net: use netdev_mc_count and netdev_mc_empty when appropriate
[linux-2.6.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
29 static int napi_weight = 128;
30 module_param(napi_weight, int, 0444);
31
32 static int csum = 1, gso = 1;
33 module_param(csum, bool, 0444);
34 module_param(gso, bool, 0444);
35
36 /* FIXME: MTU in config. */
37 #define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
38 #define GOOD_COPY_LEN   128
39
40 #define VIRTNET_SEND_COMMAND_SG_MAX    2
41
42 struct virtnet_info
43 {
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         /* Send queue. */
60         struct sk_buff_head send;
61
62         /* Work struct for refilling if we run low on memory. */
63         struct delayed_work refill;
64
65         /* Chain pages by the private ptr. */
66         struct page *pages;
67 };
68
69 struct skb_vnet_hdr {
70         union {
71                 struct virtio_net_hdr hdr;
72                 struct virtio_net_hdr_mrg_rxbuf mhdr;
73         };
74         unsigned int num_sg;
75 };
76
77 struct padded_vnet_hdr {
78         struct virtio_net_hdr hdr;
79         /*
80          * virtio_net_hdr should be in a separated sg buffer because of a
81          * QEMU bug, and data sg buffer shares same page with this header sg.
82          * This padding makes next sg 16 byte aligned after virtio_net_hdr.
83          */
84         char padding[6];
85 };
86
87 static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb)
88 {
89         return (struct skb_vnet_hdr *)skb->cb;
90 }
91
92 /*
93  * private is used to chain pages for big packets, put the whole
94  * most recent used list in the beginning for reuse
95  */
96 static void give_pages(struct virtnet_info *vi, struct page *page)
97 {
98         struct page *end;
99
100         /* Find end of list, sew whole thing into vi->pages. */
101         for (end = page; end->private; end = (struct page *)end->private);
102         end->private = (unsigned long)vi->pages;
103         vi->pages = page;
104 }
105
106 static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask)
107 {
108         struct page *p = vi->pages;
109
110         if (p) {
111                 vi->pages = (struct page *)p->private;
112                 /* clear private here, it is used to chain pages */
113                 p->private = 0;
114         } else
115                 p = alloc_page(gfp_mask);
116         return p;
117 }
118
119 static void skb_xmit_done(struct virtqueue *svq)
120 {
121         struct virtnet_info *vi = svq->vdev->priv;
122
123         /* Suppress further interrupts. */
124         svq->vq_ops->disable_cb(svq);
125
126         /* We were probably waiting for more output buffers. */
127         netif_wake_queue(vi->dev);
128 }
129
130 static void set_skb_frag(struct sk_buff *skb, struct page *page,
131                          unsigned int offset, unsigned int *len)
132 {
133         int i = skb_shinfo(skb)->nr_frags;
134         skb_frag_t *f;
135
136         f = &skb_shinfo(skb)->frags[i];
137         f->size = min((unsigned)PAGE_SIZE - offset, *len);
138         f->page_offset = offset;
139         f->page = page;
140
141         skb->data_len += f->size;
142         skb->len += f->size;
143         skb_shinfo(skb)->nr_frags++;
144         *len -= f->size;
145 }
146
147 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
148                                    struct page *page, unsigned int len)
149 {
150         struct sk_buff *skb;
151         struct skb_vnet_hdr *hdr;
152         unsigned int copy, hdr_len, offset;
153         char *p;
154
155         p = page_address(page);
156
157         /* copy small packet so we can reuse these pages for small data */
158         skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN);
159         if (unlikely(!skb))
160                 return NULL;
161
162         hdr = skb_vnet_hdr(skb);
163
164         if (vi->mergeable_rx_bufs) {
165                 hdr_len = sizeof hdr->mhdr;
166                 offset = hdr_len;
167         } else {
168                 hdr_len = sizeof hdr->hdr;
169                 offset = sizeof(struct padded_vnet_hdr);
170         }
171
172         memcpy(hdr, p, hdr_len);
173
174         len -= hdr_len;
175         p += offset;
176
177         copy = len;
178         if (copy > skb_tailroom(skb))
179                 copy = skb_tailroom(skb);
180         memcpy(skb_put(skb, copy), p, copy);
181
182         len -= copy;
183         offset += copy;
184
185         while (len) {
186                 set_skb_frag(skb, page, offset, &len);
187                 page = (struct page *)page->private;
188                 offset = 0;
189         }
190
191         if (page)
192                 give_pages(vi, page);
193
194         return skb;
195 }
196
197 static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb)
198 {
199         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
200         struct page *page;
201         int num_buf, i, len;
202
203         num_buf = hdr->mhdr.num_buffers;
