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(dev: 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(dev: vf_netdev, inc);
77	}
78
79	if (change & IFF_ALLMULTI) {
80	inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
81	dev_set_allmulti(dev: 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(dev: 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(to: vf_netdev, from: net);
95	dev_mc_sync(to: vf_netdev, from: 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(dev: ndev);
111	}
112
113	static int netvsc_open(struct net_device *net)
114	{
115	struct net_device_context *ndev_ctx = netdev_priv(dev: 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(dev: net);
122
123	/* Open up the device */
124	ret = rndis_filter_open(nvdev);
125	if (ret != 0) {
126	netdev_err(dev: net, format: "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(dev: net);
133	netvsc_tx_enable(nvscdev: nvdev, ndev: 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(dev: vf_netdev, NULL);
142	if (ret)
143	netdev_warn(dev: net,
144	format: "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(n: &nvdev->chan_table[i].napi);
168
169	aread = hv_get_bytes_to_read(rbi: &chn->inbound);
170	if (aread)
171	break;
172
173	aread = hv_get_bytes_to_read(rbi: &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(dev: ndev);
197	}
198
199	static int netvsc_close(struct net_device *net)
200	{
201	struct net_device_context *net_device_ctx = netdev_priv(dev: 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(nvscdev: nvdev, ndev: 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(dev: net, format: "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(dev: net, format: "Ring buffer not empty after closing rndis\n");
222
223	if (vf_netdev)
224	dev_close(dev: 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(dev: 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, tx_queue: 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(sk: 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, old_idx: 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(dev: 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(dev: 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(hvpfn: virt_to_hvpfn(addr: hdr),
374	offset_in_hvpage(hdr),
375	len,
376	pb: &pb[slots_used]);
377
378	packet->rmsg_size = len;
379	packet->rmsg_pgcnt = slots_used;
380
381	slots_used += fill_pg_buf(hvpfn: virt_to_hvpfn(addr: data),
382	offset_in_hvpage(data),
383	len: skb_headlen(skb),
384	pb: &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	offset: skb_frag_off(frag),
391	len: skb_frag_size(frag),
392	pb: &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(dev: net);
453	unsigned int len = skb->len;
454	int rc;
455
456	skb->dev = vf_netdev;
457	skb_record_rx_queue(skb, rx_queue: 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(syncp: &pcpu_stats->syncp);
465	pcpu_stats->tx_packets++;
466	pcpu_stats->tx_bytes += len;
467	u64_stats_update_end(syncp: &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(dev: 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(dev: vf_netdev) &&
493	netif_carrier_ok(dev: vf_netdev) && !netpoll_tx_running(dev: 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(msg: rndis_msg, NDIS_HASH_PPI_SIZE,
558	pkt_type: 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(ethertype: 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(msg: rndis_msg, NDIS_VLAN_PPI_SIZE,
590	pkt_type: 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(msg: rndis_msg, NDIS_LSO_PPI_SIZE,
603	pkt_type: 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(saddr: ip_hdr(skb)->saddr,
614	daddr: ip_hdr(skb)->daddr, len: 0, IPPROTO_TCP, sum: 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(msg: rndis_msg, NDIS_CSUM_PPI_SIZE,
628	pkt_type: 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(hdr: rndis_msg, len: 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, page_buffer: 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, net: ndev, xdp_tx: 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(dev: 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(dev: net, format: "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(dev: net, format: "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(size: 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(new: &event->list, head: &ndev_ctx->reconfig_events);
755	spin_unlock_irqrestore(lock: &ndev_ctx->lock, flags);
756
757	schedule_delayed_work(dwork: &ndev_ctx->dwork, delay: 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, net: ndev, xdp_tx: 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, ihl: 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(data: xbuf, frag_size);
805
806	if (!skb) {
807	__free_page(virt_to_page(xbuf));
808	return NULL;
809	}
810
811	skb_reserve(skb, len: hdroom);
812	skb_put(skb, len: xlen);
813	skb->dev = napi->dev;
814	} else {
815	skb = napi_alloc_skb(napi, length: 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, data: nvchan->rsc.data[i],
825	len: nvchan->rsc.len[i]);
826	}
827
828	skb->protocol = eth_type_trans(skb, dev: 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: *hash_info, type: 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(dev: 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(ndev: net, nvchan, xdp: &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(syncp: &rx_stats->syncp);
896	rx_stats->xdp_drop++;
897	u64_stats_update_end(syncp: &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: &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, rx_queue: 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(syncp: &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(syncp: &rx_stats->syncp);
929
930	if (act == XDP_TX) {
931	netvsc_xdp_xmit(skb, ndev: net);
932	return NVSP_STAT_SUCCESS;
933	}
934
935	napi_gro_receive(napi: &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(dev: 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(size: 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 = VRSS_CHANNEL_DEFAULT;
991	dev_info->send_sections = NETVSC_DEFAULT_TX;
992	dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
993	dev_info->recv_sections = NETVSC_DEFAULT_RX;
994	dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
995	}
996
997	return dev_info;
998	}
999
1000	/* Free struct netvsc_device_info */
1001	static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1002	{
1003	if (dev_info->bprog) {
1004	ASSERT_RTNL();
1005	bpf_prog_put(prog: dev_info->bprog);
1006	}
1007
1008	kfree(objp: dev_info);
1009	}
1010
1011	static int netvsc_detach(struct net_device *ndev,
1012	struct netvsc_device *nvdev)
1013	{
1014	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1015	struct hv_device *hdev = ndev_ctx->device_ctx;
1016	int ret;
1017
1018	/* Don't try continuing to try and setup sub channels */
1019	if (cancel_work_sync(work: &nvdev->subchan_work))
1020	nvdev->num_chn = 1;
1021
1022	netvsc_xdp_set(dev: ndev, NULL, NULL, nvdev);
1023
1024	/* If device was up (receiving) then shutdown */
1025	if (netif_running(dev: ndev)) {
1026	netvsc_tx_disable(nvscdev: nvdev, ndev);
1027
1028	ret = rndis_filter_close(nvdev);
1029	if (ret) {
1030	netdev_err(dev: ndev,
1031	format: "unable to close device (ret %d).\n", ret);
1032	return ret;
1033	}
1034
1035	ret = netvsc_wait_until_empty(nvdev);
1036	if (ret) {
1037	netdev_err(dev: ndev,
1038	format: "Ring buffer not empty after closing rndis\n");
1039	return ret;
1040	}
1041	}
1042
1043	netif_device_detach(dev: ndev);
1044
1045	rndis_filter_device_remove(dev: hdev, nvdev);
1046
1047	return 0;
1048	}
1049
1050	static int netvsc_attach(struct net_device *ndev,
1051	struct netvsc_device_info *dev_info)
1052	{
1053	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1054	struct hv_device *hdev = ndev_ctx->device_ctx;
1055	struct netvsc_device *nvdev;
1056	struct rndis_device *rdev;
1057	struct bpf_prog *prog;
1058	int ret = 0;
1059
1060	nvdev = rndis_filter_device_add(dev: hdev, info: dev_info);
1061	if (IS_ERR(ptr: nvdev))
1062	return PTR_ERR(ptr: nvdev);
1063
1064	if (nvdev->num_chn > 1) {
1065	ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1066
