xref: /linux/drivers/net/virtio_net.c (revision 9e8ba5f3ec35cba4fd8a8bebda548c4db2651e40)
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 bool csum = true, gso = true;
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 #define VIRTNET_DRIVER_VERSION "1.0.0"
43 
44 struct virtnet_stats {
45 	struct u64_stats_sync syncp;
46 	u64 tx_bytes;
47 	u64 tx_packets;
48 
49 	u64 rx_bytes;
50 	u64 rx_packets;
51 };
52 
53 struct virtnet_info {
54 	struct virtio_device *vdev;
55 	struct virtqueue *rvq, *svq, *cvq;
56 	struct net_device *dev;
57 	struct napi_struct napi;
58 	unsigned int status;
59 
60 	/* Number of input buffers, and max we've ever had. */
61 	unsigned int num, max;
62 
63 	/* I like... big packets and I cannot lie! */
64 	bool big_packets;
65 
66 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
67 	bool mergeable_rx_bufs;
68 
69 	/* Active statistics */
70 	struct virtnet_stats __percpu *stats;
71 
72 	/* Work struct for refilling if we run low on memory. */
73 	struct delayed_work refill;
74 
75 	/* Chain pages by the private ptr. */
76 	struct page *pages;
77 
78 	/* fragments + linear part + virtio header */
79 	struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
80 	struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
81 };
82 
83 struct skb_vnet_hdr {
84 	union {
85 		struct virtio_net_hdr hdr;
86 		struct virtio_net_hdr_mrg_rxbuf mhdr;
87 	};
88 	unsigned int num_sg;
89 };
90 
91 struct padded_vnet_hdr {
92 	struct virtio_net_hdr hdr;
93 	/*
94 	 * virtio_net_hdr should be in a separated sg buffer because of a
95 	 * QEMU bug, and data sg buffer shares same page with this header sg.
96 	 * This padding makes next sg 16 byte aligned after virtio_net_hdr.
97 	 */
98 	char padding[6];
99 };
100 
101 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
102 {
103 	return (struct skb_vnet_hdr *)skb->cb;
104 }
105 
106 /*
107  * private is used to chain pages for big packets, put the whole
108  * most recent used list in the beginning for reuse
109  */
110 static void give_pages(struct virtnet_info *vi, struct page *page)
111 {
112 	struct page *end;
113 
114 	/* Find end of list, sew whole thing into vi->pages. */
115 	for (end = page; end->private; end = (struct page *)end->private);
116 	end->private = (unsigned long)vi->pages;
117 	vi->pages = page;
118 }
119 
120 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
121 {
122 	struct page *p = vi->pages;
123 
124 	if (p) {
125 		vi->pages = (struct page *)p->private;
126 		/* clear private here, it is used to chain pages */
127 		p->private = 0;
128 	} else
129 		p = alloc_page(gfp_mask);
130 	return p;
131 }
132 
133 static void skb_xmit_done(struct virtqueue *svq)
134 {
135 	struct virtnet_info *vi = svq->vdev->priv;
136 
137 	/* Suppress further interrupts. */
138 	virtqueue_disable_cb(svq);
139 
140 	/* We were probably waiting for more output buffers. */
141 	netif_wake_queue(vi->dev);
142 }
143 
144 static void set_skb_frag(struct sk_buff *skb, struct page *page,
145 			 unsigned int offset, unsigned int *len)
146 {
147 	int size = min((unsigned)PAGE_SIZE - offset, *len);
148 	int i = skb_shinfo(skb)->nr_frags;
149 
150 	__skb_fill_page_desc(skb, i, page, offset, size);
151 
152 	skb->data_len += size;
153 	skb->len += size;
154 	skb->truesize += PAGE_SIZE;
155 	skb_shinfo(skb)->nr_frags++;
156 	*len -= size;
157 }
158 
159 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
160 				   struct page *page, unsigned int len)
161 {
162 	struct sk_buff *skb;
163 	struct skb_vnet_hdr *hdr;
164 	unsigned int copy, hdr_len, offset;
165 	char *p;
166 
167 	p = page_address(page);
168 
169 	/* copy small packet so we can reuse these pages for small data */
170 	skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
171 	if (unlikely(!skb))
172 		return NULL;
173 
174 	hdr = skb_vnet_hdr(skb);
175 
176 	if (vi->mergeable_rx_bufs) {
177 		hdr_len = sizeof hdr->mhdr;
178 		offset = hdr_len;
179 	} else {
180 		hdr_len = sizeof hdr->hdr;
181 		offset = sizeof(struct padded_vnet_hdr);
182 	}
183 
184 	memcpy(hdr, p, hdr_len);
185 
186 	len -= hdr_len;
187 	p += offset;
188 
189 	copy = len;
190 	if (copy > skb_tailroom(skb))
191 		copy = skb_tailroom(skb);
192 	memcpy(skb_put(skb, copy), p, copy);
193 
194 	len -= copy;
195 	offset += copy;
196 
197 	/*
198 	 * Verify that we can indeed put this data into a skb.
