?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Bluetooth: hci_uart: Use generic functionality from Broadcom module
[linux/fpc-iii.git] / drivers / net / hyperv / netvsc_drv.c
blobf9db6bc513e954b0831d881baae5048ad7cc9f87
1 /*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
19 */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
27 #include <linux/io.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/slab.h>
36 #include <net/arp.h>
37 #include <net/route.h>
38 #include <net/sock.h>
39 #include <net/pkt_sched.h>
40
41 #include "hyperv_net.h"
42
43 struct net_device_context {
44 /* point back to our device context */
45 struct hv_device *device_ctx;
46 struct delayed_work dwork;
47 struct work_struct work;
48 };
49
50 #define RING_SIZE_MIN 64
51 static int ring_size = 128;
52 module_param(ring_size, int, S_IRUGO);
53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
54
55 static void do_set_multicast(struct work_struct *w)
56 {
57 struct net_device_context *ndevctx =
58 container_of(w, struct net_device_context, work);
59 struct netvsc_device *nvdev;
60 struct rndis_device *rdev;
61
62 nvdev = hv_get_drvdata(ndevctx->device_ctx);
63 if (nvdev == NULL || nvdev->ndev == NULL)
64 return;
65
66 rdev = nvdev->extension;
67 if (rdev == NULL)
68 return;
69
70 if (nvdev->ndev->flags & IFF_PROMISC)
71 rndis_filter_set_packet_filter(rdev,
72 NDIS_PACKET_TYPE_PROMISCUOUS);
73 else
74 rndis_filter_set_packet_filter(rdev,
75 NDIS_PACKET_TYPE_BROADCAST |
76 NDIS_PACKET_TYPE_ALL_MULTICAST |
77 NDIS_PACKET_TYPE_DIRECTED);
78 }
79
80 static void netvsc_set_multicast_list(struct net_device *net)
81 {
82 struct net_device_context *net_device_ctx = netdev_priv(net);
83
84 schedule_work(&net_device_ctx->work);
85 }
86
87 static int netvsc_open(struct net_device *net)
88 {
89 struct net_device_context *net_device_ctx = netdev_priv(net);
90 struct hv_device *device_obj = net_device_ctx->device_ctx;
91 struct netvsc_device *nvdev;
92 struct rndis_device *rdev;
93 int ret = 0;
94
95 netif_carrier_off(net);
96
97 /* Open up the device */
98 ret = rndis_filter_open(device_obj);
99 if (ret != 0) {
100 netdev_err(net, "unable to open device (ret %d).\n", ret);
101 return ret;
102 }
103
104 netif_tx_start_all_queues(net);
105
106 nvdev = hv_get_drvdata(device_obj);
107 rdev = nvdev->extension;
108 if (!rdev->link_state)
109 netif_carrier_on(net);
110
111 return ret;
112 }
113
114 static int netvsc_close(struct net_device *net)
115 {
116 struct net_device_context *net_device_ctx = netdev_priv(net);
117 struct hv_device *device_obj = net_device_ctx->device_ctx;
118 int ret;
119
120 netif_tx_disable(net);
121
122 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
123 cancel_work_sync(&net_device_ctx->work);
124 ret = rndis_filter_close(device_obj);
125 if (ret != 0)
126 netdev_err(net, "unable to close device (ret %d).\n", ret);
127
128 return ret;
129 }
130
131 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
132 int pkt_type)
133 {
134 struct rndis_packet *rndis_pkt;
135 struct rndis_per_packet_info *ppi;
136
137 rndis_pkt = &msg->msg.pkt;
138 rndis_pkt->data_offset += ppi_size;
139
140 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
141 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
142
143 ppi->size = ppi_size;
144 ppi->type = pkt_type;
145 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
146
147 rndis_pkt->per_pkt_info_len += ppi_size;
148
149 return ppi;
150 }
151
152 union sub_key {
153 u64 k;
154 struct {
155 u8 pad[3];
156 u8 kb;
157 u32 ka;
158 };
159 };
160
161 /* Toeplitz hash function
162 * data: network byte order
163 * return: host byte order
164 */
165 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
166 {
167 union sub_key subk;
168 int k_next = 4;
169 u8 dt;
170 int i, j;
171 u32 ret = 0;
172
173 subk.k = 0;
174 subk.ka = ntohl(*(u32 *)key);
175
176 for (i = 0; i < dlen; i++) {
177 subk.kb = key[k_next];
178 k_next = (k_next + 1) % klen;
179 dt = ((u8 *)data)[i];
180 for (j = 0; j < 8; j++) {
181 if (dt & 0x80)
182 ret ^= subk.ka;
183 dt <<= 1;
184 subk.k <<= 1;
185 }
186 }
187
188 return ret;
189 }
190
191 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
