1 #include "kvm/virtio-net.h" 2 #include "kvm/virtio-pci-dev.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/ioeventfd.h" 12 13 #include <linux/virtio_net.h> 14 #include <linux/if_tun.h> 15 16 #include <arpa/inet.h> 17 #include <net/if.h> 18 19 #include <unistd.h> 20 #include <assert.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 28 #define VIRTIO_NET_QUEUE_SIZE 128 29 #define VIRTIO_NET_NUM_QUEUES 2 30 #define VIRTIO_NET_RX_QUEUE 0 31 #define VIRTIO_NET_TX_QUEUE 1 32 33 static struct pci_device_header pci_header = { 34 .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET, 35 .device_id = PCI_DEVICE_ID_VIRTIO_NET, 36 .header_type = PCI_HEADER_TYPE_NORMAL, 37 .revision_id = 0, 38 .class = 0x020000, 39 .subsys_vendor_id = PCI_SUBSYSTEM_VENDOR_ID_REDHAT_QUMRANET, 40 .subsys_id = VIRTIO_ID_NET, 41 }; 42 43 struct net_device { 44 pthread_mutex_t mutex; 45 46 struct virt_queue vqs[VIRTIO_NET_NUM_QUEUES]; 47 struct virtio_net_config config; 48 u32 host_features; 49 u32 guest_features; 50 u16 config_vector; 51 u8 status; 52 u8 isr; 53 u16 queue_selector; 54 u16 base_addr; 55 56 pthread_t io_rx_thread; 57 pthread_mutex_t io_rx_lock; 58 pthread_cond_t io_rx_cond; 59 60 pthread_t io_tx_thread; 61 pthread_mutex_t io_tx_lock; 62 pthread_cond_t io_tx_cond; 63 64 int tap_fd; 65 char tap_name[IFNAMSIZ]; 66 67 int mode; 68 69 }; 70 71 static struct net_device ndev = { 72 .mutex = PTHREAD_MUTEX_INITIALIZER, 73 74 .config = { 75 .mac = { 0x00, 0x11, 0x22, 0x33, 0x44, 0x55 }, 76 .status = VIRTIO_NET_S_LINK_UP, 77 }, 78 .host_features = 1UL << VIRTIO_NET_F_MAC 79 | 1UL << VIRTIO_NET_F_CSUM 80 | 1UL << VIRTIO_NET_F_HOST_UFO 81 | 1UL << VIRTIO_NET_F_HOST_TSO4 82 | 1UL << VIRTIO_NET_F_HOST_TSO6 83 | 1UL << VIRTIO_NET_F_GUEST_UFO 84 | 1UL << VIRTIO_NET_F_GUEST_TSO4 85 | 1UL << VIRTIO_NET_F_GUEST_TSO6, 86 }; 87 88 static void *virtio_net_rx_thread(void *p) 89 { 90 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 91 struct virt_queue *vq; 92 struct kvm *kvm; 93 u16 out, in; 94 u16 head; 95 int len; 96 97 kvm = p; 98 vq = &ndev.vqs[VIRTIO_NET_RX_QUEUE]; 99 100 while (1) { 101 mutex_lock(&ndev.io_rx_lock); 102 if (!virt_queue__available(vq)) 103 pthread_cond_wait(&ndev.io_rx_cond, &ndev.io_rx_lock); 104 mutex_unlock(&ndev.io_rx_lock); 105 106 while (virt_queue__available(vq)) { 107 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 108 len = readv(ndev.tap_fd, iov, in); 109 virt_queue__set_used_elem(vq, head, len); 110 111 /* We should interrupt guest right now, otherwise latency is huge. */ 112 virt_queue__trigger_irq(vq, pci_header.irq_line, &ndev.isr, kvm); 113 } 114 115 } 116 117 pthread_exit(NULL); 118 return NULL; 119 120 } 121 122 static void *virtio_net_tx_thread(void *p) 123 { 124 struct iovec iov[VIRTIO_NET_QUEUE_SIZE]; 125 struct virt_queue *vq; 126 struct kvm *kvm; 127 u16 out, in; 128 u16 head; 129 int len; 130 131 kvm = p; 132 vq = &ndev.vqs[VIRTIO_NET_TX_QUEUE]; 133 134 while (1) { 135 mutex_lock(&ndev.io_tx_lock); 136 if (!virt_queue__available(vq)) 137 pthread_cond_wait(&ndev.io_tx_cond, &ndev.io_tx_lock); 138 mutex_unlock(&ndev.io_tx_lock); 139 140 while (virt_queue__available(vq)) { 141 head = virt_queue__get_iov(vq, iov, &out, &in, kvm); 142 len = writev(ndev.tap_fd, iov, out); 143 virt_queue__set_used_elem(vq, head, len); 144 } 145 146 virt_queue__trigger_irq(vq, pci_header.irq_line, &ndev.isr, kvm); 147 148 } 149 150 pthread_exit(NULL); 151 152 return NULL; 153 154 } 155 156 static bool virtio_net_pci_io_device_specific_in(void *data, unsigned long offset, int size, u32 count) 157 { 158 u8 *config_space = (u8 *)&ndev.config; 159 160 if (size != 1 || count != 1) 161 return false; 162 163 if ((offset - VIRTIO_MSI_CONFIG_VECTOR) > sizeof(struct virtio_net_config)) 164 pr_error("config offset is too big: %li", offset - VIRTIO_MSI_CONFIG_VECTOR); 165 166 ioport__write8(data, config_space[offset - VIRTIO_MSI_CONFIG_VECTOR]); 167 168 return true; 169 } 170 171 static bool virtio_net_pci_io_in(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size, u32 count) 172 { 173 unsigned long offset = port - ndev.base_addr; 174 bool ret = true; 175 176 mutex_lock(&ndev.mutex); 177 178 switch (offset) { 179 case VIRTIO_PCI_HOST_FEATURES: 180 ioport__write32(data, ndev.host_features); 181 break; 182 case VIRTIO_PCI_GUEST_FEATURES: 183 ret = false; 184 break; 185 case VIRTIO_PCI_QUEUE_PFN: 186 ioport__write32(data, ndev.vqs[ndev.queue_selector].pfn); 187 break; 188 case VIRTIO_PCI_QUEUE_NUM: 189 ioport__write16(data, VIRTIO_NET_QUEUE_SIZE); 190 break; 191 case VIRTIO_PCI_QUEUE_SEL: 192 case VIRTIO_PCI_QUEUE_NOTIFY: 193 ret = false; 194 break; 195 case VIRTIO_PCI_STATUS: 196 ioport__write8(data, ndev.status); 197 break; 198 case VIRTIO_PCI_ISR: 199 ioport__write8(data, ndev.isr); 200 kvm__irq_line(kvm, pci_header.irq_line, VIRTIO_IRQ_LOW); 201 ndev.isr = VIRTIO_IRQ_LOW; 202 break; 203 case VIRTIO_MSI_CONFIG_VECTOR: 204 ioport__write16(data, ndev.config_vector); 205 break; 206 default: 207 ret = virtio_net_pci_io_device_specific_in(data, offset, size, count); 208 }; 209 210 mutex_unlock(&ndev.mutex); 211 212 return ret; 213 } 214 215 static void virtio_net_handle_callback(struct kvm *kvm, u16 queue_index) 216 { 217 switch (queue_index) { 218 case VIRTIO_NET_TX_QUEUE: { 219 mutex_lock(&ndev.io_tx_lock); 220 pthread_cond_signal(&ndev.io_tx_cond); 221 mutex_unlock(&ndev.io_tx_lock); 222 break; 223 } 224 case VIRTIO_NET_RX_QUEUE: { 225 mutex_lock(&ndev.io_rx_lock); 226 pthread_cond_signal(&ndev.io_rx_cond); 227 mutex_unlock(&ndev.io_rx_lock); 228 break; 229 } 230 default: 231 pr_warning("Unknown