xref: /linux/drivers/net/virtio_net.c (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
1 /* A simple network driver using virtio.
2  *
3  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19 //#define DEBUG
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/scatterlist.h>
26 
27 static int napi_weight = 128;
28 module_param(napi_weight, int, 0444);
29 
30 static int csum = 1, gso = 1;
31 module_param(csum, bool, 0444);
32 module_param(gso, bool, 0444);
33 
34 /* FIXME: MTU in config. */
35 #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
36 
37 struct virtnet_info
38 {
39 	struct virtio_device *vdev;
40 	struct virtqueue *rvq, *svq;
41 	struct net_device *dev;
42 	struct napi_struct napi;
43 
44 	/* Number of input buffers, and max we've ever had. */
45 	unsigned int num, max;
46 
47 	/* Receive & send queues. */
48 	struct sk_buff_head recv;
49 	struct sk_buff_head send;
50 };
51 
52 static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
53 {
54 	return (struct virtio_net_hdr *)skb->cb;
55 }
56 
57 static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
58 {
59 	sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
60 }
61 
62 static void skb_xmit_done(struct virtqueue *svq)
63 {
64 	struct virtnet_info *vi = svq->vdev->priv;
65 
66 	/* Suppress further interrupts. */
67 	svq->vq_ops->disable_cb(svq);
68 	/* We were waiting for more output buffers. */
69 	netif_wake_queue(vi->dev);
70 }
71 
72 static void receive_skb(struct net_device *dev, struct sk_buff *skb,
73 			unsigned len)
74 {
75 	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
76 
77 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
78 		pr_debug("%s: short packet %i\n", dev->name, len);
79 		dev->stats.rx_length_errors++;
80 		goto drop;
81 	}
82 	len -= sizeof(struct virtio_net_hdr);
83 	BUG_ON(len > MAX_PACKET_LEN);
84 
85 	skb_trim(skb, len);
86 	skb->protocol = eth_type_trans(skb, dev);
87 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
88 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
89 	dev->stats.rx_bytes += skb->len;
90 	dev->stats.rx_packets++;
91 
92 	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
93 		pr_debug("Needs csum!\n");
94 		if (!skb_partial_csum_set(skb,hdr->csum_start,hdr->csum_offset))
95 			goto frame_err;
96 	}
97 
98 	if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
99 		pr_debug("GSO!\n");
100 		switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
101 		case VIRTIO_NET_HDR_GSO_TCPV4:
102 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
103 			break;
104 		case VIRTIO_NET_HDR_GSO_UDP:
105 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
106 			break;
107 		case VIRTIO_NET_HDR_GSO_TCPV6:
108 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
109 			break;
110 		default:
111 			if (net_ratelimit())
112 				printk(KERN_WARNING "%s: bad gso type %u.\n",
113 				       dev->name, hdr->gso_type);
114 			goto frame_err;
115 		}
116 
117 		if (hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
118 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
119 
120 		skb_shinfo(skb)->gso_size = hdr->gso_size;
121 		if (skb_shinfo(skb)->gso_size == 0) {
122 			if (net_ratelimit())
123 				printk(KERN_WARNING "%s: zero gso size.\n",
124 				       dev->name);
125 			goto frame_err;
126 		}
127 
128 		/* Header must be checked, and gso_segs computed. */
129 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
130 		skb_shinfo(skb)->gso_segs = 0;
131 	}
132 
133 	netif_receive_skb(skb);
134 	return;
135 
136 frame_err:
137 	dev->stats.rx_frame_errors++;
138 drop:
139 	dev_kfree_skb(skb);
140 }
141 
142 static void try_fill_recv(struct virtnet_info *vi)
143 {
144 	struct sk_buff *skb;
145 	struct scatterlist sg[1+MAX_SKB_FRAGS];
146 	int num, err;
147 
148 	sg_init_table(sg, 1+MAX_SKB_FRAGS);
149 	for (;;) {
150 		skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
151 		if (unlikely(!skb))
152 			break;
153 
154 		skb_put(skb, MAX_PACKET_LEN);
155 		vnet_hdr_to_sg(sg, skb);
156 		num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
157 		skb_queue_head(&vi->recv, skb);
158 
159 		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
160 		if (err) {
161 			skb_unlink(skb, &vi->recv);
162 			kfree_skb(skb);
163 			break;
164 		}
165 		vi->num++;
166 	}
167 	if (unlikely(vi->num > vi->max))
168 		vi->max = vi->num;
169 	vi->rvq->vq_ops->kick(vi->rvq);
