xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision 206204a1)
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, u8 *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 = 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 iphdr *iphdr;
194 	int data_len;
195 	bool ret = false;
196 
197 	if (eth_hdr(skb)->h_proto != htons(ETH_P_IP))
198 		return false;
199 
200 	iphdr = ip_hdr(skb);
201 
202 	if (iphdr->version == 4) {
203 		if (iphdr->protocol == IPPROTO_TCP)
204 			data_len = 12;
205 		else
206 			data_len = 8;
207 		*hash = comp_hash(netvsc_hash_key, HASH_KEYLEN,
208 				  (u8 *)&iphdr->saddr, data_len);
209 		ret = true;
210 	}
211 
212 	return ret;
213 }
214 
215 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
216 			void *accel_priv, select_queue_fallback_t fallback)
217 {
218 	struct net_device_context *net_device_ctx = netdev_priv(ndev);
219 	struct hv_device *hdev =  net_device_ctx->device_ctx;
220 	struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
221 	u32 hash;
222 	u16 q_idx = 0;
223 
224 	if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
225 		return 0;
226 
227 	if (netvsc_set_hash(&hash, skb)) {
228 		q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
229 			ndev->real_num_tx_queues;
230 		skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
231 	}
232 
233 	return q_idx;
234 }
235 
236 static void netvsc_xmit_completion(void *context)
237 {
238 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
239 	struct sk_buff *skb = (struct sk_buff *)
240 		(unsigned long)packet->send_completion_tid;
241 	u32 index = packet->send_buf_index;
242 
243 	kfree(packet);
244 
245 	if (skb && (index == NETVSC_INVALID_INDEX))
246 		dev_kfree_skb_any(skb);
247 }
248 
249 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
250 			struct hv_page_buffer *pb)
251 {
252 	int j = 0;
253 
254 	/* Deal with compund pages by ignoring unused part
255 	 * of the page.
256 	 */
257 	page += (offset >> PAGE_SHIFT);
258 	offset &= ~PAGE_MASK;
259 
260 	while (len > 0) {
261 		unsigned long bytes;
262 
263 		bytes = PAGE_SIZE - offset;
264 		if (bytes > len)
265 			bytes = len;
266 		pb[j].pfn = page_to_pfn(page);
267 		pb[j].offset = offset;
268 		pb[j].len = bytes;
269 
270 		offset += bytes;
271 		len -= bytes;
272 
273 		if (offset == PAGE_SIZE && len) {
274 			page++;
275 			offset = 0;
276 			j++;
277 		}
278 	}
279 
280 	return j + 1;
281 }
282 
283 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
284 			   struct hv_page_buffer *pb)
285 {
286 	u32 slots_used = 0;
287 	char *data = skb->data;
288 	int frags = skb_shinfo(skb)->nr_frags;
289 	int i;
290 
291 	/* The packet is laid out thus:
292 	 * 1. hdr
293 	 * 2. skb linear data
294 	 * 3. skb fragment data
295 	 */
296 	if (hdr != NULL)
297 		slots_used += fill_pg_buf(virt_to_page(hdr),
298 					offset_in_page(hdr),
299 					len, &pb[slots_used]);
300 
301 	slots_used += fill_pg_buf(virt_to_page(data),
302 				offset_in_page(data),
303 				skb_headlen(skb), &pb[slots_used]);
304 
305 	for (i = 0; i < frags; i++) {
306 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
307 
308 		slots_used += fill_pg_buf(skb_frag_page(frag),
309 					frag->page_offset,
310 					skb_frag_size(frag), &pb[slots_used]);
311 	}
312 	return slots_used;
313 }
314 
315 static int count_skb_frag_slots(struct sk_buff *skb)
316 {
317 	int i, frags = skb_shinfo(skb)->nr_frags;
318 	int pages = 0;
319 
320 	for (i = 0; i < frags; i++) {
321 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
322 		unsigned long size = skb_frag_size(frag);
323 		unsigned long offset = frag->page_offset;
324 
325 		/* Skip unused frames from start of page */
326 		offset &= ~PAGE_MASK;
327 		pages += PFN_UP(offset + size);
328 	}
329 	return pages;
330 }
331 
332 static int netvsc_get_slots(struct sk_buff *skb)
333 {
334 	char *data = skb->data;
335 	unsigned int offset = offset_in_page(data);
336 	unsigned int len = skb_headlen(skb);
337 	int slots;
338 	int frag_slots;
339 
340 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
341 	frag_slots = count_skb_frag_slots(skb);
