Commit | Line | Data |
---|---|---|
296f96fc RR |
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 | /* FIXME: MTU in config. */ | |
28 | #define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN) | |
29 | ||
30 | struct virtnet_info | |
31 | { | |
32 | struct virtio_device *vdev; | |
33 | struct virtqueue *rvq, *svq; | |
34 | struct net_device *dev; | |
35 | struct napi_struct napi; | |
36 | ||
37 | /* Number of input buffers, and max we've ever had. */ | |
38 | unsigned int num, max; | |
39 | ||
40 | /* Receive & send queues. */ | |
41 | struct sk_buff_head recv; | |
42 | struct sk_buff_head send; | |
43 | }; | |
44 | ||
45 | static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb) | |
46 | { | |
47 | return (struct virtio_net_hdr *)skb->cb; | |
48 | } | |
49 | ||
50 | static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb) | |
51 | { | |
52 | sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr)); | |
53 | } | |
54 | ||
55 | static bool skb_xmit_done(struct virtqueue *rvq) | |
56 | { | |
57 | struct virtnet_info *vi = rvq->vdev->priv; | |
58 | ||
59 | /* In case we were waiting for output buffers. */ | |
60 | netif_wake_queue(vi->dev); | |
61 | return true; | |
62 | } | |
63 | ||
64 | static void receive_skb(struct net_device *dev, struct sk_buff *skb, | |
65 | unsigned len) | |
66 | { | |
67 | struct virtio_net_hdr *hdr = skb_vnet_hdr(skb); | |
68 | ||
69 | if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { | |
70 | pr_debug("%s: short packet %i\n", dev->name, len); | |
71 | dev->stats.rx_length_errors++; | |
72 | goto drop; | |
73 | } | |
74 | len -= sizeof(struct virtio_net_hdr); | |
75 | BUG_ON(len > MAX_PACKET_LEN); | |
76 | ||
77 | skb_trim(skb, len); | |
78 | skb->protocol = eth_type_trans(skb, dev); | |
79 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
80 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
81 | dev->stats.rx_bytes += skb->len; | |
82 | dev->stats.rx_packets++; | |
83 | ||
84 | if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { | |
85 | pr_debug("Needs csum!\n"); | |
86 | skb->ip_summed = CHECKSUM_PARTIAL; | |
87 | skb->csum_start = hdr->csum_start; | |
88 | skb->csum_offset = hdr->csum_offset; | |
89 | if (skb->csum_start > skb->len - 2 | |
90 | || skb->csum_offset > skb->len - 2) { | |
91 | if (net_ratelimit()) | |
92 | printk(KERN_WARNING "%s: csum=%u/%u len=%u\n", | |
93 | dev->name, skb->csum_start, | |
94 | skb->csum_offset, skb->len); | |
95 | goto frame_err; | |
96 | } | |
97 | } | |
98 | ||
99 | if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) { | |
100 | pr_debug("GSO!\n"); | |
101 | switch (hdr->gso_type) { | |
102 | case VIRTIO_NET_HDR_GSO_TCPV4: | |
103 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
104 | break; | |
105 | case VIRTIO_NET_HDR_GSO_TCPV4_ECN: | |
106 | skb_shinfo(skb)->gso_type = SKB_GSO_TCP_ECN; | |
107 | break; | |
108 | case VIRTIO_NET_HDR_GSO_UDP: | |
109 | skb_shinfo(skb)->gso_type = SKB_GSO_UDP; | |
110 | break; | |
111 | case VIRTIO_NET_HDR_GSO_TCPV6: | |
112 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; | |
113 | break; | |
114 | default: | |
115 | if (net_ratelimit()) | |
116 | printk(KERN_WARNING "%s: bad gso type %u.\n", | |
117 | dev->name, hdr->gso_type); | |
118 | goto frame_err; | |
119 | } | |
120 | ||
121 | skb_shinfo(skb)->gso_size = hdr->gso_size; | |
122 | if (skb_shinfo(skb)->gso_size == 0) { | |
123 | if (net_ratelimit()) | |
124 | printk(KERN_WARNING "%s: zero gso size.\n", | |
125 | dev->name); | |
