xref: /kvmtool/virtio/net.c (revision 05755b29e63a9c73ba29c8b7c632570727d7afe7)
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 u32 get_host_features(struct kvm *kvm, void *dev)
484 {
485 	u32 features;
486 	struct net_dev *ndev = dev;
487 
488 	features = 1UL << VIRTIO_NET_F_MAC
489 		| 1UL << VIRTIO_NET_F_CSUM
490 		| 1UL << VIRTIO_NET_F_HOST_TSO4
491 		| 1UL << VIRTIO_NET_F_HOST_TSO6
492 		| 1UL << VIRTIO_NET_F_GUEST_TSO4
493 		| 1UL << VIRTIO_NET_F_GUEST_TSO6
494 		| 1UL << VIRTIO_RING_F_EVENT_IDX
495 		| 1UL << VIRTIO_RING_F_INDIRECT_DESC
496 		| 1UL << VIRTIO_NET_F_CTRL_VQ
497 		| 1UL << VIRTIO_NET_F_MRG_RXBUF
498 		| 1UL << (ndev->queue_pairs > 1 ? VIRTIO_NET_F_MQ : 0);
499 
500 	/*
501 	 * The UFO feature for host and guest only can be enabled when the
502 	 * kernel has TAP UFO support.
503 	 */
504 	if (ndev->tap_ufo)
505 		features |= (1UL << VIRTIO_NET_F_HOST_UFO
506 				| 1UL << VIRTIO_NET_F_GUEST_UFO);
507 
508 	return features;
509 }
510 
511 static int virtio_net__vhost_set_features(struct net_dev *ndev)
512 {
513 	u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX;
514 	u64 vhost_features;
515 
516 	if (ioctl(ndev->vhost_fd, VHOST_GET_FEATURES, &vhost_features) != 0)
517 		die_perror("VHOST_GET_FEATURES failed");
518 
519 	/* make sure both side support mergable rx buffers */
520 	if (vhost_features & 1UL << VIRTIO_NET_F_MRG_RXBUF &&
521 			has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF))
522 		features |= 1UL << VIRTIO_NET_F_MRG_RXBUF;
523 
524 	return ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features);
525 }
526 
527 static void set_guest_features(struct kvm *kvm, void *dev, u32 features)
528 {
529 	struct net_dev *ndev = dev;
530 	struct virtio_net_config *conf = &ndev->config;
531 
532 	ndev->features = features;
533 
534 	conf->status = virtio_host_to_guest_u16(&ndev->vdev, conf->status);
535 	conf->max_virtqueue_pairs = virtio_host_to_guest_u16(&ndev->vdev,
536 							     conf->max_virtqueue_pairs);
537 }
538 
539 static void virtio_net_start(struct net_dev *ndev)
540 {
541 	if (ndev->mode == NET_MODE_TAP) {
542 		if (!virtio_net__tap_init(ndev))
543 			die_perror("TAP device initialized failed because");
544 
545 		if (ndev->vhost_fd &&
546 				virtio_net__vhost_set_features(ndev) != 0)
547 			die_perror("VHOST_SET_FEATURES failed");
548 	} else {
549 		ndev->info.vnet_hdr_len = has_virtio_feature(ndev, VIRTIO_NET_F_MRG_RXBUF) ?
