- Linux 3.0.89.
[opensuse:kernel.git] / drivers / net / virtio_net.c
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         struct virtio_device *vdev;
45         struct virtqueue *rvq, *svq, *cvq;
46         struct net_device *dev;
47         struct napi_struct napi;
48         unsigned int status;
49
50         /* Number of input buffers, and max we've ever had. */
51         unsigned int num, max;
52
53         /* I like... big packets and I cannot lie! */
54         bool big_packets;
55
56         /* Host will merge rx buffers for big packets (shake it! shake it!) */
57         bool mergeable_rx_bufs;
58
59         /* Work struct for refilling if we run low on memory. */
60         struct delayed_work refill;
61
62         /* Chain pages by the private ptr. */
63         struct page *pages;
64
65         /* fragments + linear part + virtio header */
66         struct scatterlist rx_sg[MAX_SKB_FRAGS + 2];
67         struct scatterlist tx_sg[MAX_SKB_FRAGS + 2];
68 };
69
70 struct skb_vnet_hdr {
71         union {
72                 struct virtio_net_hdr hdr;
73                 struct virtio_net_hdr_mrg_rxbuf mhdr;
74         };
75         unsigned int num_sg;
76 };
77
78 struct padded_vnet_hdr {
79         struct virtio_net_hdr hdr;
80         /*
81          * virtio_net_hdr should be in a separated sg buffer because of a
82          * QEMU bug, and data sg buffer shares same page with this header sg.
83          * This padding makes next sg 16 byte aligned after virtio_net_hdr.
84          */
85         char padding[6];
86 };
87
88 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
89 {
90         return (struct skb_vnet_hdr *)skb->cb;
91 }
92
93 /*
94  * private is used to chain pages for big packets, put the whole
95  * most recent used list in the beginning for reuse
96  */
97 static void give_pages(struct virtnet_info *vi, struct page *page)
98 {
99         struct page *end;
100
101         /* Find end of list, sew whole thing into vi->pages. */
102         for (end = page; end->private; end = (struct page *)end->private);
103         end->private = (unsigned long)vi->pages;
104         vi->pages = page;
105 }
106
107 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
108 {
109         struct page *p = vi->pages;
110
111         if (p) {
112                 vi->pages = (struct page *)p->private;
113                 /* clear private here, it is used to chain pages */
114                 p->private = 0;
115         } else
116                 p = alloc_page(gfp_mask);
117         return p;
118 }
119
120 static void skb_xmit_done(struct virtqueue *svq)
121 {
122         struct virtnet_info *vi = svq->vdev->priv;
123
124         /* Suppress further interrupts. */
125         virtqueue_disable_cb(svq);
126
127         /* We were probably waiting for more output buffers. */
128         netif_wake_queue(vi->dev);
129 }
130
131 static void set_skb_frag(struct sk_buff *skb, struct page *page,
132                          unsigned int offset, unsigned int *len)
133 {
134         int i = skb_shinfo(skb)->nr_frags;
135         skb_frag_t *f;
136
137         f = &skb_shinfo(skb)->frags[i];
138         f->size = min((unsigned)PAGE_SIZE - offset, *len);
139         f->page_offset = offset;
140         f->page = page;
141
142         skb->data_len += f->size;
143         skb->len += f->size;
144         skb_shinfo(skb)->nr_frags++;
145         *len -= f->size;
146 }
147
148 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
149                                    struct page *page, unsigned int len)
150 {
151         struct sk_buff *skb;
152         struct skb_vnet_hdr *hdr;
153         unsigned int copy, hdr_len, offset;
154         char *p;
155
156         p = page_address(page);
157
158         /* copy small packet so we can reuse these pages for small data */
159         skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
160         if (unlikely(!skb))
161                 return NULL;
162
163         hdr = skb_vnet_hdr(skb);
164
165         if (vi->mergeable_rx_bufs) {
166                 hdr_len = sizeof hdr->mhdr;
167                 offset = hdr_len;
168         } else {
169                 hdr_len = sizeof hdr->hdr;
170                 offset = sizeof(struct padded_vnet_hdr);
171         }
172
173         memcpy(hdr, p, hdr_len);
174
175         len -= hdr_len;
176         p += offset;
177
178         copy = len;
179         if (copy > skb_tailroom(skb))
180                 copy = skb_tailroom(skb);
181         memcpy(skb_put(skb, copy), p, copy);
182
183         len -= copy;
184         offset += copy;
185
186         while (len) {
187                 set_skb_frag(skb, page, offset, &len);
