xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision aac5987a)
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 #define RING_SIZE_MIN 64
44 #define LINKCHANGE_INT (2 * HZ)
45 
46 static int ring_size = 128;
47 module_param(ring_size, int, S_IRUGO);
48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
49 
50 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
51 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
52 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
53 				NETIF_MSG_TX_ERR;
54 
55 static int debug = -1;
56 module_param(debug, int, S_IRUGO);
57 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
58 
59 static void do_set_multicast(struct work_struct *w)
60 {
61 	struct net_device_context *ndevctx =
62 		container_of(w, struct net_device_context, work);
63 	struct hv_device *device_obj = ndevctx->device_ctx;
64 	struct net_device *ndev = hv_get_drvdata(device_obj);
65 	struct netvsc_device *nvdev = ndevctx->nvdev;
66 	struct rndis_device *rdev;
67 
68 	if (!nvdev)
69 		return;
70 
71 	rdev = nvdev->extension;
72 	if (rdev == NULL)
73 		return;
74 
75 	if (ndev->flags & IFF_PROMISC)
76 		rndis_filter_set_packet_filter(rdev,
77 			NDIS_PACKET_TYPE_PROMISCUOUS);
78 	else
79 		rndis_filter_set_packet_filter(rdev,
80 			NDIS_PACKET_TYPE_BROADCAST |
81 			NDIS_PACKET_TYPE_ALL_MULTICAST |
82 			NDIS_PACKET_TYPE_DIRECTED);
83 }
84 
85 static void netvsc_set_multicast_list(struct net_device *net)
86 {
87 	struct net_device_context *net_device_ctx = netdev_priv(net);
88 
89 	schedule_work(&net_device_ctx->work);
90 }
91 
92 static int netvsc_open(struct net_device *net)
93 {
94 	struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
95 	struct rndis_device *rdev;
96 	int ret = 0;
97 
98 	netif_carrier_off(net);
99 
100 	/* Open up the device */
101 	ret = rndis_filter_open(nvdev);
102 	if (ret != 0) {
103 		netdev_err(net, "unable to open device (ret %d).\n", ret);
104 		return ret;
105 	}
106 
107 	netif_tx_wake_all_queues(net);
108 
109 	rdev = nvdev->extension;
110 	if (!rdev->link_state)
111 		netif_carrier_on(net);
112 
113 	return ret;
114 }
115 
116 static int netvsc_close(struct net_device *net)
117 {
118 	struct net_device_context *net_device_ctx = netdev_priv(net);
119 	struct netvsc_device *nvdev = net_device_ctx->nvdev;
120 	int ret;
121 	u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
122 	struct vmbus_channel *chn;
123 
124 	netif_tx_disable(net);
125 
126 	/* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
127 	cancel_work_sync(&net_device_ctx->work);
128 	ret = rndis_filter_close(nvdev);
129 	if (ret != 0) {
130 		netdev_err(net, "unable to close device (ret %d).\n", ret);
131 		return ret;
132 	}
133 
134 	/* Ensure pending bytes in ring are read */
135 	while (true) {
136 		aread = 0;
137 		for (i = 0; i < nvdev->num_chn; i++) {
138 			chn = nvdev->chan_table[i].channel;
139 			if (!chn)
140 				continue;
141 
142 			hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
143 						     &awrite);
144 
145 			if (aread)
146 				break;
147 
148 			hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
149 						     &awrite);
150 
151 			if (aread)
152 				break;
153 		}
154 
155 		retry++;
156 		if (retry > retry_max || aread == 0)
157 			break;
158 
159 		msleep(msec);
160 
161 		if (msec < 1000)
162 			msec *= 2;
163 	}
164 
165 	if (aread) {
166 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
167 		ret = -ETIMEDOUT;
168 	}
169 
170 	return ret;
171 }
172 
173 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
174 				int pkt_type)
175 {
176 	struct rndis_packet *rndis_pkt;
177 	struct rndis_per_packet_info *ppi;
178 
179 	rndis_pkt = &msg->msg.pkt;
180 	rndis_pkt->data_offset += ppi_size;
181 
182 	ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
183 		rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
184 
185 	ppi->size = ppi_size;
186 	ppi->type = pkt_type;
187 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
188 
189 	rndis_pkt->per_pkt_info_len += ppi_size;
190 
191 	return ppi;
192 }
193 
194 /*
195  * Select queue for transmit.
196  *
197  * If a valid queue has already been assigned, then use that.
198  * Otherwise compute tx queue based on hash and the send table.
199  *
200  * This is basically similar to default (__netdev_pick_tx) with the added step
201  * of using the host send_table when no other queue has been assigned.
202  *
203  * TODO support XPS - but get_xps_queue not exported
204  */
205 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
206 			void *accel_priv, select_queue_fallback_t fallback)
207 {
208 	struct net_device_context *net_device_ctx = netdev_priv(ndev);
209 	struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
210 	struct sock *sk = skb->sk;
211 	int q_idx = sk_tx_queue_get(sk);
212 
213 	if (q_idx < 0 || skb->ooo_okay ||
214 	    q_idx >= ndev->real_num_tx_queues) {
215 		u16 hash = __skb_tx_hash(ndev, skb, VRSS_SEND_TAB_SIZE);
216 		int new_idx;
217 
218 		new_idx = nvsc_dev->send_table[hash]
219 			% nvsc_dev->num_chn;
220 
221 		if (q_idx != new_idx && sk &&
222 		    sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
223 			sk_tx_queue_set(sk, new_idx);
224 
225 		q_idx = new_idx;
226 	}
227 
228 	if (unlikely(!nvsc_dev->chan_table[q_idx].channel))
229 		q_idx = 0;
230 
231 	return q_idx;
232 }
233 
234 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
235 			struct hv_page_buffer *pb)
236 {
237 	int j = 0;
238 
239 	/* Deal with compund pages by ignoring unused part
240 	 * of the page.
