xref: /linux/drivers/net/virtio_net.c (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
1 /* A 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/ethtool.h>
23 #include <linux/module.h>
24 #include <linux/virtio.h>
25 #include <linux/virtio_net.h>
26 #include <linux/scatterlist.h>
27 #include <linux/if_vlan.h>
28 #include <linux/slab.h>
29 
30 static int napi_weight = 128;
31 module_param(napi_weight, int, 0444);
32 
33 static int csum = 1, gso = 1;
34 module_param(csum, bool, 0444);
35 module_param(gso, bool, 0444);
36 
37 /* FIXME: MTU in config. */
38 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
39 #define GOOD_COPY_LEN	128
40 
41 #define VIRTNET_SEND_COMMAND_SG_MAX    2
42 
43 struct virtnet_info
44 {
45 	struct virtio_device *vdev;
46 	struct virtqueue *rvq, *svq, *cvq;
47 	struct net_device *dev;
48 	struct napi_struct napi;
49 	unsigned int status;
50 
51 	/* Number of input buffers, and max we've ever had. */
52 	unsigned int num, max;
53 
54 	/* I like... big packets and I cannot lie! */
55 	bool big_packets;
56 
57 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
58 	bool mergeable_rx_bufs;
59 
60 	/* Work struct for refilling if we run low on memory. */
61 	struct delayed_work refill;
62 
63 	/* Chain pages by the private ptr. */
64 	struct page *pages;
65 };
66 
67 struct skb_vnet_hdr {
68 	union {
69 		struct virtio_net_hdr hdr;
70 		struct virtio_net_hdr_mrg_rxbuf mhdr;
71 	};
72 	unsigned int num_sg;
73 };
74 
75 struct padded_vnet_hdr {
76 	struct virtio_net_hdr hdr;
77 	/*
78 	 * virtio_net_hdr should be in a separated sg buffer because of a
79 	 * QEMU bug, and data sg buffer shares same page with this header sg.
80 	 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
81 	 */
82 	char padding[6];
83 };
84 
85 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
86 {
87 	return (struct skb_vnet_hdr *)skb->cb;
88 }
89 
90 /*
91  * private is used to chain pages for big packets, put the whole
92  * most recent used list in the beginning for reuse
93  */
94 static void give_pages(struct virtnet_info *vi, struct page *page)
95 {
96 	struct page *end;
97 
98 	/* Find end of list, sew whole thing into vi->pages. */
99 	for (end = page; end->private; end = (struct page *)end->private);
100 	end->private = (unsigned long)vi->pages;
101 	vi->pages = page;
102 }
103 
104 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
105 {
106 	struct page *p = vi->pages;
107 
108 	if (p) {
109 		vi->pages = (struct page *)p->private;
110 		/* clear private here, it is used to chain pages */
111 		p->private = 0;
112 	} else
113 		p = alloc_page(gfp_mask);
114 	return p;
115 }
116 
117 static void skb_xmit_done(struct virtqueue *svq)
118 {
119 	struct virtnet_info *vi = svq->vdev->priv;
120 
121 	/* Suppress further interrupts. */
122 	svq->vq_ops->disable_cb(svq);
123 
124 	/* We were probably waiting for more output buffers. */
125 	netif_wake_queue(vi->dev);
126 }
127 
128 static void set_skb_frag(struct sk_buff *skb, struct page *page,
129 			 unsigned int offset, unsigned int *len)
130 {
131 	int i = skb_shinfo(skb)->nr_frags;
132 	skb_frag_t *f;
133 
134 	f = &skb_shinfo(skb)->frags[i];
135 	f->size = min((unsigned)PAGE_SIZE - offset, *len);
136 	f->page_offset = offset;
137 	f->page = page;
138 
139 	skb->data_len += f->size;
140 	skb->len += f->size;
141 	skb_shinfo(skb)->nr_frags++;
142 	*len -= f->size;
143 }
144 
145 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
146 				   struct page *page, unsigned int len)
147 {
148 	struct sk_buff *skb;
149 	struct skb_vnet_hdr *hdr;
150 	unsigned int copy, hdr_len, offset;
151 	char *p;
152 
153 	p = page_address(page);
154 
155 	/* copy small packet so we can reuse these pages for small data */
156 	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
157 	if (unlikely(!skb))
