xref: /linux/drivers/net/virtio_net.c (revision 4dc7ccf7e9d9bca1989b840be9e8e84911387cf2)
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 	skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
331 	if (unlikely(!skb))
332 		return -ENOMEM;
333 
334 	skb_put(skb, MAX_PACKET_LEN);
335 
336 	hdr = skb_vnet_hdr(skb);
337 	sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
338 
339 	skb_to_sgvec(skb, sg + 1, 0, skb->len);
340 
341 	err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 2, skb);
342 	if (err < 0)
343 		dev_kfree_skb(skb);
344 
345 	return err;
346 }
347 
348 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
349 {
350 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
351 	struct page *first, *list = NULL;
352 	char *p;
353 	int i, err, offset;
354 
355 	/* page in sg[MAX_SKB_FRAGS + 1] is list tail */
356 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
357 		first = get_a_page(vi, gfp);
358 		if (!first) {
359 			if (list)
360 				give_pages(vi, list);
361 			return -ENOMEM;
362 		}
363 		sg_set_buf(&sg[i], page_address(first), PAGE_SIZE);
364 
365 		/* chain new page in list head to match sg */
366 		first->private = (unsigned long)list;
367 		list = first;
368 	}
369 
370 	first = get_a_page(vi, gfp);
371 	if (!first) {
372 		give_pages(vi, list);
373 		return -ENOMEM;
374 	}
375 	p = page_address(first);
376 
377 	/* sg[0], sg[1] share the same page */
378 	/* a separated sg[0] for  virtio_net_hdr only during to QEMU bug*/
379 	sg_set_buf(&sg[0], p, sizeof(struct virtio_net_hdr));
380 
381 	/* sg[1] for data packet, from offset */
382 	offset = sizeof(struct padded_vnet_hdr);
383 	sg_set_buf(&sg[1], p + offset, PAGE_SIZE - offset);
384 
385 	/* chain first in list head */
386 	first->private = (unsigned long)list;
387 	err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, MAX_SKB_FRAGS + 2,
388 				       first);
389 	if (err < 0)
390 		give_pages(vi, first);
391 
392 	return err;
393 }
394 
395 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
396 {
397 	struct page *page;
398 	struct scatterlist sg;
399 	int err;
400 
401 	page = get_a_page(vi, gfp);
402 	if (!page)
403 		return -ENOMEM;
404 
405 	sg_init_one(&sg, page_address(page), PAGE_SIZE);
406 
407 	err = vi->rvq->vq_ops->add_buf(vi->rvq, &sg, 0, 1, page);
408 	if (err < 0)
409 		give_pages(vi, page);
410 
411 	return err;
412 }
413 
414 /* Returns false if we couldn't fill entirely (OOM). */
415 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
416 {
417 	int err;
418 	bool oom = false;
419 
420 	do {
421 		if (vi->mergeable_rx_bufs)
422 			err = add_recvbuf_mergeable(vi, gfp);
423 		else if (vi->big_packets)
424 			err = add_recvbuf_big(vi, gfp);
425 		else
426 			err = add_recvbuf_small(vi, gfp);
427 
428 		if (err < 0) {
429 			oom = true;
430 			break;
431 		}
432 		++vi->num;
433 	} while (err > 0);
434 	if (unlikely(vi->num > vi->max))
435 		vi->max = vi->num;
436 	vi->rvq->vq_ops->kick(vi->rvq);
437 	return !oom;
438 }
439 
440 static void skb_recv_done(struct virtqueue *rvq)
441 {
442 	struct virtnet_info *vi = rvq->vdev->priv;
443 	/* Schedule NAPI, Suppress further interrupts if successful. */
444 	if (napi_schedule_prep(&vi->napi)) {
445 		rvq->vq_ops->disable_cb(rvq);
446 		__napi_schedule(&vi->napi);
447 	}
448 }
449 
450 static void refill_work(struct work_struct *work)
451 {
452 	struct virtnet_info *vi;
453 	bool still_empty;
454 
455 	vi = container_of(work, struct virtnet_info, refill.work);
456 	napi_disable(&vi->napi);
457 	still_empty = !try_fill_recv(vi, GFP_KERNEL);
458 	napi_enable(&vi->napi);
459 
460 	/* In theory, this can happen: if we don't get any buffers in
461 	 * we will *never* try to fill again. */
462 	if (still_empty)
463 		schedule_delayed_work(&vi->refill, HZ/2);
464 }
465 
466 static int virtnet_poll(struct napi_struct *napi, int budget)
467 {
468 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
469 	void *buf;
470 	unsigned int len, received = 0;
471 
472 again:
473 	while (received < budget &&
474 	       (buf = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
475 		receive_buf(vi->dev, buf, len);
476 		--vi->num;
477 		received++;
478 	}
479 
480 	if (vi->num < vi->max / 2) {
481 		if (!try_fill_recv(vi, GFP_ATOMIC))
482 			schedule_delayed_work(&vi->refill, 0);
483 	}
484 
485 	/* Out of packets? */
486 	if (received < budget) {
487 		napi_complete(napi);
488 		if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
489 		    napi_schedule_prep(napi)) {
490 			vi->rvq->vq_ops->disable_cb(vi->rvq);
491 			__napi_schedule(napi);
492 			goto again;
493 		}
494 	}
495 
496 	return received;
497 }
498 
499 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
500 {
501 	struct sk_buff *skb;
502 	unsigned int len, tot_sgs = 0;
503 
504 	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
505 		pr_debug("Sent skb %p\n", skb);