241 	 */
242 	page += (offset >> PAGE_SHIFT);
243 	offset &= ~PAGE_MASK;
244 
245 	while (len > 0) {
246 		unsigned long bytes;
247 
248 		bytes = PAGE_SIZE - offset;
249 		if (bytes > len)
250 			bytes = len;
251 		pb[j].pfn = page_to_pfn(page);
252 		pb[j].offset = offset;
253 		pb[j].len = bytes;
254 
255 		offset += bytes;
256 		len -= bytes;
257 
258 		if (offset == PAGE_SIZE && len) {
259 			page++;
260 			offset = 0;
261 			j++;
262 		}
263 	}
264 
265 	return j + 1;
266 }
267 
268 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
269 			   struct hv_netvsc_packet *packet,
270 			   struct hv_page_buffer **page_buf)
271 {
272 	struct hv_page_buffer *pb = *page_buf;
273 	u32 slots_used = 0;
274 	char *data = skb->data;
275 	int frags = skb_shinfo(skb)->nr_frags;
276 	int i;
277 
278 	/* The packet is laid out thus:
279 	 * 1. hdr: RNDIS header and PPI
280 	 * 2. skb linear data
281 	 * 3. skb fragment data
282 	 */
283 	if (hdr != NULL)
284 		slots_used += fill_pg_buf(virt_to_page(hdr),
285 					offset_in_page(hdr),
286 					len, &pb[slots_used]);
287 
288 	packet->rmsg_size = len;
289 	packet->rmsg_pgcnt = slots_used;
290 
291 	slots_used += fill_pg_buf(virt_to_page(data),
292 				offset_in_page(data),
293 				skb_headlen(skb), &pb[slots_used]);
294 
295 	for (i = 0; i < frags; i++) {
296 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
297 
298 		slots_used += fill_pg_buf(skb_frag_page(frag),
299 					frag->page_offset,
300 					skb_frag_size(frag), &pb[slots_used]);
301 	}
302 	return slots_used;
303 }
304 
305 static int count_skb_frag_slots(struct sk_buff *skb)
306 {
307 	int i, frags = skb_shinfo(skb)->nr_frags;
308 	int pages = 0;
309 
310 	for (i = 0; i < frags; i++) {
311 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
312 		unsigned long size = skb_frag_size(frag);
313 		unsigned long offset = frag->page_offset;
314 
315 		/* Skip unused frames from start of page */
316 		offset &= ~PAGE_MASK;
317 		pages += PFN_UP(offset + size);
318 	}
319 	return pages;
320 }
321 
322 static int netvsc_get_slots(struct sk_buff *skb)
323 {
324 	char *data = skb->data;
325 	unsigned int offset = offset_in_page(data);
326 	unsigned int len = skb_headlen(skb);
327 	int slots;
328 	int frag_slots;
329 
330 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
331 	frag_slots = count_skb_frag_slots(skb);
332 	return slots + frag_slots;
333 }
334 
335 static u32 net_checksum_info(struct sk_buff *skb)
336 {
337 	if (skb->protocol == htons(ETH_P_IP)) {
338 		struct iphdr *ip = ip_hdr(skb);
339 
340 		if (ip->protocol == IPPROTO_TCP)
341 			return TRANSPORT_INFO_IPV4_TCP;
342 		else if (ip->protocol == IPPROTO_UDP)
343 			return TRANSPORT_INFO_IPV4_UDP;
344 	} else {
345 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
346 
347 		if (ip6->nexthdr == IPPROTO_TCP)
348 			return TRANSPORT_INFO_IPV6_TCP;
349 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
350 			return TRANSPORT_INFO_IPV6_UDP;
351 	}
352 
353 	return TRANSPORT_INFO_NOT_IP;
354 }
355 
356 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
357 {
358 	struct net_device_context *net_device_ctx = netdev_priv(net);
359 	struct hv_netvsc_packet *packet = NULL;
360 	int ret;
361 	unsigned int num_data_pgs;
362 	struct rndis_message *rndis_msg;
363 	struct rndis_packet *rndis_pkt;
364 	u32 rndis_msg_size;
365 	struct rndis_per_packet_info *ppi;
366 	u32 hash;
367 	struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
368 	struct hv_page_buffer *pb = page_buf;
369 
370 	/* We will atmost need two pages to describe the rndis
371 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
372 	 * of pages in a single packet. If skb is scattered around
373 	 * more pages we try linearizing it.
374 	 */
375 
376 	num_data_pgs = netvsc_get_slots(skb) + 2;
377 
378 	if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
379 		++net_device_ctx->eth_stats.tx_scattered;
380 
381 		if (skb_linearize(skb))
382 			goto no_memory;
383 
384 		num_data_pgs = netvsc_get_slots(skb) + 2;
385 		if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
386 			++net_device_ctx->eth_stats.tx_too_big;
387 			goto drop;
388 		}
389 	}
390 
391 	/*
392 	 * Place the rndis header in the skb head room and
393 	 * the skb->cb will be used for hv_netvsc_packet
394 	 * structure.
395 	 */
396 	ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
397 	if (ret)
398 		goto no_memory;
399 
400 	/* Use the skb control buffer for building up the packet */
401 	BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
402 			FIELD_SIZEOF(struct sk_buff, cb));
403 	packet = (struct hv_netvsc_packet *)skb->cb;
404 
405 	packet->q_idx = skb_get_queue_mapping(skb);
406 
407 	packet->total_data_buflen = skb->len;
408 	packet->total_bytes = skb->len;
409 	packet->total_packets = 1;
410 
411 	rndis_msg = (struct rndis_message *)skb->head;
412 
413 	memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
414 
415 	/* Add the rndis header */
416 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
417 	rndis_msg->msg_len = packet->total_data_buflen;
418 	rndis_pkt = &rndis_msg->msg.pkt;
419 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
420 	rndis_pkt->data_len = packet->total_data_buflen;
421 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
422 
423 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
424 
425 	hash = skb_get_hash_raw(skb);
426 	if (hash != 0 && net->real_num_tx_queues > 1) {
427 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
428 		ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
429 				    NBL_HASH_VALUE);
430 		*(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
431 	}
432 
433 	if (skb_vlan_tag_present(skb)) {
434 		struct ndis_pkt_8021q_info *vlan;
435 
436 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
437 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
438 					IEEE_8021Q_INFO);
439 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
440 						ppi->ppi_offset);
441 		vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
442 		vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
443 				VLAN_PRIO_SHIFT;
444 	}
445 
446 	if (skb_is_gso(skb)) {
447 		struct ndis_tcp_lso_info *lso_info;
448 
449 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
450 		ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
451 				    TCP_LARGESEND_PKTINFO);
452 
453 		lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
454 							ppi->ppi_offset);
455 
456 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
457 		if (skb->protocol == htons(ETH_P_IP)) {
458 			lso_info->lso_v2_transmit.ip_version =
459 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
460 			ip_hdr(skb)->tot_len = 0;
461 			ip_hdr(skb)->check = 0;
462 			tcp_hdr(skb)->check =
463 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
464 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