1067	/* if unavailable, just proceed with one queue */
1068	if (ret) {
1069	nvdev->max_chn = 1;
1070	nvdev->num_chn = 1;
1071	}
1072	}
1073
1074	prog = dev_info->bprog;
1075	if (prog) {
1076	bpf_prog_inc(prog);
1077	ret = netvsc_xdp_set(dev: ndev, prog, NULL, nvdev);
1078	if (ret) {
1079	bpf_prog_put(prog);
1080	goto err1;
1081	}
1082	}
1083
1084	/* In any case device is now ready */
1085	nvdev->tx_disable = false;
1086	netif_device_attach(dev: ndev);
1087
1088	/* Note: enable and attach happen when sub-channels setup */
1089	netif_carrier_off(dev: ndev);
1090
1091	if (netif_running(dev: ndev)) {
1092	ret = rndis_filter_open(nvdev);
1093	if (ret)
1094	goto err2;
1095
1096	rdev = nvdev->extension;
1097	if (!rdev->link_state)
1098	netif_carrier_on(dev: ndev);
1099	}
1100
1101	return 0;
1102
1103	err2:
1104	netif_device_detach(dev: ndev);
1105
1106	err1:
1107	rndis_filter_device_remove(dev: hdev, nvdev);
1108
1109	return ret;
1110	}
1111
1112	static int netvsc_set_channels(struct net_device *net,
1113	struct ethtool_channels *channels)
1114	{
1115	struct net_device_context *net_device_ctx = netdev_priv(dev: net);
1116	struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1117	unsigned int orig, count = channels->combined_count;
1118	struct netvsc_device_info *device_info;
1119	int ret;
1120
1121	/* We do not support separate count for rx, tx, or other */
1122	if (count == 0 ||
1123	channels->rx_count || channels->tx_count || channels->other_count)
1124	return -EINVAL;
1125
1126	if (!nvdev || nvdev->destroy)
1127	return -ENODEV;
1128
1129	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1130	return -EINVAL;
1131
1132	if (count > nvdev->max_chn)
1133	return -EINVAL;
1134
1135	orig = nvdev->num_chn;
1136
1137	device_info = netvsc_devinfo_get(nvdev);
1138
1139	if (!device_info)
1140	return -ENOMEM;
1141
1142	device_info->num_chn = count;
1143
1144	ret = netvsc_detach(ndev: net, nvdev);
1145	if (ret)
1146	goto out;
1147
1148	ret = netvsc_attach(ndev: net, dev_info: device_info);
1149	if (ret) {
1150	device_info->num_chn = orig;
1151	if (netvsc_attach(ndev: net, dev_info: device_info))
1152	netdev_err(dev: net, format: "restoring channel setting failed\n");
1153	}
1154
1155	out:
1156	netvsc_devinfo_put(dev_info: device_info);
1157	return ret;
1158	}
1159
1160	static void netvsc_init_settings(struct net_device *dev)
1161	{
1162	struct net_device_context *ndc = netdev_priv(dev);
1163
1164	ndc->l4_hash = HV_DEFAULT_L4HASH;
1165
1166	ndc->speed = SPEED_UNKNOWN;
1167	ndc->duplex = DUPLEX_FULL;
1168
1169	dev->features = NETIF_F_LRO;
1170	}
1171
1172	static int netvsc_get_link_ksettings(struct net_device *dev,
1173	struct ethtool_link_ksettings *cmd)
1174	{
1175	struct net_device_context *ndc = netdev_priv(dev);
1176	struct net_device *vf_netdev;
1177
1178	vf_netdev = rtnl_dereference(ndc->vf_netdev);
1179
1180	if (vf_netdev)
1181	return __ethtool_get_link_ksettings(dev: vf_netdev, link_ksettings: cmd);
1182
1183	cmd->base.speed = ndc->speed;
1184	cmd->base.duplex = ndc->duplex;
1185	cmd->base.port = PORT_OTHER;
1186
1187	return 0;
1188	}
1189
1190	static int netvsc_set_link_ksettings(struct net_device *dev,
1191	const struct ethtool_link_ksettings *cmd)
1192	{
1193	struct net_device_context *ndc = netdev_priv(dev);
1194	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1195
1196	if (vf_netdev) {
1197	if (!vf_netdev->ethtool_ops->set_link_ksettings)
1198	return -EOPNOTSUPP;
1199
1200	return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1201	cmd);
1202	}
1203
1204	return ethtool_virtdev_set_link_ksettings(dev, cmd,
1205	dev_speed: &ndc->speed, dev_duplex: &ndc->duplex);
1206	}
1207
1208	static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1209	{
1210	struct net_device_context *ndevctx = netdev_priv(dev: ndev);
1211	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1212	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1213	int orig_mtu = ndev->mtu;
1214	struct netvsc_device_info *device_info;
1215	int ret = 0;
1216
1217	if (!nvdev || nvdev->destroy)
1218	return -ENODEV;
1219
1220	device_info = netvsc_devinfo_get(nvdev);
1221
1222	if (!device_info)
1223	return -ENOMEM;
1224
1225	/* Change MTU of underlying VF netdev first. */
1226	if (vf_netdev) {
1227	ret = dev_set_mtu(vf_netdev, mtu);
1228	if (ret)
1229	goto out;
1230	}
1231
1232	ret = netvsc_detach(ndev, nvdev);
1233	if (ret)
1234	goto rollback_vf;
1235
1236	ndev->mtu = mtu;
1237
1238	ret = netvsc_attach(ndev, dev_info: device_info);
1239	if (!ret)
1240	goto out;
1241
1242	/* Attempt rollback to original MTU */
1243	ndev->mtu = orig_mtu;
1244
1245	if (netvsc_attach(ndev, dev_info: device_info))
1246	netdev_err(dev: ndev, format: "restoring mtu failed\n");
1247	rollback_vf:
1248	if (vf_netdev)
1249	dev_set_mtu(vf_netdev, orig_mtu);
1250
1251	out:
1252	netvsc_devinfo_put(dev_info: device_info);
1253	return ret;
1254	}
1255
1256	static void netvsc_get_vf_stats(struct net_device *net,
1257	struct netvsc_vf_pcpu_stats *tot)
1258	{
1259	struct net_device_context *ndev_ctx = netdev_priv(dev: net);
1260	int i;
1261
1262	memset(tot, 0, sizeof(*tot));
1263
1264	for_each_possible_cpu(i) {
1265	const struct netvsc_vf_pcpu_stats *stats
1266	= per_cpu_ptr(ndev_ctx->vf_stats, i);
1267	u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1268	unsigned int start;
1269
1270	do {
1271	start = u64_stats_fetch_begin(syncp: &stats->syncp);
1272	rx_packets = stats->rx_packets;
1273	tx_packets = stats->tx_packets;
1274	rx_bytes = stats->rx_bytes;
1275	tx_bytes = stats->tx_bytes;
1276	} while (u64_stats_fetch_retry(syncp: &stats->syncp, start));
1277
1278	tot->rx_packets += rx_packets;
1279	tot->tx_packets += tx_packets;
1280	tot->rx_bytes += rx_bytes;
1281	tot->tx_bytes += tx_bytes;
1282	tot->tx_dropped += stats->tx_dropped;
1283	}
1284	}
1285
1286	static void netvsc_get_pcpu_stats(struct net_device *net,
1287	struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1288	{
1289	struct net_device_context *ndev_ctx = netdev_priv(dev: net);
1290	struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1291	int i;
1292
1293	/* fetch percpu stats of vf */
1294	for_each_possible_cpu(i) {
1295	const struct netvsc_vf_pcpu_stats *stats =
1296	per_cpu_ptr(ndev_ctx->vf_stats, i);
1297	struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1298	unsigned int start;
1299
1300	do {
1301	start = u64_stats_fetch_begin(syncp: &stats->syncp);
1302	this_tot->vf_rx_packets = stats->rx_packets;
1303	this_tot->vf_tx_packets = stats->tx_packets;
1304	this_tot->vf_rx_bytes = stats->rx_bytes;
1305	this_tot->vf_tx_bytes = stats->tx_bytes;
1306	} while (u64_stats_fetch_retry(syncp: &stats->syncp, start));
1307	this_tot->rx_packets = this_tot->vf_rx_packets;
1308	this_tot->tx_packets = this_tot->vf_tx_packets;
1309	this_tot->rx_bytes = this_tot->vf_rx_bytes;
1310	this_tot->tx_bytes = this_tot->vf_tx_bytes;
1311	}
1312
1313	/* fetch percpu stats of netvsc */
1314	for (i = 0; i < nvdev->num_chn; i++) {
1315	const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1316	const struct netvsc_stats_tx *tx_stats;
1317	const struct netvsc_stats_rx *rx_stats;
1318	struct netvsc_ethtool_pcpu_stats *this_tot =
1319	&pcpu_tot[nvchan->channel->target_cpu];
1320	u64 packets, bytes;
1321	unsigned int start;
1322
1323	tx_stats = &nvchan->tx_stats;
1324	do {
1325	start = u64_stats_fetch_begin(syncp: &tx_stats->syncp);
1326	packets = tx_stats->packets;
1327	bytes = tx_stats->bytes;
1328	} while (u64_stats_fetch_retry(syncp: &tx_stats->syncp, start));
1329
1330	this_tot->tx_bytes += bytes;
1331	this_tot->tx_packets += packets;
1332
1333	rx_stats = &nvchan->rx_stats;
1334	do {
1335	start = u64_stats_fetch_begin(syncp: &rx_stats->syncp);
1336	packets = rx_stats->packets;
1337	bytes = rx_stats->bytes;
1338	} while (u64_stats_fetch_retry(syncp: &rx_stats->syncp, start));
1339
1340	this_tot->rx_bytes += bytes;
1341	this_tot->rx_packets += packets;
1342	}
1343	}
1344
1345	static void netvsc_get_stats64(struct net_device *net,
1346	struct rtnl_link_stats64 *t)
1347	{
1348	struct net_device_context *ndev_ctx = netdev_priv(dev: net);
1349	struct netvsc_device *nvdev;
1350	struct netvsc_vf_pcpu_stats vf_tot;
1351	int i;
