xref: /linux/drivers/net/hyperv/netvsc_drv.c (revision c3b999cad7ec39a5487b20e8b6e737d2ab0c5393)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2009, Microsoft Corporation.
4  *
5  * Authors:
6  *   Haiyang Zhang <haiyangz@microsoft.com>
7  *   Hank Janssen  <hjanssen@microsoft.com>
8  */
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 
11 #include <linux/init.h>
12 #include <linux/atomic.h>
13 #include <linux/ethtool.h>
14 #include <linux/module.h>
15 #include <linux/highmem.h>
16 #include <linux/device.h>
17 #include <linux/io.h>
18 #include <linux/delay.h>
19 #include <linux/netdevice.h>
20 #include <linux/inetdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/pci.h>
23 #include <linux/skbuff.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/slab.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/netpoll.h>
29 #include <linux/bpf.h>
30 
31 #include <net/arp.h>
32 #include <net/netdev_lock.h>
33 #include <net/route.h>
34 #include <net/sock.h>
35 #include <net/pkt_sched.h>
36 #include <net/checksum.h>
37 #include <net/ip6_checksum.h>
38 
39 #include "hyperv_net.h"
40 
41 #define RING_SIZE_MIN	64
42 
43 #define LINKCHANGE_INT (2 * HZ)
44 #define VF_TAKEOVER_INT (HZ / 10)
45 
46 /* Macros to define the context of vf registration */
47 #define VF_REG_IN_PROBE		1
48 #define VF_REG_IN_NOTIFIER	2
49 
50 static unsigned int ring_size __ro_after_init = 128;
51 module_param(ring_size, uint, 0444);
52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of 4K pages)");
53 unsigned int netvsc_ring_bytes __ro_after_init;
54 
55 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
56 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
57 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
58 				NETIF_MSG_TX_ERR;
59 
60 static int debug = -1;
61 module_param(debug, int, 0444);
62 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
63 
64 static LIST_HEAD(netvsc_dev_list);
65 
66 static void netvsc_change_rx_flags(struct net_device *net, int change)
67 {
68 	struct net_device_context *ndev_ctx = netdev_priv(net);
69 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
70 	int inc;
71 
72 	if (!vf_netdev)
73 		return;
74 
75 	if (change & IFF_PROMISC) {
76 		inc = (net->flags & IFF_PROMISC) ? 1 : -1;
77 		dev_set_promiscuity(vf_netdev, inc);
78 	}
79 
80 	if (change & IFF_ALLMULTI) {
81 		inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
82 		dev_set_allmulti(vf_netdev, inc);
83 	}
84 }
85 
86 static void netvsc_set_rx_mode(struct net_device *net)
87 {
88 	struct net_device_context *ndev_ctx = netdev_priv(net);
89 	struct net_device *vf_netdev;
90 	struct netvsc_device *nvdev;
91 
92 	rcu_read_lock();
93 	vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
94 	if (vf_netdev) {
95 		dev_uc_sync(vf_netdev, net);
96 		dev_mc_sync(vf_netdev, net);
97 	}
98 
99 	nvdev = rcu_dereference(ndev_ctx->nvdev);
100 	if (nvdev)
101 		rndis_filter_update(nvdev);
102 	rcu_read_unlock();
103 }
104 
105 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
106 			     struct net_device *ndev)
107 {
108 	nvscdev->tx_disable = false;
109 	virt_wmb(); /* ensure queue wake up mechanism is on */
110 
111 	netif_tx_wake_all_queues(ndev);
112 }
113 
114 static int netvsc_open(struct net_device *net)
115 {
116 	struct net_device_context *ndev_ctx = netdev_priv(net);
117 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
118 	struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
119 	struct rndis_device *rdev;
120 	int ret = 0;
121 
122 	netif_carrier_off(net);
123 
124 	/* Open up the device */
125 	ret = rndis_filter_open(nvdev);
126 	if (ret != 0) {
127 		netdev_err(net, "unable to open device (ret %d).\n", ret);
128 		return ret;
129 	}
130 
131 	rdev = nvdev->extension;
132 	if (!rdev->link_state) {
133 		netif_carrier_on(net);
134 		netvsc_tx_enable(nvdev, net);
135 	}
136 
137 	if (vf_netdev) {
138 		/* Setting synthetic device up transparently sets
139 		 * slave as up. If open fails, then slave will be
140 		 * still be offline (and not used).
141 		 */
142 		ret = dev_open(vf_netdev, NULL);
143 		if (ret)
144 			netdev_warn(net,
145 				    "unable to open slave: %s: %d\n",
146 				    vf_netdev->name, ret);
147 	}
148 	return 0;
149 }
150 
151 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
152 {
153 	unsigned int retry = 0;
154 	int i;
155 
156 	/* Ensure pending bytes in ring are read */
157 	for (;;) {
158 		u32 aread = 0;
159 
160 		for (i = 0; i < nvdev->num_chn; i++) {
161 			struct vmbus_channel *chn
162 				= nvdev->chan_table[i].channel;
163 
164 			if (!chn)
165 				continue;
166 
167 			/* make sure receive not running now */
168 			napi_synchronize(&nvdev->chan_table[i].napi);
169 
170 			aread = hv_get_bytes_to_read(&chn->inbound);
171 			if (aread)
172 				break;
173 
174 			aread = hv_get_bytes_to_read(&chn->outbound);
175 			if (aread)
176 				break;
177 		}
178 
179 		if (aread == 0)
180 			return 0;
181 
182 		if (++retry > RETRY_MAX)
183 			return -ETIMEDOUT;
184 
185 		usleep_range(RETRY_US_LO, RETRY_US_HI);
186 	}
187 }
188 
189 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
190 			      struct net_device *ndev)
191 {
192 	if (nvscdev) {
193 		nvscdev->tx_disable = true;
194 		virt_wmb(); /* ensure txq will not wake up after stop */
195 	}
196 
197 	netif_tx_disable(ndev);
198 }
199 
200 static int netvsc_close(struct net_device *net)
201 {
202 	struct net_device_context *net_device_ctx = netdev_priv(net);
203 	struct net_device *vf_netdev
204 		= rtnl_dereference(net_device_ctx->vf_netdev);
205 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
206 	int ret;
207 
208 	netvsc_tx_disable(nvdev, net);
209 
210 	/* No need to close rndis filter if it is removed already */
211 	if (!nvdev)
212 		return 0;
213 
214 	ret = rndis_filter_close(nvdev);
215 	if (ret != 0) {
216 		netdev_err(net, "unable to close device (ret %d).\n", ret);
217 		return ret;
218 	}
219 
220 	ret = netvsc_wait_until_empty(nvdev);
221 	if (ret)
222 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
223 
224 	if (vf_netdev)
225 		dev_close(vf_netdev);
226 
227 	return ret;
228 }
229 
230 static inline void *init_ppi_data(struct rndis_message *msg,
231 				  u32 ppi_size, u32 pkt_type)
232 {
233 	struct rndis_packet *rndis_pkt = &msg->msg.pkt;
234 	struct rndis_per_packet_info *ppi;
235 
236 	rndis_pkt->data_offset += ppi_size;
237 	ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
238 		+ rndis_pkt->per_pkt_info_len;
239 
240 	ppi->size = ppi_size;
241 	ppi->type = pkt_type;
242 	ppi->internal = 0;
243 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
244 
245 	rndis_pkt->per_pkt_info_len += ppi_size;
246 
247 	return ppi + 1;
248 }
249 
250 static inline int netvsc_get_tx_queue(struct net_device *ndev,
251 				      struct sk_buff *skb, int old_idx)
252 {
253 	const struct net_device_context *ndc = netdev_priv(ndev);
254 	struct sock *sk = skb->sk;
255 	int q_idx;
256 
257 	q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
258 			      (VRSS_SEND_TAB_SIZE - 1)];
259 
260 	/* If queue index changed record the new value */
261 	if (q_idx != old_idx &&
262 	    sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
263 		sk_tx_queue_set(sk, q_idx);
264 
265 	return q_idx;
266 }
267 
268 /*
269  * Select queue for transmit.
270  *
271  * If a valid queue has already been assigned, then use that.
272  * Otherwise compute tx queue based on hash and the send table.
273  *
274  * This is basically similar to default (netdev_pick_tx) with the added step
275  * of using the host send_table when no other queue has been assigned.
276  *
277  * TODO support XPS - but get_xps_queue not exported
278  */
279 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
280 {
281 	int q_idx = sk_tx_queue_get(skb->sk);
282 
283 	if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
284 		/* If forwarding a packet, we use the recorded queue when
285 		 * available for better cache locality.
286 		 */
287 		if (skb_rx_queue_recorded(skb))
288 			q_idx = skb_get_rx_queue(skb);
289 		else
290 			q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
291 	}
292 
293 	return q_idx;
294 }
295 
296 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
297 			       struct net_device *sb_dev)
298 {
299 	struct net_device_context *ndc = netdev_priv(ndev);
300 	struct net_device *vf_netdev;
301 	u16 txq;
302 
303 	rcu_read_lock();
304 	vf_netdev = rcu_dereference(ndc->vf_netdev);
305 	if (vf_netdev) {
306 		const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
307 
308 		if (vf_ops->ndo_select_queue)
309 			txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
310 		else
311 			txq = netdev_pick_tx(vf_netdev, skb, NULL);
312 
313 		/* Record the queue selected by VF so that it can be
314 		 * used for common case where VF has more queues than
315 		 * the synthetic device.
