drivers/net/hyperv/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/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 #define RETRY_US_LO     5000
  42 #define RETRY_US_HI     10000
  43 #define RETRY_MAX       2000    /* >10 sec */
  44
  45 #define LINKCHANGE_INT (2 * HZ)
  46 #define VF_TAKEOVER_INT (HZ / 10)
  47
  48 static unsigned int ring_size __ro_after_init = 128;
  49 module_param(ring_size, uint, 0444);
  50 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  51 unsigned int netvsc_ring_bytes __ro_after_init;
  52
  53 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
  54                                 NETIF_MSG_LINK | NETIF_MSG_IFUP |
  55                                 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
  56                                 NETIF_MSG_TX_ERR;
  57
  58 static int debug = -1;
  59 module_param(debug, int, 0444);
  60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  61
  62 static LIST_HEAD(netvsc_dev_list);
  63
  64 static void netvsc_change_rx_flags(struct net_device *net, int change)
  65 {
  66         struct net_device_context *ndev_ctx = netdev_priv(net);
  67         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
  68         int inc;
  69
  70         if (!vf_netdev)
  71                 return;
  72
  73         if (change & IFF_PROMISC) {
  74                 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
  75                 dev_set_promiscuity(vf_netdev, inc);
  76         }
  77
  78         if (change & IFF_ALLMULTI) {
  79                 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
  80                 dev_set_allmulti(vf_netdev, inc);
  81         }
  82 }
  83
  84 static void netvsc_set_rx_mode(struct net_device *net)
  85 {
  86         struct net_device_context *ndev_ctx = netdev_priv(net);
  87         struct net_device *vf_netdev;
  88         struct netvsc_device *nvdev;
  89
  90         rcu_read_lock();
  91         vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
  92         if (vf_netdev) {
  93                 dev_uc_sync(vf_netdev, net);
  94                 dev_mc_sync(vf_netdev, net);
  95         }
  96
  97         nvdev = rcu_dereference(ndev_ctx->nvdev);
  98         if (nvdev)
  99                 rndis_filter_update(nvdev);
 100         rcu_read_unlock();
 101 }
 102
 103 static void netvsc_tx_enable(struct netvsc_device *nvscdev,
 104                              struct net_device *ndev)
 105 {
 106         nvscdev->tx_disable = false;
 107         virt_wmb(); /* ensure queue wake up mechanism is on */
 108
 109         netif_tx_wake_all_queues(ndev);
 110 }
 111
 112 static int netvsc_open(struct net_device *net)
 113 {
 114         struct net_device_context *ndev_ctx = netdev_priv(net);
 115         struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
 116         struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
 117         struct rndis_device *rdev;
 118         int ret = 0;
 119
 120         netif_carrier_off(net);
 121
 122         /* Open up the device */
 123         ret = rndis_filter_open(nvdev);
 124         if (ret != 0) {
 125                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 126                 return ret;
 127         }
 128
 129         rdev = nvdev->extension;
 130         if (!rdev->link_state) {
 131                 netif_carrier_on(net);
 132                 netvsc_tx_enable(nvdev, net);
 133         }
 134
 135         if (vf_netdev) {
 136                 /* Setting synthetic device up transparently sets
 137                  * slave as up. If open fails, then slave will be
 138                  * still be offline (and not used).
 139                  */
 140                 ret = dev_open(vf_netdev, NULL);
 141                 if (ret)
 142                         netdev_warn(net,
 143                                     "unable to open slave: %s: %d\n",
 144                                     vf_netdev->name, ret);
 145         }
 146         return 0;
 147 }
 148
 149 static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
 150 {
 151         unsigned int retry = 0;
 152         int i;
 153
 154         /* Ensure pending bytes in ring are read */
 155         for (;;) {
 156                 u32 aread = 0;
 157
 158                 for (i = 0; i < nvdev->num_chn; i++) {
 159                         struct vmbus_channel *chn
 160                                 = nvdev->chan_table[i].channel;
 161
 162                         if (!chn)
 163                                 continue;
 164
 165                         /* make sure receive not running now */
 166                         napi_synchronize(&nvdev->chan_table[i].napi);
 167
 168                         aread = hv_get_bytes_to_read(&chn->inbound);
 169                         if (aread)
 170                                 break;
 171
 172                         aread = hv_get_bytes_to_read(&chn->outbound);
 173                         if (aread)
 174                                 break;
 175                 }
 176
 177                 if (aread == 0)
 178                         return 0;
 179
 180                 if (++retry > RETRY_MAX)
 181                         return -ETIMEDOUT;
 182
 183                 usleep_range(RETRY_US_LO, RETRY_US_HI);
 184         }
 185 }
 186
 187 static void netvsc_tx_disable(struct netvsc_device *nvscdev,
 188                               struct net_device *ndev)
 189 {
 190         if (nvscdev) {
 191                 nvscdev->tx_disable = true;
 192                 virt_wmb(); /* ensure txq will not wake up after stop */
 193         }
 194
 195         netif_tx_disable(ndev);
 196 }
 197
 198 static int netvsc_close(struct net_device *net)
 199 {
 200         struct net_device_context *net_device_ctx = netdev_priv(net);
 201         struct net_device *vf_netdev
 202                 = rtnl_dereference(net_device_ctx->vf_netdev);
 203         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 204         int ret;
 205
 206         netvsc_tx_disable(nvdev, net);
 207
 208         /* No need to close rndis filter if it is removed already */
 209         if (!nvdev)
 210                 return 0;
 211
 212         ret = rndis_filter_close(nvdev);
 213         if (ret != 0) {
 214                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 215                 return ret;
 216         }
 217
 218         ret = netvsc_wait_until_empty(nvdev);
 219         if (ret)
 220                 netdev_err(net, "Ring buffer not empty after closing rndis\n");
 221
 222         if (vf_netdev)
 223                 dev_close(vf_netdev);
 224
 225         return ret;
 226 }
 227
 228 static inline void *init_ppi_data(struct rndis_message *msg,
 229                                   u32 ppi_size, u32 pkt_type)
 230 {
 231         struct rndis_packet *rndis_pkt = &msg->msg.pkt;
 232         struct rndis_per_packet_info *ppi;
 233
 234         rndis_pkt->data_offset += ppi_size;
 235         ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
 236                 + rndis_pkt->per_pkt_info_len;
 237
 238         ppi->size = ppi_size;
 239         ppi->type = pkt_type;
 240         ppi->internal = 0;
 241         ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
 242
 243         rndis_pkt->per_pkt_info_len += ppi_size;
 244
 245         return ppi + 1;
 246 }
 247
 248 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented
 249  * packets. We can use ethtool to change UDP hash level when necessary.