1352
1353	rcu_read_lock();
1354
1355	nvdev = rcu_dereference(ndev_ctx->nvdev);
1356	if (!nvdev)
1357	goto out;
1358
1359	netdev_stats_to_stats64(stats64: t, netdev_stats: &net->stats);
1360
1361	netvsc_get_vf_stats(net, tot: &vf_tot);
1362	t->rx_packets += vf_tot.rx_packets;
1363	t->tx_packets += vf_tot.tx_packets;
1364	t->rx_bytes += vf_tot.rx_bytes;
1365	t->tx_bytes += vf_tot.tx_bytes;
1366	t->tx_dropped += vf_tot.tx_dropped;
1367
1368	for (i = 0; i < nvdev->num_chn; i++) {
1369	const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1370	const struct netvsc_stats_tx *tx_stats;
1371	const struct netvsc_stats_rx *rx_stats;
1372	u64 packets, bytes, multicast;
1373	unsigned int start;
1374
1375	tx_stats = &nvchan->tx_stats;
1376	do {
1377	start = u64_stats_fetch_begin(syncp: &tx_stats->syncp);
1378	packets = tx_stats->packets;
1379	bytes = tx_stats->bytes;
1380	} while (u64_stats_fetch_retry(syncp: &tx_stats->syncp, start));
1381
1382	t->tx_bytes += bytes;
1383	t->tx_packets += packets;
1384
1385	rx_stats = &nvchan->rx_stats;
1386	do {
1387	start = u64_stats_fetch_begin(syncp: &rx_stats->syncp);
1388	packets = rx_stats->packets;
1389	bytes = rx_stats->bytes;
1390	multicast = rx_stats->multicast + rx_stats->broadcast;
1391	} while (u64_stats_fetch_retry(syncp: &rx_stats->syncp, start));
1392
1393	t->rx_bytes += bytes;
1394	t->rx_packets += packets;
1395	t->multicast += multicast;
1396	}
1397	out:
1398	rcu_read_unlock();
1399	}
1400
1401	static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1402	{
1403	struct net_device_context *ndc = netdev_priv(dev: ndev);
1404	struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1405	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1406	struct sockaddr *addr = p;
1407	int err;
1408
1409	err = eth_prepare_mac_addr_change(dev: ndev, p);
1410	if (err)
1411	return err;
1412
1413	if (!nvdev)
1414	return -ENODEV;
1415
1416	if (vf_netdev) {
1417	err = dev_set_mac_address(dev: vf_netdev, sa: addr, NULL);
1418	if (err)
1419	return err;
1420	}
1421
1422	err = rndis_filter_set_device_mac(ndev: nvdev, mac: addr->sa_data);
1423	if (!err) {
1424	eth_commit_mac_addr_change(dev: ndev, p);
1425	} else if (vf_netdev) {
1426	/* rollback change on VF */
1427	memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1428	dev_set_mac_address(dev: vf_netdev, sa: addr, NULL);
1429	}
1430
1431	return err;
1432	}
1433
1434	static const struct {
1435	char name[ETH_GSTRING_LEN];
1436	u16 offset;
1437	} netvsc_stats[] = {
1438	{ "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1439	{ "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1440	{ "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1441	{ "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1442	{ "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
1443	{ "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1444	{ "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1445	{ "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1446	{ "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1447	{ "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1448	{ "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1449	}, pcpu_stats[] = {
1450	{ "cpu%u_rx_packets",
1451	offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1452	{ "cpu%u_rx_bytes",
1453	offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1454	{ "cpu%u_tx_packets",
1455	offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1456	{ "cpu%u_tx_bytes",
1457	offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1458	{ "cpu%u_vf_rx_packets",
1459	offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1460	{ "cpu%u_vf_rx_bytes",
1461	offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1462	{ "cpu%u_vf_tx_packets",
1463	offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1464	{ "cpu%u_vf_tx_bytes",
1465	offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1466	}, vf_stats[] = {
1467	{ "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1468	{ "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1469	{ "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1470	{ "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1471	{ "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1472	};
1473
1474	#define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1475	#define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
1476
1477	/* statistics per queue (rx/tx packets/bytes) */
1478	#define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1479
1480	/* 8 statistics per queue (rx/tx packets/bytes, XDP actions) */
1481	#define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 8)
1482
1483	static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1484	{
1485	struct net_device_context *ndc = netdev_priv(dev);
1486	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1487
1488	if (!nvdev)
1489	return -ENODEV;
1490
1491	switch (string_set) {
1492	case ETH_SS_STATS:
1493	return NETVSC_GLOBAL_STATS_LEN
1494	+ NETVSC_VF_STATS_LEN
1495	+ NETVSC_QUEUE_STATS_LEN(nvdev)
1496	+ NETVSC_PCPU_STATS_LEN;
1497	default:
1498	return -EINVAL;
1499	}
1500	}
1501
1502	static void netvsc_get_ethtool_stats(struct net_device *dev,
1503	struct ethtool_stats *stats, u64 *data)
1504	{
1505	struct net_device_context *ndc = netdev_priv(dev);
1506	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1507	const void *nds = &ndc->eth_stats;
1508	const struct netvsc_stats_tx *tx_stats;
1509	const struct netvsc_stats_rx *rx_stats;
1510	struct netvsc_vf_pcpu_stats sum;
1511	struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1512	unsigned int start;
1513	u64 packets, bytes;
1514	u64 xdp_drop;
1515	u64 xdp_redirect;
1516	u64 xdp_tx;
1517	u64 xdp_xmit;
1518	int i, j, cpu;
1519
1520	if (!nvdev)
1521	return;
1522
1523	for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1524	data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1525
1526	netvsc_get_vf_stats(net: dev, tot: &sum);
1527	for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1528	data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1529
1530	for (j = 0; j < nvdev->num_chn; j++) {
1531	tx_stats = &nvdev->chan_table[j].tx_stats;
1532
1533	do {
1534	start = u64_stats_fetch_begin(syncp: &tx_stats->syncp);
1535	packets = tx_stats->packets;
1536	bytes = tx_stats->bytes;
1537	xdp_xmit = tx_stats->xdp_xmit;
1538	} while (u64_stats_fetch_retry(syncp: &tx_stats->syncp, start));
1539	data[i++] = packets;
1540	data[i++] = bytes;
1541	data[i++] = xdp_xmit;
1542
1543	rx_stats = &nvdev->chan_table[j].rx_stats;
1544	do {
1545	start = u64_stats_fetch_begin(syncp: &rx_stats->syncp);
1546	packets = rx_stats->packets;
1547	bytes = rx_stats->bytes;
1548	xdp_drop = rx_stats->xdp_drop;
1549	xdp_redirect = rx_stats->xdp_redirect;
1550	xdp_tx = rx_stats->xdp_tx;
1551	} while (u64_stats_fetch_retry(syncp: &rx_stats->syncp, start));
1552	data[i++] = packets;
1553	data[i++] = bytes;
1554	data[i++] = xdp_drop;
1555	data[i++] = xdp_redirect;
1556	data[i++] = xdp_tx;
1557	}
1558
1559	pcpu_sum = kvmalloc_array(num_possible_cpus(),
1560	size: sizeof(struct netvsc_ethtool_pcpu_stats),
1561	GFP_KERNEL);
1562	if (!pcpu_sum)
1563	return;
1564
1565	netvsc_get_pcpu_stats(net: dev, pcpu_tot: pcpu_sum);
1566	for_each_present_cpu(cpu) {
1567	struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1568
1569	for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1570	data[i++] = *(u64 *)((void *)this_sum
1571	+ pcpu_stats[j].offset);
1572	}
1573	kvfree(addr: pcpu_sum);
1574	}
1575
1576	static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1577	{
1578	struct net_device_context *ndc = netdev_priv(dev);
1579	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1580	u8 *p = data;
1581	int i, cpu;
1582
1583	if (!nvdev)
1584	return;
1585
1586	switch (stringset) {
1587	case ETH_SS_STATS:
1588	for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++)
1589	ethtool_puts(data: &p, str: netvsc_stats[i].name);
1590
1591	for (i = 0; i < ARRAY_SIZE(vf_stats); i++)
1592	ethtool_puts(data: &p, str: vf_stats[i].name);
1593
1594	for (i = 0; i < nvdev->num_chn; i++) {
1595	ethtool_sprintf(data: &p, fmt: "tx_queue_%u_packets", i);