342 	return slots + frag_slots;
343 }
344 
345 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
346 {
347 	u32 ret_val = TRANSPORT_INFO_NOT_IP;
348 
349 	if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
350 		(eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
351 		goto not_ip;
352 	}
353 
354 	*trans_off = skb_transport_offset(skb);
355 
356 	if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
357 		struct iphdr *iphdr = ip_hdr(skb);
358 
359 		if (iphdr->protocol == IPPROTO_TCP)
360 			ret_val = TRANSPORT_INFO_IPV4_TCP;
361 		else if (iphdr->protocol == IPPROTO_UDP)
362 			ret_val = TRANSPORT_INFO_IPV4_UDP;
363 	} else {
364 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
365 			ret_val = TRANSPORT_INFO_IPV6_TCP;
366 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
367 			ret_val = TRANSPORT_INFO_IPV6_UDP;
368 	}
369 
370 not_ip:
371 	return ret_val;
372 }
373 
374 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
375 {
376 	struct net_device_context *net_device_ctx = netdev_priv(net);
377 	struct hv_netvsc_packet *packet;
378 	int ret;
379 	unsigned int num_data_pgs;
380 	struct rndis_message *rndis_msg;
381 	struct rndis_packet *rndis_pkt;
382 	u32 rndis_msg_size;
383 	bool isvlan;
384 	struct rndis_per_packet_info *ppi;
385 	struct ndis_tcp_ip_checksum_info *csum_info;
386 	struct ndis_tcp_lso_info *lso_info;
387 	int  hdr_offset;
388 	u32 net_trans_info;
389 	u32 hash;
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 	/* Allocate a netvsc packet based on # of frags. */
405 	packet = kzalloc(sizeof(struct hv_netvsc_packet) +
406 			 (num_data_pgs * sizeof(struct hv_page_buffer)) +
407 			 sizeof(struct rndis_message) +
408 			 NDIS_VLAN_PPI_SIZE + NDIS_CSUM_PPI_SIZE +
409 			 NDIS_LSO_PPI_SIZE + NDIS_HASH_PPI_SIZE, GFP_ATOMIC);
410 	if (!packet) {
411 		/* out of memory, drop packet */
412 		netdev_err(net, "unable to allocate hv_netvsc_packet\n");
413 
414 		dev_kfree_skb(skb);
415 		net->stats.tx_dropped++;
416 		return NETDEV_TX_OK;
417 	}
418 
419 	packet->vlan_tci = skb->vlan_tci;
420 
421 	packet->q_idx = skb_get_queue_mapping(skb);
422 
423 	packet->is_data_pkt = true;
424 	packet->total_data_buflen = skb->len;
425 
426 	packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
427 				sizeof(struct hv_netvsc_packet) +
428 				(num_data_pgs * sizeof(struct hv_page_buffer)));
429 
430 	/* Set the completion routine */
431 	packet->send_completion = netvsc_xmit_completion;
432 	packet->send_completion_ctx = packet;
433 	packet->send_completion_tid = (unsigned long)skb;
434 
435 	isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
436 
437 	/* Add the rndis header */
438 	rndis_msg = packet->rndis_msg;
439 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
440 	rndis_msg->msg_len = packet->total_data_buflen;
441 	rndis_pkt = &rndis_msg->msg.pkt;
442 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
443 	rndis_pkt->data_len = packet->total_data_buflen;
444 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
445 
446 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
447 
448 	hash = skb_get_hash_raw(skb);
449 	if (hash != 0 && net->real_num_tx_queues > 1) {
450 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
451 		ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
452 				    NBL_HASH_VALUE);
453 		*(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
454 	}
455 
456 	if (isvlan) {
457 		struct ndis_pkt_8021q_info *vlan;
458 
459 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
460 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
461 					IEEE_8021Q_INFO);
462 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
463 						ppi->ppi_offset);
464 		vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
465 		vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
466 				VLAN_PRIO_SHIFT;
467 	}
468 
469 	net_trans_info = get_net_transport_info(skb, &hdr_offset);
470 	if (net_trans_info == TRANSPORT_INFO_NOT_IP)
471 		goto do_send;
472 
473 	/*
474 	 * Setup the sendside checksum offload only if this is not a
475 	 * GSO packet.