465 		} else {
466 			lso_info->lso_v2_transmit.ip_version =
467 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
468 			ipv6_hdr(skb)->payload_len = 0;
469 			tcp_hdr(skb)->check =
470 				~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
471 						 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
472 		}
473 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
474 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
475 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
476 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
477 			struct ndis_tcp_ip_checksum_info *csum_info;
478 
479 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
480 			ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
481 					    TCPIP_CHKSUM_PKTINFO);
482 
483 			csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
484 									 ppi->ppi_offset);
485 
486 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
487 
488 			if (skb->protocol == htons(ETH_P_IP)) {
489 				csum_info->transmit.is_ipv4 = 1;
490 
491 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
492 					csum_info->transmit.tcp_checksum = 1;
493 				else
494 					csum_info->transmit.udp_checksum = 1;
495 			} else {
496 				csum_info->transmit.is_ipv6 = 1;
497 
498 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
499 					csum_info->transmit.tcp_checksum = 1;
500 				else
501 					csum_info->transmit.udp_checksum = 1;
502 			}
503 		} else {
504 			/* Can't do offload of this type of checksum */
505 			if (skb_checksum_help(skb))
506 				goto drop;
507 		}
508 	}
509 
510 	/* Start filling in the page buffers with the rndis hdr */
511 	rndis_msg->msg_len += rndis_msg_size;
512 	packet->total_data_buflen = rndis_msg->msg_len;
513 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
514 					       skb, packet, &pb);
515 
516 	/* timestamp packet in software */
517 	skb_tx_timestamp(skb);
518 	ret = netvsc_send(net_device_ctx->device_ctx, packet,
519 			  rndis_msg, &pb, skb);
520 	if (likely(ret == 0))
521 		return NETDEV_TX_OK;
522 
523 	if (ret == -EAGAIN) {
524 		++net_device_ctx->eth_stats.tx_busy;
525 		return NETDEV_TX_BUSY;
526 	}
527 
528 	if (ret == -ENOSPC)
529 		++net_device_ctx->eth_stats.tx_no_space;
530 
531 drop:
532 	dev_kfree_skb_any(skb);
533 	net->stats.tx_dropped++;
534 
535 	return NETDEV_TX_OK;
536 
537 no_memory:
538 	++net_device_ctx->eth_stats.tx_no_memory;
539 	goto drop;
540 }
541 /*
542  * netvsc_linkstatus_callback - Link up/down notification
543  */
544 void netvsc_linkstatus_callback(struct hv_device *device_obj,
545 				struct rndis_message *resp)
546 {
547 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
548 	struct net_device *net;
549 	struct net_device_context *ndev_ctx;
550 	struct netvsc_reconfig *event;
551 	unsigned long flags;
552 
553 	net = hv_get_drvdata(device_obj);
554 
555 	if (!net)
556 		return;
557 
558 	ndev_ctx = netdev_priv(net);
559 
560 	/* Update the physical link speed when changing to another vSwitch */
561 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
562 		u32 speed;
563 
564 		speed = *(u32 *)((void *)indicate + indicate->
565 				 status_buf_offset) / 10000;
566 		ndev_ctx->speed = speed;
567 		return;
568 	}
569 
570 	/* Handle these link change statuses below */
571 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
572 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
573 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
574 		return;
575 
576 	if (net->reg_state != NETREG_REGISTERED)
577 		return;
578 
579 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
580 	if (!event)
581 		return;
582 	event->event = indicate->status;
583 
584 	spin_lock_irqsave(&ndev_ctx->lock, flags);
585 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
586 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
587 
588 	schedule_delayed_work(&ndev_ctx->dwork, 0);
589 }
590 
591 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
592 					     const struct ndis_tcp_ip_checksum_info *csum_info,
593 					     const struct ndis_pkt_8021q_info *vlan,
594 					     void *data, u32 buflen)
595 {
596 	struct sk_buff *skb;
597 
598 	skb = netdev_alloc_skb_ip_align(net, buflen);
599 	if (!skb)
600 		return skb;
601 
602 	/*
603 	 * Copy to skb. This copy is needed here since the memory pointed by
604 	 * hv_netvsc_packet cannot be deallocated
605 	 */
606 	memcpy(skb_put(skb, buflen), data, buflen);
607 
608 	skb->protocol = eth_type_trans(skb, net);
609 
610 	/* skb is already created with CHECKSUM_NONE */
611 	skb_checksum_none_assert(skb);
612 
613 	/*
614 	 * In Linux, the IP checksum is always checked.
615 	 * Do L4 checksum offload if enabled and present.
616 	 */
617 	if (csum_info && (net->features & NETIF_F_RXCSUM)) {
618 		if (csum_info->receive.tcp_checksum_succeeded ||
619 		    csum_info->receive.udp_checksum_succeeded)
620 			skb->ip_summed = CHECKSUM_UNNECESSARY;
621 	}
622 
623 	if (vlan) {
624 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT);
625 
626 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
627 				       vlan_tci);
628 	}
629 
630 	return skb;
631 }
632 
633 /*
634  * netvsc_recv_callback -  Callback when we receive a packet from the
635  * "wire" on the specified device.
636  */
637 int netvsc_recv_callback(struct net_device *net,
638 			 struct vmbus_channel *channel,
639 			 void  *data, u32 len,
640 			 const struct ndis_tcp_ip_checksum_info *csum_info,
641 			 const struct ndis_pkt_8021q_info *vlan)
642 {
643 	struct net_device_context *net_device_ctx = netdev_priv(net);
644 	struct netvsc_device *net_device = net_device_ctx->nvdev;
645 	struct net_device *vf_netdev;
646 	struct sk_buff *skb;
647 	struct netvsc_stats *rx_stats;
648 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
649 
650 
651 	if (net->reg_state != NETREG_REGISTERED)
652 		return NVSP_STAT_FAIL;
653 
654 	/*
655 	 * If necessary, inject this packet into the VF interface.
656 	 * On Hyper-V, multicast and brodcast packets are only delivered
657 	 * to the synthetic interface (after subjecting these to
658 	 * policy filters on the host). Deliver these via the VF
659 	 * interface in the guest.
660 	 */
661 	rcu_read_lock();
662 	vf_netdev = rcu_dereference(net_device_ctx->vf_netdev);
663 	if (vf_netdev && (vf_netdev->flags & IFF_UP))
664 		net = vf_netdev;
665 
666 	/* Allocate a skb - TODO direct I/O to pages? */
667 	skb = netvsc_alloc_recv_skb(net, csum_info, vlan, data, len);
668 	if (unlikely(!skb)) {
669 		++net->stats.rx_dropped;
670 		rcu_read_unlock();
671 		return NVSP_STAT_FAIL;
672 	}
673 
674 	if (net != vf_netdev)
675 		skb_record_rx_queue(skb, q_idx);
676 
677 	/*
678 	 * Even if injecting the packet, record the statistics
679 	 * on the synthetic device because modifying the VF device
680 	 * statistics will not work correctly.
