xref: /kvmtool/virtio/net.c (revision 4059ad8bb8bc1a7dd9760bbb157bda4ba8fdf56c)
131638bcaSCyrill Gorcunov #include "kvm/virtio-pci-dev.h"
2b5ee1ea7SAsias He #include "kvm/virtio-net.h"
34f56d42cSAsias He #include "kvm/virtio.h"
44f56d42cSAsias He #include "kvm/types.h"
54f56d42cSAsias He #include "kvm/mutex.h"
64f56d42cSAsias He #include "kvm/util.h"
74f56d42cSAsias He #include "kvm/kvm.h"
82449f6e3SSasha Levin #include "kvm/irq.h"
9b5ee1ea7SAsias He #include "kvm/uip.h"
10cb83de6fSSasha Levin #include "kvm/guest_compat.h"
114f56d42cSAsias He 
129ed67cdcSSasha Levin #include <linux/vhost.h>
134f56d42cSAsias He #include <linux/virtio_net.h>
144f56d42cSAsias He #include <linux/if_tun.h>
1578a2a3e2SSasha Levin #include <linux/types.h>
16c229370aSIngo Molnar 
17c229370aSIngo Molnar #include <arpa/inet.h>
184f56d42cSAsias He #include <net/if.h>
19c229370aSIngo Molnar 
20c229370aSIngo Molnar #include <unistd.h>
214f56d42cSAsias He #include <fcntl.h>
22c229370aSIngo Molnar 
23cb7202c1SSasha Levin #include <sys/socket.h>
24c229370aSIngo Molnar #include <sys/ioctl.h>
25c229370aSIngo Molnar #include <sys/types.h>
2673b7d038SAmos Kong #include <sys/wait.h>
279ed67cdcSSasha Levin #include <sys/eventfd.h>
284f56d42cSAsias He 
29*4059ad8bSAsias He #define VIRTIO_NET_QUEUE_SIZE		256
304f56d42cSAsias He #define VIRTIO_NET_NUM_QUEUES		2
314f56d42cSAsias He #define VIRTIO_NET_RX_QUEUE		0
324f56d42cSAsias He #define VIRTIO_NET_TX_QUEUE		1
334f56d42cSAsias He 
34b4fdde6dSAsias He struct net_dev;
35b4fdde6dSAsias He 
3678a2a3e2SSasha Levin extern struct kvm *kvm;
3778a2a3e2SSasha Levin 
38b4fdde6dSAsias He struct net_dev_operations {
39b4fdde6dSAsias He 	int (*rx)(struct iovec *iov, u16 in, struct net_dev *ndev);
40b4fdde6dSAsias He 	int (*tx)(struct iovec *iov, u16 in, struct net_dev *ndev);
41b4fdde6dSAsias He };
42b4fdde6dSAsias He 
438626798bSAsias He struct net_dev {
444f56d42cSAsias He 	pthread_mutex_t			mutex;
4502eca50cSAsias He 	struct virtio_device		vdev;
469a6d73f1SSasha Levin 	struct list_head		list;
474f56d42cSAsias He 
484f56d42cSAsias He 	struct virt_queue		vqs[VIRTIO_NET_NUM_QUEUES];
49c229370aSIngo Molnar 	struct virtio_net_config	config;
5078a2a3e2SSasha Levin 	u32				features;
514f56d42cSAsias He 
52c4aa7c02SPekka Enberg 	pthread_t			io_rx_thread;
53c229370aSIngo Molnar 	pthread_mutex_t			io_rx_lock;
54c4aa7c02SPekka Enberg 	pthread_cond_t			io_rx_cond;
55c4aa7c02SPekka Enberg 
56c4aa7c02SPekka Enberg 	pthread_t			io_tx_thread;
57c229370aSIngo Molnar 	pthread_mutex_t			io_tx_lock;
58c4aa7c02SPekka Enberg 	pthread_cond_t			io_tx_cond;
59c4aa7c02SPekka Enberg 
609ed67cdcSSasha Levin 	int				vhost_fd;
614f56d42cSAsias He 	int				tap_fd;
624f56d42cSAsias He 	char				tap_name[IFNAMSIZ];
63bb1a32f1SAsias He 
64bb1a32f1SAsias He 	int				mode;
65bb1a32f1SAsias He 
66b5ee1ea7SAsias He 	struct uip_info			info;
67b4fdde6dSAsias He 	struct net_dev_operations	*ops;
689a6d73f1SSasha Levin 	struct kvm			*kvm;
694f56d42cSAsias He };
704f56d42cSAsias He 
719a6d73f1SSasha Levin static LIST_HEAD(ndevs);
72312c62d1SSasha Levin static int compat_id = -1;
734f56d42cSAsias He 
74c4aa7c02SPekka Enberg static void *virtio_net_rx_thread(void *p)
754f56d42cSAsias He {
764f56d42cSAsias He 	struct iovec iov[VIRTIO_NET_QUEUE_SIZE];
