xref: /linux/drivers/net/hyperv/netvsc_drv.c (revision 55d0969c451159cff86949b38c39171cab962069)
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/route.h>
33 #include <net/sock.h>
34 #include <net/pkt_sched.h>
35 #include <net/checksum.h>
36 #include <net/ip6_checksum.h>
37 
38 #include "hyperv_net.h"
39 
40 #define RING_SIZE_MIN	64
41 
42 #define LINKCHANGE_INT (2 * HZ)
43 #define VF_TAKEOVER_INT (HZ / 10)
44 
45 /* Macros to define the context of vf registration */
46 #define VF_REG_IN_PROBE		1
47 #define VF_REG_IN_NOTIFIER	2
48 
49 static unsigned int ring_size __ro_after_init = 128;
50 module_param(ring_size, uint, 0444);
51 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of 4K pages)");
52 unsigned int netvsc_ring_bytes __ro_after_init;
53 
54 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
55 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
56 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
57 				NETIF_MSG_TX_ERR;
58 
59 static int debug = -1;
60 module_param(debug, int, 0444);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62 
63 static LIST_HEAD(netvsc_dev_list);
64 
65 static void netvsc_change_rx_flags(struct net_device *net, int change)
66 {
67 	struct net_device_context *ndev_ctx = netdev_priv(net);
68 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
69 	int inc;
70 
71 	if (!vf_netdev)
72 		return;
73 
74 	if (change & IFF_PROMISC) {
75 		inc = (net->flags & IFF_PROMISC) ? 1 : -1;
76 		dev_set_promiscuity(vf_netdev, inc);
77 	}
78 
79 	if (change & IFF_ALLMULTI) {
80 		inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
81 		dev_set_allmulti(vf_netdev, inc);
82 	}
83 }
84 
85 static void netvsc_set_rx_mode(struct net_device *net)
86 {
87 	struct net_device_context *ndev_ctx = netdev_priv(net);
88 	struct net_device *vf_netdev;
89 	struct netvsc_device *nvdev;
90 
91 	rcu_read_lock();
92 	vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
93 	if (vf_netdev) {
94 		dev_uc_sync(vf_netdev, net);
95 		dev_mc_sync(vf_netdev, net);
96 	}
97 
98 	nvdev = rcu_dereference(ndev_ctx->nvdev);
99 	if (nvdev)
100 		rndis_filter_update(nvdev);
101 	rcu_read_unlock();
102 }
103 
104 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
105 			     struct net_device *ndev)
106 {
107 	nvscdev->tx_disable = false;
108 	virt_wmb(); /* ensure queue wake up mechanism is on */
109 
110 	netif_tx_wake_all_queues(ndev);
111 }
112 
113 static int netvsc_open(struct net_device *net)
114 {
115 	struct net_device_context *ndev_ctx = netdev_priv(net);
116 	struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
117 	struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
118 	struct rndis_device *rdev;
119 	int ret = 0;
120 
121 	netif_carrier_off(net);
122 
123 	/* Open up the device */
124 	ret = rndis_filter_open(nvdev);
125 	if (ret != 0) {
126 		netdev_err(net, "unable to open device (ret %d).\n", ret);
127 		return ret;
128 	}
129 
130 	rdev = nvdev->extension;
131 	if (!rdev->link_state) {
132 		netif_carrier_on(net);
133 		netvsc_tx_enable(nvdev, net);
134 	}
135 
136 	if (vf_netdev) {
137 		/* Setting synthetic device up transparently sets
138 		 * slave as up. If open fails, then slave will be
139 		 * still be offline (and not used).
140 		 */
141 		ret = dev_open(vf_netdev, NULL);
142 		if (ret)
143 			netdev_warn(net,
144 				    "unable to open slave: %s: %d\n",
145 				    vf_netdev->name, ret);
146 	}
147 	return 0;
148 }
149 
150 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
151 {
152 	unsigned int retry = 0;
153 	int i;
154 
155 	/* Ensure pending bytes in ring are read */
156 	for (;;) {
157 		u32 aread = 0;
158 
159 		for (i = 0; i < nvdev->num_chn; i++) {
160 			struct vmbus_channel *chn
161 				= nvdev->chan_table[i].channel;
162 
163 			if (!chn)
164 				continue;
165 
166 			/* make sure receive not running now */
167 			napi_synchronize(&nvdev->chan_table[i].napi);
168 
169 			aread = hv_get_bytes_to_read(&chn->inbound);
170 			if (aread)
171 				break;
172 
173 			aread = hv_get_bytes_to_read(&chn->outbound);
174 			if (aread)
175 				break;
176 		}
177 
178 		if (aread == 0)
179 			return 0;
180 
181 		if (++retry > RETRY_MAX)
182 			return -ETIMEDOUT;
183 
184 		usleep_range(RETRY_US_LO, RETRY_US_HI);
185 	}
186 }
187 
188 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
189 			      struct net_device *ndev)
190 {
191 	if (nvscdev) {
192 		nvscdev->tx_disable = true;
193 		virt_wmb(); /* ensure txq will not wake up after stop */
194 	}
195 
196 	netif_tx_disable(ndev);
197 }
198 
199 static int netvsc_close(struct net_device *net)
200 {
201 	struct net_device_context *net_device_ctx = netdev_priv(net);
202 	struct net_device *vf_netdev
203 		= rtnl_dereference(net_device_ctx->vf_netdev);
204 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
205 	int ret;
206 
207 	netvsc_tx_disable(nvdev, net);
208 
209 	/* No need to close rndis filter if it is removed already */
210 	if (!nvdev)
211 		return 0;
212 
213 	ret = rndis_filter_close(nvdev);
214 	if (ret != 0) {
215 		netdev_err(net, "unable to close device (ret %d).\n", ret);
216 		return ret;
217 	}
218 
219 	ret = netvsc_wait_until_empty(nvdev);
220 	if (ret)
221 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
222 
223 	if (vf_netdev)
224 		dev_close(vf_netdev);
225 
226 	return ret;
227 }
228 
229 static inline void *init_ppi_data(struct rndis_message *msg,
230 				  u32 ppi_size, u32 pkt_type)
231 {
232 	struct rndis_packet *rndis_pkt = &msg->msg.pkt;
233 	struct rndis_per_packet_info *ppi;
234 
235 	rndis_pkt->data_offset += ppi_size;
236 	ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
237 		+ rndis_pkt->per_pkt_info_len;
238 
239 	ppi->size = ppi_size;
240 	ppi->type = pkt_type;
241 	ppi->internal = 0;
242 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
243 
244 	rndis_pkt->per_pkt_info_len += ppi_size;
245 
246 	return ppi + 1;
247 }
248 
249 static inline int netvsc_get_tx_queue(struct net_device *ndev,
250 				      struct sk_buff *skb, int old_idx)
251 {
252 	const struct net_device_context *ndc = netdev_priv(ndev);
253 	struct sock *sk = skb->sk;
254 	int q_idx;
255 
256 	q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
257 			      (VRSS_SEND_TAB_SIZE - 1)];
258 
259 	/* If queue index changed record the new value */
260 	if (q_idx != old_idx &&
261 	    sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
262 		sk_tx_queue_set(sk, q_idx);
263 
264 	return q_idx;
265 }
266 
267 /*
268  * Select queue for transmit.
269  *
270  * If a valid queue has already been assigned, then use that.
271  * Otherwise compute tx queue based on hash and the send table.
272  *
273  * This is basically similar to default (netdev_pick_tx) with the added step
274  * of using the host send_table when no other queue has been assigned.
275  *
276  * TODO support XPS - but get_xps_queue not exported
277  */
278 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
279 {
280 	int q_idx = sk_tx_queue_get(skb->sk);
281 
282 	if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
283 		/* If forwarding a packet, we use the recorded queue when
284 		 * available for better cache locality.
285 		 */
286 		if (skb_rx_queue_recorded(skb))
287 			q_idx = skb_get_rx_queue(skb);
288 		else
289 			q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
290 	}
291 
292 	return q_idx;
293 }
294 
295 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
296 			       struct net_device *sb_dev)
297 {
298 	struct net_device_context *ndc = netdev_priv(ndev);
299 	struct net_device *vf_netdev;
300 	u16 txq;
301 
302 	rcu_read_lock();
303 	vf_netdev = rcu_dereference(ndc->vf_netdev);
304 	if (vf_netdev) {
305 		const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
306 
307 		if (vf_ops->ndo_select_queue)
308 			txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
309 		else
310 			txq = netdev_pick_tx(vf_netdev, skb, NULL);
311 
312 		/* Record the queue selected by VF so that it can be
313 		 * used for common case where VF has more queues than
314 		 * the synthetic device.
315 		 */
316 		qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
317 	} else {
318 		txq = netvsc_pick_tx(ndev, skb);
319 	}
320 	rcu_read_unlock();
321 
322 	while (txq >= ndev->real_num_tx_queues)
323 		txq -= ndev->real_num_tx_queues;
324 
325 	return txq;
326 }
327 
328 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
329 		       struct hv_page_buffer *pb)
330 {
331 	int j = 0;
332 
333 	hvpfn += offset >> HV_HYP_PAGE_SHIFT;
334 	offset = offset & ~HV_HYP_PAGE_MASK;
335 
336 	while (len > 0) {
337 		unsigned long bytes;
338 
339 		bytes = HV_HYP_PAGE_SIZE - offset;
340 		if (bytes > len)
341 			bytes = len;
342 		pb[j].pfn = hvpfn;
343 		pb[j].offset = offset;
344 		pb[j].len = bytes;
345 
346 		offset += bytes;
347 		len -= bytes;
348 
349 		if (offset == HV_HYP_PAGE_SIZE && len) {
350 			hvpfn++;
351 			offset = 0;
352 			j++;
353 		}
354 	}
355 
356 	return j + 1;
357 }
358 
359 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
360 			   struct hv_netvsc_packet *packet,
361 			   struct hv_page_buffer *pb)
362 {
363 	u32 slots_used = 0;
364 	char *data = skb->data;
365 	int frags = skb_shinfo(skb)->nr_frags;
366 	int i;
367 
368 	/* The packet is laid out thus:
369 	 * 1. hdr: RNDIS header and PPI
370 	 * 2. skb linear data
371 	 * 3. skb fragment data
372 	 */
373 	slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
374 				  offset_in_hvpage(hdr),
375 				  len,
376 				  &pb[slots_used]);
377 
378 	packet->rmsg_size = len;
379 	packet->rmsg_pgcnt = slots_used;
380 
381 	slots_used += fill_pg_buf(virt_to_hvpfn(data),
382 				  offset_in_hvpage(data),
383 				  skb_headlen(skb),
384 				  &pb[slots_used]);
385 
386 	for (i = 0; i < frags; i++) {
387 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
388 
389 		slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
390 					  skb_frag_off(frag),
391 					  skb_frag_size(frag),
392 					  &pb[slots_used]);
393 	}
394 	return slots_used;
395 }
396 
397 static int count_skb_frag_slots(struct sk_buff *skb)
398 {
399 	int i, frags = skb_shinfo(skb)->nr_frags;
400 	int pages = 0;
401 
402 	for (i = 0; i < frags; i++) {
403 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
404 		unsigned long size = skb_frag_size(frag);
405 		unsigned long offset = skb_frag_off(frag);
406 
407 		/* Skip unused frames from start of page */
408 		offset &= ~HV_HYP_PAGE_MASK;
409 		pages += HVPFN_UP(offset + size);
410 	}
411 	return pages;
412 }
413 
414 static int netvsc_get_slots(struct sk_buff *skb)
415 {
416 	char *data = skb->data;
417 	unsigned int offset = offset_in_hvpage(data);
418 	unsigned int len = skb_headlen(skb);
419 	int slots;
420 	int frag_slots;
421 
422 	slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
423 	frag_slots = count_skb_frag_slots(skb);
424 	return slots + frag_slots;
425 }
426 
427 static u32 net_checksum_info(struct sk_buff *skb)
428 {
429 	if (skb->protocol == htons(ETH_P_IP)) {
430 		struct iphdr *ip = ip_hdr(skb);
431 
432 		if (ip->protocol == IPPROTO_TCP)
433 			return TRANSPORT_INFO_IPV4_TCP;
434 		else if (ip->protocol == IPPROTO_UDP)
435 			return TRANSPORT_INFO_IPV4_UDP;
436 	} else {
437 		struct ipv6hdr *ip6 = ipv6_hdr(skb);
438 
439 		if (ip6->nexthdr == IPPROTO_TCP)
440 			return TRANSPORT_INFO_IPV6_TCP;
441 		else if (ip6->nexthdr == IPPROTO_UDP)
442 			return TRANSPORT_INFO_IPV6_UDP;
443 	}
444 
445 	return TRANSPORT_INFO_NOT_IP;
446 }
447 
448 /* Send skb on the slave VF device. */
449 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
450 			  struct sk_buff *skb)
451 {
452 	struct net_device_context *ndev_ctx = netdev_priv(net);
453 	unsigned int len = skb->len;
454 	int rc;
455 
456 	skb->dev = vf_netdev;
457 	skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
458 
459 	rc = dev_queue_xmit(skb);
460 	if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
461 		struct netvsc_vf_pcpu_stats *pcpu_stats
462 			= this_cpu_ptr(ndev_ctx->vf_stats);
463 
464 		u64_stats_update_begin(&pcpu_stats->syncp);
465 		pcpu_stats->tx_packets++;
466 		pcpu_stats->tx_bytes += len;
467 		u64_stats_update_end(&pcpu_stats->syncp);
468 	} else {
469 		this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
470 	}
471 
472 	return rc;
473 }
474 
475 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
476 {
477 	struct net_device_context *net_device_ctx = netdev_priv(net);
478 	struct hv_netvsc_packet *packet = NULL;
479 	int ret;
480 	unsigned int num_data_pgs;
481 	struct rndis_message *rndis_msg;
482 	struct net_device *vf_netdev;
483 	u32 rndis_msg_size;
484 	u32 hash;
485 	struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
486 
487 	/* If VF is present and up then redirect packets to it.