188                 page = (struct page *)page->private;
189                 offset = 0;
190         }
191
192         if (page)
193                 give_pages(vi, page);
194
195         return skb;
196 }
197
198 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
199 {
200         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
201         struct page *page;
202         int num_buf, i, len;
203
204         num_buf = hdr->mhdr.num_buffers;
205         while (--num_buf) {
206                 i = skb_shinfo(skb)->nr_frags;
207                 if (i >= MAX_SKB_FRAGS) {
208                         pr_debug("%s: packet too long\n", skb->dev->name);
209                         skb->dev->stats.rx_length_errors++;
210                         return -EINVAL;
211                 }
212
213                 page = virtqueue_get_buf(vi->rvq, &len);
214                 if (!page) {
215                         pr_debug("%s: rx error: %d buffers missing\n",
216                                  skb->dev->name, hdr->mhdr.num_buffers);
217                         skb->dev->stats.rx_length_errors++;
218                         return -EINVAL;
219                 }
220                 if (len > PAGE_SIZE)
221                         len = PAGE_SIZE;
222
223                 set_skb_frag(skb, page, 0, &len);
224
225                 --vi->num;
226         }
227         return 0;
228 }
229
230 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
231 {
232         struct virtnet_info *vi = netdev_priv(dev);
233         struct sk_buff *skb;
234         struct page *page;
235         struct skb_vnet_hdr *hdr;
236
237         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
238                 pr_debug("%s: short packet %i\n", dev->name, len);
239                 dev->stats.rx_length_errors++;
240                 if (vi->mergeable_rx_bufs || vi->big_packets)
241                         give_pages(vi, buf);
242                 else
243                         dev_kfree_skb(buf);
244                 return;
245         }
246
247         if (!vi->mergeable_rx_bufs && !vi->big_packets) {
248                 skb = buf;
249                 len -= sizeof(struct virtio_net_hdr);
250                 skb_trim(skb, len);
251         } else {
252                 page = buf;
253                 skb = page_to_skb(vi, page, len);
254                 if (unlikely(!skb)) {
255                         dev->stats.rx_dropped++;
256                         give_pages(vi, page);
257                         return;
258                 }
259                 if (vi->mergeable_rx_bufs)
260                         if (receive_mergeable(vi, skb)) {
261                                 dev_kfree_skb(skb);
262                                 return;
263                         }
264         }
265
266         hdr = skb_vnet_hdr(skb);
267         skb->truesize += skb->data_len;
268         dev->stats.rx_bytes += skb->len;
269         dev->stats.rx_packets++;
270
271         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
272                 pr_debug("Needs csum!\n");
273                 if (!skb_partial_csum_set(skb,
274                                           hdr->hdr.csum_start,
275                                           hdr->hdr.csum_offset))
276                         goto frame_err;
277         } else if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) {
278                 skb->ip_summed = CHECKSUM_UNNECESSARY;
279         }
280
281         skb->protocol = eth_type_trans(skb, dev);
282         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
283                  ntohs(skb->protocol), skb->len, skb->pkt_type);
284
285         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
286                 pr_debug("GSO!\n");
287                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
288                 case VIRTIO_NET_HDR_GSO_TCPV4:
289                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
290                         break;
291                 case VIRTIO_NET_HDR_GSO_UDP:
292                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
293                         break;
294                 case VIRTIO_NET_HDR_GSO_TCPV6:
295                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
296                         break;
297                 default:
298                         if (net_ratelimit())
299                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
300                                        dev->name, hdr->hdr.gso_type);
301                         goto frame_err;
302                 }
303
304                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
305                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