199 	 * This is here to handle cases when the device erroneously
200 	 * tries to receive more than is possible. This is usually
201 	 * the case of a broken device.
202 	 */
203 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
204 		if (net_ratelimit())
205 			pr_debug("%s: too much data\n", skb->dev->name);
206 		dev_kfree_skb(skb);
207 		return NULL;
208 	}
209 
210 	while (len) {
211 		set_skb_frag(skb, page, offset, &len);
212 		page = (struct page *)page->private;
213 		offset = 0;
214 	}
215 
216 	if (page)
217 		give_pages(vi, page);
218 
219 	return skb;
220 }
221 
222 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
223 {
224 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
225 	struct page *page;
226 	int num_buf, i, len;
227 
228 	num_buf = hdr->mhdr.num_buffers;
229 	while (--num_buf) {
230 		i = skb_shinfo(skb)->nr_frags;
231 		if (i >= MAX_SKB_FRAGS) {
232 			pr_debug("%s: packet too long\n", skb->dev->name);
233 			skb->dev->stats.rx_length_errors++;
234 			return -EINVAL;
235 		}
236 		page = virtqueue_get_buf(vi->rvq, &len);
237 		if (!page) {
238 			pr_debug("%s: rx error: %d buffers missing\n",
239 				 skb->dev->name, hdr->mhdr.num_buffers);
240 			skb->dev->stats.rx_length_errors++;
241 			return -EINVAL;
242 		}
243 
244 		if (len > PAGE_SIZE)
245 			len = PAGE_SIZE;
246 
247 		set_skb_frag(skb, page, 0, &len);
248 
249 		--vi->num;
250 	}
251 	return 0;
252 }
253 
254 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
255 {
256 	struct virtnet_info *vi = netdev_priv(dev);
257 	struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
258 	struct sk_buff *skb;
259 	struct page *page;
260 	struct skb_vnet_hdr *hdr;
261 
262 	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
263 		pr_debug("%s: short packet %i\n", dev->name, len);
264 		dev->stats.rx_length_errors++;
265 		if (vi->mergeable_rx_bufs || vi->big_packets)
266 			give_pages(vi, buf);
267 		else
268 			dev_kfree_skb(buf);
269 		return;
270 	}
271 
272 	if (!vi->mergeable_rx_bufs && !vi->big_packets) {
273 		skb = buf;
274 		len -= sizeof(struct virtio_net_hdr);
275 		skb_trim(skb, len);
276 	} else {
277 		page = buf;
278 		skb = page_to_skb(vi, page, len);
279 		if (unlikely(!skb)) {
280 			dev->stats.rx_dropped++;
281 			give_pages(vi, page);
282 			return;
283 		}
284 		if (vi->mergeable_rx_bufs)
285 			if (receive_mergeable(vi, skb)) {
286 				dev_kfree_skb(skb);
287 				return;
288 			}
289 	}
290 
291 	hdr = skb_vnet_hdr(skb);
292 
293 	u64_stats_update_begin(&stats->syncp);
294 	stats->rx_bytes += skb->len;
295 	stats->rx_packets++;
296 	u64_stats_update_end(&stats->syncp);
297 
298 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
299 		pr_debug("Needs csum!\n");
300 		if (!skb_partial_csum_set(skb,
301 					  hdr->hdr.csum_start,
302 					  hdr->hdr.csum_offset))
303 			goto frame_err;
304 	} else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
305 		skb->ip_summed = CHECKSUM_UNNECESSARY;
306 	}
307 
308 	skb->protocol = eth_type_trans(skb, dev);
309 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
310 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
311 
312 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
313 		pr_debug("GSO!\n");
314 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
315 		case VIRTIO_NET_HDR_GSO_TCPV4:
316 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
317 			break;
318 		case VIRTIO_NET_HDR_GSO_UDP:
319 			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
320 			break;
321 		case VIRTIO_NET_HDR_GSO_TCPV6:
322 			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
323 			break;
324 		default:
325 			if (net_ratelimit())
326 				printk(KERN_WARNING "%s: bad gso type %u.\n",
327 				       dev->name, hdr->hdr.gso_type);
328 			goto frame_err;
329 		}
330 
331 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
332 			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
333 
334 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