550 						sizeof(struct virtio_net_hdr_mrg_rxbuf) :
551 						sizeof(struct virtio_net_hdr);
552 		uip_init(&ndev->info);
553 	}
554 }
555 
556 static void virtio_net_stop(struct net_dev *ndev)
557 {
558 	/* Undo whatever start() did */
559 	if (ndev->mode == NET_MODE_TAP)
560 		virtio_net__tap_exit(ndev);
561 	else
562 		uip_exit(&ndev->info);
563 }
564 
565 static void notify_status(struct kvm *kvm, void *dev, u32 status)
566 {
567 	if (status & VIRTIO__STATUS_START)
568 		virtio_net_start(dev);
569 	else if (status & VIRTIO__STATUS_STOP)
570 		virtio_net_stop(dev);
571 }
572 
573 static bool is_ctrl_vq(struct net_dev *ndev, u32 vq)
574 {
575 	return vq == (u32)(ndev->queue_pairs * 2);
576 }
577 
578 static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 page_size, u32 align,
579 		   u32 pfn)
580 {
581 	struct vhost_vring_state state = { .index = vq };
582 	struct net_dev_queue *net_queue;
583 	struct vhost_vring_addr addr;
584 	struct net_dev *ndev = dev;
585 	struct virt_queue *queue;
586 	void *p;
587 	int r;
588 
589 	compat__remove_message(compat_id);
590 
591 	net_queue	= &ndev->queues[vq];
592 	net_queue->id	= vq;
593 	net_queue->ndev	= ndev;
594 	queue		= &net_queue->vq;
595 	queue->pfn	= pfn;
596 	p		= virtio_get_vq(kvm, queue->pfn, page_size);
597 
598 	vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, align);
599 	virtio_init_device_vq(&ndev->vdev, queue);
600 
601 	mutex_init(&net_queue->lock);
602 	pthread_cond_init(&net_queue->cond, NULL);
603 	if (is_ctrl_vq(ndev, vq)) {
604 		pthread_create(&net_queue->thread, NULL, virtio_net_ctrl_thread,
605 			       net_queue);
606 
607 		return 0;
608 	} else if (ndev->vhost_fd == 0 ) {
609 		if (vq & 1)
610 			pthread_create(&net_queue->thread, NULL,
611 				       virtio_net_tx_thread, net_queue);
612 		else
613 			pthread_create(&net_queue->thread, NULL,
614 				       virtio_net_rx_thread, net_queue);
615 
616 		return 0;
617 	}
618 
619 	if (queue->endian != VIRTIO_ENDIAN_HOST)
620 		die_perror("VHOST requires the same endianness in guest and host");
621 
622 	state.num = queue->vring.num;
623 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state);
624 	if (r < 0)
625 		die_perror("VHOST_SET_VRING_NUM failed");
626 	state.num = 0;
627 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state);
628 	if (r < 0)
629 		die_perror("VHOST_SET_VRING_BASE failed");
630 
631 	addr = (struct vhost_vring_addr) {
632 		.index = vq,
633 		.desc_user_addr = (u64)(unsigned long)queue->vring.desc,
634 		.avail_user_addr = (u64)(unsigned long)queue->vring.avail,
635 		.used_user_addr = (u64)(unsigned long)queue->vring.used,
636 	};
637 
638 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr);
639 	if (r < 0)
640 		die_perror("VHOST_SET_VRING_ADDR failed");
641 
642 	return 0;
643 }
644 
645 static void exit_vq(struct kvm *kvm, void *dev, u32 vq)
646 {
647 	struct net_dev *ndev = dev;
648 	struct net_dev_queue *queue = &ndev->queues[vq];
649 
650 	if (!is_ctrl_vq(ndev, vq) && queue->gsi) {
651 		irq__del_irqfd(kvm, queue->gsi, queue->irqfd);
652 		close(queue->irqfd);
653 		queue->gsi = queue->irqfd = 0;
654 	}
655 
656 	/*
657 	 * TODO: vhost reset owner. It's the only way to cleanly stop vhost, but
658 	 * we can't restart it at the moment.
659 	 */
660 	if (ndev->vhost_fd && !is_ctrl_vq(ndev, vq)) {
661 		pr_warning("Cannot reset VHOST queue");
662 		ioctl(ndev->vhost_fd, VHOST_RESET_OWNER);
663 		return;
664 	}
665 
666 	/*
667 	 * Threads are waiting on cancellation points (readv or
668 	 * pthread_cond_wait) and should stop gracefully.