queue index %u", queue_index); 232 } 233 } 234 235 static bool virtio_net_pci_io_out(struct ioport *ioport, struct kvm *kvm, u16 port, void *data, int size, u32 count) 236 { 237 unsigned long offset = port - ndev.base_addr; 238 bool ret = true; 239 240 mutex_lock(&ndev.mutex); 241 242 switch (offset) { 243 case VIRTIO_PCI_GUEST_FEATURES: 244 ndev.guest_features = ioport__read32(data); 245 break; 246 case VIRTIO_PCI_QUEUE_PFN: { 247 struct virt_queue *queue; 248 void *p; 249 250 assert(ndev.queue_selector < VIRTIO_NET_NUM_QUEUES); 251 252 queue = &ndev.vqs[ndev.queue_selector]; 253 queue->pfn = ioport__read32(data); 254 p = guest_pfn_to_host(kvm, queue->pfn); 255 256 vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, VIRTIO_PCI_VRING_ALIGN); 257 258 break; 259 } 260 case VIRTIO_PCI_QUEUE_SEL: 261 ndev.queue_selector = ioport__read16(data); 262 break; 263 case VIRTIO_PCI_QUEUE_NOTIFY: { 264 u16 queue_index; 265 266 queue_index = ioport__read16(data); 267 virtio_net_handle_callback(kvm, queue_index); 268 break; 269 } 270 case VIRTIO_PCI_STATUS: 271 ndev.status = ioport__read8(data); 272 break; 273 case VIRTIO_MSI_CONFIG_VECTOR: 274 ndev.config_vector = VIRTIO_MSI_NO_VECTOR; 275 break; 276 case VIRTIO_MSI_QUEUE_VECTOR: 277 break; 278 default: 279 ret = false; 280 }; 281 282 mutex_unlock(&ndev.mutex); 283 284 return ret; 285 } 286 287 static void ioevent_callback(struct kvm *kvm, void *param) 288 { 289 virtio_net_handle_callback(kvm, (u64)(long)param); 290 } 291 292 static struct ioport_operations virtio_net_io_ops = { 293 .io_in = virtio_net_pci_io_in, 294 .io_out = virtio_net_pci_io_out, 295 }; 296 297 static bool virtio_net__tap_init(const struct virtio_net_parameters *params) 298 { 299 int sock = socket(AF_INET, SOCK_STREAM, 0); 300 int i, pid, status, offload, hdr_len; 301 struct sockaddr_in sin = {0}; 302 struct ifreq ifr; 303 304 for (i = 0 ; i < 6 ; i++) 305 ndev.config.mac[i] = params->guest_mac[i]; 306 307 ndev.tap_fd = open("/dev/net/tun", O_RDWR); 308 if (ndev.tap_fd < 0) { 309 pr_warning("Unable to open /dev/net/tun"); 310 goto fail; 311 } 312 313 memset(&ifr, 0, sizeof(ifr)); 314 ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR; 315 if (ioctl(ndev.tap_fd, TUNSETIFF, &ifr) < 0) { 316 pr_warning("Config tap device error. Are you root?"); 317 goto fail; 318 } 319 320 strncpy(ndev.tap_name, ifr.ifr_name, sizeof(ndev.tap_name)); 321 322 if (ioctl(ndev.tap_fd, TUNSETNOCSUM, 1) < 0) { 323 pr_warning("Config tap device TUNSETNOCSUM error"); 324 goto fail; 325 } 326 327 hdr_len = sizeof(struct virtio_net_hdr); 328 if (ioctl(ndev.tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0) { 329 pr_warning("Config tap device TUNSETVNETHDRSZ error"); 330 goto fail; 331 } 332 333 offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO; 334 