158 		return NULL;
159 
160 	hdr = skb_vnet_hdr(skb);
161 
162 	if (vi->mergeable_rx_bufs) {
163 		hdr_len = sizeof hdr->mhdr;
164 		offset = hdr_len;
165 	} else {
166 		hdr_len = sizeof hdr->hdr;
167 		offset = sizeof(struct padded_vnet_hdr);
168 	}
169 
170 	memcpy(hdr, p, hdr_len);
171 
172 	len -= hdr_len;
173 	p += offset;
174 
175 	copy = len;
176 	if (copy > skb_tailroom(skb))
177 		copy = skb_tailroom(skb);
178 	memcpy(skb_put(skb, copy), p, copy);
179 
180 	len -= copy;
181 	offset += copy;
182 
183 	while (len) {
184 		set_skb_frag(skb, page, offset, &len);
185 		page = (struct page *)page->private;
186 		offset = 0;
187 	}
188 
189 	if (page)
190 		give_pages(vi, page);
191 
192 	return skb;
193 }
194 
195 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
196 {
197 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
198 	struct page *page;
199 	int num_buf, i, len;
200 
201 	num_buf = hdr->mhdr.num_buffers;
202 	while (--num_buf) {
203 		i = skb_shinfo(skb)->nr_frags;
204 		if (i >= MAX_SKB_FRAGS) {
205 			pr_debug("%s: packet too long\n", skb->dev->name);
206 			skb->dev->stats.rx_length_errors++;
207 			return -EINVAL;
208 		}
209 
210 		page = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
211 		if (!page) {
212 			pr_debug("%s: rx error: %d buffers missing\n",
213 				 skb->dev->name, hdr->mhdr.num_buffers);
214 			skb->dev->stats.rx_length_errors++;
215 			return -EINVAL;
216 		}
217 		if (len > PAGE_SIZE)
218 			len = PAGE_SIZE;
219 
220 		set_skb_frag(skb, page, 0, &len);
221 
222 		--vi->num;
223 	}
224 	return 0;
225 }
226 
227 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
228 {
229 	struct virtnet_info *vi = netdev_priv(dev);
230 	struct sk_buff *skb;
231 	struct page *page;
232 	struct skb_vnet_hdr *hdr;
233 
234 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
235 		pr_debug("%s: short packet %i\n", dev->name, len);
236 		dev->stats.rx_length_errors++;
237 		if (vi->mergeable_rx_bufs || vi->big_packets)
238 			give_pages(vi, buf);
239 		else
240 			dev_kfree_skb(buf);
241 		return;
242 	}
243 
244 	if (!vi->mergeable_rx_bufs && !vi->big_packets) {
245 		skb = buf;
246 		len -= sizeof(struct virtio_net_hdr);
247 		skb_trim(skb, len);
248 	} else {
249 		page = buf;
250 		skb = page_to_skb(vi, page, len);
251 		if (unlikely(!skb)) {
252 			dev->stats.rx_dropped++;
253 			give_pages(vi, page);
254 			return;
255 		}
256 		if (vi->mergeable_rx_bufs)
257 			if (receive_mergeable(vi, skb)) {
258 				dev_kfree_skb(skb);
259 				return;
260 			}
261 	}
262 
263 	hdr = skb_vnet_hdr(skb);
264 	skb->truesize += skb->data_len;
265 	dev->stats.rx_bytes += skb->len;
266 	dev->stats.rx_packets++;
267 
268 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
269 		pr_debug("Needs csum!\n");
270 		if (!skb_partial_csum_set(skb,
271 					  hdr->hdr.csum_start,
272 					  hdr->hdr.csum_offset))
273 			goto frame_err;
274 	}
275 
276 	skb->protocol = eth_type_trans(skb, dev);
277 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
278 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
279 
280 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
281 		pr_debug("GSO!\n");
282 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
283 		case VIRTIO_NET_HDR_GSO_TCPV4:
284 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
285 			break;
286 		case VIRTIO_NET_HDR_GSO_UDP:
287 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
288 			break;
289 		case VIRTIO_NET_HDR_GSO_TCPV6:
290 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
291 			break;
292 		default:
293 			if (net_ratelimit())
294 				printk(KERN_WARNING "%s: bad gso type %u.\n",
295 				       dev->name, hdr->hdr.gso_type);
296 			goto frame_err;
297 		}
298 
299 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
300 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