681 	 */
682 	rx_stats = &net_device->chan_table[q_idx].rx_stats;
683 	u64_stats_update_begin(&rx_stats->syncp);
684 	rx_stats->packets++;
685 	rx_stats->bytes += len;
686 
687 	if (skb->pkt_type == PACKET_BROADCAST)
688 		++rx_stats->broadcast;
689 	else if (skb->pkt_type == PACKET_MULTICAST)
690 		++rx_stats->multicast;
691 	u64_stats_update_end(&rx_stats->syncp);
692 
693 	/*
694 	 * Pass the skb back up. Network stack will deallocate the skb when it
695 	 * is done.
696 	 * TODO - use NAPI?
697 	 */
698 	netif_receive_skb(skb);
699 	rcu_read_unlock();
700 
701 	return 0;
702 }
703 
704 static void netvsc_get_drvinfo(struct net_device *net,
705 			       struct ethtool_drvinfo *info)
706 {
707 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
708 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
709 }
710 
711 static void netvsc_get_channels(struct net_device *net,
712 				struct ethtool_channels *channel)
713 {
714 	struct net_device_context *net_device_ctx = netdev_priv(net);
715 	struct netvsc_device *nvdev = net_device_ctx->nvdev;
716 
717 	if (nvdev) {
718 		channel->max_combined	= nvdev->max_chn;
719 		channel->combined_count = nvdev->num_chn;
720 	}
721 }
722 
723 static int netvsc_set_queues(struct net_device *net, struct hv_device *dev,
724 			     u32 num_chn)
725 {
726 	struct netvsc_device_info device_info;
727 	int ret;
728 
729 	memset(&device_info, 0, sizeof(device_info));
730 	device_info.num_chn = num_chn;
731 	device_info.ring_size = ring_size;
732 	device_info.max_num_vrss_chns = num_chn;
733 
734 	ret = rndis_filter_device_add(dev, &device_info);
735 	if (ret)
736 		return ret;
737 
738 	ret = netif_set_real_num_tx_queues(net, num_chn);
739 	if (ret)
740 		return ret;
741 
742 	ret = netif_set_real_num_rx_queues(net, num_chn);
743 
744 	return ret;
745 }
746 
747 static int netvsc_set_channels(struct net_device *net,
748 			       struct ethtool_channels *channels)
749 {
750 	struct net_device_context *net_device_ctx = netdev_priv(net);
751 	struct hv_device *dev = net_device_ctx->device_ctx;
752 	struct netvsc_device *nvdev = net_device_ctx->nvdev;
753 	unsigned int count = channels->combined_count;
754 	int ret;
755 
756 	/* We do not support separate count for rx, tx, or other */
757 	if (count == 0 ||
758 	    channels->rx_count || channels->tx_count || channels->other_count)
759 		return -EINVAL;
760 
761 	if (count > net->num_tx_queues || count > net->num_rx_queues)
762 		return -EINVAL;
763 
764 	if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
765 		return -ENODEV;
766 
767 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
768 		return -EINVAL;
769 
770 	if (count > nvdev->max_chn)
771 		return -EINVAL;
772 
773 	ret = netvsc_close(net);
774 	if (ret)
775 		return ret;
776 
777 	net_device_ctx->start_remove = true;
778 	rndis_filter_device_remove(dev, nvdev);
779 
780 	ret = netvsc_set_queues(net, dev, count);
781 	if (ret == 0)
782 		nvdev->num_chn = count;
783 	else
784 		netvsc_set_queues(net, dev, nvdev->num_chn);
785 
786 	netvsc_open(net);
787 	net_device_ctx->start_remove = false;
788 
789 	/* We may have missed link change notifications */
790 	schedule_delayed_work(&net_device_ctx->dwork, 0);
791 
792 	return ret;
793 }
794 
795 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
796 {
797 	struct ethtool_cmd diff1 = *cmd;
798 	struct ethtool_cmd diff2 = {};
799 
800 	ethtool_cmd_speed_set(&diff1, 0);
801 	diff1.duplex = 0;
802 	/* advertising and cmd are usually set */
803 	diff1.advertising = 0;
804 	diff1.cmd = 0;
805 	/* We set port to PORT_OTHER */
806 	diff2.port = PORT_OTHER;
807 
808 	return !memcmp(&diff1, &diff2, sizeof(diff1));
809 }
810 
811 static void netvsc_init_settings(struct net_device *dev)
812 {
813 	struct net_device_context *ndc = netdev_priv(dev);
814 
815 	ndc->speed = SPEED_UNKNOWN;
816 	ndc->duplex = DUPLEX_UNKNOWN;
817 }
818 
819 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
820 {
821 	struct net_device_context *ndc = netdev_priv(dev);
822 
823 	ethtool_cmd_speed_set(cmd, ndc->speed);
824 	cmd->duplex = ndc->duplex;
825 	cmd->port = PORT_OTHER;
826 
827 	return 0;
828 }
829 
830 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
831 {
832 	struct net_device_context *ndc = netdev_priv(dev);
833 	u32 speed;
834 
835 	speed = ethtool_cmd_speed(cmd);
836 	if (!ethtool_validate_speed(speed) ||
837 	    !ethtool_validate_duplex(cmd->duplex) ||
838 	    !netvsc_validate_ethtool_ss_cmd(cmd))
839 		return -EINVAL;
840 
841 	ndc->speed = speed;
842 	ndc->duplex = cmd->duplex;
843 
844 	return 0;
845 }
846 
847 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
848 {
849 	struct net_device_context *ndevctx = netdev_priv(ndev);
850 	struct netvsc_device *nvdev = ndevctx->nvdev;
851 	struct hv_device *hdev = ndevctx->device_ctx;
852 	struct netvsc_device_info device_info;
853 	int ret;
854 
855 	if (ndevctx->start_remove || !nvdev || nvdev->destroy)
856 		return -ENODEV;
857 
858 	ret = netvsc_close(ndev);
859 	if (ret)
860 		goto out;
861 
862 	memset(&device_info, 0, sizeof(device_info));
863 	device_info.ring_size = ring_size;
864 	device_info.num_chn = nvdev->num_chn;
865 	device_info.max_num_vrss_chns = nvdev->num_chn;
866 
867 	ndevctx->start_remove = true;
868 	rndis_filter_device_remove(hdev, nvdev);
869 
870 	/* 'nvdev' has been freed in rndis_filter_device_remove() ->
871 	 * netvsc_device_remove () -> free_netvsc_device().
872 	 * We mustn't access it before it's re-created in
873 	 * rndis_filter_device_add() -> netvsc_device_add().