1596	ethtool_sprintf(data: &p, fmt: "tx_queue_%u_bytes", i);
1597	ethtool_sprintf(data: &p, fmt: "tx_queue_%u_xdp_xmit", i);
1598	ethtool_sprintf(data: &p, fmt: "rx_queue_%u_packets", i);
1599	ethtool_sprintf(data: &p, fmt: "rx_queue_%u_bytes", i);
1600	ethtool_sprintf(data: &p, fmt: "rx_queue_%u_xdp_drop", i);
1601	ethtool_sprintf(data: &p, fmt: "rx_queue_%u_xdp_redirect", i);
1602	ethtool_sprintf(data: &p, fmt: "rx_queue_%u_xdp_tx", i);
1603	}
1604
1605	for_each_present_cpu(cpu) {
1606	for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++)
1607	ethtool_sprintf(data: &p, fmt: pcpu_stats[i].name, cpu);
1608	}
1609
1610	break;
1611	}
1612	}
1613
1614	static int
1615	netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1616	struct ethtool_rxnfc *info)
1617	{
1618	const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1619
1620	info->data = RXH_IP_SRC | RXH_IP_DST;
1621
1622	switch (info->flow_type) {
1623	case TCP_V4_FLOW:
1624	if (ndc->l4_hash & HV_TCP4_L4HASH)
1625	info->data |= l4_flag;
1626
1627	break;
1628
1629	case TCP_V6_FLOW:
1630	if (ndc->l4_hash & HV_TCP6_L4HASH)
1631	info->data |= l4_flag;
1632
1633	break;
1634
1635	case UDP_V4_FLOW:
1636	if (ndc->l4_hash & HV_UDP4_L4HASH)
1637	info->data |= l4_flag;
1638
1639	break;
1640
1641	case UDP_V6_FLOW:
1642	if (ndc->l4_hash & HV_UDP6_L4HASH)
1643	info->data |= l4_flag;
1644
1645	break;
1646
1647	case IPV4_FLOW:
1648	case IPV6_FLOW:
1649	break;
1650	default:
1651	info->data = 0;
1652	break;
1653	}
1654
1655	return 0;
1656	}
1657
1658	static int
1659	netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1660	u32 *rules)
1661	{
1662	struct net_device_context *ndc = netdev_priv(dev);
1663	struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1664
1665	if (!nvdev)
1666	return -ENODEV;
1667
1668	switch (info->cmd) {
1669	case ETHTOOL_GRXRINGS:
1670	info->data = nvdev->num_chn;
1671	return 0;
1672
1673	case ETHTOOL_GRXFH:
1674	return netvsc_get_rss_hash_opts(ndc, info);
1675	}
1676	return -EOPNOTSUPP;
1677	}
1678
1679	static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1680	struct ethtool_rxnfc *info)
1681	{
1682	if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1683	RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1684	switch (info->flow_type) {
1685	case TCP_V4_FLOW:
1686	ndc->l4_hash |= HV_TCP4_L4HASH;
1687	break;
1688
1689	case TCP_V6_FLOW:
1690	ndc->l4_hash |= HV_TCP6_L4HASH;
1691	break;
1692
1693	case UDP_V4_FLOW:
1694	ndc->l4_hash |= HV_UDP4_L4HASH;
1695	break;
1696
1697	case UDP_V6_FLOW:
1698	ndc->l4_hash |= HV_UDP6_L4HASH;
1699	break;
1700
1701	default:
1702	return -EOPNOTSUPP;
1703	}
1704
1705	return 0;
1706	}
1707
1708	if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1709	switch (info->flow_type) {
1710	case TCP_V4_FLOW:
1711	ndc->l4_hash &= ~HV_TCP4_L4HASH;
1712	break;
1713
1714	case TCP_V6_FLOW:
1715	ndc->l4_hash &= ~HV_TCP6_L4HASH;
1716	break;
1717
1718	case UDP_V4_FLOW:
1719	ndc->l4_hash &= ~HV_UDP4_L4HASH;
1720	break;
1721
1722	case UDP_V6_FLOW:
1723	ndc->l4_hash &= ~HV_UDP6_L4HASH;
1724	break;
1725
1726	default:
1727	return -EOPNOTSUPP;
1728	}
1729
1730	return 0;
1731	}
1732
1733	return -EOPNOTSUPP;
1734	}
1735
1736	static int
1737	netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1738	{
1739	struct net_device_context *ndc = netdev_priv(dev: ndev);
1740
1741	if (info->cmd == ETHTOOL_SRXFH)
1742	return netvsc_set_rss_hash_opts(ndc, info);
1743
1744	return -EOPNOTSUPP;
1745	}
1746
1747	static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1748	{
1749	return NETVSC_HASH_KEYLEN;
1750	}
1751
1752	static u32 netvsc_rss_indir_size(struct net_device *dev)
1753	{
1754	struct net_device_context *ndc = netdev_priv(dev);
1755
1756	return ndc->rx_table_sz;
1757	}
1758
1759	static int netvsc_get_rxfh(struct net_device *dev,
1760	struct ethtool_rxfh_param *rxfh)
1761	{
1762	struct net_device_context *ndc = netdev_priv(dev);
1763	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1764	struct rndis_device *rndis_dev;
1765	int i;
1766
1767	if (!ndev)
1768	return -ENODEV;
1769
1770	rxfh->hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1771
1772	rndis_dev = ndev->extension;
1773	if (rxfh->indir) {
1774	for (i = 0; i < ndc->rx_table_sz; i++)
1775	rxfh->indir[i] = ndc->rx_table[i];
1776	}
1777
1778	if (rxfh->key)
1779	memcpy(rxfh->key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1780
1781	return 0;
1782	}
1783
1784	static int netvsc_set_rxfh(struct net_device *dev,
1785	struct ethtool_rxfh_param *rxfh,
1786	struct netlink_ext_ack *extack)
1787	{
1788	struct net_device_context *ndc = netdev_priv(dev);
1789	struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1790	struct rndis_device *rndis_dev;
1791	u8 *key = rxfh->key;
1792	int i;
1793
1794	if (!ndev)
1795	return -ENODEV;
1796
1797	if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
1798	rxfh->hfunc != ETH_RSS_HASH_TOP)
1799	return -EOPNOTSUPP;
1800
1801	rndis_dev = ndev->extension;
1802	if (rxfh->indir) {
1803	for (i = 0; i < ndc->rx_table_sz; i++)
1804	if (rxfh->indir[i] >= ndev->num_chn)
1805	return -EINVAL;
1806
1807	for (i = 0; i < ndc->rx_table_sz; i++)
1808	ndc->rx_table[i] = rxfh->indir[i];
1809	}
1810
1811	if (!key) {
1812	if (!rxfh->indir)
1813	return 0;
1814
1815	key = rndis_dev->rss_key;
1816	}
1817
1818	return rndis_filter_set_rss_param(rdev: rndis_dev, key);
1819	}
1820
1821	/* Hyper-V RNDIS protocol does not have ring in the HW sense.
1822	* It does have pre-allocated receive area which is divided into sections.
1823	*/
1824	static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1825	struct ethtool_ringparam *ring)
1826	{
1827	u32 max_buf_size;
1828
1829	ring->rx_pending = nvdev->recv_section_cnt;
1830	ring->tx_pending = nvdev->send_section_cnt;
1831
1832	if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1833	max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1834	else
1835	max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1836
1837	ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1838	ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1839	/ nvdev->send_section_size;
1840	}
1841
1842	static void netvsc_get_ringparam(struct net_device *ndev,
1843	struct ethtool_ringparam *ring,
1844	struct kernel_ethtool_ringparam *kernel_ring,
1845	struct netlink_ext_ack *extack)
1846	{
1847	struct net_device_context *ndevctx = netdev_priv(dev: ndev);
1848	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1849
1850	if (!nvdev)
1851	return;
1852
1853	__netvsc_get_ringparam(nvdev, ring);
1854	}
1855
1856	static int netvsc_set_ringparam(struct net_device *ndev,
1857	struct ethtool_ringparam *ring,
1858	struct kernel_ethtool_ringparam *kernel_ring,
1859	struct netlink_ext_ack *extack)
1860	{
1861	struct net_device_context *ndevctx = netdev_priv(dev: ndev);
1862	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1863	struct netvsc_device_info *device_info;
1864	struct ethtool_ringparam orig;
1865	u32 new_tx, new_rx;
1866	int ret = 0;
1867
1868	if (!nvdev || nvdev->destroy)
1869	return -ENODEV;
1870
1871	memset(&orig, 0, sizeof(orig));
1872	__netvsc_get_ringparam(nvdev, ring: &orig);
1873
1874	new_tx = clamp_t(u32, ring->tx_pending,
1875	NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1876	new_rx = clamp_t(u32, ring->rx_pending,
1877	NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1878
1879	if (new_tx == orig.tx_pending &&
1880	new_rx == orig.rx_pending)
1881	return 0; /* no change */
1882
1883	device_info = netvsc_devinfo_get(nvdev);
1884
1885	if (!device_info)
1886	return -ENOMEM;
1887
1888	device_info->send_sections = new_tx;
1889	device_info->recv_sections = new_rx;
1890
1891	ret = netvsc_detach(ndev, nvdev);
1892	if (ret)
1893	goto out;
1894
1895	ret = netvsc_attach(ndev, dev_info: device_info);
1896	if (ret) {
1897	device_info->send_sections = orig.tx_pending;
1898	device_info->recv_sections = orig.rx_pending;
1899
1900	if (netvsc_attach(ndev, dev_info: device_info))
1901	netdev_err(dev: ndev, format: "restoring ringparam failed");
1902	}
1903
1904	out:
1905	netvsc_devinfo_put(dev_info: device_info);
1906	return ret;
1907	}
1908
1909	static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1910	netdev_features_t features)
