xref: /kvmtool/virtio/net.c (revision 92c1f37b753eea0b73ddc7eee56eb737bb629b00) 
1 #include "kvm/virtio-pci-dev.h"
2 #include "kvm/virtio-net.h"
3 #include "kvm/virtio.h"
4 #include "kvm/types.h"
5 #include "kvm/mutex.h"
6 #include "kvm/util.h"
7 #include "kvm/kvm.h"
8 #include "kvm/irq.h"
9 #include "kvm/uip.h"
10 #include "kvm/guest_compat.h"
11 #include "kvm/virtio-trans.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 <assert.h>
23 #include <fcntl.h>
24 
25 #include <sys/socket.h>
26 #include <sys/ioctl.h>
27 #include <sys/types.h>
28 #include <sys/wait.h>
29 #include <sys/eventfd.h>
30 
31 #define VIRTIO_NET_QUEUE_SIZE		128
32 #define VIRTIO_NET_NUM_QUEUES		2
33 #define VIRTIO_NET_RX_QUEUE		0
34 #define VIRTIO_NET_TX_QUEUE		1
35 
36 struct net_dev;
37 
38 extern struct kvm *kvm;
39 
40 struct net_dev_operations {
41 	int (*rx)(struct iovec *iov, u16 in, struct net_dev *ndev);
42 	int (*tx)(struct iovec *iov, u16 in, struct net_dev *ndev);
43 };
44 
45 struct net_dev {
46 	pthread_mutex_t			mutex;
47 	struct virtio_trans		vtrans;
48 	struct list_head		list;
49 
50 	struct virt_queue		vqs[VIRTIO_NET_NUM_QUEUES];
51 	struct virtio_net_config	config;
52 	u32				features;
53 
54 	pthread_t			io_rx_thread;
55 	pthread_mutex_t			io_rx_lock;
56 	pthread_cond_t			io_rx_cond;
57 
58 	pthread_t			io_tx_thread;
59 	pthread_mutex_t			io_tx_lock;
60 	pthread_cond_t			io_tx_cond;
61 
62 	int				vhost_fd;
63 	int				tap_fd;
64 	char				tap_name[IFNAMSIZ];
65 
66 	int				mode;
67 
68 	struct uip_info			info;
69 	struct net_dev_operations	*ops;
70 	struct kvm			*kvm;
71 };
72 
73 static LIST_HEAD(ndevs);
74 static int compat_id = -1;
75 
76 static void *virtio_net_rx_thread(void *p)
77 {
78 	struct iovec iov[VIRTIO_NET_QUEUE_SIZE];
79 	struct virt_queue *vq;
80 	struct kvm *kvm;
81 	struct net_dev *ndev = p;
82 	u16 out, in;
83 	u16 head;
84 	int len;
85 
86 	kvm	= ndev->kvm;
87 	vq	= &ndev->vqs[VIRTIO_NET_RX_QUEUE];
88 
89 	while (1) {
90 
91 		mutex_lock(&ndev->io_rx_lock);
92 		if (!virt_queue__available(vq))
93 			pthread_cond_wait(&ndev->io_rx_cond, &ndev->io_rx_lock);
94 		mutex_unlock(&ndev->io_rx_lock);
95 
96 		while (virt_queue__available(vq)) {
97 
98 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
99 
100 			len = ndev->ops->rx(iov, in, ndev);
101 
102 			virt_queue__set_used_elem(vq, head, len);
103 
104 			/* We should interrupt guest right now, otherwise latency is huge. */
105 			if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_RX_QUEUE]))
106 				ndev->vtrans.trans_ops->signal_vq(kvm, &ndev->vtrans,
107 								VIRTIO_NET_RX_QUEUE);
108 		}
109 
110 	}
111 
112 	pthread_exit(NULL);
113 	return NULL;
114 
115 }
116 
117 static void *virtio_net_tx_thread(void *p)
118 {
119 	struct iovec iov[VIRTIO_NET_QUEUE_SIZE];
120 	struct virt_queue *vq;
121 	struct kvm *kvm;
122 	struct net_dev *ndev = p;
123 	u16 out, in;
124 	u16 head;
125 	int len;
126 
127 	kvm	= ndev->kvm;
128 	vq	= &ndev->vqs[VIRTIO_NET_TX_QUEUE];
129 
130 	while (1) {
131 		mutex_lock(&ndev->io_tx_lock);