192 {
193 struct flow_keys flow;
194 int data_len;
195
196 if (!skb_flow_dissect(skb, &flow) ||
197 !(flow.n_proto == htons(ETH_P_IP) ||
198 flow.n_proto == htons(ETH_P_IPV6)))
199 return false;
200
201 if (flow.ip_proto == IPPROTO_TCP)
202 data_len = 12;
203 else
204 data_len = 8;
205
206 *hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
207
208 return true;
209 }
210
211 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
212 void *accel_priv, select_queue_fallback_t fallback)
213 {
214 struct net_device_context *net_device_ctx = netdev_priv(ndev);
215 struct hv_device *hdev = net_device_ctx->device_ctx;
216 struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
217 u32 hash;
218 u16 q_idx = 0;
219
220 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
221 return 0;
222
223 if (netvsc_set_hash(&hash, skb)) {
224 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
225 ndev->real_num_tx_queues;
226 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
227 }
228
229 return q_idx;
230 }
231
232 static void netvsc_xmit_completion(void *context)
233 {
234 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
235 struct sk_buff *skb = (struct sk_buff *)
236 (unsigned long)packet->send_completion_tid;
237
238 if (!packet->part_of_skb)
239 kfree(packet);
240
241 if (skb)
242 dev_kfree_skb_any(skb);
243 }
244
245 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
246 struct hv_page_buffer *pb)
247 {
248 int j = 0;
249
250 /* Deal with compund pages by ignoring unused part
251 * of the page.
252 */
253 page += (offset >> PAGE_SHIFT);
254 offset &= ~PAGE_MASK;
255
256 while (len > 0) {
257 unsigned long bytes;
258
259 bytes = PAGE_SIZE - offset;
260 if (bytes > len)
261 bytes = len;
262 pb[j].pfn = page_to_pfn(page);
263 pb[j].offset = offset;
264 pb[j].len = bytes;
265
266 offset += bytes;
267 len -= bytes;
268
269 if (offset == PAGE_SIZE && len) {
270 page++;
271 offset = 0;
272 j++;
273 }
274 }
275
276 return j + 1;
277 }
278
279 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
280 struct hv_page_buffer *pb)
281 {
282 u32 slots_used = 0;
283 char *data = skb->data;
284 int frags = skb_shinfo(skb)->nr_frags;
285 int i;
286
287 /* The packet is laid out thus:
288 * 1. hdr
289 * 2. skb linear data
290 * 3. skb fragment data
291 */
292 if (hdr != NULL)
293 slots_used += fill_pg_buf(virt_to_page(hdr),
294 offset_in_page(hdr),
295 len, &pb[slots_used]);
296
297 slots_used += fill_pg_buf(virt_to_page(data),
298 offset_in_page(data),
299 skb_headlen(skb), &pb[slots_used]);
300
301 for (i = 0; i < frags; i++) {
302 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
303
304 slots_used += fill_pg_buf(skb_frag_page(frag),
305 frag->page_offset,
306 skb_frag_size(frag), &pb[slots_used]);
307 }
308 return slots_used;
309 }
310
311 static int count_skb_frag_slots(struct sk_buff *skb)
312 {
313 int i, frags = skb_shinfo(skb)->nr_frags;
314 int pages = 0;
315
316 for (i = 0; i < frags; i++) {
317 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
318 unsigned long size = skb_frag_size(frag);
319 unsigned long offset = frag->page_offset;
320
321 /* Skip unused frames from start of page */
322 offset &= ~PAGE_MASK;
323 pages += PFN_UP(offset + size);
324 }
325 return pages;
326 }
327
328 static int netvsc_get_slots(struct sk_buff *skb)
329 {
330 char *data = skb->data;
331 unsigned int offset = offset_in_page(data);
332 unsigned int len = skb_headlen(skb);
333 int slots;
334 int frag_slots;
335
336 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
337 frag_slots = count_skb_frag_slots(skb);
338 return slots + frag_slots;
339 }
340
341 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
342 {
343 u32 ret_val = TRANSPORT_INFO_NOT_IP;
344
345 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
346 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
347 goto not_ip;
348 }
349
350 *trans_off = skb_transport_offset(skb);
351
352 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
353 struct iphdr *iphdr = ip_hdr(skb);
354
355 if (iphdr->protocol == IPPROTO_TCP)
356 ret_val = TRANSPORT_INFO_IPV4_TCP;
357 else if (iphdr->protocol == IPPROTO_UDP)
358 ret_val = TRANSPORT_INFO_IPV4_UDP;