316 		 */
317 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
318 	} else {
319 		txq = netvsc_pick_tx(ndev, skb);
320 	}
321 	rcu_read_unlock();
322 
323 	while (txq >= ndev->real_num_tx_queues)
324 		txq -= ndev->real_num_tx_queues;
325 
326 	return txq;
327 }
328 
329 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
330 		       struct hv_page_buffer *pb)
331 {
332 	int j = 0;
333 
334 	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
335 	offset = offset & ~HV_HYP_PAGE_MASK;
336 
337 	while (len > 0) {
338 		unsigned long bytes;
339 
340 		bytes = HV_HYP_PAGE_SIZE - offset;
341 		if (bytes > len)
342 			bytes = len;
343 		pb[j].pfn = hvpfn;
344 		pb[j].offset = offset;
345 		pb[j].len = bytes;
346 
347 		offset += bytes;
348 		len -= bytes;
349 
350 		if (offset == HV_HYP_PAGE_SIZE && len) {
351 			hvpfn++;
352 			offset = 0;
353 			j++;
354 		}
355 	}
356 
357 	return j + 1;
358 }
359 
360 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
361 			   struct hv_netvsc_packet *packet,
362 			   struct hv_page_buffer *pb)
363 {
364 	u32 slots_used = 0;
365 	char *data = skb->data;
366 	int frags = skb_shinfo(skb)->nr_frags;
367 	int i;
368 
369 	/* The packet is laid out thus:
370 	 * 1. hdr: RNDIS header and PPI
371 	 * 2. skb linear data
372 	 * 3. skb fragment data
373 	 */
374 	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
375 				  offset_in_hvpage(hdr),
376 				  len,
377 				  &pb[slots_used]);
378 
379 	packet->rmsg_size = len;
380 	packet->rmsg_pgcnt = slots_used;
381 
382 	slots_used += fill_pg_buf(virt_to_hvpfn(data),
383 				  offset_in_hvpage(data),
384 				  skb_headlen(skb),
385 				  &pb[slots_used]);
386 
387 	for (i = 0; i < frags; i++) {
388 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
389 
390 		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
391 					  skb_frag_off(frag),
392 					  skb_frag_size(frag),
393 					  &pb[slots_used]);
394 	}
395 	return slots_used;
396 }
397 
398 static int count_skb_frag_slots(struct sk_buff *skb)
399 {
400 	int i, frags = skb_shinfo(skb)->nr_frags;
401 	int pages = 0;
402 
403 	for (i = 0; i < frags; i++) {
404 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
405 		unsigned long size = skb_frag_size(frag);
406 		unsigned long offset = skb_frag_off(frag);
407 
408 		/* Skip unused frames from start of page */
409 		offset &= ~HV_HYP_PAGE_MASK;
410 		pages += HVPFN_UP(offset + size);
411 	}
412 	return pages;
413 }
414 
415 static int netvsc_get_slots(struct sk_buff *skb)
416 {
417 	char *data = skb->data;
418 	unsigned int offset = offset_in_hvpage(data);
419 	unsigned int len = skb_headlen(skb);
420 	int slots;
421 	int frag_slots;
422 
423 	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
424 	frag_slots = count_skb_frag_slots(skb);
425 	return slots + frag_slots;
426 }
427 
428 static u32 net_checksum_info(struct sk_buff *skb)
429 {
430 	if (skb->protocol == htons(ETH_P_IP)) {
431 		struct iphdr *ip = ip_hdr(skb);
432 
433 		if (ip->protocol == IPPROTO_TCP)
434 			return TRANSPORT_INFO_IPV4_TCP;
435 		else if (ip->protocol == IPPROTO_UDP)
436 			return TRANSPORT_INFO_IPV4_UDP;
437 	} else {
438 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
439 
440 		if (ip6->nexthdr == IPPROTO_TCP)
441 			return TRANSPORT_INFO_IPV6_TCP;
442 		else if (ip6->nexthdr == IPPROTO_UDP)
443 			return TRANSPORT_INFO_IPV6_UDP;
444 	}
445 
446 	return TRANSPORT_INFO_NOT_IP;
447 }
448 
449 /* Send skb on the slave VF device. */
450 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
451 			  struct sk_buff *skb)
452 {
453 	struct net_device_context *ndev_ctx = netdev_priv(net);
454 	unsigned int len = skb->len;
455 	int rc;
456 
457 	skb->dev = vf_netdev;
458 	skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
459 
460 	rc = dev_queue_xmit(skb);
461 	if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
462 		struct netvsc_vf_pcpu_stats *pcpu_stats
463 			= this_cpu_ptr(ndev_ctx->vf_stats);
464 
465 		u64_stats_update_begin(&pcpu_stats->syncp);
466 		pcpu_stats->tx_packets++;
467 		pcpu_stats->tx_bytes += len;
468 		u64_stats_update_end(&pcpu_stats->syncp);
469 	} else {
470 		this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
471 	}
472 
473 	return rc;
474 }
475 
476 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
477 {
478 	struct net_device_context *net_device_ctx = netdev_priv(net);
479 	struct hv_netvsc_packet *packet = NULL;
480 	int ret;
481 	unsigned int num_data_pgs;
482 	struct rndis_message *rndis_msg;
483 	struct net_device *vf_netdev;
484 	u32 rndis_msg_size;
485 	u32 hash;
486 	struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
487 
488 	/* If VF is present and up then redirect packets to it.
489 	 * Skip the VF if it is marked down or has no carrier.
490 	 * If netpoll is in uses, then VF can not be used either.
491 	 */
492 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
493 	if (vf_netdev && netif_running(vf_netdev) &&
494 	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
495 	    net_device_ctx->data_path_is_vf)
496 		return netvsc_vf_xmit(net, vf_netdev, skb);
497 
498 	/* We will atmost need two pages to describe the rndis
499 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
500 	 * of pages in a single packet. If skb is scattered around
501 	 * more pages we try linearizing it.
502 	 */
503 
504 	num_data_pgs = netvsc_get_slots(skb) + 2;
505 
506 	if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
507 		++net_device_ctx->eth_stats.tx_scattered;
508 
509 		if (skb_linearize(skb))
510 			goto no_memory;
511 
512 		num_data_pgs = netvsc_get_slots(skb) + 2;
513 		if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
514 			++net_device_ctx->eth_stats.tx_too_big;
515 			goto drop;
516 		}
517 	}
518 
519 	/*
520 	 * Place the rndis header in the skb head room and
521 	 * the skb->cb will be used for hv_netvsc_packet
522 	 * structure.
523 	 */
524 	ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
525 	if (ret)
526 		goto no_memory;
527 
528 	/* Use the skb control buffer for building up the packet */
529 	BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
530 			sizeof_field(struct sk_buff, cb));
531 	packet = (struct hv_netvsc_packet *)skb->cb;
532 
533 	packet->q_idx = skb_get_queue_mapping(skb);
534 
535 	packet->total_data_buflen = skb->len;
536 	packet->total_bytes = skb->len;
537 	packet->total_packets = 1;
538 
539 	rndis_msg = (struct rndis_message *)skb->head;
540 
541 	/* Add the rndis header */
542 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
543 	rndis_msg->msg_len = packet->total_data_buflen;
544 
545 	rndis_msg->msg.pkt = (struct rndis_packet) {
546 		.data_offset = sizeof(struct rndis_packet),
547 		.data_len = packet->total_data_buflen,
548 		.per_pkt_info_offset = sizeof(struct rndis_packet),
549 	};
550 
551 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
552 
553 	hash = skb_get_hash_raw(skb);
554 	if (hash != 0 && net->real_num_tx_queues > 1) {
555 		u32 *hash_info;
556 
557 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
558 		hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
559 					  NBL_HASH_VALUE);
560 		*hash_info = hash;
561 	}
562 
563 	/* When using AF_PACKET we need to drop VLAN header from
564 	 * the frame and update the SKB to allow the HOST OS
565 	 * to transmit the 802.1Q packet
566 	 */
567 	if (skb->protocol == htons(ETH_P_8021Q)) {
568 		u16 vlan_tci;
569 
570 		skb_reset_mac_header(skb);
571 		if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
572 			if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
573 				++net_device_ctx->eth_stats.vlan_error;
574 				goto drop;
575 			}
576 
577 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
578 			/* Update the NDIS header pkt lengths */
579 			packet->total_data_buflen -= VLAN_HLEN;
580 			packet->total_bytes -= VLAN_HLEN;
581 			rndis_msg->msg_len = packet->total_data_buflen;
582 			rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
583 		}
584 	}
585 
586 	if (skb_vlan_tag_present(skb)) {
587 		struct ndis_pkt_8021q_info *vlan;
588 
589 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
590 		vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
591 				     IEEE_8021Q_INFO);
592 
593 		vlan->value = 0;
594 		vlan->vlanid = skb_vlan_tag_get_id(skb);
595 		vlan->cfi = skb_vlan_tag_get_cfi(skb);
596 		vlan->pri = skb_vlan_tag_get_prio(skb);
597 	}
598 
599 	if (skb_is_gso(skb)) {
600 		struct ndis_tcp_lso_info *lso_info;
601 
602 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
603 		lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
604 					 TCP_LARGESEND_PKTINFO);
605 
606 		lso_info->value = 0;
607 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
608 		if (skb->protocol == htons(ETH_P_IP)) {
609 			lso_info->lso_v2_transmit.ip_version =
610 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
611 			ip_hdr(skb)->tot_len = 0;
612 			ip_hdr(skb)->check = 0;
613 			tcp_hdr(skb)->check =
614 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
615 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
616 		} else {
617 			lso_info->lso_v2_transmit.ip_version =
618 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
619 			tcp_v6_gso_csum_prep(skb);
620 		}
621 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
622 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
623 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
624 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
625 			struct ndis_tcp_ip_checksum_info *csum_info;
626 
627 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
628 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
629 						  TCPIP_CHKSUM_PKTINFO);
630 
631 			csum_info->value = 0;
632 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
633 
634 			if (skb->protocol == htons(ETH_P_IP)) {
635 				csum_info->transmit.is_ipv4 = 1;
636 
637 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
638 					csum_info->transmit.tcp_checksum = 1;
639 				else
640 					csum_info->transmit.udp_checksum = 1;
641 			} else {
642 				csum_info->transmit.is_ipv6 = 1;
643 
644 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
645 					csum_info->transmit.tcp_checksum = 1;
646 				else
647 					csum_info->transmit.udp_checksum = 1;
648 			}
649 		} else {
650 			/* Can't do offload of this type of checksum */
651 			if (skb_checksum_help(skb))
652 				goto drop;
653 		}
654 	}
655 
656 	/* Start filling in the page buffers with the rndis hdr */
657 	rndis_msg->msg_len += rndis_msg_size;
658 	packet->total_data_buflen = rndis_msg->msg_len;
659 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
660 					       skb, packet, pb);
661 
662 	/* timestamp packet in software */
663 	skb_tx_timestamp(skb);
664 
665 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
666 	if (likely(ret == 0))
667 		return NETDEV_TX_OK;
668 
669 	if (ret == -EAGAIN) {
670 		++net_device_ctx->eth_stats.tx_busy;
671 		return NETDEV_TX_BUSY;
672 	}
673 
674 	if (ret == -ENOSPC)
675 		++net_device_ctx->eth_stats.tx_no_space;
676 
677 drop:
678 	dev_kfree_skb_any(skb);
679 	net->stats.tx_dropped++;
680 
681 	return NETDEV_TX_OK;
682 
683 no_memory:
684 	++net_device_ctx->eth_stats.tx_no_memory;
685 	goto drop;
686 }
687 
688 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
689 				     struct net_device *ndev)
690 {
691 	return netvsc_xmit(skb, ndev, false);
692 }
693 
694 /*
695  * netvsc_linkstatus_callback - Link up/down notification
696  */
697 void netvsc_linkstatus_callback(struct net_device *net,
698 				struct rndis_message *resp,
699 				void *data, u32 data_buflen)
700 {
701 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
702 	struct net_device_context *ndev_ctx = netdev_priv(net);
703 	struct netvsc_reconfig *event;
704 	unsigned long flags;
705 
706 	/* Ensure the packet is big enough to access its fields */
707 	if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
708 		netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
709 			   resp->msg_len);
710 		return;
711 	}
712 
713 	/* Copy the RNDIS indicate status into nvchan->recv_buf */
714 	memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
715 
716 	/* Update the physical link speed when changing to another vSwitch */
717 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
718 		u32 speed;
719 
720 		/* Validate status_buf_offset and status_buflen.
721 		 *
722 		 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
723 		 * for the status buffer field in resp->msg_len; perform the validation
724 		 * using data_buflen (>= resp->msg_len).