774f56d42cSAsias He 	struct virt_queue *vq;
7843835ac9SSasha Levin 	struct kvm *kvm;
799a6d73f1SSasha Levin 	struct net_dev *ndev = p;
803fdf659dSSasha Levin 	u16 out, in;
813fdf659dSSasha Levin 	u16 head;
824f56d42cSAsias He 	int len;
834f56d42cSAsias He 
849a6d73f1SSasha Levin 	kvm	= ndev->kvm;
859a6d73f1SSasha Levin 	vq	= &ndev->vqs[VIRTIO_NET_RX_QUEUE];
86c4aa7c02SPekka Enberg 
87c4aa7c02SPekka Enberg 	while (1) {
889a6d73f1SSasha Levin 		mutex_lock(&ndev->io_rx_lock);
89c4aa7c02SPekka Enberg 		if (!virt_queue__available(vq))
909a6d73f1SSasha Levin 			pthread_cond_wait(&ndev->io_rx_cond, &ndev->io_rx_lock);
919a6d73f1SSasha Levin 		mutex_unlock(&ndev->io_rx_lock);
924f56d42cSAsias He 
934f56d42cSAsias He 		while (virt_queue__available(vq)) {
9443835ac9SSasha Levin 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
959a6d73f1SSasha Levin 			len = ndev->ops->rx(iov, in, ndev);
96246c8347SAsias He 			virt_queue__set_used_elem(vq, head, len);
977f5ffaf5SAsias He 
98c4aa7c02SPekka Enberg 			/* We should interrupt guest right now, otherwise latency is huge. */
9992c1f37bSSasha Levin 			if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_RX_QUEUE]))
10002eca50cSAsias He 				ndev->vdev.ops->signal_vq(kvm, &ndev->vdev,
10192c1f37bSSasha Levin 							   VIRTIO_NET_RX_QUEUE);
1024f56d42cSAsias He 		}
103c4aa7c02SPekka Enberg 	}
104c4aa7c02SPekka Enberg 
105c4aa7c02SPekka Enberg 	pthread_exit(NULL);
106c4aa7c02SPekka Enberg 	return NULL;
107c4aa7c02SPekka Enberg 
108c4aa7c02SPekka Enberg }
109c4aa7c02SPekka Enberg 
110c4aa7c02SPekka Enberg static void *virtio_net_tx_thread(void *p)
1114f56d42cSAsias He {
1124f56d42cSAsias He 	struct iovec iov[VIRTIO_NET_QUEUE_SIZE];
1134f56d42cSAsias He 	struct virt_queue *vq;
11443835ac9SSasha Levin 	struct kvm *kvm;
1159a6d73f1SSasha Levin 	struct net_dev *ndev = p;
1163fdf659dSSasha Levin 	u16 out, in;
1173fdf659dSSasha Levin 	u16 head;
1184f56d42cSAsias He 	int len;
1194f56d42cSAsias He 
1209a6d73f1SSasha Levin 	kvm	= ndev->kvm;
1219a6d73f1SSasha Levin 	vq	= &ndev->vqs[VIRTIO_NET_TX_QUEUE];
122c4aa7c02SPekka Enberg 
123c4aa7c02SPekka Enberg 	while (1) {
1249a6d73f1SSasha Levin 		mutex_lock(&ndev->io_tx_lock);
125c4aa7c02SPekka Enberg 		if (!virt_queue__available(vq))
1269a6d73f1SSasha Levin 			pthread_cond_wait(&ndev->io_tx_cond, &ndev->io_tx_lock);
1279a6d73f1SSasha Levin 		mutex_unlock(&ndev->io_tx_lock);
1284f56d42cSAsias He 
1294f56d42cSAsias He 		while (virt_queue__available(vq)) {
13043835ac9SSasha Levin 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
1319a6d73f1SSasha Levin 			len = ndev->ops->tx(iov, out, ndev);
1324f56d42cSAsias He 			virt_queue__set_used_elem(vq, head, len);
1334f56d42cSAsias He 		}
1344f56d42cSAsias He 
13592c1f37bSSasha Levin 		if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_TX_QUEUE]))
13602eca50cSAsias He 			ndev->vdev.ops->signal_vq(kvm, &ndev->vdev, VIRTIO_NET_TX_QUEUE);
1374f56d42cSAsias He 	}
1384f56d42cSAsias He 
139c4aa7c02SPekka Enberg 	pthread_exit(NULL);
140407475bfSPekka Enberg 
141c4aa7c02SPekka Enberg 	return NULL;
142c4aa7c02SPekka Enberg 
143c4aa7c02SPekka Enberg }
144407475bfSPekka Enberg 
1459a6d73f1SSasha Levin static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue)
1464f56d42cSAsias He {