488 	 * Skip the VF if it is marked down or has no carrier.
489 	 * If netpoll is in uses, then VF can not be used either.
490 	 */
491 	vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
492 	if (vf_netdev && netif_running(vf_netdev) &&
493 	    netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
494 	    net_device_ctx->data_path_is_vf)
495 		return netvsc_vf_xmit(net, vf_netdev, skb);
496 
497 	/* We will atmost need two pages to describe the rndis
498 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
499 	 * of pages in a single packet. If skb is scattered around
500 	 * more pages we try linearizing it.
501 	 */
502 
503 	num_data_pgs = netvsc_get_slots(skb) + 2;
504 
505 	if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
506 		++net_device_ctx->eth_stats.tx_scattered;
507 
508 		if (skb_linearize(skb))
509 			goto no_memory;
510 
511 		num_data_pgs = netvsc_get_slots(skb) + 2;
512 		if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
513 			++net_device_ctx->eth_stats.tx_too_big;
514 			goto drop;
515 		}
516 	}
517 
518 	/*
519 	 * Place the rndis header in the skb head room and
520 	 * the skb->cb will be used for hv_netvsc_packet
521 	 * structure.
522 	 */
523 	ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
524 	if (ret)
525 		goto no_memory;
526 
527 	/* Use the skb control buffer for building up the packet */
528 	BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
529 			sizeof_field(struct sk_buff, cb));
530 	packet = (struct hv_netvsc_packet *)skb->cb;
531 
532 	packet->q_idx = skb_get_queue_mapping(skb);
533 
534 	packet->total_data_buflen = skb->len;
535 	packet->total_bytes = skb->len;
536 	packet->total_packets = 1;
537 
538 	rndis_msg = (struct rndis_message *)skb->head;
539 
540 	/* Add the rndis header */
541 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
542 	rndis_msg->msg_len = packet->total_data_buflen;
543 
544 	rndis_msg->msg.pkt = (struct rndis_packet) {
545 		.data_offset = sizeof(struct rndis_packet),
546 		.data_len = packet->total_data_buflen,
547 		.per_pkt_info_offset = sizeof(struct rndis_packet),
548 	};
549 
550 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
551 
552 	hash = skb_get_hash_raw(skb);
553 	if (hash != 0 && net->real_num_tx_queues > 1) {
554 		u32 *hash_info;
555 
556 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
557 		hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
558 					  NBL_HASH_VALUE);
559 		*hash_info = hash;
560 	}
561 
562 	/* When using AF_PACKET we need to drop VLAN header from
563 	 * the frame and update the SKB to allow the HOST OS
564 	 * to transmit the 802.1Q packet
565 	 */
566 	if (skb->protocol == htons(ETH_P_8021Q)) {
567 		u16 vlan_tci;
568 
569 		skb_reset_mac_header(skb);
570 		if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
571 			if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
572 				++net_device_ctx->eth_stats.vlan_error;
573 				goto drop;
574 			}
575 
576 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
577 			/* Update the NDIS header pkt lengths */
578 			packet->total_data_buflen -= VLAN_HLEN;
579 			packet->total_bytes -= VLAN_HLEN;
580 			rndis_msg->msg_len = packet->total_data_buflen;
581 			rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
582 		}
583 	}
584 
585 	if (skb_vlan_tag_present(skb)) {
586 		struct ndis_pkt_8021q_info *vlan;
587 
588 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
589 		vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
590 				     IEEE_8021Q_INFO);
591 
592 		vlan->value = 0;
593 		vlan->vlanid = skb_vlan_tag_get_id(skb);
594 		vlan->cfi = skb_vlan_tag_get_cfi(skb);
595 		vlan->pri = skb_vlan_tag_get_prio(skb);
596 	}
597 
598 	if (skb_is_gso(skb)) {
599 		struct ndis_tcp_lso_info *lso_info;
600 
601 		rndis_msg_size += NDIS_LSO_PPI_SIZE;
602 		lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
603 					 TCP_LARGESEND_PKTINFO);
604 
605 		lso_info->value = 0;
606 		lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
607 		if (skb->protocol == htons(ETH_P_IP)) {
608 			lso_info->lso_v2_transmit.ip_version =
609 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
610 			ip_hdr(skb)->tot_len = 0;
611 			ip_hdr(skb)->check = 0;
612 			tcp_hdr(skb)->check =
613 				~csum_tcpudp_magic(ip_hdr(skb)->saddr,
614 						   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
615 		} else {
616 			lso_info->lso_v2_transmit.ip_version =
617 				NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
618 			tcp_v6_gso_csum_prep(skb);
619 		}
620 		lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
621 		lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
622 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
623 		if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
624 			struct ndis_tcp_ip_checksum_info *csum_info;
625 
626 			rndis_msg_size += NDIS_CSUM_PPI_SIZE;
627 			csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
628 						  TCPIP_CHKSUM_PKTINFO);
629 
630 			csum_info->value = 0;
631 			csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
632 
633 			if (skb->protocol == htons(ETH_P_IP)) {
634 				csum_info->transmit.is_ipv4 = 1;
635 
636 				if (ip_hdr(skb)->protocol == IPPROTO_TCP)
637 					csum_info->transmit.tcp_checksum = 1;
638 				else
639 					csum_info->transmit.udp_checksum = 1;
640 			} else {
641 				csum_info->transmit.is_ipv6 = 1;
642 
643 				if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
644 					csum_info->transmit.tcp_checksum = 1;
645 				else
646 					csum_info->transmit.udp_checksum = 1;
647 			}
648 		} else {
649 			/* Can't do offload of this type of checksum */
650 			if (skb_checksum_help(skb))
651 				goto drop;
652 		}
653 	}
654 
655 	/* Start filling in the page buffers with the rndis hdr */
656 	rndis_msg->msg_len += rndis_msg_size;
657 	packet->total_data_buflen = rndis_msg->msg_len;
658 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
659 					       skb, packet, pb);
660 
661 	/* timestamp packet in software */
662 	skb_tx_timestamp(skb);
663 
664 	ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
665 	if (likely(ret == 0))
666 		return NETDEV_TX_OK;
667 
668 	if (ret == -EAGAIN) {
669 		++net_device_ctx->eth_stats.tx_busy;
670 		return NETDEV_TX_BUSY;
671 	}
672 
673 	if (ret == -ENOSPC)
674 		++net_device_ctx->eth_stats.tx_no_space;
675 
676 drop:
677 	dev_kfree_skb_any(skb);
678 	net->stats.tx_dropped++;
679 
680 	return NETDEV_TX_OK;
681 
682 no_memory:
683 	++net_device_ctx->eth_stats.tx_no_memory;
684 	goto drop;
685 }
686 
687 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
688 				     struct net_device *ndev)
689 {
690 	return netvsc_xmit(skb, ndev, false);
691 }
692 
693 /*
694  * netvsc_linkstatus_callback - Link up/down notification
695  */
696 void netvsc_linkstatus_callback(struct net_device *net,
697 				struct rndis_message *resp,
698 				void *data, u32 data_buflen)
699 {
700 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
701 	struct net_device_context *ndev_ctx = netdev_priv(net);
702 	struct netvsc_reconfig *event;
703 	unsigned long flags;
704 
705 	/* Ensure the packet is big enough to access its fields */
706 	if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
707 		netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
708 			   resp->msg_len);
709 		return;
710 	}
711 
712 	/* Copy the RNDIS indicate status into nvchan->recv_buf */
713 	memcpy(indicate, data + RNDIS_HEADER_SIZE, sizeof(*indicate));
714 
715 	/* Update the physical link speed when changing to another vSwitch */
716 	if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
717 		u32 speed;
718 
719 		/* Validate status_buf_offset and status_buflen.
720 		 *
721 		 * Certain (pre-Fe) implementations of Hyper-V's vSwitch didn't account
722 		 * for the status buffer field in resp->msg_len; perform the validation
723 		 * using data_buflen (>= resp->msg_len).
724 		 */
725 		if (indicate->status_buflen < sizeof(speed) ||
726 		    indicate->status_buf_offset < sizeof(*indicate) ||
727 		    data_buflen - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
728 		    data_buflen - RNDIS_HEADER_SIZE - indicate->status_buf_offset
729 				< indicate->status_buflen) {
730 			netdev_err(net, "invalid rndis_indicate_status packet\n");
731 			return;
732 		}
733 
734 		speed = *(u32 *)(data + RNDIS_HEADER_SIZE + indicate->status_buf_offset) / 10000;
735 		ndev_ctx->speed = speed;
736 		return;
737 	}
738 
739 	/* Handle these link change statuses below */
740 	if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
741 	    indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
742 	    indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
743 		return;
744 
745 	if (net->reg_state != NETREG_REGISTERED)
746 		return;
747 
748 	event = kzalloc(sizeof(*event), GFP_ATOMIC);
749 	if (!event)
750 		return;
751 	event->event = indicate->status;
752 
753 	spin_lock_irqsave(&ndev_ctx->lock, flags);
754 	list_add_tail(&event->list, &ndev_ctx->reconfig_events);
755 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
756 
757 	schedule_delayed_work(&ndev_ctx->dwork, 0);
758 }
759 
760 /* This function should only be called after skb_record_rx_queue() */
761 void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
762 {
763 	int rc;
764 
765 	skb->queue_mapping = skb_get_rx_queue(skb);
766 	__skb_push(skb, ETH_HLEN);
767 
768 	rc = netvsc_xmit(skb, ndev, true);
769 
770 	if (dev_xmit_complete(rc))
771 		return;
772 
773 	dev_kfree_skb_any(skb);
774 	ndev->stats.tx_dropped++;
775 }
776 
777 static void netvsc_comp_ipcsum(struct sk_buff *skb)
778 {
779 	struct iphdr *iph = (struct iphdr *)skb->data;
780 
781 	iph->check = 0;
782 	iph->check = ip_fast_csum(iph, iph->ihl);
783 }
784 
785 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
786 					     struct netvsc_channel *nvchan,
787 					     struct xdp_buff *xdp)
788 {
789 	struct napi_struct *napi = &nvchan->napi;
790 	const struct ndis_pkt_8021q_info *vlan = &nvchan->rsc.vlan;
791 	const struct ndis_tcp_ip_checksum_info *csum_info =
792 						&nvchan->rsc.csum_info;
793 	const u32 *hash_info = &nvchan->rsc.hash_info;
794 	u8 ppi_flags = nvchan->rsc.ppi_flags;
795 	struct sk_buff *skb;
796 	void *xbuf = xdp->data_hard_start;
797 	int i;
798 
799 	if (xbuf) {
800 		unsigned int hdroom = xdp->data - xdp->data_hard_start;
801 		unsigned int xlen = xdp->data_end - xdp->data;
802 		unsigned int frag_size = xdp->frame_sz;
803 
804 		skb = build_skb(xbuf, frag_size);
805 
806 		if (!skb) {
807 			__free_page(virt_to_page(xbuf));
808 			return NULL;
809 		}
810 
811 		skb_reserve(skb, hdroom);
812 		skb_put(skb, xlen);
813 		skb->dev = napi->dev;
814 	} else {
815 		skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
816 
817 		if (!skb)
818 			return NULL;
819 
820 		/* Copy to skb. This copy is needed here since the memory
821 		 * pointed by hv_netvsc_packet cannot be deallocated.