874 	 */
875 
876 	ndev->mtu = mtu;
877 
878 	rndis_filter_device_add(hdev, &device_info);
879 
880 out:
881 	netvsc_open(ndev);
882 	ndevctx->start_remove = false;
883 
884 	/* We may have missed link change notifications */
885 	schedule_delayed_work(&ndevctx->dwork, 0);
886 
887 	return ret;
888 }
889 
890 static void netvsc_get_stats64(struct net_device *net,
891 			       struct rtnl_link_stats64 *t)
892 {
893 	struct net_device_context *ndev_ctx = netdev_priv(net);
894 	struct netvsc_device *nvdev = ndev_ctx->nvdev;
895 	int i;
896 
897 	if (!nvdev)
898 		return;
899 
900 	for (i = 0; i < nvdev->num_chn; i++) {
901 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
902 		const struct netvsc_stats *stats;
903 		u64 packets, bytes, multicast;
904 		unsigned int start;
905 
906 		stats = &nvchan->tx_stats;
907 		do {
908 			start = u64_stats_fetch_begin_irq(&stats->syncp);
909 			packets = stats->packets;
910 			bytes = stats->bytes;
911 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
912 
913 		t->tx_bytes	+= bytes;
914 		t->tx_packets	+= packets;
915 
916 		stats = &nvchan->rx_stats;
917 		do {
918 			start = u64_stats_fetch_begin_irq(&stats->syncp);
919 			packets = stats->packets;
920 			bytes = stats->bytes;
921 			multicast = stats->multicast + stats->broadcast;
922 		} while (u64_stats_fetch_retry_irq(&stats->syncp, start));
923 
924 		t->rx_bytes	+= bytes;
925 		t->rx_packets	+= packets;
926 		t->multicast	+= multicast;
927 	}
928 
929 	t->tx_dropped	= net->stats.tx_dropped;
930 	t->tx_errors	= net->stats.tx_errors;
931 
932 	t->rx_dropped	= net->stats.rx_dropped;
933 	t->rx_errors	= net->stats.rx_errors;
934 }
935 
936 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
937 {
938 	struct sockaddr *addr = p;
939 	char save_adr[ETH_ALEN];
940 	unsigned char save_aatype;
941 	int err;
942 
943 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
944 	save_aatype = ndev->addr_assign_type;
945 
946 	err = eth_mac_addr(ndev, p);
947 	if (err != 0)
948 		return err;
949 
950 	err = rndis_filter_set_device_mac(ndev, addr->sa_data);
951 	if (err != 0) {
952 		/* roll back to saved MAC */
953 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
954 		ndev->addr_assign_type = save_aatype;
955 	}
956 
957 	return err;
958 }
959 
960 static const struct {
961 	char name[ETH_GSTRING_LEN];
962 	u16 offset;
963 } netvsc_stats[] = {
964 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
965 	{ "tx_no_memory",  offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
966 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
967 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
968 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
969 };
970 
971 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
972 
973 /* 4 statistics per queue (rx/tx packets/bytes) */
974 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
975 
976 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
977 {
978 	struct net_device_context *ndc = netdev_priv(dev);
979 	struct netvsc_device *nvdev = ndc->nvdev;
980 
981 	switch (string_set) {
982 	case ETH_SS_STATS:
983 		return NETVSC_GLOBAL_STATS_LEN + NETVSC_QUEUE_STATS_LEN(nvdev);
984 	default:
985 		return -EINVAL;
986 	}
987 }
988 
989 static void netvsc_get_ethtool_stats(struct net_device *dev,
990 				     struct ethtool_stats *stats, u64 *data)
991 {
992 	struct net_device_context *ndc = netdev_priv(dev);
993 	struct netvsc_device *nvdev = ndc->nvdev;
994 	const void *nds = &ndc->eth_stats;
995 	const struct netvsc_stats *qstats;
996 	unsigned int start;
997 	u64 packets, bytes;
998 	int i, j;
999 
1000 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1001 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1002 
1003 	for (j = 0; j < nvdev->num_chn; j++) {
1004 		qstats = &nvdev->chan_table[j].tx_stats;
1005 
1006 		do {
1007 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1008 			packets = qstats->packets;
1009 			bytes = qstats->bytes;
1010 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1011 		data[i++] = packets;
1012 		data[i++] = bytes;
1013 
1014 		qstats = &nvdev->chan_table[j].rx_stats;
1015 		do {
1016 			start = u64_stats_fetch_begin_irq(&qstats->syncp);
1017 			packets = qstats->packets;
1018 			bytes = qstats->bytes;
1019 		} while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1020 		data[i++] = packets;
1021 		data[i++] = bytes;
1022 	}
1023 }
1024 
1025 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1026 {
1027 	struct net_device_context *ndc = netdev_priv(dev);
1028 	struct netvsc_device *nvdev = ndc->nvdev;
1029 	u8 *p = data;
1030 	int i;
1031 
1032 	switch (stringset) {
1033 	case ETH_SS_STATS:
1034 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1035 			memcpy(p + i * ETH_GSTRING_LEN,
1036 			       netvsc_stats[i].name, ETH_GSTRING_LEN);
1037 
1038 		p += i * ETH_GSTRING_LEN;
1039 		for (i = 0; i < nvdev->num_chn; i++) {
1040 			sprintf(p, "tx_queue_%u_packets", i);
1041 			p += ETH_GSTRING_LEN;
1042 			sprintf(p, "tx_queue_%u_bytes", i);
1043 			p += ETH_GSTRING_LEN;
1044 			sprintf(p, "rx_queue_%u_packets", i);
1045 			p += ETH_GSTRING_LEN;
1046 			sprintf(p, "rx_queue_%u_bytes", i);
1047 			p += ETH_GSTRING_LEN;
1048 		}
1049 
1050 		break;
1051 	}
1052 }
1053 
1054 static int
1055 netvsc_get_rss_hash_opts(struct netvsc_device *nvdev,
1056 			 struct ethtool_rxnfc *info)
1057 {
1058 	info->data = RXH_IP_SRC | RXH_IP_DST;
1059 
1060 	switch (info->flow_type) {
1061 	case TCP_V4_FLOW:
1062 	case TCP_V6_FLOW:
1063 		info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
1064 		/* fallthrough */
1065 	case UDP_V4_FLOW:
1066 	case UDP_V6_FLOW:
1067 	case IPV4_FLOW:
1068 	case IPV6_FLOW:
1069 		break;
1070 	default:
1071 		info->data = 0;
1072 		break;
1073 	}
1074 
1075 	return 0;
1076 }
1077 
1078 static int
1079 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1080 		 u32 *rules)
1081 {
1082 	struct net_device_context *ndc = netdev_priv(dev);
1083 	struct netvsc_device *nvdev = ndc->nvdev;
1084 
1085 	switch (info->cmd) {
1086 	case ETHTOOL_GRXRINGS:
1087 		info->data = nvdev->num_chn;
1088 		return 0;
1089 
1090 	case ETHTOOL_GRXFH:
1091 		return netvsc_get_rss_hash_opts(nvdev, info);
1092 	}
1093 	return -EOPNOTSUPP;
1094 }
1095 
1096 #ifdef CONFIG_NET_POLL_CONTROLLER
1097 static void netvsc_poll_controller(struct net_device *net)
1098 {
1099 	/* As netvsc_start_xmit() works synchronous we don't have to
1100 	 * trigger anything here.