1479a6d73f1SSasha Levin 	switch (queue) {
148b5ee1ea7SAsias He 	case VIRTIO_NET_TX_QUEUE:
1499a6d73f1SSasha Levin 		mutex_lock(&ndev->io_tx_lock);
1509a6d73f1SSasha Levin 		pthread_cond_signal(&ndev->io_tx_cond);
1519a6d73f1SSasha Levin 		mutex_unlock(&ndev->io_tx_lock);
152407475bfSPekka Enberg 		break;
153b5ee1ea7SAsias He 	case VIRTIO_NET_RX_QUEUE:
1549a6d73f1SSasha Levin 		mutex_lock(&ndev->io_rx_lock);
1559a6d73f1SSasha Levin 		pthread_cond_signal(&ndev->io_rx_cond);
1569a6d73f1SSasha Levin 		mutex_unlock(&ndev->io_rx_lock);
157407475bfSPekka Enberg 		break;
158407475bfSPekka Enberg 	default:
1599a6d73f1SSasha Levin 		pr_warning("Unknown queue index %u", queue);
160c4aa7c02SPekka Enberg 	}
1614f56d42cSAsias He }
1624f56d42cSAsias He 
1639a6d73f1SSasha Levin static bool virtio_net__tap_init(const struct virtio_net_params *params,
1649a6d73f1SSasha Levin 					struct net_dev *ndev)
1654f56d42cSAsias He {
166cb7202c1SSasha Levin 	int sock = socket(AF_INET, SOCK_STREAM, 0);
167f715177dSAsias He 	int pid, status, offload, hdr_len;
168cb7202c1SSasha Levin 	struct sockaddr_in sin = {0};
169246c8347SAsias He 	struct ifreq ifr;
1704f56d42cSAsias He 
171f19edd1eSSasha Levin 	/* Did the user already gave us the FD? */
172f19edd1eSSasha Levin 	if (params->fd) {
173f19edd1eSSasha Levin 		ndev->tap_fd = params->fd;
174f19edd1eSSasha Levin 		return 1;
175f19edd1eSSasha Levin 	}
176f19edd1eSSasha Levin 
1779a6d73f1SSasha Levin 	ndev->tap_fd = open("/dev/net/tun", O_RDWR);
1789a6d73f1SSasha Levin 	if (ndev->tap_fd < 0) {
1794542f276SCyrill Gorcunov 		pr_warning("Unable to open /dev/net/tun");
1803b02f580SSasha Levin 		goto fail;
1813b02f580SSasha Levin 	}
1824f56d42cSAsias He 
1834f56d42cSAsias He 	memset(&ifr, 0, sizeof(ifr));
184246c8347SAsias He 	ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
1859a6d73f1SSasha Levin 	if (ioctl(ndev->tap_fd, TUNSETIFF, &ifr) < 0) {
1864542f276SCyrill Gorcunov 		pr_warning("Config tap device error. Are you root?");
1873b02f580SSasha Levin 		goto fail;
1883b02f580SSasha Levin 	}
1894f56d42cSAsias He 
1909a6d73f1SSasha Levin 	strncpy(ndev->tap_name, ifr.ifr_name, sizeof(ndev->tap_name));
1914f56d42cSAsias He 
1929a6d73f1SSasha Levin 	if (ioctl(ndev->tap_fd, TUNSETNOCSUM, 1) < 0) {
1934542f276SCyrill Gorcunov 		pr_warning("Config tap device TUNSETNOCSUM error");
194246c8347SAsias He 		goto fail;
195246c8347SAsias He 	}
196246c8347SAsias He 
197246c8347SAsias He 	hdr_len = sizeof(struct virtio_net_hdr);
1989a6d73f1SSasha Levin 	if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0)
1994542f276SCyrill Gorcunov 		pr_warning("Config tap device TUNSETVNETHDRSZ error");
200246c8347SAsias He 
201246c8347SAsias He 	offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO;
2029a6d73f1SSasha Levin 	if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) {
2034542f276SCyrill Gorcunov 		pr_warning("Config tap device TUNSETOFFLOAD error");
204246c8347SAsias He 		goto fail;
205246c8347SAsias He 	}
2064f56d42cSAsias He 
20773b7d038SAmos Kong 	if (strcmp(params->script, "none")) {
20873b7d038SAmos Kong 		pid = fork();
20973b7d038SAmos Kong 		if (pid == 0) {
2109a6d73f1SSasha Levin 			execl(params->script, params->script, ndev->tap_name, NULL);
21173b7d038SAmos Kong 			_exit(1);
21273b7d038SAmos Kong 		} else {