506 		vi->dev->stats.tx_bytes += skb->len;
507 		vi->dev->stats.tx_packets++;
508 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
509 		dev_kfree_skb_any(skb);
510 	}
511 	return tot_sgs;
512 }
513 
514 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
515 {
516 	struct scatterlist sg[2+MAX_SKB_FRAGS];
517 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
518 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
519 
520 	sg_init_table(sg, 2+MAX_SKB_FRAGS);
521 
522 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
523 
524 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
525 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
526 		hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
527 		hdr->hdr.csum_offset = skb->csum_offset;
528 	} else {
529 		hdr->hdr.flags = 0;
530 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
531 	}
532 
533 	if (skb_is_gso(skb)) {
534 		hdr->hdr.hdr_len = skb_headlen(skb);
535 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
536 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
537 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
538 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
539 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
540 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
541 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
542 		else
543 			BUG();
544 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
545 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
546 	} else {
547 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
548 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
549 	}
550 
551 	hdr->mhdr.num_buffers = 0;
552 
553 	/* Encode metadata header at front. */
554 	if (vi->mergeable_rx_bufs)
555 		sg_set_buf(sg, &hdr->mhdr, sizeof hdr->mhdr);
556 	else
557 		sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
558 
559 	hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
560 	return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
561 }
562 
563 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
564 {
565 	struct virtnet_info *vi = netdev_priv(dev);
566 	int capacity;
567 
568 again:
569 	/* Free up any pending old buffers before queueing new ones. */
570 	free_old_xmit_skbs(vi);
571 
572 	/* Try to transmit */
573 	capacity = xmit_skb(vi, skb);
574 
575 	/* This can happen with OOM and indirect buffers. */
576 	if (unlikely(capacity < 0)) {
577 		netif_stop_queue(dev);
578 		dev_warn(&dev->dev, "Unexpected full queue\n");
579 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
580 			vi->svq->vq_ops->disable_cb(vi->svq);
581 			netif_start_queue(dev);
582 			goto again;
583 		}
584 		return NETDEV_TX_BUSY;
585 	}
586 	vi->svq->vq_ops->kick(vi->svq);
587 
588 	/* Don't wait up for transmitted skbs to be freed. */
589 	skb_orphan(skb);
590 	nf_reset(skb);
591 
592 	/* Apparently nice girls don't return TX_BUSY; stop the queue
593 	 * before it gets out of hand.  Naturally, this wastes entries. */
594 	if (capacity < 2+MAX_SKB_FRAGS) {
595 		netif_stop_queue(dev);
596 		if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
597 			/* More just got used, free them then recheck. */
598 			capacity += free_old_xmit_skbs(vi);
599 			if (capacity >= 2+MAX_SKB_FRAGS) {
600 				netif_start_queue(dev);
601 				vi->svq->vq_ops->disable_cb(vi->svq);
602 			}
603 		}
604 	}
605 
606 	return NETDEV_TX_OK;
607 }
608 
609 static int virtnet_set_mac_address(struct net_device *dev, void *p)
610 {
611 	struct virtnet_info *vi = netdev_priv(dev);
612 	struct virtio_device *vdev = vi->vdev;
613 	int ret;
614 
615 	ret = eth_mac_addr(dev, p);
616 	if (ret)
617 		return ret;
618 
619 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
620 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
621 		                  dev->dev_addr, dev->addr_len);
622 
623 	return 0;
624 }
625 
626 #ifdef CONFIG_NET_POLL_CONTROLLER
627 static void virtnet_netpoll(struct net_device *dev)
628 {
629 	struct virtnet_info *vi = netdev_priv(dev);
630 
631 	napi_schedule(&vi->napi);
632 }
633 #endif
634 
635 static int virtnet_open(struct net_device *dev)
636 {
637 	struct virtnet_info *vi = netdev_priv(dev);
638 
639 	napi_enable(&vi->napi);
640 
641 	/* If all buffers were filled by other side before we napi_enabled, we
642 	 * won't get another interrupt, so process any outstanding packets
643 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
644 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
645 	if (napi_schedule_prep(&vi->napi)) {
646 		vi->rvq->vq_ops->disable_cb(vi->rvq);
647 		__napi_schedule(&vi->napi);
648 	}
649 	return 0;
650 }
651 
652 /*
653  * Send command via the control virtqueue and check status.  Commands
654  * supported by the hypervisor, as indicated by feature bits, should
655  * never fail unless improperly formated.
