xref: /kvmtool/virtio/net.c (revision 3a60be06942944e25fe39726b1849d7e3bf252b6)
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 		mutex_lock(&ndev->io_rx_lock);
91 		if (!virt_queue__available(vq))
92 			pthread_cond_wait(&ndev->io_rx_cond, &ndev->io_rx_lock);
93 		mutex_unlock(&ndev->io_rx_lock);
94 
95 		while (virt_queue__available(vq)) {
96 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
97 			len = ndev->ops->rx(iov, in, ndev);
98 			virt_queue__set_used_elem(vq, head, len);
99 
100 			/* We should interrupt guest right now, otherwise latency is huge. */
101 			if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_RX_QUEUE]))
102 				ndev->vtrans.trans_ops->signal_vq(kvm, &ndev->vtrans,
103 								VIRTIO_NET_RX_QUEUE);
104 		}
105 	}
106 
107 	pthread_exit(NULL);
108 	return NULL;
109 
110 }
111 
112 static void *virtio_net_tx_thread(void *p)
113 {
114 	struct iovec iov[VIRTIO_NET_QUEUE_SIZE];
115 	struct virt_queue *vq;
116 	struct kvm *kvm;
117 	struct net_dev *ndev = p;
118 	u16 out, in;
119 	u16 head;
120 	int len;
121 
122 	kvm	= ndev->kvm;
123 	vq	= &ndev->vqs[VIRTIO_NET_TX_QUEUE];
124 
125 	while (1) {
126 		mutex_lock(&ndev->io_tx_lock);
127 		if (!virt_queue__available(vq))
128 			pthread_cond_wait(&ndev->io_tx_cond, &ndev->io_tx_lock);
129 		mutex_unlock(&ndev->io_tx_lock);
130 
131 		while (virt_queue__available(vq)) {
132 			head = virt_queue__get_iov(vq, iov, &out, &in, kvm);
133 			len = ndev->ops->tx(iov, out, ndev);
134 			virt_queue__set_used_elem(vq, head, len);
135 		}
136 
137 		if (virtio_queue__should_signal(&ndev->vqs[VIRTIO_NET_TX_QUEUE]))
138 			ndev->vtrans.trans_ops->signal_vq(kvm, &ndev->vtrans, VIRTIO_NET_TX_QUEUE);
139 	}
140 
141 	pthread_exit(NULL);
142 
143 	return NULL;
144 
145 }
146 
147 static void virtio_net_handle_callback(struct kvm *kvm, struct net_dev *ndev, int queue)
148 {
149 	switch (queue) {
150 	case VIRTIO_NET_TX_QUEUE:
151 		mutex_lock(&ndev->io_tx_lock);
152 		pthread_cond_signal(&ndev->io_tx_cond);
153 		mutex_unlock(&ndev->io_tx_lock);
154 		break;
155 	case VIRTIO_NET_RX_QUEUE:
156 		mutex_lock(&ndev->io_rx_lock);
157 		pthread_cond_signal(&ndev->io_rx_cond);
158 		mutex_unlock(&ndev->io_rx_lock);
159 		break;
160 	default:
161 		pr_warning("Unknown queue index %u", queue);
162 	}
163 }
164 
165 static bool virtio_net__tap_init(const struct virtio_net_params *params,
166 					struct net_dev *ndev)
167 {
168 	int sock = socket(AF_INET, SOCK_STREAM, 0);
169 	int pid, status, offload, hdr_len;
170 	struct sockaddr_in sin = {0};
171 	struct ifreq ifr;
172 
173 	/* Did the user already gave us the FD? */
174 	if (params->fd) {
175 		ndev->tap_fd = params->fd;
176 		return 1;
177 	}
178 
179 	ndev->tap_fd = open("/dev/net/tun", O_RDWR);
180 	if (ndev->tap_fd < 0) {
181 		pr_warning("Unable to open /dev/net/tun");
182 		goto fail;
183 	}
184 
185 	memset(&ifr, 0, sizeof(ifr));
186 	ifr.ifr_flags = IFF_TAP | IFF_NO_PI | IFF_VNET_HDR;
187 	if (ioctl(ndev->tap_fd, TUNSETIFF, &ifr) < 0) {
188 		pr_warning("Config tap device error. Are you root?");
189 		goto fail;
190 	}
191 
192 	strncpy(ndev->tap_name, ifr.ifr_name, sizeof(ndev->tap_name));
193 
194 	if (ioctl(ndev->tap_fd, TUNSETNOCSUM, 1) < 0) {
195 		pr_warning("Config tap device TUNSETNOCSUM error");
196 		goto fail;
197 	}
198 
199 	hdr_len = sizeof(struct virtio_net_hdr);
200 	if (ioctl(ndev->tap_fd, TUNSETVNETHDRSZ, &hdr_len) < 0)