21373b7d038SAmos Kong 			waitpid(pid, &status, 0);
21473b7d038SAmos Kong 			if (WIFEXITED(status) && WEXITSTATUS(status) != 0) {
2154542f276SCyrill Gorcunov 				pr_warning("Fail to setup tap by %s", params->script);
21673b7d038SAmos Kong 				goto fail;
21773b7d038SAmos Kong 			}
21873b7d038SAmos Kong 		}
21973b7d038SAmos Kong 	} else {
220cb7202c1SSasha Levin 		memset(&ifr, 0, sizeof(ifr));
2219a6d73f1SSasha Levin 		strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
222bdfcfca6SSasha Levin 		sin.sin_addr.s_addr = inet_addr(params->host_ip);
223cb7202c1SSasha Levin 		memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr));
224cb7202c1SSasha Levin 		ifr.ifr_addr.sa_family = AF_INET;
2253b02f580SSasha Levin 		if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) {
2264542f276SCyrill Gorcunov 			pr_warning("Could not set ip address on tap device");
2273b02f580SSasha Levin 			goto fail;
2283b02f580SSasha Levin 		}
22973b7d038SAmos Kong 	}
230cb7202c1SSasha Levin 
231cb7202c1SSasha Levin 	memset(&ifr, 0, sizeof(ifr));
2329a6d73f1SSasha Levin 	strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
233cb7202c1SSasha Levin 	ioctl(sock, SIOCGIFFLAGS, &ifr);
234cb7202c1SSasha Levin 	ifr.ifr_flags |= IFF_UP | IFF_RUNNING;
235cb7202c1SSasha Levin 	if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0)
2364542f276SCyrill Gorcunov 		pr_warning("Could not bring tap device up");
237cb7202c1SSasha Levin 
238cb7202c1SSasha Levin 	close(sock);
2393b02f580SSasha Levin 
2403b02f580SSasha Levin 	return 1;
2413b02f580SSasha Levin 
2423b02f580SSasha Levin fail:
2433b02f580SSasha Levin 	if (sock >= 0)
2443b02f580SSasha Levin 		close(sock);
2459a6d73f1SSasha Levin 	if (ndev->tap_fd >= 0)
2469a6d73f1SSasha Levin 		close(ndev->tap_fd);
2473b02f580SSasha Levin 
2483b02f580SSasha Levin 	return 0;
2494f56d42cSAsias He }
2504f56d42cSAsias He 
2519a6d73f1SSasha Levin static void virtio_net__io_thread_init(struct kvm *kvm, struct net_dev *ndev)
252c4aa7c02SPekka Enberg {
25306e5512fSAsias He 	pthread_mutex_init(&ndev->io_tx_lock, NULL);
2549a6d73f1SSasha Levin 	pthread_mutex_init(&ndev->io_rx_lock, NULL);
255c4aa7c02SPekka Enberg 
2569a6d73f1SSasha Levin 	pthread_cond_init(&ndev->io_tx_cond, NULL);
25706e5512fSAsias He 	pthread_cond_init(&ndev->io_rx_cond, NULL);
258c4aa7c02SPekka Enberg 
2599a6d73f1SSasha Levin 	pthread_create(&ndev->io_tx_thread, NULL, virtio_net_tx_thread, ndev);
26006e5512fSAsias He 	pthread_create(&ndev->io_rx_thread, NULL, virtio_net_rx_thread, ndev);
261c4aa7c02SPekka Enberg }
262c4aa7c02SPekka Enberg 
263b4fdde6dSAsias He static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
264b4fdde6dSAsias He {
265b4fdde6dSAsias He 	return writev(ndev->tap_fd, iov, out);
266b4fdde6dSAsias He }
267b4fdde6dSAsias He 
268b4fdde6dSAsias He static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
269b4fdde6dSAsias He {
270b4fdde6dSAsias He 	return readv(ndev->tap_fd, iov, in);
271b4fdde6dSAsias He }
272b4fdde6dSAsias He 
273b4fdde6dSAsias He static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
274b4fdde6dSAsias He {
275b4fdde6dSAsias He 	return uip_tx(iov, out, &ndev->info);
276b4fdde6dSAsias He }
277b4fdde6dSAsias He 
278b4fdde6dSAsias He static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