476 	 */
477 	if (skb_is_gso(skb))
478 		goto do_lso;
479 
480 	if ((skb->ip_summed == CHECKSUM_NONE) ||
481 	    (skb->ip_summed == CHECKSUM_UNNECESSARY))
482 		goto do_send;
483 
484 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
485 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
486 			    TCPIP_CHKSUM_PKTINFO);
487 
488 	csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
489 			ppi->ppi_offset);
490 
491 	if (net_trans_info & (INFO_IPV4 << 16))
492 		csum_info->transmit.is_ipv4 = 1;
493 	else
494 		csum_info->transmit.is_ipv6 = 1;
495 
496 	if (net_trans_info & INFO_TCP) {
497 		csum_info->transmit.tcp_checksum = 1;
498 		csum_info->transmit.tcp_header_offset = hdr_offset;
499 	} else if (net_trans_info & INFO_UDP) {
500 		/* UDP checksum offload is not supported on ws2008r2.
501 		 * Furthermore, on ws2012 and ws2012r2, there are some
502 		 * issues with udp checksum offload from Linux guests.
503 		 * (these are host issues).
504 		 * For now compute the checksum here.
505 		 */
506 		struct udphdr *uh;
507 		u16 udp_len;
508 
509 		ret = skb_cow_head(skb, 0);
510 		if (ret)
511 			goto drop;
512 
513 		uh = udp_hdr(skb);
514 		udp_len = ntohs(uh->len);
515 		uh->check = 0;
516 		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
517 					      ip_hdr(skb)->daddr,
518 					      udp_len, IPPROTO_UDP,
519 					      csum_partial(uh, udp_len, 0));
520 		if (uh->check == 0)
521 			uh->check = CSUM_MANGLED_0;
522 
523 		csum_info->transmit.udp_checksum = 0;
524 	}
525 	goto do_send;
526 
527 do_lso:
528 	rndis_msg_size += NDIS_LSO_PPI_SIZE;
529 	ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
530 			    TCP_LARGESEND_PKTINFO);
531 
532 	lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
533 			ppi->ppi_offset);
534 
535 	lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
536 	if (net_trans_info & (INFO_IPV4 << 16)) {
537 		lso_info->lso_v2_transmit.ip_version =
538 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
539 		ip_hdr(skb)->tot_len = 0;
540 		ip_hdr(skb)->check = 0;
541 		tcp_hdr(skb)->check =
542 		~csum_tcpudp_magic(ip_hdr(skb)->saddr,
543 				   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
544 	} else {
545 		lso_info->lso_v2_transmit.ip_version =
546 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
547 		ipv6_hdr(skb)->payload_len = 0;
548 		tcp_hdr(skb)->check =
549 		~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
550 				&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
551 	}
552 	lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
553 	lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
554 
555 do_send:
556 	/* Start filling in the page buffers with the rndis hdr */
557 	rndis_msg->msg_len += rndis_msg_size;
558 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
559 					skb, &packet->page_buf[0]);
560 
561 	ret = netvsc_send(net_device_ctx->device_ctx, packet);
562 
563 drop:
564 	if (ret == 0) {