301 
302 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
303 		if (skb_shinfo(skb)->gso_size == 0) {
304 			if (net_ratelimit())
305 				printk(KERN_WARNING "%s: zero gso size.\n",
306 				       dev->name);
307 			goto frame_err;
308 		}
309 
310 		/* Header must be checked, and gso_segs computed. */
311 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
312 		skb_shinfo(skb)->gso_segs = 0;
313 	}
314 
315 	netif_receive_skb(skb);
316 	return;
317 
318 frame_err:
319 	dev->stats.rx_frame_errors++;
320 	dev_kfree_skb(skb);
321 }
322 
323 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
324 {
325 	struct sk_buff *skb;
326 	struct skb_vnet_hdr *hdr;
327 	struct scatterlist sg[2];
328 	int err;
329 
330 	sg_init_table(sg, 2);
331 	skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
332 	if (unlikely(!skb))
333 		return -ENOMEM;
334 
335 	skb_put(skb, MAX_PACKET_LEN);
336 
337 	hdr = skb_vnet_hdr(skb);
338 	sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
339 
340 	skb_to_sgvec(skb, sg + 1, 0, skb->len);
341 
342 	err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 2, skb);
343 	if (err < 0)
344 		dev_kfree_skb(skb);
345 
346 	return err;
347 }
348 
349 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
350 {
351 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
352 	struct page *first, *list = NULL;
353 	char *p;
354 	int i, err, offset;
355 
356 	sg_init_table(sg, MAX_SKB_FRAGS + 2);
357 	/* page in sg[MAX_SKB_FRAGS + 1] is list tail */
358 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
359 		first = get_a_page(vi, gfp);
360 		if (!first) {
361 			if (list)
362 				give_pages(vi, list);
363 			return -ENOMEM;
364 		}
365 		sg_set_buf(&sg[i], page_address(first), PAGE_SIZE);
366 
367 		/* chain new page in list head to match sg */
368 		first->private = (unsigned long)list;
369 		list = first;
370 	}
371 
372 	first = get_a_page(vi, gfp);
373 	if (!first) {
374 		give_pages(vi, list);
375 		return -ENOMEM;
376 	}
377 	p = page_address(first);
378 
379 	/* sg[0], sg[1] share the same page */
380 	/* a separated sg[0] for  virtio_net_hdr only during to QEMU bug*/
381 	sg_set_buf(&sg[0], p, sizeof(struct virtio_net_hdr));
382 
383 	/* sg[1] for data packet, from offset */
384 	offset = sizeof(struct padded_vnet_hdr);
385 	sg_set_buf(&sg[1], p + offset, PAGE_SIZE - offset);
386 
387 	/* chain first in list head */
388 	first->private = (unsigned long)list;
389 	err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, MAX_SKB_FRAGS + 2,
390 				       first);
391 	if (err < 0)
392 		give_pages(vi, first);
393 
394 	return err;
395 }
396 
397 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
398 {
399 	struct page *page;
400 	struct scatterlist sg;
401 	int err;
402 
403 	page = get_a_page(vi, gfp);
404 	if (!page)
405 		return -ENOMEM;
406 
407 	sg_init_one(&sg, page_address(page), PAGE_SIZE);
408 
409 	err = vi->rvq->vq_ops->add_buf(vi->rvq, &sg, 0, 1, page);
410 	if (err < 0)
411 		give_pages(vi, page);
412 
413 	return err;
414 }
415 
416 /* Returns false if we couldn't fill entirely (OOM). */
417 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
418 {
419 	int err;
420 	bool oom = false;
421 
422 	do {
423 		if (vi->mergeable_rx_bufs)
424 			err = add_recvbuf_mergeable(vi, gfp);
425 		else if (vi->big_packets)
426 			err = add_recvbuf_big(vi, gfp);
427 		else
428 			err = add_recvbuf_small(vi, gfp);
429 
430 		if (err < 0) {
431 			oom = true;
432 			break;
433 		}
434 		++vi->num;
435 	} while (err > 0);
436 	if (unlikely(vi->num > vi->max))
437 		vi->max = vi->num;
438 	vi->rvq->vq_ops->kick(vi->rvq);
439 	return !oom;
440 }
441 
442 static void skb_recv_done(struct virtqueue *rvq)
443 {
444 	struct virtnet_info *vi = rvq->vdev->priv;
445 	/* Schedule NAPI, Suppress further interrupts if successful. */