359 } else {
360 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
361 ret_val = TRANSPORT_INFO_IPV6_TCP;
362 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
363 ret_val = TRANSPORT_INFO_IPV6_UDP;
364 }
365
366 not_ip:
367 return ret_val;
368 }
369
370 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
371 {
372 struct net_device_context *net_device_ctx = netdev_priv(net);
373 struct hv_netvsc_packet *packet;
374 int ret;
375 unsigned int num_data_pgs;
376 struct rndis_message *rndis_msg;
377 struct rndis_packet *rndis_pkt;
378 u32 rndis_msg_size;
379 bool isvlan;
380 struct rndis_per_packet_info *ppi;
381 struct ndis_tcp_ip_checksum_info *csum_info;
382 struct ndis_tcp_lso_info *lso_info;
383 int hdr_offset;
384 u32 net_trans_info;
385 u32 hash;
386 u32 skb_length = skb->len;
387 u32 head_room = skb_headroom(skb);
388 u32 pkt_sz;
389 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
390
391
392 /* We will atmost need two pages to describe the rndis
393 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
394 * of pages in a single packet.
395 */
396 num_data_pgs = netvsc_get_slots(skb) + 2;
397 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
398 netdev_err(net, "Packet too big: %u\n", skb->len);
399 dev_kfree_skb(skb);
400 net->stats.tx_dropped++;
401 return NETDEV_TX_OK;
402 }
403
404 pkt_sz = sizeof(struct hv_netvsc_packet) +
405 sizeof(struct rndis_message) +
406 NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
407 NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE;
408
409 if (head_room < pkt_sz) {
410 packet = kmalloc(pkt_sz, GFP_ATOMIC);
411 if (!packet) {
412 /* out of memory, drop packet */
413 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
414 dev_kfree_skb(skb);
415 net->stats.tx_dropped++;
416 return NETDEV_TX_OK;
417 }
418 packet->part_of_skb = false;
419 } else {
420 /* Use the headroom for building up the packet */
421 packet = (struct hv_netvsc_packet *)skb->head;
422 packet->part_of_skb = true;
423 }
424
425 packet->status = 0;
426 packet->xmit_more = skb->xmit_more;
427
428 packet->vlan_tci = skb->vlan_tci;
429 packet->page_buf = page_buf;
430
431 packet->q_idx = skb_get_queue_mapping(skb);
432
433 packet->is_data_pkt = true;
434 packet->total_data_buflen = skb->len;
435
436 packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
437 sizeof(struct hv_netvsc_packet));
438
439 memset(packet->rndis_msg, 0, sizeof(struct rndis_message) +
440 NDIS_VLAN_PPI_SIZE +
441 NDIS_CSUM_PPI_SIZE +
442 NDIS_LSO_PPI_SIZE +
443 NDIS_HASH_PPI_SIZE);
444
445 /* Set the completion routine */
446 packet->send_completion = netvsc_xmit_completion;
447 packet->send_completion_ctx = packet;
448 packet->send_completion_tid = (unsigned long)skb;
449
450 isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
451
452 /* Add the rndis header */
453 rndis_msg = packet->rndis_msg;
454 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
455 rndis_msg->msg_len = packet->total_data_buflen;
456 rndis_pkt = &rndis_msg->msg.pkt;
457 rndis_pkt->data_offset = sizeof(struct rndis_packet);
458 rndis_pkt->data_len = packet->total_data_buflen;
459 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
460
461 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
462
463 hash = skb_get_hash_raw(skb);
464 if (hash != 0 && net->real_num_tx_queues > 1) {
465 rndis_msg_size += NDIS_HASH_PPI_SIZE;
466 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
467 NBL_HASH_VALUE);
468 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
469 }
470
471 if (isvlan) {
472 struct ndis_pkt_8021q_info *vlan;
473
474 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
475 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
476 IEEE_8021Q_INFO);
477 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
478 ppi->ppi_offset);
479 vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
480 vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
481 VLAN_PRIO_SHIFT;
482 }
483
484 net_trans_info = get_net_transport_info(skb, &hdr_offset);
485 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
486 goto do_send;
487
488 /*
489 * Setup the sendside checksum offload only if this is not a
490 * GSO packet.