201 		pr_warning("Config tap device TUNSETVNETHDRSZ error");
202 
203 	offload = TUN_F_CSUM | TUN_F_TSO4 | TUN_F_TSO6 | TUN_F_UFO;
204 	if (ioctl(ndev->tap_fd, TUNSETOFFLOAD, offload) < 0) {
205 		pr_warning("Config tap device TUNSETOFFLOAD error");
206 		goto fail;
207 	}
208 
209 	if (strcmp(params->script, "none")) {
210 		pid = fork();
211 		if (pid == 0) {
212 			execl(params->script, params->script, ndev->tap_name, NULL);
213 			_exit(1);
214 		} else {
215 			waitpid(pid, &status, 0);
216 			if (WIFEXITED(status) && WEXITSTATUS(status) != 0) {
217 				pr_warning("Fail to setup tap by %s", params->script);
218 				goto fail;
219 			}
220 		}
221 	} else {
222 		memset(&ifr, 0, sizeof(ifr));
223 		strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
224 		sin.sin_addr.s_addr = inet_addr(params->host_ip);
225 		memcpy(&(ifr.ifr_addr), &sin, sizeof(ifr.ifr_addr));
226 		ifr.ifr_addr.sa_family = AF_INET;
227 		if (ioctl(sock, SIOCSIFADDR, &ifr) < 0) {
228 			pr_warning("Could not set ip address on tap device");
229 			goto fail;
230 		}
231 	}
232 
233 	memset(&ifr, 0, sizeof(ifr));
234 	strncpy(ifr.ifr_name, ndev->tap_name, sizeof(ndev->tap_name));
235 	ioctl(sock, SIOCGIFFLAGS, &ifr);
236 	ifr.ifr_flags |= IFF_UP | IFF_RUNNING;
237 	if (ioctl(sock, SIOCSIFFLAGS, &ifr) < 0)
238 		pr_warning("Could not bring tap device up");
239 
240 	close(sock);
241 
242 	return 1;
243 
244 fail:
245 	if (sock >= 0)
246 		close(sock);
247 	if (ndev->tap_fd >= 0)
248 		close(ndev->tap_fd);
249 
250 	return 0;
251 }
252 
253 static void virtio_net__io_thread_init(struct kvm *kvm, struct net_dev *ndev)
254 {
255 	pthread_mutex_init(&ndev->io_tx_lock, NULL);
256 	pthread_mutex_init(&ndev->io_rx_lock, NULL);
257 
258 	pthread_cond_init(&ndev->io_tx_cond, NULL);
259 	pthread_cond_init(&ndev->io_rx_cond, NULL);
260 
261 	pthread_create(&ndev->io_tx_thread, NULL, virtio_net_tx_thread, ndev);
262 	pthread_create(&ndev->io_rx_thread, NULL, virtio_net_rx_thread, ndev);
263 }
264 
265 static inline int tap_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
266 {
267 	return writev(ndev->tap_fd, iov, out);
268 }
269 
270 static inline int tap_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
271 {
272 	return readv(ndev->tap_fd, iov, in);
273 }
274 
275 static inline int uip_ops_tx(struct iovec *iov, u16 out, struct net_dev *ndev)
276 {
277 	return uip_tx(iov, out, &ndev->info);
278 }
279 
280 static inline int uip_ops_rx(struct iovec *iov, u16 in, struct net_dev *ndev)
281 {
282 	return uip_rx(iov, in, &ndev->info);
283 }
284 
285 static struct net_dev_operations tap_ops = {
286 	.rx	= tap_ops_rx,
287 	.tx	= tap_ops_tx,
288 };
289 
290 static struct net_dev_operations uip_ops = {
291 	.rx	= uip_ops_rx,
292 	.tx	= uip_ops_tx,
293 };
294 
295 static void set_config(struct kvm *kvm, void *dev, u8 data, u32 offset)
296 {
297 	struct net_dev *ndev = dev;
298 
299 	((u8 *)(&ndev->config))[offset] = data;
300 }
301 
302 static u8 get_config(struct kvm *kvm, void *dev, u32 offset)
303 {
304 	struct net_dev *ndev = dev;
305 
306 	return ((u8 *)(&ndev->config))[offset];
307 }
308 
309 static u32 get_host_features(struct kvm *kvm, void *dev)
310 {
311 	return 1UL << VIRTIO_NET_F_MAC
312 		| 1UL << VIRTIO_NET_F_CSUM
313 		| 1UL << VIRTIO_NET_F_HOST_UFO
314 		| 1UL << VIRTIO_NET_F_HOST_TSO4
315 		| 1UL << VIRTIO_NET_F_HOST_TSO6
316 		| 1UL << VIRTIO_NET_F_GUEST_UFO