446 	if (napi_schedule_prep(&vi->napi)) {
447 		rvq->vq_ops->disable_cb(rvq);
448 		__napi_schedule(&vi->napi);
449 	}
450 }
451 
452 static void refill_work(struct work_struct *work)
453 {
454 	struct virtnet_info *vi;
455 	bool still_empty;
456 
457 	vi = container_of(work, struct virtnet_info, refill.work);
458 	napi_disable(&vi->napi);
459 	still_empty = !try_fill_recv(vi, GFP_KERNEL);
460 	napi_enable(&vi->napi);
461 
462 	/* In theory, this can happen: if we don't get any buffers in
463 	 * we will *never* try to fill again. */
464 	if (still_empty)
465 		schedule_delayed_work(&vi->refill, HZ/2);
466 }
467 
468 static int virtnet_poll(struct napi_struct *napi, int budget)
469 {
470 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
471 	void *buf;
472 	unsigned int len, received = 0;
473 
474 again:
475 	while (received < budget &&
476 	       (buf = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
477 		receive_buf(vi->dev, buf, len);
478 		--vi->num;
479 		received++;
480 	}
481 
482 	if (vi->num < vi->max / 2) {
483 		if (!try_fill_recv(vi, GFP_ATOMIC))
484 			schedule_delayed_work(&vi->refill, 0);
485 	}
486 
487 	/* Out of packets? */
488 	if (received < budget) {
489 		napi_complete(napi);
490 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
491 		    napi_schedule_prep(napi)) {
492 			vi->rvq->vq_ops->disable_cb(vi->rvq);
493 			__napi_schedule(napi);
494 			goto again;
495 		}
496 	}
497 
498 	return received;
499 }
500 
501 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
502 {
503 	struct sk_buff *skb;
504 	unsigned int len, tot_sgs = 0;
505 
506 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
507 		pr_debug("Sent skb %p\n", skb);
508 		vi->dev->stats.tx_bytes += skb->len;
509 		vi->dev->stats.tx_packets++;
510 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
511 		dev_kfree_skb_any(skb);
512 	}
513 	return tot_sgs;
514 }
515 
516 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
517 {
518 	struct scatterlist sg[2+MAX_SKB_FRAGS];
519 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
520 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
521 
522 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
523 
524 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
525 
526 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
527 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
528 		hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
529 		hdr->hdr.csum_offset = skb->csum_offset;
530 	} else {
531 		hdr->hdr.flags = 0;
532 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
533 	}
534 
535 	if (skb_is_gso(skb)) {
536 		hdr->hdr.hdr_len = skb_headlen(skb);
537 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
538 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
539 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
540 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
541 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
542 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
543 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
544 		else
545 			BUG();
546 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
547 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
548 	} else {
549 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
550 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
551 	}
552 
553 	hdr->mhdr.num_buffers = 0;
554 
555 	/* Encode metadata header at front. */
556 	if (vi->mergeable_rx_bufs)
557 		sg_set_buf(sg, &hdr->mhdr, sizeof hdr->mhdr);
558 	else
559 		sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
560 
561 	hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
562 	return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
563 }
564 
565 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
566 {
567 	struct virtnet_info *vi = netdev_priv(dev);