335 		if (skb_shinfo(skb)->gso_size == 0) {
336 			if (net_ratelimit())
337 				printk(KERN_WARNING "%s: zero gso size.\n",
338 				       dev->name);
339 			goto frame_err;
340 		}
341 
342 		/* Header must be checked, and gso_segs computed. */
343 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
344 		skb_shinfo(skb)->gso_segs = 0;
345 	}
346 
347 	netif_receive_skb(skb);
348 	return;
349 
350 frame_err:
351 	dev->stats.rx_frame_errors++;
352 	dev_kfree_skb(skb);
353 }
354 
355 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
356 {
357 	struct sk_buff *skb;
358 	struct skb_vnet_hdr *hdr;
359 	int err;
360 
361 	skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
362 	if (unlikely(!skb))
363 		return -ENOMEM;
364 
365 	skb_put(skb, MAX_PACKET_LEN);
366 
367 	hdr = skb_vnet_hdr(skb);
368 	sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
369 
370 	skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
371 
372 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
373 	if (err < 0)
374 		dev_kfree_skb(skb);
375 
376 	return err;
377 }
378 
379 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
380 {
381 	struct page *first, *list = NULL;
382 	char *p;
383 	int i, err, offset;
384 
385 	/* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
386 	for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
387 		first = get_a_page(vi, gfp);
388 		if (!first) {
389 			if (list)
390 				give_pages(vi, list);
391 			return -ENOMEM;
392 		}
393 		sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
394 
395 		/* chain new page in list head to match sg */
396 		first->private = (unsigned long)list;
397 		list = first;
398 	}
399 
400 	first = get_a_page(vi, gfp);
401 	if (!first) {
402 		give_pages(vi, list);
403 		return -ENOMEM;
404 	}
405 	p = page_address(first);
406 
407 	/* vi->rx_sg[0], vi->rx_sg[1] share the same page */
408 	/* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
409 	sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
410 
411 	/* vi->rx_sg[1] for data packet, from offset */
412 	offset = sizeof(struct padded_vnet_hdr);
413 	sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
414 
415 	/* chain first in list head */
416 	first->private = (unsigned long)list;
417 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
418 				    first, gfp);
419 	if (err < 0)
420 		give_pages(vi, first);
421 
422 	return err;
423 }
424 
425 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
426 {
427 	struct page *page;
428 	int err;
429 
430 	page = get_a_page(vi, gfp);
431 	if (!page)
432 		return -ENOMEM;
433 
434 	sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
435 
436 	err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
437 	if (err < 0)
438 		give_pages(vi, page);
439 
440 	return err;
441 }
442 
443 /* Returns false if we couldn't fill entirely (OOM). */
444 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
445 {
446 	int err;
447 	bool oom;
448 
449 	do {
450 		if (vi->mergeable_rx_bufs)
451 			err = add_recvbuf_mergeable(vi, gfp);
452 		else if (vi->big_packets)
453 			err = add_recvbuf_big(vi, gfp);
454 		else
455 			err = add_recvbuf_small(vi, gfp);
456 
457 		oom = err == -ENOMEM;
458 		if (err < 0)
459 			break;
460 		++vi->num;
461 	} while (err > 0);
462 	if (unlikely(vi->num > vi->max))
463 		vi->max = vi->num;
464 	virtqueue_kick(vi->rvq);
465 	return !oom;
466 }
467 
468 static void skb_recv_done(struct virtqueue *rvq)
469 {
470 	struct virtnet_info *vi = rvq->vdev->priv;
471 	/* Schedule NAPI, Suppress further interrupts if successful. */
472 	if (napi_schedule_prep(&vi->napi)) {
473 		virtqueue_disable_cb(rvq);
474 		__napi_schedule(&vi->napi);
475 	}
476 }
477 
478 static void virtnet_napi_enable(struct virtnet_info *vi)
479 {
480 	napi_enable(&vi->napi);
481 
482 	/* If all buffers were filled by other side before we napi_enabled, we
483 	 * won't get another interrupt, so process any outstanding packets
484 	 * now.  virtnet_poll wants re-enable the queue, so we disable here.