132 		if (!virt_queue__available(vq))
133 			pthread_cond_wait(&ndev->io_tx_cond, &ndev->io_tx_lock);
134 		mutex_unlock(&ndev->io_tx_lock);
135 
136 		while (virt_queue__available(vq)) {
137 
138 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
139 
140 			len = ndev->ops->tx(iov, out, ndev);
141 
142 			virt_queue__set_used_elem(vq, head, len);
143 		}
144 
145 		if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_TX_QUEUE]))
146 			ndev->vtrans.trans_ops->signal_vq(kvm, &ndev->vtrans, VIRTIO_NET_TX_QUEUE);
147 	}
148 
149 	pthread_exit(NULL);
150 
151 	return NULL;
152 
153 }
154 
155 static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue)
156 {
157 	switch (queue) {
158 	case VIRTIO_NET_TX_QUEUE:
159 		mutex_lock(&ndev->io_tx_lock);
160 		pthread_cond_signal(&ndev->io_tx_cond);
161 		mutex_unlock(&ndev->io_tx_lock);
162 		break;
163 	case VIRTIO_NET_RX_QUEUE:
164 		mutex_lock(&ndev->io_rx_lock);
165 		pthread_cond_signal(&ndev->io_rx_cond);
166 		mutex_unlock(&ndev->io_rx_lock);
167 		break;
168 	default:
169 		pr_warning("Unknown queue index %u", queue);
170 	}
171 }
172 
173 static bool virtio_net__tap_init(const struct virtio_net_params *params,
174 					struct net_dev *ndev)
175 {
176 	int sock = socket(AF_INET, SOCK_STREAM, 0);
177 	int pid, status, offload, hdr_len;
178 	struct sockaddr_in sin = {0};
179 	struct ifreq ifr;
180 
181 	ndev->tap_fd = open("/dev/net/tun", O_RDWR);
182 	if (ndev->tap_fd < 0) {
183 		pr_warning("Unable to open /dev/net/tun");
184 		goto fail;
185 	}
186 
187 	memset(&ifr, 0, sizeof(ifr));
188 	ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
189 	if (ioctl(ndev->tap_fd, TUNSETIFF, &ifr) < 0) {
190 		pr_warning("Config tap device error. Are you root?");
191 		goto fail;
192 	}
193 
194 	strncpy(ndev->tap_name, ifr.ifr_name, sizeof(ndev->tap_name));
195 
196 	if (ioctl(ndev->tap_fd, TUNSETNOCSUM, 1) < 0) {
197 		pr_warning("Config tap device TUNSETNOCSUM error");
198 		goto fail;
199 	}
200 
201 	hdr_len = sizeof(struct virtio_net_hdr);
202 	if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0)
203 		pr_warning("Config tap device TUNSETVNETHDRSZ error");
204 
205 	offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO;
206 	if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) {
207 		pr_warning("Config tap device TUNSETOFFLOAD error");
208 		goto fail;
209 	}
210 
211 	if (strcmp(params->script, "none")) {
212 		pid = fork();
213 		if (pid == 0) {
214 			execl(params->script, params->script, ndev->tap_name, NULL);
215 			_exit(1);
216 		} else {
217 			waitpid(pid, &status, 0);
218 			if (WIFEXITED(status) && WEXITSTATUS(status) != 0) {
219 				pr_warning("Fail to setup tap by %s", params->script);
220 				goto fail;
221 			}
222 		}
223 	} else {
224 		memset(&ifr, 0, sizeof(ifr));
225 		strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
226 		sin.sin_addr.s_addr = inet_addr(params->host_ip);
227 		memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr));
228 		ifr.ifr_addr.sa_family = AF_INET;