317 		| 1UL << VIRTIO_NET_F_GUEST_TSO4
318 		| 1UL << VIRTIO_NET_F_GUEST_TSO6
319 		| 1UL << VIRTIO_RING_F_EVENT_IDX
320 		| 1UL << VIRTIO_RING_F_INDIRECT_DESC;
321 }
322 
323 static void set_guest_features(struct kvm *kvm, void *dev, u32 features)
324 {
325 	struct net_dev *ndev = dev;
326 
327 	ndev->features = features;
328 }
329 
330 static int init_vq(struct kvm *kvm, void *dev, u32 vq, u32 pfn)
331 {
332 	struct vhost_vring_state state = { .index = vq };
333 	struct vhost_vring_addr addr;
334 	struct net_dev *ndev = dev;
335 	struct virt_queue *queue;
336 	void *p;
337 	int r;
338 
339 	compat__remove_message(compat_id);
340 
341 	queue		= &ndev->vqs[vq];
342 	queue->pfn	= pfn;
343 	p		= guest_pfn_to_host(kvm, queue->pfn);
344 
345 	vring_init(&queue->vring, VIRTIO_NET_QUEUE_SIZE, p, VIRTIO_PCI_VRING_ALIGN);
346 
347 	if (ndev->vhost_fd == 0)
348 		return 0;
349 
350 	state.num = queue->vring.num;
351 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_NUM, &state);
352 	if (r < 0)
353 		die_perror("VHOST_SET_VRING_NUM failed");
354 	state.num = 0;
355 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_BASE, &state);
356 	if (r < 0)
357 		die_perror("VHOST_SET_VRING_BASE failed");
358 
359 	addr = (struct vhost_vring_addr) {
360 		.index = vq,
361 		.desc_user_addr = (u64)(unsigned long)queue->vring.desc,
362 		.avail_user_addr = (u64)(unsigned long)queue->vring.avail,
363 		.used_user_addr = (u64)(unsigned long)queue->vring.used,
364 	};
365 
366 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_ADDR, &addr);
367 	if (r < 0)
368 		die_perror("VHOST_SET_VRING_ADDR failed");
369 
370 	return 0;
371 }
372 
373 static void notify_vq_gsi(struct kvm *kvm, void *dev, u32 vq, u32 gsi)
374 {
375 	struct net_dev *ndev = dev;
376 	struct kvm_irqfd irq;
377 	struct vhost_vring_file file;
378 	int r;
379 
380 	if (ndev->vhost_fd == 0)
381 		return;
382 
383 	irq = (struct kvm_irqfd) {
384 		.gsi	= gsi,
385 		.fd	= eventfd(0, 0),
386 	};
387 	file = (struct vhost_vring_file) {
388 		.index	= vq,
389 		.fd	= irq.fd,
390 	};
391 
392 	r = ioctl(kvm->vm_fd, KVM_IRQFD, &irq);
393 	if (r < 0)
394 		die_perror("KVM_IRQFD failed");
395 
396 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_CALL, &file);
397 	if (r < 0)
398 		die_perror("VHOST_SET_VRING_CALL failed");
399 	file.fd = ndev->tap_fd;
400 	r = ioctl(ndev->vhost_fd, VHOST_NET_SET_BACKEND, &file);
401 	if (r != 0)
402 		die("VHOST_NET_SET_BACKEND failed %d", errno);
403 
404 }
405 
406 static void notify_vq_eventfd(struct kvm *kvm, void *dev, u32 vq, u32 efd)
407 {
408 	struct net_dev *ndev = dev;
409 	struct vhost_vring_file file = {
410 		.index	= vq,
411 		.fd	= efd,
412 	};
413 	int r;
414 
415 	if (ndev->vhost_fd == 0)
416 		return;
417 
418 	r = ioctl(ndev->vhost_fd, VHOST_SET_VRING_KICK, &file);
419 	if (r < 0)
420 		die_perror("VHOST_SET_VRING_KICK failed");
421 }
422 
423 static int notify_vq(struct kvm *kvm, void *dev, u32 vq)
424 {
425 	struct net_dev *ndev = dev;
426 
427 	virtio_net_handle_callback(kvm, ndev, vq);
428 
429 	return 0;
430 }
431 
432 static int get_pfn_vq(struct kvm *kvm, void *dev, u32 vq)
433 {
434 	struct net_dev *ndev = dev;
435 
436 	return ndev->vqs[vq].pfn;
437 }
438 
439 static int get_size_vq(struct kvm *kvm, void *dev, u32 vq)
440 {
441 	return VIRTIO_NET_QUEUE_SIZE;
442 }
443 
444 static struct virtio_ops net_dev_virtio_ops = (struct virtio_ops) {