1911	{
1912	struct net_device_context *ndevctx = netdev_priv(dev: ndev);
1913	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1914
1915	if (!nvdev || nvdev->destroy)
1916	return features;
1917
1918	if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1919	features ^= NETIF_F_LRO;
1920	netdev_info(dev: ndev, format: "Skip LRO - unsupported with XDP\n");
1921	}
1922
1923	return features;
1924	}
1925
1926	static int netvsc_set_features(struct net_device *ndev,
1927	netdev_features_t features)
1928	{
1929	netdev_features_t change = features ^ ndev->features;
1930	struct net_device_context *ndevctx = netdev_priv(dev: ndev);
1931	struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1932	struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1933	struct ndis_offload_params offloads;
1934	int ret = 0;
1935
1936	if (!nvdev || nvdev->destroy)
1937	return -ENODEV;
1938
1939	if (!(change & NETIF_F_LRO))
1940	goto syncvf;
1941
1942	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1943
1944	if (features & NETIF_F_LRO) {
1945	offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1946	offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1947	} else {
1948	offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1949	offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1950	}
1951
1952	ret = rndis_filter_set_offload_params(ndev, nvdev, req_offloads: &offloads);
1953
1954	if (ret) {
1955	features ^= NETIF_F_LRO;
1956	ndev->features = features;
1957	}
1958
1959	syncvf:
1960	if (!vf_netdev)
1961	return ret;
1962
1963	vf_netdev->wanted_features = features;
1964	netdev_update_features(dev: vf_netdev);
1965
1966	return ret;
1967	}
1968
1969	static int netvsc_get_regs_len(struct net_device *netdev)
1970	{
1971	return VRSS_SEND_TAB_SIZE * sizeof(u32);
1972	}
1973
1974	static void netvsc_get_regs(struct net_device *netdev,
1975	struct ethtool_regs *regs, void *p)
1976	{
1977	struct net_device_context *ndc = netdev_priv(dev: netdev);
1978	u32 *regs_buff = p;
1979
1980	/* increase the version, if buffer format is changed. */
1981	regs->version = 1;
1982
1983	memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
1984	}
1985
1986	static u32 netvsc_get_msglevel(struct net_device *ndev)
1987	{
1988	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1989
1990	return ndev_ctx->msg_enable;
1991	}
1992
1993	static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1994	{
1995	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1996
1997	ndev_ctx->msg_enable = val;
1998	}
1999
2000	static const struct ethtool_ops ethtool_ops = {
2001	.get_drvinfo = netvsc_get_drvinfo,
2002	.get_regs_len = netvsc_get_regs_len,
2003	.get_regs = netvsc_get_regs,
2004	.get_msglevel = netvsc_get_msglevel,
2005	.set_msglevel = netvsc_set_msglevel,
2006	.get_link = ethtool_op_get_link,
2007	.get_ethtool_stats = netvsc_get_ethtool_stats,
2008	.get_sset_count = netvsc_get_sset_count,
2009	.get_strings = netvsc_get_strings,
2010	.get_channels = netvsc_get_channels,
2011	.set_channels = netvsc_set_channels,
2012	.get_ts_info = ethtool_op_get_ts_info,
2013	.get_rxnfc = netvsc_get_rxnfc,
2014	.set_rxnfc = netvsc_set_rxnfc,
2015	.get_rxfh_key_size = netvsc_get_rxfh_key_size,
2016	.get_rxfh_indir_size = netvsc_rss_indir_size,
2017	.get_rxfh = netvsc_get_rxfh,
2018	.set_rxfh = netvsc_set_rxfh,
2019	.get_link_ksettings = netvsc_get_link_ksettings,
2020	.set_link_ksettings = netvsc_set_link_ksettings,
2021	.get_ringparam = netvsc_get_ringparam,
2022	.set_ringparam = netvsc_set_ringparam,
2023	};
2024
2025	static const struct net_device_ops device_ops = {
2026	.ndo_open = netvsc_open,
2027	.ndo_stop = netvsc_close,
2028	.ndo_start_xmit = netvsc_start_xmit,
2029	.ndo_change_rx_flags = netvsc_change_rx_flags,
2030	.ndo_set_rx_mode = netvsc_set_rx_mode,
2031	.ndo_fix_features = netvsc_fix_features,
2032	.ndo_set_features = netvsc_set_features,
2033	.ndo_change_mtu = netvsc_change_mtu,
2034	.ndo_validate_addr = eth_validate_addr,
2035	.ndo_set_mac_address = netvsc_set_mac_addr,
2036	.ndo_select_queue = netvsc_select_queue,
2037	.ndo_get_stats64 = netvsc_get_stats64,
2038	.ndo_bpf = netvsc_bpf,
2039	.ndo_xdp_xmit = netvsc_ndoxdp_xmit,
2040	};
2041
2042	/*
2043	* Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2044	* down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2045	* present send GARP packet to network peers with netif_notify_peers().
2046	*/
2047	static void netvsc_link_change(struct work_struct *w)
2048	{
2049	struct net_device_context *ndev_ctx =
2050	container_of(w, struct net_device_context, dwork.work);
2051	struct hv_device *device_obj = ndev_ctx->device_ctx;
2052	struct net_device *net = hv_get_drvdata(dev: device_obj);
2053	unsigned long flags, next_reconfig, delay;
2054	struct netvsc_reconfig *event = NULL;
2055	struct netvsc_device *net_device;
2056	struct rndis_device *rdev;
2057	bool reschedule = false;
2058
2059	/* if changes are happening, comeback later */
2060	if (!rtnl_trylock()) {
2061	schedule_delayed_work(dwork: &ndev_ctx->dwork, LINKCHANGE_INT);
2062	return;
2063	}
2064
2065	net_device = rtnl_dereference(ndev_ctx->nvdev);
2066	if (!net_device)
2067	goto out_unlock;
2068
2069	rdev = net_device->extension;
2070
2071	next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2072	if (time_is_after_jiffies(next_reconfig)) {
2073	/* link_watch only sends one notification with current state
2074	* per second, avoid doing reconfig more frequently. Handle
2075	* wrap around.
2076	*/
2077	delay = next_reconfig - jiffies;
2078	delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2079	schedule_delayed_work(dwork: &ndev_ctx->dwork, delay);
2080	goto out_unlock;
2081	}
2082	ndev_ctx->last_reconfig = jiffies;
2083
2084	spin_lock_irqsave(&ndev_ctx->lock, flags);
2085	if (!list_empty(head: &ndev_ctx->reconfig_events)) {
2086	event = list_first_entry(&ndev_ctx->reconfig_events,
2087	struct netvsc_reconfig, list);
2088	list_del(entry: &event->list);
2089	reschedule = !list_empty(head: &ndev_ctx->reconfig_events);
2090	}
2091	spin_unlock_irqrestore(lock: &ndev_ctx->lock, flags);
2092
2093	if (!event)
2094	goto out_unlock;
2095
2096	switch (event->event) {
2097	/* Only the following events are possible due to the check in
2098	* netvsc_linkstatus_callback()
2099	*/
2100	case RNDIS_STATUS_MEDIA_CONNECT:
2101	if (rdev->link_state) {
2102	rdev->link_state = false;
2103	netif_carrier_on(dev: net);
2104	netvsc_tx_enable(nvscdev: net_device, ndev: net);
2105	} else {
2106	__netdev_notify_peers(dev: net);
2107	}
2108	kfree(objp: event);
2109	break;
2110	case RNDIS_STATUS_MEDIA_DISCONNECT:
2111	if (!rdev->link_state) {
2112	rdev->link_state = true;
2113	netif_carrier_off(dev: net);
2114	netvsc_tx_disable(nvscdev: net_device, ndev: net);
2115	}
2116	kfree(objp: event);
2117	break;
2118	case RNDIS_STATUS_NETWORK_CHANGE:
2119	/* Only makes sense if carrier is present */
2120	if (!rdev->link_state) {
2121	rdev->link_state = true;
2122	netif_carrier_off(dev: net);
2123	netvsc_tx_disable(nvscdev: net_device, ndev: net);
2124	event->event = RNDIS_STATUS_MEDIA_CONNECT;
2125	spin_lock_irqsave(&ndev_ctx->lock, flags);
2126	list_add(new: &event->list, head: &ndev_ctx->reconfig_events);
2127	spin_unlock_irqrestore(lock: &ndev_ctx->lock, flags);
2128	reschedule = true;
2129	}
2130	break;
2131	}
2132
2133	rtnl_unlock();
2134
2135	/* link_watch only sends one notification with current state per
2136	* second, handle next reconfig event in 2 seconds.
2137	*/
2138	if (reschedule)
2139	schedule_delayed_work(dwork: &ndev_ctx->dwork, LINKCHANGE_INT);
2140
2141	return;
2142
2143	out_unlock:
2144	rtnl_unlock();
2145	}
2146
2147	static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2148	{
2149	struct net_device_context *net_device_ctx;
2150	struct net_device *dev;
2151
2152	dev = netdev_master_upper_dev_get(dev: vf_netdev);
2153	if (!dev || dev->netdev_ops != &device_ops)
2154	return NULL; /* not a netvsc device */
2155
2156	net_device_ctx = netdev_priv(dev);
2157	if (!rtnl_dereference(net_device_ctx->nvdev))
2158	return NULL; /* device is removed */
2159
2160	return dev;
2161	}
2162
2163	/* Called when VF is injecting data into network stack.