170 }
171 
172 static void skb_recv_done(struct virtqueue *rvq)
173 {
174 	struct virtnet_info *vi = rvq->vdev->priv;
175 	/* Schedule NAPI, Suppress further interrupts if successful. */
176 	if (netif_rx_schedule_prep(vi->dev, &vi->napi)) {
177 		rvq->vq_ops->disable_cb(rvq);
178 		__netif_rx_schedule(vi->dev, &vi->napi);
179 	}
180 }
181 
182 static int virtnet_poll(struct napi_struct *napi, int budget)
183 {
184 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
185 	struct sk_buff *skb = NULL;
186 	unsigned int len, received = 0;
187 
188 again:
189 	while (received < budget &&
190 	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
191 		__skb_unlink(skb, &vi->recv);
192 		receive_skb(vi->dev, skb, len);
193 		vi->num--;
194 		received++;
195 	}
196 
197 	/* FIXME: If we oom and completely run out of inbufs, we need
198 	 * to start a timer trying to fill more. */
199 	if (vi->num < vi->max / 2)
200 		try_fill_recv(vi);
201 
202 	/* Out of packets? */
203 	if (received < budget) {
204 		netif_rx_complete(vi->dev, napi);
205 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq))
206 		    && netif_rx_reschedule(vi->dev, napi))
207 			goto again;
208 	}
209 
210 	return received;
211 }
212 
213 static void free_old_xmit_skbs(struct virtnet_info *vi)
214 {
215 	struct sk_buff *skb;
216 	unsigned int len;
217 
218 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
219 		pr_debug("Sent skb %p\n", skb);
220 		__skb_unlink(skb, &vi->send);
221 		vi->dev->stats.tx_bytes += len;
222 		vi->dev->stats.tx_packets++;
223 		kfree_skb(skb);
224 	}
225 }
226 
227 static int start_xmit(struct sk_buff *skb, struct net_device *dev)
228 {
229 	struct virtnet_info *vi = netdev_priv(dev);
230 	int num, err;
231 	struct scatterlist sg[1+MAX_SKB_FRAGS];
232 	struct virtio_net_hdr *hdr;
233 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
234 	DECLARE_MAC_BUF(mac);
235 
236 	sg_init_table(sg, 1+MAX_SKB_FRAGS);
237 
238 	pr_debug("%s: xmit %p %s\n", dev->name, skb, print_mac(mac, dest));
239 
240 	/* Encode metadata header at front. */
241 	hdr = skb_vnet_hdr(skb);
242 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
243 		hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
244 		hdr->csum_start = skb->csum_start - skb_headroom(skb);
245 		hdr->csum_offset = skb->csum_offset;
246 	} else {
247 		hdr->flags = 0;
248 		hdr->csum_offset = hdr->csum_start = 0;
249 	}
250 
251 	if (skb_is_gso(skb)) {
252 		hdr->hdr_len = skb_transport_header(skb) - skb->data;
253 		hdr->gso_size = skb_shinfo(skb)->gso_size;
254 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
255 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
256 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
257 			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
258 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
259 			hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
260 		else
261 			BUG();
262 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
263 			hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
264 	} else {
265 		hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
266 		hdr->gso_size = hdr->hdr_len = 0;
267 	}
268 
269 	vnet_hdr_to_sg(sg, skb);
270 	num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
271 	__skb_queue_head(&vi->send, skb);
272 
273 again:
274 	/* Free up any pending old buffers before queueing new ones. */
275 	free_old_xmit_skbs(vi);
276 	err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
277 	if (err) {
278 		pr_debug("%s: virtio not prepared to send\n", dev->name);
279 		netif_stop_queue(dev);
280 
281 		/* Activate callback for using skbs: if this fails it
282 		 * means some were used in the meantime. */
283 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
284 			printk("Unlikely: restart svq failed\n");
285 			netif_start_queue(dev);
286 			goto again;
287 		}
288 		__skb_unlink(skb, &vi->send);
289 
290 		return NETDEV_TX_BUSY;
291 	}
292 	vi->svq->vq_ops->kick(vi->svq);
293 
294 	return 0;
295 }
296 
297 static int virtnet_open(struct net_device *dev)
298 {
299 	struct virtnet_info *vi = netdev_priv(dev);
300 
301 	napi_enable(&vi->napi);
302 
303 	/* If all buffers were filled by other side before we napi_enabled, we
304 	 * won't get another interrupt, so process any outstanding packets
305 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