656  */
657 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
658 				 struct scatterlist *data, int out, int in)
659 {
660 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
661 	struct virtio_net_ctrl_hdr ctrl;
662 	virtio_net_ctrl_ack status = ~0;
663 	unsigned int tmp;
664 	int i;
665 
666 	/* Caller should know better */
667 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
668 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
669 
670 	out++; /* Add header */
671 	in++; /* Add return status */
672 
673 	ctrl.class = class;
674 	ctrl.cmd = cmd;
675 
676 	sg_init_table(sg, out + in);
677 
678 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
679 	for_each_sg(data, s, out + in - 2, i)
680 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
681 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
682 
683 	BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
684 
685 	vi->cvq->vq_ops->kick(vi->cvq);
686 
687 	/*
688 	 * Spin for a response, the kick causes an ioport write, trapping
689 	 * into the hypervisor, so the request should be handled immediately.
690 	 */
691 	while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
692 		cpu_relax();
693 
694 	return status == VIRTIO_NET_OK;
695 }
696 
697 static int virtnet_close(struct net_device *dev)
698 {
699 	struct virtnet_info *vi = netdev_priv(dev);
700 
701 	napi_disable(&vi->napi);
702 
703 	return 0;
704 }
705 
706 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
707 {
708 	struct virtnet_info *vi = netdev_priv(dev);
709 	struct virtio_device *vdev = vi->vdev;
710 
711 	if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
712 		return -ENOSYS;
713 
714 	return ethtool_op_set_tx_hw_csum(dev, data);
715 }
716 
717 static void virtnet_set_rx_mode(struct net_device *dev)
718 {
719 	struct virtnet_info *vi = netdev_priv(dev);
720 	struct scatterlist sg[2];
721 	u8 promisc, allmulti;
722 	struct virtio_net_ctrl_mac *mac_data;
723 	struct dev_addr_list *addr;
724 	struct netdev_hw_addr *ha;
725 	int uc_count;
726 	int mc_count;
727 	void *buf;
728 	int i;
729 
730 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
731 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
732 		return;
733 
734 	promisc = ((dev->flags & IFF_PROMISC) != 0);
735 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
736 
737 	sg_init_one(sg, &promisc, sizeof(promisc));
738 
739 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
740 				  VIRTIO_NET_CTRL_RX_PROMISC,
741 				  sg, 1, 0))
742 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
743 			 promisc ? "en" : "dis");
744 
745 	sg_init_one(sg, &allmulti, sizeof(allmulti));
746 
747 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
748 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
749 				  sg, 1, 0))
750 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
751 			 allmulti ? "en" : "dis");
752 
753 	uc_count = netdev_uc_count(dev);
754 	mc_count = netdev_mc_count(dev);
755 	/* MAC filter - use one buffer for both lists */
756 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
757 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
758 	mac_data = buf;
759 	if (!buf) {
760 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
761 		return;
762 	}
763 
764 	sg_init_table(sg, 2);
765 
766 	/* Store the unicast list and count in the front of the buffer */
767 	mac_data->entries = uc_count;
768 	i = 0;
769 	netdev_for_each_uc_addr(ha, dev)
770 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
771 
772 	sg_set_buf(&sg[0], mac_data,
773 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
774 
775 	/* multicast list and count fill the end */
776 	mac_data = (void *)&mac_data->macs[uc_count][0];
777 
778 	mac_data->entries = mc_count;
779 	i = 0;
780 	netdev_for_each_mc_addr(addr, dev)
781 		memcpy(&mac_data->macs[i++][0], addr->da_addr, ETH_ALEN);
782 
783 	sg_set_buf(&sg[1], mac_data,
784 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
785 
786 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
787 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
788 				  sg, 2, 0))
789 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
790 
791 	kfree(buf);
792 }
793 
794 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
795 {
796 	struct virtnet_info *vi = netdev_priv(dev);
797 	struct scatterlist sg;
798 
799 	sg_init_one(&sg, &vid, sizeof(vid));