306
307                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
308                 if (skb_shinfo(skb)->gso_size == 0) {
309                         if (net_ratelimit())
310                                 printk(KERN_WARNING "%s: zero gso size.\n",
311                                        dev->name);
312                         goto frame_err;
313                 }
314
315                 /* Header must be checked, and gso_segs computed. */
316                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
317                 skb_shinfo(skb)->gso_segs = 0;
318         }
319
320         netif_receive_skb(skb);
321         return;
322
323 frame_err:
324         dev->stats.rx_frame_errors++;
325         dev_kfree_skb(skb);
326 }
327
328 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
329 {
330         struct sk_buff *skb;
331         struct skb_vnet_hdr *hdr;
332         int err;
333
334         skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
335         if (unlikely(!skb))
336                 return -ENOMEM;
337
338         skb_put(skb, MAX_PACKET_LEN);
339
340         hdr = skb_vnet_hdr(skb);
341         sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr);
342
343         skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len);
344
345         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp);
346         if (err < 0)
347                 dev_kfree_skb(skb);
348
349         return err;
350 }
351
352 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
353 {
354         struct page *first, *list = NULL;
355         char *p;
356         int i, err, offset;
357
358         /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */
359         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
360                 first = get_a_page(vi, gfp);
361                 if (!first) {
362                         if (list)
363                                 give_pages(vi, list);
364                         return -ENOMEM;
365                 }
366                 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE);
367
368                 /* chain new page in list head to match sg */
369                 first->private = (unsigned long)list;
370                 list = first;
371         }
372
373         first = get_a_page(vi, gfp);
374         if (!first) {
375                 give_pages(vi, list);
376                 return -ENOMEM;
377         }
378         p = page_address(first);
379
380         /* vi->rx_sg[0], vi->rx_sg[1] share the same page */
381         /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */
382         sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr));
383
384         /* vi->rx_sg[1] for data packet, from offset */
385         offset = sizeof(struct padded_vnet_hdr);
386         sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset);
387
388         /* chain first in list head */
389         first->private = (unsigned long)list;
390         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2,
391                                     first, gfp);
392         if (err < 0)
393                 give_pages(vi, first);
394
395         return err;
396 }
397
398 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
399 {
400         struct page *page;
401         int err;
402
403         page = get_a_page(vi, gfp);
404         if (!page)
405                 return -ENOMEM;
406
407         sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE);
408
409         err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp);
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;
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                 oom = err == -ENOMEM;
431                 if (err < 0)
432                         break;
433                 ++vi->num;
434         } while (err > 0);
435         if (unlikely(vi->num > vi->max))
436                 vi->max = vi->num;
437         virtqueue_kick(vi->rvq);
438         return !oom;
439 }
440
441 static void skb_recv_done(struct virtqueue *rvq)
442 {
443         struct virtnet_info *vi = rvq->vdev->priv;
444         /* Schedule NAPI, Suppress further interrupts if successful. */
445         if (napi_schedule_prep(&vi->napi)) {
446                 virtqueue_disable_cb(rvq);
447                 __napi_schedule(&vi->napi);
448         }
449 }
450
451 static void virtnet_napi_enable(struct virtnet_info *vi)
452 {
453         napi_enable(&vi->napi);
454
455         /* If all buffers were filled by other side before we napi_enabled, we
456          * won't get another interrupt, so process any outstanding packets
457          * now.  virtnet_poll wants re-enable the queue, so we disable here.