445 	.set_config		= set_config,
446 	.get_config		= get_config,
447 	.get_host_features	= get_host_features,
448 	.set_guest_features	= set_guest_features,
449 	.init_vq		= init_vq,
450 	.notify_vq		= notify_vq,
451 	.get_pfn_vq		= get_pfn_vq,
452 	.get_size_vq		= get_size_vq,
453 	.notify_vq_gsi		= notify_vq_gsi,
454 	.notify_vq_eventfd	= notify_vq_eventfd,
455 };
456 
457 static void virtio_net__vhost_init(struct kvm *kvm, struct net_dev *ndev)
458 {
459 	u64 features = 1UL << VIRTIO_RING_F_EVENT_IDX;
460 	struct vhost_memory *mem;
461 	int r;
462 
463 	ndev->vhost_fd = open("/dev/vhost-net", O_RDWR);
464 	if (ndev->vhost_fd < 0)
465 		die_perror("Failed openning vhost-net device");
466 
467 	mem = malloc(sizeof(*mem) + sizeof(struct vhost_memory_region));
468 	if (mem == NULL)
469 		die("Failed allocating memory for vhost memory map");
470 
471 	mem->nregions = 1;
472 	mem->regions[0] = (struct vhost_memory_region) {
473 		.guest_phys_addr	= 0,
474 		.memory_size		= kvm->ram_size,
475 		.userspace_addr		= (unsigned long)kvm->ram_start,
476 	};
477 
478 	r = ioctl(ndev->vhost_fd, VHOST_SET_OWNER);
479 	if (r != 0)
480 		die_perror("VHOST_SET_OWNER failed");
481 
482 	r = ioctl(ndev->vhost_fd, VHOST_SET_FEATURES, &features);
483 	if (r != 0)
484 		die_perror("VHOST_SET_FEATURES failed");
485 	r = ioctl(ndev->vhost_fd, VHOST_SET_MEM_TABLE, mem);
486 	if (r != 0)
487 		die_perror("VHOST_SET_MEM_TABLE failed");
488 	free(mem);
489 }
490 
491 void virtio_net__init(const struct virtio_net_params *params)
492 {
493 	int i;
494 	struct net_dev *ndev;
495 
496 	if (!params)
497 		return;
498 
499 	ndev = calloc(1, sizeof(struct net_dev));
500 	if (ndev == NULL)
501 		die("Failed allocating ndev");
502 
503 	list_add_tail(&ndev->list, &ndevs);
504 
505 	ndev->kvm = params->kvm;
506 
507 	mutex_init(&ndev->mutex);
508 	ndev->config.status = VIRTIO_NET_S_LINK_UP;
509 
510 	for (i = 0 ; i < 6 ; i++) {
511 		ndev->config.mac[i]		= params->guest_mac[i];
512 		ndev->info.guest_mac.addr[i]	= params->guest_mac[i];
513 		ndev->info.host_mac.addr[i]	= params->host_mac[i];
514 	}
515 
516 	ndev->mode = params->mode;
517 	if (ndev->mode == NET_MODE_TAP) {
518 		if (!virtio_net__tap_init(params, ndev))
519 			die_perror("You have requested a TAP device, but creation of one has"
520 					"failed because:");
521 		ndev->ops = &tap_ops;
522 	} else {
523 		ndev->info.host_ip		= ntohl(inet_addr(params->host_ip));
524 		ndev->info.guest_ip		= ntohl(inet_addr(params->guest_ip));
525 		ndev->info.guest_netmask	= ntohl(inet_addr("255.255.255.0"));
526 		ndev->info.buf_nr		= 20,
527 		uip_init(&ndev->info);
528 		ndev->ops = &uip_ops;
529 	}
530 
531 	virtio_trans_init(&ndev->vtrans, VIRTIO_PCI);
532 	ndev->vtrans.trans_ops->init(kvm, &ndev->vtrans, ndev, PCI_DEVICE_ID_VIRTIO_NET,
533 					VIRTIO_ID_NET, PCI_CLASS_NET);
534 	ndev->vtrans.virtio_ops = &net_dev_virtio_ops;
535 
536 	if (params->vhost)
537 		virtio_net__vhost_init(params->kvm, ndev);
538 	else
539 		virtio_net__io_thread_init(params->kvm, ndev);
540 
541 	if (compat_id != -1)
542 		compat_id = compat__add_message("virtio-net device was not detected",
543 						"While you have requested a virtio-net device, "
544 						"the guest kernel did not initialize it.\n"
545 						"Please make sure that the guest kernel was "
546 						"compiled with CONFIG_VIRTIO_NET=y enabled "
547 						"in its .config");
548 }
549