1 #include "kvm/virtio-pci-dev.h" 2 #include "kvm/virtio-net.h" 3 #include "kvm/virtio.h" 4 #include "kvm/ioport.h" 5 #include "kvm/types.h" 6 #include "kvm/mutex.h" 7 #include "kvm/util.h" 8 #include "kvm/kvm.h" 9 #include "kvm/pci.h" 10 #include "kvm/irq.h" 11 #include "kvm/uip.h" 12 #include "kvm/ioeventfd.h" 13 14 #include <linux/virtio_net.h> 15 #include <linux/if_tun.h> 16 17 #include <arpa/inet.h> 18 #include <net/if.h> 19 20 #include <unistd.h> 21 #include <assert.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 29 #define VIRTIO_NET_QUEUE_SIZE 128 30 #define VIRTIO_NET_NUM_QUEUES 2 31 #define VIRTIO_NET_RX_QUEUE 0 32 #define VIRTIO_NET_TX_QUEUE 1 33 34 static struct pci_device_header pci_header = { 35 .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET, 36 .device_id = PCI_DEVICE_ID_VIRTIO_NET, 37 .header_type = PCI_HEADER_TYPE_NORMAL, 38 .revision_id = 0, 39 .class = 0x020000, 40 .subsys_vendor_id = PCI_SUBSYSTEM_VENDOR_ID_REDHAT_QUMRANET, 41 .subsys_id = VIRTIO_ID_NET, 42 }; 43 44 struct net_dev; 45 46 struct net_dev_operations { 47 int (*rx)(struct iovec *iov, u16 in, struct net_dev *ndev); 48 int (*tx)(struct iovec *iov, u16 in, struct net_dev *ndev); 49 }; 50 51 struct net_dev { 52 pthread_mutex_t mutex; 53 54 struct virt_queue vqs[VIRTIO_NET_NUM_QUEUES]; 55 struct virtio_net_config config; 56 u32 host_features; 57 u32 guest_features; 58 u16 config_vector; 59 u8 status; 60 u8 isr; 61 u16 queue_selector; 62 u16 base_addr; 63 64 pthread_t io_rx_thread; 65 pthread_mutex_t io_rx_lock; 66 pthread_cond_t io_rx_cond; 67 68 pthread_t io_tx_thread; 69 pthread_mutex_t io_tx_lock; 70 pthread_cond_t io_tx_cond; 71 72 int tap_fd; 73 char tap_name[IFNAMSIZ]; 74 75 int mode; 76 77 struct uip_info info; 78 struct net_dev_operations *ops; 79 }; 80 81 static struct net_dev ndev = { 82 .mutex = PTHREAD_MUTEX_INITIALIZER, 83 84 .config = { 85 .status = VIRTIO_NET_S_LINK_UP, 86 }, 87 .host_features = 1UL << VIRTIO_NET_F_MAC 88 | 1UL << VIRTIO_NET_F_CSUM 89 | 1UL << VIRTIO_NET_F_HOST_UFO 90 | 1UL << VIRTIO_NET_F_HOST_TSO4 91 | 1UL << VIRTIO_NET_F_HOST_TSO6 92 | 1UL << VIRTIO_NET_F_GUEST_UFO 93 | 1UL << VIRTIO_NET_F_GUEST_TSO4 94 | 1UL << VIRTIO_NET_F_GUEST_TSO6, 95 .info = { 96 .host_mac.addr = {0x00, 0x01, 0x01, 0x01, 0x01, 0x01}, 97 .guest_mac.addr = {0x00, 0x15, 0x15, 0x15, 0x15, 0x15}, 98 .host_ip = 0xc0a82101, 99 .buf_nr = 20, 100 } 101 }; 102 103 static void *virtio_net_rx_thread(void *p) 104 { 105 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 106 struct virt_queue *vq; 107 struct kvm *kvm; 108 u16 out, in; 109 u16 head; 110 int len; 111 112 kvm = p; 113 vq = &ndev.vqs[VIRTIO_NET_RX_QUEUE]; 114 115 while (1) { 116 117 mutex_lock(&ndev.io_rx_lock); 118 if (!virt_queue__available(vq)) 119 pthread_cond_wait(&ndev.io_rx_cond, &ndev.io_rx_lock); 120 mutex_unlock(&ndev.io_rx_lock); 121 122 while (virt_queue__available(vq)) { 123 124 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 125 126 len = ndev.ops->rx(iov, in, &ndev); 127 128 virt_queue__set_used_elem(vq, head, len); 129 130 /* We should interrupt guest right now, otherwise latency is huge. */ 131 