndev->vdev, ops, trans, 973 PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET); 974 if (r < 0) { 975 free(ops); 976 return r; 977 } 978 979 if (params->vhost) 980 virtio_net__vhost_init(params->kvm, ndev); 981 982 if (compat_id == -1) 983 compat_id = virtio_compat_add_message("virtio-net", "CONFIG_VIRTIO_NET"); 984 985 return 0; 986 } 987 988 int virtio_net__init(struct kvm *kvm) 989 { 990 int i, r; 991 992 for (i = 0; i < kvm->cfg.num_net_devices; i++) { 993 kvm->cfg.net_params[i].kvm = kvm; 994 r = virtio_net__init_one(&kvm->cfg.net_params[i]); 995 if (r < 0) 996 goto cleanup; 997 } 998 999 if (kvm->cfg.num_net_devices == 0 && kvm->cfg.no_net == 0) { 1000 static struct virtio_net_params net_params; 1001 1002 net_params = (struct virtio_net_params) { 1003 .guest_ip = kvm->cfg.guest_ip, 1004 .host_ip = kvm->cfg.host_ip, 1005 .kvm = kvm, 1006 .script = kvm->cfg.script, 1007 .mode = NET_MODE_USER, 1008 }; 1009 str_to_mac(kvm->cfg.guest_mac, net_params.guest_mac); 1010 str_to_mac(kvm->cfg.host_mac, net_params.host_mac); 1011 1012 r = virtio_net__init_one(&net_params); 1013 if (r < 0) 1014 goto cleanup; 1015 } 1016 1017 return 0; 1018 1019 cleanup: 1020 virtio_net__exit(kvm); 1021 return r; 1022 } 1023 virtio_dev_init(virtio_net__init); 1024 1025 int virtio_net__exit(struct kvm *kvm) 1026 { 1027 struct virtio_net_params *params; 1028 struct net_dev *ndev; 1029 struct list_head *ptr, *n; 1030 1031 list_for_each_safe(ptr, n, &ndevs) { 1032 ndev = list_entry(ptr, struct net_dev, list); 1033 params = ndev->params; 1034 /* Cleanup any tap device which attached to bridge */ 1035 if (ndev->mode == NET_MODE_TAP && 1036 strcmp(params->downscript, "none")) 1037 virtio_net_exec_script(params->downscript, ndev->tap_name); 1038 1039 list_del(&ndev->list); 1040 free(ndev); 1041 } 1042 return 0; 1043 } 1044 virtio_dev_exit(virtio_net__exit); 1045