458          * We synchronize against interrupts via NAPI_STATE_SCHED */
459         if (napi_schedule_prep(&vi->napi)) {
460                 virtqueue_disable_cb(vi->rvq);
461                 __napi_schedule(&vi->napi);
462         }
463 }
464
465 static void refill_work(struct work_struct *work)
466 {
467         struct virtnet_info *vi;
468         bool still_empty;
469
470         vi = container_of(work, struct virtnet_info, refill.work);
471         napi_disable(&vi->napi);
472         still_empty = !try_fill_recv(vi, GFP_KERNEL);
473         virtnet_napi_enable(vi);
474
475         /* In theory, this can happen: if we don't get any buffers in
476          * we will *never* try to fill again. */
477         if (still_empty)
478                 schedule_delayed_work(&vi->refill, HZ/2);
479 }
480
481 static int virtnet_poll(struct napi_struct *napi, int budget)
482 {
483         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
484         void *buf;
485         unsigned int r, len, received = 0;
486
487 again:
488         while (received < budget &&
489                (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) {
490                 receive_buf(vi->dev, buf, len);
491                 --vi->num;
492                 received++;
493         }
494
495         if (vi->num < vi->max / 2) {
496                 if (!try_fill_recv(vi, GFP_ATOMIC))
497                         schedule_delayed_work(&vi->refill, 0);
498         }
499
500         /* Out of packets? */
501         if (received < budget) {
502                 r = virtqueue_enable_cb_prepare(vi->rvq);
503                 napi_complete(napi);
504                 if (unlikely(virtqueue_poll(vi->rvq, r)) &&
505                     napi_schedule_prep(napi)) {
506                         virtqueue_disable_cb(vi->rvq);
507                         __napi_schedule(napi);
508                         goto again;
509                 }
510         }
511
512         return received;
513 }
514
515 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
516 {
517         struct sk_buff *skb;
518         unsigned int len, tot_sgs = 0;
519
520         while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) {
521                 pr_debug("Sent skb %p\n", skb);
522                 vi->dev->stats.tx_bytes += skb->len;
523                 vi->dev->stats.tx_packets++;
524                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
525                 dev_kfree_skb_any(skb);
526         }
527         return tot_sgs;
528 }
529
530 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
531 {
532         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
533         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
534
535         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
536
537         if (skb->ip_summed == CHECKSUM_PARTIAL) {
538                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
539                 hdr->hdr.csum_start = skb_checksum_start_offset(skb);
540                 hdr->hdr.csum_offset = skb->csum_offset;
541         } else {
542                 hdr->hdr.flags = 0;
543                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
544         }
545
546         if (skb_is_gso(skb)) {
547                 hdr->hdr.hdr_len = skb_headlen(skb);
548                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
549                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
550                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
551                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
552                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
553                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
554                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
555                 else
556                         BUG();
557                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
558                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
559         } else {
560                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
561                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
562         }
563
564         hdr->mhdr.num_buffers = 0;
565
566         /* Encode metadata header at front. */
567         if (vi->mergeable_rx_bufs)
568                 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr);
569         else
570                 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr);
571
572         hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1;
573         return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg,
574                                         0, skb);
575 }
576
577 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
578 {
579         struct virtnet_info *vi = netdev_priv(dev);
580         int capacity;
581
582         /* Free up any pending old buffers before queueing new ones. */
583         free_old_xmit_skbs(vi);
584
585         /* Try to transmit */
586         capacity = xmit_skb(vi, skb);
587
588         /* This can happen with OOM and indirect buffers. */
589         if (unlikely(capacity < 0)) {
590                 if (net_ratelimit()) {
591                         if (likely(capacity == -ENOMEM)) {
592                                 dev_warn(&dev->dev,
593                                          "TX queue failure: out of memory\n");
594                         } else {
595                                 dev->stats.tx_fifo_errors++;
596                                 dev_warn(&dev->dev,
597                                          "Unexpected TX queue failure: %d\n",
598                                          capacity);
599                         }
600                 }
601                 dev->stats.tx_dropped++;
602                 kfree_skb(skb);
603                 return NETDEV_TX_OK;
604         }
605         virtqueue_kick(vi->svq);
606
607         /* Don't wait up for transmitted skbs to be freed. */
608         skb_orphan(skb);
609         nf_reset(skb);
610
611         /* Apparently nice girls don't return TX_BUSY; stop the queue
612          * before it gets out of hand.  Naturally, this wastes entries. */
613         if (capacity < 2+MAX_SKB_FRAGS) {
614                 netif_stop_queue(dev);
615                 if (unlikely(!virtqueue_enable_cb_delayed(vi->svq))) {
616                         /* More just got used, free them then recheck. */
617                         capacity += free_old_xmit_skbs(vi);
618                         if (capacity >= 2+MAX_SKB_FRAGS) {
619                                 netif_start_queue(dev);
620                                 virtqueue_disable_cb(vi->svq);
621                         }
622                 }
623         }
624
625         return NETDEV_TX_OK;
626 }
627
628 static int virtnet_set_mac_address(struct net_device *dev, void *p)
629 {
630         struct virtnet_info *vi = netdev_priv(dev);
631         struct virtio_device *vdev = vi->vdev;
632         int ret;
633
634         ret = eth_mac_addr(dev, p);
635         if (ret)
636                 return ret;
637
638         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
639                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
640                                   dev->dev_addr, dev->addr_len);
641
642         return 0;
643 }
644
645 #ifdef CONFIG_NET_POLL_CONTROLLER
646 static void virtnet_netpoll(struct net_device *dev)
647 {
648         struct virtnet_info *vi = netdev_priv(dev);
649
650         napi_schedule(&vi->napi);
651 }
652 #endif
653
654 static int virtnet_open(struct net_device *dev)
655 {
656         struct virtnet_info *vi = netdev_priv(dev);
657
658         virtnet_napi_enable(vi);
659         return 0;
660 }
661
662 /*
663  * Send command via the control virtqueue and check status.  Commands
664  * supported by the hypervisor, as indicated by feature bits, should
665  * never fail unless improperly formated.