568 	int capacity;
569 
570 again:
571 	/* Free up any pending old buffers before queueing new ones. */
572 	free_old_xmit_skbs(vi);
573 
574 	/* Try to transmit */
575 	capacity = xmit_skb(vi, skb);
576 
577 	/* This can happen with OOM and indirect buffers. */
578 	if (unlikely(capacity < 0)) {
579 		netif_stop_queue(dev);
580 		dev_warn(&dev->dev, "Unexpected full queue\n");
581 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
582 			vi->svq->vq_ops->disable_cb(vi->svq);
583 			netif_start_queue(dev);
584 			goto again;
585 		}
586 		return NETDEV_TX_BUSY;
587 	}
588 	vi->svq->vq_ops->kick(vi->svq);
589 
590 	/* Don't wait up for transmitted skbs to be freed. */
591 	skb_orphan(skb);
592 	nf_reset(skb);
593 
594 	/* Apparently nice girls don't return TX_BUSY; stop the queue
595 	 * before it gets out of hand.  Naturally, this wastes entries. */
596 	if (capacity < 2+MAX_SKB_FRAGS) {
597 		netif_stop_queue(dev);
598 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
599 			/* More just got used, free them then recheck. */
600 			capacity += free_old_xmit_skbs(vi);
601 			if (capacity >= 2+MAX_SKB_FRAGS) {
602 				netif_start_queue(dev);
603 				vi->svq->vq_ops->disable_cb(vi->svq);
604 			}
605 		}
606 	}
607 
608 	return NETDEV_TX_OK;
609 }
610 
611 static int virtnet_set_mac_address(struct net_device *dev, void *p)
612 {
613 	struct virtnet_info *vi = netdev_priv(dev);
614 	struct virtio_device *vdev = vi->vdev;
615 	int ret;
616 
617 	ret = eth_mac_addr(dev, p);
618 	if (ret)
619 		return ret;
620 
621 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
622 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
623 		                  dev->dev_addr, dev->addr_len);
624 
625 	return 0;
626 }
627 
628 #ifdef CONFIG_NET_POLL_CONTROLLER
629 static void virtnet_netpoll(struct net_device *dev)
630 {
631 	struct virtnet_info *vi = netdev_priv(dev);
632 
633 	napi_schedule(&vi->napi);
634 }
635 #endif
636 
637 static int virtnet_open(struct net_device *dev)
638 {
639 	struct virtnet_info *vi = netdev_priv(dev);
640 
641 	napi_enable(&vi->napi);
642 
643 	/* If all buffers were filled by other side before we napi_enabled, we
644 	 * won't get another interrupt, so process any outstanding packets
645 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
646 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
647 	if (napi_schedule_prep(&vi->napi)) {
648 		vi->rvq->vq_ops->disable_cb(vi->rvq);
649 		__napi_schedule(&vi->napi);
650 	}
651 	return 0;
652 }
653 
654 /*
655  * Send command via the control virtqueue and check status.  Commands
656  * supported by the hypervisor, as indicated by feature bits, should
657  * never fail unless improperly formated.
658  */
659 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
660 				 struct scatterlist *data, int out, int in)
661 {
662 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
663 	struct virtio_net_ctrl_hdr ctrl;
664 	virtio_net_ctrl_ack status = ~0;
665 	unsigned int tmp;
666 	int i;
667 
668 	/* Caller should know better */
669 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
670 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
671 
672 	out++; /* Add header */
673 	in++; /* Add return status */
674 
675 	ctrl.class = class;
676 	ctrl.cmd = cmd;
677 
678 	sg_init_table(sg, out + in);
679 
680 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
681 	for_each_sg(data, s, out + in - 2, i)
682 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
683 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
684 
685 	BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
686 
687 	vi->cvq->vq_ops->kick(vi->cvq);
688 
689 	/*
690 	 * Spin for a response, the kick causes an ioport write, trapping
691 	 * into the hypervisor, so the request should be handled immediately.