485 	 * We synchronize against interrupts via NAPI_STATE_SCHED */
486 	if (napi_schedule_prep(&vi->napi)) {
487 		virtqueue_disable_cb(vi->rvq);
488 		__napi_schedule(&vi->napi);
489 	}
490 }
491 
492 static void refill_work(struct work_struct *work)
493 {
494 	struct virtnet_info *vi;
495 	bool still_empty;
496 
497 	vi = container_of(work, struct virtnet_info, refill.work);
498 	napi_disable(&vi->napi);
499 	still_empty = !try_fill_recv(vi, GFP_KERNEL);
500 	virtnet_napi_enable(vi);
501 
502 	/* In theory, this can happen: if we don't get any buffers in
503 	 * we will *never* try to fill again. */
504 	if (still_empty)
505 		schedule_delayed_work(&vi->refill, HZ/2);
506 }
507 
508 static int virtnet_poll(struct napi_struct *napi, int budget)
509 {
510 	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
511 	void *buf;
512 	unsigned int len, received = 0;
513 
514 again:
515 	while (received < budget &&
516 	       (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
517 		receive_buf(vi->dev, buf, len);
518 		--vi->num;
519 		received++;
520 	}
521 
522 	if (vi->num < vi->max / 2) {
523 		if (!try_fill_recv(vi, GFP_ATOMIC))
524 			schedule_delayed_work(&vi->refill, 0);
525 	}
526 
527 	/* Out of packets? */
528 	if (received < budget) {
529 		napi_complete(napi);
530 		if (unlikely(!virtqueue_enable_cb(vi->rvq)) &&
531 		    napi_schedule_prep(napi)) {
532 			virtqueue_disable_cb(vi->rvq);
533 			__napi_schedule(napi);
534 			goto again;
535 		}
536 	}
537 
538 	return received;
539 }
540 
541 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
542 {
543 	struct sk_buff *skb;
544 	unsigned int len, tot_sgs = 0;
545 	struct virtnet_stats __percpu *stats = this_cpu_ptr(vi->stats);
546 
547 	while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
548 		pr_debug("Sent skb %p\n", skb);
549 
550 		u64_stats_update_begin(&stats->syncp);
551 		stats->tx_bytes += skb->len;
552 		stats->tx_packets++;
553 		u64_stats_update_end(&stats->syncp);
554 
555 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
556 		dev_kfree_skb_any(skb);
557 	}
558 	return tot_sgs;
559 }
560 
561 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
562 {
563 	struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
564 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
565 
566 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
567 
568 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
569 		hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
570 		hdr->hdr.csum_start = skb_checksum_start_offset(skb);
571 		hdr->hdr.csum_offset = skb->csum_offset;
572 	} else {
573 		hdr->hdr.flags = 0;
574 		hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
575 	}
576 
577 	if (skb_is_gso(skb)) {
578 		hdr->hdr.hdr_len = skb_headlen(skb);
579 		hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
580 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
581 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
582 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
583 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
584 		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
585 			hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
586 		else
587 			BUG();
588 		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
589 			hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
590 	} else {
591 		hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
592 		hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
593 	}
594 
595 	hdr->mhdr.num_buffers = 0;
596 
597 	/* Encode metadata header at front. */
598 	if (vi->mergeable_rx_bufs)
599 		sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
600 	else
601 		sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
602 
603 	hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
604 	return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
605 					0, skb);
606 }
607 
608 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
609 {
610 	struct virtnet_info *vi = netdev_priv(dev);
611 	int capacity;
612 
613 	/* Free up any pending old buffers before queueing new ones. */
614 	free_old_xmit_skbs(vi);
615 
616 	/* Try to transmit */
617 	capacity = xmit_skb(vi, skb);
618 
619 	/* This can happen with OOM and indirect buffers. */
620 	if (unlikely(capacity < 0)) {
621 		if (net_ratelimit()) {