229 		if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) {
230 			pr_warning("Could not set ip address on tap device");
231 			goto fail;
232 		}
233 	}
234 
235 	memset(&ifr, 0, sizeof(ifr));
236 	strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
237 	ioctl(sock, SIOCGIFFLAGS, &ifr);
238 	ifr.ifr_flags |= IFF_UP | IFF_RUNNING;
239 	if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0)
240 		pr_warning("Could not bring tap device up");
241 
242 	close(sock);
243 
244 	return 1;
245 
246 fail:
247 	if (sock >= 0)
248 		close(sock);
249 	if (ndev->tap_fd >= 0)
250 		close(ndev->tap_fd);
251 
252 	return 0;
253 }
254 
255 static void virtio_net__io_thread_init(struct kvm *kvm, struct net_dev *ndev)
256 {
257 	pthread_mutex_init(&ndev->io_tx_lock, NULL);
258 	pthread_mutex_init(&ndev->io_rx_lock, NULL);
259 
260 	pthread_cond_init(&ndev->io_tx_cond, NULL);
261 	pthread_cond_init(&ndev->io_rx_cond, NULL);
262 
263 	pthread_create(&ndev->io_tx_thread, NULL, virtio_net_tx_thread, ndev);
264 	pthread_create(&ndev->io_rx_thread, NULL, virtio_net_rx_thread, ndev);
265 }
266 
267 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
268 {
269 	return writev(ndev->tap_fd, iov, out);
270 }
271 
272 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
273 {
274 	return readv(ndev->tap_fd, iov, in);
275 }
276 
277 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
278 {
279 	return uip_tx(iov, out, &ndev->info);
280 }
281 
282 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
283 {
284 	return uip_rx(iov, in, &ndev->info);
285 }
286 
287 static struct net_dev_operations tap_ops = {
288 	.rx	= tap_ops_rx,
289 	.tx	= tap_ops_tx,
290 };
291 
292 static struct net_dev_operations uip_ops = {
293 	.rx	= uip_ops_rx,
294 	.tx	= uip_ops_tx,
295 };
296 
297 static void set_config(struct kvm *kvm, void *dev, u8 data, u32 offset)
298 {
299 	struct net_dev *ndev = dev;
300 
301 	((u8 *)(&ndev->config))[offset] = data;
302 }
303 
304 static u8 get_config(struct kvm *kvm, void *dev, u32 offset)
305 {
306 	struct net_dev *ndev = dev;
307 
308 	return ((u8 *)(&ndev->config))[offset];
309 }
310 
311 static u32 get_host_features(struct kvm *kvm, void *dev)
312 {
313 	return 1UL << VIRTIO_NET_F_MAC
314 		| 1UL << VIRTIO_NET_F_CSUM
315 		| 1UL << VIRTIO_NET_F_HOST_UFO
316 		| 1UL << VIRTIO_NET_F_HOST_TSO4
317 		| 1UL << VIRTIO_NET_F_HOST_TSO6
318 		| 1UL << VIRTIO_NET_F_GUEST_UFO
319 		| 1UL << VIRTIO_NET_F_GUEST_TSO4
320 		| 1UL << VIRTIO_NET_F_GUEST_TSO6
321 		| 1UL << VIRTIO_RING_F_EVENT_IDX;
322 }
323 
324 static void set_guest_features(struct kvm *kvm, void *dev, u32 features)
325 {
326 	struct net_dev *ndev = dev;
327 
328 	ndev->features = features;
329 }
330 
331 static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 pfn)
332 {
333 	struct vhost_vring_state state = { .index = vq };
334 	struct vhost_vring_addr addr;
335 	struct net_dev *ndev = dev;
336 	struct virt_queue *queue;
337 	void *p;
338 	int r;
339 
340 	compat__remove_message(compat_id);
341 
342 	queue			= &ndev->vqs[vq];
343 	queue->pfn		= pfn;