1101 	 */
1102 }
1103 #endif
1104 
1105 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1106 {
1107 	return NETVSC_HASH_KEYLEN;
1108 }
1109 
1110 static u32 netvsc_rss_indir_size(struct net_device *dev)
1111 {
1112 	return ITAB_NUM;
1113 }
1114 
1115 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1116 			   u8 *hfunc)
1117 {
1118 	struct net_device_context *ndc = netdev_priv(dev);
1119 	struct netvsc_device *ndev = ndc->nvdev;
1120 	struct rndis_device *rndis_dev = ndev->extension;
1121 	int i;
1122 
1123 	if (hfunc)
1124 		*hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1125 
1126 	if (indir) {
1127 		for (i = 0; i < ITAB_NUM; i++)
1128 			indir[i] = rndis_dev->ind_table[i];
1129 	}
1130 
1131 	if (key)
1132 		memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1133 
1134 	return 0;
1135 }
1136 
1137 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1138 			   const u8 *key, const u8 hfunc)
1139 {
1140 	struct net_device_context *ndc = netdev_priv(dev);
1141 	struct netvsc_device *ndev = ndc->nvdev;
1142 	struct rndis_device *rndis_dev = ndev->extension;
1143 	int i;
1144 
1145 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1146 		return -EOPNOTSUPP;
1147 
1148 	if (indir) {
1149 		for (i = 0; i < ITAB_NUM; i++)
1150 			if (indir[i] >= dev->num_rx_queues)
1151 				return -EINVAL;
1152 
1153 		for (i = 0; i < ITAB_NUM; i++)
1154 			rndis_dev->ind_table[i] = indir[i];
1155 	}
1156 
1157 	if (!key) {
1158 		if (!indir)
1159 			return 0;
1160 
1161 		key = rndis_dev->rss_key;
1162 	}
1163 
1164 	return rndis_filter_set_rss_param(rndis_dev, key, ndev->num_chn);
1165 }
1166 
1167 static const struct ethtool_ops ethtool_ops = {
1168 	.get_drvinfo	= netvsc_get_drvinfo,
1169 	.get_link	= ethtool_op_get_link,
1170 	.get_ethtool_stats = netvsc_get_ethtool_stats,
1171 	.get_sset_count = netvsc_get_sset_count,
1172 	.get_strings	= netvsc_get_strings,
1173 	.get_channels   = netvsc_get_channels,
1174 	.set_channels   = netvsc_set_channels,
1175 	.get_ts_info	= ethtool_op_get_ts_info,
1176 	.get_settings	= netvsc_get_settings,
1177 	.set_settings	= netvsc_set_settings,
1178 	.get_rxnfc	= netvsc_get_rxnfc,
1179 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
1180 	.get_rxfh_indir_size = netvsc_rss_indir_size,
1181 	.get_rxfh	= netvsc_get_rxfh,
1182 	.set_rxfh	= netvsc_set_rxfh,
1183 };
1184 
1185 static const struct net_device_ops device_ops = {
1186 	.ndo_open =			netvsc_open,
1187 	.ndo_stop =			netvsc_close,
1188 	.ndo_start_xmit =		netvsc_start_xmit,
1189 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
1190 	.ndo_change_mtu =		netvsc_change_mtu,
1191 	.ndo_validate_addr =		eth_validate_addr,
1192 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1193 	.ndo_select_queue =		netvsc_select_queue,
1194 	.ndo_get_stats64 =		netvsc_get_stats64,
1195 #ifdef CONFIG_NET_POLL_CONTROLLER
1196 	.ndo_poll_controller =		netvsc_poll_controller,
1197 #endif
1198 };
1199 
1200 /*
1201  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1202  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1203  * present send GARP packet to network peers with netif_notify_peers().
1204  */
1205 static void netvsc_link_change(struct work_struct *w)
1206 {
1207 	struct net_device_context *ndev_ctx =
1208 		container_of(w, struct net_device_context, dwork.work);
1209 	struct hv_device *device_obj = ndev_ctx->device_ctx;
1210 	struct net_device *net = hv_get_drvdata(device_obj);
1211 	struct netvsc_device *net_device;
1212 	struct rndis_device *rdev;
1213 	struct netvsc_reconfig *event = NULL;
1214 	bool notify = false, reschedule = false;
1215 	unsigned long flags, next_reconfig, delay;
1216 
1217 	rtnl_lock();
1218 	if (ndev_ctx->start_remove)
1219 		goto out_unlock;
1220 
1221 	net_device = ndev_ctx->nvdev;
1222 	rdev = net_device->extension;
1223 
1224 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1225 	if (time_is_after_jiffies(next_reconfig)) {
1226 		/* link_watch only sends one notification with current state
1227 		 * per second, avoid doing reconfig more frequently. Handle
1228 		 * wrap around.
1229 		 */
1230 		delay = next_reconfig - jiffies;
1231 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1232 		schedule_delayed_work(&ndev_ctx->dwork, delay);
1233 		goto out_unlock;
1234 	}
1235 	ndev_ctx->last_reconfig = jiffies;
1236 
1237 	spin_lock_irqsave(&ndev_ctx->lock, flags);
1238 	if (!list_empty(&ndev_ctx->reconfig_events)) {
1239 		event = list_first_entry(&ndev_ctx->reconfig_events,
1240 					 struct netvsc_reconfig, list);
1241 		list_del(&event->list);
1242 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
1243 	}
1244 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1245 
1246 	if (!event)
1247 		goto out_unlock;
1248 
1249 	switch (event->event) {
1250 		/* Only the following events are possible due to the check in
1251 		 * netvsc_linkstatus_callback()
1252 		 */
1253 	case RNDIS_STATUS_MEDIA_CONNECT:
1254 		if (rdev->link_state) {
1255 			rdev->link_state = false;
1256 			netif_carrier_on(net);
1257 			netif_tx_wake_all_queues(net);
1258 		} else {
1259 			notify = true;
1260 		}
1261 		kfree(event);
1262 		break;
1263 	case RNDIS_STATUS_MEDIA_DISCONNECT:
1264 		if (!rdev->link_state) {
1265 			rdev->link_state = true;
1266 			netif_carrier_off(net);
1267 			netif_tx_stop_all_queues(net);
1268 		}
1269 		kfree(event);
1270 		break;
1271 	case RNDIS_STATUS_NETWORK_CHANGE:
1272 		/* Only makes sense if carrier is present */
1273 		if (!rdev->link_state) {
1274 			rdev->link_state = true;
1275 			netif_carrier_off(net);
1276 			netif_tx_stop_all_queues(net);
1277 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
1278 			spin_lock_irqsave(&ndev_ctx->lock, flags);
1279 			list_add(&event->list, &ndev_ctx->reconfig_events);
1280 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1281 			reschedule = true;
1282 		}
1283 		break;
1284 	}
1285 
1286 	rtnl_unlock();
1287 
1288 	if (notify)
1289 		netdev_notify_peers(net);
1290 
1291 	/* link_watch only sends one notification with current state per
1292 	 * second, handle next reconfig event in 2 seconds.