565 		net->stats.tx_bytes += skb->len;
566 		net->stats.tx_packets++;
567 	} else {
568 		kfree(packet);
569 		if (ret != -EAGAIN) {
570 			dev_kfree_skb_any(skb);
571 			net->stats.tx_dropped++;
572 		}
573 	}
574 
575 	return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
576 }
577 
578 /*
579  * netvsc_linkstatus_callback - Link up/down notification
580  */
581 void netvsc_linkstatus_callback(struct hv_device *device_obj,
582 				       unsigned int status)
583 {
584 	struct net_device *net;
585 	struct net_device_context *ndev_ctx;
586 	struct netvsc_device *net_device;
587 	struct rndis_device *rdev;
588 
589 	net_device = hv_get_drvdata(device_obj);
590 	rdev = net_device->extension;
591 
592 	rdev->link_state = status != 1;
593 
594 	net = net_device->ndev;
595 
596 	if (!net || net->reg_state != NETREG_REGISTERED)
597 		return;
598 
599 	ndev_ctx = netdev_priv(net);
600 	if (status == 1) {
601 		schedule_delayed_work(&ndev_ctx->dwork, 0);
602 		schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
603 	} else {
604 		schedule_delayed_work(&ndev_ctx->dwork, 0);
605 	}
606 }
607 
608 /*
609  * netvsc_recv_callback -  Callback when we receive a packet from the
610  * "wire" on the specified device.
611  */
612 int netvsc_recv_callback(struct hv_device *device_obj,
613 				struct hv_netvsc_packet *packet,
614 				struct ndis_tcp_ip_checksum_info *csum_info)
615 {
616 	struct net_device *net;
617 	struct sk_buff *skb;
618 
619 	net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
620 	if (!net || net->reg_state != NETREG_REGISTERED) {
621 		packet->status = NVSP_STAT_FAIL;
622 		return 0;
623 	}
624 
625 	/* Allocate a skb - TODO direct I/O to pages? */
626 	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
627 	if (unlikely(!skb)) {
628 		++net->stats.rx_dropped;
629 		packet->status = NVSP_STAT_FAIL;
630 		return 0;
631 	}
632 
633 	/*
634 	 * Copy to skb. This copy is needed here since the memory pointed by
635 	 * hv_netvsc_packet cannot be deallocated
636 	 */
637 	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
638 		packet->total_data_buflen);
639 
640 	skb->protocol = eth_type_trans(skb, net);
641 	if (csum_info) {
642 		/* We only look at the IP checksum here.
643 		 * Should we be dropping the packet if checksum
644 		 * failed? How do we deal with other checksums - TCP/UDP?
645 		 */
646 		if (csum_info->receive.ip_checksum_succeeded)
647 			skb->ip_summed = CHECKSUM_UNNECESSARY;
648 		else
649 			skb->ip_summed = CHECKSUM_NONE;
650 	}
651 
652 	if (packet->vlan_tci & VLAN_TAG_PRESENT)
653 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
654 				       packet->vlan_tci);
655 
656 	skb_record_rx_queue(skb, packet->channel->
657 			    offermsg.offer.sub_channel_index);
658 
659 	net->stats.rx_packets++;
660 	net->stats.rx_bytes += packet->total_data_buflen;
661 
662 	/*
663 	 * Pass the skb back up. Network stack will deallocate the skb when it
664 	 * is done.
665 	 * TODO - use NAPI?