800 
801 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
802 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
803 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
804 }
805 
806 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
807 {
808 	struct virtnet_info *vi = netdev_priv(dev);
809 	struct scatterlist sg;
810 
811 	sg_init_one(&sg, &vid, sizeof(vid));
812 
813 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
814 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
815 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
816 }
817 
818 static const struct ethtool_ops virtnet_ethtool_ops = {
819 	.set_tx_csum = virtnet_set_tx_csum,
820 	.set_sg = ethtool_op_set_sg,
821 	.set_tso = ethtool_op_set_tso,
822 	.set_ufo = ethtool_op_set_ufo,
823 	.get_link = ethtool_op_get_link,
824 };
825 
826 #define MIN_MTU 68
827 #define MAX_MTU 65535
828 
829 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
830 {
831 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
832 		return -EINVAL;
833 	dev->mtu = new_mtu;
834 	return 0;
835 }
836 
837 static const struct net_device_ops virtnet_netdev = {
838 	.ndo_open            = virtnet_open,
839 	.ndo_stop   	     = virtnet_close,
840 	.ndo_start_xmit      = start_xmit,
841 	.ndo_validate_addr   = eth_validate_addr,
842 	.ndo_set_mac_address = virtnet_set_mac_address,
843 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
844 	.ndo_change_mtu	     = virtnet_change_mtu,
845 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
846 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
847 #ifdef CONFIG_NET_POLL_CONTROLLER
848 	.ndo_poll_controller = virtnet_netpoll,
849 #endif
850 };
851 
852 static void virtnet_update_status(struct virtnet_info *vi)
853 {
854 	u16 v;
855 
856 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
857 		return;
858 
859 	vi->vdev->config->get(vi->vdev,
860 			      offsetof(struct virtio_net_config, status),
861 			      &v, sizeof(v));
862 
863 	/* Ignore unknown (future) status bits */
864 	v &= VIRTIO_NET_S_LINK_UP;
865 
866 	if (vi->status == v)
867 		return;
868 
869 	vi->status = v;
870 
871 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
872 		netif_carrier_on(vi->dev);
873 		netif_wake_queue(vi->dev);
874 	} else {
875 		netif_carrier_off(vi->dev);
876 		netif_stop_queue(vi->dev);
877 	}
878 }
879 
880 static void virtnet_config_changed(struct virtio_device *vdev)
881 {
882 	struct virtnet_info *vi = vdev->priv;
883 
884 	virtnet_update_status(vi);
885 }
886 
887 static int virtnet_probe(struct virtio_device *vdev)
888 {
889 	int err;
890 	struct net_device *dev;
891 	struct virtnet_info *vi;
892 	struct virtqueue *vqs[3];
893 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
894 	const char *names[] = { "input", "output", "control" };
895 	int nvqs;
896 
897 	/* Allocate ourselves a network device with room for our info */
898 	dev = alloc_etherdev(sizeof(struct virtnet_info));
899 	if (!dev)
900 		return -ENOMEM;
901 
902 	/* Set up network device as normal. */
903 	dev->netdev_ops = &virtnet_netdev;
904 	dev->features = NETIF_F_HIGHDMA;
905 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
906 	SET_NETDEV_DEV(dev, &vdev->dev);
907 
908 	/* Do we support "hardware" checksums? */
909 	if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
910 		/* This opens up the world of extra features. */
911 		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
912 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
913 			dev->features |= NETIF_F_TSO | NETIF_F_UFO
914 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
915 		}
916 		/* Individual feature bits: what can host handle? */
917 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
918 			dev->features |= NETIF_F_TSO;
919 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
920 			dev->features |= NETIF_F_TSO6;
921 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
922 			dev->features |= NETIF_F_TSO_ECN;
923 		if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
924 			dev->features |= NETIF_F_UFO;
925 	}
926 
927 	/* Configuration may specify what MAC to use.  Otherwise random. */
928 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
929 		vdev->config->get(vdev,
930 				  offsetof(struct virtio_net_config, mac),
931 				  dev->dev_addr, dev->addr_len);
932 	} else
933 		random_ether_addr(dev->dev_addr);
934 