2164	* Change the associated network device from VF to netvsc.
2165	* note: already called with rcu_read_lock
2166	*/
2167	static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2168	{
2169	struct sk_buff *skb = *pskb;
2170	struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2171	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
2172	struct netvsc_vf_pcpu_stats *pcpu_stats
2173	= this_cpu_ptr(ndev_ctx->vf_stats);
2174
2175	skb = skb_share_check(skb, GFP_ATOMIC);
2176	if (unlikely(!skb))
2177	return RX_HANDLER_CONSUMED;
2178
2179	*pskb = skb;
2180
2181	skb->dev = ndev;
2182
2183	u64_stats_update_begin(syncp: &pcpu_stats->syncp);
2184	pcpu_stats->rx_packets++;
2185	pcpu_stats->rx_bytes += skb->len;
2186	u64_stats_update_end(syncp: &pcpu_stats->syncp);
2187
2188	return RX_HANDLER_ANOTHER;
2189	}
2190
2191	static int netvsc_vf_join(struct net_device *vf_netdev,
2192	struct net_device *ndev, int context)
2193	{
2194	struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
2195	int ret;
2196
2197	ret = netdev_rx_handler_register(dev: vf_netdev,
2198	rx_handler: netvsc_vf_handle_frame, rx_handler_data: ndev);
2199	if (ret != 0) {
2200	netdev_err(dev: vf_netdev,
2201	format: "can not register netvsc VF receive handler (err = %d)\n",
2202	ret);
2203	goto rx_handler_failed;
2204	}
2205
2206	ret = netdev_master_upper_dev_link(dev: vf_netdev, upper_dev: ndev,
2207	NULL, NULL, NULL);
2208	if (ret != 0) {
2209	netdev_err(dev: vf_netdev,
2210	format: "can not set master device %s (err = %d)\n",
2211	ndev->name, ret);
2212	goto upper_link_failed;
2213	}
2214
2215	/* If this registration is called from probe context vf_takeover
2216	* is taken care of later in probe itself.
2217	*/
2218	if (context == VF_REG_IN_NOTIFIER)
2219	schedule_delayed_work(dwork: &ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2220
2221	call_netdevice_notifiers(val: NETDEV_JOIN, dev: vf_netdev);
2222
2223	netdev_info(dev: vf_netdev, format: "joined to %s\n", ndev->name);
2224	return 0;
2225
2226	upper_link_failed:
2227	netdev_rx_handler_unregister(dev: vf_netdev);
2228	rx_handler_failed:
2229	return ret;
2230	}
2231
2232	static void __netvsc_vf_setup(struct net_device *ndev,
2233	struct net_device *vf_netdev)
2234	{
2235	int ret;
2236
2237	/* Align MTU of VF with master */
2238	ret = dev_set_mtu(vf_netdev, ndev->mtu);
2239	if (ret)
2240	netdev_warn(dev: vf_netdev,
2241	format: "unable to change mtu to %u\n", ndev->mtu);
2242
2243	/* set multicast etc flags on VF */
2244	dev_change_flags(dev: vf_netdev, flags: ndev->flags | IFF_SLAVE, NULL);
2245
2246	/* sync address list from ndev to VF */
2247	netif_addr_lock_bh(dev: ndev);
2248	dev_uc_sync(to: vf_netdev, from: ndev);
2249	dev_mc_sync(to: vf_netdev, from: ndev);
2250	netif_addr_unlock_bh(dev: ndev);
2251
2252	if (netif_running(dev: ndev)) {
2253	ret = dev_open(dev: vf_netdev, NULL);
2254	if (ret)
2255	netdev_warn(dev: vf_netdev,
2256	format: "unable to open: %d\n", ret);
2257	}
2258	}
2259
2260	/* Setup VF as slave of the synthetic device.
2261	* Runs in workqueue to avoid recursion in netlink callbacks.
2262	*/
2263	static void netvsc_vf_setup(struct work_struct *w)
2264	{
2265	struct net_device_context *ndev_ctx
2266	= container_of(w, struct net_device_context, vf_takeover.work);
2267	struct net_device *ndev = hv_get_drvdata(dev: ndev_ctx->device_ctx);
2268	struct net_device *vf_netdev;
2269
2270	if (!rtnl_trylock()) {
2271	schedule_delayed_work(dwork: &ndev_ctx->vf_takeover, delay: 0);
2272	return;
2273	}
2274
2275	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2276	if (vf_netdev)
2277	__netvsc_vf_setup(ndev, vf_netdev);
2278
2279	rtnl_unlock();
2280	}
2281
2282	/* Find netvsc by VF serial number.
2283	* The PCI hyperv controller records the serial number as the slot kobj name.
2284	*/
2285	static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2286	{
2287	struct device *parent = vf_netdev->dev.parent;
2288	struct net_device_context *ndev_ctx;
2289	struct net_device *ndev;
2290	struct pci_dev *pdev;
2291	u32 serial;
2292
2293	if (!parent || !dev_is_pci(parent))
2294	return NULL; /* not a PCI device */
2295
2296	pdev = to_pci_dev(parent);
2297	if (!pdev->slot) {
2298	netdev_notice(dev: vf_netdev, format: "no PCI slot information\n");
2299	return NULL;
2300	}
2301
2302	if (kstrtou32(s: pci_slot_name(slot: pdev->slot), base: 10, res: &serial)) {
2303	netdev_notice(dev: vf_netdev, format: "Invalid vf serial:%s\n",
2304	pci_slot_name(slot: pdev->slot));
2305	return NULL;
2306	}
2307
2308	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2309	if (!ndev_ctx->vf_alloc)
2310	continue;
2311
2312	if (ndev_ctx->vf_serial != serial)
2313	continue;
2314
2315	ndev = hv_get_drvdata(dev: ndev_ctx->device_ctx);
2316	if (ndev->addr_len != vf_netdev->addr_len ||
2317	memcmp(p: ndev->perm_addr, q: vf_netdev->perm_addr,
2318	size: ndev->addr_len) != 0)
2319	continue;
2320
2321	return ndev;
2322
2323	}
2324
2325	/* Fallback path to check synthetic vf with help of mac addr.
2326	* Because this function can be called before vf_netdev is
2327	* initialized (NETDEV_POST_INIT) when its perm_addr has not been copied
2328	* from dev_addr, also try to match to its dev_addr.
2329	* Note: On Hyper-V and Azure, it's not possible to set a MAC address
2330	* on a VF that matches to the MAC of a unrelated NETVSC device.
2331	*/
2332	list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2333	ndev = hv_get_drvdata(dev: ndev_ctx->device_ctx);
2334	if (ether_addr_equal(addr1: vf_netdev->perm_addr, addr2: ndev->perm_addr) ||
2335	ether_addr_equal(addr1: vf_netdev->dev_addr, addr2: ndev->perm_addr))
2336	return ndev;
2337	}
2338
2339	netdev_notice(dev: vf_netdev,
2340	format: "no netdev found for vf serial:%u\n", serial);
2341	return NULL;
2342	}
2343
2344	static int netvsc_prepare_bonding(struct net_device *vf_netdev)
2345	{
2346	struct net_device *ndev;
2347
2348	ndev = get_netvsc_byslot(vf_netdev);
2349	if (!ndev)
2350	return NOTIFY_DONE;
2351
2352	/* set slave flag before open to prevent IPv6 addrconf */
2353	vf_netdev->flags |= IFF_SLAVE;
2354	return NOTIFY_DONE;
2355	}
2356
2357	static int netvsc_register_vf(struct net_device *vf_netdev, int context)
2358	{
2359	struct net_device_context *net_device_ctx;
2360	struct netvsc_device *netvsc_dev;
2361	struct bpf_prog *prog;
2362	struct net_device *ndev;
2363	int ret;
2364
2365	if (vf_netdev->addr_len != ETH_ALEN)
2366	return NOTIFY_DONE;
2367
2368	ndev = get_netvsc_byslot(vf_netdev);
2369	if (!ndev)
2370	return NOTIFY_DONE;
2371
2372	net_device_ctx = netdev_priv(dev: ndev);
2373	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2374	if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2375	return NOTIFY_DONE;
2376
2377	/* if synthetic interface is a different namespace,
2378	* then move the VF to that namespace; join will be
2379	* done again in that context.