692 	 */
693 	while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
694 		cpu_relax();
695 
696 	return status == VIRTIO_NET_OK;
697 }
698 
699 static int virtnet_close(struct net_device *dev)
700 {
701 	struct virtnet_info *vi = netdev_priv(dev);
702 
703 	napi_disable(&vi->napi);
704 
705 	return 0;
706 }
707 
708 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
709 {
710 	struct virtnet_info *vi = netdev_priv(dev);
711 	struct virtio_device *vdev = vi->vdev;
712 
713 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
714 		return -ENOSYS;
715 
716 	return ethtool_op_set_tx_hw_csum(dev, data);
717 }
718 
719 static void virtnet_set_rx_mode(struct net_device *dev)
720 {
721 	struct virtnet_info *vi = netdev_priv(dev);
722 	struct scatterlist sg[2];
723 	u8 promisc, allmulti;
724 	struct virtio_net_ctrl_mac *mac_data;
725 	struct dev_addr_list *addr;
726 	struct netdev_hw_addr *ha;
727 	int uc_count;
728 	int mc_count;
729 	void *buf;
730 	int i;
731 
732 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
733 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
734 		return;
735 
736 	promisc = ((dev->flags & IFF_PROMISC) != 0);
737 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
738 
739 	sg_init_one(sg, &promisc, sizeof(promisc));
740 
741 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
742 				  VIRTIO_NET_CTRL_RX_PROMISC,
743 				  sg, 1, 0))
744 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
745 			 promisc ? "en" : "dis");
746 
747 	sg_init_one(sg, &allmulti, sizeof(allmulti));
748 
749 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
750 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
751 				  sg, 1, 0))
752 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
753 			 allmulti ? "en" : "dis");
754 
755 	uc_count = netdev_uc_count(dev);
756 	mc_count = netdev_mc_count(dev);
757 	/* MAC filter - use one buffer for both lists */
758 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
759 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
760 	mac_data = buf;
761 	if (!buf) {
762 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
763 		return;
764 	}
765 
766 	sg_init_table(sg, 2);
767 
768 	/* Store the unicast list and count in the front of the buffer */
769 	mac_data->entries = uc_count;
770 	i = 0;
771 	netdev_for_each_uc_addr(ha, dev)
772 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
773 
774 	sg_set_buf(&sg[0], mac_data,
775 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
776 
777 	/* multicast list and count fill the end */
778 	mac_data = (void *)&mac_data->macs[uc_count][0];
779 
780 	mac_data->entries = mc_count;
781 	i = 0;
782 	netdev_for_each_mc_addr(addr, dev)
783 		memcpy(&mac_data->macs[i++][0], addr->da_addr, ETH_ALEN);
784 
785 	sg_set_buf(&sg[1], mac_data,
786 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
787 
788 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
789 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
790 				  sg, 2, 0))
791 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
792 
793 	kfree(buf);
794 }
795 
796 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
797 {
798 	struct virtnet_info *vi = netdev_priv(dev);
799 	struct scatterlist sg;
800 
801 	sg_init_one(&sg, &vid, sizeof(vid));
802 
803 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
804 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
805 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
806 }
807 
808 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
809 {
810 	struct virtnet_info *vi = netdev_priv(dev);
811 	struct scatterlist sg;
812 
813 	sg_init_one(&sg, &vid, sizeof(vid));
814 
815 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
816 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
817 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
818 }
819 
820 static const struct ethtool_ops virtnet_ethtool_ops = {
821 	.set_tx_csum = virtnet_set_tx_csum,
822 	.set_sg = ethtool_op_set_sg,
823 	.set_tso = ethtool_op_set_tso,