822 		 */
823 		for (i = 0; i < nvchan->rsc.cnt; i++)
824 			skb_put_data(skb, nvchan->rsc.data[i],
825 				     nvchan->rsc.len[i]);
826 	}
827 
828 	skb->protocol = eth_type_trans(skb, net);
829 
830 	/* skb is already created with CHECKSUM_NONE */
831 	skb_checksum_none_assert(skb);
832 
833 	/* Incoming packets may have IP header checksum verified by the host.
834 	 * They may not have IP header checksum computed after coalescing.
835 	 * We compute it here if the flags are set, because on Linux, the IP
836 	 * checksum is always checked.
837 	 */
838 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && csum_info->receive.ip_checksum_value_invalid &&
839 	    csum_info->receive.ip_checksum_succeeded &&
840 	    skb->protocol == htons(ETH_P_IP)) {
841 		/* Check that there is enough space to hold the IP header. */
842 		if (skb_headlen(skb) < sizeof(struct iphdr)) {
843 			kfree_skb(skb);
844 			return NULL;
845 		}
846 		netvsc_comp_ipcsum(skb);
847 	}
848 
849 	/* Do L4 checksum offload if enabled and present. */
850 	if ((ppi_flags & NVSC_RSC_CSUM_INFO) && (net->features & NETIF_F_RXCSUM)) {
851 		if (csum_info->receive.tcp_checksum_succeeded ||
852 		    csum_info->receive.udp_checksum_succeeded)
853 			skb->ip_summed = CHECKSUM_UNNECESSARY;
854 	}
855 
856 	if ((ppi_flags & NVSC_RSC_HASH_INFO) && (net->features & NETIF_F_RXHASH))
857 		skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
858 
859 	if (ppi_flags & NVSC_RSC_VLAN) {
860 		u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
861 			(vlan->cfi ? VLAN_CFI_MASK : 0);
862 
863 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
864 				       vlan_tci);
865 	}
866 
867 	return skb;
868 }
869 
870 /*
871  * netvsc_recv_callback -  Callback when we receive a packet from the
872  * "wire" on the specified device.
873  */
874 int netvsc_recv_callback(struct net_device *net,
875 			 struct netvsc_device *net_device,
876 			 struct netvsc_channel *nvchan)
877 {
878 	struct net_device_context *net_device_ctx = netdev_priv(net);
879 	struct vmbus_channel *channel = nvchan->channel;
880 	u16 q_idx = channel->offermsg.offer.sub_channel_index;
881 	struct sk_buff *skb;
882 	struct netvsc_stats_rx *rx_stats = &nvchan->rx_stats;
883 	struct xdp_buff xdp;
884 	u32 act;
885 
886 	if (net->reg_state != NETREG_REGISTERED)
887 		return NVSP_STAT_FAIL;
888 
889 	act = netvsc_run_xdp(net, nvchan, &xdp);
890 
891 	if (act == XDP_REDIRECT)
892 		return NVSP_STAT_SUCCESS;
893 
894 	if (act != XDP_PASS && act != XDP_TX) {
895 		u64_stats_update_begin(&rx_stats->syncp);
896 		rx_stats->xdp_drop++;
897 		u64_stats_update_end(&rx_stats->syncp);
898 
899 		return NVSP_STAT_SUCCESS; /* consumed by XDP */
900 	}
901 
902 	/* Allocate a skb - TODO direct I/O to pages? */
903 	skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
904 
905 	if (unlikely(!skb)) {
906 		++net_device_ctx->eth_stats.rx_no_memory;
907 		return NVSP_STAT_FAIL;
908 	}
909 
910 	skb_record_rx_queue(skb, q_idx);
911 
912 	/*
913 	 * Even if injecting the packet, record the statistics
914 	 * on the synthetic device because modifying the VF device
915 	 * statistics will not work correctly.
916 	 */
917 	u64_stats_update_begin(&rx_stats->syncp);
918 	if (act == XDP_TX)
919 		rx_stats->xdp_tx++;
920 
921 	rx_stats->packets++;
922 	rx_stats->bytes += nvchan->rsc.pktlen;
923 
924 	if (skb->pkt_type == PACKET_BROADCAST)
925 		++rx_stats->broadcast;
926 	else if (skb->pkt_type == PACKET_MULTICAST)
927 		++rx_stats->multicast;
928 	u64_stats_update_end(&rx_stats->syncp);
929 
930 	if (act == XDP_TX) {
931 		netvsc_xdp_xmit(skb, net);
932 		return NVSP_STAT_SUCCESS;
933 	}
934 
935 	napi_gro_receive(&nvchan->napi, skb);
936 	return NVSP_STAT_SUCCESS;
937 }
938 
939 static void netvsc_get_drvinfo(struct net_device *net,
940 			       struct ethtool_drvinfo *info)
941 {
942 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
943 	strscpy(info->fw_version, "N/A", sizeof(info->fw_version));
944 }
945 
946 static void netvsc_get_channels(struct net_device *net,
947 				struct ethtool_channels *channel)
948 {
949 	struct net_device_context *net_device_ctx = netdev_priv(net);
950 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
951 
952 	if (nvdev) {
953 		channel->max_combined	= nvdev->max_chn;
954 		channel->combined_count = nvdev->num_chn;
955 	}
956 }
957 
958 /* Alloc struct netvsc_device_info, and initialize it from either existing
959  * struct netvsc_device, or from default values.
960  */
961 static
962 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
963 {
964 	struct netvsc_device_info *dev_info;
965 	struct bpf_prog *prog;
966 
967 	dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
968 
969 	if (!dev_info)
970 		return NULL;
971 
972 	if (nvdev) {
973 		ASSERT_RTNL();
974 
975 		dev_info->num_chn = nvdev->num_chn;
976 		dev_info->send_sections = nvdev->send_section_cnt;
977 		dev_info->send_section_size = nvdev->send_section_size;
978 		dev_info->recv_sections = nvdev->recv_section_cnt;
979 		dev_info->recv_section_size = nvdev->recv_section_size;
980 
981 		memcpy(dev_info->rss_key, nvdev->extension->rss_key,
982 		       NETVSC_HASH_KEYLEN);
983 
984 		prog = netvsc_xdp_get(nvdev);
985 		if (prog) {
986 			bpf_prog_inc(prog);
987 			dev_info->bprog = prog;
988 		}
989 	} else {
990 		dev_info->num_chn = max(VRSS_CHANNEL_DEFAULT,
991 					netif_get_num_default_rss_queues());
992 		dev_info->send_sections = NETVSC_DEFAULT_TX;
993 		dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
994 		dev_info->recv_sections = NETVSC_DEFAULT_RX;
995 		dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
996 	}
997 
998 	return dev_info;
999 }
1000 
1001 /* Free struct netvsc_device_info */
1002 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1003 {
1004 	if (dev_info->bprog) {
1005 		ASSERT_RTNL();
1006 		bpf_prog_put(dev_info->bprog);
1007 	}
1008 
1009 	kfree(dev_info);
1010 }
1011 
1012 static int netvsc_detach(struct net_device *ndev,
1013 			 struct netvsc_device *nvdev)
1014 {
1015 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1016 	struct hv_device *hdev = ndev_ctx->device_ctx;
1017 	int ret;
1018 
1019 	/* Don't try continuing to try and setup sub channels */
1020 	if (cancel_work_sync(&nvdev->subchan_work))
1021 		nvdev->num_chn = 1;
1022 
1023 	netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1024 
1025 	/* If device was up (receiving) then shutdown */
1026 	if (netif_running(ndev)) {
1027 		netvsc_tx_disable(nvdev, ndev);
1028 
1029 		ret = rndis_filter_close(nvdev);
1030 		if (ret) {
1031 			netdev_err(ndev,
1032 				   "unable to close device (ret %d).\n", ret);
1033 			return ret;
1034 		}
1035 
1036 		ret = netvsc_wait_until_empty(nvdev);
1037 		if (ret) {
1038 			netdev_err(ndev,
1039 				   "Ring buffer not empty after closing rndis\n");
1040 			return ret;
1041 		}
1042 	}
1043 
1044 	netif_device_detach(ndev);
1045 
1046 	rndis_filter_device_remove(hdev, nvdev);
1047 
1048 	return 0;
1049 }
1050 
1051 static int netvsc_attach(struct net_device *ndev,
1052 			 struct netvsc_device_info *dev_info)
1053 {
1054 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1055 	struct hv_device *hdev = ndev_ctx->device_ctx;
1056 	struct netvsc_device *nvdev;
1057 	struct rndis_device *rdev;
1058 	struct bpf_prog *prog;
1059 	int ret = 0;
1060 
1061 	nvdev = rndis_filter_device_add(hdev, dev_info);
1062 	if (IS_ERR(nvdev))
1063 		return PTR_ERR(nvdev);
1064 
1065 	if (nvdev->num_chn > 1) {
1066 		ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1067 
1068 		/* if unavailable, just proceed with one queue */
1069 		if (ret) {
1070 			nvdev->max_chn = 1;
1071 			nvdev->num_chn = 1;
1072 		}
1073 	}
1074 
1075 	prog = dev_info->bprog;
1076 	if (prog) {
1077 		bpf_prog_inc(prog);
1078 		ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1079 		if (ret) {
1080 			bpf_prog_put(prog);
1081 			goto err1;
1082 		}
1083 	}
1084 
1085 	/* In any case device is now ready */
1086 	nvdev->tx_disable = false;
1087 	netif_device_attach(ndev);
1088 
1089 	/* Note: enable and attach happen when sub-channels setup */
1090 	netif_carrier_off(ndev);
1091 
1092 	if (netif_running(ndev)) {
1093 		ret = rndis_filter_open(nvdev);
1094 		if (ret)
1095 			goto err2;
1096 
1097 		rdev = nvdev->extension;
1098 		if (!rdev->link_state)
1099 			netif_carrier_on(ndev);
1100 	}
1101 
1102 	return 0;
1103 
1104 err2:
1105 	netif_device_detach(ndev);
1106 
1107 err1:
1108 	rndis_filter_device_remove(hdev, nvdev);
1109 
1110 	return ret;
1111 }
1112 
1113 static int netvsc_set_channels(struct net_device *net,
1114 			       struct ethtool_channels *channels)
1115 {
1116 	struct net_device_context *net_device_ctx = netdev_priv(net);
1117 	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1118 	unsigned int orig, count = channels->combined_count;
1119 	struct netvsc_device_info *device_info;
1120 	int ret;
1121 
1122 	/* We do not support separate count for rx, tx, or other */
1123 	if (count == 0 ||
1124 	    channels->rx_count || channels->tx_count || channels->other_count)
1125 		return -EINVAL;
1126 
1127 	if (!nvdev || nvdev->destroy)
1128 		return -ENODEV;
1129 
1130 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1131 		return -EINVAL;
1132 
1133 	if (count > nvdev->max_chn)
1134 		return -EINVAL;
1135 
1136 	orig = nvdev->num_chn;
1137 
1138 	device_info = netvsc_devinfo_get(nvdev);
1139 
1140 	if (!device_info)
1141 		return -ENOMEM;
1142 
1143 	device_info->num_chn = count;
1144 
1145 	ret = netvsc_detach(net, nvdev);
1146 	if (ret)
1147 		goto out;
1148 
1149 	ret = netvsc_attach(net, device_info);
1150 	if (ret) {
1151 		device_info->num_chn = orig;
1152 		if (netvsc_attach(net, device_info))
1153 			netdev_err(net, "restoring channel setting failed\n");
1154 	}
1155 
1156 out:
1157 	netvsc_devinfo_put(device_info);
1158 	return ret;
1159 }
1160 
1161 static void netvsc_init_settings(struct net_device *dev)