 250  */
 251 static inline u32 netvsc_get_hash(
 252         struct sk_buff *skb,
 253         const struct net_device_context *ndc)
 254 {
 255         struct flow_keys flow;
 256         u32 hash, pkt_proto = 0;
 257         static u32 hashrnd __read_mostly;
 258
 259         net_get_random_once(&hashrnd, sizeof(hashrnd));
 260
 261         if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
 262                 return 0;
 263
 264         switch (flow.basic.ip_proto) {
 265         case IPPROTO_TCP:
 266                 if (flow.basic.n_proto == htons(ETH_P_IP))
 267                         pkt_proto = HV_TCP4_L4HASH;
 268                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 269                         pkt_proto = HV_TCP6_L4HASH;
 270
 271                 break;
 272
 273         case IPPROTO_UDP:
 274                 if (flow.basic.n_proto == htons(ETH_P_IP))
 275                         pkt_proto = HV_UDP4_L4HASH;
 276                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 277                         pkt_proto = HV_UDP6_L4HASH;
 278
 279                 break;
 280         }
 281
 282         if (pkt_proto & ndc->l4_hash) {
 283                 return skb_get_hash(skb);
 284         } else {
 285                 if (flow.basic.n_proto == htons(ETH_P_IP))
 286                         hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
 287                 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
 288                         hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
 289                 else
 290                         return 0;
 291
 292                 __skb_set_sw_hash(skb, hash, false);
 293         }
 294
 295         return hash;
 296 }
 297
 298 static inline int netvsc_get_tx_queue(struct net_device *ndev,
 299                                       struct sk_buff *skb, int old_idx)
 300 {
 301         const struct net_device_context *ndc = netdev_priv(ndev);
 302         struct sock *sk = skb->sk;
 303         int q_idx;
 304
 305         q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
 306                               (VRSS_SEND_TAB_SIZE - 1)];
 307
 308         /* If queue index changed record the new value */
 309         if (q_idx != old_idx &&
 310             sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
 311                 sk_tx_queue_set(sk, q_idx);
 312
 313         return q_idx;
 314 }
 315
 316 /*
 317  * Select queue for transmit.
 318  *
 319  * If a valid queue has already been assigned, then use that.
 320  * Otherwise compute tx queue based on hash and the send table.
 321  *
 322  * This is basically similar to default (netdev_pick_tx) with the added step
 323  * of using the host send_table when no other queue has been assigned.
 324  *
 325  * TODO support XPS - but get_xps_queue not exported
 326  */
 327 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
 328 {
 329         int q_idx = sk_tx_queue_get(skb->sk);
 330
 331         if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
 332                 /* If forwarding a packet, we use the recorded queue when
 333                  * available for better cache locality.
 334                  */
 335                 if (skb_rx_queue_recorded(skb))
 336                         q_idx = skb_get_rx_queue(skb);
 337                 else
 338                         q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
 339         }
 340
 341         return q_idx;
 342 }
 343
 344 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
 345                                struct net_device *sb_dev)
 346 {
 347         struct net_device_context *ndc = netdev_priv(ndev);
 348         struct net_device *vf_netdev;
 349         u16 txq;
 350
 351         rcu_read_lock();
 352         vf_netdev = rcu_dereference(ndc->vf_netdev);
 353         if (vf_netdev) {
 354                 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
 355
 356                 if (vf_ops->ndo_select_queue)
 357                         txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
 358                 else
 359                         txq = netdev_pick_tx(vf_netdev, skb, NULL);
 360
 361                 /* Record the queue selected by VF so that it can be
 362                  * used for common case where VF has more queues than
 363                  * the synthetic device.
 364                  */
 365                 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
 366         } else {
 367                 txq = netvsc_pick_tx(ndev, skb);
 368         }
 369         rcu_read_unlock();
 370
 371         while (txq >= ndev->real_num_tx_queues)
 372                 txq -= ndev->real_num_tx_queues;
 373
 374         return txq;
 375 }
 376
 377 static u32 fill_pg_buf(unsigned long hvpfn, u32 offset, u32 len,
 378                        struct hv_page_buffer *pb)
 379 {
 380         int j = 0;
 381
 382         hvpfn += offset >> HV_HYP_PAGE_SHIFT;
 383         offset = offset & ~HV_HYP_PAGE_MASK;
 384
 385         while (len > 0) {
 386                 unsigned long bytes;
 387
 388                 bytes = HV_HYP_PAGE_SIZE - offset;
 389                 if (bytes > len)
 390                         bytes = len;
 391                 pb[j].pfn = hvpfn;
 392                 pb[j].offset = offset;
 393                 pb[j].len = bytes;
 394
 395                 offset += bytes;
 396                 len -= bytes;
 397
 398                 if (offset == HV_HYP_PAGE_SIZE && len) {
 399                         hvpfn++;
 400                         offset = 0;
 401                         j++;
 402                 }
 403         }
 404
 405         return j + 1;
 406 }
 407
 408 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
 409                            struct hv_netvsc_packet *packet,
 410                            struct hv_page_buffer *pb)
 411 {
 412         u32 slots_used = 0;
 413         char *data = skb->data;
 414         int frags = skb_shinfo(skb)->nr_frags;
 415         int i;
 416
 417         /* The packet is laid out thus:
 418          * 1. hdr: RNDIS header and PPI
 419          * 2. skb linear data
 420          * 3. skb fragment data
 421          */
 422         slots_used += fill_pg_buf(virt_to_hvpfn(hdr),
 423                                   offset_in_hvpage(hdr),
 424                                   len,
 425                                   &pb[slots_used]);
 426
 427         packet->rmsg_size = len;
 428         packet->rmsg_pgcnt = slots_used;
 429
 430         slots_used += fill_pg_buf(virt_to_hvpfn(data),
 431                                   offset_in_hvpage(data),
 432                                   skb_headlen(skb),
 433                                   &pb[slots_used]);
 434
 435         for (i = 0; i < frags; i++) {
 436                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 437
 438                 slots_used += fill_pg_buf(page_to_hvpfn(skb_frag_page(frag)),
 439                                           skb_frag_off(frag),
 440                                           skb_frag_size(frag),
 441                                           &pb[slots_used]);
 442         }
 443         return slots_used;
 444 }
 445
 446 static int count_skb_frag_slots(struct sk_buff *skb)
 447 {
 448         int i, frags = skb_shinfo(skb)->nr_frags;
 449         int pages = 0;
 450
 451         for (i = 0; i < frags; i++) {
 452                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 453                 unsigned long size = skb_frag_size(frag);
 454                 unsigned long offset = skb_frag_off(frag);
 455
 456                 /* Skip unused frames from start of page */
 457                 offset &= ~HV_HYP_PAGE_MASK;
 458                 pages += HVPFN_UP(offset + size);
 459         }
 460         return pages;
 461 }
 462
 463 static int netvsc_get_slots(struct sk_buff *skb)
 464 {
 465         char *data = skb->data;
 466         unsigned int offset = offset_in_hvpage(data);
 467         unsigned int len = skb_headlen(skb);
 468         int slots;
 469         int frag_slots;
 470
 471         slots = DIV_ROUND_UP(offset + len, HV_HYP_PAGE_SIZE);
 472         frag_slots = count_skb_frag_slots(skb);
 473         return slots + frag_slots;
 474 }
 475
 476 static u32 net_checksum_info(struct sk_buff *skb)
 477 {
 478         if (skb->protocol == htons(ETH_P_IP)) {
 479                 struct iphdr *ip = ip_hdr(skb);
 480
 481                 if (ip->protocol == IPPROTO_TCP)
 482                         return TRANSPORT_INFO_IPV4_TCP;
 483                 else if (ip->protocol == IPPROTO_UDP)
 484                         return TRANSPORT_INFO_IPV4_UDP;
 485         } else {
 486                 struct ipv6hdr *ip6 = ipv6_hdr(skb);
 487
 488                 if (ip6->nexthdr == IPPROTO_TCP)
 489                         return TRANSPORT_INFO_IPV6_TCP;
 490                 else if (ip6->nexthdr == IPPROTO_UDP)
 491                         return TRANSPORT_INFO_IPV6_UDP;
 492         }
 493
 494         return TRANSPORT_INFO_NOT_IP;
 495 }
 496
 497 /* Send skb on the slave VF device. */
 498 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
 499                           struct sk_buff *skb)
 500 {
 501         struct net_device_context *ndev_ctx = netdev_priv(net);
 502         unsigned int len = skb->len;
 503         int rc;
 504
 505         skb->dev = vf_netdev;
 506         skb_record_rx_queue(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
 507
 508         rc = dev_queue_xmit(skb);
 509         if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
 510                 struct netvsc_vf_pcpu_stats *pcpu_stats
 511                         = this_cpu_ptr(ndev_ctx->vf_stats);
 512
 513                 u64_stats_update_begin(&pcpu_stats->syncp);
 514                 pcpu_stats->tx_packets++;
 515                 pcpu_stats->tx_bytes += len;
 516                 u64_stats_update_end(&pcpu_stats->syncp);
 517         } else {
 518                 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
 519         }
 520
 521         return rc;
 522 }
 523
 524 static int netvsc_xmit(struct sk_buff *skb, struct net_device *net, bool xdp_tx)
 525 {
 526         struct net_device_context *net_device_ctx = netdev_priv(net);
 527         struct hv_netvsc_packet *packet = NULL;
 528         int ret;
 529         unsigned int num_data_pgs;
 530         struct rndis_message *rndis_msg;
 531         struct net_device *vf_netdev;
 532         u32 rndis_msg_size;
 533         u32 hash;
 534         struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
 535
 536         /* If VF is present and up then redirect packets to it.