725 		 */
726 		if (indicate->status_buflen < sizeof(speed) ||
727 		    indicate->status_buf_offset < sizeof(*indicate) ||
728 		    data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
729 		    data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
730 				< indicate->status_buflen) {
731 			netdev_err(net, "invalid rndis_indicate_status packet\n");
732 			return;
733 		}
734 
735 		speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
736 		ndev_ctx->speed = speed;
737 		return;
738 	}
739 
740 	/* Handle these link change statuses below */
741 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
742 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
743 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
744 		return;
745 
746 	if (net->reg_state != NETREG_REGISTERED)
747 		return;
748 
749 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
750 	if (!event)
751 		return;
752 	event->event = indicate->status;
753 
754 	spin_lock_irqsave(&ndev_ctx->lock, flags);
755 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
756 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
757 
758 	schedule_delayed_work(&ndev_ctx->dwork, 0);
759 }
760 
761 /* This function should only be called after skb_record_rx_queue() */
762 void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
763 {
764 	int rc;
765 
766 	skb->queue_mapping = skb_get_rx_queue(skb);
767 	__skb_push(skb, ETH_HLEN);
768 
769 	rc = netvsc_xmit(skb, ndev, true);
770 
771 	if (dev_xmit_complete(rc))
772 		return;
773 
774 	dev_kfree_skb_any(skb);
775 	ndev->stats.tx_dropped++;
776 }
777 
778 static void netvsc_comp_ipcsum(struct sk_buff *skb)
779 {
780 	struct iphdr *iph = (struct iphdr *)skb->data;
781 
782 	iph->check = 0;
783 	iph->check = ip_fast_csum(iph, iph->ihl);
784 }
785 
786 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
787 					     struct netvsc_channel *nvchan,
788 					     struct xdp_buff *xdp)
789 {
790 	struct napi_struct *napi = &nvchan->napi;
791 	const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
792 	const struct ndis_tcp_ip_checksum_info *csum_info =
793 						&nvchan->rsc.csum_info;
794 	const u32 *hash_info = &nvchan->rsc.hash_info;
795 	u8 ppi_flags = nvchan->rsc.ppi_flags;
796 	struct sk_buff *skb;
797 	void *xbuf = xdp->data_hard_start;
798 	int i;
799 
800 	if (xbuf) {
801 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
802 		unsigned int xlen = xdp->data_end - xdp->data;
803 		unsigned int frag_size = xdp->frame_sz;
804 
805 		skb = build_skb(xbuf, frag_size);
806 
807 		if (!skb) {
808 			__free_page(virt_to_page(xbuf));
809 			return NULL;
810 		}
811 
812 		skb_reserve(skb, hdroom);
813 		skb_put(skb, xlen);
814 		skb->dev = napi->dev;
815 	} else {
816 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
817 
818 		if (!skb)
819 			return NULL;
820 
821 		/* Copy to skb. This copy is needed here since the memory
822 		 * pointed by hv_netvsc_packet cannot be deallocated.
823 		 */
824 		for (i = 0; i < nvchan->rsc.cnt; i++)
825 			skb_put_data(skb, nvchan->rsc.data[i],
826 				     nvchan->rsc.len[i]);
827 	}
828 
829 	skb->protocol = eth_type_trans(skb, net);
830 
831 	/* skb is already created with CHECKSUM_NONE */
832 	skb_checksum_none_assert(skb);
833 
834 	/* Incoming packets may have IP header checksum verified by the host.
835 	 * They may not have IP header checksum computed after coalescing.
836 	 * We compute it here if the flags are set, because on Linux, the IP
837 	 * checksum is always checked.
838 	 */
839 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
840 	    csum_info->receive.ip_checksum_succeeded &&
841 	    skb->protocol == htons(ETH_P_IP)) {
842 		/* Check that there is enough space to hold the IP header. */
843 		if (skb_headlen(skb) < sizeof(struct iphdr)) {
844 			kfree_skb(skb);
845 			return NULL;
846 		}
847 		netvsc_comp_ipcsum(skb);
848 	}
849 
850 	/* Do L4 checksum offload if enabled and present. */
851 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
852 		if (csum_info->receive.tcp_checksum_succeeded ||
853 		    csum_info->receive.udp_checksum_succeeded)
854 			skb->ip_summed = CHECKSUM_UNNECESSARY;
855 	}
856 
857 	if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
858 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
859 
860 	if (ppi_flags & NVSC_RSC_VLAN) {
861 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
862 			(vlan->cfi ? VLAN_CFI_MASK : 0);
863 
864 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
865 				       vlan_tci);
866 	}
867 
868 	return skb;
869 }
870 
871 /*
872  * netvsc_recv_callback -  Callback when we receive a packet from the
873  * "wire" on the specified device.
874  */
875 int netvsc_recv_callback(struct net_device *net,
876 			 struct netvsc_device *net_device,
877 			 struct netvsc_channel *nvchan)
878 {
879 	struct net_device_context *net_device_ctx = netdev_priv(net);
880 	struct vmbus_channel *channel = nvchan->channel;
881 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
882 	struct sk_buff *skb;
883 	struct netvsc_stats_rx *rx_stats = &nvchan->rx_stats;
884 	struct xdp_buff xdp;
885 	u32 act;
886 
887 	if (net->reg_state != NETREG_REGISTERED)
888 		return NVSP_STAT_FAIL;
889 
890 	act = netvsc_run_xdp(net, nvchan, &xdp);
891 
892 	if (act == XDP_REDIRECT)
893 		return NVSP_STAT_SUCCESS;
894 
895 	if (act != XDP_PASS && act != XDP_TX) {
896 		u64_stats_update_begin(&rx_stats->syncp);
897 		rx_stats->xdp_drop++;
898 		u64_stats_update_end(&rx_stats->syncp);
899 
900 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
901 	}
902 
903 	/* Allocate a skb - TODO direct I/O to pages? */
904 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
905 
906 	if (unlikely(!skb)) {
907 		++net_device_ctx->eth_stats.rx_no_memory;
908 		return NVSP_STAT_FAIL;
909 	}
910 
911 	skb_record_rx_queue(skb, q_idx);
912 
913 	/*
914 	 * Even if injecting the packet, record the statistics
915 	 * on the synthetic device because modifying the VF device
916 	 * statistics will not work correctly.
917 	 */
918 	u64_stats_update_begin(&rx_stats->syncp);
919 	if (act == XDP_TX)
920 		rx_stats->xdp_tx++;
921 
922 	rx_stats->packets++;
923 	rx_stats->bytes += nvchan->rsc.pktlen;
924 
925 	if (skb->pkt_type == PACKET_BROADCAST)
926 		++rx_stats->broadcast;
927 	else if (skb->pkt_type == PACKET_MULTICAST)
928 		++rx_stats->multicast;
929 	u64_stats_update_end(&rx_stats->syncp);
930 
931 	if (act == XDP_TX) {
932 		netvsc_xdp_xmit(skb, net);
933 		return NVSP_STAT_SUCCESS;
934 	}
935 
936 	napi_gro_receive(&nvchan->napi, skb);
937 	return NVSP_STAT_SUCCESS;
938 }
939 
940 static void netvsc_get_drvinfo(struct net_device *net,
941 			       struct ethtool_drvinfo *info)
942 {
943 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
944 	strscpy(info->fw_version, "N/A", sizeof(info->fw_version));
945 }
946 
947 static void netvsc_get_channels(struct net_device *net,
948 				struct ethtool_channels *channel)
949 {
950 	struct net_device_context *net_device_ctx = netdev_priv(net);
951 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
952 
953 	if (nvdev) {
954 		channel->max_combined	= nvdev->max_chn;
955 		channel->combined_count = nvdev->num_chn;
956 	}
957 }
958 
959 /* Alloc struct netvsc_device_info, and initialize it from either existing
960  * struct netvsc_device, or from default values.
961  */
962 static
963 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
964 {
965 	struct netvsc_device_info *dev_info;
966 	struct bpf_prog *prog;
967 
968 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
969 
970 	if (!dev_info)
971 		return NULL;
972 
973 	if (nvdev) {
974 		ASSERT_RTNL();
975 
976 		dev_info->num_chn = nvdev->num_chn;
977 		dev_info->send_sections = nvdev->send_section_cnt;
978 		dev_info->send_section_size = nvdev->send_section_size;
979 		dev_info->recv_sections = nvdev->recv_section_cnt;
980 		dev_info->recv_section_size = nvdev->recv_section_size;
981 
982 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
983 		       NETVSC_HASH_KEYLEN);
984 
985 		prog = netvsc_xdp_get(nvdev);
986 		if (prog) {
987 			bpf_prog_inc(prog);
988 			dev_info->bprog = prog;
989 		}
990 	} else {
991 		dev_info->num_chn = max(VRSS_CHANNEL_DEFAULT,
992 					netif_get_num_default_rss_queues());
993 		dev_info->send_sections = NETVSC_DEFAULT_TX;
994 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
995 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
996 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
997 	}
998 
999 	return dev_info;
1000 }
1001 
1002 /* Free struct netvsc_device_info */
1003 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1004 {
1005 	if (dev_info->bprog) {
1006 		ASSERT_RTNL();
1007 		bpf_prog_put(dev_info->bprog);
1008 	}
1009 
1010 	kfree(dev_info);
1011 }
1012 
1013 static int netvsc_detach(struct net_device *ndev,
1014 			 struct netvsc_device *nvdev)
1015 {
1016 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1017 	struct hv_device *hdev = ndev_ctx->device_ctx;
1018 	int ret;
1019 
1020 	/* Don't try continuing to try and setup sub channels */
1021 	if (cancel_work_sync(&nvdev->subchan_work))
1022 		nvdev->num_chn = 1;
1023 
1024 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1025 
1026 	/* If device was up (receiving) then shutdown */
1027 	if (netif_running(ndev)) {
1028 		netvsc_tx_disable(nvdev, ndev);
1029 
1030 		ret = rndis_filter_close(nvdev);
1031 		if (ret) {
1032 			netdev_err(ndev,
1033 				   "unable to close device (ret %d).\n", ret);
1034 			return ret;
1035 		}
1036 
1037 		ret = netvsc_wait_until_empty(nvdev);
1038 		if (ret) {
1039 			netdev_err(ndev,
1040 				   "Ring buffer not empty after closing rndis\n");
1041 			return ret;
1042 		}
1043 	}
1044 
1045 	netif_device_detach(ndev);
1046 
1047 	rndis_filter_device_remove(hdev, nvdev);
1048 
1049 	return 0;
1050 }
1051 
1052 static int netvsc_attach(struct net_device *ndev,
1053 			 struct netvsc_device_info *dev_info)
1054 {
1055 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1056 	struct hv_device *hdev = ndev_ctx->device_ctx;
1057 	struct netvsc_device *nvdev;
1058 	struct rndis_device *rdev;
1059 	struct bpf_prog *prog;
1060 	int ret = 0;
1061 
1062 	nvdev = rndis_filter_device_add(hdev, dev_info);
1063 	if (IS_ERR(nvdev))
1064 		return PTR_ERR(nvdev);
1065 
1066 	if (nvdev->num_chn > 1) {
1067 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1068 
1069 		/* if unavailable, just proceed with one queue */
1070 		if (ret) {
1071 			nvdev->max_chn = 1;
1072 			nvdev->num_chn = 1;
1073 		}
1074 	}
1075 
1076 	prog = dev_info->bprog;
1077 	if (prog) {
1078 		bpf_prog_inc(prog);
1079 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1080 		if (ret) {
1081 			bpf_prog_put(prog);
1082 			goto err1;
1083 		}
1084 	}
1085 
1086 	/* In any case device is now ready */
1087 	nvdev->tx_disable = false;
1088 	netif_device_attach(ndev);
1089 