491 */
492 if (skb_is_gso(skb))
493 goto do_lso;
494
495 if ((skb->ip_summed == CHECKSUM_NONE) ||
496 (skb->ip_summed == CHECKSUM_UNNECESSARY))
497 goto do_send;
498
499 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
500 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
501 TCPIP_CHKSUM_PKTINFO);
502
503 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
504 ppi->ppi_offset);
505
506 if (net_trans_info & (INFO_IPV4 << 16))
507 csum_info->transmit.is_ipv4 = 1;
508 else
509 csum_info->transmit.is_ipv6 = 1;
510
511 if (net_trans_info & INFO_TCP) {
512 csum_info->transmit.tcp_checksum = 1;
513 csum_info->transmit.tcp_header_offset = hdr_offset;
514 } else if (net_trans_info & INFO_UDP) {
515 /* UDP checksum offload is not supported on ws2008r2.
516 * Furthermore, on ws2012 and ws2012r2, there are some
517 * issues with udp checksum offload from Linux guests.
518 * (these are host issues).
519 * For now compute the checksum here.
520 */
521 struct udphdr *uh;
522 u16 udp_len;
523
524 ret = skb_cow_head(skb, 0);
525 if (ret)
526 goto drop;
527
528 uh = udp_hdr(skb);
529 udp_len = ntohs(uh->len);
530 uh->check = 0;
531 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
532 ip_hdr(skb)->daddr,
533 udp_len, IPPROTO_UDP,
534 csum_partial(uh, udp_len, 0));
535 if (uh->check == 0)
536 uh->check = CSUM_MANGLED_0;
537
538 csum_info->transmit.udp_checksum = 0;
539 }
540 goto do_send;
541
542 do_lso:
543 rndis_msg_size += NDIS_LSO_PPI_SIZE;
544 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
545 TCP_LARGESEND_PKTINFO);
546
547 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
548 ppi->ppi_offset);
549
550 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
551 if (net_trans_info & (INFO_IPV4 << 16)) {
552 lso_info->lso_v2_transmit.ip_version =
553 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
554 ip_hdr(skb)->tot_len = 0;
555 ip_hdr(skb)->check = 0;
556 tcp_hdr(skb)->check =
557 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
558 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
559 } else {
560 lso_info->lso_v2_transmit.ip_version =
561 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
562 ipv6_hdr(skb)->payload_len = 0;
563 tcp_hdr(skb)->check =
564 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
565 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
566 }
567 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
568 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
569
570 do_send:
571 /* Start filling in the page buffers with the rndis hdr */
572 rndis_msg->msg_len += rndis_msg_size;
573 packet->total_data_buflen = rndis_msg->msg_len;
574 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
575 skb, &page_buf[0]);
576
577 ret = netvsc_send(net_device_ctx->device_ctx, packet);
578
579 drop:
580 if (ret == 0) {
581 net->stats.tx_bytes += skb_length;
582 net->stats.tx_packets++;
583 } else {
584 if (!packet->part_of_skb)
585 kfree(packet);
586 if (ret != -EAGAIN) {
587 dev_kfree_skb_any(skb);