669 	 */
670 	pthread_cancel(queue->thread);
671 	pthread_join(queue->thread, NULL);
672 }
673 
674 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi)
675 {
676 	struct net_dev *ndev = dev;
677 	struct net_dev_queue *queue = &ndev->queues[vq];
678 	struct vhost_vring_file file;
679 	int r;
680 
681 	if (ndev->vhost_fd == 0)
682 		return;
683 
684 	file = (struct vhost_vring_file) {
685 		.index	= vq,
686 		.fd	= eventfd(0, 0),
687 	};
688 
689 	r = irq__add_irqfd(kvm, gsi, file.fd, -1);
690 	if (r < 0)
691 		die_perror("KVM_IRQFD failed");
692 
693 	queue->irqfd = file.fd;
694 	queue->gsi = gsi;
695 
696 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file);
697 	if (r < 0)
698 		die_perror("VHOST_SET_VRING_CALL failed");
699 	file.fd = ndev->tap_fd;
700 	r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file);
701 	if (r != 0)
702 		die("VHOST_NET_SET_BACKEND failed %d", errno);
703 
704 }
705 
706 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd)
707 {
708 	struct net_dev *ndev = dev;
709 	struct vhost_vring_file file = {
710 		.index	= vq,
711 		.fd	= efd,
712 	};
713 	int r;
714 
715 	if (ndev->vhost_fd == 0 || is_ctrl_vq(ndev, vq))
716 		return;
717 
718 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file);
719 	if (r < 0)
720 		die_perror("VHOST_SET_VRING_KICK failed");
721 }
722 
723 static int notify_vq(struct kvm *kvm, void *dev, u32 vq)
724 {
725 	struct net_dev *ndev = dev;
726 
727 	virtio_net_handle_callback(kvm, ndev, vq);
728 
729 	return 0;
730 }
731 
732 static struct virt_queue *get_vq(struct kvm *kvm, void *dev, u32 vq)
733 {
734 	struct net_dev *ndev = dev;
735 
736 	return &ndev->queues[vq].vq;
737 }
738 
739 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq)
740 {
741 	/* FIXME: dynamic */
742 	return VIRTIO_NET_QUEUE_SIZE;
743 }
744 
745 static int set_size_vq(struct kvm *kvm, void *dev, u32 vq, int size)
746 {
747 	/* FIXME: dynamic */
748 	return size;
749 }
750 
751 static int get_vq_count(struct kvm *kvm, void *dev)
752 {
753 	struct net_dev *ndev = dev;
754 
755 	return ndev->queue_pairs * 2 + 1;
756 }
757 
758 static struct virtio_ops net_dev_virtio_ops = {
759 	.get_config		= get_config,
760 	.get_host_features	= get_host_features,
761 	.set_guest_features	= set_guest_features,
762 	.get_vq_count		= get_vq_count,
763 	.init_vq		= init_vq,
764 	.exit_vq		= exit_vq,
765 	.get_vq			= get_vq,
766 	.get_size_vq		= get_size_vq,
767 	.set_size_vq		= set_size_vq,
768 	.notify_vq		= notify_vq,
769 	.notify_vq_gsi		= notify_vq_gsi,
770 	.notify_vq_eventfd	= notify_vq_eventfd,
771 	.notify_status		= notify_status,
772 };
773 
774 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev)
775 {
776 	struct kvm_mem_bank *bank;
777 	struct vhost_memory *mem;
778 	int r, i;
779 
780 	ndev->vhost_fd = open("/dev/vhost-net", O_RDWR);
781 	if (ndev->vhost_fd < 0)
782 		die_perror("Failed openning vhost-net device");
783 
784 	mem = calloc(1, sizeof(*mem) + kvm->mem_slots * sizeof(struct vhost_memory_region));
785 	if (mem == NULL)