1090 	/* Note: enable and attach happen when sub-channels setup */
1091 	netif_carrier_off(ndev);
1092 
1093 	if (netif_running(ndev)) {
1094 		ret = rndis_filter_open(nvdev);
1095 		if (ret)
1096 			goto err2;
1097 
1098 		rdev = nvdev->extension;
1099 		if (!rdev->link_state)
1100 			netif_carrier_on(ndev);
1101 	}
1102 
1103 	return 0;
1104 
1105 err2:
1106 	netif_device_detach(ndev);
1107 
1108 err1:
1109 	rndis_filter_device_remove(hdev, nvdev);
1110 
1111 	return ret;
1112 }
1113 
1114 static int netvsc_set_channels(struct net_device *net,
1115 			       struct ethtool_channels *channels)
1116 {
1117 	struct net_device_context *net_device_ctx = netdev_priv(net);
1118 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1119 	unsigned int orig, count = channels->combined_count;
1120 	struct netvsc_device_info *device_info;
1121 	int ret;
1122 
1123 	/* We do not support separate count for rx, tx, or other */
1124 	if (count == 0 ||
1125 	    channels->rx_count || channels->tx_count || channels->other_count)
1126 		return -EINVAL;
1127 
1128 	if (!nvdev || nvdev->destroy)
1129 		return -ENODEV;
1130 
1131 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1132 		return -EINVAL;
1133 
1134 	if (count > nvdev->max_chn)
1135 		return -EINVAL;
1136 
1137 	orig = nvdev->num_chn;
1138 
1139 	device_info = netvsc_devinfo_get(nvdev);
1140 
1141 	if (!device_info)
1142 		return -ENOMEM;
1143 
1144 	device_info->num_chn = count;
1145 
1146 	ret = netvsc_detach(net, nvdev);
1147 	if (ret)
1148 		goto out;
1149 
1150 	ret = netvsc_attach(net, device_info);
1151 	if (ret) {
1152 		device_info->num_chn = orig;
1153 		if (netvsc_attach(net, device_info))
1154 			netdev_err(net, "restoring channel setting failed\n");
1155 	}
1156 
1157 out:
1158 	netvsc_devinfo_put(device_info);
1159 	return ret;
1160 }
1161 
1162 static void netvsc_init_settings(struct net_device *dev)
1163 {
1164 	struct net_device_context *ndc = netdev_priv(dev);
1165 
1166 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1167 
1168 	ndc->speed = SPEED_UNKNOWN;
1169 	ndc->duplex = DUPLEX_FULL;
1170 
1171 	dev->features = NETIF_F_LRO;
1172 }
1173 
1174 static int netvsc_get_link_ksettings(struct net_device *dev,
1175 				     struct ethtool_link_ksettings *cmd)
1176 {
1177 	struct net_device_context *ndc = netdev_priv(dev);
1178 	struct net_device *vf_netdev;
1179 
1180 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1181 
1182 	if (vf_netdev)
1183 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
1184 
1185 	cmd->base.speed = ndc->speed;
1186 	cmd->base.duplex = ndc->duplex;
1187 	cmd->base.port = PORT_OTHER;
1188 
1189 	return 0;
1190 }
1191 
1192 static int netvsc_set_link_ksettings(struct net_device *dev,
1193 				     const struct ethtool_link_ksettings *cmd)
1194 {
1195 	struct net_device_context *ndc = netdev_priv(dev);
1196 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1197 
1198 	if (vf_netdev) {
1199 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1200 			return -EOPNOTSUPP;
1201 
1202 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1203 								  cmd);
1204 	}
1205 
1206 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1207 						  &ndc->speed, &ndc->duplex);
1208 }
1209 
1210 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1211 {
1212 	struct net_device_context *ndevctx = netdev_priv(ndev);
1213 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1214 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1215 	int orig_mtu = ndev->mtu;
1216 	struct netvsc_device_info *device_info;
1217 	int ret = 0;
1218 
1219 	if (!nvdev || nvdev->destroy)
1220 		return -ENODEV;
1221 
1222 	device_info = netvsc_devinfo_get(nvdev);
1223 
1224 	if (!device_info)
1225 		return -ENOMEM;
1226 
1227 	/* Change MTU of underlying VF netdev first. */
1228 	if (vf_netdev) {
1229 		ret = dev_set_mtu(vf_netdev, mtu);
1230 		if (ret)
1231 			goto out;
1232 	}
1233 
1234 	ret = netvsc_detach(ndev, nvdev);
1235 	if (ret)
1236 		goto rollback_vf;
1237 
1238 	WRITE_ONCE(ndev->mtu, mtu);
1239 
1240 	ret = netvsc_attach(ndev, device_info);
1241 	if (!ret)
1242 		goto out;
1243 
1244 	/* Attempt rollback to original MTU */
1245 	WRITE_ONCE(ndev->mtu, orig_mtu);
1246 
1247 	if (netvsc_attach(ndev, device_info))
1248 		netdev_err(ndev, "restoring mtu failed\n");
1249 rollback_vf:
1250 	if (vf_netdev)
1251 		dev_set_mtu(vf_netdev, orig_mtu);
1252 
1253 out:
1254 	netvsc_devinfo_put(device_info);
1255 	return ret;
1256 }
1257 
1258 static void netvsc_get_vf_stats(struct net_device *net,
1259 				struct netvsc_vf_pcpu_stats *tot)
1260 {
1261 	struct net_device_context *ndev_ctx = netdev_priv(net);
1262 	int i;
1263 
1264 	memset(tot, 0, sizeof(*tot));
1265 
1266 	for_each_possible_cpu(i) {
1267 		const struct netvsc_vf_pcpu_stats *stats
1268 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1269 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1270 		unsigned int start;
1271 
1272 		do {
1273 			start = u64_stats_fetch_begin(&stats->syncp);
1274 			rx_packets = stats->rx_packets;
1275 			tx_packets = stats->tx_packets;
1276 			rx_bytes = stats->rx_bytes;
1277 			tx_bytes = stats->tx_bytes;
1278 		} while (u64_stats_fetch_retry(&stats->syncp, start));
1279 
1280 		tot->rx_packets += rx_packets;
1281 		tot->tx_packets += tx_packets;
1282 		tot->rx_bytes   += rx_bytes;
1283 		tot->tx_bytes   += tx_bytes;
1284 		tot->tx_dropped += stats->tx_dropped;
1285 	}
1286 }
1287 
1288 static void netvsc_get_pcpu_stats(struct net_device *net,
1289 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1290 {
1291 	struct net_device_context *ndev_ctx = netdev_priv(net);
1292 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1293 	int i;
1294 
1295 	/* fetch percpu stats of vf */
1296 	for_each_possible_cpu(i) {
1297 		const struct netvsc_vf_pcpu_stats *stats =
1298 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1299 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1300 		unsigned int start;
1301 
1302 		do {
1303 			start = u64_stats_fetch_begin(&stats->syncp);
1304 			this_tot->vf_rx_packets = stats->rx_packets;
1305 			this_tot->vf_tx_packets = stats->tx_packets;
1306 			this_tot->vf_rx_bytes = stats->rx_bytes;
1307 			this_tot->vf_tx_bytes = stats->tx_bytes;
1308 		} while (u64_stats_fetch_retry(&stats->syncp, start));
1309 		this_tot->rx_packets = this_tot->vf_rx_packets;
1310 		this_tot->tx_packets = this_tot->vf_tx_packets;
1311 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1312 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1313 	}
1314 
1315 	/* fetch percpu stats of netvsc */
1316 	for (i = 0; i < nvdev->num_chn; i++) {
1317 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1318 		const struct netvsc_stats_tx *tx_stats;
1319 		const struct netvsc_stats_rx *rx_stats;
1320 		struct netvsc_ethtool_pcpu_stats *this_tot =
1321 			&pcpu_tot[nvchan->channel->target_cpu];
1322 		u64 packets, bytes;
1323 		unsigned int start;
1324 
1325 		tx_stats = &nvchan->tx_stats;
1326 		do {
1327 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1328 			packets = tx_stats->packets;
1329 			bytes = tx_stats->bytes;
1330 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1331 
1332 		this_tot->tx_bytes	+= bytes;
1333 		this_tot->tx_packets	+= packets;
1334 
1335 		rx_stats = &nvchan->rx_stats;
1336 		do {
1337 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1338 			packets = rx_stats->packets;
1339 			bytes = rx_stats->bytes;
1340 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1341 
1342 		this_tot->rx_bytes	+= bytes;
1343 		this_tot->rx_packets	+= packets;
1344 	}
1345 }
1346 
1347 static void netvsc_get_stats64(struct net_device *net,
1348 			       struct rtnl_link_stats64 *t)
1349 {
1350 	struct net_device_context *ndev_ctx = netdev_priv(net);
1351 	struct netvsc_device *nvdev;
1352 	struct netvsc_vf_pcpu_stats vf_tot;
1353 	int i;
1354 
1355 	rcu_read_lock();
1356 
1357 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1358 	if (!nvdev)
1359 		goto out;
1360 
1361 	netdev_stats_to_stats64(t, &net->stats);
1362 
1363 	netvsc_get_vf_stats(net, &vf_tot);
1364 	t->rx_packets += vf_tot.rx_packets;
1365 	t->tx_packets += vf_tot.tx_packets;
1366 	t->rx_bytes   += vf_tot.rx_bytes;
1367 	t->tx_bytes   += vf_tot.tx_bytes;
1368 	t->tx_dropped += vf_tot.tx_dropped;
1369 
1370 	for (i = 0; i < nvdev->num_chn; i++) {
1371 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1372 		const struct netvsc_stats_tx *tx_stats;
1373 		const struct netvsc_stats_rx *rx_stats;
1374 		u64 packets, bytes, multicast;
1375 		unsigned int start;
1376 
1377 		tx_stats = &nvchan->tx_stats;
1378 		do {
1379 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1380 			packets = tx_stats->packets;
1381 			bytes = tx_stats->bytes;
1382 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1383 
1384 		t->tx_bytes	+= bytes;
1385 		t->tx_packets	+= packets;
1386 
1387 		rx_stats = &nvchan->rx_stats;
1388 		do {
1389 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1390 			packets = rx_stats->packets;
1391 			bytes = rx_stats->bytes;
1392 			multicast = rx_stats->multicast + rx_stats->broadcast;
1393 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1394 
1395 		t->rx_bytes	+= bytes;
1396 		t->rx_packets	+= packets;
1397 		t->multicast	+= multicast;
1398 	}
1399 out:
1400 	rcu_read_unlock();
1401 }
1402 
1403 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1404 {
1405 	struct net_device_context *ndc = netdev_priv(ndev);
1406 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1407 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1408 	struct sockaddr *addr = p;
1409 	int err;
1410 
1411 	err = eth_prepare_mac_addr_change(ndev, p);
1412 	if (err)
1413 		return err;
1414 
1415 	if (!nvdev)
1416 		return -ENODEV;
1417 
1418 	if (vf_netdev) {
1419 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1420 		if (err)
1421 			return err;
1422 	}
1423 
1424 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1425 	if (!err) {
1426 		eth_commit_mac_addr_change(ndev, p);
1427 	} else if (vf_netdev) {
1428 		/* rollback change on VF */
1429 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1430 		dev_set_mac_address(vf_netdev, addr, NULL);
1431 	}
1432 
1433 	return err;
1434 }
1435 
1436 static const struct {
1437 	char name[ETH_GSTRING_LEN];
1438 	u16 offset;
1439 } netvsc_stats[] = {
1440 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1441 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1442 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1443 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1444 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1445 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1446 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1447 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1448 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1449 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1450 	{ "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1451 }, pcpu_stats[] = {