204         while (--num_buf) {
205                 i = skb_shinfo(skb)->nr_frags;
206                 if (i >= MAX_SKB_FRAGS) {
207                         pr_debug("%s: packet too long\n", skb->dev->name);
208                         skb->dev->stats.rx_length_errors++;
209                         return -EINVAL;
210                 }
211
212                 page = vi->rvq->vq_ops->get_buf(vi->rvq, &len);
213                 if (!page) {
214                         pr_debug("%s: rx error: %d buffers missing\n",
215                                  skb->dev->name, hdr->mhdr.num_buffers);
216                         skb->dev->stats.rx_length_errors++;
217                         return -EINVAL;
218                 }
219                 if (len > PAGE_SIZE)
220                         len = PAGE_SIZE;
221
222                 set_skb_frag(skb, page, 0, &len);
223
224                 --vi->num;
225         }
226         return 0;
227 }
228
229 static void receive_buf(struct net_device *dev, void *buf, unsigned int len)
230 {
231         struct virtnet_info *vi = netdev_priv(dev);
232         struct sk_buff *skb;
233         struct page *page;
234         struct skb_vnet_hdr *hdr;
235
236         if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
237                 pr_debug("%s: short packet %i\n", dev->name, len);
238                 dev->stats.rx_length_errors++;
239                 if (vi->mergeable_rx_bufs || vi->big_packets)
240                         give_pages(vi, buf);
241                 else
242                         dev_kfree_skb(buf);
243                 return;
244         }
245
246         if (!vi->mergeable_rx_bufs && !vi->big_packets) {
247                 skb = buf;
248                 len -= sizeof(struct virtio_net_hdr);
249                 skb_trim(skb, len);
250         } else {
251                 page = buf;
252                 skb = page_to_skb(vi, page, len);
253                 if (unlikely(!skb)) {
254                         dev->stats.rx_dropped++;
255                         give_pages(vi, page);
256                         return;
257                 }
258                 if (vi->mergeable_rx_bufs)
259                         if (receive_mergeable(vi, skb)) {
260                                 dev_kfree_skb(skb);
261                                 return;
262                         }
263         }
264
265         hdr = skb_vnet_hdr(skb);
266         skb->truesize += skb->data_len;
267         dev->stats.rx_bytes += skb->len;
268         dev->stats.rx_packets++;
269
270         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
271                 pr_debug("Needs csum!\n");
272                 if (!skb_partial_csum_set(skb,
273                                           hdr->hdr.csum_start,
274                                           hdr->hdr.csum_offset))
275                         goto frame_err;
276         }
277
278         skb->protocol = eth_type_trans(skb, dev);
279         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
280                  ntohs(skb->protocol), skb->len, skb->pkt_type);
281
282         if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
283                 pr_debug("GSO!\n");
284                 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
285                 case VIRTIO_NET_HDR_GSO_TCPV4:
286                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
287                         break;
288                 case VIRTIO_NET_HDR_GSO_UDP:
289                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
290                         break;
291                 case VIRTIO_NET_HDR_GSO_TCPV6:
292                         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
293                         break;
294                 default:
295                         if (net_ratelimit())
296                                 printk(KERN_WARNING "%s: bad gso type %u.\n",
297                                        dev->name, hdr->hdr.gso_type);
298                         goto frame_err;
299                 }
300
301                 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
302                         skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
303
304                 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
305                 if (skb_shinfo(skb)->gso_size == 0) {
306                         if (net_ratelimit())
307                                 printk(KERN_WARNING "%s: zero gso size.\n",
308                                        dev->name);
309                         goto frame_err;
310                 }
311
312                 /* Header must be checked, and gso_segs computed. */
313                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
314                 skb_shinfo(skb)->gso_segs = 0;
315         }
316
317         netif_receive_skb(skb);
318         return;
319
320 frame_err:
321         dev->stats.rx_frame_errors++;
322         dev_kfree_skb(skb);
323 }
324
325 static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp)