824 	.set_ufo = ethtool_op_set_ufo,
825 	.get_link = ethtool_op_get_link,
826 };
827 
828 #define MIN_MTU 68
829 #define MAX_MTU 65535
830 
831 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
832 {
833 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
834 		return -EINVAL;
835 	dev->mtu = new_mtu;
836 	return 0;
837 }
838 
839 static const struct net_device_ops virtnet_netdev = {
840 	.ndo_open            = virtnet_open,
841 	.ndo_stop   	     = virtnet_close,
842 	.ndo_start_xmit      = start_xmit,
843 	.ndo_validate_addr   = eth_validate_addr,
844 	.ndo_set_mac_address = virtnet_set_mac_address,
845 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
846 	.ndo_change_mtu	     = virtnet_change_mtu,
847 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
848 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
849 #ifdef CONFIG_NET_POLL_CONTROLLER
850 	.ndo_poll_controller = virtnet_netpoll,
851 #endif
852 };
853 
854 static void virtnet_update_status(struct virtnet_info *vi)
855 {
856 	u16 v;
857 
858 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
859 		return;
860 
861 	vi->vdev->config->get(vi->vdev,
862 			      offsetof(struct virtio_net_config, status),
863 			      &v, sizeof(v));
864 
865 	/* Ignore unknown (future) status bits */
866 	v &= VIRTIO_NET_S_LINK_UP;
867 
868 	if (vi->status == v)
869 		return;
870 
871 	vi->status = v;
872 
873 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
874 		netif_carrier_on(vi->dev);
875 		netif_wake_queue(vi->dev);
876 	} else {
877 		netif_carrier_off(vi->dev);
878 		netif_stop_queue(vi->dev);
879 	}
880 }
881 
882 static void virtnet_config_changed(struct virtio_device *vdev)
883 {
884 	struct virtnet_info *vi = vdev->priv;
885 
886 	virtnet_update_status(vi);
887 }
888 
889 static int virtnet_probe(struct virtio_device *vdev)
890 {
891 	int err;
892 	struct net_device *dev;
893 	struct virtnet_info *vi;
894 	struct virtqueue *vqs[3];
895 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
896 	const char *names[] = { "input", "output", "control" };
897 	int nvqs;
898 
899 	/* Allocate ourselves a network device with room for our info */
900 	dev = alloc_etherdev(sizeof(struct virtnet_info));
901 	if (!dev)
902 		return -ENOMEM;
903 
904 	/* Set up network device as normal. */
905 	dev->netdev_ops = &virtnet_netdev;
906 	dev->features = NETIF_F_HIGHDMA;
907 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
908 	SET_NETDEV_DEV(dev, &vdev->dev);
909 
910 	/* Do we support "hardware" checksums? */
911 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
912 		/* This opens up the world of extra features. */
913 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
914 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
915 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
916 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
917 		}
918 		/* Individual feature bits: what can host handle? */
919 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
920 			dev->features |= NETIF_F_TSO;
921 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
922 			dev->features |= NETIF_F_TSO6;
923 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
924 			dev->features |= NETIF_F_TSO_ECN;
925 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
926 			dev->features |= NETIF_F_UFO;
927 	}
928 
929 	/* Configuration may specify what MAC to use.  Otherwise random. */
930 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
931 		vdev->config->get(vdev,
932 				  offsetof(struct virtio_net_config, mac),
933 				  dev->dev_addr, dev->addr_len);
934 	} else
935 		random_ether_addr(dev->dev_addr);
936 
937 	/* Set up our device-specific information */
938 	vi = netdev_priv(dev);
939 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
940 	vi->dev = dev;
941 	vi->vdev = vdev;
942 	vdev->priv = vi;
943 	vi->pages = NULL;
944 	INIT_DELAYED_WORK(&vi->refill, refill_work);
945 
946 	/* If we can receive ANY GSO packets, we must allocate large ones. */