1162 {
1163 	struct net_device_context *ndc = netdev_priv(dev);
1164 
1165 	ndc->l4_hash = HV_DEFAULT_L4HASH;
1166 
1167 	ndc->speed = SPEED_UNKNOWN;
1168 	ndc->duplex = DUPLEX_FULL;
1169 
1170 	dev->features = NETIF_F_LRO;
1171 }
1172 
1173 static int netvsc_get_link_ksettings(struct net_device *dev,
1174 				     struct ethtool_link_ksettings *cmd)
1175 {
1176 	struct net_device_context *ndc = netdev_priv(dev);
1177 	struct net_device *vf_netdev;
1178 
1179 	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1180 
1181 	if (vf_netdev)
1182 		return __ethtool_get_link_ksettings(vf_netdev, cmd);
1183 
1184 	cmd->base.speed = ndc->speed;
1185 	cmd->base.duplex = ndc->duplex;
1186 	cmd->base.port = PORT_OTHER;
1187 
1188 	return 0;
1189 }
1190 
1191 static int netvsc_set_link_ksettings(struct net_device *dev,
1192 				     const struct ethtool_link_ksettings *cmd)
1193 {
1194 	struct net_device_context *ndc = netdev_priv(dev);
1195 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1196 
1197 	if (vf_netdev) {
1198 		if (!vf_netdev->ethtool_ops->set_link_ksettings)
1199 			return -EOPNOTSUPP;
1200 
1201 		return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1202 								  cmd);
1203 	}
1204 
1205 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1206 						  &ndc->speed, &ndc->duplex);
1207 }
1208 
1209 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1210 {
1211 	struct net_device_context *ndevctx = netdev_priv(ndev);
1212 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1213 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1214 	int orig_mtu = ndev->mtu;
1215 	struct netvsc_device_info *device_info;
1216 	int ret = 0;
1217 
1218 	if (!nvdev || nvdev->destroy)
1219 		return -ENODEV;
1220 
1221 	device_info = netvsc_devinfo_get(nvdev);
1222 
1223 	if (!device_info)
1224 		return -ENOMEM;
1225 
1226 	/* Change MTU of underlying VF netdev first. */
1227 	if (vf_netdev) {
1228 		ret = dev_set_mtu(vf_netdev, mtu);
1229 		if (ret)
1230 			goto out;
1231 	}
1232 
1233 	ret = netvsc_detach(ndev, nvdev);
1234 	if (ret)
1235 		goto rollback_vf;
1236 
1237 	WRITE_ONCE(ndev->mtu, mtu);
1238 
1239 	ret = netvsc_attach(ndev, device_info);
1240 	if (!ret)
1241 		goto out;
1242 
1243 	/* Attempt rollback to original MTU */
1244 	WRITE_ONCE(ndev->mtu, orig_mtu);
1245 
1246 	if (netvsc_attach(ndev, device_info))
1247 		netdev_err(ndev, "restoring mtu failed\n");
1248 rollback_vf:
1249 	if (vf_netdev)
1250 		dev_set_mtu(vf_netdev, orig_mtu);
1251 
1252 out:
1253 	netvsc_devinfo_put(device_info);
1254 	return ret;
1255 }
1256 
1257 static void netvsc_get_vf_stats(struct net_device *net,
1258 				struct netvsc_vf_pcpu_stats *tot)
1259 {
1260 	struct net_device_context *ndev_ctx = netdev_priv(net);
1261 	int i;
1262 
1263 	memset(tot, 0, sizeof(*tot));
1264 
1265 	for_each_possible_cpu(i) {
1266 		const struct netvsc_vf_pcpu_stats *stats
1267 			= per_cpu_ptr(ndev_ctx->vf_stats, i);
1268 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1269 		unsigned int start;
1270 
1271 		do {
1272 			start = u64_stats_fetch_begin(&stats->syncp);
1273 			rx_packets = stats->rx_packets;
1274 			tx_packets = stats->tx_packets;
1275 			rx_bytes = stats->rx_bytes;
1276 			tx_bytes = stats->tx_bytes;
1277 		} while (u64_stats_fetch_retry(&stats->syncp, start));
1278 
1279 		tot->rx_packets += rx_packets;
1280 		tot->tx_packets += tx_packets;
1281 		tot->rx_bytes   += rx_bytes;
1282 		tot->tx_bytes   += tx_bytes;
1283 		tot->tx_dropped += stats->tx_dropped;
1284 	}
1285 }
1286 
1287 static void netvsc_get_pcpu_stats(struct net_device *net,
1288 				  struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1289 {
1290 	struct net_device_context *ndev_ctx = netdev_priv(net);
1291 	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1292 	int i;
1293 
1294 	/* fetch percpu stats of vf */
1295 	for_each_possible_cpu(i) {
1296 		const struct netvsc_vf_pcpu_stats *stats =
1297 			per_cpu_ptr(ndev_ctx->vf_stats, i);
1298 		struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1299 		unsigned int start;
1300 
1301 		do {
1302 			start = u64_stats_fetch_begin(&stats->syncp);
1303 			this_tot->vf_rx_packets = stats->rx_packets;
1304 			this_tot->vf_tx_packets = stats->tx_packets;
1305 			this_tot->vf_rx_bytes = stats->rx_bytes;
1306 			this_tot->vf_tx_bytes = stats->tx_bytes;
1307 		} while (u64_stats_fetch_retry(&stats->syncp, start));
1308 		this_tot->rx_packets = this_tot->vf_rx_packets;
1309 		this_tot->tx_packets = this_tot->vf_tx_packets;
1310 		this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1311 		this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1312 	}
1313 
1314 	/* fetch percpu stats of netvsc */
1315 	for (i = 0; i < nvdev->num_chn; i++) {
1316 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1317 		const struct netvsc_stats_tx *tx_stats;
1318 		const struct netvsc_stats_rx *rx_stats;
1319 		struct netvsc_ethtool_pcpu_stats *this_tot =
1320 			&pcpu_tot[nvchan->channel->target_cpu];
1321 		u64 packets, bytes;
1322 		unsigned int start;
1323 
1324 		tx_stats = &nvchan->tx_stats;
1325 		do {
1326 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1327 			packets = tx_stats->packets;
1328 			bytes = tx_stats->bytes;
1329 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1330 
1331 		this_tot->tx_bytes	+= bytes;
1332 		this_tot->tx_packets	+= packets;
1333 
1334 		rx_stats = &nvchan->rx_stats;
1335 		do {
1336 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1337 			packets = rx_stats->packets;
1338 			bytes = rx_stats->bytes;
1339 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1340 
1341 		this_tot->rx_bytes	+= bytes;
1342 		this_tot->rx_packets	+= packets;
1343 	}
1344 }
1345 
1346 static void netvsc_get_stats64(struct net_device *net,
1347 			       struct rtnl_link_stats64 *t)
1348 {
1349 	struct net_device_context *ndev_ctx = netdev_priv(net);
1350 	struct netvsc_device *nvdev;
1351 	struct netvsc_vf_pcpu_stats vf_tot;
1352 	int i;
1353 
1354 	rcu_read_lock();
1355 
1356 	nvdev = rcu_dereference(ndev_ctx->nvdev);
1357 	if (!nvdev)
1358 		goto out;
1359 
1360 	netdev_stats_to_stats64(t, &net->stats);
1361 
1362 	netvsc_get_vf_stats(net, &vf_tot);
1363 	t->rx_packets += vf_tot.rx_packets;
1364 	t->tx_packets += vf_tot.tx_packets;
1365 	t->rx_bytes   += vf_tot.rx_bytes;
1366 	t->tx_bytes   += vf_tot.tx_bytes;
1367 	t->tx_dropped += vf_tot.tx_dropped;
1368 
1369 	for (i = 0; i < nvdev->num_chn; i++) {
1370 		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1371 		const struct netvsc_stats_tx *tx_stats;
1372 		const struct netvsc_stats_rx *rx_stats;
1373 		u64 packets, bytes, multicast;
1374 		unsigned int start;
1375 
1376 		tx_stats = &nvchan->tx_stats;
1377 		do {
1378 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1379 			packets = tx_stats->packets;
1380 			bytes = tx_stats->bytes;
1381 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1382 
1383 		t->tx_bytes	+= bytes;
1384 		t->tx_packets	+= packets;
1385 
1386 		rx_stats = &nvchan->rx_stats;
1387 		do {
1388 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1389 			packets = rx_stats->packets;
1390 			bytes = rx_stats->bytes;
1391 			multicast = rx_stats->multicast + rx_stats->broadcast;
1392 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1393 
1394 		t->rx_bytes	+= bytes;
1395 		t->rx_packets	+= packets;
1396 		t->multicast	+= multicast;
1397 	}
1398 out:
1399 	rcu_read_unlock();
1400 }
1401 
1402 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1403 {
1404 	struct net_device_context *ndc = netdev_priv(ndev);
1405 	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1406 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1407 	struct sockaddr *addr = p;
1408 	int err;
1409 
1410 	err = eth_prepare_mac_addr_change(ndev, p);
1411 	if (err)
1412 		return err;
1413 
1414 	if (!nvdev)
1415 		return -ENODEV;
1416 
1417 	if (vf_netdev) {
1418 		err = dev_set_mac_address(vf_netdev, addr, NULL);
1419 		if (err)
1420 			return err;
1421 	}
1422 
1423 	err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1424 	if (!err) {
1425 		eth_commit_mac_addr_change(ndev, p);
1426 	} else if (vf_netdev) {
1427 		/* rollback change on VF */
1428 		memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1429 		dev_set_mac_address(vf_netdev, addr, NULL);
1430 	}
1431 
1432 	return err;
1433 }
1434 
1435 static const struct {
1436 	char name[ETH_GSTRING_LEN];
1437 	u16 offset;
1438 } netvsc_stats[] = {
1439 	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1440 	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1441 	{ "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1442 	{ "tx_too_big",	  offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1443 	{ "tx_busy",	  offsetof(struct netvsc_ethtool_stats, tx_busy) },
1444 	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1445 	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1446 	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1447 	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1448 	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1449 	{ "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1450 }, pcpu_stats[] = {
1451 	{ "cpu%u_rx_packets",
1452 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1453 	{ "cpu%u_rx_bytes",
1454 		offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1455 	{ "cpu%u_tx_packets",
1456 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1457 	{ "cpu%u_tx_bytes",
1458 		offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1459 	{ "cpu%u_vf_rx_packets",
1460 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1461 	{ "cpu%u_vf_rx_bytes",
1462 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1463 	{ "cpu%u_vf_tx_packets",
1464 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1465 	{ "cpu%u_vf_tx_bytes",
1466 		offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1467 }, vf_stats[] = {