126 | goto frame_err; | |
127 | } | |
128 | ||
129 | /* Header must be checked, and gso_segs computed. */ | |
130 | skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; | |
131 | skb_shinfo(skb)->gso_segs = 0; | |
132 | } | |
133 | ||
134 | netif_receive_skb(skb); | |
135 | return; | |
136 | ||
137 | frame_err: | |
138 | dev->stats.rx_frame_errors++; | |
139 | drop: | |
140 | dev_kfree_skb(skb); | |
141 | } | |
142 | ||
143 | static void try_fill_recv(struct virtnet_info *vi) | |
144 | { | |
145 | struct sk_buff *skb; | |
146 | struct scatterlist sg[1+MAX_SKB_FRAGS]; | |
147 | int num, err; | |
148 | ||
4d125de3 | 149 | sg_init_table(sg, 1+MAX_SKB_FRAGS); |
296f96fc RR |
150 | for (;;) { |
151 | skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN); | |
152 | if (unlikely(!skb)) | |
153 | break; | |
154 | ||
155 | skb_put(skb, MAX_PACKET_LEN); | |
156 | vnet_hdr_to_sg(sg, skb); | |
157 | num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1; | |
158 | skb_queue_head(&vi->recv, skb); | |
159 | ||
160 | err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb); | |
161 | if (err) { | |
162 | skb_unlink(skb, &vi->recv); | |
163 | kfree_skb(skb); | |
164 | break; | |
165 | } | |
166 | vi->num++; | |
167 | } | |
168 | if (unlikely(vi->num > vi->max)) | |
169 | vi->max = vi->num; | |
170 | vi->rvq->vq_ops->kick(vi->rvq); | |
171 | } | |
172 | ||
173 | static bool skb_recv_done(struct virtqueue *rvq) | |
174 | { | |
175 | struct virtnet_info *vi = rvq->vdev->priv; | |
176 | netif_rx_schedule(vi->dev, &vi->napi); | |
177 | /* Suppress further interrupts. */ | |
178 | return false; | |
179 | } | |
180 | ||
181 | static int virtnet_poll(struct napi_struct *napi, int budget) | |
182 | { | |
183 | struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi); | |
184 | struct sk_buff *skb = NULL; | |
185 | unsigned int len, received = 0; | |
186 | ||
187 | again: | |
188 | while (received < budget && | |
189 | (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) { | |
190 | __skb_unlink(skb, &vi->recv); | |
191 | receive_skb(vi->dev, skb, len); | |
192 | vi->num--; | |
193 | received++; | |
194 | } | |
195 | ||
196 | /* FIXME: If we oom and completely run out of inbufs, we need | |
197 | * to start a timer trying to fill more. */ | |
198 | if (vi->num < vi->max / 2) | |
199 | try_fill_recv(vi); | |
200 | ||
8329d98e RR |
201 | /* Out of packets? */ |
202 | if (received < budget) { | |
296f96fc RR |
203 | netif_rx_complete(vi->dev, napi); |
204 | if (unlikely(!vi->rvq->vq_ops->restart(vi->rvq)) | |
205 | && netif_rx_reschedule(vi->dev, napi)) | |
206 | goto again; | |
207 | } | |
208 | ||
209 | return received; | |
210 | } | |
211 | ||
212 | static void free_old_xmit_skbs(struct virtnet_info *vi) | |
213 | { | |
214 | struct sk_buff *skb; | |
215 | unsigned int len; | |
216 | ||
217 | while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) { | |
218 | pr_debug("Sent skb %p\n", skb); | |
219 | __skb_unlink(skb, &vi->send); | |
220 | vi->dev->stats.tx_bytes += len; | |
221 | vi->dev->stats.tx_packets++; | |
222 | kfree_skb(skb); | |
223 | } | |
224 | } | |
225 | ||
226 | static int start_xmit(struct sk_buff *skb, struct net_device *dev) | |
227 | { | |
228 | struct virtnet_info *vi = netdev_priv(dev); | |
229 | int num, err; | |
230 | struct scatterlist sg[1+MAX_SKB_FRAGS]; | |
231 | struct virtio_net_hdr *hdr; | |
232 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; | |
233 | DECLARE_MAC_BUF(mac); | |
234 | ||