279b4fdde6dSAsias He {
280b4fdde6dSAsias He 	return uip_rx(iov, in, &ndev->info);
281b4fdde6dSAsias He }
282b4fdde6dSAsias He 
283b4fdde6dSAsias He static struct net_dev_operations tap_ops = {
284b4fdde6dSAsias He 	.rx	= tap_ops_rx,
285b4fdde6dSAsias He 	.tx	= tap_ops_tx,
286b4fdde6dSAsias He };
287b4fdde6dSAsias He 
288b4fdde6dSAsias He static struct net_dev_operations uip_ops = {
289b4fdde6dSAsias He 	.rx	= uip_ops_rx,
290b4fdde6dSAsias He 	.tx	= uip_ops_tx,
291b4fdde6dSAsias He };
292b4fdde6dSAsias He 
29378a2a3e2SSasha Levin static void set_config(struct kvm *kvm, void *dev, u8 data, u32 offset)
29478a2a3e2SSasha Levin {
29578a2a3e2SSasha Levin 	struct net_dev *ndev = dev;
29678a2a3e2SSasha Levin 
29778a2a3e2SSasha Levin 	((u8 *)(&ndev->config))[offset] = data;
29878a2a3e2SSasha Levin }
29978a2a3e2SSasha Levin 
30078a2a3e2SSasha Levin static u8 get_config(struct kvm *kvm, void *dev, u32 offset)
30178a2a3e2SSasha Levin {
30278a2a3e2SSasha Levin 	struct net_dev *ndev = dev;
30378a2a3e2SSasha Levin 
30478a2a3e2SSasha Levin 	return ((u8 *)(&ndev->config))[offset];
30578a2a3e2SSasha Levin }
30678a2a3e2SSasha Levin 
30778a2a3e2SSasha Levin static u32 get_host_features(struct kvm *kvm, void *dev)
30878a2a3e2SSasha Levin {
30978a2a3e2SSasha Levin 	return 1UL << VIRTIO_NET_F_MAC
31078a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_CSUM
31178a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_HOST_UFO
31278a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_HOST_TSO4
31378a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_HOST_TSO6
31478a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_GUEST_UFO
31578a2a3e2SSasha Levin 		| 1UL << VIRTIO_NET_F_GUEST_TSO4
31692c1f37bSSasha Levin 		| 1UL << VIRTIO_NET_F_GUEST_TSO6
317754c8ce3SSasha Levin 		| 1UL << VIRTIO_RING_F_EVENT_IDX
318754c8ce3SSasha Levin 		| 1UL << VIRTIO_RING_F_INDIRECT_DESC;
31978a2a3e2SSasha Levin }
32078a2a3e2SSasha Levin 
32178a2a3e2SSasha Levin static void set_guest_features(struct kvm *kvm, void *dev, u32 features)
32278a2a3e2SSasha Levin {
32378a2a3e2SSasha Levin 	struct net_dev *ndev = dev;
32478a2a3e2SSasha Levin 
32578a2a3e2SSasha Levin 	ndev->features = features;
32678a2a3e2SSasha Levin }
32778a2a3e2SSasha Levin 
32878a2a3e2SSasha Levin static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 pfn)
32978a2a3e2SSasha Levin {
3309ed67cdcSSasha Levin 	struct vhost_vring_state state = { .index = vq };
3319ed67cdcSSasha Levin 	struct vhost_vring_addr addr;
33278a2a3e2SSasha Levin 	struct net_dev *ndev = dev;
33378a2a3e2SSasha Levin 	struct virt_queue *queue;
33478a2a3e2SSasha Levin 	void *p;
3359ed67cdcSSasha Levin 	int r;
33678a2a3e2SSasha Levin 
337312c62d1SSasha Levin 	compat__remove_message(compat_id);
33878a2a3e2SSasha Levin 
33978a2a3e2SSasha Levin 	queue		= &ndev->vqs[vq];
34078a2a3e2SSasha Levin 	queue->pfn	= pfn;
34178a2a3e2SSasha Levin 	p		= guest_pfn_to_host(kvm, queue->pfn);
34278a2a3e2SSasha Levin 
34378a2a3e2SSasha Levin 	vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, VIRTIO_PCI_VRING_ALIGN);
34478a2a3e2SSasha Levin 
3459ed67cdcSSasha Levin 	if (ndev->vhost_fd == 0)
34678a2a3e2SSasha Levin 		return 0;
3479ed67cdcSSasha Levin 
3489ed67cdcSSasha Levin 	state.num = queue->vring.num;