947 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
948 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
949 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
950 		vi->big_packets = true;
951 
952 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
953 		vi->mergeable_rx_bufs = true;
954 
955 	/* We expect two virtqueues, receive then send,
956 	 * and optionally control. */
957 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
958 
959 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
960 	if (err)
961 		goto free;
962 
963 	vi->rvq = vqs[0];
964 	vi->svq = vqs[1];
965 
966 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
967 		vi->cvq = vqs[2];
968 
969 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
970 			dev->features |= NETIF_F_HW_VLAN_FILTER;
971 	}
972 
973 	err = register_netdev(dev);
974 	if (err) {
975 		pr_debug("virtio_net: registering device failed\n");
976 		goto free_vqs;
977 	}
978 
979 	/* Last of all, set up some receive buffers. */
980 	try_fill_recv(vi, GFP_KERNEL);
981 
982 	/* If we didn't even get one input buffer, we're useless. */
983 	if (vi->num == 0) {
984 		err = -ENOMEM;
985 		goto unregister;
986 	}
987 
988 	vi->status = VIRTIO_NET_S_LINK_UP;
989 	virtnet_update_status(vi);
990 	netif_carrier_on(dev);
991 
992 	pr_debug("virtnet: registered device %s\n", dev->name);
993 	return 0;
994 
995 unregister:
996 	unregister_netdev(dev);
997 	cancel_delayed_work_sync(&vi->refill);
998 free_vqs:
999 	vdev->config->del_vqs(vdev);
1000 free:
1001 	free_netdev(dev);
1002 	return err;
1003 }
1004 
1005 static void free_unused_bufs(struct virtnet_info *vi)
1006 {
1007 	void *buf;
1008 	while (1) {
1009 		buf = vi->svq->vq_ops->detach_unused_buf(vi->svq);
1010 		if (!buf)
1011 			break;
1012 		dev_kfree_skb(buf);
1013 	}
1014 	while (1) {
1015 		buf = vi->rvq->vq_ops->detach_unused_buf(vi->rvq);
1016 		if (!buf)
1017 			break;
1018 		if (vi->mergeable_rx_bufs || vi->big_packets)
1019 			give_pages(vi, buf);
1020 		else
1021 			dev_kfree_skb(buf);
1022 		--vi->num;
1023 	}
1024 	BUG_ON(vi->num != 0);
1025 }
1026 
1027 static void __devexit virtnet_remove(struct virtio_device *vdev)
1028 {
1029 	struct virtnet_info *vi = vdev->priv;
1030 
1031 	/* Stop all the virtqueues. */
1032 	vdev->config->reset(vdev);
1033 
1034 
1035 	unregister_netdev(vi->dev);
1036 	cancel_delayed_work_sync(&vi->refill);
1037 
1038 	/* Free unused buffers in both send and recv, if any. */
1039 	free_unused_bufs(vi);
1040 
1041 	vdev->config->del_vqs(vi->vdev);
1042 
1043 	while (vi->pages)
1044 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
1045 
1046 	free_netdev(vi->dev);
1047 }
1048 
1049 static struct virtio_device_id id_table[] = {
1050 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1051 	{ 0 },
1052 };
1053 
1054 static unsigned int features[] = {
1055 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1056 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1057 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1058 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1059 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1060 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1061 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1062 };
1063 
1064 static struct virtio_driver virtio_net_driver = {
1065 	.feature_table = features,
1066 	.feature_table_size = ARRAY_SIZE(features),
1067 	.driver.name =	KBUILD_MODNAME,
1068 	.driver.owner =	THIS_MODULE,
1069 	.id_table =	id_table,
1070 	.probe =	virtnet_probe,
1071 	.remove =	__devexit_p(virtnet_remove),
1072 	.config_changed = virtnet_config_changed,
1073 };
1074 
1075 static int __init init(void)
1076 {
1077 	return register_virtio_driver(&virtio_net_driver);
1078 }
1079 
1080 static void __exit fini(void)
1081 {
1082 	unregister_virtio_driver(&virtio_net_driver);
1083 }
1084 module_init(init);
1085 module_exit(fini);
1086 
1087 MODULE_DEVICE_TABLE(virtio, id_table);
1088 MODULE_DESCRIPTION("Virtio network driver");
1089 MODULE_LICENSE("GPL");
1090