588 net->stats.tx_dropped++;
589 }
590 }
591
592 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
593 }
594
595 /*
596 * netvsc_linkstatus_callback - Link up/down notification
597 */
598 void netvsc_linkstatus_callback(struct hv_device *device_obj,
599 struct rndis_message *resp)
600 {
601 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
602 struct net_device *net;
603 struct net_device_context *ndev_ctx;
604 struct netvsc_device *net_device;
605 struct rndis_device *rdev;
606
607 net_device = hv_get_drvdata(device_obj);
608 rdev = net_device->extension;
609
610 switch (indicate->status) {
611 case RNDIS_STATUS_MEDIA_CONNECT:
612 rdev->link_state = false;
613 break;
614 case RNDIS_STATUS_MEDIA_DISCONNECT:
615 rdev->link_state = true;
616 break;
617 case RNDIS_STATUS_NETWORK_CHANGE:
618 rdev->link_change = true;
619 break;
620 default:
621 return;
622 }
623
624 net = net_device->ndev;
625
626 if (!net || net->reg_state != NETREG_REGISTERED)
627 return;
628
629 ndev_ctx = netdev_priv(net);
630 if (!rdev->link_state) {
631 schedule_delayed_work(&ndev_ctx->dwork, 0);
632 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
633 } else {
634 schedule_delayed_work(&ndev_ctx->dwork, 0);
635 }
636 }
637
638 /*
639 * netvsc_recv_callback - Callback when we receive a packet from the
640 * "wire" on the specified device.
641 */
642 int netvsc_recv_callback(struct hv_device *device_obj,
643 struct hv_netvsc_packet *packet,
644 struct ndis_tcp_ip_checksum_info *csum_info)
645 {
646 struct net_device *net;
647 struct sk_buff *skb;
648
649 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
650 if (!net || net->reg_state != NETREG_REGISTERED) {
651 packet->status = NVSP_STAT_FAIL;
652 return 0;
653 }
654
655 /* Allocate a skb - TODO direct I/O to pages? */
656 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
657 if (unlikely(!skb)) {
658 ++net->stats.rx_dropped;
659 packet->status = NVSP_STAT_FAIL;
660 return 0;
661 }
662
663 /*
664 * Copy to skb. This copy is needed here since the memory pointed by
665 * hv_netvsc_packet cannot be deallocated
666 */
667 memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
668 packet->total_data_buflen);
669
670 skb->protocol = eth_type_trans(skb, net);
671 if (csum_info) {
672 /* We only look at the IP checksum here.
673 * Should we be dropping the packet if checksum
674 * failed? How do we deal with other checksums - TCP/UDP?
675 */
676 if (csum_info->receive.ip_checksum_succeeded)
677 skb->ip_summed = CHECKSUM_UNNECESSARY;
678 else
679 skb->ip_summed = CHECKSUM_NONE;
680 }
681
682 if (packet->vlan_tci & VLAN_TAG_PRESENT)
683 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
684 packet->vlan_tci);
685
686 skb_record_rx_queue(skb, packet->channel->
687 offermsg.offer.sub_channel_index);
688
689 net->stats.rx_packets++;
690 net->stats.rx_bytes += packet->total_data_buflen;
691
692 /*
693 * Pass the skb back up. Network stack will deallocate the skb when it
694 * is done.
695 * TODO - use NAPI?