1468 	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1469 	{ "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1470 	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1471 	{ "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1472 	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1473 };
1474 
1475 #define NETVSC_GLOBAL_STATS_LEN	ARRAY_SIZE(netvsc_stats)
1476 #define NETVSC_VF_STATS_LEN	ARRAY_SIZE(vf_stats)
1477 
1478 /* statistics per queue (rx/tx packets/bytes) */
1479 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1480 
1481 /* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1482 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
1483 
1484 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1485 {
1486 	struct net_device_context *ndc = netdev_priv(dev);
1487 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1488 
1489 	if (!nvdev)
1490 		return -ENODEV;
1491 
1492 	switch (string_set) {
1493 	case ETH_SS_STATS:
1494 		return NETVSC_GLOBAL_STATS_LEN
1495 			+ NETVSC_VF_STATS_LEN
1496 			+ NETVSC_QUEUE_STATS_LEN(nvdev)
1497 			+ NETVSC_PCPU_STATS_LEN;
1498 	default:
1499 		return -EINVAL;
1500 	}
1501 }
1502 
1503 static void netvsc_get_ethtool_stats(struct net_device *dev,
1504 				     struct ethtool_stats *stats, u64 *data)
1505 {
1506 	struct net_device_context *ndc = netdev_priv(dev);
1507 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1508 	const void *nds = &ndc->eth_stats;
1509 	const struct netvsc_stats_tx *tx_stats;
1510 	const struct netvsc_stats_rx *rx_stats;
1511 	struct netvsc_vf_pcpu_stats sum;
1512 	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1513 	unsigned int start;
1514 	u64 packets, bytes;
1515 	u64 xdp_drop;
1516 	u64 xdp_redirect;
1517 	u64 xdp_tx;
1518 	u64 xdp_xmit;
1519 	int i, j, cpu;
1520 
1521 	if (!nvdev)
1522 		return;
1523 
1524 	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1525 		data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1526 
1527 	netvsc_get_vf_stats(dev, &sum);
1528 	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1529 		data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1530 
1531 	for (j = 0; j < nvdev->num_chn; j++) {
1532 		tx_stats = &nvdev->chan_table[j].tx_stats;
1533 
1534 		do {
1535 			start = u64_stats_fetch_begin(&tx_stats->syncp);
1536 			packets = tx_stats->packets;
1537 			bytes = tx_stats->bytes;
1538 			xdp_xmit = tx_stats->xdp_xmit;
1539 		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1540 		data[i++] = packets;
1541 		data[i++] = bytes;
1542 		data[i++] = xdp_xmit;
1543 
1544 		rx_stats = &nvdev->chan_table[j].rx_stats;
1545 		do {
1546 			start = u64_stats_fetch_begin(&rx_stats->syncp);
1547 			packets = rx_stats->packets;
1548 			bytes = rx_stats->bytes;
1549 			xdp_drop = rx_stats->xdp_drop;
1550 			xdp_redirect = rx_stats->xdp_redirect;
1551 			xdp_tx = rx_stats->xdp_tx;
1552 		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1553 		data[i++] = packets;
1554 		data[i++] = bytes;
1555 		data[i++] = xdp_drop;
1556 		data[i++] = xdp_redirect;
1557 		data[i++] = xdp_tx;
1558 	}
1559 
1560 	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1561 				  sizeof(struct netvsc_ethtool_pcpu_stats),
1562 				  GFP_KERNEL);
1563 	if (!pcpu_sum)
1564 		return;
1565 
1566 	netvsc_get_pcpu_stats(dev, pcpu_sum);
1567 	for_each_present_cpu(cpu) {
1568 		struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1569 
1570 		for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1571 			data[i++] = *(u64 *)((void *)this_sum
1572 					     + pcpu_stats[j].offset);
1573 	}
1574 	kvfree(pcpu_sum);
1575 }
1576 
1577 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1578 {
1579 	struct net_device_context *ndc = netdev_priv(dev);
1580 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1581 	u8 *p = data;
1582 	int i, cpu;
1583 
1584 	if (!nvdev)
1585 		return;
1586 
1587 	switch (stringset) {
1588 	case ETH_SS_STATS:
1589 		for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1590 			ethtool_puts(&p, netvsc_stats[i].name);
1591 
1592 		for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1593 			ethtool_puts(&p, vf_stats[i].name);
1594 
1595 		for (i = 0; i < nvdev->num_chn; i++) {
1596 			ethtool_sprintf(&p, "tx_queue_%u_packets", i);
1597 			ethtool_sprintf(&p, "tx_queue_%u_bytes", i);
1598 			ethtool_sprintf(&p, "tx_queue_%u_xdp_xmit", i);
1599 			ethtool_sprintf(&p, "rx_queue_%u_packets", i);
1600 			ethtool_sprintf(&p, "rx_queue_%u_bytes", i);
1601 			ethtool_sprintf(&p, "rx_queue_%u_xdp_drop", i);
1602 			ethtool_sprintf(&p, "rx_queue_%u_xdp_redirect", i);
1603 			ethtool_sprintf(&p, "rx_queue_%u_xdp_tx", i);
1604 		}
1605 
1606 		for_each_present_cpu(cpu) {
1607 			for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1608 				ethtool_sprintf(&p, pcpu_stats[i].name, cpu);
1609 		}
1610 
1611 		break;
1612 	}
1613 }
1614 
1615 static int
1616 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1617 			 struct ethtool_rxnfc *info)
1618 {
1619 	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1620 
1621 	info->data = RXH_IP_SRC | RXH_IP_DST;
1622 
1623 	switch (info->flow_type) {
1624 	case TCP_V4_FLOW:
1625 		if (ndc->l4_hash & HV_TCP4_L4HASH)
1626 			info->data |= l4_flag;
1627 
1628 		break;
1629 
1630 	case TCP_V6_FLOW:
1631 		if (ndc->l4_hash & HV_TCP6_L4HASH)
1632 			info->data |= l4_flag;
1633 
1634 		break;
1635 
1636 	case UDP_V4_FLOW:
1637 		if (ndc->l4_hash & HV_UDP4_L4HASH)
1638 			info->data |= l4_flag;
1639 
1640 		break;
1641 
1642 	case UDP_V6_FLOW:
1643 		if (ndc->l4_hash & HV_UDP6_L4HASH)
1644 			info->data |= l4_flag;
1645 
1646 		break;
1647 
1648 	case IPV4_FLOW:
1649 	case IPV6_FLOW:
1650 		break;
1651 	default:
1652 		info->data = 0;
1653 		break;
1654 	}
1655 
1656 	return 0;
1657 }
1658 
1659 static int
1660 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1661 		 u32 *rules)
1662 {
1663 	struct net_device_context *ndc = netdev_priv(dev);
1664 	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1665 
1666 	if (!nvdev)
1667 		return -ENODEV;
1668 
1669 	switch (info->cmd) {
1670 	case ETHTOOL_GRXRINGS:
1671 		info->data = nvdev->num_chn;
1672 		return 0;
1673 
1674 	case ETHTOOL_GRXFH:
1675 		return netvsc_get_rss_hash_opts(ndc, info);
1676 	}
1677 	return -EOPNOTSUPP;
1678 }
1679 
1680 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1681 				    struct ethtool_rxnfc *info)
1682 {
1683 	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1684 			   RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1685 		switch (info->flow_type) {
1686 		case TCP_V4_FLOW:
1687 			ndc->l4_hash |= HV_TCP4_L4HASH;
1688 			break;
1689 
1690 		case TCP_V6_FLOW:
1691 			ndc->l4_hash |= HV_TCP6_L4HASH;
1692 			break;
1693 
1694 		case UDP_V4_FLOW:
1695 			ndc->l4_hash |= HV_UDP4_L4HASH;
1696 			break;
1697 
1698 		case UDP_V6_FLOW:
1699 			ndc->l4_hash |= HV_UDP6_L4HASH;
1700 			break;
1701 
1702 		default:
1703 			return -EOPNOTSUPP;
1704 		}
1705 
1706 		return 0;
1707 	}
1708 
1709 	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1710 		switch (info->flow_type) {
1711 		case TCP_V4_FLOW:
1712 			ndc->l4_hash &= ~HV_TCP4_L4HASH;
1713 			break;
1714 
1715 		case TCP_V6_FLOW:
1716 			ndc->l4_hash &= ~HV_TCP6_L4HASH;
1717 			break;
1718 
1719 		case UDP_V4_FLOW:
1720 			ndc->l4_hash &= ~HV_UDP4_L4HASH;
1721 			break;
1722 
1723 		case UDP_V6_FLOW:
1724 			ndc->l4_hash &= ~HV_UDP6_L4HASH;
1725 			break;
1726 
1727 		default:
1728 			return -EOPNOTSUPP;
1729 		}
1730 
1731 		return 0;
1732 	}
1733 
1734 	return -EOPNOTSUPP;
1735 }
1736 
1737 static int
1738 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1739 {
1740 	struct net_device_context *ndc = netdev_priv(ndev);
1741 
1742 	if (info->cmd == ETHTOOL_SRXFH)
1743 		return netvsc_set_rss_hash_opts(ndc, info);
1744 
1745 	return -EOPNOTSUPP;
1746 }
1747 
1748 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1749 {
1750 	return NETVSC_HASH_KEYLEN;
1751 }
1752 
1753 static u32 netvsc_rss_indir_size(struct net_device *dev)
1754 {
1755 	struct net_device_context *ndc = netdev_priv(dev);
1756 
1757 	return ndc->rx_table_sz;
1758 }
1759 
1760 static int netvsc_get_rxfh(struct net_device *dev,
1761 			   struct ethtool_rxfh_param *rxfh)
1762 {
1763 	struct net_device_context *ndc = netdev_priv(dev);
1764 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1765 	struct rndis_device *rndis_dev;
1766 	int i;
1767 
1768 	if (!ndev)
1769 		return -ENODEV;
1770 
1771 	rxfh->hfunc = ETH_RSS_HASH_TOP;	/* Toeplitz */
1772 
1773 	rndis_dev = ndev->extension;
1774 	if (rxfh->indir) {
1775 		for (i = 0; i < ndc->rx_table_sz; i++)
1776 			rxfh->indir[i] = ndc->rx_table[i];
1777 	}
1778 
1779 	if (rxfh->key)
1780 		memcpy(rxfh->key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1781 
1782 	return 0;
1783 }
1784 
1785 static int netvsc_set_rxfh(struct net_device *dev,
1786 			   struct ethtool_rxfh_param *rxfh,
1787 			   struct netlink_ext_ack *extack)
1788 {
1789 	struct net_device_context *ndc = netdev_priv(dev);
1790 	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1791 	struct rndis_device *rndis_dev;
1792 	u8 *key = rxfh->key;
1793 	int i;
1794 
1795 	if (!ndev)
1796 		return -ENODEV;
1797 
1798 	if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
1799 	    rxfh->hfunc != ETH_RSS_HASH_TOP)
1800 		return -EOPNOTSUPP;
1801 
1802 	rndis_dev = ndev->extension;
1803 	if (rxfh->indir) {
1804 		for (i = 0; i < ndc->rx_table_sz; i++)
1805 			if (rxfh->indir[i] >= ndev->num_chn)
1806 				return -EINVAL;
1807 
1808 		for (i = 0; i < ndc->rx_table_sz; i++)
1809 			ndc->rx_table[i] = rxfh->indir[i];
1810 	}
1811 
1812 	if (!key) {
1813 		if (!rxfh->indir)
1814 			return 0;
1815 
1816 		key = rndis_dev->rss_key;
1817 	}
1818 
1819 	return rndis_filter_set_rss_param(rndis_dev, key);
1820 }
1821 
1822 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1823  * It does have pre-allocated receive area which is divided into sections.