if (ioctl(ndev.tap_fd, TUNSETOFFLOAD, offload) < 0) { 335 pr_warning("Config tap device TUNSETOFFLOAD error"); 336 goto fail; 337 } 338 339 if (strcmp(params->script, "none")) { 340 pid = fork(); 341 if (pid == 0) { 342 execl(params->script, params->script, ndev.tap_name, NULL); 343 _exit(1); 344 } else { 345 waitpid(pid, &status, 0); 346 if (WIFEXITED(status) && WEXITSTATUS(status) != 0) { 347 pr_warning("Fail to setup tap by %s", params->script); 348 goto fail; 349 } 350 } 351 } else { 352 memset(&ifr, 0, sizeof(ifr)); 353 strncpy(ifr.ifr_name, ndev.tap_name, sizeof(ndev.tap_name)); 354 sin.sin_addr.s_addr = inet_addr(params->host_ip); 355 memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr)); 356 ifr.ifr_addr.sa_family = AF_INET; 357 if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) { 358 pr_warning("Could not set ip address on tap device"); 359 goto fail; 360 } 361 } 362 363 memset(&ifr, 0, sizeof(ifr)); 364 strncpy(ifr.ifr_name, ndev.tap_name, sizeof(ndev.tap_name)); 365 ioctl(sock, SIOCGIFFLAGS, &ifr); 366 ifr.ifr_flags |= IFF_UP | IFF_RUNNING; 367 if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0) 368 pr_warning("Could not bring tap device up"); 369 370 close(sock); 371 372 return 1; 373 374 fail: 375 if (sock >= 0) 376 close(sock); 377 if (ndev.tap_fd >= 0) 378 close(ndev.tap_fd); 379 380 return 0; 381 } 382 383 static void virtio_net__io_thread_init(struct kvm *kvm) 384 { 385 pthread_mutex_init(&ndev.io_rx_lock, NULL); 386 pthread_cond_init(&ndev.io_tx_cond, NULL); 387 388 pthread_mutex_init(&ndev.io_rx_lock, NULL); 389 pthread_cond_init(&ndev.io_tx_cond, NULL); 390 391 pthread_create(&ndev.io_rx_thread, NULL, virtio_net_rx_thread, (void *)kvm); 392 pthread_create(&ndev.io_tx_thread, NULL, virtio_net_tx_thread, (void *)kvm); 393 } 394 395 void virtio_net__init(const struct virtio_net_parameters *params) 396 { 397 if (virtio_net__tap_init(params)) { 398 u8 dev, line, pin; 399 u16 net_base_addr; 400 u64 i; 401 struct ioevent ioevent; 402 403 if (irq__register_device(VIRTIO_ID_NET, &dev, &pin, &line) < 0) 404 return; 405 406 pci_header.irq_pin = pin; 407 pci_header.irq_line = line; 408 net_base_addr = ioport__register(IOPORT_EMPTY, &virtio_net_io_ops, IOPORT_SIZE, NULL); 409 pci_header.bar[0] = net_base_addr | PCI_BASE_ADDRESS_SPACE_IO; 410 ndev.base_addr = net_base_addr; 411 412 pci__register(&pci_header, dev); 413 414 virtio_net__io_thread_init(params->kvm); 415 416 for (i = 0; i < VIRTIO_NET_NUM_QUEUES; i++) { 417 ioevent = (struct ioevent) { 418 .io_addr = net_base_addr + VIRTIO_PCI_QUEUE_NOTIFY, 419 .io_len = sizeof(u16), 420 .fn = ioevent_callback, 421 .datamatch = i, 422 .fn_ptr = (void *)(long)i, 423 .fn_kvm = params->kvm, 424 .fd = eventfd(0, 0), 425 }; 426 427 ioeventfd__add_event(&ioevent); 428 } 429 } 430 } 431