622 			if (likely(capacity == -ENOMEM)) {
623 				dev_warn(&dev->dev,
624 					 "TX queue failure: out of memory\n");
625 			} else {
626 				dev->stats.tx_fifo_errors++;
627 				dev_warn(&dev->dev,
628 					 "Unexpected TX queue failure: %d\n",
629 					 capacity);
630 			}
631 		}
632 		dev->stats.tx_dropped++;
633 		kfree_skb(skb);
634 		return NETDEV_TX_OK;
635 	}
636 	virtqueue_kick(vi->svq);
637 
638 	/* Don't wait up for transmitted skbs to be freed. */
639 	skb_orphan(skb);
640 	nf_reset(skb);
641 
642 	/* Apparently nice girls don't return TX_BUSY; stop the queue
643 	 * before it gets out of hand.  Naturally, this wastes entries. */
644 	if (capacity < 2+MAX_SKB_FRAGS) {
645 		netif_stop_queue(dev);
646 		if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
647 			/* More just got used, free them then recheck. */
648 			capacity += free_old_xmit_skbs(vi);
649 			if (capacity >= 2+MAX_SKB_FRAGS) {
650 				netif_start_queue(dev);
651 				virtqueue_disable_cb(vi->svq);
652 			}
653 		}
654 	}
655 
656 	return NETDEV_TX_OK;
657 }
658 
659 static int virtnet_set_mac_address(struct net_device *dev, void *p)
660 {
661 	struct virtnet_info *vi = netdev_priv(dev);
662 	struct virtio_device *vdev = vi->vdev;
663 	int ret;
664 
665 	ret = eth_mac_addr(dev, p);
666 	if (ret)
667 		return ret;
668 
669 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
670 		vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
671 		                  dev->dev_addr, dev->addr_len);
672 
673 	return 0;
674 }
675 
676 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
677 					       struct rtnl_link_stats64 *tot)
678 {
679 	struct virtnet_info *vi = netdev_priv(dev);
680 	int cpu;
681 	unsigned int start;
682 
683 	for_each_possible_cpu(cpu) {
684 		struct virtnet_stats __percpu *stats
685 			= per_cpu_ptr(vi->stats, cpu);
686 		u64 tpackets, tbytes, rpackets, rbytes;
687 
688 		do {
689 			start = u64_stats_fetch_begin(&stats->syncp);
690 			tpackets = stats->tx_packets;
691 			tbytes   = stats->tx_bytes;
692 			rpackets = stats->rx_packets;
693 			rbytes   = stats->rx_bytes;
694 		} while (u64_stats_fetch_retry(&stats->syncp, start));
695 
696 		tot->rx_packets += rpackets;
697 		tot->tx_packets += tpackets;
698 		tot->rx_bytes   += rbytes;
699 		tot->tx_bytes   += tbytes;
700 	}
701 
702 	tot->tx_dropped = dev->stats.tx_dropped;
703 	tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
704 	tot->rx_dropped = dev->stats.rx_dropped;
705 	tot->rx_length_errors = dev->stats.rx_length_errors;
706 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
707 
708 	return tot;
709 }
710 
711 #ifdef CONFIG_NET_POLL_CONTROLLER
712 static void virtnet_netpoll(struct net_device *dev)
713 {
714 	struct virtnet_info *vi = netdev_priv(dev);
715 
716 	napi_schedule(&vi->napi);
717 }
718 #endif
719 
720 static int virtnet_open(struct net_device *dev)
721 {
722 	struct virtnet_info *vi = netdev_priv(dev);
723 
724 	virtnet_napi_enable(vi);
725 	return 0;
726 }
727 
728 /*
729  * Send command via the control virtqueue and check status.  Commands
730  * supported by the hypervisor, as indicated by feature bits, should
731  * never fail unless improperly formated.
732  */
733 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
734 				 struct scatterlist *data, int out, int in)
735 {
736 	struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
737 	struct virtio_net_ctrl_hdr ctrl;
738 	virtio_net_ctrl_ack status = ~0;
739 	unsigned int tmp;
740 	int i;
741 
742 	/* Caller should know better */
743 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
744 		(out + in > VIRTNET_SEND_COMMAND_SG_MAX));
745 
746 	out++; /* Add header */
747 	in++; /* Add return status */
748 
749 	ctrl.class = class;
750 	ctrl.cmd = cmd;
751 
752 	sg_init_table(sg, out + in);
753 
754 	sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
755 	for_each_sg(data, s, out + in - 2, i)
756 		sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
757 	sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
758 
759 	BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0);
760 
761 	virtqueue_kick(vi->cvq);
762 
763 	/*
764 	 * Spin for a response, the kick causes an ioport write, trapping
765 	 * into the hypervisor, so the request should be handled immediately.