1293 	 */
1294 	if (reschedule)
1295 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1296 
1297 	return;
1298 
1299 out_unlock:
1300 	rtnl_unlock();
1301 }
1302 
1303 static struct net_device *get_netvsc_bymac(const u8 *mac)
1304 {
1305 	struct net_device *dev;
1306 
1307 	ASSERT_RTNL();
1308 
1309 	for_each_netdev(&init_net, dev) {
1310 		if (dev->netdev_ops != &device_ops)
1311 			continue;	/* not a netvsc device */
1312 
1313 		if (ether_addr_equal(mac, dev->perm_addr))
1314 			return dev;
1315 	}
1316 
1317 	return NULL;
1318 }
1319 
1320 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1321 {
1322 	struct net_device *dev;
1323 
1324 	ASSERT_RTNL();
1325 
1326 	for_each_netdev(&init_net, dev) {
1327 		struct net_device_context *net_device_ctx;
1328 
1329 		if (dev->netdev_ops != &device_ops)
1330 			continue;	/* not a netvsc device */
1331 
1332 		net_device_ctx = netdev_priv(dev);
1333 		if (net_device_ctx->nvdev == NULL)
1334 			continue;	/* device is removed */
1335 
1336 		if (rtnl_dereference(net_device_ctx->vf_netdev) == vf_netdev)
1337 			return dev;	/* a match */
1338 	}
1339 
1340 	return NULL;
1341 }
1342 
1343 static int netvsc_register_vf(struct net_device *vf_netdev)
1344 {
1345 	struct net_device *ndev;
1346 	struct net_device_context *net_device_ctx;
1347 	struct netvsc_device *netvsc_dev;
1348 
1349 	if (vf_netdev->addr_len != ETH_ALEN)
1350 		return NOTIFY_DONE;
1351 
1352 	/*
1353 	 * We will use the MAC address to locate the synthetic interface to
1354 	 * associate with the VF interface. If we don't find a matching
1355 	 * synthetic interface, move on.
1356 	 */
1357 	ndev = get_netvsc_bymac(vf_netdev->perm_addr);
1358 	if (!ndev)
1359 		return NOTIFY_DONE;
1360 
1361 	net_device_ctx = netdev_priv(ndev);
1362 	netvsc_dev = net_device_ctx->nvdev;
1363 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
1364 		return NOTIFY_DONE;
1365 
1366 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1367 	/*
1368 	 * Take a reference on the module.
1369 	 */
1370 	try_module_get(THIS_MODULE);
1371 
1372 	dev_hold(vf_netdev);
1373 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
1374 	return NOTIFY_OK;
1375 }
1376 
1377 static int netvsc_vf_up(struct net_device *vf_netdev)
1378 {
1379 	struct net_device *ndev;
1380 	struct netvsc_device *netvsc_dev;
1381 	struct net_device_context *net_device_ctx;
1382 
1383 	ndev = get_netvsc_byref(vf_netdev);
1384 	if (!ndev)
1385 		return NOTIFY_DONE;
1386 
1387 	net_device_ctx = netdev_priv(ndev);
1388 	netvsc_dev = net_device_ctx->nvdev;
1389 
1390 	netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1391 
1392 	/*
1393 	 * Open the device before switching data path.
1394 	 */
1395 	rndis_filter_open(netvsc_dev);
1396 
1397 	/*
1398 	 * notify the host to switch the data path.
1399 	 */
1400 	netvsc_switch_datapath(ndev, true);
1401 	netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1402 
1403 	netif_carrier_off(ndev);
1404 
1405 	/* Now notify peers through VF device. */
1406 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1407 
1408 	return NOTIFY_OK;
1409 }
1410 
1411 static int netvsc_vf_down(struct net_device *vf_netdev)
1412 {
1413 	struct net_device *ndev;
1414 	struct netvsc_device *netvsc_dev;
1415 	struct net_device_context *net_device_ctx;
1416 
1417 	ndev = get_netvsc_byref(vf_netdev);
1418 	if (!ndev)
1419 		return NOTIFY_DONE;
1420 
1421 	net_device_ctx = netdev_priv(ndev);
1422 	netvsc_dev = net_device_ctx->nvdev;
1423 
1424 	netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1425 	netvsc_switch_datapath(ndev, false);
1426 	netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1427 	rndis_filter_close(netvsc_dev);
1428 	netif_carrier_on(ndev);
1429 
1430 	/* Now notify peers through netvsc device. */
1431 	call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1432 
1433 	return NOTIFY_OK;
1434 }
1435 
1436 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1437 {
1438 	struct net_device *ndev;
1439 	struct net_device_context *net_device_ctx;
1440 
1441 	ndev = get_netvsc_byref(vf_netdev);
1442 	if (!ndev)
1443 		return NOTIFY_DONE;
1444 
1445 	net_device_ctx = netdev_priv(ndev);
1446 
1447 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1448 
1449 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
1450 	dev_put(vf_netdev);
1451 	module_put(THIS_MODULE);
1452 	return NOTIFY_OK;
1453 }
1454 
1455 static int netvsc_probe(struct hv_device *dev,
1456 			const struct hv_vmbus_device_id *dev_id)
1457 {
1458 	struct net_device *net = NULL;
1459 	struct net_device_context *net_device_ctx;
1460 	struct netvsc_device_info device_info;
1461 	struct netvsc_device *nvdev;
1462 	int ret;
1463 
1464 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
1465 				VRSS_CHANNEL_MAX);
1466 	if (!net)
1467 		return -ENOMEM;
1468 
1469 	netif_carrier_off(net);
1470 
1471 	netvsc_init_settings(net);
1472 
1473 	net_device_ctx = netdev_priv(net);
1474 	net_device_ctx->device_ctx = dev;
1475 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1476 	if (netif_msg_probe(net_device_ctx))
1477 		netdev_dbg(net, "netvsc msg_enable: %d\n",
1478 			   net_device_ctx->msg_enable);
1479 
1480 	hv_set_drvdata(dev, net);
1481 
1482 	net_device_ctx->start_remove = false;
1483 
1484 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1485 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
1486 