666 	 */
667 	netif_rx(skb);
668 
669 	return 0;
670 }
671 
672 static void netvsc_get_drvinfo(struct net_device *net,
673 			       struct ethtool_drvinfo *info)
674 {
675 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
676 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
677 }
678 
679 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
680 {
681 	struct net_device_context *ndevctx = netdev_priv(ndev);
682 	struct hv_device *hdev =  ndevctx->device_ctx;
683 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
684 	struct netvsc_device_info device_info;
685 	int limit = ETH_DATA_LEN;
686 
687 	if (nvdev == NULL || nvdev->destroy)
688 		return -ENODEV;
689 
690 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
691 		limit = NETVSC_MTU;
692 
693 	if (mtu < 68 || mtu > limit)
694 		return -EINVAL;
695 
696 	nvdev->start_remove = true;
697 	cancel_work_sync(&ndevctx->work);
698 	netif_tx_disable(ndev);
699 	rndis_filter_device_remove(hdev);
700 
701 	ndev->mtu = mtu;
702 
703 	ndevctx->device_ctx = hdev;
704 	hv_set_drvdata(hdev, ndev);
705 	device_info.ring_size = ring_size;
706 	rndis_filter_device_add(hdev, &device_info);
707 	netif_tx_wake_all_queues(ndev);
708 
709 	return 0;
710 }
711 
712 
713 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
714 {
715 	struct net_device_context *ndevctx = netdev_priv(ndev);
716 	struct hv_device *hdev =  ndevctx->device_ctx;
717 	struct sockaddr *addr = p;
718 	char save_adr[ETH_ALEN];
719 	unsigned char save_aatype;
720 	int err;
721 
722 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
723 	save_aatype = ndev->addr_assign_type;
724 
725 	err = eth_mac_addr(ndev, p);
726 	if (err != 0)
727 		return err;
728 
729 	err = rndis_filter_set_device_mac(hdev, addr->sa_data);
730 	if (err != 0) {
731 		/* roll back to saved MAC */
732 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
733 		ndev->addr_assign_type = save_aatype;
734 	}
735 
736 	return err;
737 }
738 
739 
740 static const struct ethtool_ops ethtool_ops = {
741 	.get_drvinfo	= netvsc_get_drvinfo,
742 	.get_link	= ethtool_op_get_link,
743 };
744 
745 static const struct net_device_ops device_ops = {
746 	.ndo_open =			netvsc_open,
747 	.ndo_stop =			netvsc_close,
748 	.ndo_start_xmit =		netvsc_start_xmit,
749 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
750 	.ndo_change_mtu =		netvsc_change_mtu,
751 	.ndo_validate_addr =		eth_validate_addr,
752 	.ndo_set_mac_address =		netvsc_set_mac_addr,
753 	.ndo_select_queue =		netvsc_select_queue,
754 };
755 
756 /*
757  * Send GARP packet to network peers after migrations.
758  * After Quick Migration, the network is not immediately operational in the
759  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
760  * another netif_notify_peers() into a delayed work, otherwise GARP packet
761  * will not be sent after quick migration, and cause network disconnection.
762  * Also, we update the carrier status here.