326 {
327         struct sk_buff *skb;
328         struct skb_vnet_hdr *hdr;
329         struct scatterlist sg[2];
330         int err;
331
332         skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN);
333         if (unlikely(!skb))
334                 return -ENOMEM;
335
336         skb_put(skb, MAX_PACKET_LEN);
337
338         hdr = skb_vnet_hdr(skb);
339         sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
340
341         skb_to_sgvec(skb, sg + 1, 0, skb->len);
342
343         err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, 2, skb);
344         if (err < 0)
345                 dev_kfree_skb(skb);
346
347         return err;
348 }
349
350 static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp)
351 {
352         struct scatterlist sg[MAX_SKB_FRAGS + 2];
353         struct page *first, *list = NULL;
354         char *p;
355         int i, err, offset;
356
357         /* page in sg[MAX_SKB_FRAGS + 1] is list tail */
358         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
359                 first = get_a_page(vi, gfp);
360                 if (!first) {
361                         if (list)
362                                 give_pages(vi, list);
363                         return -ENOMEM;
364                 }
365                 sg_set_buf(&sg[i], page_address(first), PAGE_SIZE);
366
367                 /* chain new page in list head to match sg */
368                 first->private = (unsigned long)list;
369                 list = first;
370         }
371
372         first = get_a_page(vi, gfp);
373         if (!first) {
374                 give_pages(vi, list);
375                 return -ENOMEM;
376         }
377         p = page_address(first);
378
379         /* sg[0], sg[1] share the same page */
380         /* a separated sg[0] for  virtio_net_hdr only during to QEMU bug*/
381         sg_set_buf(&sg[0], p, sizeof(struct virtio_net_hdr));
382
383         /* sg[1] for data packet, from offset */
384         offset = sizeof(struct padded_vnet_hdr);
385         sg_set_buf(&sg[1], p + offset, PAGE_SIZE - offset);
386
387         /* chain first in list head */
388         first->private = (unsigned long)list;
389         err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, MAX_SKB_FRAGS + 2,
390                                        first);
391         if (err < 0)
392                 give_pages(vi, first);
393
394         return err;
395 }
396
397 static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp)
398 {
399         struct page *page;
400         struct scatterlist sg;
401         int err;
402
403         page = get_a_page(vi, gfp);
404         if (!page)
405                 return -ENOMEM;
406
407         sg_init_one(&sg, page_address(page), PAGE_SIZE);
408
409         err = vi->rvq->vq_ops->add_buf(vi->rvq, &sg, 0, 1, page);
410         if (err < 0)
411                 give_pages(vi, page);
412
413         return err;
414 }
415
416 /* Returns false if we couldn't fill entirely (OOM). */
417 static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp)
418 {
419         int err;
420         bool oom = false;
421
422         do {
423                 if (vi->mergeable_rx_bufs)
424                         err = add_recvbuf_mergeable(vi, gfp);
425                 else if (vi->big_packets)
426                         err = add_recvbuf_big(vi, gfp);
427                 else
428                         err = add_recvbuf_small(vi, gfp);
429
430                 if (err < 0) {
431                         oom = true;
432                         break;
433                 }
434                 ++vi->num;
435         } while (err > 0);
436         if (unlikely(vi->num > vi->max))
437                 vi->max = vi->num;
438         vi->rvq->vq_ops->kick(vi->rvq);
439         return !oom;
440 }
441
442 static void skb_recv_done(struct virtqueue *rvq)
443 {
444         struct virtnet_info *vi = rvq->vdev->priv;
445         /* Schedule NAPI, Suppress further interrupts if successful. */
446         if (napi_schedule_prep(&vi->napi)) {
447                 rvq->vq_ops->disable_cb(rvq);
448                 __napi_schedule(&vi->napi);
449         }
450 }
451
452 static void refill_work(struct work_struct *work)
453 {
454         struct virtnet_info *vi;
455         bool still_empty;
456
457         vi = container_of(work, struct virtnet_info, refill.work);
458         napi_disable(&vi->napi);
459         still_empty = !try_fill_recv(vi, GFP_KERNEL);
460         napi_enable(&vi->napi);
461
462         /* In theory, this can happen: if we don't get any buffers in
463          * we will *never* try to fill again. */
464         if (still_empty)
465                 schedule_delayed_work(&vi->refill, HZ/2);
466 }
467
468 static int virtnet_poll(struct napi_struct *napi, int budget)
469 {
470         struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
471         void *buf;
472         unsigned int len, received = 0;
473
474 again:
475         while (received < budget &&