666  */
667 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
668                                  struct scatterlist *data, int out, int in)
669 {
670         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
671         struct virtio_net_ctrl_hdr ctrl;
672         virtio_net_ctrl_ack status = ~0;
673         unsigned int tmp;
674         int i;
675
676         /* Caller should know better */
677         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
678                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
679
680         out++; /* Add header */
681         in++; /* Add return status */
682
683         ctrl.class = class;
684         ctrl.cmd = cmd;
685
686         sg_init_table(sg, out + in);
687
688         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
689         for_each_sg(data, s, out + in - 2, i)
690                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
691         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
692
693         BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0);
694
695         virtqueue_kick(vi->cvq);
696
697         /*
698          * Spin for a response, the kick causes an ioport write, trapping
699          * into the hypervisor, so the request should be handled immediately.
700          */
701         while (!virtqueue_get_buf(vi->cvq, &tmp))
702                 cpu_relax();
703
704         return status == VIRTIO_NET_OK;
705 }
706
707 static int virtnet_close(struct net_device *dev)
708 {
709         struct virtnet_info *vi = netdev_priv(dev);
710
711         napi_disable(&vi->napi);
712
713         return 0;
714 }
715
716 static void virtnet_set_rx_mode(struct net_device *dev)
717 {
718         struct virtnet_info *vi = netdev_priv(dev);
719         struct scatterlist sg[2];
720         u8 promisc, allmulti;
721         struct virtio_net_ctrl_mac *mac_data;
722         struct netdev_hw_addr *ha;
723         int uc_count;
724         int mc_count;
725         void *buf;
726         int i;
727
728         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
729         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
730                 return;
731
732         promisc = ((dev->flags & IFF_PROMISC) != 0);
733         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
734
735         sg_init_one(sg, &promisc, sizeof(promisc));
736
737         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
738                                   VIRTIO_NET_CTRL_RX_PROMISC,
739                                   sg, 1, 0))
740                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
741                          promisc ? "en" : "dis");
742
743         sg_init_one(sg, &allmulti, sizeof(allmulti));
744
745         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
746                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
747                                   sg, 1, 0))
748                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
749                          allmulti ? "en" : "dis");
750
751         uc_count = netdev_uc_count(dev);
752         mc_count = netdev_mc_count(dev);
753         /* MAC filter - use one buffer for both lists */
754         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
755                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
756         mac_data = buf;
757         if (!buf) {
758                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
759                 return;
760         }
761
762         sg_init_table(sg, 2);
763
764         /* Store the unicast list and count in the front of the buffer */
765         mac_data->entries = uc_count;
766         i = 0;
767         netdev_for_each_uc_addr(ha, dev)
768                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
769
770         sg_set_buf(&sg[0], mac_data,
771                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
772
773         /* multicast list and count fill the end */
774         mac_data = (void *)&mac_data->macs[uc_count][0];
775
776         mac_data->entries = mc_count;
777         i = 0;
778         netdev_for_each_mc_addr(ha, dev)
779                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
780
781         sg_set_buf(&sg[1], mac_data,
782                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
783
784         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
785                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
786                                   sg, 2, 0))
787                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
788
789         kfree(buf);
790 }
791
792 static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid)
793 {
794         struct virtnet_info *vi = netdev_priv(dev);
795         struct scatterlist sg;
796
797         sg_init_one(&sg, &vid, sizeof(vid));
798
799         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
800                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0))
801                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
802 }
803
804 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
805 {
806         struct virtnet_info *vi = netdev_priv(dev);
807         struct scatterlist sg;
808
809         sg_init_one(&sg, &vid, sizeof(vid));
810