1452 	{ "cpu%u_rx_packets",
1453 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1454 	{ "cpu%u_rx_bytes",
1455 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1456 	{ "cpu%u_tx_packets",
1457 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1458 	{ "cpu%u_tx_bytes",
1459 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1460 	{ "cpu%u_vf_rx_packets",
1461 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1462 	{ "cpu%u_vf_rx_bytes",
1463 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1464 	{ "cpu%u_vf_tx_packets",
1465 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1466 	{ "cpu%u_vf_tx_bytes",
1467 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1468 }, vf_stats[] = {
1469 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1470 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1471 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1472 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1473 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1474 };
1475 
1476 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1477 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1478 
1479 /* statistics per queue (rx/tx packets/bytes) */
1480 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1481 
1482 /* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1483 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
1484 
1485 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1486 {
1487 	struct net_device_context *ndc = netdev_priv(dev);
1488 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1489 
1490 	if (!nvdev)
1491 		return -ENODEV;
1492 
1493 	switch (string_set) {
1494 	case ETH_SS_STATS:
1495 		return NETVSC_GLOBAL_STATS_LEN
1496 			+ NETVSC_VF_STATS_LEN
1497 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1498 			+ NETVSC_PCPU_STATS_LEN;
1499 	default:
1500 		return -EINVAL;
1501 	}
1502 }
1503 
1504 static void netvsc_get_ethtool_stats(struct net_device *dev,
1505 				     struct ethtool_stats *stats, u64 *data)
1506 {
1507 	struct net_device_context *ndc = netdev_priv(dev);
1508 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1509 	const void *nds = &ndc->eth_stats;
1510 	const struct netvsc_stats_tx *tx_stats;
1511 	const struct netvsc_stats_rx *rx_stats;
1512 	struct netvsc_vf_pcpu_stats sum;
1513 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1514 	unsigned int start;
1515 	u64 packets, bytes;
1516 	u64 xdp_drop;
1517 	u64 xdp_redirect;
1518 	u64 xdp_tx;
1519 	u64 xdp_xmit;
1520 	int i, j, cpu;
1521 
1522 	if (!nvdev)
1523 		return;
1524 
1525 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1526 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1527 
1528 	netvsc_get_vf_stats(dev, &sum);
1529 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1530 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1531 
1532 	for (j = 0; j < nvdev->num_chn; j++) {
1533 		tx_stats = &nvdev->chan_table[j].tx_stats;
1534 
1535 		do {
1536 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1537 			packets = tx_stats->packets;
1538 			bytes = tx_stats->bytes;
1539 			xdp_xmit = tx_stats->xdp_xmit;
1540 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1541 		data[i++] = packets;
1542 		data[i++] = bytes;
1543 		data[i++] = xdp_xmit;
1544 
1545 		rx_stats = &nvdev->chan_table[j].rx_stats;
1546 		do {
1547 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1548 			packets = rx_stats->packets;
1549 			bytes = rx_stats->bytes;
1550 			xdp_drop = rx_stats->xdp_drop;
1551 			xdp_redirect = rx_stats->xdp_redirect;
1552 			xdp_tx = rx_stats->xdp_tx;
1553 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1554 		data[i++] = packets;
1555 		data[i++] = bytes;
1556 		data[i++] = xdp_drop;
1557 		data[i++] = xdp_redirect;
1558 		data[i++] = xdp_tx;
1559 	}
1560 
1561 	pcpu_sum = kvmalloc_array(nr_cpu_ids,
1562 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1563 				  GFP_KERNEL);
1564 	if (!pcpu_sum)
1565 		return;
1566 
1567 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1568 	for_each_present_cpu(cpu) {
1569 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1570 
1571 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1572 			data[i++] = *(u64 *)((void *)this_sum
1573 					     + pcpu_stats[j].offset);
1574 	}
1575 	kvfree(pcpu_sum);
1576 }
1577 
1578 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1579 {
1580 	struct net_device_context *ndc = netdev_priv(dev);
1581 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1582 	u8 *p = data;
1583 	int i, cpu;
1584 
1585 	if (!nvdev)
1586 		return;
1587 
1588 	switch (stringset) {
1589 	case ETH_SS_STATS:
1590 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1591 			ethtool_puts(&p, netvsc_stats[i].name);
1592 
1593 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1594 			ethtool_puts(&p, vf_stats[i].name);
1595 
1596 		for (i = 0; i < nvdev->num_chn; i++) {
1597 			ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1598 			ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1599 			ethtool_sprintf(&p, "tx_queue_%u_xdp_xmit", i);
1600 			ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1601 			ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1602 			ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1603 			ethtool_sprintf(&p, "rx_queue_%u_xdp_redirect", i);
1604 			ethtool_sprintf(&p, "rx_queue_%u_xdp_tx", i);
1605 		}
1606 
1607 		for_each_present_cpu(cpu) {
1608 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1609 				ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1610 		}
1611 
1612 		break;
1613 	}
1614 }
1615 
1616 static int
1617 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1618 			 struct ethtool_rxnfc *info)
1619 {
1620 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1621 
1622 	info->data = RXH_IP_SRC | RXH_IP_DST;
1623 
1624 	switch (info->flow_type) {
1625 	case TCP_V4_FLOW:
1626 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1627 			info->data |= l4_flag;
1628 
1629 		break;
1630 
1631 	case TCP_V6_FLOW:
1632 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1633 			info->data |= l4_flag;
1634 
1635 		break;
1636 
1637 	case UDP_V4_FLOW:
1638 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1639 			info->data |= l4_flag;
1640 
1641 		break;
1642 
1643 	case UDP_V6_FLOW:
1644 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1645 			info->data |= l4_flag;
1646 
1647 		break;
1648 
1649 	case IPV4_FLOW:
1650 	case IPV6_FLOW:
1651 		break;
1652 	default:
1653 		info->data = 0;
1654 		break;
1655 	}
1656 
1657 	return 0;
1658 }
1659 
1660 static int
1661 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1662 		 u32 *rules)
1663 {
1664 	struct net_device_context *ndc = netdev_priv(dev);
1665 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1666 
1667 	if (!nvdev)
1668 		return -ENODEV;
1669 
1670 	switch (info->cmd) {
1671 	case ETHTOOL_GRXRINGS:
1672 		info->data = nvdev->num_chn;
1673 		return 0;
1674 
1675 	case ETHTOOL_GRXFH:
1676 		return netvsc_get_rss_hash_opts(ndc, info);
1677 	}
1678 	return -EOPNOTSUPP;
1679 }
1680 
1681 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1682 				    struct ethtool_rxnfc *info)
1683 {
1684 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1685 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1686 		switch (info->flow_type) {
1687 		case TCP_V4_FLOW:
1688 			ndc->l4_hash |= HV_TCP4_L4HASH;
1689 			break;
1690 
1691 		case TCP_V6_FLOW:
1692 			ndc->l4_hash |= HV_TCP6_L4HASH;
1693 			break;
1694 
1695 		case UDP_V4_FLOW:
1696 			ndc->l4_hash |= HV_UDP4_L4HASH;
1697 			break;
1698 
1699 		case UDP_V6_FLOW:
1700 			ndc->l4_hash |= HV_UDP6_L4HASH;
1701 			break;
1702 
1703 		default:
1704 			return -EOPNOTSUPP;
1705 		}
1706 
1707 		return 0;
1708 	}
1709 
1710 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1711 		switch (info->flow_type) {
1712 		case TCP_V4_FLOW:
1713 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1714 			break;
1715 
1716 		case TCP_V6_FLOW:
1717 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1718 			break;
1719 
1720 		case UDP_V4_FLOW:
1721 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1722 			break;
1723 
1724 		case UDP_V6_FLOW:
1725 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1726 			break;
1727 
1728 		default:
1729 			return -EOPNOTSUPP;
1730 		}
1731 
1732 		return 0;
1733 	}
1734 
1735 	return -EOPNOTSUPP;
1736 }
1737 
1738 static int
1739 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1740 {
1741 	struct net_device_context *ndc = netdev_priv(ndev);
1742 
1743 	if (info->cmd == ETHTOOL_SRXFH)
1744 		return netvsc_set_rss_hash_opts(ndc, info);
1745 
1746 	return -EOPNOTSUPP;
1747 }
1748 
1749 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1750 {
1751 	return NETVSC_HASH_KEYLEN;
1752 }
1753 
1754 static u32 netvsc_rss_indir_size(struct net_device *dev)
1755 {
1756 	struct net_device_context *ndc = netdev_priv(dev);
1757 
1758 	return ndc->rx_table_sz;
1759 }
1760 
1761 static int netvsc_get_rxfh(struct net_device *dev,
1762 			   struct ethtool_rxfh_param *rxfh)
1763 {
1764 	struct net_device_context *ndc = netdev_priv(dev);
1765 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1766 	struct rndis_device *rndis_dev;
1767 	int i;
1768 
1769 	if (!ndev)
1770 		return -ENODEV;
1771 
1772 	rxfh->hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1773 
1774 	rndis_dev = ndev->extension;
1775 	if (rxfh->indir) {
1776 		for (i = 0; i < ndc->rx_table_sz; i++)
1777 			rxfh->indir[i] = ndc->rx_table[i];
1778 	}
1779 
1780 	if (rxfh->key)
1781 		memcpy(rxfh->key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1782 
1783 	return 0;
1784 }
1785 
1786 static int netvsc_set_rxfh(struct net_device *dev,
1787 			   struct ethtool_rxfh_param *rxfh,
1788 			   struct netlink_ext_ack *extack)
1789 {
1790 	struct net_device_context *ndc = netdev_priv(dev);
1791 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1792 	struct rndis_device *rndis_dev;
1793 	u8 *key = rxfh->key;
1794 	int i;
1795 
1796 	if (!ndev)
1797 		return -ENODEV;
1798 
1799 	if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
1800 	    rxfh->hfunc != ETH_RSS_HASH_TOP)
1801 		return -EOPNOTSUPP;
1802 
1803 	rndis_dev = ndev->extension;
1804 	if (rxfh->indir) {
1805 		for (i = 0; i < ndc->rx_table_sz; i++)
1806 			if (rxfh->indir[i] >= ndev->num_chn)
1807 				return -EINVAL;
1808 
1809 		for (i = 0; i < ndc->rx_table_sz; i++)
1810 			ndc->rx_table[i] = rxfh->indir[i];
1811 	}
1812 
1813 	if (!key) {
1814 		if (!rxfh->indir)
1815 			return 0;
1816 
1817 		key = rndis_dev->rss_key;
1818 	}
1819 
1820 	return rndis_filter_set_rss_param(rndis_dev, key);
1821 }
1822 
1823 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1824  * It does have pre-allocated receive area which is divided into sections.