763  */
764 static void netvsc_link_change(struct work_struct *w)
765 {
766 	struct net_device_context *ndev_ctx;
767 	struct net_device *net;
768 	struct netvsc_device *net_device;
769 	struct rndis_device *rdev;
770 	bool notify;
771 
772 	rtnl_lock();
773 
774 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
775 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
776 	rdev = net_device->extension;
777 	net = net_device->ndev;
778 
779 	if (rdev->link_state) {
780 		netif_carrier_off(net);
781 		notify = false;
782 	} else {
783 		netif_carrier_on(net);
784 		notify = true;
785 	}
786 
787 	rtnl_unlock();
788 
789 	if (notify)
790 		netdev_notify_peers(net);
791 }
792 
793 
794 static int netvsc_probe(struct hv_device *dev,
795 			const struct hv_vmbus_device_id *dev_id)
796 {
797 	struct net_device *net = NULL;
798 	struct net_device_context *net_device_ctx;
799 	struct netvsc_device_info device_info;
800 	struct netvsc_device *nvdev;
801 	int ret;
802 
803 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
804 				num_online_cpus());
805 	if (!net)
806 		return -ENOMEM;
807 
808 	netif_carrier_off(net);
809 
810 	net_device_ctx = netdev_priv(net);
811 	net_device_ctx->device_ctx = dev;
812 	hv_set_drvdata(dev, net);
813 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
814 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
815 
816 	net->netdev_ops = &device_ops;
817 
818 	net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
819 				NETIF_F_TSO;
820 	net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
821 			NETIF_F_IP_CSUM | NETIF_F_TSO;
822 
823 	net->ethtool_ops = &ethtool_ops;
824 	SET_NETDEV_DEV(net, &dev->device);
825 
826 	/* Notify the netvsc driver of the new device */
827 	device_info.ring_size = ring_size;
828 	ret = rndis_filter_device_add(dev, &device_info);
829 	if (ret != 0) {
830 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
831 		free_netdev(net);
832 		hv_set_drvdata(dev, NULL);
833 		return ret;
834 	}
835 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
836 
837 	nvdev = hv_get_drvdata(dev);
838 	netif_set_real_num_tx_queues(net, nvdev->num_chn);
839 	netif_set_real_num_rx_queues(net, nvdev->num_chn);
840 
841 	ret = register_netdev(net);
842 	if (ret != 0) {
843 		pr_err("Unable to register netdev.\n");
844 		rndis_filter_device_remove(dev);
845 		free_netdev(net);
846 	} else {
847 		schedule_delayed_work(&net_device_ctx->dwork, 0);
848 	}
849 
850 	return ret;
851 }
852 
853 static int netvsc_remove(struct hv_device *dev)
854 {
855 	struct net_device *net;
856 	struct net_device_context *ndev_ctx;
857 	struct netvsc_device *net_device;
858 
859 	net_device = hv_get_drvdata(dev);
860 	net = net_device->ndev;
861 
862 	if (net == NULL) {
863 		dev_err(&dev->device, "No net device to remove\n");
864 		return 0;
865 	}
866 
867 	net_device->start_remove = true;
868 
869 	ndev_ctx = netdev_priv(net);
870 	cancel_delayed_work_sync(&ndev_ctx->dwork);
871 	cancel_work_sync(&ndev_ctx->work);
872 
873 	/* Stop outbound asap */
874 	netif_tx_disable(net);
875 
876 	unregister_netdev(net);
877 
878 	/*
879 	 * Call to the vsc driver to let it know that the device is being
880 	 * removed
881 	 */
882 	rndis_filter_device_remove(dev);
883 
884 	free_netdev(net);
885 	return 0;
886 }
887 
888 static const struct hv_vmbus_device_id id_table[] = {
889 	/* Network guid */
890 	{ HV_NIC_GUID, },
891 	{ },
892 };
893 
894 MODULE_DEVICE_TABLE(vmbus, id_table);
895 
896 /* The one and only one */
897 static struct  hv_driver netvsc_drv = {
898 	.name = KBUILD_MODNAME,
899 	.id_table = id_table,
900 	.probe = netvsc_probe,
901 	.remove = netvsc_remove,
902 };
903 
904 static void __exit netvsc_drv_exit(void)
905 {
906 	vmbus_driver_unregister(&netvsc_drv);
907 }
908 
909 static int __init netvsc_drv_init(void)
910 {
911 	if (ring_size < RING_SIZE_MIN) {
912 		ring_size = RING_SIZE_MIN;
913 		pr_info("Increased ring_size to %d (min allowed)\n",
914 			ring_size);
915 	}
916 	return vmbus_driver_register(&netvsc_drv);
917 }
918 
919 MODULE_LICENSE("GPL");
920 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
921 
922 module_init(netvsc_drv_init);
923 module_exit(netvsc_drv_exit);
924