 537          * Skip the VF if it is marked down or has no carrier.
 538          * If netpoll is in uses, then VF can not be used either.
 539          */
 540         vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
 541         if (vf_netdev && netif_running(vf_netdev) &&
 542             netif_carrier_ok(vf_netdev) && !netpoll_tx_running(net) &&
 543             net_device_ctx->data_path_is_vf)
 544                 return netvsc_vf_xmit(net, vf_netdev, skb);
 545
 546         /* We will atmost need two pages to describe the rndis
 547          * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
 548          * of pages in a single packet. If skb is scattered around
 549          * more pages we try linearizing it.
 550          */
 551
 552         num_data_pgs = netvsc_get_slots(skb) + 2;
 553
 554         if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
 555                 ++net_device_ctx->eth_stats.tx_scattered;
 556
 557                 if (skb_linearize(skb))
 558                         goto no_memory;
 559
 560                 num_data_pgs = netvsc_get_slots(skb) + 2;
 561                 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
 562                         ++net_device_ctx->eth_stats.tx_too_big;
 563                         goto drop;
 564                 }
 565         }
 566
 567         /*
 568          * Place the rndis header in the skb head room and
 569          * the skb->cb will be used for hv_netvsc_packet
 570          * structure.
 571          */
 572         ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
 573         if (ret)
 574                 goto no_memory;
 575
 576         /* Use the skb control buffer for building up the packet */
 577         BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
 578                         sizeof_field(struct sk_buff, cb));
 579         packet = (struct hv_netvsc_packet *)skb->cb;
 580
 581         packet->q_idx = skb_get_queue_mapping(skb);
 582
 583         packet->total_data_buflen = skb->len;
 584         packet->total_bytes = skb->len;
 585         packet->total_packets = 1;
 586
 587         rndis_msg = (struct rndis_message *)skb->head;
 588
 589         /* Add the rndis header */
 590         rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
 591         rndis_msg->msg_len = packet->total_data_buflen;
 592
 593         rndis_msg->msg.pkt = (struct rndis_packet) {
 594                 .data_offset = sizeof(struct rndis_packet),
 595                 .data_len = packet->total_data_buflen,
 596                 .per_pkt_info_offset = sizeof(struct rndis_packet),
 597         };
 598
 599         rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
 600
 601         hash = skb_get_hash_raw(skb);
 602         if (hash != 0 && net->real_num_tx_queues > 1) {
 603                 u32 *hash_info;
 604
 605                 rndis_msg_size += NDIS_HASH_PPI_SIZE;
 606                 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
 607                                           NBL_HASH_VALUE);
 608                 *hash_info = hash;
 609         }
 610
 611         /* When using AF_PACKET we need to drop VLAN header from
 612          * the frame and update the SKB to allow the HOST OS
 613          * to transmit the 802.1Q packet
 614          */
 615         if (skb->protocol == htons(ETH_P_8021Q)) {
 616                 u16 vlan_tci;
 617
 618                 skb_reset_mac_header(skb);
 619                 if (eth_type_vlan(eth_hdr(skb)->h_proto)) {
 620                         if (unlikely(__skb_vlan_pop(skb, &vlan_tci) != 0)) {
 621                                 ++net_device_ctx->eth_stats.vlan_error;
 622                                 goto drop;
 623                         }
 624
 625                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
 626                         /* Update the NDIS header pkt lengths */
 627                         packet->total_data_buflen -= VLAN_HLEN;
 628                         packet->total_bytes -= VLAN_HLEN;
 629                         rndis_msg->msg_len = packet->total_data_buflen;
 630                         rndis_msg->msg.pkt.data_len = packet->total_data_buflen;
 631                 }
 632         }
 633
 634         if (skb_vlan_tag_present(skb)) {
 635                 struct ndis_pkt_8021q_info *vlan;
 636
 637                 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
 638                 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
 639                                      IEEE_8021Q_INFO);
 640
 641                 vlan->value = 0;
 642                 vlan->vlanid = skb_vlan_tag_get_id(skb);
 643                 vlan->cfi = skb_vlan_tag_get_cfi(skb);
 644                 vlan->pri = skb_vlan_tag_get_prio(skb);
 645         }
 646
 647         if (skb_is_gso(skb)) {
 648                 struct ndis_tcp_lso_info *lso_info;
 649
 650                 rndis_msg_size += NDIS_LSO_PPI_SIZE;
 651                 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
 652                                          TCP_LARGESEND_PKTINFO);
 653
 654                 lso_info->value = 0;
 655                 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
 656                 if (skb->protocol == htons(ETH_P_IP)) {
 657                         lso_info->lso_v2_transmit.ip_version =
 658                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
 659                         ip_hdr(skb)->tot_len = 0;
 660                         ip_hdr(skb)->check = 0;
 661                         tcp_hdr(skb)->check =
 662                                 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 663                                                    ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 664                 } else {
 665                         lso_info->lso_v2_transmit.ip_version =
 666                                 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
 667                         tcp_v6_gso_csum_prep(skb);
 668                 }
 669                 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
 670                 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
 671         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 672                 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
 673                         struct ndis_tcp_ip_checksum_info *csum_info;
 674
 675                         rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 676                         csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 677                                                   TCPIP_CHKSUM_PKTINFO);
 678
 679                         csum_info->value = 0;
 680                         csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
 681
 682                         if (skb->protocol == htons(ETH_P_IP)) {
 683                                 csum_info->transmit.is_ipv4 = 1;
 684
 685                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
 686                                         csum_info->transmit.tcp_checksum = 1;
 687                                 else
 688                                         csum_info->transmit.udp_checksum = 1;
 689                         } else {
 690                                 csum_info->transmit.is_ipv6 = 1;
 691
 692                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
 693                                         csum_info->transmit.tcp_checksum = 1;
 694                                 else
 695                                         csum_info->transmit.udp_checksum = 1;
 696                         }
 697                 } else {
 698                         /* Can't do offload of this type of checksum */
 699                         if (skb_checksum_help(skb))
 700                                 goto drop;
 701                 }
 702         }
 703
 704         /* Start filling in the page buffers with the rndis hdr */
 705         rndis_msg->msg_len += rndis_msg_size;
 706         packet->total_data_buflen = rndis_msg->msg_len;