1825  */
1826 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1827 				   struct ethtool_ringparam *ring)
1828 {
1829 	u32 max_buf_size;
1830 
1831 	ring->rx_pending = nvdev->recv_section_cnt;
1832 	ring->tx_pending = nvdev->send_section_cnt;
1833 
1834 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1835 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1836 	else
1837 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1838 
1839 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1840 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1841 		/ nvdev->send_section_size;
1842 }
1843 
1844 static void netvsc_get_ringparam(struct net_device *ndev,
1845 				 struct ethtool_ringparam *ring,
1846 				 struct kernel_ethtool_ringparam *kernel_ring,
1847 				 struct netlink_ext_ack *extack)
1848 {
1849 	struct net_device_context *ndevctx = netdev_priv(ndev);
1850 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1851 
1852 	if (!nvdev)
1853 		return;
1854 
1855 	__netvsc_get_ringparam(nvdev, ring);
1856 }
1857 
1858 static int netvsc_set_ringparam(struct net_device *ndev,
1859 				struct ethtool_ringparam *ring,
1860 				struct kernel_ethtool_ringparam *kernel_ring,
1861 				struct netlink_ext_ack *extack)
1862 {
1863 	struct net_device_context *ndevctx = netdev_priv(ndev);
1864 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1865 	struct netvsc_device_info *device_info;
1866 	struct ethtool_ringparam orig;
1867 	u32 new_tx, new_rx;
1868 	int ret = 0;
1869 
1870 	if (!nvdev || nvdev->destroy)
1871 		return -ENODEV;
1872 
1873 	memset(&orig, 0, sizeof(orig));
1874 	__netvsc_get_ringparam(nvdev, &orig);
1875 
1876 	new_tx = clamp_t(u32, ring->tx_pending,
1877 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1878 	new_rx = clamp_t(u32, ring->rx_pending,
1879 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1880 
1881 	if (new_tx == orig.tx_pending &&
1882 	    new_rx == orig.rx_pending)
1883 		return 0;	 /* no change */
1884 
1885 	device_info = netvsc_devinfo_get(nvdev);
1886 
1887 	if (!device_info)
1888 		return -ENOMEM;
1889 
1890 	device_info->send_sections = new_tx;
1891 	device_info->recv_sections = new_rx;
1892 
1893 	ret = netvsc_detach(ndev, nvdev);
1894 	if (ret)
1895 		goto out;
1896 
1897 	ret = netvsc_attach(ndev, device_info);
1898 	if (ret) {
1899 		device_info->send_sections = orig.tx_pending;
1900 		device_info->recv_sections = orig.rx_pending;
1901 
1902 		if (netvsc_attach(ndev, device_info))
1903 			netdev_err(ndev, "restoring ringparam failed");
1904 	}
1905 
1906 out:
1907 	netvsc_devinfo_put(device_info);
1908 	return ret;
1909 }
1910 
1911 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1912 					     netdev_features_t features)
1913 {
1914 	struct net_device_context *ndevctx = netdev_priv(ndev);
1915 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1916 
1917 	if (!nvdev || nvdev->destroy)
1918 		return features;
1919 
1920 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1921 		features ^= NETIF_F_LRO;
1922 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1923 	}
1924 
1925 	return features;
1926 }
1927 
1928 static int netvsc_set_features(struct net_device *ndev,
1929 			       netdev_features_t features)
1930 {
1931 	netdev_features_t change = features ^ ndev->features;
1932 	struct net_device_context *ndevctx = netdev_priv(ndev);
1933 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1934 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1935 	struct ndis_offload_params offloads;
1936 	int ret = 0;
1937 
1938 	if (!nvdev || nvdev->destroy)
1939 		return -ENODEV;
1940 
1941 	if (!(change & NETIF_F_LRO))
1942 		goto syncvf;
1943 
1944 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1945 
1946 	if (features & NETIF_F_LRO) {
1947 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1948 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1949 	} else {
1950 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1951 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1952 	}
1953 
1954 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1955 
1956 	if (ret) {
1957 		features ^= NETIF_F_LRO;
1958 		ndev->features = features;
1959 	}
1960 
1961 syncvf:
1962 	if (!vf_netdev)
1963 		return ret;
1964 
1965 	vf_netdev->wanted_features = features;
1966 	netdev_update_features(vf_netdev);
1967 
1968 	return ret;
1969 }
1970 
1971 static int netvsc_get_regs_len(struct net_device *netdev)
1972 {
1973 	return VRSS_SEND_TAB_SIZE * sizeof(u32);
1974 }
1975 
1976 static void netvsc_get_regs(struct net_device *netdev,
1977 			    struct ethtool_regs *regs, void *p)
1978 {
1979 	struct net_device_context *ndc = netdev_priv(netdev);
1980 	u32 *regs_buff = p;
1981 
1982 	/* increase the version, if buffer format is changed. */
1983 	regs->version = 1;
1984 
1985 	memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1986 }
1987 
1988 static u32 netvsc_get_msglevel(struct net_device *ndev)
1989 {
1990 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1991 
1992 	return ndev_ctx->msg_enable;
1993 }
1994 
1995 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1996 {
1997 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1998 
1999 	ndev_ctx->msg_enable = val;
2000 }
2001 
2002 static const struct ethtool_ops ethtool_ops = {
2003 	.get_drvinfo	= netvsc_get_drvinfo,
2004 	.get_regs_len	= netvsc_get_regs_len,
2005 	.get_regs	= netvsc_get_regs,
2006 	.get_msglevel	= netvsc_get_msglevel,
2007 	.set_msglevel	= netvsc_set_msglevel,
2008 	.get_link	= ethtool_op_get_link,
2009 	.get_ethtool_stats = netvsc_get_ethtool_stats,
2010 	.get_sset_count = netvsc_get_sset_count,
2011 	.get_strings	= netvsc_get_strings,
2012 	.get_channels   = netvsc_get_channels,
2013 	.set_channels   = netvsc_set_channels,
2014 	.get_ts_info	= ethtool_op_get_ts_info,
2015 	.get_rxnfc	= netvsc_get_rxnfc,
2016 	.set_rxnfc	= netvsc_set_rxnfc,
2017 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
2018 	.get_rxfh_indir_size = netvsc_rss_indir_size,
2019 	.get_rxfh	= netvsc_get_rxfh,
2020 	.set_rxfh	= netvsc_set_rxfh,
2021 	.get_link_ksettings = netvsc_get_link_ksettings,
2022 	.set_link_ksettings = netvsc_set_link_ksettings,
2023 	.get_ringparam	= netvsc_get_ringparam,
2024 	.set_ringparam	= netvsc_set_ringparam,
2025 };
2026 
2027 static const struct net_device_ops device_ops = {
2028 	.ndo_open =			netvsc_open,
2029 	.ndo_stop =			netvsc_close,
2030 	.ndo_start_xmit =		netvsc_start_xmit,
2031 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
2032 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
2033 	.ndo_fix_features =		netvsc_fix_features,
2034 	.ndo_set_features =		netvsc_set_features,
2035 	.ndo_change_mtu =		netvsc_change_mtu,
2036 	.ndo_validate_addr =		eth_validate_addr,
2037 	.ndo_set_mac_address =		netvsc_set_mac_addr,
2038 	.ndo_select_queue =		netvsc_select_queue,
2039 	.ndo_get_stats64 =		netvsc_get_stats64,
2040 	.ndo_bpf =			netvsc_bpf,
2041 	.ndo_xdp_xmit =			netvsc_ndoxdp_xmit,
2042 };
2043 
2044 /*
2045  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2046  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2047  * present send GARP packet to network peers with netif_notify_peers().
2048  */
2049 static void netvsc_link_change(struct work_struct *w)
2050 {
2051 	struct net_device_context *ndev_ctx =
2052 		container_of(w, struct net_device_context, dwork.work);
2053 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2054 	struct net_device *net = hv_get_drvdata(device_obj);
2055 	unsigned long flags, next_reconfig, delay;
2056 	struct netvsc_reconfig *event = NULL;
2057 	struct netvsc_device *net_device;
2058 	struct rndis_device *rdev;
2059 	bool reschedule = false;
2060 
2061 	/* if changes are happening, comeback later */
2062 	if (!rtnl_trylock()) {
2063 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2064 		return;
2065 	}
2066 
2067 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2068 	if (!net_device)
2069 		goto out_unlock;
2070 
2071 	rdev = net_device->extension;
2072 
2073 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2074 	if (time_is_after_jiffies(next_reconfig)) {
2075 		/* link_watch only sends one notification with current state
2076 		 * per second, avoid doing reconfig more frequently. Handle
2077 		 * wrap around.
2078 		 */
2079 		delay = next_reconfig - jiffies;
2080 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2081 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2082 		goto out_unlock;
2083 	}
2084 	ndev_ctx->last_reconfig = jiffies;
2085 
2086 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2087 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2088 		event = list_first_entry(&ndev_ctx->reconfig_events,
2089 					 struct netvsc_reconfig, list);
2090 		list_del(&event->list);
2091 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2092 	}
2093 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2094 
2095 	if (!event)
2096 		goto out_unlock;
2097 
2098 	switch (event->event) {
2099 		/* Only the following events are possible due to the check in
2100 		 * netvsc_linkstatus_callback()
2101 		 */
2102 	case RNDIS_STATUS_MEDIA_CONNECT:
2103 		if (rdev->link_state) {
2104 			rdev->link_state = false;
2105 			netif_carrier_on(net);
2106 			netvsc_tx_enable(net_device, net);
2107 		} else {
2108 			__netdev_notify_peers(net);
2109 		}
2110 		kfree(event);
2111 		break;
2112 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2113 		if (!rdev->link_state) {
2114 			rdev->link_state = true;
2115 			netif_carrier_off(net);
2116 			netvsc_tx_disable(net_device, net);
2117 		}
2118 		kfree(event);
2119 		break;
2120 	case RNDIS_STATUS_NETWORK_CHANGE:
2121 		/* Only makes sense if carrier is present */
2122 		if (!rdev->link_state) {
2123 			rdev->link_state = true;
2124 			netif_carrier_off(net);
2125 			netvsc_tx_disable(net_device, net);
2126 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2127 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2128 			list_add(&event->list, &ndev_ctx->reconfig_events);
2129 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2130 			reschedule = true;
2131 		}
2132 		break;
2133 	}
2134 
2135 	rtnl_unlock();
2136 
2137 	/* link_watch only sends one notification with current state per
2138 	 * second, handle next reconfig event in 2 seconds.
2139 	 */
2140 	if (reschedule)
2141 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2142 
2143 	return;
2144 
2145 out_unlock:
2146 	rtnl_unlock();
2147 }
2148 
2149 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2150 {
2151 	struct net_device_context *net_device_ctx;
2152 	struct net_device *dev;
2153 
2154 	dev = netdev_master_upper_dev_get(vf_netdev);
2155 	if (!dev || dev->netdev_ops != &device_ops)
2156 		return NULL;	/* not a netvsc device */
2157 
2158 	net_device_ctx = netdev_priv(dev);
2159 	if (!rtnl_dereference(net_device_ctx->nvdev))
2160 		return NULL;	/* device is removed */
2161 
2162 	return dev;
2163 }
2164 
2165 /* Called when VF is injecting data into network stack.
2166  * Change the associated network device from VF to netvsc.
2167  * note: already called with rcu_read_lock
2168  */
2169 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2170 {
2171 	struct sk_buff *skb = *pskb;
2172 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2173 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2174 	struct netvsc_vf_pcpu_stats *pcpu_stats
2175 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2176 
2177 	skb = skb_share_check(skb, GFP_ATOMIC);
2178 	if (unlikely(!skb))
2179 		return RX_HANDLER_CONSUMED;
2180 
2181 	*pskb = skb;
2182 
2183 	skb->dev = ndev;
2184 
2185 	u64_stats_update_begin(&pcpu_stats->syncp);
2186 	pcpu_stats->rx_packets++;
2187 	pcpu_stats->rx_bytes += skb->len;
2188 	u64_stats_update_end(&pcpu_stats->syncp);
2189 
2190 	return RX_HANDLER_ANOTHER;
2191 }
2192 
2193 static int netvsc_vf_join(struct net_device *vf_netdev,
2194 			  struct net_device *ndev, int context)
2195 {
2196 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2197 	int ret;
2198 
2199 	ret = netdev_rx_handler_register(vf_netdev,
2200 					 netvsc_vf_handle_frame, ndev);
2201 	if (ret != 0) {
2202 		netdev_err(vf_netdev,
2203 			   "can not register netvsc VF receive handler (err = %d)\n",
2204 			   ret);
2205 		goto rx_handler_failed;
2206 	}
2207 
2208 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2209 					   NULL, NULL, NULL);
2210 	if (ret != 0) {
2211 		netdev_err(vf_netdev,
2212 			   "can not set master device %s (err = %d)\n",
2213 			   ndev->name, ret);
2214 		goto upper_link_failed;
2215 	}
2216 
2217 	/* If this registration is called from probe context vf_takeover
2218 	 * is taken care of later in probe itself.
2219 	 */
2220 	if (context == VF_REG_IN_NOTIFIER)
2221 		schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2222 
2223 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2224 
2225 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2226 	return 0;
2227 
2228 upper_link_failed:
2229 	netdev_rx_handler_unregister(vf_netdev);
2230 rx_handler_failed:
2231 	return ret;
2232 }
2233 
2234 static void __netvsc_vf_setup(struct net_device *ndev,
2235 			      struct net_device *vf_netdev)
2236 {
2237 	int ret;
2238 
2239 	/* Align MTU of VF with master */
2240 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2241 	if (ret)
2242 		netdev_warn(vf_netdev,
2243 			    "unable to change mtu to %u\n", ndev->mtu);
2244 
2245 	/* set multicast etc flags on VF */
2246 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2247 
2248 	/* sync address list from ndev to VF */
2249 	netif_addr_lock_bh(ndev);
2250 	dev_uc_sync(vf_netdev, ndev);
2251 	dev_mc_sync(vf_netdev, ndev);
2252 	netif_addr_unlock_bh(ndev);
2253 
2254 	if (netif_running(ndev)) {
2255 		ret = dev_open(vf_netdev, NULL);
2256 		if (ret)
2257 			netdev_warn(vf_netdev,
2258 				    "unable to open: %d\n", ret);
2259 	}
2260 }
2261 
2262 /* Setup VF as slave of the synthetic device.
