netvsc_drv.c

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