306 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
307 	if (netif_rx_schedule_prep(dev, &vi->napi)) {
308 		vi->rvq->vq_ops->disable_cb(vi->rvq);
309 		__netif_rx_schedule(dev, &vi->napi);
310 	}
311 	return 0;
312 }
313 
314 static int virtnet_close(struct net_device *dev)
315 {
316 	struct virtnet_info *vi = netdev_priv(dev);
317 
318 	napi_disable(&vi->napi);
319 
320 	return 0;
321 }
322 
323 static int virtnet_probe(struct virtio_device *vdev)
324 {
325 	int err;
326 	struct net_device *dev;
327 	struct virtnet_info *vi;
328 
329 	/* Allocate ourselves a network device with room for our info */
330 	dev = alloc_etherdev(sizeof(struct virtnet_info));
331 	if (!dev)
332 		return -ENOMEM;
333 
334 	/* Set up network device as normal. */
335 	dev->open = virtnet_open;
336 	dev->stop = virtnet_close;
337 	dev->hard_start_xmit = start_xmit;
338 	dev->features = NETIF_F_HIGHDMA;
339 	SET_NETDEV_DEV(dev, &vdev->dev);
340 
341 	/* Do we support "hardware" checksums? */
342 	if (csum && vdev->config->feature(vdev, VIRTIO_NET_F_CSUM)) {
343 		/* This opens up the world of extra features. */
344 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
345 		if (gso && vdev->config->feature(vdev, VIRTIO_NET_F_GSO)) {
346 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
347 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
348 		}
349 	}
350 
351 	/* Configuration may specify what MAC to use.  Otherwise random. */
352 	if (vdev->config->feature(vdev, VIRTIO_NET_F_MAC)) {
353 		vdev->config->get(vdev,
354 				  offsetof(struct virtio_net_config, mac),
355 				  dev->dev_addr, dev->addr_len);
356 	} else
357 		random_ether_addr(dev->dev_addr);
358 
359 	/* Set up our device-specific information */
360 	vi = netdev_priv(dev);
361 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
362 	vi->dev = dev;
363 	vi->vdev = vdev;
364 
365 	/* We expect two virtqueues, receive then send. */
366 	vi->rvq = vdev->config->find_vq(vdev, 0, skb_recv_done);
367 	if (IS_ERR(vi->rvq)) {
368 		err = PTR_ERR(vi->rvq);
369 		goto free;
370 	}
371 
372 	vi->svq = vdev->config->find_vq(vdev, 1, skb_xmit_done);
373 	if (IS_ERR(vi->svq)) {
374 		err = PTR_ERR(vi->svq);
375 		goto free_recv;
376 	}
377 
378 	/* Initialize our empty receive and send queues. */
379 	skb_queue_head_init(&vi->recv);
380 	skb_queue_head_init(&vi->send);
381 
382 	err = register_netdev(dev);
383 	if (err) {
384 		pr_debug("virtio_net: registering device failed\n");
385 		goto free_send;
386 	}
387 
388 	/* Last of all, set up some receive buffers. */
389 	try_fill_recv(vi);
390 
391 	/* If we didn't even get one input buffer, we're useless. */
392 	if (vi->num == 0) {
393 		err = -ENOMEM;
394 		goto unregister;
395 	}
396 
397 	pr_debug("virtnet: registered device %s\n", dev->name);
398 	vdev->priv = vi;
399 	return 0;
400 
401 unregister:
402 	unregister_netdev(dev);
403 free_send:
404 	vdev->config->del_vq(vi->svq);
405 free_recv:
406 	vdev->config->del_vq(vi->rvq);
407 free:
408 	free_netdev(dev);
409 	return err;
410 }
411 
412 static void virtnet_remove(struct virtio_device *vdev)
413 {
414 	struct virtnet_info *vi = vdev->priv;
415 	struct sk_buff *skb;
416 
417 	/* Stop all the virtqueues. */
418 	vdev->config->reset(vdev);
419 
420 	/* Free our skbs in send and recv queues, if any. */
421 	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
422 		kfree_skb(skb);
423 		vi->num--;
424 	}
425 	while ((skb = __skb_dequeue(&vi->send)) != NULL)
426 		kfree_skb(skb);
427 
428 	BUG_ON(vi->num != 0);
429 
430 	vdev->config->del_vq(vi->svq);
431 	vdev->config->del_vq(vi->rvq);
432 	unregister_netdev(vi->dev);
433 	free_netdev(vi->dev);
434 }
435 
436 static struct virtio_device_id id_table[] = {
437 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
438 	{ 0 },
439 };
440 
441 static struct virtio_driver virtio_net = {
442 	.driver.name =	KBUILD_MODNAME,
443 	.driver.owner =	THIS_MODULE,
444 	.id_table =	id_table,
445 	.probe =	virtnet_probe,
446 	.remove =	__devexit_p(virtnet_remove),
447 };
448 
449 static int __init init(void)
450 {
451 	return register_virtio_driver(&virtio_net);
452 }
453 
454 static void __exit fini(void)
455 {
456 	unregister_virtio_driver(&virtio_net);
457 }
458 module_init(init);
459 module_exit(fini);
460 
461 MODULE_DEVICE_TABLE(virtio, id_table);
462 MODULE_DESCRIPTION("Virtio network driver");
463 MODULE_LICENSE("GPL");
464