4d125de3 RR |
235 | sg_init_table(sg, 1+MAX_SKB_FRAGS); |
236 | ||
296f96fc RR |
237 | pr_debug("%s: xmit %p %s\n", dev->name, skb, print_mac(mac, dest)); |
238 | ||
239 | free_old_xmit_skbs(vi); | |
240 | ||
241 | /* Encode metadata header at front. */ | |
242 | hdr = skb_vnet_hdr(skb); | |
243 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
244 | hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; | |
245 | hdr->csum_start = skb->csum_start - skb_headroom(skb); | |
246 | hdr->csum_offset = skb->csum_offset; | |
247 | } else { | |
248 | hdr->flags = 0; | |
249 | hdr->csum_offset = hdr->csum_start = 0; | |
250 | } | |
251 | ||
252 | if (skb_is_gso(skb)) { | |
253 | hdr->gso_size = skb_shinfo(skb)->gso_size; | |
254 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) | |
255 | hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4_ECN; | |
256 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) | |
257 | hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4; | |
258 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) | |
259 | hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6; | |
260 | else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) | |
261 | hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP; | |
262 | else | |
263 | BUG(); | |
264 | } else { | |
265 | hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE; | |
266 | hdr->gso_size = 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 | err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb); | |
273 | if (err) { | |
274 | pr_debug("%s: virtio not prepared to send\n", dev->name); | |
275 | skb_unlink(skb, &vi->send); | |
276 | netif_stop_queue(dev); | |
277 | return NETDEV_TX_BUSY; | |
278 | } | |
279 | vi->svq->vq_ops->kick(vi->svq); | |
280 | ||
281 | return 0; | |
282 | } | |
283 | ||
284 | static int virtnet_open(struct net_device *dev) | |
285 | { | |
286 | struct virtnet_info *vi = netdev_priv(dev); | |
287 | ||
288 | try_fill_recv(vi); | |
289 | ||
290 | /* If we didn't even get one input buffer, we're useless. */ | |
291 | if (vi->num == 0) | |
292 | return -ENOMEM; | |
293 | ||
294 | napi_enable(&vi->napi); | |
295 | return 0; | |
296 | } | |
297 | ||
298 | static int virtnet_close(struct net_device *dev) | |
299 | { | |
300 | struct virtnet_info *vi = netdev_priv(dev); | |
301 | struct sk_buff *skb; | |
302 | ||
303 | napi_disable(&vi->napi); | |
304 | ||
305 | /* networking core has neutered skb_xmit_done/skb_recv_done, so don't | |
306 | * worry about races vs. get(). */ | |
307 | vi->rvq->vq_ops->shutdown(vi->rvq); | |
308 | while ((skb = __skb_dequeue(&vi->recv)) != NULL) { | |
309 | kfree_skb(skb); | |
310 | vi->num--; | |
311 | } | |
312 | vi->svq->vq_ops->shutdown(vi->svq); | |
313 | while ((skb = __skb_dequeue(&vi->send)) != NULL) | |
314 | kfree_skb(skb); | |
315 | ||
316 | BUG_ON(vi->num != 0); | |
317 | return 0; | |
318 | } | |
319 | ||
320 | static int virtnet_probe(struct virtio_device *vdev) | |
321 | { | |
322 | int err; | |
323 | unsigned int len; | |
324 | struct net_device *dev; | |
325 | struct virtnet_info *vi; | |
326 | void *token; | |
327 | ||
328 | /* Allocate ourselves a network device with room for our info */ | |
329 | dev = alloc_etherdev(sizeof(struct virtnet_info)); | |
330 | if (!dev) | |
331 | return -ENOMEM; | |
332 | ||
333 | /* Set up network device as normal. */ | |
334 | ether_setup(dev); | |
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 | token = vdev->config->find(vdev, VIRTIO_CONFIG_NET_F, &len); | |