3499ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state);
3509ed67cdcSSasha Levin 	if (r < 0)
3519ed67cdcSSasha Levin 		die_perror("VHOST_SET_VRING_NUM failed");
3529ed67cdcSSasha Levin 	state.num = 0;
3539ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state);
3549ed67cdcSSasha Levin 	if (r < 0)
3559ed67cdcSSasha Levin 		die_perror("VHOST_SET_VRING_BASE failed");
3569ed67cdcSSasha Levin 
3579ed67cdcSSasha Levin 	addr = (struct vhost_vring_addr) {
3589ed67cdcSSasha Levin 		.index = vq,
3599ed67cdcSSasha Levin 		.desc_user_addr = (u64)(unsigned long)queue->vring.desc,
3609ed67cdcSSasha Levin 		.avail_user_addr = (u64)(unsigned long)queue->vring.avail,
3619ed67cdcSSasha Levin 		.used_user_addr = (u64)(unsigned long)queue->vring.used,
3629ed67cdcSSasha Levin 	};
3639ed67cdcSSasha Levin 
3649ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr);
3659ed67cdcSSasha Levin 	if (r < 0)
3669ed67cdcSSasha Levin 		die_perror("VHOST_SET_VRING_ADDR failed");
3679ed67cdcSSasha Levin 
3689ed67cdcSSasha Levin 	return 0;
3699ed67cdcSSasha Levin }
3709ed67cdcSSasha Levin 
3719ed67cdcSSasha Levin static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi)
3729ed67cdcSSasha Levin {
3739ed67cdcSSasha Levin 	struct net_dev *ndev = dev;
3749ed67cdcSSasha Levin 	struct kvm_irqfd irq;
3759ed67cdcSSasha Levin 	struct vhost_vring_file file;
3769ed67cdcSSasha Levin 	int r;
3779ed67cdcSSasha Levin 
3789ed67cdcSSasha Levin 	if (ndev->vhost_fd == 0)
3799ed67cdcSSasha Levin 		return;
3809ed67cdcSSasha Levin 
3819ed67cdcSSasha Levin 	irq = (struct kvm_irqfd) {
3829ed67cdcSSasha Levin 		.gsi	= gsi,
3839ed67cdcSSasha Levin 		.fd	= eventfd(0, 0),
3849ed67cdcSSasha Levin 	};
3859ed67cdcSSasha Levin 	file = (struct vhost_vring_file) {
3869ed67cdcSSasha Levin 		.index	= vq,
3879ed67cdcSSasha Levin 		.fd	= irq.fd,
3889ed67cdcSSasha Levin 	};
3899ed67cdcSSasha Levin 
3909ed67cdcSSasha Levin 	r = ioctl(kvm->vm_fd, KVM_IRQFD, &irq);
3919ed67cdcSSasha Levin 	if (r < 0)
3929ed67cdcSSasha Levin 		die_perror("KVM_IRQFD failed");
3939ed67cdcSSasha Levin 
3949ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file);
3959ed67cdcSSasha Levin 	if (r < 0)
3969ed67cdcSSasha Levin 		die_perror("VHOST_SET_VRING_CALL failed");
3979ed67cdcSSasha Levin 	file.fd = ndev->tap_fd;
3989ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file);
3999ed67cdcSSasha Levin 	if (r != 0)
4009ed67cdcSSasha Levin 		die("VHOST_NET_SET_BACKEND failed %d", errno);
4019ed67cdcSSasha Levin 
4029ed67cdcSSasha Levin }
4039ed67cdcSSasha Levin 
4049ed67cdcSSasha Levin static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd)
4059ed67cdcSSasha Levin {
4069ed67cdcSSasha Levin 	struct net_dev *ndev = dev;
4079ed67cdcSSasha Levin 	struct vhost_vring_file file = {
4089ed67cdcSSasha Levin 		.index	= vq,
4099ed67cdcSSasha Levin 		.fd	= efd,
4109ed67cdcSSasha Levin 	};
4119ed67cdcSSasha Levin 	int r;
4129ed67cdcSSasha Levin 
4139ed67cdcSSasha Levin 	if (ndev->vhost_fd == 0)
4149ed67cdcSSasha Levin 		return;
4159ed67cdcSSasha Levin 
4169ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file);
4179ed67cdcSSasha Levin 	if (r < 0)
4189ed67cdcSSasha Levin 		die_perror("VHOST_SET_VRING_KICK failed");
41978a2a3e2SSasha Levin }
42078a2a3e2SSasha Levin 