1824  */
1825 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1826 				   struct ethtool_ringparam *ring)
1827 {
1828 	u32 max_buf_size;
1829 
1830 	ring->rx_pending = nvdev->recv_section_cnt;
1831 	ring->tx_pending = nvdev->send_section_cnt;
1832 
1833 	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1834 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1835 	else
1836 		max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1837 
1838 	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1839 	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1840 		/ nvdev->send_section_size;
1841 }
1842 
1843 static void netvsc_get_ringparam(struct net_device *ndev,
1844 				 struct ethtool_ringparam *ring,
1845 				 struct kernel_ethtool_ringparam *kernel_ring,
1846 				 struct netlink_ext_ack *extack)
1847 {
1848 	struct net_device_context *ndevctx = netdev_priv(ndev);
1849 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1850 
1851 	if (!nvdev)
1852 		return;
1853 
1854 	__netvsc_get_ringparam(nvdev, ring);
1855 }
1856 
1857 static int netvsc_set_ringparam(struct net_device *ndev,
1858 				struct ethtool_ringparam *ring,
1859 				struct kernel_ethtool_ringparam *kernel_ring,
1860 				struct netlink_ext_ack *extack)
1861 {
1862 	struct net_device_context *ndevctx = netdev_priv(ndev);
1863 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1864 	struct netvsc_device_info *device_info;
1865 	struct ethtool_ringparam orig;
1866 	u32 new_tx, new_rx;
1867 	int ret = 0;
1868 
1869 	if (!nvdev || nvdev->destroy)
1870 		return -ENODEV;
1871 
1872 	memset(&orig, 0, sizeof(orig));
1873 	__netvsc_get_ringparam(nvdev, &orig);
1874 
1875 	new_tx = clamp_t(u32, ring->tx_pending,
1876 			 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1877 	new_rx = clamp_t(u32, ring->rx_pending,
1878 			 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1879 
1880 	if (new_tx == orig.tx_pending &&
1881 	    new_rx == orig.rx_pending)
1882 		return 0;	 /* no change */
1883 
1884 	device_info = netvsc_devinfo_get(nvdev);
1885 
1886 	if (!device_info)
1887 		return -ENOMEM;
1888 
1889 	device_info->send_sections = new_tx;
1890 	device_info->recv_sections = new_rx;
1891 
1892 	ret = netvsc_detach(ndev, nvdev);
1893 	if (ret)
1894 		goto out;
1895 
1896 	ret = netvsc_attach(ndev, device_info);
1897 	if (ret) {
1898 		device_info->send_sections = orig.tx_pending;
1899 		device_info->recv_sections = orig.rx_pending;
1900 
1901 		if (netvsc_attach(ndev, device_info))
1902 			netdev_err(ndev, "restoring ringparam failed");
1903 	}
1904 
1905 out:
1906 	netvsc_devinfo_put(device_info);
1907 	return ret;
1908 }
1909 
1910 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1911 					     netdev_features_t features)
1912 {
1913 	struct net_device_context *ndevctx = netdev_priv(ndev);
1914 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1915 
1916 	if (!nvdev || nvdev->destroy)
1917 		return features;
1918 
1919 	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1920 		features ^= NETIF_F_LRO;
1921 		netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1922 	}
1923 
1924 	return features;
1925 }
1926 
1927 static int netvsc_set_features(struct net_device *ndev,
1928 			       netdev_features_t features)
1929 {
1930 	netdev_features_t change = features ^ ndev->features;
1931 	struct net_device_context *ndevctx = netdev_priv(ndev);
1932 	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1933 	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1934 	struct ndis_offload_params offloads;
1935 	int ret = 0;
1936 
1937 	if (!nvdev || nvdev->destroy)
1938 		return -ENODEV;
1939 
1940 	if (!(change & NETIF_F_LRO))
1941 		goto syncvf;
1942 
1943 	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1944 
1945 	if (features & NETIF_F_LRO) {
1946 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1947 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1948 	} else {
1949 		offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1950 		offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1951 	}
1952 
1953 	ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1954 
1955 	if (ret) {
1956 		features ^= NETIF_F_LRO;
1957 		ndev->features = features;
1958 	}
1959 
1960 syncvf:
1961 	if (!vf_netdev)
1962 		return ret;
1963 
1964 	vf_netdev->wanted_features = features;
1965 	netdev_update_features(vf_netdev);
1966 
1967 	return ret;
1968 }
1969 
1970 static int netvsc_get_regs_len(struct net_device *netdev)
1971 {
1972 	return VRSS_SEND_TAB_SIZE * sizeof(u32);
1973 }
1974 
1975 static void netvsc_get_regs(struct net_device *netdev,
1976 			    struct ethtool_regs *regs, void *p)
1977 {
1978 	struct net_device_context *ndc = netdev_priv(netdev);
1979 	u32 *regs_buff = p;
1980 
1981 	/* increase the version, if buffer format is changed. */
1982 	regs->version = 1;
1983 
1984 	memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1985 }
1986 
1987 static u32 netvsc_get_msglevel(struct net_device *ndev)
1988 {
1989 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1990 
1991 	return ndev_ctx->msg_enable;
1992 }
1993 
1994 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1995 {
1996 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1997 
1998 	ndev_ctx->msg_enable = val;
1999 }
2000 
2001 static const struct ethtool_ops ethtool_ops = {
2002 	.get_drvinfo	= netvsc_get_drvinfo,
2003 	.get_regs_len	= netvsc_get_regs_len,
2004 	.get_regs	= netvsc_get_regs,
2005 	.get_msglevel	= netvsc_get_msglevel,
2006 	.set_msglevel	= netvsc_set_msglevel,
2007 	.get_link	= ethtool_op_get_link,
2008 	.get_ethtool_stats = netvsc_get_ethtool_stats,
2009 	.get_sset_count = netvsc_get_sset_count,
2010 	.get_strings	= netvsc_get_strings,
2011 	.get_channels   = netvsc_get_channels,
2012 	.set_channels   = netvsc_set_channels,
2013 	.get_ts_info	= ethtool_op_get_ts_info,
2014 	.get_rxnfc	= netvsc_get_rxnfc,
2015 	.set_rxnfc	= netvsc_set_rxnfc,
2016 	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
2017 	.get_rxfh_indir_size = netvsc_rss_indir_size,
2018 	.get_rxfh	= netvsc_get_rxfh,
2019 	.set_rxfh	= netvsc_set_rxfh,
2020 	.get_link_ksettings = netvsc_get_link_ksettings,
2021 	.set_link_ksettings = netvsc_set_link_ksettings,
2022 	.get_ringparam	= netvsc_get_ringparam,
2023 	.set_ringparam	= netvsc_set_ringparam,
2024 };
2025 
2026 static const struct net_device_ops device_ops = {
2027 	.ndo_open =			netvsc_open,
2028 	.ndo_stop =			netvsc_close,
2029 	.ndo_start_xmit =		netvsc_start_xmit,
2030 	.ndo_change_rx_flags =		netvsc_change_rx_flags,
2031 	.ndo_set_rx_mode =		netvsc_set_rx_mode,
2032 	.ndo_fix_features =		netvsc_fix_features,
2033 	.ndo_set_features =		netvsc_set_features,
2034 	.ndo_change_mtu =		netvsc_change_mtu,
2035 	.ndo_validate_addr =		eth_validate_addr,
2036 	.ndo_set_mac_address =		netvsc_set_mac_addr,
2037 	.ndo_select_queue =		netvsc_select_queue,
2038 	.ndo_get_stats64 =		netvsc_get_stats64,
2039 	.ndo_bpf =			netvsc_bpf,
2040 	.ndo_xdp_xmit =			netvsc_ndoxdp_xmit,
2041 };
2042 
2043 /*
2044  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2045  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2046  * present send GARP packet to network peers with netif_notify_peers().
2047  */
2048 static void netvsc_link_change(struct work_struct *w)
2049 {
2050 	struct net_device_context *ndev_ctx =
2051 		container_of(w, struct net_device_context, dwork.work);
2052 	struct hv_device *device_obj = ndev_ctx->device_ctx;
2053 	struct net_device *net = hv_get_drvdata(device_obj);
2054 	unsigned long flags, next_reconfig, delay;
2055 	struct netvsc_reconfig *event = NULL;
2056 	struct netvsc_device *net_device;
2057 	struct rndis_device *rdev;
2058 	bool reschedule = false;
2059 
2060 	/* if changes are happening, comeback later */
2061 	if (!rtnl_trylock()) {
2062 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2063 		return;
2064 	}
2065 
2066 	net_device = rtnl_dereference(ndev_ctx->nvdev);
2067 	if (!net_device)
2068 		goto out_unlock;
2069 
2070 	rdev = net_device->extension;
2071 
2072 	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2073 	if (time_is_after_jiffies(next_reconfig)) {
2074 		/* link_watch only sends one notification with current state
2075 		 * per second, avoid doing reconfig more frequently. Handle
2076 		 * wrap around.
2077 		 */
2078 		delay = next_reconfig - jiffies;
2079 		delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2080 		schedule_delayed_work(&ndev_ctx->dwork, delay);
2081 		goto out_unlock;
2082 	}
2083 	ndev_ctx->last_reconfig = jiffies;
2084 
2085 	spin_lock_irqsave(&ndev_ctx->lock, flags);
2086 	if (!list_empty(&ndev_ctx->reconfig_events)) {
2087 		event = list_first_entry(&ndev_ctx->reconfig_events,
2088 					 struct netvsc_reconfig, list);
2089 		list_del(&event->list);
2090 		reschedule = !list_empty(&ndev_ctx->reconfig_events);
2091 	}
2092 	spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2093 
2094 	if (!event)
2095 		goto out_unlock;
2096 
2097 	switch (event->event) {
2098 		/* Only the following events are possible due to the check in
2099 		 * netvsc_linkstatus_callback()
2100 		 */
2101 	case RNDIS_STATUS_MEDIA_CONNECT:
2102 		if (rdev->link_state) {
2103 			rdev->link_state = false;
2104 			netif_carrier_on(net);
2105 			netvsc_tx_enable(net_device, net);
2106 		} else {
2107 			__netdev_notify_peers(net);
2108 		}
2109 		kfree(event);
2110 		break;
2111 	case RNDIS_STATUS_MEDIA_DISCONNECT:
2112 		if (!rdev->link_state) {
2113 			rdev->link_state = true;
2114 			netif_carrier_off(net);
2115 			netvsc_tx_disable(net_device, net);
2116 		}
2117 		kfree(event);
2118 		break;
2119 	case RNDIS_STATUS_NETWORK_CHANGE:
2120 		/* Only makes sense if carrier is present */
2121 		if (!rdev->link_state) {
2122 			rdev->link_state = true;
2123 			netif_carrier_off(net);
2124 			netvsc_tx_disable(net_device, net);
2125 			event->event = RNDIS_STATUS_MEDIA_CONNECT;
2126 			spin_lock_irqsave(&ndev_ctx->lock, flags);
2127 			list_add(&event->list, &ndev_ctx->reconfig_events);
2128 			spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2129 			reschedule = true;
2130 		}
2131 		break;
2132 	}
2133 
2134 	rtnl_unlock();
2135 
2136 	/* link_watch only sends one notification with current state per
2137 	 * second, handle next reconfig event in 2 seconds.
2138 	 */
2139 	if (reschedule)
2140 		schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2141 
2142 	return;
2143 
2144 out_unlock:
2145 	rtnl_unlock();
2146 }
2147 
2148 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2149 {
2150 	struct net_device_context *net_device_ctx;
2151 	struct net_device *dev;
2152 
2153 	dev = netdev_master_upper_dev_get(vf_netdev);
2154 	if (!dev || dev->netdev_ops != &device_ops)
2155 		return NULL;	/* not a netvsc device */
2156 
2157 	net_device_ctx = netdev_priv(dev);
2158 	if (!rtnl_dereference(net_device_ctx->nvdev))
2159 		return NULL;	/* device is removed */
2160 
2161 	return dev;
2162 }
2163 
2164 /* Called when VF is injecting data into network stack.
2165  * Change the associated network device from VF to netvsc.