935 	/* Set up our device-specific information */
936 	vi = netdev_priv(dev);
937 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
938 	vi->dev = dev;
939 	vi->vdev = vdev;
940 	vdev->priv = vi;
941 	vi->pages = NULL;
942 	INIT_DELAYED_WORK(&vi->refill, refill_work);
943 
944 	/* If we can receive ANY GSO packets, we must allocate large ones. */
945 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
946 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
947 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
948 		vi->big_packets = true;
949 
950 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
951 		vi->mergeable_rx_bufs = true;
952 
953 	/* We expect two virtqueues, receive then send,
954 	 * and optionally control. */
955 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
956 
957 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
958 	if (err)
959 		goto free;
960 
961 	vi->rvq = vqs[0];
962 	vi->svq = vqs[1];
963 
964 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
965 		vi->cvq = vqs[2];
966 
967 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
968 			dev->features |= NETIF_F_HW_VLAN_FILTER;
969 	}
970 
971 	err = register_netdev(dev);
972 	if (err) {
973 		pr_debug("virtio_net: registering device failed\n");
974 		goto free_vqs;
975 	}
976 
977 	/* Last of all, set up some receive buffers. */
978 	try_fill_recv(vi, GFP_KERNEL);
979 
980 	/* If we didn't even get one input buffer, we're useless. */
981 	if (vi->num == 0) {
982 		err = -ENOMEM;
983 		goto unregister;
984 	}
985 
986 	vi->status = VIRTIO_NET_S_LINK_UP;
987 	virtnet_update_status(vi);
988 	netif_carrier_on(dev);
989 
990 	pr_debug("virtnet: registered device %s\n", dev->name);
991 	return 0;
992 
993 unregister:
994 	unregister_netdev(dev);
995 	cancel_delayed_work_sync(&vi->refill);
996 free_vqs:
997 	vdev->config->del_vqs(vdev);
998 free:
999 	free_netdev(dev);
1000 	return err;
1001 }
1002 
1003 static void free_unused_bufs(struct virtnet_info *vi)
1004 {
1005 	void *buf;
1006 	while (1) {
1007 		buf = vi->svq->vq_ops->detach_unused_buf(vi->svq);
1008 		if (!buf)
1009 			break;
1010 		dev_kfree_skb(buf);
1011 	}
1012 	while (1) {
1013 		buf = vi->rvq->vq_ops->detach_unused_buf(vi->rvq);
1014 		if (!buf)
1015 			break;
1016 		if (vi->mergeable_rx_bufs || vi->big_packets)
1017 			give_pages(vi, buf);
1018 		else
1019 			dev_kfree_skb(buf);
1020 		--vi->num;
1021 	}
1022 	BUG_ON(vi->num != 0);
1023 }
1024 
1025 static void __devexit virtnet_remove(struct virtio_device *vdev)
1026 {
1027 	struct virtnet_info *vi = vdev->priv;
1028 
1029 	/* Stop all the virtqueues. */
1030 	vdev->config->reset(vdev);
1031 
1032 
1033 	unregister_netdev(vi->dev);
1034 	cancel_delayed_work_sync(&vi->refill);
1035 
1036 	/* Free unused buffers in both send and recv, if any. */
1037 	free_unused_bufs(vi);
1038 
1039 	vdev->config->del_vqs(vi->vdev);
1040 
1041 	while (vi->pages)
1042 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
1043 
1044 	free_netdev(vi->dev);
1045 }
1046 
1047 static struct virtio_device_id id_table[] = {
1048 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1049 	{ 0 },
1050 };
1051 
1052 static unsigned int features[] = {
1053 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1054 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1055 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1056 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1057 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1058 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1059 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1060 };
1061 
1062 static struct virtio_driver virtio_net_driver = {
1063 	.feature_table = features,
1064 	.feature_table_size = ARRAY_SIZE(features),
1065 	.driver.name =	KBUILD_MODNAME,
1066 	.driver.owner =	THIS_MODULE,
1067 	.id_table =	id_table,
1068 	.probe =	virtnet_probe,
1069 	.remove =	__devexit_p(virtnet_remove),
1070 	.config_changed = virtnet_config_changed,
1071 };
1072 
1073 static int __init init(void)
1074 {
1075 	return register_virtio_driver(&virtio_net_driver);
1076 }
1077 
1078 static void __exit fini(void)
1079 {
1080 	unregister_virtio_driver(&virtio_net_driver);
1081 }
1082 module_init(init);
1083 module_exit(fini);
1084 
1085 MODULE_DEVICE_TABLE(virtio, id_table);
1086 MODULE_DESCRIPTION("Virtio network driver");
1087 MODULE_LICENSE("GPL");
1088