xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision afb46f79)
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_start_queue(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 static void netvsc_xmit_completion(void *context)
153 {
154 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
155 	struct sk_buff *skb = (struct sk_buff *)
156 		(unsigned long)packet->completion.send.send_completion_tid;
157 
158 	kfree(packet);
159 
160 	if (skb)
161 		dev_kfree_skb_any(skb);
162 }
163 
164 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
165 			struct hv_page_buffer *pb)
166 {
167 	int j = 0;
168 
169 	/* Deal with compund pages by ignoring unused part
170 	 * of the page.
171 	 */
172 	page += (offset >> PAGE_SHIFT);
173 	offset &= ~PAGE_MASK;
174 
175 	while (len > 0) {
176 		unsigned long bytes;
177 
178 		bytes = PAGE_SIZE - offset;
179 		if (bytes > len)
180 			bytes = len;
181 		pb[j].pfn = page_to_pfn(page);
182 		pb[j].offset = offset;
183 		pb[j].len = bytes;
184 
185 		offset += bytes;
186 		len -= bytes;
187 
188 		if (offset == PAGE_SIZE && len) {
189 			page++;
190 			offset = 0;
191 			j++;
192 		}
193 	}
194 
195 	return j + 1;
196 }
197 
198 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
199 			   struct hv_page_buffer *pb)
200 {
201 	u32 slots_used = 0;
202 	char *data = skb->data;
203 	int frags = skb_shinfo(skb)->nr_frags;
204 	int i;
205 
206 	/* The packet is laid out thus:
207 	 * 1. hdr
208 	 * 2. skb linear data
209 	 * 3. skb fragment data
210 	 */
211 	if (hdr != NULL)
212 		slots_used += fill_pg_buf(virt_to_page(hdr),
213 					offset_in_page(hdr),
214 					len, &pb[slots_used]);
215 
216 	slots_used += fill_pg_buf(virt_to_page(data),
217 				offset_in_page(data),
218 				skb_headlen(skb), &pb[slots_used]);
219 
220 	for (i = 0; i < frags; i++) {
221 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
222 
223 		slots_used += fill_pg_buf(skb_frag_page(frag),
224 					frag->page_offset,
225 					skb_frag_size(frag), &pb[slots_used]);
226 	}
227 	return slots_used;
228 }
229 
230 static int count_skb_frag_slots(struct sk_buff *skb)
231 {
232 	int i, frags = skb_shinfo(skb)->nr_frags;
233 	int pages = 0;
234 
235 	for (i = 0; i < frags; i++) {
236 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
237 		unsigned long size = skb_frag_size(frag);
238 		unsigned long offset = frag->page_offset;
239 
240 		/* Skip unused frames from start of page */
241 		offset &= ~PAGE_MASK;
242 		pages += PFN_UP(offset + size);
243 	}
244 	return pages;
245 }
246 
247 static int netvsc_get_slots(struct sk_buff *skb)
248 {
249 	char *data = skb->data;
250 	unsigned int offset = offset_in_page(data);
251 	unsigned int len = skb_headlen(skb);
252 	int slots;
253 	int frag_slots;
254 
255 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
256 	frag_slots = count_skb_frag_slots(skb);
257 	return slots + frag_slots;
258 }
259 
260 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
261 {
262 	u32 ret_val = TRANSPORT_INFO_NOT_IP;
263 
264 	if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
265 		(eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
266 		goto not_ip;
267 	}
268 
269 	*trans_off = skb_transport_offset(skb);
270 
271 	if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
272 		struct iphdr *iphdr = ip_hdr(skb);
273 
274 		if (iphdr->protocol == IPPROTO_TCP)
275 			ret_val = TRANSPORT_INFO_IPV4_TCP;
276 		else if (iphdr->protocol == IPPROTO_UDP)
277 			ret_val = TRANSPORT_INFO_IPV4_UDP;
278 	} else {
279 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
280 			ret_val = TRANSPORT_INFO_IPV6_TCP;
281 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
282 			ret_val = TRANSPORT_INFO_IPV6_UDP;
283 	}
284 
285 not_ip:
286 	return ret_val;
287 }
288 
289 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
290 {
291 	struct net_device_context *net_device_ctx = netdev_priv(net);
292 	struct hv_netvsc_packet *packet;
293 	int ret;
294 	unsigned int num_data_pgs;
295 	struct rndis_message *rndis_msg;
296 	struct rndis_packet *rndis_pkt;
297 	u32 rndis_msg_size;
298 	bool isvlan;
299 	struct rndis_per_packet_info *ppi;
300 	struct ndis_tcp_ip_checksum_info *csum_info;
301 	struct ndis_tcp_lso_info *lso_info;
302 	int  hdr_offset;
303 	u32 net_trans_info;
304 
305 
306 	/* We will atmost need two pages to describe the rndis
307 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
308 	 * of pages in a single packet.