476                (buf = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
477                 receive_buf(vi->dev, buf, len);
478                 --vi->num;
479                 received++;
480         }
481
482         if (vi->num < vi->max / 2) {
483                 if (!try_fill_recv(vi, GFP_ATOMIC))
484                         schedule_delayed_work(&vi->refill, 0);
485         }
486
487         /* Out of packets? */
488         if (received < budget) {
489                 napi_complete(napi);
490                 if (unlikely(!vi->rvq->vq_ops->enable_cb(vi->rvq)) &&
491                     napi_schedule_prep(napi)) {
492                         vi->rvq->vq_ops->disable_cb(vi->rvq);
493                         __napi_schedule(napi);
494                         goto again;
495                 }
496         }
497
498         return received;
499 }
500
501 static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
502 {
503         struct sk_buff *skb;
504         unsigned int len, tot_sgs = 0;
505
506         while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
507                 pr_debug("Sent skb %p\n", skb);
508                 __skb_unlink(skb, &vi->send);
509                 vi->dev->stats.tx_bytes += skb->len;
510                 vi->dev->stats.tx_packets++;
511                 tot_sgs += skb_vnet_hdr(skb)->num_sg;
512                 dev_kfree_skb_any(skb);
513         }
514         return tot_sgs;
515 }
516
517 static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb)
518 {
519         struct scatterlist sg[2+MAX_SKB_FRAGS];
520         struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb);
521         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
522
523         sg_init_table(sg, 2+MAX_SKB_FRAGS);
524
525         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
526
527         if (skb->ip_summed == CHECKSUM_PARTIAL) {
528                 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
529                 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb);
530                 hdr->hdr.csum_offset = skb->csum_offset;
531         } else {
532                 hdr->hdr.flags = 0;
533                 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0;
534         }
535
536         if (skb_is_gso(skb)) {
537                 hdr->hdr.hdr_len = skb_headlen(skb);
538                 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size;
539                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
540                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
541                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
542                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
543                 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
544                         hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
545                 else
546                         BUG();
547                 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
548                         hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
549         } else {
550                 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
551                 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0;
552         }
553
554         hdr->mhdr.num_buffers = 0;
555
556         /* Encode metadata header at front. */
557         if (vi->mergeable_rx_bufs)
558                 sg_set_buf(sg, &hdr->mhdr, sizeof hdr->mhdr);
559         else
560                 sg_set_buf(sg, &hdr->hdr, sizeof hdr->hdr);
561
562         hdr->num_sg = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
563         return vi->svq->vq_ops->add_buf(vi->svq, sg, hdr->num_sg, 0, skb);
564 }
565
566 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
567 {
568         struct virtnet_info *vi = netdev_priv(dev);
569         int capacity;
570
571 again:
572         /* Free up any pending old buffers before queueing new ones. */
573         free_old_xmit_skbs(vi);
574
575         /* Try to transmit */
576         capacity = xmit_skb(vi, skb);
577
578         /* This can happen with OOM and indirect buffers. */
579         if (unlikely(capacity < 0)) {
580                 netif_stop_queue(dev);
581                 dev_warn(&dev->dev, "Unexpected full queue\n");
582                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
583                         vi->svq->vq_ops->disable_cb(vi->svq);
584                         netif_start_queue(dev);
585                         goto again;
586                 }
587                 return NETDEV_TX_BUSY;
588         }
589         vi->svq->vq_ops->kick(vi->svq);
590
591         /*
592          * Put new one in send queue.  You'd expect we'd need this before
593          * xmit_skb calls add_buf(), since the callback can be triggered
594          * immediately after that.  But since the callback just triggers
595          * another call back here, normal network xmit locking prevents the
596          * race.