811         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
812                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
813                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
814 }
815
816 static const struct ethtool_ops virtnet_ethtool_ops = {
817         .get_link = ethtool_op_get_link,
818 };
819
820 #define MIN_MTU 68
821 #define MAX_MTU 65535
822
823 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
824 {
825         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
826                 return -EINVAL;
827         dev->mtu = new_mtu;
828         return 0;
829 }
830
831 static const struct net_device_ops virtnet_netdev = {
832         .ndo_open            = virtnet_open,
833         .ndo_stop            = virtnet_close,
834         .ndo_start_xmit      = start_xmit,
835         .ndo_validate_addr   = eth_validate_addr,
836         .ndo_set_mac_address = virtnet_set_mac_address,
837         .ndo_set_rx_mode     = virtnet_set_rx_mode,
838         .ndo_change_mtu      = virtnet_change_mtu,
839         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
840         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
841 #ifdef CONFIG_NET_POLL_CONTROLLER
842         .ndo_poll_controller = virtnet_netpoll,
843 #endif
844 };
845
846 static void virtnet_update_status(struct virtnet_info *vi)
847 {
848         u16 v;
849
850         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
851                 return;
852
853         vi->vdev->config->get(vi->vdev,
854                               offsetof(struct virtio_net_config, status),
855                               &v, sizeof(v));
856
857         /* Ignore unknown (future) status bits */
858         v &= VIRTIO_NET_S_LINK_UP;
859
860         if (vi->status == v)
861                 return;
862
863         vi->status = v;
864
865         if (vi->status & VIRTIO_NET_S_LINK_UP) {
866                 netif_carrier_on(vi->dev);
867                 netif_wake_queue(vi->dev);
868         } else {
869                 netif_carrier_off(vi->dev);
870                 netif_stop_queue(vi->dev);
871         }
872 }
873
874 static void virtnet_config_changed(struct virtio_device *vdev)
875 {
876         struct virtnet_info *vi = vdev->priv;
877
878         virtnet_update_status(vi);
879 }
880
881 static int virtnet_probe(struct virtio_device *vdev)
882 {
883         int err;
884         struct net_device *dev;
885         struct virtnet_info *vi;
886         struct virtqueue *vqs[3];
887         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
888         const char *names[] = { "input", "output", "control" };
889         int nvqs;
890
891         /* Allocate ourselves a network device with room for our info */
892         dev = alloc_etherdev(sizeof(struct virtnet_info));
893         if (!dev)
894                 return -ENOMEM;
895
896         /* Set up network device as normal. */
897         dev->netdev_ops = &virtnet_netdev;
898         dev->features = NETIF_F_HIGHDMA;
899         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
900         SET_NETDEV_DEV(dev, &vdev->dev);
901
902         /* Do we support "hardware" checksums? */
903         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
904                 /* This opens up the world of extra features. */
905                 dev->hw_features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
906                 if (csum)
907                         dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
908
909                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
910                         dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
911                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
912                 }
913                 /* Individual feature bits: what can host handle? */
914                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
915                         dev->hw_features |= NETIF_F_TSO;
916                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
917                         dev->hw_features |= NETIF_F_TSO6;
918                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
919                         dev->hw_features |= NETIF_F_TSO_ECN;
920                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
921                         dev->hw_features |= NETIF_F_UFO;
922
923                 if (gso)
924                         dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
925                 /* (!csum && gso) case will be fixed by register_netdev() */
926         }
927
928         /* Configuration may specify what MAC to use.  Otherwise random. */
929         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
930                 vdev->config->get(vdev,
931                                   offsetof(struct virtio_net_config, mac),
932                                   dev->dev_addr, dev->addr_len);
933         } else
934                 random_ether_addr(dev->dev_addr);
935
936         /* Set up our device-specific information */
937         vi = netdev_priv(dev);
938         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
939         vi->dev = dev;
940         vi->vdev = vdev;
941         vdev->priv = vi;
942         vi->pages = NULL;
943         INIT_DELAYED_WORK(&vi->refill, refill_work);