344 	p			= guest_pfn_to_host(kvm, queue->pfn);
345 
346 	vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, VIRTIO_PCI_VRING_ALIGN);
347 
348 	if (ndev->vhost_fd == 0)
349 		return 0;
350 
351 	state.num = queue->vring.num;
352 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state);
353 	if (r < 0)
354 		die_perror("VHOST_SET_VRING_NUM failed");
355 	state.num = 0;
356 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state);
357 	if (r < 0)
358 		die_perror("VHOST_SET_VRING_BASE failed");
359 
360 	addr = (struct vhost_vring_addr) {
361 		.index = vq,
362 		.desc_user_addr = (u64)(unsigned long)queue->vring.desc,
363 		.avail_user_addr = (u64)(unsigned long)queue->vring.avail,
364 		.used_user_addr = (u64)(unsigned long)queue->vring.used,
365 	};
366 
367 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr);
368 	if (r < 0)
369 		die_perror("VHOST_SET_VRING_ADDR failed");
370 
371 	return 0;
372 }
373 
374 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi)
375 {
376 	struct net_dev *ndev = dev;
377 	struct kvm_irqfd irq;
378 	struct vhost_vring_file file;
379 	int r;
380 
381 	if (ndev->vhost_fd == 0)
382 		return;
383 
384 	irq = (struct kvm_irqfd) {
385 		.gsi	= gsi,
386 		.fd	= eventfd(0, 0),
387 	};
388 	file = (struct vhost_vring_file) {
389 		.index	= vq,
390 		.fd	= irq.fd,
391 	};
392 
393 	r = ioctl(kvm->vm_fd, KVM_IRQFD, &irq);
394 	if (r < 0)
395 		die_perror("KVM_IRQFD failed");
396 
397 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file);
398 	if (r < 0)
399 		die_perror("VHOST_SET_VRING_CALL failed");
400 	file.fd = ndev->tap_fd;
401 	r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file);
402 	if (r != 0)
403 		die("VHOST_NET_SET_BACKEND failed %d", errno);
404 
405 }
406 
407 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd)
408 {
409 	struct net_dev *ndev = dev;
410 	struct vhost_vring_file file = {
411 		.index	= vq,
412 		.fd	= efd,
413 	};
414 	int r;
415 
416 	if (ndev->vhost_fd == 0)
417 		return;
418 
419 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file);
420 	if (r < 0)
421 		die_perror("VHOST_SET_VRING_KICK failed");
422 }
423 
424 static int notify_vq(struct kvm *kvm, void *dev, u32 vq)
425 {
426 	struct net_dev *ndev = dev;
427 
428 	virtio_net_handle_callback(kvm, ndev, vq);
429 
430 	return 0;
431 }
432 
433 static int get_pfn_vq(struct kvm *kvm, void *dev, u32 vq)
434 {
435 	struct net_dev *ndev = dev;
436 
437 	return ndev->vqs[vq].pfn;
438 }
439 
440 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq)
441 {
442 	return VIRTIO_NET_QUEUE_SIZE;
443 }
444 
445 static struct virtio_ops net_dev_virtio_ops = (struct virtio_ops) {
446 	.set_config		= set_config,
447 	.get_config		= get_config,
448 	.get_host_features	= get_host_features,
449 	.set_guest_features	= set_guest_features,
450 	.init_vq		= init_vq,
451 	.notify_vq		= notify_vq,
452 	.get_pfn_vq		= get_pfn_vq,
453 	.get_size_vq		= get_size_vq,
454 	.notify_vq_gsi		= notify_vq_gsi,
455 	.notify_vq_eventfd	= notify_vq_eventfd,
456 };
457 