597          */
598         __skb_queue_head(&vi->send, skb);
599
600         /* Don't wait up for transmitted skbs to be freed. */
601         skb_orphan(skb);
602         nf_reset(skb);
603
604         /* Apparently nice girls don't return TX_BUSY; stop the queue
605          * before it gets out of hand.  Naturally, this wastes entries. */
606         if (capacity < 2+MAX_SKB_FRAGS) {
607                 netif_stop_queue(dev);
608                 if (unlikely(!vi->svq->vq_ops->enable_cb(vi->svq))) {
609                         /* More just got used, free them then recheck. */
610                         capacity += free_old_xmit_skbs(vi);
611                         if (capacity >= 2+MAX_SKB_FRAGS) {
612                                 netif_start_queue(dev);
613                                 vi->svq->vq_ops->disable_cb(vi->svq);
614                         }
615                 }
616         }
617
618         return NETDEV_TX_OK;
619 }
620
621 static int virtnet_set_mac_address(struct net_device *dev, void *p)
622 {
623         struct virtnet_info *vi = netdev_priv(dev);
624         struct virtio_device *vdev = vi->vdev;
625         int ret;
626
627         ret = eth_mac_addr(dev, p);
628         if (ret)
629                 return ret;
630
631         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
632                 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac),
633                                   dev->dev_addr, dev->addr_len);
634
635         return 0;
636 }
637
638 #ifdef CONFIG_NET_POLL_CONTROLLER
639 static void virtnet_netpoll(struct net_device *dev)
640 {
641         struct virtnet_info *vi = netdev_priv(dev);
642
643         napi_schedule(&vi->napi);
644 }
645 #endif
646
647 static int virtnet_open(struct net_device *dev)
648 {
649         struct virtnet_info *vi = netdev_priv(dev);
650
651         napi_enable(&vi->napi);
652
653         /* If all buffers were filled by other side before we napi_enabled, we
654          * won't get another interrupt, so process any outstanding packets
655          * now.  virtnet_poll wants re-enable the queue, so we disable here.
656          * We synchronize against interrupts via NAPI_STATE_SCHED */
657         if (napi_schedule_prep(&vi->napi)) {
658                 vi->rvq->vq_ops->disable_cb(vi->rvq);
659                 __napi_schedule(&vi->napi);
660         }
661         return 0;
662 }
663
664 /*
665  * Send command via the control virtqueue and check status.  Commands
666  * supported by the hypervisor, as indicated by feature bits, should
667  * never fail unless improperly formated.
668  */
669 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
670                                  struct scatterlist *data, int out, int in)
671 {
672         struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2];
673         struct virtio_net_ctrl_hdr ctrl;
674         virtio_net_ctrl_ack status = ~0;
675         unsigned int tmp;
676         int i;
677
678         /* Caller should know better */
679         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ||
680                 (out + in > VIRTNET_SEND_COMMAND_SG_MAX));
681
682         out++; /* Add header */
683         in++; /* Add return status */
684
685         ctrl.class = class;
686         ctrl.cmd = cmd;
687
688         sg_init_table(sg, out + in);
689
690         sg_set_buf(&sg[0], &ctrl, sizeof(ctrl));
691         for_each_sg(data, s, out + in - 2, i)
692                 sg_set_buf(&sg[i + 1], sg_virt(s), s->length);
693         sg_set_buf(&sg[out + in - 1], &status, sizeof(status));
694
695         BUG_ON(vi->cvq->vq_ops->add_buf(vi->cvq, sg, out, in, vi) < 0);
696
697         vi->cvq->vq_ops->kick(vi->cvq);
698
699         /*
700          * Spin for a response, the kick causes an ioport write, trapping
701          * into the hypervisor, so the request should be handled immediately.