766 	 */
767 	while (!virtqueue_get_buf(vi->cvq, &tmp))
768 		cpu_relax();
769 
770 	return status == VIRTIO_NET_OK;
771 }
772 
773 static int virtnet_close(struct net_device *dev)
774 {
775 	struct virtnet_info *vi = netdev_priv(dev);
776 
777 	napi_disable(&vi->napi);
778 
779 	return 0;
780 }
781 
782 static void virtnet_set_rx_mode(struct net_device *dev)
783 {
784 	struct virtnet_info *vi = netdev_priv(dev);
785 	struct scatterlist sg[2];
786 	u8 promisc, allmulti;
787 	struct virtio_net_ctrl_mac *mac_data;
788 	struct netdev_hw_addr *ha;
789 	int uc_count;
790 	int mc_count;
791 	void *buf;
792 	int i;
793 
794 	/* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
795 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
796 		return;
797 
798 	promisc = ((dev->flags & IFF_PROMISC) != 0);
799 	allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
800 
801 	sg_init_one(sg, &promisc, sizeof(promisc));
802 
803 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
804 				  VIRTIO_NET_CTRL_RX_PROMISC,
805 				  sg, 1, 0))
806 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
807 			 promisc ? "en" : "dis");
808 
809 	sg_init_one(sg, &allmulti, sizeof(allmulti));
810 
811 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
812 				  VIRTIO_NET_CTRL_RX_ALLMULTI,
813 				  sg, 1, 0))
814 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
815 			 allmulti ? "en" : "dis");
816 
817 	uc_count = netdev_uc_count(dev);
818 	mc_count = netdev_mc_count(dev);
819 	/* MAC filter - use one buffer for both lists */
820 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
821 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
822 	mac_data = buf;
823 	if (!buf) {
824 		dev_warn(&dev->dev, "No memory for MAC address buffer\n");
825 		return;
826 	}
827 
828 	sg_init_table(sg, 2);
829 
830 	/* Store the unicast list and count in the front of the buffer */
831 	mac_data->entries = uc_count;
832 	i = 0;
833 	netdev_for_each_uc_addr(ha, dev)
834 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
835 
836 	sg_set_buf(&sg[0], mac_data,
837 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
838 
839 	/* multicast list and count fill the end */
840 	mac_data = (void *)&mac_data->macs[uc_count][0];
841 
842 	mac_data->entries = mc_count;
843 	i = 0;
844 	netdev_for_each_mc_addr(ha, dev)
845 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
846 
847 	sg_set_buf(&sg[1], mac_data,
848 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
849 
850 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
851 				  VIRTIO_NET_CTRL_MAC_TABLE_SET,
852 				  sg, 2, 0))
853 		dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
854 
855 	kfree(buf);
856 }
857 
858 static int virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
859 {
860 	struct virtnet_info *vi = netdev_priv(dev);
861 	struct scatterlist sg;
862 
863 	sg_init_one(&sg, &vid, sizeof(vid));
864 
865 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
866 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
867 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
868 	return 0;
869 }
870 
871 static int virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
872 {
873 	struct virtnet_info *vi = netdev_priv(dev);
874 	struct scatterlist sg;
875 
876 	sg_init_one(&sg, &vid, sizeof(vid));
877 
878 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
879 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
880 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
881 	return 0;
882 }
883 
884 static void virtnet_get_ringparam(struct net_device *dev,
885 				struct ethtool_ringparam *ring)
886 {
887 	struct virtnet_info *vi = netdev_priv(dev);
888 
889 	ring->rx_max_pending = virtqueue_get_vring_size(vi->rvq);
890 	ring->tx_max_pending = virtqueue_get_vring_size(vi->svq);
891 	ring->rx_pending = ring->rx_max_pending;
892 	ring->tx_pending = ring->tx_max_pending;
893 
894 }
895 
896 
897 static void virtnet_get_drvinfo(struct net_device *dev,
898 				struct ethtool_drvinfo *info)
899 {
900 	struct virtnet_info *vi = netdev_priv(dev);
901 	struct virtio_device *vdev = vi->vdev;
902 
903 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
904 	strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
905 	strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
906 
907 }
908 
909 static const struct ethtool_ops virtnet_ethtool_ops = {
910 	.get_drvinfo = virtnet_get_drvinfo,
911 	.get_link = ethtool_op_get_link,