virt_queue__trigger_irq(vq, pci_header.irq_line, &ndev.isr, kvm); 132 } 133 134 } 135 136 pthread_exit(NULL); 137 return NULL; 138 139 } 140 141 static void *virtio_net_tx_thread(void *p) 142 { 143 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 144 struct virt_queue *vq; 145 struct kvm *kvm; 146 u16 out, in; 147 u16 head; 148 int len; 149 150 kvm = p; 151 vq = &ndev.vqs[VIRTIO_NET_TX_QUEUE]; 152 153 while (1) { 154 mutex_lock(&ndev.io_tx_lock); 155 if (!virt_queue__available(vq)) 156 pthread_cond_wait(&ndev.io_tx_cond, &ndev.io_tx_lock); 157 mutex_unlock(&ndev.io_tx_lock); 158 159 while (virt_queue__available(vq)) { 160 161 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 162 163 len = ndev.ops->tx(iov, out, &ndev); 164 165 virt_queue__set_used_elem(vq, head, len); 166 } 167 168 virt_queue__trigger_irq(vq, pci_header.irq_line, &ndev.isr, kvm); 169 170 } 171 172 pthread_exit(NULL); 173 174 return NULL; 175 176 } 177 178 static bool virtio_net_pci_io_device_specific_in(void *data, unsigned long offset, int size, u32 count) 179 { 180 u8 *config_space = (u8 *)&ndev.config; 181 182 if (size != 1 || count != 1) 183 return false; 184 185 if ((offset - VIRTIO_MSI_CONFIG_VECTOR) > sizeof(struct virtio_net_config)) 186 pr_error("config offset is too big: %li", offset - VIRTIO_MSI_CONFIG_VECTOR); 187 188 ioport__write8(data, config_space[offset - VIRTIO_MSI_CONFIG_VECTOR]); 189 190 return true; 191 } 192 193 static bool virtio_net_pci_io_in(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size, u32 count) 194 { 195 unsigned long offset = port - ndev.base_addr; 196 bool ret = true; 197 198 mutex_lock(&ndev.mutex); 199 200 switch (offset) { 201 case VIRTIO_PCI_HOST_FEATURES: 202 ioport__write32(data, ndev.host_features); 203 break; 204 case VIRTIO_PCI_GUEST_FEATURES: 205 ret = false; 206 break; 207 case VIRTIO_PCI_QUEUE_PFN: 208 ioport__write32(data, ndev.vqs[ndev.queue_selector].pfn); 209 break; 210 case VIRTIO_PCI_QUEUE_NUM: 211 ioport__write16(data, VIRTIO_NET_QUEUE_SIZE); 212 break; 213 case VIRTIO_PCI_QUEUE_SEL: 214 case VIRTIO_PCI_QUEUE_NOTIFY: 215 ret = false; 216 break; 217 case VIRTIO_PCI_STATUS: 218 ioport__write8(data, ndev.status); 219 break; 220 case VIRTIO_PCI_ISR: 221 ioport__write8(data, ndev.isr); 222 kvm__irq_line(kvm, pci_header.irq_line, VIRTIO_IRQ_LOW); 223 ndev.isr = VIRTIO_IRQ_LOW; 224 break; 225 case VIRTIO_MSI_CONFIG_VECTOR: 226 ioport__write16(data, ndev.config_vector); 227 break; 228 default: 229 ret = virtio_net_pci_io_device_specific_in(data, offset, size, count); 230 }; 231 232 mutex_unlock(&ndev.mutex); 233 234 return ret; 235 } 236 237 static void virtio_net_handle_callback(struct kvm *kvm, u16 queue_index) 238 { 239 switch (queue_index) { 240 case VIRTIO_NET_TX_QUEUE: 241 mutex_lock(&ndev.io_tx_lock); 242 pthread_cond_signal(&ndev.io_tx_cond); 243 mutex_unlock(&ndev.io_tx_lock); 244 break; 245 case VIRTIO_NET_RX_QUEUE: 246 mutex_lock(&ndev.io_rx_lock); 247 pthread_cond_signal(&ndev.io_rx_cond); 248 mutex_unlock(&ndev.io_rx_lock); 249 break; 250 default: 251 pr_warning("Unknown queue index %u", queue_index); 252 } 253 } 254 255 static bool