 707         packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
 708                                                skb, packet, pb);
 709
 710         /* timestamp packet in software */
 711         skb_tx_timestamp(skb);
 712
 713         ret = netvsc_send(net, packet, rndis_msg, pb, skb, xdp_tx);
 714         if (likely(ret == 0))
 715                 return NETDEV_TX_OK;
 716
 717         if (ret == -EAGAIN) {
 718                 ++net_device_ctx->eth_stats.tx_busy;
 719                 return NETDEV_TX_BUSY;
 720         }
 721
 722         if (ret == -ENOSPC)
 723                 ++net_device_ctx->eth_stats.tx_no_space;
 724
 725 drop:
 726         dev_kfree_skb_any(skb);
 727         net->stats.tx_dropped++;
 728
 729         return NETDEV_TX_OK;
 730
 731 no_memory:
 732         ++net_device_ctx->eth_stats.tx_no_memory;
 733         goto drop;
 734 }
 735
 736 static netdev_tx_t netvsc_start_xmit(struct sk_buff *skb,
 737                                      struct net_device *ndev)
 738 {
 739         return netvsc_xmit(skb, ndev, false);
 740 }
 741
 742 /*
 743  * netvsc_linkstatus_callback - Link up/down notification
 744  */
 745 void netvsc_linkstatus_callback(struct net_device *net,
 746                                 struct rndis_message *resp)
 747 {
 748         struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
 749         struct net_device_context *ndev_ctx = netdev_priv(net);
 750         struct netvsc_reconfig *event;
 751         unsigned long flags;
 752
 753         /* Ensure the packet is big enough to access its fields */
 754         if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_indicate_status)) {
 755                 netdev_err(net, "invalid rndis_indicate_status packet, len: %u\n",
 756                            resp->msg_len);
 757                 return;
 758         }
 759
 760         /* Update the physical link speed when changing to another vSwitch */
 761         if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
 762                 u32 speed;
 763
 764                 /* Validate status_buf_offset */
 765                 if (indicate->status_buflen < sizeof(speed) ||
 766                     indicate->status_buf_offset < sizeof(*indicate) ||
 767                     resp->msg_len - RNDIS_HEADER_SIZE < indicate->status_buf_offset ||
 768                     resp->msg_len - RNDIS_HEADER_SIZE - indicate->status_buf_offset
 769                                 < indicate->status_buflen) {
 770                         netdev_err(net, "invalid rndis_indicate_status packet\n");
 771                         return;
 772                 }
 773
 774                 speed = *(u32 *)((void *)indicate
 775                                  + indicate->status_buf_offset) / 10000;
 776                 ndev_ctx->speed = speed;
 777                 return;
 778         }
 779
 780         /* Handle these link change statuses below */
 781         if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
 782             indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
 783             indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
 784                 return;
 785
 786         if (net->reg_state != NETREG_REGISTERED)
 787                 return;
 788
 789         event = kzalloc(sizeof(*event), GFP_ATOMIC);
 790         if (!event)
 791                 return;
 792         event->event = indicate->status;
 793
 794         spin_lock_irqsave(&ndev_ctx->lock, flags);
 795         list_add_tail(&event->list, &ndev_ctx->reconfig_events);
 796         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
 797
 798         schedule_delayed_work(&ndev_ctx->dwork, 0);
 799 }
 800
 801 static void netvsc_xdp_xmit(struct sk_buff *skb, struct net_device *ndev)
 802 {
 803         int rc;
 804
 805         skb->queue_mapping = skb_get_rx_queue(skb);
 806         __skb_push(skb, ETH_HLEN);
 807
 808         rc = netvsc_xmit(skb, ndev, true);
 809
 810         if (dev_xmit_complete(rc))
 811                 return;
 812
 813         dev_kfree_skb_any(skb);
 814         ndev->stats.tx_dropped++;
 815 }
 816
 817 static void netvsc_comp_ipcsum(struct sk_buff *skb)
 818 {
 819         struct iphdr *iph = (struct iphdr *)skb->data;
 820
 821         iph->check = 0;
 822         iph->check = ip_fast_csum(iph, iph->ihl);
 823 }
 824
 825 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
 826                                              struct netvsc_channel *nvchan,
 827                                              struct xdp_buff *xdp)
 828 {
 829         struct napi_struct *napi = &nvchan->napi;
 830         const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
 831         const struct ndis_tcp_ip_checksum_info *csum_info =
 832                                                 nvchan->rsc.csum_info;
 833         const u32 *hash_info = nvchan->rsc.hash_info;
 834         struct sk_buff *skb;
 835         void *xbuf = xdp->data_hard_start;
 836         int i;
 837
 838         if (xbuf) {
 839                 unsigned int hdroom = xdp->data - xdp->data_hard_start;
 840                 unsigned int xlen = xdp->data_end - xdp->data;
 841                 unsigned int frag_size = xdp->frame_sz;
 842
 843                 skb = build_skb(xbuf, frag_size);
 844
 845                 if (!skb) {
 846                         __free_page(virt_to_page(xbuf));
 847                         return NULL;
 848                 }
 849
 850                 skb_reserve(skb, hdroom);
 851                 skb_put(skb, xlen);
 852                 skb->dev = napi->dev;
 853         } else {
 854                 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
 855
 856                 if (!skb)
 857                         return NULL;
 858
 859                 /* Copy to skb. This copy is needed here since the memory
 860                  * pointed by hv_netvsc_packet cannot be deallocated.
 861                  */
 862                 for (i = 0; i < nvchan->rsc.cnt; i++)
 863                         skb_put_data(skb, nvchan->rsc.data[i],
 864                                      nvchan->rsc.len[i]);
 865         }
 866
 867         skb->protocol = eth_type_trans(skb, net);
 868
 869         /* skb is already created with CHECKSUM_NONE */
 870         skb_checksum_none_assert(skb);
 871
 872         /* Incoming packets may have IP header checksum verified by the host.
 873          * They may not have IP header checksum computed after coalescing.
 874          * We compute it here if the flags are set, because on Linux, the IP
 875          * checksum is always checked.
 876          */
 877         if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
 878             csum_info->receive.ip_checksum_succeeded &&
 879             skb->protocol == htons(ETH_P_IP)) {
 880                 /* Check that there is enough space to hold the IP header. */
 881                 if (skb_headlen(skb) < sizeof(struct iphdr)) {
 882                         kfree_skb(skb);
 883                         return NULL;
 884                 }
 885                 netvsc_comp_ipcsum(skb);
 886         }
 887
 888         /* Do L4 checksum offload if enabled and present. */
 889         if (csum_info && (net->features & NETIF_F_RXCSUM)) {
 890                 if (csum_info->receive.tcp_checksum_succeeded ||
 891                     csum_info->receive.udp_checksum_succeeded)
 892                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 893         }
 894
 895         if (hash_info && (net->features & NETIF_F_RXHASH))
 896                 skb_set_hash(skb, *hash_info, PKT_HASH_TYPE_L4);
 897
 898         if (vlan) {
 899                 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
 900                         (vlan->cfi ? VLAN_CFI_MASK : 0);
 901
 902                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 903                                        vlan_tci);
 904         }
 905
 906         return skb;
 907 }
 908
 909 /*
 910  * netvsc_recv_callback -  Callback when we receive a packet from the
 911  * "wire" on the specified device.