912 	.get_ringparam = virtnet_get_ringparam,
913 };
914 
915 #define MIN_MTU 68
916 #define MAX_MTU 65535
917 
918 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
919 {
920 	if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
921 		return -EINVAL;
922 	dev->mtu = new_mtu;
923 	return 0;
924 }
925 
926 static const struct net_device_ops virtnet_netdev = {
927 	.ndo_open            = virtnet_open,
928 	.ndo_stop   	     = virtnet_close,
929 	.ndo_start_xmit      = start_xmit,
930 	.ndo_validate_addr   = eth_validate_addr,
931 	.ndo_set_mac_address = virtnet_set_mac_address,
932 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
933 	.ndo_change_mtu	     = virtnet_change_mtu,
934 	.ndo_get_stats64     = virtnet_stats,
935 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
936 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
937 #ifdef CONFIG_NET_POLL_CONTROLLER
938 	.ndo_poll_controller = virtnet_netpoll,
939 #endif
940 };
941 
942 static void virtnet_update_status(struct virtnet_info *vi)
943 {
944 	u16 v;
945 
946 	if (virtio_config_val(vi->vdev, VIRTIO_NET_F_STATUS,
947 			      offsetof(struct virtio_net_config, status),
948 			      &v) < 0)
949 		return;
950 
951 	/* Ignore unknown (future) status bits */
952 	v &= VIRTIO_NET_S_LINK_UP;
953 
954 	if (vi->status == v)
955 		return;
956 
957 	vi->status = v;
958 
959 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
960 		netif_carrier_on(vi->dev);
961 		netif_wake_queue(vi->dev);
962 	} else {
963 		netif_carrier_off(vi->dev);
964 		netif_stop_queue(vi->dev);
965 	}
966 }
967 
968 static void virtnet_config_changed(struct virtio_device *vdev)
969 {
970 	struct virtnet_info *vi = vdev->priv;
971 
972 	virtnet_update_status(vi);
973 }
974 
975 static int virtnet_probe(struct virtio_device *vdev)
976 {
977 	int err;
978 	struct net_device *dev;
979 	struct virtnet_info *vi;
980 	struct virtqueue *vqs[3];
981 	vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
982 	const char *names[] = { "input", "output", "control" };
983 	int nvqs;
984 
985 	/* Allocate ourselves a network device with room for our info */
986 	dev = alloc_etherdev(sizeof(struct virtnet_info));
987 	if (!dev)
988 		return -ENOMEM;
989 
990 	/* Set up network device as normal. */
991 	dev->priv_flags |= IFF_UNICAST_FLT;
992 	dev->netdev_ops = &virtnet_netdev;
993 	dev->features = NETIF_F_HIGHDMA;
994 
995 	SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
996 	SET_NETDEV_DEV(dev, &vdev->dev);
997 
998 	/* Do we support "hardware" checksums? */
999 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1000 		/* This opens up the world of extra features. */
1001 		dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1002 		if (csum)
1003 			dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
1004 
1005 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1006 			dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1007 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
1008 		}
1009 		/* Individual feature bits: what can host handle? */
1010 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1011 			dev->hw_features |= NETIF_F_TSO;
1012 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1013 			dev->hw_features |= NETIF_F_TSO6;
1014 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1015 			dev->hw_features |= NETIF_F_TSO_ECN;
1016 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1017 			dev->hw_features |= NETIF_F_UFO;
1018 
1019 		if (gso)
1020 			dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1021 		/* (!csum && gso) case will be fixed by register_netdev() */
1022 	}
1023 
1024 	/* Configuration may specify what MAC to use.  Otherwise random. */
1025 	if (virtio_config_val_len(vdev, VIRTIO_NET_F_MAC,
1026 				  offsetof(struct virtio_net_config, mac),
1027 				  dev->dev_addr, dev->addr_len) < 0)
1028 		random_ether_addr(dev->dev_addr);
1029 
1030 	/* Set up our device-specific information */
1031 	vi = netdev_priv(dev);
1032 	netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
1033 	vi->dev = dev;
1034 	vi->vdev = vdev;
1035 	vdev->priv = vi;
1036 	vi->pages = NULL;
1037 	vi->stats = alloc_percpu(struct virtnet_stats);
1038 	err = -ENOMEM;
1039 	if (vi->stats == NULL)
1040 		goto free;
1041 
1042 	INIT_DELAYED_WORK(&vi->refill, refill_work);
1043 	sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
1044 	sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
1045 