2166  * note: already called with rcu_read_lock
2167  */
2168 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2169 {
2170 	struct sk_buff *skb = *pskb;
2171 	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2172 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2173 	struct netvsc_vf_pcpu_stats *pcpu_stats
2174 		 = this_cpu_ptr(ndev_ctx->vf_stats);
2175 
2176 	skb = skb_share_check(skb, GFP_ATOMIC);
2177 	if (unlikely(!skb))
2178 		return RX_HANDLER_CONSUMED;
2179 
2180 	*pskb = skb;
2181 
2182 	skb->dev = ndev;
2183 
2184 	u64_stats_update_begin(&pcpu_stats->syncp);
2185 	pcpu_stats->rx_packets++;
2186 	pcpu_stats->rx_bytes += skb->len;
2187 	u64_stats_update_end(&pcpu_stats->syncp);
2188 
2189 	return RX_HANDLER_ANOTHER;
2190 }
2191 
2192 static int netvsc_vf_join(struct net_device *vf_netdev,
2193 			  struct net_device *ndev, int context)
2194 {
2195 	struct net_device_context *ndev_ctx = netdev_priv(ndev);
2196 	int ret;
2197 
2198 	ret = netdev_rx_handler_register(vf_netdev,
2199 					 netvsc_vf_handle_frame, ndev);
2200 	if (ret != 0) {
2201 		netdev_err(vf_netdev,
2202 			   "can not register netvsc VF receive handler (err = %d)\n",
2203 			   ret);
2204 		goto rx_handler_failed;
2205 	}
2206 
2207 	ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2208 					   NULL, NULL, NULL);
2209 	if (ret != 0) {
2210 		netdev_err(vf_netdev,
2211 			   "can not set master device %s (err = %d)\n",
2212 			   ndev->name, ret);
2213 		goto upper_link_failed;
2214 	}
2215 
2216 	/* If this registration is called from probe context vf_takeover
2217 	 * is taken care of later in probe itself.
2218 	 */
2219 	if (context == VF_REG_IN_NOTIFIER)
2220 		schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2221 
2222 	call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2223 
2224 	netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2225 	return 0;
2226 
2227 upper_link_failed:
2228 	netdev_rx_handler_unregister(vf_netdev);
2229 rx_handler_failed:
2230 	return ret;
2231 }
2232 
2233 static void __netvsc_vf_setup(struct net_device *ndev,
2234 			      struct net_device *vf_netdev)
2235 {
2236 	int ret;
2237 
2238 	/* Align MTU of VF with master */
2239 	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2240 	if (ret)
2241 		netdev_warn(vf_netdev,
2242 			    "unable to change mtu to %u\n", ndev->mtu);
2243 
2244 	/* set multicast etc flags on VF */
2245 	dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2246 
2247 	/* sync address list from ndev to VF */
2248 	netif_addr_lock_bh(ndev);
2249 	dev_uc_sync(vf_netdev, ndev);
2250 	dev_mc_sync(vf_netdev, ndev);
2251 	netif_addr_unlock_bh(ndev);
2252 
2253 	if (netif_running(ndev)) {
2254 		ret = dev_open(vf_netdev, NULL);
2255 		if (ret)
2256 			netdev_warn(vf_netdev,
2257 				    "unable to open: %d\n", ret);
2258 	}
2259 }
2260 
2261 /* Setup VF as slave of the synthetic device.
2262  * Runs in workqueue to avoid recursion in netlink callbacks.
2263  */
2264 static void netvsc_vf_setup(struct work_struct *w)
2265 {
2266 	struct net_device_context *ndev_ctx
2267 		= container_of(w, struct net_device_context, vf_takeover.work);
2268 	struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2269 	struct net_device *vf_netdev;
2270 
2271 	if (!rtnl_trylock()) {
2272 		schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2273 		return;
2274 	}
2275 
2276 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2277 	if (vf_netdev)
2278 		__netvsc_vf_setup(ndev, vf_netdev);
2279 
2280 	rtnl_unlock();
2281 }
2282 
2283 /* Find netvsc by VF serial number.
2284  * The PCI hyperv controller records the serial number as the slot kobj name.
2285  */
2286 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2287 {
2288 	struct device *parent = vf_netdev->dev.parent;
2289 	struct net_device_context *ndev_ctx;
2290 	struct net_device *ndev;
2291 	struct pci_dev *pdev;
2292 	u32 serial;
2293 
2294 	if (!parent || !dev_is_pci(parent))
2295 		return NULL; /* not a PCI device */
2296 
2297 	pdev = to_pci_dev(parent);
2298 	if (!pdev->slot) {
2299 		netdev_notice(vf_netdev, "no PCI slot information\n");
2300 		return NULL;
2301 	}
2302 
2303 	if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2304 		netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2305 			      pci_slot_name(pdev->slot));
2306 		return NULL;
2307 	}
2308 
2309 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2310 		if (!ndev_ctx->vf_alloc)
2311 			continue;
2312 
2313 		if (ndev_ctx->vf_serial != serial)
2314 			continue;
2315 
2316 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2317 		if (ndev->addr_len != vf_netdev->addr_len ||
2318 		    memcmp(ndev->perm_addr, vf_netdev->perm_addr,
2319 			   ndev->addr_len) != 0)
2320 			continue;
2321 
2322 		return ndev;
2323 
2324 	}
2325 
2326 	/* Fallback path to check synthetic vf with help of mac addr.
2327 	 * Because this function can be called before vf_netdev is
2328 	 * initialized (NETDEV_POST_INIT) when its perm_addr has not been copied
2329 	 * from dev_addr, also try to match to its dev_addr.
2330 	 * Note: On Hyper-V and Azure, it's not possible to set a MAC address
2331 	 * on a VF that matches to the MAC of a unrelated NETVSC device.
2332 	 */
2333 	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2334 		ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2335 		if (ether_addr_equal(vf_netdev->perm_addr, ndev->perm_addr) ||
2336 		    ether_addr_equal(vf_netdev->dev_addr, ndev->perm_addr))
2337 			return ndev;
2338 	}
2339 
2340 	netdev_notice(vf_netdev,
2341 		      "no netdev found for vf serial:%u\n", serial);
2342 	return NULL;
2343 }
2344 
2345 static int netvsc_prepare_bonding(struct net_device *vf_netdev)
2346 {
2347 	struct net_device *ndev;
2348 
2349 	ndev = get_netvsc_byslot(vf_netdev);
2350 	if (!ndev)
2351 		return NOTIFY_DONE;
2352 
2353 	/* set slave flag before open to prevent IPv6 addrconf */
2354 	vf_netdev->flags |= IFF_SLAVE;
2355 	return NOTIFY_DONE;
2356 }
2357 
2358 static int netvsc_register_vf(struct net_device *vf_netdev, int context)
2359 {
2360 	struct net_device_context *net_device_ctx;
2361 	struct netvsc_device *netvsc_dev;
2362 	struct bpf_prog *prog;
2363 	struct net_device *ndev;
2364 	int ret;
2365 
2366 	if (vf_netdev->addr_len != ETH_ALEN)
2367 		return NOTIFY_DONE;
2368 
2369 	ndev = get_netvsc_byslot(vf_netdev);
2370 	if (!ndev)
2371 		return NOTIFY_DONE;
2372 
2373 	net_device_ctx = netdev_priv(ndev);
2374 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2375 	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2376 		return NOTIFY_DONE;
2377 
2378 	/* if synthetic interface is a different namespace,
2379 	 * then move the VF to that namespace; join will be
2380 	 * done again in that context.
2381 	 */
2382 	if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2383 		ret = dev_change_net_namespace(vf_netdev,
2384 					       dev_net(ndev), "eth%d");
2385 		if (ret)
2386 			netdev_err(vf_netdev,
2387 				   "could not move to same namespace as %s: %d\n",
2388 				   ndev->name, ret);
2389 		else
2390 			netdev_info(vf_netdev,
2391 				    "VF moved to namespace with: %s\n",
2392 				    ndev->name);
2393 		return NOTIFY_DONE;
2394 	}
2395 
2396 	netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2397 
2398 	if (netvsc_vf_join(vf_netdev, ndev, context) != 0)
2399 		return NOTIFY_DONE;
2400 
2401 	dev_hold(vf_netdev);
2402 	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2403 
2404 	if (ndev->needed_headroom < vf_netdev->needed_headroom)
2405 		ndev->needed_headroom = vf_netdev->needed_headroom;
2406 
2407 	vf_netdev->wanted_features = ndev->features;
2408 	netdev_update_features(vf_netdev);
2409 
2410 	prog = netvsc_xdp_get(netvsc_dev);
2411 	netvsc_vf_setxdp(vf_netdev, prog);
2412 
2413 	return NOTIFY_OK;
2414 }
2415 
2416 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2417  *
2418  * Typically a UP or DOWN event is followed by a CHANGE event, so
2419  * net_device_ctx->data_path_is_vf is used to cache the current data path
2420  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2421  * message.
2422  *
2423  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2424  * interface, there is only the CHANGE event and no UP or DOWN event.
2425  */
2426 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2427 {
2428 	struct net_device_context *net_device_ctx;
2429 	struct netvsc_device *netvsc_dev;
2430 	struct net_device *ndev;
2431 	bool vf_is_up = false;
2432 	int ret;
2433 
2434 	if (event != NETDEV_GOING_DOWN)
2435 		vf_is_up = netif_running(vf_netdev);
2436 
2437 	ndev = get_netvsc_byref(vf_netdev);
2438 	if (!ndev)
2439 		return NOTIFY_DONE;
2440 
2441 	net_device_ctx = netdev_priv(ndev);
2442 	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2443 	if (!netvsc_dev)
2444 		return NOTIFY_DONE;
2445 
2446 	if (net_device_ctx->data_path_is_vf == vf_is_up)
2447 		return NOTIFY_OK;
2448 
2449 	if (vf_is_up && !net_device_ctx->vf_alloc) {
2450 		netdev_info(ndev, "Waiting for the VF association from host\n");
2451 		wait_for_completion(&net_device_ctx->vf_add);
2452 	}
2453 
2454 	ret = netvsc_switch_datapath(ndev, vf_is_up);
2455 
2456 	if (ret) {
2457 		netdev_err(ndev,
2458 			   "Data path failed to switch %s VF: %s, err: %d\n",
2459 			   vf_is_up ? "to" : "from", vf_netdev->name, ret);
2460 		return NOTIFY_DONE;
2461 	} else {
2462 		netdev_info(ndev, "Data path switched %s VF: %s\n",
2463 			    vf_is_up ? "to" : "from", vf_netdev->name);
2464 	}
2465 
2466 	return NOTIFY_OK;
2467 }
2468 
2469 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2470 {
2471 	struct net_device *ndev;
2472 	struct net_device_context *net_device_ctx;
2473 
2474 	ndev = get_netvsc_byref(vf_netdev);
2475 	if (!ndev)
2476 		return NOTIFY_DONE;
2477 
2478 	net_device_ctx = netdev_priv(ndev);
2479 	cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2480 
2481 	netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2482 
2483 	netvsc_vf_setxdp(vf_netdev, NULL);
2484 
2485 	reinit_completion(&net_device_ctx->vf_add);
2486 	netdev_rx_handler_unregister(vf_netdev);
2487 	netdev_upper_dev_unlink(vf_netdev, ndev);
2488 	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2489 	dev_put(vf_netdev);
2490 
2491 	ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2492 
2493 	return NOTIFY_OK;
2494 }
2495 
2496 static int check_dev_is_matching_vf(struct net_device *event_ndev)
2497 {
2498 	/* Skip NetVSC interfaces */
2499 	if (event_ndev->netdev_ops == &device_ops)
2500 		return -ENODEV;
2501 
2502 	/* Avoid non-Ethernet type devices */
2503 	if (event_ndev->type != ARPHRD_ETHER)
2504 		return -ENODEV;
2505 
2506 	/* Avoid Vlan dev with same MAC registering as VF */
2507 	if (is_vlan_dev(event_ndev))
2508 		return -ENODEV;
2509 
2510 	/* Avoid Bonding master dev with same MAC registering as VF */
2511 	if (netif_is_bond_master(event_ndev))
2512 		return -ENODEV;
2513 
2514 	return 0;
2515 }
2516 
2517 static int netvsc_probe(struct hv_device *dev,
2518 			const struct hv_vmbus_device_id *dev_id)
2519 {
2520 	struct net_device *net = NULL, *vf_netdev;
2521 	struct net_device_context *net_device_ctx;
2522 	struct netvsc_device_info *device_info = NULL;
2523 	struct netvsc_device *nvdev;
2524 	int ret = -ENOMEM;
2525 
2526 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2527 				VRSS_CHANNEL_MAX);
2528 	if (!net)
2529 		goto no_net;
2530 
2531 	netif_carrier_off(net);
2532 
2533 	netvsc_init_settings(net);
2534 
2535 	net_device_ctx = netdev_priv(net);
2536 	net_device_ctx->device_ctx = dev;
2537 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2538 	if (netif_msg_probe(net_device_ctx))
2539 		netdev_dbg(net, "netvsc msg_enable: %d\n",
2540 			   net_device_ctx->msg_enable);
2541 
2542 	hv_set_drvdata(dev, net);
2543 
2544 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2545 
2546 	init_completion(&net_device_ctx->vf_add);
2547 	spin_lock_init(&net_device_ctx->lock);
2548 	INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2549 	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2550 
2551 	net_device_ctx->vf_stats
2552 		= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2553 	if (!net_device_ctx->vf_stats)
2554 		goto no_stats;
2555 
2556 	net->netdev_ops = &device_ops;
2557 	net->ethtool_ops = &ethtool_ops;
2558 	SET_NETDEV_DEV(net, &dev->device);
2559 	dma_set_min_align_mask(&dev->device, HV_HYP_PAGE_SIZE - 1);
2560 
2561 	/* We always need headroom for rndis header */
2562 	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2563 
2564 	/* Initialize the number of queues to be 1, we may change it if more
2565 	 * channels are offered later.