309 	 */
310 	num_data_pgs = netvsc_get_slots(skb) + 2;
311 	if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
312 		netdev_err(net, "Packet too big: %u\n", skb->len);
313 		dev_kfree_skb(skb);
314 		net->stats.tx_dropped++;
315 		return NETDEV_TX_OK;
316 	}
317 
318 	/* Allocate a netvsc packet based on # of frags. */
319 	packet = kzalloc(sizeof(struct hv_netvsc_packet) +
320 			 (num_data_pgs * sizeof(struct hv_page_buffer)) +
321 			 sizeof(struct rndis_message) +
322 			 NDIS_VLAN_PPI_SIZE +
323 			 NDIS_CSUM_PPI_SIZE +
324 			 NDIS_LSO_PPI_SIZE, GFP_ATOMIC);
325 	if (!packet) {
326 		/* out of memory, drop packet */
327 		netdev_err(net, "unable to allocate hv_netvsc_packet\n");
328 
329 		dev_kfree_skb(skb);
330 		net->stats.tx_dropped++;
331 		return NETDEV_TX_OK;
332 	}
333 
334 	packet->vlan_tci = skb->vlan_tci;
335 
336 	packet->is_data_pkt = true;
337 	packet->total_data_buflen = skb->len;
338 
339 	packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
340 				sizeof(struct hv_netvsc_packet) +
341 				(num_data_pgs * sizeof(struct hv_page_buffer)));
342 
343 	/* Set the completion routine */
344 	packet->completion.send.send_completion = netvsc_xmit_completion;
345 	packet->completion.send.send_completion_ctx = packet;
346 	packet->completion.send.send_completion_tid = (unsigned long)skb;
347 
348 	isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
349 
350 	/* Add the rndis header */
351 	rndis_msg = packet->rndis_msg;
352 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
353 	rndis_msg->msg_len = packet->total_data_buflen;
354 	rndis_pkt = &rndis_msg->msg.pkt;
355 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
356 	rndis_pkt->data_len = packet->total_data_buflen;
357 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
358 
359 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
360 
361 	if (isvlan) {
362 		struct ndis_pkt_8021q_info *vlan;
363 
364 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
365 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
366 					IEEE_8021Q_INFO);
367 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
368 						ppi->ppi_offset);
369 		vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
370 		vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
371 				VLAN_PRIO_SHIFT;
372 	}
373 
374 	net_trans_info = get_net_transport_info(skb, &hdr_offset);
375 	if (net_trans_info == TRANSPORT_INFO_NOT_IP)
376 		goto do_send;
377 
378 	/*
379 	 * Setup the sendside checksum offload only if this is not a
380 	 * GSO packet.
381 	 */
382 	if (skb_is_gso(skb))
383 		goto do_lso;
384 
385 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
386 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
387 			    TCPIP_CHKSUM_PKTINFO);
388 
389 	csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
390 			ppi->ppi_offset);
391 
392 	if (net_trans_info & (INFO_IPV4 << 16))
393 		csum_info->transmit.is_ipv4 = 1;
394 	else
395 		csum_info->transmit.is_ipv6 = 1;
396 
397 	if (net_trans_info & INFO_TCP) {
398 		csum_info->transmit.tcp_checksum = 1;
399 		csum_info->transmit.tcp_header_offset = hdr_offset;
400 	} else if (net_trans_info & INFO_UDP) {
401 		/* UDP checksum offload is not supported on ws2008r2.
402 		 * Furthermore, on ws2012 and ws2012r2, there are some
403 		 * issues with udp checksum offload from Linux guests.
404 		 * (these are host issues).
405 		 * For now compute the checksum here.