696 */
697 netif_rx(skb);
698
699 return 0;
700 }
701
702 static void netvsc_get_drvinfo(struct net_device *net,
703 struct ethtool_drvinfo *info)
704 {
705 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
706 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
707 }
708
709 static void netvsc_get_channels(struct net_device *net,
710 struct ethtool_channels *channel)
711 {
712 struct net_device_context *net_device_ctx = netdev_priv(net);
713 struct hv_device *dev = net_device_ctx->device_ctx;
714 struct netvsc_device *nvdev = hv_get_drvdata(dev);
715
716 if (nvdev) {
717 channel->max_combined = nvdev->max_chn;
718 channel->combined_count = nvdev->num_chn;
719 }
720 }
721
722 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
723 {
724 struct net_device_context *ndevctx = netdev_priv(ndev);
725 struct hv_device *hdev = ndevctx->device_ctx;
726 struct netvsc_device *nvdev = hv_get_drvdata(hdev);
727 struct netvsc_device_info device_info;
728 int limit = ETH_DATA_LEN;
729
730 if (nvdev == NULL || nvdev->destroy)
731 return -ENODEV;
732
733 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
734 limit = NETVSC_MTU - ETH_HLEN;
735
736 /* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
737 if (mtu < ETH_DATA_LEN || mtu > limit)
738 return -EINVAL;
739
740 nvdev->start_remove = true;
741 cancel_work_sync(&ndevctx->work);
742 netif_tx_disable(ndev);
743 rndis_filter_device_remove(hdev);
744
745 ndev->mtu = mtu;
746
747 ndevctx->device_ctx = hdev;
748 hv_set_drvdata(hdev, ndev);
749 device_info.ring_size = ring_size;
750 rndis_filter_device_add(hdev, &device_info);
751 netif_tx_wake_all_queues(ndev);
752
753 return 0;
754 }
755
756
757 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
758 {
759 struct net_device_context *ndevctx = netdev_priv(ndev);
760 struct hv_device *hdev = ndevctx->device_ctx;
761 struct sockaddr *addr = p;
762 char save_adr[ETH_ALEN];
763 unsigned char save_aatype;
764 int err;
765
766 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
767 save_aatype = ndev->addr_assign_type;
768
769 err = eth_mac_addr(ndev, p);
770 if (err != 0)
771 return err;
772
773 err = rndis_filter_set_device_mac(hdev, addr->sa_data);
774 if (err != 0) {
775 /* roll back to saved MAC */
776 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
777 ndev->addr_assign_type = save_aatype;
778 }
779
780 return err;
781 }
782
783 #ifdef CONFIG_NET_POLL_CONTROLLER
784 static void netvsc_poll_controller(struct net_device *net)
785 {
786 /* As netvsc_start_xmit() works synchronous we don't have to
787 * trigger anything here.
788 */
789 }
790 #endif
791
792 static const struct ethtool_ops ethtool_ops = {
793 .get_drvinfo = netvsc_get_drvinfo,
794 .get_link = ethtool_op_get_link,
795 .get_channels = netvsc_get_channels,
796 };
797
798 static const struct net_device_ops device_ops = {
799 .ndo_open = netvsc_open,
800 .ndo_stop = netvsc_close,
801 .ndo_start_xmit = netvsc_start_xmit,
802 .ndo_set_rx_mode = netvsc_set_multicast_list,
803 .ndo_change_mtu = netvsc_change_mtu,
804 .ndo_validate_addr = eth_validate_addr,
805 .ndo_set_mac_address = netvsc_set_mac_addr,
806 .ndo_select_queue = netvsc_select_queue,
807 #ifdef CONFIG_NET_POLL_CONTROLLER
808 .ndo_poll_controller = netvsc_poll_controller,
809 #endif
810 };
811
812 /*
813 * Send GARP packet to network peers after migrations.
814 * After Quick Migration, the network is not immediately operational in the
815 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
816 * another netif_notify_peers() into a delayed work, otherwise GARP packet
817 * will not be sent after quick migration, and cause network disconnection.
818 * Also, we update the carrier status here.
819 */
820 static void netvsc_link_change(struct work_struct *w)
821 {
822 struct net_device_context *ndev_ctx;
823 struct net_device *net;
824 struct netvsc_device *net_device;
825 struct rndis_device *rdev;
826 bool notify, refresh = false;
827 char *argv[] = { "/etc/init.d/network", "restart", NULL };
828 char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
829
830 rtnl_lock();
831
832 ndev_ctx = container_of(w, struct net_device_context, dwork.work);
833 net_device = hv_get_drvdata(ndev_ctx->device_ctx);
834 rdev = net_device->extension;
835 net = net_device->ndev;
836
837 if (rdev->link_state) {
838 netif_carrier_off(net);
839 notify = false;
840 } else {
841 netif_carrier_on(net);
842 notify = true;
843 if (rdev->link_change) {
844 rdev->link_change = false;
845 refresh = true;
846 }
847 }
848
849 rtnl_unlock();
850
851 if (refresh)
852 call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
853
854 if (notify)
855 netdev_notify_peers(net);
856 }
857
858
859 static int netvsc_probe(struct hv_device *dev,
860 const struct hv_vmbus_device_id *dev_id)
861 {
862 struct net_device *net = NULL;
863 struct net_device_context *net_device_ctx;
864 struct netvsc_device_info device_info;
865 struct netvsc_device *nvdev;
866 int ret;
867 u32 max_needed_headroom;
868
869 net = alloc_etherdev_mq(sizeof(struct net_device_context),
870 num_online_cpus());
871 if (!net)
872 return -ENOMEM;
873
874 max_needed_headroom = sizeof(struct hv_netvsc_packet) +
875 sizeof(struct rndis_message) +
876 NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
877 NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE;
878
879 netif_carrier_off(net);
880
881 net_device_ctx = netdev_priv(net);
882 net_device_ctx->device_ctx = dev;
883 hv_set_drvdata(dev, net);
884 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
885 INIT_WORK(&net_device_ctx->work, do_set_multicast);
886
887 net->netdev_ops = &device_ops;
888
889 net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
890 NETIF_F_TSO;
891 net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
892 NETIF_F_IP_CSUM | NETIF_F_TSO;
893
894 net->ethtool_ops = &ethtool_ops;
895 SET_NETDEV_DEV(net, &dev->device);
896
897 /*
898 * Request additional head room in the skb.