2380	*/
2381	if (!net_eq(net1: dev_net(dev: ndev), net2: dev_net(dev: vf_netdev))) {
2382	ret = dev_change_net_namespace(dev: vf_netdev,
2383	net: dev_net(dev: ndev), pat: "eth%d");
2384	if (ret)
2385	netdev_err(dev: vf_netdev,
2386	format: "could not move to same namespace as %s: %d\n",
2387	ndev->name, ret);
2388	else
2389	netdev_info(dev: vf_netdev,
2390	format: "VF moved to namespace with: %s\n",
2391	ndev->name);
2392	return NOTIFY_DONE;
2393	}
2394
2395	netdev_info(dev: ndev, format: "VF registering: %s\n", vf_netdev->name);
2396
2397	if (netvsc_vf_join(vf_netdev, ndev, context) != 0)
2398	return NOTIFY_DONE;
2399
2400	dev_hold(dev: vf_netdev);
2401	rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2402
2403	if (ndev->needed_headroom < vf_netdev->needed_headroom)
2404	ndev->needed_headroom = vf_netdev->needed_headroom;
2405
2406	vf_netdev->wanted_features = ndev->features;
2407	netdev_update_features(dev: vf_netdev);
2408
2409	prog = netvsc_xdp_get(nvdev: netvsc_dev);
2410	netvsc_vf_setxdp(vf_netdev, prog);
2411
2412	return NOTIFY_OK;
2413	}
2414
2415	/* Change the data path when VF UP/DOWN/CHANGE are detected.
2416	*
2417	* Typically a UP or DOWN event is followed by a CHANGE event, so
2418	* net_device_ctx->data_path_is_vf is used to cache the current data path
2419	* to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2420	* message.
2421	*
2422	* During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2423	* interface, there is only the CHANGE event and no UP or DOWN event.
2424	*/
2425	static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2426	{
2427	struct net_device_context *net_device_ctx;
2428	struct netvsc_device *netvsc_dev;
2429	struct net_device *ndev;
2430	bool vf_is_up = false;
2431	int ret;
2432
2433	if (event != NETDEV_GOING_DOWN)
2434	vf_is_up = netif_running(dev: vf_netdev);
2435
2436	ndev = get_netvsc_byref(vf_netdev);
2437	if (!ndev)
2438	return NOTIFY_DONE;
2439
2440	net_device_ctx = netdev_priv(dev: ndev);
2441	netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2442	if (!netvsc_dev)
2443	return NOTIFY_DONE;
2444
2445	if (net_device_ctx->data_path_is_vf == vf_is_up)
2446	return NOTIFY_OK;
2447
2448	if (vf_is_up && !net_device_ctx->vf_alloc) {
2449	netdev_info(dev: ndev, format: "Waiting for the VF association from host\n");
2450	wait_for_completion(&net_device_ctx->vf_add);
2451	}
2452
2453	ret = netvsc_switch_datapath(nv_dev: ndev, vf: vf_is_up);
2454
2455	if (ret) {
2456	netdev_err(dev: ndev,
2457	format: "Data path failed to switch %s VF: %s, err: %d\n",
2458	vf_is_up ? "to" : "from", vf_netdev->name, ret);
2459	return NOTIFY_DONE;
2460	} else {
2461	netdev_info(dev: ndev, format: "Data path switched %s VF: %s\n",
2462	vf_is_up ? "to" : "from", vf_netdev->name);
2463	}
2464
2465	return NOTIFY_OK;
2466	}
2467
2468	static int netvsc_unregister_vf(struct net_device *vf_netdev)
2469	{
2470	struct net_device *ndev;
2471	struct net_device_context *net_device_ctx;
2472
2473	ndev = get_netvsc_byref(vf_netdev);
2474	if (!ndev)
2475	return NOTIFY_DONE;
2476
2477	net_device_ctx = netdev_priv(dev: ndev);
2478	cancel_delayed_work_sync(dwork: &net_device_ctx->vf_takeover);
2479
2480	netdev_info(dev: ndev, format: "VF unregistering: %s\n", vf_netdev->name);
2481
2482	netvsc_vf_setxdp(vf_netdev, NULL);
2483
2484	reinit_completion(x: &net_device_ctx->vf_add);
2485	netdev_rx_handler_unregister(dev: vf_netdev);
2486	netdev_upper_dev_unlink(dev: vf_netdev, upper_dev: ndev);
2487	RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2488	dev_put(dev: vf_netdev);
2489
2490	ndev->needed_headroom = RNDIS_AND_PPI_SIZE;
2491
2492	return NOTIFY_OK;
2493	}
2494
2495	static int check_dev_is_matching_vf(struct net_device *event_ndev)
2496	{
2497	/* Skip NetVSC interfaces */
2498	if (event_ndev->netdev_ops == &device_ops)
2499	return -ENODEV;
2500
2501	/* Avoid non-Ethernet type devices */
2502	if (event_ndev->type != ARPHRD_ETHER)
2503	return -ENODEV;
2504
2505	/* Avoid Vlan dev with same MAC registering as VF */
2506	if (is_vlan_dev(dev: event_ndev))
2507	return -ENODEV;
2508
2509	/* Avoid Bonding master dev with same MAC registering as VF */
2510	if (netif_is_bond_master(dev: event_ndev))
2511	return -ENODEV;
2512
2513	return 0;
2514	}
2515
2516	static int netvsc_probe(struct hv_device *dev,
2517	const struct hv_vmbus_device_id *dev_id)
2518	{
2519	struct net_device *net = NULL, *vf_netdev;
2520	struct net_device_context *net_device_ctx;
2521	struct netvsc_device_info *device_info = NULL;
2522	struct netvsc_device *nvdev;
2523	int ret = -ENOMEM;
2524
2525	net = alloc_etherdev_mq(sizeof(struct net_device_context),
2526	VRSS_CHANNEL_MAX);
2527	if (!net)
2528	goto no_net;
2529
2530	netif_carrier_off(dev: net);
2531
2532	netvsc_init_settings(dev: net);
2533
2534	net_device_ctx = netdev_priv(dev: net);
2535	net_device_ctx->device_ctx = dev;
2536	net_device_ctx->msg_enable = netif_msg_init(debug_value: debug, default_msg_enable_bits: default_msg);
2537	if (netif_msg_probe(net_device_ctx))
2538	netdev_dbg(net, "netvsc msg_enable: %d\n",
2539	net_device_ctx->msg_enable);
2540
2541	hv_set_drvdata(dev, data: net);
2542
2543	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2544
2545	init_completion(x: &net_device_ctx->vf_add);
2546	spin_lock_init(&net_device_ctx->lock);
2547	INIT_LIST_HEAD(list: &net_device_ctx->reconfig_events);
2548	INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2549
2550	net_device_ctx->vf_stats
2551	= netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2552	if (!net_device_ctx->vf_stats)
2553	goto no_stats;
2554
2555	net->netdev_ops = &device_ops;
2556	net->ethtool_ops = &ethtool_ops;
2557	SET_NETDEV_DEV(net, &dev->device);
2558	dma_set_min_align_mask(dev: &dev->device, HV_HYP_PAGE_SIZE - 1);
2559
2560	/* We always need headroom for rndis header */
2561	net->needed_headroom = RNDIS_AND_PPI_SIZE;
2562
2563	/* Initialize the number of queues to be 1, we may change it if more
2564	* channels are offered later.
2565	*/
2566	netif_set_real_num_tx_queues(dev: net, txq: 1);
2567	netif_set_real_num_rx_queues(dev: net, rxq: 1);
2568
2569	/* Notify the netvsc driver of the new device */
2570	device_info = netvsc_devinfo_get(NULL);
2571
2572	if (!device_info) {
2573	ret = -ENOMEM;
2574	goto devinfo_failed;
2575	}
2576
2577	/* We must get rtnl lock before scheduling nvdev->subchan_work,
2578	* otherwise netvsc_subchan_work() can get rtnl lock first and wait
2579	* all subchannels to show up, but that may not happen because
2580	* netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2581	* -> ... -> device_add() -> ... -> __device_attach() can't get
2582	* the device lock, so all the subchannels can't be processed --
2583	* finally netvsc_subchan_work() hangs forever.
2584	*
2585	* The rtnl lock also needs to be held before rndis_filter_device_add()
2586	* which advertises nvsp_2_vsc_capability / sriov bit, and triggers
2587	* VF NIC offering and registering. If VF NIC finished register_netdev()
2588	* earlier it may cause name based config failure.
2589	*/
2590	rtnl_lock();
2591
2592	nvdev = rndis_filter_device_add(dev, info: device_info);
2593	if (IS_ERR(ptr: nvdev)) {
2594	ret = PTR_ERR(ptr: nvdev);
2595	netdev_err(dev: net, format: "unable to add netvsc device (ret %d)\n", ret);
2596	goto rndis_failed;
2597	}
2598
2599	eth_hw_addr_set(dev: net, addr: device_info->mac_adr);
2600
2601	if (nvdev->num_chn > 1)
2602	schedule_work(work: &nvdev->subchan_work);
2603
2604	/* hw_features computed in rndis_netdev_set_hwcaps() */
2605	net->features = net->hw_features |
2606	NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2607	NETIF_F_HW_VLAN_CTAG_RX;
2608	net->vlan_features = net->features;
2609
2610	netdev_lockdep_set_classes(net);
2611
2612	net->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
2613	NETDEV_XDP_ACT_NDO_XMIT;
2614
2615	/* MTU range: 68 - 1500 or 65521 */
2616	net->min_mtu = NETVSC_MTU_MIN;
2617	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2618	net->max_mtu = NETVSC_MTU - ETH_HLEN;
2619	else
2620	net->max_mtu = ETH_DATA_LEN;
2621
2622	nvdev->tx_disable = false;
2623
2624	ret = register_netdevice(dev: net);
2625	if (ret != 0) {
2626	pr_err("Unable to register netdev.\n");
2627	goto register_failed;
2628	}
2629
2630	list_add(new: &net_device_ctx->list, head: &netvsc_dev_list);
2631
2632	/* When the hv_netvsc driver is unloaded and reloaded, the
2633	* NET_DEVICE_REGISTER for the vf device is replayed before probe
2634	* is complete. This is because register_netdevice_notifier() gets
2635	* registered before vmbus_driver_register() so that callback func
2636	* is set before probe and we don't miss events like NETDEV_POST_INIT
2637	* So, in this section we try to register the matching vf device that
2638	* is present as a netdevice, knowing that its register call is not
2639	* processed in the netvsc_netdev_notifier(as probing is progress and
2640	* get_netvsc_byslot fails).