42178a2a3e2SSasha Levin static int notify_vq(struct kvm *kvm, void *dev, u32 vq)
42278a2a3e2SSasha Levin {
4239a6d73f1SSasha Levin 	struct net_dev *ndev = dev;
4249a6d73f1SSasha Levin 
4259a6d73f1SSasha Levin 	virtio_net_handle_callback(kvm, ndev, vq);
42678a2a3e2SSasha Levin 
42778a2a3e2SSasha Levin 	return 0;
42878a2a3e2SSasha Levin }
42978a2a3e2SSasha Levin 
43078a2a3e2SSasha Levin static int get_pfn_vq(struct kvm *kvm, void *dev, u32 vq)
43178a2a3e2SSasha Levin {
43278a2a3e2SSasha Levin 	struct net_dev *ndev = dev;
43378a2a3e2SSasha Levin 
43478a2a3e2SSasha Levin 	return ndev->vqs[vq].pfn;
43578a2a3e2SSasha Levin }
43678a2a3e2SSasha Levin 
43778a2a3e2SSasha Levin static int get_size_vq(struct kvm *kvm, void *dev, u32 vq)
43878a2a3e2SSasha Levin {
43978a2a3e2SSasha Levin 	return VIRTIO_NET_QUEUE_SIZE;
44078a2a3e2SSasha Levin }
44178a2a3e2SSasha Levin 
4421c47ce69SSasha Levin static struct virtio_ops net_dev_virtio_ops = (struct virtio_ops) {
4431c47ce69SSasha Levin 	.set_config		= set_config,
4441c47ce69SSasha Levin 	.get_config		= get_config,
4451c47ce69SSasha Levin 	.get_host_features	= get_host_features,
4461c47ce69SSasha Levin 	.set_guest_features	= set_guest_features,
4471c47ce69SSasha Levin 	.init_vq		= init_vq,
4481c47ce69SSasha Levin 	.notify_vq		= notify_vq,
4491c47ce69SSasha Levin 	.get_pfn_vq		= get_pfn_vq,
4501c47ce69SSasha Levin 	.get_size_vq		= get_size_vq,
4519ed67cdcSSasha Levin 	.notify_vq_gsi		= notify_vq_gsi,
4529ed67cdcSSasha Levin 	.notify_vq_eventfd	= notify_vq_eventfd,
4531c47ce69SSasha Levin };
4541c47ce69SSasha Levin 
4559ed67cdcSSasha Levin static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev)
4569ed67cdcSSasha Levin {
45792c1f37bSSasha Levin 	u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX;
4589ed67cdcSSasha Levin 	struct vhost_memory *mem;
4599ed67cdcSSasha Levin 	int r;
4609ed67cdcSSasha Levin 
4619ed67cdcSSasha Levin 	ndev->vhost_fd = open("/dev/vhost-net", O_RDWR);
4629ed67cdcSSasha Levin 	if (ndev->vhost_fd < 0)
4639ed67cdcSSasha Levin 		die_perror("Failed openning vhost-net device");
4649ed67cdcSSasha Levin 
4659ed67cdcSSasha Levin 	mem = malloc(sizeof(*mem) + sizeof(struct vhost_memory_region));
4669ed67cdcSSasha Levin 	if (mem == NULL)
4679ed67cdcSSasha Levin 		die("Failed allocating memory for vhost memory map");
4689ed67cdcSSasha Levin 
4699ed67cdcSSasha Levin 	mem->nregions = 1;
4709ed67cdcSSasha Levin 	mem->regions[0] = (struct vhost_memory_region) {
4719ed67cdcSSasha Levin 		.guest_phys_addr	= 0,
4729ed67cdcSSasha Levin 		.memory_size		= kvm->ram_size,
473b4108023SAsias He 		.userspace_addr		= (unsigned long)kvm->ram_start,
4749ed67cdcSSasha Levin 	};
4759ed67cdcSSasha Levin 
4769ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER);
4779ed67cdcSSasha Levin 	if (r != 0)
4789ed67cdcSSasha Levin 		die_perror("VHOST_SET_OWNER failed");
4799ed67cdcSSasha Levin 
4809ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features);
4819ed67cdcSSasha Levin 	if (r != 0)
4829ed67cdcSSasha Levin 		die_perror("VHOST_SET_FEATURES failed");
4839ed67cdcSSasha Levin 	r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem);
4849ed67cdcSSasha Levin 	if (r != 0)
4859ed67cdcSSasha Levin 		die_perror("VHOST_SET_MEM_TABLE failed");
4869ed67cdcSSasha Levin 	free(mem);
4879ed67cdcSSasha Levin }