343 | if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_NO_CSUM)) { | |
344 | /* This opens up the world of extra features. */ | |
345 | dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; | |
346 | if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO4)) | |
347 | dev->features |= NETIF_F_TSO; | |
348 | if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_UFO)) | |
349 | dev->features |= NETIF_F_UFO; | |
350 | if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO4_ECN)) | |
351 | dev->features |= NETIF_F_TSO_ECN; | |
352 | if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO6)) | |
353 | dev->features |= NETIF_F_TSO6; | |
354 | } | |
355 | ||
356 | /* Configuration may specify what MAC to use. Otherwise random. */ | |
357 | token = vdev->config->find(vdev, VIRTIO_CONFIG_NET_MAC_F, &len); | |
358 | if (token) { | |
359 | dev->addr_len = len; | |
360 | vdev->config->get(vdev, token, dev->dev_addr, len); | |
361 | } else | |
362 | random_ether_addr(dev->dev_addr); | |
363 | ||
364 | /* Set up our device-specific information */ | |
365 | vi = netdev_priv(dev); | |
366 | netif_napi_add(dev, &vi->napi, virtnet_poll, 16); | |
367 | vi->dev = dev; | |
368 | vi->vdev = vdev; | |
369 | ||
370 | /* We expect two virtqueues, receive then send. */ | |
371 | vi->rvq = vdev->config->find_vq(vdev, skb_recv_done); | |
372 | if (IS_ERR(vi->rvq)) { | |
373 | err = PTR_ERR(vi->rvq); | |
374 | goto free; | |
375 | } | |
376 | ||
377 | vi->svq = vdev->config->find_vq(vdev, skb_xmit_done); | |
378 | if (IS_ERR(vi->svq)) { | |
379 | err = PTR_ERR(vi->svq); | |
380 | goto free_recv; | |
381 | } | |
382 | ||
383 | /* Initialize our empty receive and send queues. */ | |
384 | skb_queue_head_init(&vi->recv); | |
385 | skb_queue_head_init(&vi->send); | |
386 | ||
387 | err = register_netdev(dev); | |
388 | if (err) { | |
389 | pr_debug("virtio_net: registering device failed\n"); | |
390 | goto free_send; | |
391 | } | |
392 | pr_debug("virtnet: registered device %s\n", dev->name); | |
393 | vdev->priv = vi; | |
394 | return 0; | |
395 | ||
396 | free_send: | |
397 | vdev->config->del_vq(vi->svq); | |
398 | free_recv: | |
399 | vdev->config->del_vq(vi->rvq); | |
400 | free: | |
401 | free_netdev(dev); | |
402 | return err; | |
403 | } | |
404 | ||
405 | static void virtnet_remove(struct virtio_device *vdev) | |
406 | { | |
74b2553f RR |
407 | struct virtnet_info *vi = vdev->priv; |
408 | ||
409 | vdev->config->del_vq(vi->svq); | |
410 | vdev->config->del_vq(vi->rvq); | |
411 | unregister_netdev(vi->dev); | |
412 | free_netdev(vi->dev); | |
296f96fc RR |
413 | } |
414 | ||
415 | static struct virtio_device_id id_table[] = { | |
416 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
417 | { 0 }, | |
418 | }; | |
419 | ||
420 | static struct virtio_driver virtio_net = { | |
421 | .driver.name = KBUILD_MODNAME, | |
422 | .driver.owner = THIS_MODULE, | |
423 | .id_table = id_table, | |
424 | .probe = virtnet_probe, | |
425 | .remove = __devexit_p(virtnet_remove), | |
426 | }; | |
427 | ||
428 | static int __init init(void) | |
429 | { | |
430 | return register_virtio_driver(&virtio_net); | |
431 | } | |
432 | ||
433 | static void __exit fini(void) | |
434 | { | |
435 | unregister_virtio_driver(&virtio_net); | |
436 | } | |
437 | module_init(init); | |
438 | module_exit(fini); | |
439 | ||
440 | MODULE_DEVICE_TABLE(virtio, id_table); | |
441 | MODULE_DESCRIPTION("Virtio network driver"); | |
442 | MODULE_LICENSE("GPL"); |