 912  */
 913 int netvsc_recv_callback(struct net_device *net,
 914                          struct netvsc_device *net_device,
 915                          struct netvsc_channel *nvchan)
 916 {
 917         struct net_device_context *net_device_ctx = netdev_priv(net);
 918         struct vmbus_channel *channel = nvchan->channel;
 919         u16 q_idx = channel->offermsg.offer.sub_channel_index;
 920         struct sk_buff *skb;
 921         struct netvsc_stats *rx_stats = &nvchan->rx_stats;
 922         struct xdp_buff xdp;
 923         u32 act;
 924
 925         if (net->reg_state != NETREG_REGISTERED)
 926                 return NVSP_STAT_FAIL;
 927
 928         act = netvsc_run_xdp(net, nvchan, &xdp);
 929
 930         if (act != XDP_PASS && act != XDP_TX) {
 931                 u64_stats_update_begin(&rx_stats->syncp);
 932                 rx_stats->xdp_drop++;
 933                 u64_stats_update_end(&rx_stats->syncp);
 934
 935                 return NVSP_STAT_SUCCESS; /* consumed by XDP */
 936         }
 937
 938         /* Allocate a skb - TODO direct I/O to pages? */
 939         skb = netvsc_alloc_recv_skb(net, nvchan, &xdp);
 940
 941         if (unlikely(!skb)) {
 942                 ++net_device_ctx->eth_stats.rx_no_memory;
 943                 return NVSP_STAT_FAIL;
 944         }
 945
 946         skb_record_rx_queue(skb, q_idx);
 947
 948         /*
 949          * Even if injecting the packet, record the statistics
 950          * on the synthetic device because modifying the VF device
 951          * statistics will not work correctly.
 952          */
 953         u64_stats_update_begin(&rx_stats->syncp);
 954         rx_stats->packets++;
 955         rx_stats->bytes += nvchan->rsc.pktlen;
 956
 957         if (skb->pkt_type == PACKET_BROADCAST)
 958                 ++rx_stats->broadcast;
 959         else if (skb->pkt_type == PACKET_MULTICAST)
 960                 ++rx_stats->multicast;
 961         u64_stats_update_end(&rx_stats->syncp);
 962
 963         if (act == XDP_TX) {
 964                 netvsc_xdp_xmit(skb, net);
 965                 return NVSP_STAT_SUCCESS;
 966         }
 967
 968         napi_gro_receive(&nvchan->napi, skb);
 969         return NVSP_STAT_SUCCESS;
 970 }
 971
 972 static void netvsc_get_drvinfo(struct net_device *net,
 973                                struct ethtool_drvinfo *info)
 974 {
 975         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 976         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 977 }
 978
 979 static void netvsc_get_channels(struct net_device *net,
 980                                 struct ethtool_channels *channel)
 981 {
 982         struct net_device_context *net_device_ctx = netdev_priv(net);
 983         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
 984
 985         if (nvdev) {
 986                 channel->max_combined   = nvdev->max_chn;
 987                 channel->combined_count = nvdev->num_chn;
 988         }
 989 }
 990
 991 /* Alloc struct netvsc_device_info, and initialize it from either existing
 992  * struct netvsc_device, or from default values.
 993  */
 994 static
 995 struct netvsc_device_info *netvsc_devinfo_get(struct netvsc_device *nvdev)
 996 {
 997         struct netvsc_device_info *dev_info;
 998         struct bpf_prog *prog;
 999
1000         dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
1001
1002         if (!dev_info)
1003                 return NULL;
1004
1005         if (nvdev) {
1006                 ASSERT_RTNL();
1007
1008                 dev_info->num_chn = nvdev->num_chn;
1009                 dev_info->send_sections = nvdev->send_section_cnt;
1010                 dev_info->send_section_size = nvdev->send_section_size;
1011                 dev_info->recv_sections = nvdev->recv_section_cnt;
1012                 dev_info->recv_section_size = nvdev->recv_section_size;
1013
1014                 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
1015                        NETVSC_HASH_KEYLEN);
1016
1017                 prog = netvsc_xdp_get(nvdev);
1018                 if (prog) {
1019                         bpf_prog_inc(prog);
1020                         dev_info->bprog = prog;
1021                 }
1022         } else {
1023                 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
1024                 dev_info->send_sections = NETVSC_DEFAULT_TX;
1025                 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
1026                 dev_info->recv_sections = NETVSC_DEFAULT_RX;
1027                 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
1028         }
1029
1030         return dev_info;
1031 }
1032
1033 /* Free struct netvsc_device_info */
1034 static void netvsc_devinfo_put(struct netvsc_device_info *dev_info)
1035 {
1036         if (dev_info->bprog) {
1037                 ASSERT_RTNL();
1038                 bpf_prog_put(dev_info->bprog);
1039         }
1040
1041         kfree(dev_info);
1042 }
1043
1044 static int netvsc_detach(struct net_device *ndev,
1045                          struct netvsc_device *nvdev)
1046 {
1047         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1048         struct hv_device *hdev = ndev_ctx->device_ctx;
1049         int ret;
1050
1051         /* Don't try continuing to try and setup sub channels */
1052         if (cancel_work_sync(&nvdev->subchan_work))
1053                 nvdev->num_chn = 1;
1054
1055         netvsc_xdp_set(ndev, NULL, NULL, nvdev);
1056
1057         /* If device was up (receiving) then shutdown */
1058         if (netif_running(ndev)) {
1059                 netvsc_tx_disable(nvdev, ndev);
1060
1061                 ret = rndis_filter_close(nvdev);
1062                 if (ret) {
1063                         netdev_err(ndev,
1064                                    "unable to close device (ret %d).\n", ret);
1065                         return ret;
1066                 }
1067
1068                 ret = netvsc_wait_until_empty(nvdev);
1069                 if (ret) {
1070                         netdev_err(ndev,
1071                                    "Ring buffer not empty after closing rndis\n");
1072                         return ret;
1073                 }
1074         }
1075
1076         netif_device_detach(ndev);
1077
1078         rndis_filter_device_remove(hdev, nvdev);
1079
1080         return 0;
1081 }
1082
1083 static int netvsc_attach(struct net_device *ndev,
1084                          struct netvsc_device_info *dev_info)
1085 {
1086         struct net_device_context *ndev_ctx = netdev_priv(ndev);
1087         struct hv_device *hdev = ndev_ctx->device_ctx;
1088         struct netvsc_device *nvdev;
1089         struct rndis_device *rdev;
1090         struct bpf_prog *prog;
1091         int ret = 0;
1092
1093         nvdev = rndis_filter_device_add(hdev, dev_info);
1094         if (IS_ERR(nvdev))
1095                 return PTR_ERR(nvdev);
1096
1097         if (nvdev->num_chn > 1) {
1098                 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
1099
1100                 /* if unavailable, just proceed with one queue */
1101                 if (ret) {
1102                         nvdev->max_chn = 1;
1103                         nvdev->num_chn = 1;
1104                 }
1105         }
1106
1107         prog = dev_info->bprog;
1108         if (prog) {
1109                 bpf_prog_inc(prog);
1110                 ret = netvsc_xdp_set(ndev, prog, NULL, nvdev);
1111                 if (ret) {
1112                         bpf_prog_put(prog);
1113                         goto err1;
1114                 }
1115         }
1116
1117         /* In any case device is now ready */
1118         nvdev->tx_disable = false;
1119         netif_device_attach(ndev);
1120
1121         /* Note: enable and attach happen when sub-channels setup */
1122         netif_carrier_off(ndev);
1123
1124         if (netif_running(ndev)) {
1125                 ret = rndis_filter_open(nvdev);
1126                 if (ret)
1127                         goto err2;
1128
1129                 rdev = nvdev->extension;
1130                 if (!rdev->link_state)
1131                         netif_carrier_on(ndev);
1132         }
1133
1134         return 0;
1135
1136 err2:
1137         netif_device_detach(ndev);
1138
1139 err1:
1140         rndis_filter_device_remove(hdev, nvdev);
1141
1142         return ret;
1143 }
1144
1145 static int netvsc_set_channels(struct net_device *net,
1146                                struct ethtool_channels *channels)
1147 {
1148         struct net_device_context *net_device_ctx = netdev_priv(net);
1149         struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
1150         unsigned int orig, count = channels->combined_count;
1151         struct netvsc_device_info *device_info;
1152         int ret;
1153
1154         /* We do not support separate count for rx, tx, or other */
1155         if (count == 0 ||
1156             channels->rx_count || channels->tx_count || channels->other_count)
1157                 return -EINVAL;
1158
1159         if (!nvdev || nvdev->destroy)
1160                 return -ENODEV;
1161
1162         if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1163                 return -EINVAL;
1164
1165         if (count > nvdev->max_chn)
1166                 return -EINVAL;
1167
1168         orig = nvdev->num_chn;
1169
1170         device_info = netvsc_devinfo_get(nvdev);
1171
1172         if (!device_info)
1173                 return -ENOMEM;
1174
1175         device_info->num_chn = count;
1176
1177         ret = netvsc_detach(net, nvdev);
1178         if (ret)
1179                 goto out;
1180
1181         ret = netvsc_attach(net, device_info);
1182         if (ret) {
1183                 device_info->num_chn = orig;
1184                 if (netvsc_attach(net, device_info))
1185                         netdev_err(net, "restoring channel setting failed\n");
1186         }
1187
1188 out:
1189         netvsc_devinfo_put(device_info);
1190         return ret;
1191 }
1192
1193 static void netvsc_init_settings(struct net_device *dev)
1194 {
1195         struct net_device_context *ndc = netdev_priv(dev);
1196
1197         ndc->l4_hash = HV_DEFAULT_L4HASH;
1198
1199         ndc->speed = SPEED_UNKNOWN;
1200         ndc->duplex = DUPLEX_FULL;
1201
1202         dev->features = NETIF_F_LRO;
1203 }
1204
1205 static int netvsc_get_link_ksettings(struct net_device *dev,
1206                                      struct ethtool_link_ksettings *cmd)
1207 {
1208         struct net_device_context *ndc = netdev_priv(dev);
1209         struct net_device *vf_netdev;
1210
1211         vf_netdev = rtnl_dereference(ndc->vf_netdev);
1212
1213         if (vf_netdev)
1214                 return __ethtool_get_link_ksettings(vf_netdev, cmd);
1215
1216         cmd->base.speed = ndc->speed;
1217         cmd->base.duplex = ndc->duplex;
1218         cmd->base.port = PORT_OTHER;
1219
1220         return 0;
1221 }
1222
1223 static int netvsc_set_link_ksettings(struct net_device *dev,
1224                                      const struct ethtool_link_ksettings *cmd)
1225 {
1226         struct net_device_context *ndc = netdev_priv(dev);
1227         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1228
1229         if (vf_netdev) {
1230                 if (!vf_netdev->ethtool_ops->set_link_ksettings)
1231                         return -EOPNOTSUPP;
1232
1233                 return vf_netdev->ethtool_ops->set_link_ksettings(vf_netdev,
1234                                                                   cmd);
1235         }
1236
1237         return ethtool_virtdev_set_link_ksettings(dev, cmd,
1238                                                   &ndc->speed, &ndc->duplex);
1239 }
1240
1241 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1242 {
1243         struct net_device_context *ndevctx = netdev_priv(ndev);
1244         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1245         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1246         int orig_mtu = ndev->mtu;
1247         struct netvsc_device_info *device_info;
1248         int ret = 0;
1249
1250         if (!nvdev || nvdev->destroy)
1251                 return -ENODEV;
1252
1253         device_info = netvsc_devinfo_get(nvdev);
1254
1255         if (!device_info)
1256                 return -ENOMEM;
1257
1258         /* Change MTU of underlying VF netdev first. */
1259         if (vf_netdev) {
1260                 ret = dev_set_mtu(vf_netdev, mtu);
1261                 if (ret)
1262                         goto out;
1263         }
1264
1265         ret = netvsc_detach(ndev, nvdev);
1266         if (ret)
1267                 goto rollback_vf;
1268
1269         ndev->mtu = mtu;
1270
1271         ret = netvsc_attach(ndev, device_info);
1272         if (!ret)
1273                 goto out;
1274
1275         /* Attempt rollback to original MTU */
1276         ndev->mtu = orig_mtu;
1277
1278         if (netvsc_attach(ndev, device_info))
1279                 netdev_err(ndev, "restoring mtu failed\n");
1280 rollback_vf:
1281         if (vf_netdev)
1282                 dev_set_mtu(vf_netdev, orig_mtu);
1283
1284 out:
1285         netvsc_devinfo_put(device_info);
1286         return ret;
1287 }
1288
1289 static void netvsc_get_vf_stats(struct net_device *net,
1290                                 struct netvsc_vf_pcpu_stats *tot)
1291 {
1292         struct net_device_context *ndev_ctx = netdev_priv(net);
1293         int i;
1294
1295         memset(tot, 0, sizeof(*tot));
1296
1297         for_each_possible_cpu(i) {
1298                 const struct netvsc_vf_pcpu_stats *stats
1299                         = per_cpu_ptr(ndev_ctx->vf_stats, i);
1300                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1301                 unsigned int start;
1302
1303                 do {
1304                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1305                         rx_packets = stats->rx_packets;
1306                         tx_packets = stats->tx_packets;
1307                         rx_bytes = stats->rx_bytes;
1308                         tx_bytes = stats->tx_bytes;
1309                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1310
1311                 tot->rx_packets += rx_packets;
1312                 tot->tx_packets += tx_packets;
1313                 tot->rx_bytes   += rx_bytes;
1314                 tot->tx_bytes   += tx_bytes;
1315                 tot->tx_dropped += stats->tx_dropped;
1316         }
1317 }
1318
1319 static void netvsc_get_pcpu_stats(struct net_device *net,
1320                                   struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1321 {
1322         struct net_device_context *ndev_ctx = netdev_priv(net);
1323         struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1324         int i;
1325
1326         /* fetch percpu stats of vf */
1327         for_each_possible_cpu(i) {
1328                 const struct netvsc_vf_pcpu_stats *stats =
1329                         per_cpu_ptr(ndev_ctx->vf_stats, i);
1330                 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1331                 unsigned int start;
1332
1333                 do {
1334                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1335                         this_tot->vf_rx_packets = stats->rx_packets;
1336                         this_tot->vf_tx_packets = stats->tx_packets;
1337                         this_tot->vf_rx_bytes = stats->rx_bytes;
1338                         this_tot->vf_tx_bytes = stats->tx_bytes;
1339                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1340                 this_tot->rx_packets = this_tot->vf_rx_packets;
1341                 this_tot->tx_packets = this_tot->vf_tx_packets;
1342                 this_tot->rx_bytes   = this_tot->vf_rx_bytes;
1343                 this_tot->tx_bytes   = this_tot->vf_tx_bytes;
1344         }
1345
1346         /* fetch percpu stats of netvsc */
1347         for (i = 0; i < nvdev->num_chn; i++) {
1348                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1349                 const struct netvsc_stats *stats;
1350                 struct netvsc_ethtool_pcpu_stats *this_tot =
1351                         &pcpu_tot[nvchan->channel->target_cpu];
1352                 u64 packets, bytes;
1353                 unsigned int start;
1354
1355                 stats = &nvchan->tx_stats;
1356                 do {