702          */
703         while (!vi->cvq->vq_ops->get_buf(vi->cvq, &tmp))
704                 cpu_relax();
705
706         return status == VIRTIO_NET_OK;
707 }
708
709 static int virtnet_close(struct net_device *dev)
710 {
711         struct virtnet_info *vi = netdev_priv(dev);
712
713         napi_disable(&vi->napi);
714
715         return 0;
716 }
717
718 static int virtnet_set_tx_csum(struct net_device *dev, u32 data)
719 {
720         struct virtnet_info *vi = netdev_priv(dev);
721         struct virtio_device *vdev = vi->vdev;
722
723         if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM))
724                 return -ENOSYS;
725
726         return ethtool_op_set_tx_hw_csum(dev, data);
727 }
728
729 static void virtnet_set_rx_mode(struct net_device *dev)
730 {
731         struct virtnet_info *vi = netdev_priv(dev);
732         struct scatterlist sg[2];
733         u8 promisc, allmulti;
734         struct virtio_net_ctrl_mac *mac_data;
735         struct dev_addr_list *addr;
736         struct netdev_hw_addr *ha;
737         int uc_count;
738         int mc_count;
739         void *buf;
740         int i;
741
742         /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */
743         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
744                 return;
745
746         promisc = ((dev->flags & IFF_PROMISC) != 0);
747         allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
748
749         sg_init_one(sg, &promisc, sizeof(promisc));
750
751         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
752                                   VIRTIO_NET_CTRL_RX_PROMISC,
753                                   sg, 1, 0))
754                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
755                          promisc ? "en" : "dis");
756
757         sg_init_one(sg, &allmulti, sizeof(allmulti));
758
759         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
760                                   VIRTIO_NET_CTRL_RX_ALLMULTI,
761                                   sg, 1, 0))
762                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
763                          allmulti ? "en" : "dis");
764
765         uc_count = netdev_uc_count(dev);
766         mc_count = netdev_mc_count(dev);
767         /* MAC filter - use one buffer for both lists */
768         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
769                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
770         mac_data = buf;
771         if (!buf) {
772                 dev_warn(&dev->dev, "No memory for MAC address buffer\n");
773                 return;
774         }
775
776         sg_init_table(sg, 2);
777
778         /* Store the unicast list and count in the front of the buffer */
779         mac_data->entries = uc_count;
780         i = 0;
781         netdev_for_each_uc_addr(ha, dev)
782                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
783
784         sg_set_buf(&sg[0], mac_data,
785                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
786
787         /* multicast list and count fill the end */
788         mac_data = (void *)&mac_data->macs[uc_count][0];
789
790         mac_data->entries = mc_count;
791         addr = dev->mc_list;
792         for (i = 0; i < mc_count; i++, addr = addr->next)
793                 memcpy(&mac_data->macs[i][0], addr->da_addr, ETH_ALEN);
794
795         sg_set_buf(&sg[1], mac_data,
796                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
797
798         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
799                                   VIRTIO_NET_CTRL_MAC_TABLE_SET,
800                                   sg, 2, 0))
801                 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n");
802
803         kfree(buf);
804 }
805
806 static void virtnet_vlan_rx_add_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_ADD, &sg, 1, 0))
815                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
816 }
817
818 static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid)
819 {
820         struct virtnet_info *vi = netdev_priv(dev);
821         struct scatterlist sg;
822
823         sg_init_one(&sg, &vid, sizeof(vid));
824
825         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
826                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0))
827                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
828 }
829
830 static const struct ethtool_ops virtnet_ethtool_ops = {
831         .set_tx_csum = virtnet_set_tx_csum,
832         .set_sg = ethtool_op_set_sg,
833         .set_tso = ethtool_op_set_tso,
834         .set_ufo = ethtool_op_set_ufo,
835         .get_link = ethtool_op_get_link,
836 };
837
838 #define MIN_MTU 68
839 #define MAX_MTU 65535
840
841 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
842 {
843         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
844                 return -EINVAL;
845         dev->mtu = new_mtu;
846         return 0;
847 }
848
849 static const struct net_device_ops virtnet_netdev = {
850         .ndo_open            = virtnet_open,
851         .ndo_stop            = virtnet_close,
852         .ndo_start_xmit      = start_xmit,
853         .ndo_validate_addr   = eth_validate_addr,
854         .ndo_set_mac_address = virtnet_set_mac_address,
855         .ndo_set_rx_mode     = virtnet_set_rx_mode,
856         .ndo_change_mtu      = virtnet_change_mtu,
857         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
858         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
859 #ifdef CONFIG_NET_POLL_CONTROLLER
860         .ndo_poll_controller = virtnet_netpoll,
861 #endif
862 };
863
864 static void virtnet_update_status(struct virtnet_info *vi)
865 {
866         u16 v;
867
868         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS))
869                 return;
870
871         vi->vdev->config->get(vi->vdev,
872                               offsetof(struct virtio_net_config, status),