786 		die("Failed allocating memory for vhost memory map");
787 
788 	i = 0;
789 	list_for_each_entry(bank, &kvm->mem_banks, list) {
790 		mem->regions[i] = (struct vhost_memory_region) {
791 			.guest_phys_addr = bank->guest_phys_addr,
792 			.memory_size	 = bank->size,
793 			.userspace_addr	 = (unsigned long)bank->host_addr,
794 		};
795 		i++;
796 	}
797 	mem->nregions = i;
798 
799 	r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER);
800 	if (r != 0)
801 		die_perror("VHOST_SET_OWNER failed");
802 
803 	r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem);
804 	if (r != 0)
805 		die_perror("VHOST_SET_MEM_TABLE failed");
806 
807 	ndev->vdev.use_vhost = true;
808 
809 	free(mem);
810 }
811 
812 static inline void str_to_mac(const char *str, char *mac)
813 {
814 	sscanf(str, "%hhx:%hhx:%hhx:%hhx:%hhx:%hhx",
815 		mac, mac+1, mac+2, mac+3, mac+4, mac+5);
816 }
817 static int set_net_param(struct kvm *kvm, struct virtio_net_params *p,
818 			const char *param, const char *val)
819 {
820 	if (strcmp(param, "guest_mac") == 0) {
821 		str_to_mac(val, p->guest_mac);
822 	} else if (strcmp(param, "mode") == 0) {
823 		if (!strncmp(val, "user", 4)) {
824 			int i;
825 
826 			for (i = 0; i < kvm->cfg.num_net_devices; i++)
827 				if (kvm->cfg.net_params[i].mode == NET_MODE_USER)
828 					die("Only one usermode network device allowed at a time");
829 			p->mode = NET_MODE_USER;
830 		} else if (!strncmp(val, "tap", 3)) {
831 			p->mode = NET_MODE_TAP;
832 		} else if (!strncmp(val, "none", 4)) {
833 			kvm->cfg.no_net = 1;
834 			return -1;
835 		} else
836 			die("Unknown network mode %s, please use user, tap or none", kvm->cfg.network);
837 	} else if (strcmp(param, "script") == 0) {
838 		p->script = strdup(val);
839 	} else if (strcmp(param, "downscript") == 0) {
840 		p->downscript = strdup(val);
841 	} else if (strcmp(param, "guest_ip") == 0) {
842 		p->guest_ip = strdup(val);
843 	} else if (strcmp(param, "host_ip") == 0) {
844 		p->host_ip = strdup(val);
845 	} else if (strcmp(param, "trans") == 0) {
846 		p->trans = strdup(val);
847 	} else if (strcmp(param, "tapif") == 0) {
848 		p->tapif = strdup(val);
849 	} else if (strcmp(param, "vhost") == 0) {
850 		p->vhost = atoi(val);
851 	} else if (strcmp(param, "fd") == 0) {
852 		p->fd = atoi(val);
853 	} else if (strcmp(param, "mq") == 0) {
854 		p->mq = atoi(val);
855 	} else
856 		die("Unknown network parameter %s", param);
857 
858 	return 0;
859 }
860 
861 int netdev_parser(const struct option *opt, const char *arg, int unset)
862 {
863 	struct virtio_net_params p;
864 	char *buf = NULL, *cmd = NULL, *cur = NULL;
865 	bool on_cmd = true;
866 	struct kvm *kvm = opt->ptr;
867 
868 	if (arg) {
869 		buf = strdup(arg);
870 		if (buf == NULL)
871 			die("Failed allocating new net buffer");
872 		cur = strtok(buf, ",=");
873 	}
874 
875 	p = (struct virtio_net_params) {
876 		.guest_ip	= DEFAULT_GUEST_ADDR,
877 		.host_ip	= DEFAULT_HOST_ADDR,
878 		.script		= DEFAULT_SCRIPT,
879 		.downscript	= DEFAULT_SCRIPT,