899 * We will use this space to build the rndis
900 * heaser and other state we need to maintain.
901 */
902 net->needed_headroom = max_needed_headroom;
903
904 /* Notify the netvsc driver of the new device */
905 device_info.ring_size = ring_size;
906 ret = rndis_filter_device_add(dev, &device_info);
907 if (ret != 0) {
908 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
909 free_netdev(net);
910 hv_set_drvdata(dev, NULL);
911 return ret;
912 }
913 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
914
915 nvdev = hv_get_drvdata(dev);
916 netif_set_real_num_tx_queues(net, nvdev->num_chn);
917 netif_set_real_num_rx_queues(net, nvdev->num_chn);
918
919 ret = register_netdev(net);
920 if (ret != 0) {
921 pr_err("Unable to register netdev.\n");
922 rndis_filter_device_remove(dev);
923 free_netdev(net);
924 } else {
925 schedule_delayed_work(&net_device_ctx->dwork, 0);
926 }
927
928 return ret;
929 }
930
931 static int netvsc_remove(struct hv_device *dev)
932 {
933 struct net_device *net;
934 struct net_device_context *ndev_ctx;
935 struct netvsc_device *net_device;
936
937 net_device = hv_get_drvdata(dev);
938 net = net_device->ndev;
939
940 if (net == NULL) {
941 dev_err(&dev->device, "No net device to remove\n");
942 return 0;
943 }
944
945 net_device->start_remove = true;
946
947 ndev_ctx = netdev_priv(net);
948 cancel_delayed_work_sync(&ndev_ctx->dwork);
949 cancel_work_sync(&ndev_ctx->work);
950
951 /* Stop outbound asap */
952 netif_tx_disable(net);
953
954 unregister_netdev(net);
955
956 /*
957 * Call to the vsc driver to let it know that the device is being
958 * removed
959 */
960 rndis_filter_device_remove(dev);
961
962 free_netdev(net);
963 return 0;
964 }
965
966 static const struct hv_vmbus_device_id id_table[] = {
967 /* Network guid */
968 { HV_NIC_GUID, },
969 { },
970 };
971
972 MODULE_DEVICE_TABLE(vmbus, id_table);
973
974 /* The one and only one */
975 static struct hv_driver netvsc_drv = {
976 .name = KBUILD_MODNAME,
977 .id_table = id_table,
978 .probe = netvsc_probe,
979 .remove = netvsc_remove,
980 };
981
982 static void __exit netvsc_drv_exit(void)
983 {
984 vmbus_driver_unregister(&netvsc_drv);
985 }
986
987 static int __init netvsc_drv_init(void)
988 {
989 if (ring_size < RING_SIZE_MIN) {
990 ring_size = RING_SIZE_MIN;
991 pr_info("Increased ring_size to %d (min allowed)\n",
992 ring_size);
993 }
994 return vmbus_driver_register(&netvsc_drv);
995 }
996
997 MODULE_LICENSE("GPL");
998 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
999
1000 module_init(netvsc_drv_init);
1001 module_exit(netvsc_drv_exit);
Linux with added FPC-III support