4889ed67cdcSSasha Levin 
4899a6d73f1SSasha Levin void virtio_net__init(const struct virtio_net_params *params)
4904f56d42cSAsias He {
491b5ee1ea7SAsias He 	int i;
4929a6d73f1SSasha Levin 	struct net_dev *ndev;
4939a6d73f1SSasha Levin 
4949a6d73f1SSasha Levin 	if (!params)
4959a6d73f1SSasha Levin 		return;
4969a6d73f1SSasha Levin 
4979a6d73f1SSasha Levin 	ndev = calloc(1, sizeof(struct net_dev));
4989a6d73f1SSasha Levin 	if (ndev == NULL)
4999a6d73f1SSasha Levin 		die("Failed allocating ndev");
5009a6d73f1SSasha Levin 
5019a6d73f1SSasha Levin 	list_add_tail(&ndev->list, &ndevs);
5029a6d73f1SSasha Levin 
5039a6d73f1SSasha Levin 	ndev->kvm = params->kvm;
5049a6d73f1SSasha Levin 
5059a6d73f1SSasha Levin 	mutex_init(&ndev->mutex);
5069a6d73f1SSasha Levin 	ndev->config.status = VIRTIO_NET_S_LINK_UP;
5072449f6e3SSasha Levin 
5080c54698eSAsias He 	for (i = 0 ; i < 6 ; i++) {
5099a6d73f1SSasha Levin 		ndev->config.mac[i]		= params->guest_mac[i];
5109a6d73f1SSasha Levin 		ndev->info.guest_mac.addr[i]	= params->guest_mac[i];
5119a6d73f1SSasha Levin 		ndev->info.host_mac.addr[i]	= params->host_mac[i];
5120c54698eSAsias He 	}
513f715177dSAsias He 
5149a6d73f1SSasha Levin 	ndev->mode = params->mode;
5159a6d73f1SSasha Levin 	if (ndev->mode == NET_MODE_TAP) {
5169a6d73f1SSasha Levin 		if (!virtio_net__tap_init(params, ndev))
5179a6d73f1SSasha Levin 			die_perror("You have requested a TAP device, but creation of one has"
5189a6d73f1SSasha Levin 					"failed because:");
5199a6d73f1SSasha Levin 		ndev->ops = &tap_ops;
520b4fdde6dSAsias He 	} else {
5219a6d73f1SSasha Levin 		ndev->info.host_ip		= ntohl(inet_addr(params->host_ip));
5229a6d73f1SSasha Levin 		ndev->info.guest_ip		= ntohl(inet_addr(params->guest_ip));
5239a6d73f1SSasha Levin 		ndev->info.guest_netmask	= ntohl(inet_addr("255.255.255.0"));
5249a6d73f1SSasha Levin 		ndev->info.buf_nr		= 20,
5259a6d73f1SSasha Levin 		uip_init(&ndev->info);
5269a6d73f1SSasha Levin 		ndev->ops = &uip_ops;
527b4fdde6dSAsias He 	}
528b5ee1ea7SAsias He 
52969205aa1SAsias He 	if (params->trans && strcmp(params->trans, "mmio") == 0)
53069205aa1SAsias He 		virtio_init(kvm, ndev, &ndev->vdev, &net_dev_virtio_ops,
53169205aa1SAsias He 			    VIRTIO_MMIO, PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET);
53269205aa1SAsias He 	else
53302eca50cSAsias He 		virtio_init(kvm, ndev, &ndev->vdev, &net_dev_virtio_ops,
53402eca50cSAsias He 			    VIRTIO_PCI, PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET);
53527ab67f5SSasha Levin 
5369ed67cdcSSasha Levin 	if (params->vhost)
5379ed67cdcSSasha Levin 		virtio_net__vhost_init(params->kvm, ndev);
5389ed67cdcSSasha Levin 	else
5399a6d73f1SSasha Levin 		virtio_net__io_thread_init(params->kvm, ndev);
540cb83de6fSSasha Levin 
541312c62d1SSasha Levin 	if (compat_id != -1)
542312c62d1SSasha Levin 		compat_id = compat__add_message("virtio-net device was not detected",
543cb83de6fSSasha Levin 						"While you have requested a virtio-net device, "
544fc835ab3SSasha Levin 						"the guest kernel did not initialize it.\n"
545fc835ab3SSasha Levin 						"Please make sure that the guest kernel was "
546fc835ab3SSasha Levin 						"compiled with CONFIG_VIRTIO_NET=y enabled "
547fc835ab3SSasha Levin 						"in its .config");
5484f56d42cSAsias He }
549