873                               &v, sizeof(v));
874
875         /* Ignore unknown (future) status bits */
876         v &= VIRTIO_NET_S_LINK_UP;
877
878         if (vi->status == v)
879                 return;
880
881         vi->status = v;
882
883         if (vi->status & VIRTIO_NET_S_LINK_UP) {
884                 netif_carrier_on(vi->dev);
885                 netif_wake_queue(vi->dev);
886         } else {
887                 netif_carrier_off(vi->dev);
888                 netif_stop_queue(vi->dev);
889         }
890 }
891
892 static void virtnet_config_changed(struct virtio_device *vdev)
893 {
894         struct virtnet_info *vi = vdev->priv;
895
896         virtnet_update_status(vi);
897 }
898
899 static int virtnet_probe(struct virtio_device *vdev)
900 {
901         int err;
902         struct net_device *dev;
903         struct virtnet_info *vi;
904         struct virtqueue *vqs[3];
905         vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL};
906         const char *names[] = { "input", "output", "control" };
907         int nvqs;
908
909         /* Allocate ourselves a network device with room for our info */
910         dev = alloc_etherdev(sizeof(struct virtnet_info));
911         if (!dev)
912                 return -ENOMEM;
913
914         /* Set up network device as normal. */
915         dev->netdev_ops = &virtnet_netdev;
916         dev->features = NETIF_F_HIGHDMA;
917         SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops);
918         SET_NETDEV_DEV(dev, &vdev->dev);
919
920         /* Do we support "hardware" checksums? */
921         if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
922                 /* This opens up the world of extra features. */
923                 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
924                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
925                         dev->features |= NETIF_F_TSO | NETIF_F_UFO
926                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
927                 }
928                 /* Individual feature bits: what can host handle? */
929                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
930                         dev->features |= NETIF_F_TSO;
931                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
932                         dev->features |= NETIF_F_TSO6;
933                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
934                         dev->features |= NETIF_F_TSO_ECN;
935                 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
936                         dev->features |= NETIF_F_UFO;
937         }
938
939         /* Configuration may specify what MAC to use.  Otherwise random. */
940         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
941                 vdev->config->get(vdev,
942                                   offsetof(struct virtio_net_config, mac),
943                                   dev->dev_addr, dev->addr_len);
944         } else
945                 random_ether_addr(dev->dev_addr);
946
947         /* Set up our device-specific information */
948         vi = netdev_priv(dev);
949         netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight);
950         vi->dev = dev;
951         vi->vdev = vdev;
952         vdev->priv = vi;
953         vi->pages = NULL;
954         INIT_DELAYED_WORK(&vi->refill, refill_work);
955
956         /* If we can receive ANY GSO packets, we must allocate large ones. */
957         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
958             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
959             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN))
960                 vi->big_packets = true;
961
962         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
963                 vi->mergeable_rx_bufs = true;
964
965         /* We expect two virtqueues, receive then send,
966          * and optionally control. */
967         nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2;
968
969         err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names);
970         if (err)
971                 goto free;
972
973         vi->rvq = vqs[0];
974         vi->svq = vqs[1];
975
976         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) {
977                 vi->cvq = vqs[2];
978
979                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
980                         dev->features |= NETIF_F_HW_VLAN_FILTER;
981         }
982
983         /* Initialize our empty send queue. */
984         skb_queue_head_init(&vi->send);
985
986         err = register_netdev(dev);
987         if (err) {
988                 pr_debug("virtio_net: registering device failed\n");
989                 goto free_vqs;
990         }
991
992         /* Last of all, set up some receive buffers. */
993         try_fill_recv(vi, GFP_KERNEL);
994
995         /* If we didn't even get one input buffer, we're useless. */
996         if (vi->num == 0) {
997                 err = -ENOMEM;
998                 goto unregister;
999         }
1000
1001         vi->status = VIRTIO_NET_S_LINK_UP;
1002         virtnet_update_status(vi);
1003         netif_carrier_on(dev);
1004
1005         pr_debug("virtnet: registered device %s\n", dev->name);
1006         return 0;
1007
1008 unregister:
1009         unregister_netdev(dev);
1010         cancel_delayed_work_sync(&vi->refill);
1011 free_vqs:
1012         vdev->config->del_vqs(vdev);
1013 free:
1014         free_netdev(dev);
1015         return err;
1016 }
1017
1018 static void free_unused_bufs(struct virtnet_info *vi)
1019 {
1020         void *buf;
1021         while (1) {
1022                 buf = vi->rvq->vq_ops->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         /* Free our skbs in send queue, if any. */
1042         __skb_queue_purge(&vi->send);
1043
1044         unregister_netdev(vi->dev);
1045         cancel_delayed_work_sync(&vi->refill);
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");