1357                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1358                         packets = stats->packets;
1359                         bytes = stats->bytes;
1360                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1361
1362                 this_tot->tx_bytes      += bytes;
1363                 this_tot->tx_packets    += packets;
1364
1365                 stats = &nvchan->rx_stats;
1366                 do {
1367                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1368                         packets = stats->packets;
1369                         bytes = stats->bytes;
1370                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1371
1372                 this_tot->rx_bytes      += bytes;
1373                 this_tot->rx_packets    += packets;
1374         }
1375 }
1376
1377 static void netvsc_get_stats64(struct net_device *net,
1378                                struct rtnl_link_stats64 *t)
1379 {
1380         struct net_device_context *ndev_ctx = netdev_priv(net);
1381         struct netvsc_device *nvdev;
1382         struct netvsc_vf_pcpu_stats vf_tot;
1383         int i;
1384
1385         rcu_read_lock();
1386
1387         nvdev = rcu_dereference(ndev_ctx->nvdev);
1388         if (!nvdev)
1389                 goto out;
1390
1391         netdev_stats_to_stats64(t, &net->stats);
1392
1393         netvsc_get_vf_stats(net, &vf_tot);
1394         t->rx_packets += vf_tot.rx_packets;
1395         t->tx_packets += vf_tot.tx_packets;
1396         t->rx_bytes   += vf_tot.rx_bytes;
1397         t->tx_bytes   += vf_tot.tx_bytes;
1398         t->tx_dropped += vf_tot.tx_dropped;
1399
1400         for (i = 0; i < nvdev->num_chn; i++) {
1401                 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1402                 const struct netvsc_stats *stats;
1403                 u64 packets, bytes, multicast;
1404                 unsigned int start;
1405
1406                 stats = &nvchan->tx_stats;
1407                 do {
1408                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1409                         packets = stats->packets;
1410                         bytes = stats->bytes;
1411                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1412
1413                 t->tx_bytes     += bytes;
1414                 t->tx_packets   += packets;
1415
1416                 stats = &nvchan->rx_stats;
1417                 do {
1418                         start = u64_stats_fetch_begin_irq(&stats->syncp);
1419                         packets = stats->packets;
1420                         bytes = stats->bytes;
1421                         multicast = stats->multicast + stats->broadcast;
1422                 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1423
1424                 t->rx_bytes     += bytes;
1425                 t->rx_packets   += packets;
1426                 t->multicast    += multicast;
1427         }
1428 out:
1429         rcu_read_unlock();
1430 }
1431
1432 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1433 {
1434         struct net_device_context *ndc = netdev_priv(ndev);
1435         struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
1436         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1437         struct sockaddr *addr = p;
1438         int err;
1439
1440         err = eth_prepare_mac_addr_change(ndev, p);
1441         if (err)
1442                 return err;
1443
1444         if (!nvdev)
1445                 return -ENODEV;
1446
1447         if (vf_netdev) {
1448                 err = dev_set_mac_address(vf_netdev, addr, NULL);
1449                 if (err)
1450                         return err;
1451         }
1452
1453         err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
1454         if (!err) {
1455                 eth_commit_mac_addr_change(ndev, p);
1456         } else if (vf_netdev) {
1457                 /* rollback change on VF */
1458                 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
1459                 dev_set_mac_address(vf_netdev, addr, NULL);
1460         }
1461
1462         return err;
1463 }
1464
1465 static const struct {
1466         char name[ETH_GSTRING_LEN];
1467         u16 offset;
1468 } netvsc_stats[] = {
1469         { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
1470         { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
1471         { "tx_no_space",  offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1472         { "tx_too_big",   offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1473         { "tx_busy",      offsetof(struct netvsc_ethtool_stats, tx_busy) },
1474         { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1475         { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
1476         { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
1477         { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1478         { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
1479         { "vlan_error", offsetof(struct netvsc_ethtool_stats, vlan_error) },
1480 }, pcpu_stats[] = {
1481         { "cpu%u_rx_packets",
1482                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1483         { "cpu%u_rx_bytes",
1484                 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1485         { "cpu%u_tx_packets",
1486                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1487         { "cpu%u_tx_bytes",
1488                 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1489         { "cpu%u_vf_rx_packets",
1490                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1491         { "cpu%u_vf_rx_bytes",
1492                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1493         { "cpu%u_vf_tx_packets",
1494                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1495         { "cpu%u_vf_tx_bytes",
1496                 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
1497 }, vf_stats[] = {
1498         { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1499         { "vf_rx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1500         { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1501         { "vf_tx_bytes",   offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1502         { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
1503 };
1504
1505 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
1506 #define NETVSC_VF_STATS_LEN     ARRAY_SIZE(vf_stats)
1507
1508 /* statistics per queue (rx/tx packets/bytes) */
1509 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1510
1511 /* 5 statistics per queue (rx/tx packets/bytes, rx xdp_drop) */
1512 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 5)
1513
1514 static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1515 {
1516         struct net_device_context *ndc = netdev_priv(dev);
1517         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1518
1519         if (!nvdev)
1520                 return -ENODEV;
1521
1522         switch (string_set) {
1523         case ETH_SS_STATS:
1524                 return NETVSC_GLOBAL_STATS_LEN
1525                         + NETVSC_VF_STATS_LEN
1526                         + NETVSC_QUEUE_STATS_LEN(nvdev)
1527                         + NETVSC_PCPU_STATS_LEN;
1528         default:
1529                 return -EINVAL;
1530         }
1531 }
1532
1533 static void netvsc_get_ethtool_stats(struct net_device *dev,
1534                                      struct ethtool_stats *stats, u64 *data)
1535 {
1536         struct net_device_context *ndc = netdev_priv(dev);
1537         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1538         const void *nds = &ndc->eth_stats;
1539         const struct netvsc_stats *qstats;
1540         struct netvsc_vf_pcpu_stats sum;
1541         struct netvsc_ethtool_pcpu_stats *pcpu_sum;
1542         unsigned int start;
1543         u64 packets, bytes;
1544         u64 xdp_drop;
1545         int i, j, cpu;
1546
1547         if (!nvdev)
1548                 return;
1549
1550         for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
1551                 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
1552
1553         netvsc_get_vf_stats(dev, &sum);