2263  * Runs in workqueue to avoid recursion in netlink callbacks.
2264  */
2265 static void netvsc_vf_setup(struct work_struct *w)
2266 {
2267 	struct net_device_context *ndev_ctx
2268 		= container_of(w, struct net_device_context, vf_takeover.work);
2269 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2270 	struct net_device *vf_netdev;
2271 
2272 	if (!rtnl_trylock()) {
2273 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2274 		return;
2275 	}
2276 
2277 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2278 	if (vf_netdev)
2279 		__netvsc_vf_setup(ndev, vf_netdev);
2280 
2281 	rtnl_unlock();
2282 }
2283 
2284 /* Find netvsc by VF serial number.
2285  * The PCI hyperv controller records the serial number as the slot kobj name.
2286  */
2287 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2288 {
2289 	struct device *parent = vf_netdev->dev.parent;
2290 	struct net_device_context *ndev_ctx;
2291 	struct net_device *ndev;
2292 	struct pci_dev *pdev;
2293 	u32 serial;
2294 
2295 	if (!parent || !dev_is_pci(parent))
2296 		return NULL; /* not a PCI device */
2297 
2298 	pdev = to_pci_dev(parent);
2299 	if (!pdev->slot) {
2300 		netdev_notice(vf_netdev, "no PCI slot information\n");
2301 		return NULL;
2302 	}
2303 
2304 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2305 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2306 			      pci_slot_name(pdev->slot));
2307 		return NULL;
2308 	}
2309 
2310 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2311 		if (!ndev_ctx->vf_alloc)
2312 			continue;
2313 
2314 		if (ndev_ctx->vf_serial != serial)
2315 			continue;
2316 
2317 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2318 		if (ndev->addr_len != vf_netdev->addr_len ||
2319 		    memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2320 			   ndev->addr_len) != 0)
2321 			continue;
2322 
2323 		return ndev;
2324 
2325 	}
2326 
2327 	/* Fallback path to check synthetic vf with help of mac addr.
2328 	 * Because this function can be called before vf_netdev is
2329 	 * initialized (NETDEV_POST_INIT) when its perm_addr has not been copied
2330 	 * from dev_addr, also try to match to its dev_addr.
2331 	 * Note: On Hyper-V and Azure, it's not possible to set a MAC address
2332 	 * on a VF that matches to the MAC of a unrelated NETVSC device.
2333 	 */
2334 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2335 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2336 		if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr) ||
2337 		    ether_addr_equal(vf_netdev->dev_addr, ndev->perm_addr))
2338 			return ndev;
2339 	}
2340 
2341 	netdev_notice(vf_netdev,
2342 		      "no netdev found for vf serial:%u\n", serial);
2343 	return NULL;
2344 }
2345 
2346 static int netvsc_prepare_bonding(struct net_device *vf_netdev)
2347 {
2348 	struct net_device *ndev;
2349 
2350 	ndev = get_netvsc_byslot(vf_netdev);
2351 	if (!ndev)
2352 		return NOTIFY_DONE;
2353 
2354 	/* set slave flag before open to prevent IPv6 addrconf */
2355 	vf_netdev->flags |= IFF_SLAVE;
2356 	return NOTIFY_DONE;
2357 }
2358 
2359 static int netvsc_register_vf(struct net_device *vf_netdev, int context)
2360 {
2361 	struct net_device_context *net_device_ctx;
2362 	struct netvsc_device *netvsc_dev;
2363 	struct bpf_prog *prog;
2364 	struct net_device *ndev;
2365 	int ret;
2366 
2367 	if (vf_netdev->addr_len != ETH_ALEN)
2368 		return NOTIFY_DONE;
2369 
2370 	ndev = get_netvsc_byslot(vf_netdev);
2371 	if (!ndev)
2372 		return NOTIFY_DONE;
2373 
2374 	net_device_ctx = netdev_priv(ndev);
2375 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2376 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2377 		return NOTIFY_DONE;
2378 
2379 	/* if synthetic interface is a different namespace,
2380 	 * then move the VF to that namespace; join will be
2381 	 * done again in that context.
2382 	 */
2383 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2384 		ret = dev_change_net_namespace(vf_netdev,
2385 					       dev_net(ndev), "eth%d");
2386 		if (ret)
2387 			netdev_err(vf_netdev,
2388 				   "could not move to same namespace as %s: %d\n",
2389 				   ndev->name, ret);
2390 		else
2391 			netdev_info(vf_netdev,
2392 				    "VF moved to namespace with: %s\n",
2393 				    ndev->name);
2394 		return NOTIFY_DONE;
2395 	}
2396 
2397 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2398 
2399 	if (netvsc_vf_join(vf_netdev, ndev, context) != 0)
2400 		return NOTIFY_DONE;
2401 
2402 	dev_hold(vf_netdev);
2403 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2404 
2405 	if (ndev->needed_headroom < vf_netdev->needed_headroom)
2406 		ndev->needed_headroom = vf_netdev->needed_headroom;
2407 
2408 	vf_netdev->wanted_features = ndev->features;
2409 	netdev_update_features(vf_netdev);
2410 
2411 	prog = netvsc_xdp_get(netvsc_dev);
2412 	netvsc_vf_setxdp(vf_netdev, prog);
2413 
2414 	return NOTIFY_OK;
2415 }
2416 
2417 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2418  *
2419  * Typically a UP or DOWN event is followed by a CHANGE event, so
2420  * net_device_ctx->data_path_is_vf is used to cache the current data path
2421  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2422  * message.
2423  *
2424  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2425  * interface, there is only the CHANGE event and no UP or DOWN event.
2426  */
2427 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2428 {
2429 	struct net_device_context *net_device_ctx;
2430 	struct netvsc_device *netvsc_dev;
2431 	struct net_device *ndev;
2432 	bool vf_is_up = false;
2433 	int ret;
2434 
2435 	if (event != NETDEV_GOING_DOWN)
2436 		vf_is_up = netif_running(vf_netdev);
2437 
2438 	ndev = get_netvsc_byref(vf_netdev);
2439 	if (!ndev)
2440 		return NOTIFY_DONE;
2441 
2442 	net_device_ctx = netdev_priv(ndev);
2443 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2444 	if (!netvsc_dev)
2445 		return NOTIFY_DONE;
2446 
2447 	if (net_device_ctx->data_path_is_vf == vf_is_up)
2448 		return NOTIFY_OK;
2449 
2450 	if (vf_is_up && !net_device_ctx->vf_alloc) {
2451 		netdev_info(ndev, "Waiting for the VF association from host\n");
2452 		wait_for_completion(&net_device_ctx->vf_add);
2453 	}
2454 
2455 	ret = netvsc_switch_datapath(ndev, vf_is_up);
2456 
2457 	if (ret) {
2458 		netdev_err(ndev,
2459 			   "Data path failed to switch %s VF: %s, err: %d\n",
2460 			   vf_is_up ? "to" : "from", vf_netdev->name, ret);
2461 		return NOTIFY_DONE;
2462 	} else {
2463 		netdev_info(ndev, "Data path switched %s VF: %s\n",
2464 			    vf_is_up ? "to" : "from", vf_netdev->name);
2465 
2466 		/* In Azure, when accelerated networking in enabled, other NICs
2467 		 * like MANA, MLX, are configured as a bonded nic with
2468 		 * Netvsc(failover) NIC. For bonded NICs, the min of the max
2469 		 * pkt aggregate size of the members is propagated in the stack.
2470 		 * In order to allow these NICs (MANA/MLX) to use up to
2471 		 * GSO_MAX_SIZE gso packet size, we need to allow Netvsc NIC to
2472 		 * also support this in the guest.
2473 		 * This value is only increased for netvsc NIC when datapath is
2474 		 * switched over to the VF
2475 		 */
2476 		if (vf_is_up)
2477 			netif_set_tso_max_size(ndev, vf_netdev->tso_max_size);
2478 		else
2479 			netif_set_tso_max_size(ndev, netvsc_dev->netvsc_gso_max_size);
2480 	}
2481 
2482 	return NOTIFY_OK;
2483 }
2484 
2485 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2486 {
2487 	struct net_device *ndev;
2488 	struct net_device_context *net_device_ctx;
2489 
2490 	ndev = get_netvsc_byref(vf_netdev);
2491 	if (!ndev)
2492 		return NOTIFY_DONE;
2493 
2494 	net_device_ctx = netdev_priv(ndev);
2495 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2496 
2497 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2498 
2499 	netvsc_vf_setxdp(vf_netdev, NULL);
2500 
2501 	reinit_completion(&net_device_ctx->vf_add);
2502 	netdev_rx_handler_unregister(vf_netdev);
2503 	netdev_upper_dev_unlink(vf_netdev, ndev);
2504 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2505 	dev_put(vf_netdev);
2506 
2507 	ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2508 
2509 	return NOTIFY_OK;
2510 }
2511 
2512 static int check_dev_is_matching_vf(struct net_device *event_ndev)
2513 {
2514 	/* Skip NetVSC interfaces */
2515 	if (event_ndev->netdev_ops == &device_ops)
2516 		return -ENODEV;
2517 
2518 	/* Avoid non-Ethernet type devices */
2519 	if (event_ndev->type != ARPHRD_ETHER)
2520 		return -ENODEV;
2521 
2522 	/* Avoid Vlan dev with same MAC registering as VF */
2523 	if (is_vlan_dev(event_ndev))
2524 		return -ENODEV;
2525 
2526 	/* Avoid Bonding master dev with same MAC registering as VF */
2527 	if (netif_is_bond_master(event_ndev))
2528 		return -ENODEV;
2529 
2530 	return 0;
2531 }
2532 
2533 static int netvsc_probe(struct hv_device *dev,
2534 			const struct hv_vmbus_device_id *dev_id)
2535 {
2536 	struct net_device *net = NULL, *vf_netdev;
2537 	struct net_device_context *net_device_ctx;
2538 	struct netvsc_device_info *device_info = NULL;
2539 	struct netvsc_device *nvdev;
2540 	int ret = -ENOMEM;
2541 
2542 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2543 				VRSS_CHANNEL_MAX);
2544 	if (!net)
2545 		goto no_net;
2546 
2547 	netif_carrier_off(net);
2548 
2549 	netvsc_init_settings(net);
2550 
2551 	net_device_ctx = netdev_priv(net);
2552 	net_device_ctx->device_ctx = dev;
2553 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2554 	if (netif_msg_probe(net_device_ctx))
2555 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2556 			   net_device_ctx->msg_enable);
2557 
2558 	hv_set_drvdata(dev, net);
2559 
2560 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2561 
2562 	init_completion(&net_device_ctx->vf_add);
2563 	spin_lock_init(&net_device_ctx->lock);
2564 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2565 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2566 
2567 	net_device_ctx->vf_stats
2568 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2569 	if (!net_device_ctx->vf_stats)
2570 		goto no_stats;
2571 
2572 	net->netdev_ops = &device_ops;
2573 	net->ethtool_ops = &ethtool_ops;
2574 	SET_NETDEV_DEV(net, &dev->device);
2575 	dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
2576 
2577 	/* We always need headroom for rndis header */
2578 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2579 
2580 	/* Initialize the number of queues to be 1, we may change it if more
2581 	 * channels are offered later.