virtio_net_pci_io_out(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size, u32 count) 256 { 257 unsigned long offset = port - ndev.base_addr; 258 bool ret = true; 259 260 mutex_lock(&ndev.mutex); 261 262 switch (offset) { 263 case VIRTIO_PCI_GUEST_FEATURES: 264 ndev.guest_features = ioport__read32(data); 265 break; 266 case VIRTIO_PCI_QUEUE_PFN: { 267 struct virt_queue *queue; 268 void *p; 269 270 assert(ndev.queue_selector < VIRTIO_NET_NUM_QUEUES); 271 272 queue = &ndev.vqs[ndev.queue_selector]; 273 queue->pfn = ioport__read32(data); 274 p = guest_pfn_to_host(kvm, queue->pfn); 275 276 vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, VIRTIO_PCI_VRING_ALIGN); 277 278 break; 279 } 280 case VIRTIO_PCI_QUEUE_SEL: 281 ndev.queue_selector = ioport__read16(data); 282 break; 283 case VIRTIO_PCI_QUEUE_NOTIFY: { 284 u16 queue_index; 285 286 queue_index = ioport__read16(data); 287 virtio_net_handle_callback(kvm, queue_index); 288 break; 289 } 290 case VIRTIO_PCI_STATUS: 291 ndev.status = ioport__read8(data); 292 break; 293 case VIRTIO_MSI_CONFIG_VECTOR: 294 ndev.config_vector = VIRTIO_MSI_NO_VECTOR; 295 break; 296 case VIRTIO_MSI_QUEUE_VECTOR: 297 break; 298 default: 299 ret = false; 300 }; 301 302 mutex_unlock(&ndev.mutex); 303 304 return ret; 305 } 306 307 static void ioevent_callback(struct kvm *kvm, void *param) 308 { 309 virtio_net_handle_callback(kvm, (u64)(long)param); 310 } 311 312 static struct ioport_operations virtio_net_io_ops = { 313 .io_in = virtio_net_pci_io_in, 314 .io_out = virtio_net_pci_io_out, 315 }; 316 317 static bool virtio_net__tap_init(const struct virtio_net_parameters *params) 318 { 319 int sock = socket(AF_INET, SOCK_STREAM, 0); 320 int pid, status, offload, hdr_len; 321 struct sockaddr_in sin = {0}; 322 struct ifreq ifr; 323 324 ndev.tap_fd = open("/dev/net/tun", O_RDWR); 325 if (ndev.tap_fd < 0) { 326 pr_warning("Unable to open /dev/net/tun"); 327 goto fail; 328 } 329 330 memset(&ifr, 0, sizeof(ifr)); 331 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; 332 if (ioctl(ndev.tap_fd, TUNSETIFF, &ifr) < 0) { 333 pr_warning("Config tap device error. Are you root?"); 334 goto fail; 335 } 336 337 strncpy(ndev.tap_name, ifr.ifr_name, sizeof(ndev.tap_name)); 338 339 if (ioctl(ndev.tap_fd, TUNSETNOCSUM, 1) < 0) { 340 pr_warning("Config tap device TUNSETNOCSUM error"); 341 goto fail; 342 } 343 344 hdr_len = sizeof(struct virtio_net_hdr); 345 if (ioctl(ndev.tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0) { 346 pr_warning("Config tap device TUNSETVNETHDRSZ error"); 347 } 348 349 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO; 350 if (ioctl(ndev.tap_fd, TUNSETOFFLOAD, offload) < 0) { 351 pr_warning("Config tap device TUNSETOFFLOAD error"); 352 goto fail; 353 } 354 355 if (strcmp(params->script, "none")) { 356 pid = fork(); 357 if (pid == 0) { 358 execl(params->script, params->script, ndev.tap_name, NULL); 359 _exit(1); 360 } else { 361 waitpid(pid, &status, 0); 362 if (WIFEXITED(status) && WEXITSTATUS(status) != 0) { 363 pr_warning("Fail to setup tap by %s", params->script); 364 goto fail; 365 } 366 } 367 } else { 368 memset(&ifr, 0, sizeof(ifr)); 369 