2641	*/
2642	for_each_netdev(dev_net(net), vf_netdev) {
2643	ret = check_dev_is_matching_vf(event_ndev: vf_netdev);
2644	if (ret != 0)
2645	continue;
2646
2647	if (net != get_netvsc_byslot(vf_netdev))
2648	continue;
2649
2650	netvsc_prepare_bonding(vf_netdev);
2651	netvsc_register_vf(vf_netdev, VF_REG_IN_PROBE);
2652	__netvsc_vf_setup(ndev: net, vf_netdev);
2653	break;
2654	}
2655	rtnl_unlock();
2656
2657	netvsc_devinfo_put(dev_info: device_info);
2658	return 0;
2659
2660	register_failed:
2661	rndis_filter_device_remove(dev, nvdev);
2662	rndis_failed:
2663	rtnl_unlock();
2664	netvsc_devinfo_put(dev_info: device_info);
2665	devinfo_failed:
2666	free_percpu(pdata: net_device_ctx->vf_stats);
2667	no_stats:
2668	hv_set_drvdata(dev, NULL);
2669	free_netdev(dev: net);
2670	no_net:
2671	return ret;
2672	}
2673
2674	static void netvsc_remove(struct hv_device *dev)
2675	{
2676	struct net_device_context *ndev_ctx;
2677	struct net_device *vf_netdev, *net;
2678	struct netvsc_device *nvdev;
2679
2680	net = hv_get_drvdata(dev);
2681	if (net == NULL) {
2682	dev_err(&dev->device, "No net device to remove\n");
2683	return;
2684	}
2685
2686	ndev_ctx = netdev_priv(dev: net);
2687
2688	cancel_delayed_work_sync(dwork: &ndev_ctx->dwork);
2689
2690	rtnl_lock();
2691	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2692	if (nvdev) {
2693	cancel_work_sync(work: &nvdev->subchan_work);
2694	netvsc_xdp_set(dev: net, NULL, NULL, nvdev);
2695	}
2696
2697	/*
2698	* Call to the vsc driver to let it know that the device is being
2699	* removed. Also blocks mtu and channel changes.
2700	*/
2701	vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2702	if (vf_netdev)
2703	netvsc_unregister_vf(vf_netdev);
2704
2705	if (nvdev)
2706	rndis_filter_device_remove(dev, nvdev);
2707
2708	unregister_netdevice(dev: net);
2709	list_del(entry: &ndev_ctx->list);
2710
2711	rtnl_unlock();
2712
2713	hv_set_drvdata(dev, NULL);
2714
2715	free_percpu(pdata: ndev_ctx->vf_stats);
2716	free_netdev(dev: net);
2717	}
2718
2719	static int netvsc_suspend(struct hv_device *dev)
2720	{
2721	struct net_device_context *ndev_ctx;
2722	struct netvsc_device *nvdev;
2723	struct net_device *net;
2724	int ret;
2725
2726	net = hv_get_drvdata(dev);
2727
2728	ndev_ctx = netdev_priv(dev: net);
2729	cancel_delayed_work_sync(dwork: &ndev_ctx->dwork);
2730
2731	rtnl_lock();
2732
2733	nvdev = rtnl_dereference(ndev_ctx->nvdev);
2734	if (nvdev == NULL) {
2735	ret = -ENODEV;
2736	goto out;
2737	}
2738
2739	/* Save the current config info */
2740	ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2741	if (!ndev_ctx->saved_netvsc_dev_info) {
2742	ret = -ENOMEM;
2743	goto out;
2744	}
2745	ret = netvsc_detach(ndev: net, nvdev);
2746	out:
2747	rtnl_unlock();
2748
2749	return ret;
2750	}
2751
2752	static int netvsc_resume(struct hv_device *dev)
2753	{
2754	struct net_device *net = hv_get_drvdata(dev);
2755	struct net_device_context *net_device_ctx;
2756	struct netvsc_device_info *device_info;
2757	int ret;
2758
2759	rtnl_lock();
2760
2761	net_device_ctx = netdev_priv(dev: net);
2762
2763	/* Reset the data path to the netvsc NIC before re-opening the vmbus
2764	* channel. Later netvsc_netdev_event() will switch the data path to
2765	* the VF upon the UP or CHANGE event.
2766	*/
2767	net_device_ctx->data_path_is_vf = false;
2768	device_info = net_device_ctx->saved_netvsc_dev_info;
2769
2770	ret = netvsc_attach(ndev: net, dev_info: device_info);
2771
2772	netvsc_devinfo_put(dev_info: device_info);
2773	net_device_ctx->saved_netvsc_dev_info = NULL;
2774
2775	rtnl_unlock();
2776
2777	return ret;
2778	}
2779	static const struct hv_vmbus_device_id id_table[] = {
2780	/* Network guid */
2781	{ HV_NIC_GUID, },
2782	{ },
2783	};
2784
2785	MODULE_DEVICE_TABLE(vmbus, id_table);
2786
2787	/* The one and only one */
2788	static struct hv_driver netvsc_drv = {
2789	.name = KBUILD_MODNAME,
2790	.id_table = id_table,
2791	.probe = netvsc_probe,
2792	.remove = netvsc_remove,
2793	.suspend = netvsc_suspend,
2794	.resume = netvsc_resume,
2795	.driver = {
2796	.probe_type = PROBE_FORCE_SYNCHRONOUS,
2797	},
2798	};
2799
2800	/*
2801	* On Hyper-V, every VF interface is matched with a corresponding
2802	* synthetic interface. The synthetic interface is presented first
2803	* to the guest. When the corresponding VF instance is registered,
2804	* we will take care of switching the data path.
2805	*/
2806	static int netvsc_netdev_event(struct notifier_block *this,
2807	unsigned long event, void *ptr)
2808	{
2809	struct net_device *event_dev = netdev_notifier_info_to_dev(info: ptr);
2810	int ret = 0;
2811
2812	ret = check_dev_is_matching_vf(event_ndev: event_dev);
2813	if (ret != 0)
2814	return NOTIFY_DONE;
2815
2816	switch (event) {
2817	case NETDEV_POST_INIT:
2818	return netvsc_prepare_bonding(vf_netdev: event_dev);
2819	case NETDEV_REGISTER:
2820	return netvsc_register_vf(vf_netdev: event_dev, VF_REG_IN_NOTIFIER);
2821	case NETDEV_UNREGISTER:
2822	return netvsc_unregister_vf(vf_netdev: event_dev);
2823	case NETDEV_UP:
2824	case NETDEV_DOWN:
2825	case NETDEV_CHANGE:
2826	case NETDEV_GOING_DOWN:
2827	return netvsc_vf_changed(vf_netdev: event_dev, event);
2828	default:
2829	return NOTIFY_DONE;
2830	}
2831	}
2832
2833	static struct notifier_block netvsc_netdev_notifier = {
2834	.notifier_call = netvsc_netdev_event,
2835	};
2836
2837	static void __exit netvsc_drv_exit(void)
2838	{
2839	unregister_netdevice_notifier(nb: &netvsc_netdev_notifier);
2840	vmbus_driver_unregister(hv_driver: &netvsc_drv);
2841	}
2842
2843	static int __init netvsc_drv_init(void)
2844	{
2845	int ret;
2846
2847	if (ring_size < RING_SIZE_MIN) {
2848	ring_size = RING_SIZE_MIN;
2849	pr_info("Increased ring_size to %u (min allowed)\n",
2850	ring_size);
2851	}
2852	netvsc_ring_bytes = VMBUS_RING_SIZE(ring_size * 4096);
2853
2854	register_netdevice_notifier(nb: &netvsc_netdev_notifier);
2855
2856	ret = vmbus_driver_register(&netvsc_drv);
2857	if (ret)
2858	goto err_vmbus_reg;
2859
2860	return 0;
2861
2862	err_vmbus_reg:
2863	unregister_netdevice_notifier(nb: &netvsc_netdev_notifier);
2864	return ret;
2865	}
2866
2867	MODULE_LICENSE("GPL");
2868	MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2869
2870	module_init(netvsc_drv_init);
2871	module_exit(netvsc_drv_exit);
2872