2582 	 */
2583 	netif_set_real_num_tx_queues(net, 1);
2584 	netif_set_real_num_rx_queues(net, 1);
2585 
2586 	/* Notify the netvsc driver of the new device */
2587 	device_info = netvsc_devinfo_get(NULL);
2588 
2589 	if (!device_info) {
2590 		ret = -ENOMEM;
2591 		goto devinfo_failed;
2592 	}
2593 
2594 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2595 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2596 	 * all subchannels to show up, but that may not happen because
2597 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2598 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2599 	 * the device lock, so all the subchannels can't be processed --
2600 	 * finally netvsc_subchan_work() hangs forever.
2601 	 *
2602 	 * The rtnl lock also needs to be held before rndis_filter_device_add()
2603 	 * which advertises nvsp_2_vsc_capability / sriov bit, and triggers
2604 	 * VF NIC offering and registering. If VF NIC finished register_netdev()
2605 	 * earlier it may cause name based config failure.
2606 	 */
2607 	rtnl_lock();
2608 
2609 	nvdev = rndis_filter_device_add(dev, device_info);
2610 	if (IS_ERR(nvdev)) {
2611 		ret = PTR_ERR(nvdev);
2612 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2613 		goto rndis_failed;
2614 	}
2615 
2616 	eth_hw_addr_set(net, device_info->mac_adr);
2617 
2618 	if (nvdev->num_chn > 1)
2619 		schedule_work(&nvdev->subchan_work);
2620 
2621 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2622 	net->features = net->hw_features |
2623 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2624 		NETIF_F_HW_VLAN_CTAG_RX;
2625 	net->vlan_features = net->features;
2626 
2627 	netdev_lockdep_set_classes(net);
2628 
2629 	net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2630 			    NETDEV_XDP_ACT_NDO_XMIT;
2631 
2632 	/* MTU range: 68 - 1500 or 65521 */
2633 	net->min_mtu = NETVSC_MTU_MIN;
2634 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2635 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2636 	else
2637 		net->max_mtu = ETH_DATA_LEN;
2638 
2639 	nvdev->tx_disable = false;
2640 
2641 	ret = register_netdevice(net);
2642 	if (ret != 0) {
2643 		pr_err("Unable to register netdev.\n");
2644 		goto register_failed;
2645 	}
2646 
2647 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2648 
2649 	/* When the hv_netvsc driver is unloaded and reloaded, the
2650 	 * NET_DEVICE_REGISTER for the vf device is replayed before probe
2651 	 * is complete. This is because register_netdevice_notifier() gets
2652 	 * registered before vmbus_driver_register() so that callback func
2653 	 * is set before probe and we don't miss events like NETDEV_POST_INIT
2654 	 * So, in this section we try to register the matching vf device that
2655 	 * is present as a netdevice, knowing that its register call is not
2656 	 * processed in the netvsc_netdev_notifier(as probing is progress and
2657 	 * get_netvsc_byslot fails).
2658 	 */
2659 	for_each_netdev(dev_net(net), vf_netdev) {
2660 		ret = check_dev_is_matching_vf(vf_netdev);
2661 		if (ret != 0)
2662 			continue;
2663 
2664 		if (net != get_netvsc_byslot(vf_netdev))
2665 			continue;
2666 
2667 		netvsc_prepare_bonding(vf_netdev);
2668 		netvsc_register_vf(vf_netdev, VF_REG_IN_PROBE);
2669 		__netvsc_vf_setup(net, vf_netdev);
2670 		break;
2671 	}
2672 	rtnl_unlock();
2673 
2674 	netvsc_devinfo_put(device_info);
2675 	return 0;
2676 
2677 register_failed:
2678 	rndis_filter_device_remove(dev, nvdev);
2679 rndis_failed:
2680 	rtnl_unlock();
2681 	netvsc_devinfo_put(device_info);
2682 devinfo_failed:
2683 	free_percpu(net_device_ctx->vf_stats);
2684 no_stats:
2685 	hv_set_drvdata(dev, NULL);
2686 	free_netdev(net);
2687 no_net:
2688 	return ret;
2689 }
2690 
2691 static void netvsc_remove(struct hv_device *dev)
2692 {
2693 	struct net_device_context *ndev_ctx;
2694 	struct net_device *vf_netdev, *net;
2695 	struct netvsc_device *nvdev;
2696 
2697 	net = hv_get_drvdata(dev);
2698 	if (net == NULL) {
2699 		dev_err(&dev->device, "No net device to remove\n");
2700 		return;
2701 	}
2702 
2703 	ndev_ctx = netdev_priv(net);
2704 
2705 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2706 
2707 	rtnl_lock();
2708 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2709 	if (nvdev) {
2710 		cancel_work_sync(&nvdev->subchan_work);
2711 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2712 	}
2713 
2714 	/*
2715 	 * Call to the vsc driver to let it know that the device is being
2716 	 * removed. Also blocks mtu and channel changes.
2717 	 */
2718 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2719 	if (vf_netdev)
2720 		netvsc_unregister_vf(vf_netdev);
2721 
2722 	if (nvdev)
2723 		rndis_filter_device_remove(dev, nvdev);
2724 
2725 	unregister_netdevice(net);
2726 	list_del(&ndev_ctx->list);
2727 
2728 	rtnl_unlock();
2729 
2730 	hv_set_drvdata(dev, NULL);
2731 
2732 	free_percpu(ndev_ctx->vf_stats);
2733 	free_netdev(net);
2734 }
2735 
2736 static int netvsc_suspend(struct hv_device *dev)
2737 {
2738 	struct net_device_context *ndev_ctx;
2739 	struct netvsc_device *nvdev;
2740 	struct net_device *net;
2741 	int ret;
2742 
2743 	net = hv_get_drvdata(dev);
2744 
2745 	ndev_ctx = netdev_priv(net);
2746 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2747 
2748 	rtnl_lock();
2749 
2750 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2751 	if (nvdev == NULL) {
2752 		ret = -ENODEV;
2753 		goto out;
2754 	}
2755 
2756 	/* Save the current config info */
2757 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2758 	if (!ndev_ctx->saved_netvsc_dev_info) {
2759 		ret = -ENOMEM;
2760 		goto out;
2761 	}
2762 	ret = netvsc_detach(net, nvdev);
2763 out:
2764 	rtnl_unlock();
2765 
2766 	return ret;
2767 }
2768 
2769 static int netvsc_resume(struct hv_device *dev)
2770 {
2771 	struct net_device *net = hv_get_drvdata(dev);
2772 	struct net_device_context *net_device_ctx;
2773 	struct netvsc_device_info *device_info;
2774 	int ret;
2775 
2776 	rtnl_lock();
2777 
2778 	net_device_ctx = netdev_priv(net);
2779 
2780 	/* Reset the data path to the netvsc NIC before re-opening the vmbus
2781 	 * channel. Later netvsc_netdev_event() will switch the data path to
2782 	 * the VF upon the UP or CHANGE event.
2783 	 */
2784 	net_device_ctx->data_path_is_vf = false;
2785 	device_info = net_device_ctx->saved_netvsc_dev_info;
2786 
2787 	ret = netvsc_attach(net, device_info);
2788 
2789 	netvsc_devinfo_put(device_info);
2790 	net_device_ctx->saved_netvsc_dev_info = NULL;
2791 
2792 	rtnl_unlock();
2793 
2794 	return ret;
2795 }
2796 static const struct hv_vmbus_device_id id_table[] = {
2797 	/* Network guid */
2798 	{ HV_NIC_GUID, },
2799 	{ },
2800 };
2801 
2802 MODULE_DEVICE_TABLE(vmbus, id_table);
2803 
2804 /* The one and only one */
2805 static struct  hv_driver netvsc_drv = {
2806 	.name = KBUILD_MODNAME,
2807 	.id_table = id_table,
2808 	.probe = netvsc_probe,
2809 	.remove = netvsc_remove,
2810 	.suspend = netvsc_suspend,
2811 	.resume = netvsc_resume,
2812 	.driver = {
2813 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2814 	},
2815 };
2816 
2817 /* Set VF's namespace same as the synthetic NIC */
2818 static void netvsc_event_set_vf_ns(struct net_device *ndev)
2819 {
2820 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2821 	struct net_device *vf_netdev;
2822 	int ret;
2823 
2824 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2825 	if (!vf_netdev)
2826 		return;
2827 
2828 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2829 		ret = dev_change_net_namespace(vf_netdev, dev_net(ndev),
2830 					       "eth%d");
2831 		if (ret)
2832 			netdev_err(vf_netdev,
2833 				   "Cannot move to same namespace as %s: %d\n",
2834 				   ndev->name, ret);
2835 		else
2836 			netdev_info(vf_netdev,
2837 				    "Moved VF to namespace with: %s\n",
2838 				    ndev->name);
2839 	}
2840 }
2841 
2842 /*
2843  * On Hyper-V, every VF interface is matched with a corresponding
2844  * synthetic interface. The synthetic interface is presented first
2845  * to the guest. When the corresponding VF instance is registered,
2846  * we will take care of switching the data path.
2847  */
2848 static int netvsc_netdev_event(struct notifier_block *this,
2849 			       unsigned long event, void *ptr)
2850 {
2851 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2852 	int ret = 0;
2853 
2854 	if (event_dev->netdev_ops == &device_ops && event == NETDEV_REGISTER) {
2855 		netvsc_event_set_vf_ns(event_dev);
2856 		return NOTIFY_DONE;
2857 	}
2858 
2859 	ret = check_dev_is_matching_vf(event_dev);
2860 	if (ret != 0)
2861 		return NOTIFY_DONE;
2862 
2863 	switch (event) {
2864 	case NETDEV_POST_INIT:
2865 		return netvsc_prepare_bonding(event_dev);
2866 	case NETDEV_REGISTER:
2867 		return netvsc_register_vf(event_dev, VF_REG_IN_NOTIFIER);
2868 	case NETDEV_UNREGISTER:
2869 		return netvsc_unregister_vf(event_dev);
2870 	case NETDEV_UP:
2871 	case NETDEV_DOWN:
2872 	case NETDEV_CHANGE:
2873 	case NETDEV_GOING_DOWN:
2874 		return netvsc_vf_changed(event_dev, event);
2875 	default:
2876 		return NOTIFY_DONE;
2877 	}
2878 }
2879 
2880 static struct notifier_block netvsc_netdev_notifier = {
2881 	.notifier_call = netvsc_netdev_event,
2882 };
2883 
2884 static void __exit netvsc_drv_exit(void)
2885 {
2886 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2887 	vmbus_driver_unregister(&netvsc_drv);
2888 }
2889 
2890 static int __init netvsc_drv_init(void)
2891 {
2892 	int ret;
2893 
2894 	if (ring_size < RING_SIZE_MIN) {
2895 		ring_size = RING_SIZE_MIN;
2896 		pr_info("Increased ring_size to %u (min allowed)\n",
2897 			ring_size);
2898 	}
2899 	netvsc_ring_bytes = VMBUS_RING_SIZE(ring_size * 4096);
2900 
2901 	register_netdevice_notifier(&netvsc_netdev_notifier);
2902 
2903 	ret = vmbus_driver_register(&netvsc_drv);
2904 	if (ret)
2905 		goto err_vmbus_reg;
2906 
2907 	return 0;
2908 
2909 err_vmbus_reg:
2910 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2911 	return ret;
2912 }
2913 
2914 MODULE_LICENSE("GPL");
2915 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2916 
2917 module_init(netvsc_drv_init);
2918 module_exit(netvsc_drv_exit);
2919