strncpy(ifr.ifr_name, ndev.tap_name, sizeof(ndev.tap_name)); 370 sin.sin_addr.s_addr = inet_addr(params->host_ip); 371 memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr)); 372 ifr.ifr_addr.sa_family = AF_INET; 373 if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) { 374 pr_warning("Could not set ip address on tap device"); 375 goto fail; 376 } 377 } 378 379 memset(&ifr, 0, sizeof(ifr)); 380 strncpy(ifr.ifr_name, ndev.tap_name, sizeof(ndev.tap_name)); 381 ioctl(sock, SIOCGIFFLAGS, &ifr); 382 ifr.ifr_flags |= IFF_UP | IFF_RUNNING; 383 if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0) 384 pr_warning("Could not bring tap device up"); 385 386 close(sock); 387 388 return 1; 389 390 fail: 391 if (sock >= 0) 392 close(sock); 393 if (ndev.tap_fd >= 0) 394 close(ndev.tap_fd); 395 396 return 0; 397 } 398 399 static void virtio_net__io_thread_init(struct kvm *kvm) 400 { 401 pthread_mutex_init(&ndev.io_rx_lock, NULL); 402 pthread_cond_init(&ndev.io_tx_cond, NULL); 403 404 pthread_mutex_init(&ndev.io_rx_lock, NULL); 405 pthread_cond_init(&ndev.io_tx_cond, NULL); 406 407 pthread_create(&ndev.io_rx_thread, NULL, virtio_net_rx_thread, (void *)kvm); 408 pthread_create(&ndev.io_tx_thread, NULL, virtio_net_tx_thread, (void *)kvm); 409 } 410 411 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 412 { 413 return writev(ndev->tap_fd, iov, out); 414 } 415 416 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 417 { 418 return readv(ndev->tap_fd, iov, in); 419 } 420 421 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev) 422 { 423 return uip_tx(iov, out, &ndev->info); 424 } 425 426 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev) 427 { 428 return uip_rx(iov, in, &ndev->info); 429 } 430 431 static struct net_dev_operations tap_ops = { 432 .rx = tap_ops_rx, 433 .tx = tap_ops_tx, 434 }; 435 436 static struct net_dev_operations uip_ops = { 437 .rx = uip_ops_rx, 438 .tx = uip_ops_tx, 439 }; 440 441 void virtio_net__init(const struct virtio_net_parameters *params) 442 { 443 struct ioevent ioevent; 444 u8 dev, line, pin; 445 u16 net_base_addr; 446 int i; 447 448 if (irq__register_device(VIRTIO_ID_NET, &dev, &pin, &line) < 0) 449 return; 450 451 pci_header.irq_pin = pin; 452 pci_header.irq_line = line; 453 net_base_addr = ioport__register(IOPORT_EMPTY, &virtio_net_io_ops, IOPORT_SIZE, NULL); 454 pci_header.bar[0] = net_base_addr | PCI_BASE_ADDRESS_SPACE_IO; 455 ndev.base_addr = net_base_addr; 456 pci__register(&pci_header, dev); 457 458 for (i = 0 ; i < 6 ; i++) 459 ndev.config.mac[i] = params->guest_mac[i]; 460 461 ndev.mode = params->mode; 462 if (ndev.mode == NET_MODE_TAP) { 463 virtio_net__tap_init(params); 464 ndev.ops = &tap_ops; 465 } else { 466 uip_init(&ndev.info); 467 ndev.ops = &uip_ops; 468 } 469 470 virtio_net__io_thread_init(params->kvm); 471 472 for (i = 0; i < VIRTIO_NET_NUM_QUEUES; i++) { 473 ioevent = (struct ioevent) { 474 .io_addr = net_base_addr + VIRTIO_PCI_QUEUE_NOTIFY, 475 .io_len = sizeof(u16), 476 .fn = ioevent_callback, 477 .datamatch = i, 478 .fn_ptr = (void *)(long)i, 479 .fn_kvm = params->kvm, 480 .fd = eventfd(0, 0), 481 }; 482 483 ioeventfd__add_event(&ioevent); 484 } 485 } 486