880 		.mode		= NET_MODE_TAP,
881 	};
882 
883 	str_to_mac(DEFAULT_GUEST_MAC, p.guest_mac);
884 	p.guest_mac[5] += kvm->cfg.num_net_devices;
885 
886 	while (cur) {
887 		if (on_cmd) {
888 			cmd = cur;
889 		} else {
890 			if (set_net_param(kvm, &p, cmd, cur) < 0)
891 				goto done;
892 		}
893 		on_cmd = !on_cmd;
894 
895 		cur = strtok(NULL, ",=");
896 	};
897 
898 	kvm->cfg.num_net_devices++;
899 
900 	kvm->cfg.net_params = realloc(kvm->cfg.net_params, kvm->cfg.num_net_devices * sizeof(*kvm->cfg.net_params));
901 	if (kvm->cfg.net_params == NULL)
902 		die("Failed adding new network device");
903 
904 	kvm->cfg.net_params[kvm->cfg.num_net_devices - 1] = p;
905 
906 done:
907 	free(buf);
908 	return 0;
909 }
910 
911 static int virtio_net__init_one(struct virtio_net_params *params)
912 {
913 	int i, err;
914 	struct net_dev *ndev;
915 	struct virtio_ops *ops;
916 	enum virtio_trans trans = VIRTIO_DEFAULT_TRANS(params->kvm);
917 
918 	ndev = calloc(1, sizeof(struct net_dev));
919 	if (ndev == NULL)
920 		return -ENOMEM;
921 
922 	ops = malloc(sizeof(*ops));
923 	if (ops == NULL) {
924 		err = -ENOMEM;
925 		goto err_free_ndev;
926 	}
927 
928 	list_add_tail(&ndev->list, &ndevs);
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 	virtio_init(params->kvm, ndev, &ndev->vdev, ops, trans,
973 		    PCI_DEVICE_ID_VIRTIO_NET, VIRTIO_ID_NET, PCI_CLASS_NET);
974 
975 	if (params->vhost)
976 		virtio_net__vhost_init(params->kvm, ndev);
977 
978 	if (compat_id == -1)
979 		compat_id = virtio_compat_add_message("virtio-net", "CONFIG_VIRTIO_NET");
980 
981 	return 0;
982 
983 err_free_ndev:
984 	free(ndev);
985 	return err;
986 }
987 
988 int virtio_net__init(struct kvm *kvm)
989 {
990 	int i;
991 
992 	for (i = 0; i < kvm->cfg.num_net_devices; i++) {
993 		kvm->cfg.net_params[i].kvm = kvm;
994 		virtio_net__init_one(&kvm->cfg.net_params[i]);
995 	}
996 
997 	if (kvm->cfg.num_net_devices == 0 && kvm->cfg.no_net == 0) {
998 		static struct virtio_net_params net_params;
999 
1000 		net_params = (struct virtio_net_params) {
1001 			.guest_ip	= kvm->cfg.guest_ip,
1002 			.host_ip	= kvm->cfg.host_ip,
1003 			.kvm		= kvm,
1004 			.script		= kvm->cfg.script,
1005 			.mode		= NET_MODE_USER,
1006 		};
1007 		str_to_mac(kvm->cfg.guest_mac, net_params.guest_mac);
1008 		str_to_mac(kvm->cfg.host_mac, net_params.host_mac);
1009 
1010 		virtio_net__init_one(&net_params);
1011 	}
1012 
1013 	return 0;
1014 }
1015 virtio_dev_init(virtio_net__init);
1016 
1017 int virtio_net__exit(struct kvm *kvm)
1018 {
1019 	struct virtio_net_params *params;
1020 	struct net_dev *ndev;
1021 	struct list_head *ptr;
1022 
1023 	list_for_each(ptr, &ndevs) {
1024 		ndev = list_entry(ptr, struct net_dev, list);
1025 		params = ndev->params;
1026 		/* Cleanup any tap device which attached to bridge */
1027 		if (ndev->mode == NET_MODE_TAP &&
1028 		    strcmp(params->downscript, "none"))
1029 			virtio_net_exec_script(params->downscript, ndev->tap_name);
1030 	}
1031 	return 0;
1032 }
1033 virtio_dev_exit(virtio_net__exit);
1034