458 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev)
459 {
460 	u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX;
461 	struct vhost_memory *mem;
462 	int r;
463 
464 	ndev->vhost_fd = open("/dev/vhost-net", O_RDWR);
465 	if (ndev->vhost_fd < 0)
466 		die_perror("Failed openning vhost-net device");
467 
468 	mem = malloc(sizeof(*mem) + sizeof(struct vhost_memory_region));
469 	if (mem == NULL)
470 		die("Failed allocating memory for vhost memory map");
471 
472 	mem->nregions = 1;
473 	mem->regions[0] = (struct vhost_memory_region) {
474 		.guest_phys_addr	= 0,
475 		.memory_size		= kvm->ram_size,
476 		.userspace_addr		= (u64)kvm->ram_start,
477 	};
478 
479 	r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER);
480 	if (r != 0)
481 		die_perror("VHOST_SET_OWNER failed");
482 
483 	r = ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features);
484 	if (r != 0)
485 		die_perror("VHOST_SET_FEATURES failed");
486 	r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem);
487 	if (r != 0)
488 		die_perror("VHOST_SET_MEM_TABLE failed");
489 	free(mem);
490 }
491 
492 void virtio_net__init(const struct virtio_net_params *params)
493 {
494 	int i;
495 	struct net_dev *ndev;
496 
497 	if (!params)
498 		return;
499 
500 	ndev = calloc(1, sizeof(struct net_dev));
501 	if (ndev == NULL)
502 		die("Failed allocating ndev");
503 
504 	list_add_tail(&ndev->list, &ndevs);
505 
506 	ndev->kvm = params->kvm;
507 
508 	mutex_init(&ndev->mutex);
509 	ndev->config.status = VIRTIO_NET_S_LINK_UP;
510 
511 	for (i = 0 ; i < 6 ; i++) {
512 		ndev->config.mac[i]		= params->guest_mac[i];
513 		ndev->info.guest_mac.addr[i]	= params->guest_mac[i];
514 		ndev->info.host_mac.addr[i]	= params->host_mac[i];
515 	}
516 
517 	ndev->mode = params->mode;
518 	if (ndev->mode == NET_MODE_TAP) {
519 		if (!virtio_net__tap_init(params, ndev))
520 			die_perror("You have requested a TAP device, but creation of one has"
521 					"failed because:");
522 		ndev->ops = &tap_ops;
523 	} else {
524 		ndev->info.host_ip		= ntohl(inet_addr(params->host_ip));
525 		ndev->info.guest_ip		= ntohl(inet_addr(params->guest_ip));
526 		ndev->info.guest_netmask	= ntohl(inet_addr("255.255.255.0"));
527 		ndev->info.buf_nr		= 20,
528 		uip_init(&ndev->info);
529 		ndev->ops = &uip_ops;
530 	}
531 
532 	virtio_trans_init(&ndev->vtrans, VIRTIO_PCI);
533 	ndev->vtrans.trans_ops->init(kvm, &ndev->vtrans, ndev, PCI_DEVICE_ID_VIRTIO_NET,
534 					VIRTIO_ID_NET, PCI_CLASS_NET);
535 	ndev->vtrans.virtio_ops = &net_dev_virtio_ops;
536 
537 	if (params->vhost)
538 		virtio_net__vhost_init(params->kvm, ndev);
539 	else
540 		virtio_net__io_thread_init(params->kvm, ndev);
541 
542 	if (compat_id != -1)
543 		compat_id = compat__add_message("virtio-net device was not detected",
544 						"While you have requested a virtio-net device, "
545 						"the guest kernel did not initialize it.\n"
546 						"Please make sure that the guest kernel was "
547 						"compiled with CONFIG_VIRTIO_NET=y enabled "
548 						"in its .config");
549 }
550