406 		 */
407 		struct udphdr *uh;
408 		u16 udp_len;
409 
410 		ret = skb_cow_head(skb, 0);
411 		if (ret)
412 			goto drop;
413 
414 		uh = udp_hdr(skb);
415 		udp_len = ntohs(uh->len);
416 		uh->check = 0;
417 		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
418 					      ip_hdr(skb)->daddr,
419 					      udp_len, IPPROTO_UDP,
420 					      csum_partial(uh, udp_len, 0));
421 		if (uh->check == 0)
422 			uh->check = CSUM_MANGLED_0;
423 
424 		csum_info->transmit.udp_checksum = 0;
425 	}
426 	goto do_send;
427 
428 do_lso:
429 	rndis_msg_size += NDIS_LSO_PPI_SIZE;
430 	ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
431 			    TCP_LARGESEND_PKTINFO);
432 
433 	lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
434 			ppi->ppi_offset);
435 
436 	lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
437 	if (net_trans_info & (INFO_IPV4 << 16)) {
438 		lso_info->lso_v2_transmit.ip_version =
439 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
440 		ip_hdr(skb)->tot_len = 0;
441 		ip_hdr(skb)->check = 0;
442 		tcp_hdr(skb)->check =
443 		~csum_tcpudp_magic(ip_hdr(skb)->saddr,
444 				   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
445 	} else {
446 		lso_info->lso_v2_transmit.ip_version =
447 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
448 		ipv6_hdr(skb)->payload_len = 0;
449 		tcp_hdr(skb)->check =
450 		~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
451 				&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
452 	}
453 	lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
454 	lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
455 
456 do_send:
457 	/* Start filling in the page buffers with the rndis hdr */
458 	rndis_msg->msg_len += rndis_msg_size;
459 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
460 					skb, &packet->page_buf[0]);
461 
462 	ret = netvsc_send(net_device_ctx->device_ctx, packet);
463 
464 drop:
465 	if (ret == 0) {
466 		net->stats.tx_bytes += skb->len;
467 		net->stats.tx_packets++;
468 	} else {
469 		kfree(packet);
470 		if (ret != -EAGAIN) {
471 			dev_kfree_skb_any(skb);
472 			net->stats.tx_dropped++;
473 		}
474 	}
475 
476 	return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
477 }
478 
479 /*
480  * netvsc_linkstatus_callback - Link up/down notification
481  */
482 void netvsc_linkstatus_callback(struct hv_device *device_obj,
483 				       unsigned int status)
484 {
485 	struct net_device *net;
486 	struct net_device_context *ndev_ctx;
487 	struct netvsc_device *net_device;
488 	struct rndis_device *rdev;
489 
490 	net_device = hv_get_drvdata(device_obj);
491 	rdev = net_device->extension;
492 
493 	rdev->link_state = status != 1;
494 
495 	net = net_device->ndev;
496 
497 	if (!net || net->reg_state != NETREG_REGISTERED)
498 		return;
499 
500 	ndev_ctx = netdev_priv(net);
501 	if (status == 1) {
502 		schedule_delayed_work(&ndev_ctx->dwork, 0);
503 		schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
504 	} else {
505 		schedule_delayed_work(&ndev_ctx->dwork, 0);
506 	}
507 }
508 
509 /*
510  * netvsc_recv_callback -  Callback when we receive a packet from the
511  * "wire" on the specified device.
512  */
513 int netvsc_recv_callback(struct hv_device *device_obj,
514 				struct hv_netvsc_packet *packet,
515 				struct ndis_tcp_ip_checksum_info *csum_info)
516 {
517 	struct net_device *net;
518 	struct sk_buff *skb;
519 
520 	net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
521 	if (!net || net->reg_state != NETREG_REGISTERED) {
522 		packet->status = NVSP_STAT_FAIL;
523 		return 0;
524 	}
525 
526 	/* Allocate a skb - TODO direct I/O to pages? */
527 	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
528 	if (unlikely(!skb)) {
529 		++net->stats.rx_dropped;
530 		packet->status = NVSP_STAT_FAIL;
531 		return 0;
532 	}
533 
534 	/*
535 	 * Copy to skb. This copy is needed here since the memory pointed by
536 	 * hv_netvsc_packet cannot be deallocated
537 	 */
538 	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
539 		packet->total_data_buflen);
540 
541 	skb->protocol = eth_type_trans(skb, net);
542 	if (csum_info) {
543 		/* We only look at the IP checksum here.