944         sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg));
945         sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg));
946
947         /* If we can receive ANY GSO packets, we must allocate large ones. */
948         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
949             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
950             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
951                 vi->big_packets = true;
952
953         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
954                 vi->mergeable_rx_bufs = true;
955
956         /* We expect two virtqueues, receive then send,
957          * and optionally control. */
958         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
959
960         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
961         if (err)
962                 goto free;
963
964         vi->rvq = vqs[0];
965         vi->svq = vqs[1];
966
967         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
968                 vi->cvq = vqs[2];
969
970                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
971                         dev->features |= NETIF_F_HW_VLAN_FILTER;
972         }
973
974         err = register_netdev(dev);
975         if (err) {
976                 pr_debug("virtio_net: registering device failed\n");
977                 goto free_vqs;
978         }
979
980         /* Last of all, set up some receive buffers. */
981         try_fill_recv(vi, GFP_KERNEL);
982
983         /* If we didn't even get one input buffer, we're useless. */
984         if (vi->num == 0) {
985                 err = -ENOMEM;
986                 goto unregister;
987         }
988
989         /* Assume link up if device can't report link status,
990            otherwise get link status from config. */
991         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
992                 netif_carrier_off(dev);
993                 virtnet_update_status(vi);
994         } else {
995                 vi->status = VIRTIO_NET_S_LINK_UP;
996                 netif_carrier_on(dev);
997         }
998
999         pr_debug("virtnet: registered device %s\n", dev->name);
1000         return 0;
1001
1002 unregister:
1003         unregister_netdev(dev);
1004         cancel_delayed_work_sync(&vi->refill);
1005 free_vqs:
1006         vdev->config->del_vqs(vdev);
1007 free:
1008         free_netdev(dev);
1009         return err;
1010 }
1011
1012 static void free_unused_bufs(struct virtnet_info *vi)
1013 {
1014         void *buf;
1015         while (1) {
1016                 buf = virtqueue_detach_unused_buf(vi->svq);
1017                 if (!buf)
1018                         break;
1019                 dev_kfree_skb(buf);
1020         }
1021         while (1) {
1022                 buf = virtqueue_detach_unused_buf(vi->rvq);
1023                 if (!buf)
1024                         break;
1025                 if (vi->mergeable_rx_bufs || vi->big_packets)
1026                         give_pages(vi, buf);
1027                 else
1028                         dev_kfree_skb(buf);
1029                 --vi->num;
1030         }
1031         BUG_ON(vi->num != 0);
1032 }
1033
1034 static void __devexit virtnet_remove(struct virtio_device *vdev)
1035 {
1036         struct virtnet_info *vi = vdev->priv;
1037
1038         /* Stop all the virtqueues. */
1039         vdev->config->reset(vdev);
1040
1041
1042         unregister_netdev(vi->dev);
1043         cancel_delayed_work_sync(&vi->refill);
1044
1045         /* Free unused buffers in both send and recv, if any. */
1046         free_unused_bufs(vi);
1047
1048         vdev->config->del_vqs(vi->vdev);
1049
1050         while (vi->pages)
1051                 __free_pages(get_a_page(vi, GFP_KERNEL), 0);
1052
1053         free_netdev(vi->dev);
1054 }
1055
1056 static struct virtio_device_id id_table[] = {
1057         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
1058         { 0 },
1059 };
1060
1061 static unsigned int features[] = {
1062         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM,
1063         VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC,
1064         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6,
1065         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6,
1066         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO,
1067         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ,
1068         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN,
1069 };
1070
1071 static struct virtio_driver virtio_net_driver = {
1072         .feature_table = features,
1073         .feature_table_size = ARRAY_SIZE(features),
1074         .driver.name =  KBUILD_MODNAME,
1075         .driver.owner = THIS_MODULE,
1076         .id_table =     id_table,
1077         .probe =        virtnet_probe,
1078         .remove =       __devexit_p(virtnet_remove),
1079         .config_changed = virtnet_config_changed,
1080 };
1081
1082 static int __init init(void)
1083 {
1084         return register_virtio_driver(&virtio_net_driver);
1085 }
1086
1087 static void __exit fini(void)
1088 {
1089         unregister_virtio_driver(&virtio_net_driver);
1090 }
1091 module_init(init);
1092 module_exit(fini);
1093
1094 MODULE_DEVICE_TABLE(virtio, id_table);
1095 MODULE_DESCRIPTION("Virtio network driver");
1096 MODULE_LICENSE("GPL");