1046 	/* If we can receive ANY GSO packets, we must allocate large ones. */
1047 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1048 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1049 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
1050 		vi->big_packets = true;
1051 
1052 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1053 		vi->mergeable_rx_bufs = true;
1054 
1055 	/* We expect two virtqueues, receive then send,
1056 	 * and optionally control. */
1057 	nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
1058 
1059 	err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
1060 	if (err)
1061 		goto free_stats;
1062 
1063 	vi->rvq = vqs[0];
1064 	vi->svq = vqs[1];
1065 
1066 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
1067 		vi->cvq = vqs[2];
1068 
1069 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1070 			dev->features |= NETIF_F_HW_VLAN_FILTER;
1071 	}
1072 
1073 	err = register_netdev(dev);
1074 	if (err) {
1075 		pr_debug("virtio_net: registering device failed\n");
1076 		goto free_vqs;
1077 	}
1078 
1079 	/* Last of all, set up some receive buffers. */
1080 	try_fill_recv(vi, GFP_KERNEL);
1081 
1082 	/* If we didn't even get one input buffer, we're useless. */
1083 	if (vi->num == 0) {
1084 		err = -ENOMEM;
1085 		goto unregister;
1086 	}
1087 
1088 	/* Assume link up if device can't report link status,
1089 	   otherwise get link status from config. */
1090 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1091 		netif_carrier_off(dev);
1092 		virtnet_update_status(vi);
1093 	} else {
1094 		vi->status = VIRTIO_NET_S_LINK_UP;
1095 		netif_carrier_on(dev);
1096 	}
1097 
1098 	pr_debug("virtnet: registered device %s\n", dev->name);
1099 	return 0;
1100 
1101 unregister:
1102 	unregister_netdev(dev);
1103 	cancel_delayed_work_sync(&vi->refill);
1104 free_vqs:
1105 	vdev->config->del_vqs(vdev);
1106 free_stats:
1107 	free_percpu(vi->stats);
1108 free:
1109 	free_netdev(dev);
1110 	return err;
1111 }
1112 
1113 static void free_unused_bufs(struct virtnet_info *vi)
1114 {
1115 	void *buf;
1116 	while (1) {
1117 		buf = virtqueue_detach_unused_buf(vi->svq);
1118 		if (!buf)
1119 			break;
1120 		dev_kfree_skb(buf);
1121 	}
1122 	while (1) {
1123 		buf = virtqueue_detach_unused_buf(vi->rvq);
1124 		if (!buf)
1125 			break;
1126 		if (vi->mergeable_rx_bufs || vi->big_packets)
1127 			give_pages(vi, buf);
1128 		else
1129 			dev_kfree_skb(buf);
1130 		--vi->num;
1131 	}
1132 	BUG_ON(vi->num != 0);
1133 }
1134 
1135 static void __devexit virtnet_remove(struct virtio_device *vdev)
1136 {
1137 	struct virtnet_info *vi = vdev->priv;
1138 
1139 	/* Stop all the virtqueues. */
1140 	vdev->config->reset(vdev);
1141 
1142 
1143 	unregister_netdev(vi->dev);
1144 	cancel_delayed_work_sync(&vi->refill);
1145 
1146 	/* Free unused buffers in both send and recv, if any. */
1147 	free_unused_bufs(vi);
1148 
1149 	vdev->config->del_vqs(vi->vdev);
1150 
1151 	while (vi->pages)
1152 		__free_pages(get_a_page(vi, GFP_KERNEL), 0);
1153 
1154 	free_percpu(vi->stats);
1155 	free_netdev(vi->dev);
1156 }
1157 
1158 static struct virtio_device_id id_table[] = {
1159 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1160 	{ 0 },
1161 };
1162 
1163 static unsigned int features[] = {
1164 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1165 	VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1166 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1167 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1168 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1169 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1170 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1171 };
1172 
1173 static struct virtio_driver virtio_net_driver = {
1174 	.feature_table = features,
1175 	.feature_table_size = ARRAY_SIZE(features),
1176 	.driver.name =	KBUILD_MODNAME,
1177 	.driver.owner =	THIS_MODULE,
1178 	.id_table =	id_table,
1179 	.probe =	virtnet_probe,
1180 	.remove =	__devexit_p(virtnet_remove),
1181 	.config_changed = virtnet_config_changed,
1182 };
1183 
1184 static int __init init(void)
1185 {
1186 	return register_virtio_driver(&virtio_net_driver);
1187 }
1188 
1189 static void __exit fini(void)
1190 {
1191 	unregister_virtio_driver(&virtio_net_driver);
1192 }
1193 module_init(init);
1194 module_exit(fini);
1195 
1196 MODULE_DEVICE_TABLE(virtio, id_table);
1197 MODULE_DESCRIPTION("Virtio network driver");
1198 MODULE_LICENSE("GPL");
1199