544 		 * Should we be dropping the packet if checksum
545 		 * failed? How do we deal with other checksums - TCP/UDP?
546 		 */
547 		if (csum_info->receive.ip_checksum_succeeded)
548 			skb->ip_summed = CHECKSUM_UNNECESSARY;
549 		else
550 			skb->ip_summed = CHECKSUM_NONE;
551 	}
552 
553 	if (packet->vlan_tci & VLAN_TAG_PRESENT)
554 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
555 				       packet->vlan_tci);
556 
557 	net->stats.rx_packets++;
558 	net->stats.rx_bytes += packet->total_data_buflen;
559 
560 	/*
561 	 * Pass the skb back up. Network stack will deallocate the skb when it
562 	 * is done.
563 	 * TODO - use NAPI?
564 	 */
565 	netif_rx(skb);
566 
567 	return 0;
568 }
569 
570 static void netvsc_get_drvinfo(struct net_device *net,
571 			       struct ethtool_drvinfo *info)
572 {
573 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
574 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
575 }
576 
577 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
578 {
579 	struct net_device_context *ndevctx = netdev_priv(ndev);
580 	struct hv_device *hdev =  ndevctx->device_ctx;
581 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
582 	struct netvsc_device_info device_info;
583 	int limit = ETH_DATA_LEN;
584 
585 	if (nvdev == NULL || nvdev->destroy)
586 		return -ENODEV;
587 
588 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
589 		limit = NETVSC_MTU;
590 
591 	if (mtu < 68 || mtu > limit)
592 		return -EINVAL;
593 
594 	nvdev->start_remove = true;
595 	cancel_work_sync(&ndevctx->work);
596 	netif_tx_disable(ndev);
597 	rndis_filter_device_remove(hdev);
598 
599 	ndev->mtu = mtu;
600 
601 	ndevctx->device_ctx = hdev;
602 	hv_set_drvdata(hdev, ndev);
603 	device_info.ring_size = ring_size;
604 	rndis_filter_device_add(hdev, &device_info);
605 	netif_wake_queue(ndev);
606 
607 	return 0;
608 }
609 
610 
611 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
612 {
613 	struct net_device_context *ndevctx = netdev_priv(ndev);
614 	struct hv_device *hdev =  ndevctx->device_ctx;
615 	struct sockaddr *addr = p;
616 	char save_adr[ETH_ALEN];
617 	unsigned char save_aatype;
618 	int err;
619 
620 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
621 	save_aatype = ndev->addr_assign_type;
622 
623 	err = eth_mac_addr(ndev, p);
624 	if (err != 0)
625 		return err;
626 
627 	err = rndis_filter_set_device_mac(hdev, addr->sa_data);
628 	if (err != 0) {
629 		/* roll back to saved MAC */
630 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
631 		ndev->addr_assign_type = save_aatype;
632 	}
633 
634 	return err;
635 }
636 
637 
638 static const struct ethtool_ops ethtool_ops = {
639 	.get_drvinfo	= netvsc_get_drvinfo,
640 	.get_link	= ethtool_op_get_link,
641 };
642 
643 static const struct net_device_ops device_ops = {
644 	.ndo_open =			netvsc_open,
645 	.ndo_stop =			netvsc_close,
646 	.ndo_start_xmit =		netvsc_start_xmit,
647 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
648 	.ndo_change_mtu =		netvsc_change_mtu,
649 	.ndo_validate_addr =		eth_validate_addr,
650 	.ndo_set_mac_address =		netvsc_set_mac_addr,
651 };
652 
653 /*
654  * Send GARP packet to network peers after migrations.
655  * After Quick Migration, the network is not immediately operational in the
656  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
657  * another netif_notify_peers() into a delayed work, otherwise GARP packet
658  * will not be sent after quick migration, and cause network disconnection.
659  * Also, we update the carrier status here.