2566 	 */
2567 	netif_set_real_num_tx_queues(net, 1);
2568 	netif_set_real_num_rx_queues(net, 1);
2569 
2570 	/* Notify the netvsc driver of the new device */
2571 	device_info = netvsc_devinfo_get(NULL);
2572 
2573 	if (!device_info) {
2574 		ret = -ENOMEM;
2575 		goto devinfo_failed;
2576 	}
2577 
2578 	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2579 	 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2580 	 * all subchannels to show up, but that may not happen because
2581 	 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2582 	 * -> ... -> device_add() -> ... -> __device_attach() can't get
2583 	 * the device lock, so all the subchannels can't be processed --
2584 	 * finally netvsc_subchan_work() hangs forever.
2585 	 *
2586 	 * The rtnl lock also needs to be held before rndis_filter_device_add()
2587 	 * which advertises nvsp_2_vsc_capability / sriov bit, and triggers
2588 	 * VF NIC offering and registering. If VF NIC finished register_netdev()
2589 	 * earlier it may cause name based config failure.
2590 	 */
2591 	rtnl_lock();
2592 
2593 	nvdev = rndis_filter_device_add(dev, device_info);
2594 	if (IS_ERR(nvdev)) {
2595 		ret = PTR_ERR(nvdev);
2596 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2597 		goto rndis_failed;
2598 	}
2599 
2600 	eth_hw_addr_set(net, device_info->mac_adr);
2601 
2602 	if (nvdev->num_chn > 1)
2603 		schedule_work(&nvdev->subchan_work);
2604 
2605 	/* hw_features computed in rndis_netdev_set_hwcaps() */
2606 	net->features = net->hw_features |
2607 		NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2608 		NETIF_F_HW_VLAN_CTAG_RX;
2609 	net->vlan_features = net->features;
2610 
2611 	netdev_lockdep_set_classes(net);
2612 
2613 	net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2614 			    NETDEV_XDP_ACT_NDO_XMIT;
2615 
2616 	/* MTU range: 68 - 1500 or 65521 */
2617 	net->min_mtu = NETVSC_MTU_MIN;
2618 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2619 		net->max_mtu = NETVSC_MTU - ETH_HLEN;
2620 	else
2621 		net->max_mtu = ETH_DATA_LEN;
2622 
2623 	nvdev->tx_disable = false;
2624 
2625 	ret = register_netdevice(net);
2626 	if (ret != 0) {
2627 		pr_err("Unable to register netdev.\n");
2628 		goto register_failed;
2629 	}
2630 
2631 	list_add(&net_device_ctx->list, &netvsc_dev_list);
2632 
2633 	/* When the hv_netvsc driver is unloaded and reloaded, the
2634 	 * NET_DEVICE_REGISTER for the vf device is replayed before probe
2635 	 * is complete. This is because register_netdevice_notifier() gets
2636 	 * registered before vmbus_driver_register() so that callback func
2637 	 * is set before probe and we don't miss events like NETDEV_POST_INIT
2638 	 * So, in this section we try to register the matching vf device that
2639 	 * is present as a netdevice, knowing that its register call is not
2640 	 * processed in the netvsc_netdev_notifier(as probing is progress and
2641 	 * get_netvsc_byslot fails).
2642 	 */
2643 	for_each_netdev(dev_net(net), vf_netdev) {
2644 		ret = check_dev_is_matching_vf(vf_netdev);
2645 		if (ret != 0)
2646 			continue;
2647 
2648 		if (net != get_netvsc_byslot(vf_netdev))
2649 			continue;
2650 
2651 		netvsc_prepare_bonding(vf_netdev);
2652 		netvsc_register_vf(vf_netdev, VF_REG_IN_PROBE);
2653 		__netvsc_vf_setup(net, vf_netdev);
2654 		break;
2655 	}
2656 	rtnl_unlock();
2657 
2658 	netvsc_devinfo_put(device_info);
2659 	return 0;
2660 
2661 register_failed:
2662 	rndis_filter_device_remove(dev, nvdev);
2663 rndis_failed:
2664 	rtnl_unlock();
2665 	netvsc_devinfo_put(device_info);
2666 devinfo_failed:
2667 	free_percpu(net_device_ctx->vf_stats);
2668 no_stats:
2669 	hv_set_drvdata(dev, NULL);
2670 	free_netdev(net);
2671 no_net:
2672 	return ret;
2673 }
2674 
2675 static void netvsc_remove(struct hv_device *dev)
2676 {
2677 	struct net_device_context *ndev_ctx;
2678 	struct net_device *vf_netdev, *net;
2679 	struct netvsc_device *nvdev;
2680 
2681 	net = hv_get_drvdata(dev);
2682 	if (net == NULL) {
2683 		dev_err(&dev->device, "No net device to remove\n");
2684 		return;
2685 	}
2686 
2687 	ndev_ctx = netdev_priv(net);
2688 
2689 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2690 
2691 	rtnl_lock();
2692 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2693 	if (nvdev) {
2694 		cancel_work_sync(&nvdev->subchan_work);
2695 		netvsc_xdp_set(net, NULL, NULL, nvdev);
2696 	}
2697 
2698 	/*
2699 	 * Call to the vsc driver to let it know that the device is being
2700 	 * removed. Also blocks mtu and channel changes.
2701 	 */
2702 	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2703 	if (vf_netdev)
2704 		netvsc_unregister_vf(vf_netdev);
2705 
2706 	if (nvdev)
2707 		rndis_filter_device_remove(dev, nvdev);
2708 
2709 	unregister_netdevice(net);
2710 	list_del(&ndev_ctx->list);
2711 
2712 	rtnl_unlock();
2713 
2714 	hv_set_drvdata(dev, NULL);
2715 
2716 	free_percpu(ndev_ctx->vf_stats);
2717 	free_netdev(net);
2718 }
2719 
2720 static int netvsc_suspend(struct hv_device *dev)
2721 {
2722 	struct net_device_context *ndev_ctx;
2723 	struct netvsc_device *nvdev;
2724 	struct net_device *net;
2725 	int ret;
2726 
2727 	net = hv_get_drvdata(dev);
2728 
2729 	ndev_ctx = netdev_priv(net);
2730 	cancel_delayed_work_sync(&ndev_ctx->dwork);
2731 
2732 	rtnl_lock();
2733 
2734 	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2735 	if (nvdev == NULL) {
2736 		ret = -ENODEV;
2737 		goto out;
2738 	}
2739 
2740 	/* Save the current config info */
2741 	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2742 	if (!ndev_ctx->saved_netvsc_dev_info) {
2743 		ret = -ENOMEM;
2744 		goto out;
2745 	}
2746 	ret = netvsc_detach(net, nvdev);
2747 out:
2748 	rtnl_unlock();
2749 
2750 	return ret;
2751 }
2752 
2753 static int netvsc_resume(struct hv_device *dev)
2754 {
2755 	struct net_device *net = hv_get_drvdata(dev);
2756 	struct net_device_context *net_device_ctx;
2757 	struct netvsc_device_info *device_info;
2758 	int ret;
2759 
2760 	rtnl_lock();
2761 
2762 	net_device_ctx = netdev_priv(net);
2763 
2764 	/* Reset the data path to the netvsc NIC before re-opening the vmbus
2765 	 * channel. Later netvsc_netdev_event() will switch the data path to
2766 	 * the VF upon the UP or CHANGE event.
2767 	 */
2768 	net_device_ctx->data_path_is_vf = false;
2769 	device_info = net_device_ctx->saved_netvsc_dev_info;
2770 
2771 	ret = netvsc_attach(net, device_info);
2772 
2773 	netvsc_devinfo_put(device_info);
2774 	net_device_ctx->saved_netvsc_dev_info = NULL;
2775 
2776 	rtnl_unlock();
2777 
2778 	return ret;
2779 }
2780 static const struct hv_vmbus_device_id id_table[] = {
2781 	/* Network guid */
2782 	{ HV_NIC_GUID, },
2783 	{ },
2784 };
2785 
2786 MODULE_DEVICE_TABLE(vmbus, id_table);
2787 
2788 /* The one and only one */
2789 static struct  hv_driver netvsc_drv = {
2790 	.name = KBUILD_MODNAME,
2791 	.id_table = id_table,
2792 	.probe = netvsc_probe,
2793 	.remove = netvsc_remove,
2794 	.suspend = netvsc_suspend,
2795 	.resume = netvsc_resume,
2796 	.driver = {
2797 		.probe_type = PROBE_FORCE_SYNCHRONOUS,
2798 	},
2799 };
2800 
2801 /*
2802  * On Hyper-V, every VF interface is matched with a corresponding
2803  * synthetic interface. The synthetic interface is presented first
2804  * to the guest. When the corresponding VF instance is registered,
2805  * we will take care of switching the data path.
2806  */
2807 static int netvsc_netdev_event(struct notifier_block *this,
2808 			       unsigned long event, void *ptr)
2809 {
2810 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2811 	int ret = 0;
2812 
2813 	ret = check_dev_is_matching_vf(event_dev);
2814 	if (ret != 0)
2815 		return NOTIFY_DONE;
2816 
2817 	switch (event) {
2818 	case NETDEV_POST_INIT:
2819 		return netvsc_prepare_bonding(event_dev);
2820 	case NETDEV_REGISTER:
2821 		return netvsc_register_vf(event_dev, VF_REG_IN_NOTIFIER);
2822 	case NETDEV_UNREGISTER:
2823 		return netvsc_unregister_vf(event_dev);
2824 	case NETDEV_UP:
2825 	case NETDEV_DOWN:
2826 	case NETDEV_CHANGE:
2827 	case NETDEV_GOING_DOWN:
2828 		return netvsc_vf_changed(event_dev, event);
2829 	default:
2830 		return NOTIFY_DONE;
2831 	}
2832 }
2833 
2834 static struct notifier_block netvsc_netdev_notifier = {
2835 	.notifier_call = netvsc_netdev_event,
2836 };
2837 
2838 static void __exit netvsc_drv_exit(void)
2839 {
2840 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2841 	vmbus_driver_unregister(&netvsc_drv);
2842 }
2843 
2844 static int __init netvsc_drv_init(void)
2845 {
2846 	int ret;
2847 
2848 	if (ring_size < RING_SIZE_MIN) {
2849 		ring_size = RING_SIZE_MIN;
2850 		pr_info("Increased ring_size to %u (min allowed)\n",
2851 			ring_size);
2852 	}
2853 	netvsc_ring_bytes = VMBUS_RING_SIZE(ring_size * 4096);
2854 
2855 	register_netdevice_notifier(&netvsc_netdev_notifier);
2856 
2857 	ret = vmbus_driver_register(&netvsc_drv);
2858 	if (ret)
2859 		goto err_vmbus_reg;
2860 
2861 	return 0;
2862 
2863 err_vmbus_reg:
2864 	unregister_netdevice_notifier(&netvsc_netdev_notifier);
2865 	return ret;
2866 }
2867 
2868 MODULE_LICENSE("GPL");
2869 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2870 
2871 module_init(netvsc_drv_init);
2872 module_exit(netvsc_drv_exit);
2873