1487 	spin_lock_init(&net_device_ctx->lock);
1488 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1489 
1490 	net->netdev_ops = &device_ops;
1491 	net->ethtool_ops = &ethtool_ops;
1492 	SET_NETDEV_DEV(net, &dev->device);
1493 
1494 	/* We always need headroom for rndis header */
1495 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
1496 
1497 	/* Notify the netvsc driver of the new device */
1498 	memset(&device_info, 0, sizeof(device_info));
1499 	device_info.ring_size = ring_size;
1500 	device_info.max_num_vrss_chns = min_t(u32, VRSS_CHANNEL_DEFAULT,
1501 					      num_online_cpus());
1502 	ret = rndis_filter_device_add(dev, &device_info);
1503 	if (ret != 0) {
1504 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1505 		free_netdev(net);
1506 		hv_set_drvdata(dev, NULL);
1507 		return ret;
1508 	}
1509 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1510 
1511 	/* hw_features computed in rndis_filter_device_add */
1512 	net->features = net->hw_features |
1513 		NETIF_F_HIGHDMA | NETIF_F_SG |
1514 		NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
1515 	net->vlan_features = net->features;
1516 
1517 	nvdev = net_device_ctx->nvdev;
1518 	netif_set_real_num_tx_queues(net, nvdev->num_chn);
1519 	netif_set_real_num_rx_queues(net, nvdev->num_chn);
1520 
1521 	/* MTU range: 68 - 1500 or 65521 */
1522 	net->min_mtu = NETVSC_MTU_MIN;
1523 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
1524 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
1525 	else
1526 		net->max_mtu = ETH_DATA_LEN;
1527 
1528 	ret = register_netdev(net);
1529 	if (ret != 0) {
1530 		pr_err("Unable to register netdev.\n");
1531 		rndis_filter_device_remove(dev, nvdev);
1532 		free_netdev(net);
1533 	}
1534 
1535 	return ret;
1536 }
1537 
1538 static int netvsc_remove(struct hv_device *dev)
1539 {
1540 	struct net_device *net;
1541 	struct net_device_context *ndev_ctx;
1542 
1543 	net = hv_get_drvdata(dev);
1544 
1545 	if (net == NULL) {
1546 		dev_err(&dev->device, "No net device to remove\n");
1547 		return 0;
1548 	}
1549 
1550 	ndev_ctx = netdev_priv(net);
1551 
1552 	/* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1553 	 * removing the device.
1554 	 */
1555 	rtnl_lock();
1556 	ndev_ctx->start_remove = true;
1557 	rtnl_unlock();
1558 
1559 	cancel_delayed_work_sync(&ndev_ctx->dwork);
1560 	cancel_work_sync(&ndev_ctx->work);
1561 
1562 	/* Stop outbound asap */
1563 	netif_tx_disable(net);
1564 
1565 	unregister_netdev(net);
1566 
1567 	/*
1568 	 * Call to the vsc driver to let it know that the device is being
1569 	 * removed
1570 	 */
1571 	rndis_filter_device_remove(dev, ndev_ctx->nvdev);
1572 
1573 	hv_set_drvdata(dev, NULL);
1574 
1575 	free_netdev(net);
1576 	return 0;
1577 }
1578 
1579 static const struct hv_vmbus_device_id id_table[] = {
1580 	/* Network guid */
1581 	{ HV_NIC_GUID, },
1582 	{ },
1583 };
1584 
1585 MODULE_DEVICE_TABLE(vmbus, id_table);
1586 
1587 /* The one and only one */
1588 static struct  hv_driver netvsc_drv = {
1589 	.name = KBUILD_MODNAME,
1590 	.id_table = id_table,
1591 	.probe = netvsc_probe,
1592 	.remove = netvsc_remove,
1593 };
1594 
1595 /*
1596  * On Hyper-V, every VF interface is matched with a corresponding
1597  * synthetic interface. The synthetic interface is presented first
1598  * to the guest. When the corresponding VF instance is registered,
1599  * we will take care of switching the data path.
1600  */
1601 static int netvsc_netdev_event(struct notifier_block *this,
1602 			       unsigned long event, void *ptr)
1603 {
1604 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1605 
1606 	/* Skip our own events */
1607 	if (event_dev->netdev_ops == &device_ops)
1608 		return NOTIFY_DONE;
1609 
1610 	/* Avoid non-Ethernet type devices */
1611 	if (event_dev->type != ARPHRD_ETHER)
1612 		return NOTIFY_DONE;
1613 
1614 	/* Avoid Vlan dev with same MAC registering as VF */
1615 	if (is_vlan_dev(event_dev))
1616 		return NOTIFY_DONE;
1617 
1618 	/* Avoid Bonding master dev with same MAC registering as VF */
1619 	if ((event_dev->priv_flags & IFF_BONDING) &&
1620 	    (event_dev->flags & IFF_MASTER))
1621 		return NOTIFY_DONE;
1622 
1623 	switch (event) {
1624 	case NETDEV_REGISTER:
1625 		return netvsc_register_vf(event_dev);
1626 	case NETDEV_UNREGISTER:
1627 		return netvsc_unregister_vf(event_dev);
1628 	case NETDEV_UP:
1629 		return netvsc_vf_up(event_dev);
1630 	case NETDEV_DOWN:
1631 		return netvsc_vf_down(event_dev);
1632 	default:
1633 		return NOTIFY_DONE;
1634 	}
1635 }
1636 
1637 static struct notifier_block netvsc_netdev_notifier = {
1638 	.notifier_call = netvsc_netdev_event,
1639 };
1640 
1641 static void __exit netvsc_drv_exit(void)
1642 {
1643 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
1644 	vmbus_driver_unregister(&netvsc_drv);
1645 }
1646 
1647 static int __init netvsc_drv_init(void)
1648 {
1649 	int ret;
1650 
1651 	if (ring_size < RING_SIZE_MIN) {
1652 		ring_size = RING_SIZE_MIN;
1653 		pr_info("Increased ring_size to %d (min allowed)\n",
1654 			ring_size);
1655 	}
1656 	ret = vmbus_driver_register(&netvsc_drv);
1657 
1658 	if (ret)
1659 		return ret;
1660 
1661 	register_netdevice_notifier(&netvsc_netdev_notifier);
1662 	return 0;
1663 }
1664 
1665 MODULE_LICENSE("GPL");
1666 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1667 
1668 module_init(netvsc_drv_init);
1669 module_exit(netvsc_drv_exit);
1670