660  */
661 static void netvsc_link_change(struct work_struct *w)
662 {
663 	struct net_device_context *ndev_ctx;
664 	struct net_device *net;
665 	struct netvsc_device *net_device;
666 	struct rndis_device *rdev;
667 	bool notify;
668 
669 	rtnl_lock();
670 
671 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
672 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
673 	rdev = net_device->extension;
674 	net = net_device->ndev;
675 
676 	if (rdev->link_state) {
677 		netif_carrier_off(net);
678 		notify = false;
679 	} else {
680 		netif_carrier_on(net);
681 		notify = true;
682 	}
683 
684 	rtnl_unlock();
685 
686 	if (notify)
687 		netdev_notify_peers(net);
688 }
689 
690 
691 static int netvsc_probe(struct hv_device *dev,
692 			const struct hv_vmbus_device_id *dev_id)
693 {
694 	struct net_device *net = NULL;
695 	struct net_device_context *net_device_ctx;
696 	struct netvsc_device_info device_info;
697 	int ret;
698 
699 	net = alloc_etherdev(sizeof(struct net_device_context));
700 	if (!net)
701 		return -ENOMEM;
702 
703 	netif_carrier_off(net);
704 
705 	net_device_ctx = netdev_priv(net);
706 	net_device_ctx->device_ctx = dev;
707 	hv_set_drvdata(dev, net);
708 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
709 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
710 
711 	net->netdev_ops = &device_ops;
712 
713 	net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
714 				NETIF_F_TSO;
715 	net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
716 			NETIF_F_IP_CSUM | NETIF_F_TSO;
717 
718 	SET_ETHTOOL_OPS(net, &ethtool_ops);
719 	SET_NETDEV_DEV(net, &dev->device);
720 
721 	/* Notify the netvsc driver of the new device */
722 	device_info.ring_size = ring_size;
723 	ret = rndis_filter_device_add(dev, &device_info);
724 	if (ret != 0) {
725 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
726 		free_netdev(net);
727 		hv_set_drvdata(dev, NULL);
728 		return ret;
729 	}
730 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
731 
732 	ret = register_netdev(net);
733 	if (ret != 0) {
734 		pr_err("Unable to register netdev.\n");
735 		rndis_filter_device_remove(dev);
736 		free_netdev(net);
737 	} else {
738 		schedule_delayed_work(&net_device_ctx->dwork, 0);
739 	}
740 
741 	return ret;
742 }
743 
744 static int netvsc_remove(struct hv_device *dev)
745 {
746 	struct net_device *net;
747 	struct net_device_context *ndev_ctx;
748 	struct netvsc_device *net_device;
749 
750 	net_device = hv_get_drvdata(dev);
751 	net = net_device->ndev;
752 
753 	if (net == NULL) {
754 		dev_err(&dev->device, "No net device to remove\n");
755 		return 0;
756 	}
757 
758 	net_device->start_remove = true;
759 
760 	ndev_ctx = netdev_priv(net);
761 	cancel_delayed_work_sync(&ndev_ctx->dwork);
762 	cancel_work_sync(&ndev_ctx->work);
763 
764 	/* Stop outbound asap */
765 	netif_tx_disable(net);
766 
767 	unregister_netdev(net);
768 
769 	/*
770 	 * Call to the vsc driver to let it know that the device is being
771 	 * removed
772 	 */
773 	rndis_filter_device_remove(dev);
774 
775 	free_netdev(net);
776 	return 0;
777 }
778 
779 static const struct hv_vmbus_device_id id_table[] = {
780 	/* Network guid */
781 	{ HV_NIC_GUID, },
782 	{ },
783 };
784 
785 MODULE_DEVICE_TABLE(vmbus, id_table);
786 
787 /* The one and only one */
788 static struct  hv_driver netvsc_drv = {
789 	.name = KBUILD_MODNAME,
790 	.id_table = id_table,
791 	.probe = netvsc_probe,
792 	.remove = netvsc_remove,
793 };
794 
795 static void __exit netvsc_drv_exit(void)
796 {
797 	vmbus_driver_unregister(&netvsc_drv);
798 }
799 
800 static int __init netvsc_drv_init(void)
801 {
802 	if (ring_size < RING_SIZE_MIN) {
803 		ring_size = RING_SIZE_MIN;
804 		pr_info("Increased ring_size to %d (min allowed)\n",
805 			ring_size);
806 	}
807 	return vmbus_driver_register(&netvsc_drv);
808 }
809 
810 MODULE_LICENSE("GPL");
811 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
812 
813 module_init(netvsc_drv_init);
814 module_exit(netvsc_drv_exit);
815