1554         for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1555                 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1556
1557         for (j = 0; j < nvdev->num_chn; j++) {
1558                 qstats = &nvdev->chan_table[j].tx_stats;
1559
1560                 do {
1561                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1562                         packets = qstats->packets;
1563                         bytes = qstats->bytes;
1564                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1565                 data[i++] = packets;
1566                 data[i++] = bytes;
1567
1568                 qstats = &nvdev->chan_table[j].rx_stats;
1569                 do {
1570                         start = u64_stats_fetch_begin_irq(&qstats->syncp);
1571                         packets = qstats->packets;
1572                         bytes = qstats->bytes;
1573                         xdp_drop = qstats->xdp_drop;
1574                 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1575                 data[i++] = packets;
1576                 data[i++] = bytes;
1577                 data[i++] = xdp_drop;
1578         }
1579
1580         pcpu_sum = kvmalloc_array(num_possible_cpus(),
1581                                   sizeof(struct netvsc_ethtool_pcpu_stats),
1582                                   GFP_KERNEL);
1583         netvsc_get_pcpu_stats(dev, pcpu_sum);
1584         for_each_present_cpu(cpu) {
1585                 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1586
1587                 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1588                         data[i++] = *(u64 *)((void *)this_sum
1589                                              + pcpu_stats[j].offset);
1590         }
1591         kvfree(pcpu_sum);
1592 }
1593
1594 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1595 {
1596         struct net_device_context *ndc = netdev_priv(dev);
1597         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1598         u8 *p = data;
1599         int i, cpu;
1600
1601         if (!nvdev)
1602                 return;
1603
1604         switch (stringset) {
1605         case ETH_SS_STATS:
1606                 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1607                         memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1608                         p += ETH_GSTRING_LEN;
1609                 }
1610
1611                 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1612                         memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1613                         p += ETH_GSTRING_LEN;
1614                 }
1615
1616                 for (i = 0; i < nvdev->num_chn; i++) {
1617                         sprintf(p, "tx_queue_%u_packets", i);
1618                         p += ETH_GSTRING_LEN;
1619                         sprintf(p, "tx_queue_%u_bytes", i);
1620                         p += ETH_GSTRING_LEN;
1621                         sprintf(p, "rx_queue_%u_packets", i);
1622                         p += ETH_GSTRING_LEN;
1623                         sprintf(p, "rx_queue_%u_bytes", i);
1624                         p += ETH_GSTRING_LEN;
1625                         sprintf(p, "rx_queue_%u_xdp_drop", i);
1626                         p += ETH_GSTRING_LEN;
1627                 }
1628
1629                 for_each_present_cpu(cpu) {
1630                         for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1631                                 sprintf(p, pcpu_stats[i].name, cpu);
1632                                 p += ETH_GSTRING_LEN;
1633                         }
1634                 }
1635
1636                 break;
1637         }
1638 }
1639
1640 static int
1641 netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1642                          struct ethtool_rxnfc *info)
1643 {
1644         const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1645
1646         info->data = RXH_IP_SRC | RXH_IP_DST;
1647
1648         switch (info->flow_type) {
1649         case TCP_V4_FLOW:
1650                 if (ndc->l4_hash & HV_TCP4_L4HASH)
1651                         info->data |= l4_flag;
1652
1653                 break;
1654
1655         case TCP_V6_FLOW:
1656                 if (ndc->l4_hash & HV_TCP6_L4HASH)
1657                         info->data |= l4_flag;
1658
1659                 break;
1660
1661         case UDP_V4_FLOW:
1662                 if (ndc->l4_hash & HV_UDP4_L4HASH)
1663                         info->data |= l4_flag;
1664
1665                 break;
1666
1667         case UDP_V6_FLOW:
1668                 if (ndc->l4_hash & HV_UDP6_L4HASH)
1669                         info->data |= l4_flag;
1670
1671                 break;
1672
1673         case IPV4_FLOW:
1674         case IPV6_FLOW:
1675                 break;
1676         default:
1677                 info->data = 0;
1678                 break;
1679         }
1680
1681         return 0;
1682 }
1683
1684 static int
1685 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1686                  u32 *rules)
1687 {
1688         struct net_device_context *ndc = netdev_priv(dev);
1689         struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1690
1691         if (!nvdev)
1692                 return -ENODEV;
1693
1694         switch (info->cmd) {
1695         case ETHTOOL_GRXRINGS:
1696                 info->data = nvdev->num_chn;
1697                 return 0;
1698
1699         case ETHTOOL_GRXFH:
1700                 return netvsc_get_rss_hash_opts(ndc, info);
1701         }
1702         return -EOPNOTSUPP;
1703 }
1704
1705 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1706                                     struct ethtool_rxnfc *info)
1707 {
1708         if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1709                            RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
1710                 switch (info->flow_type) {
1711                 case TCP_V4_FLOW:
1712                         ndc->l4_hash |= HV_TCP4_L4HASH;
1713                         break;
1714
1715                 case TCP_V6_FLOW:
1716                         ndc->l4_hash |= HV_TCP6_L4HASH;
1717                         break;
1718
1719                 case UDP_V4_FLOW:
1720                         ndc->l4_hash |= HV_UDP4_L4HASH;
1721                         break;
1722
1723                 case UDP_V6_FLOW:
1724                         ndc->l4_hash |= HV_UDP6_L4HASH;
1725                         break;
1726
1727                 default:
1728                         return -EOPNOTSUPP;
1729                 }
1730
1731                 return 0;
1732         }
1733
1734         if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
1735                 switch (info->flow_type) {
1736                 case TCP_V4_FLOW:
1737                         ndc->l4_hash &= ~HV_TCP4_L4HASH;
1738                         break;
1739
1740                 case TCP_V6_FLOW:
1741                         ndc->l4_hash &= ~HV_TCP6_L4HASH;
1742                         break;
1743
1744                 case UDP_V4_FLOW:
1745                         ndc->l4_hash &= ~HV_UDP4_L4HASH;
1746                         break;
1747
1748                 case UDP_V6_FLOW:
1749                         ndc->l4_hash &= ~HV_UDP6_L4HASH;
1750                         break;
1751
1752                 default:
1753                         return -EOPNOTSUPP;
1754                 }
1755
1756                 return 0;
1757         }
1758
1759         return -EOPNOTSUPP;
1760 }
1761
1762 static int
1763 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1764 {
1765         struct net_device_context *ndc = netdev_priv(ndev);
1766
1767         if (info->cmd == ETHTOOL_SRXFH)
1768                 return netvsc_set_rss_hash_opts(ndc, info);
1769
1770         return -EOPNOTSUPP;
1771 }
1772
1773 static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1774 {
1775         return NETVSC_HASH_KEYLEN;
1776 }
1777
1778 static u32 netvsc_rss_indir_size(struct net_device *dev)
1779 {
1780         return ITAB_NUM;
1781 }
1782
1783 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1784                            u8 *hfunc)
1785 {
1786         struct net_device_context *ndc = netdev_priv(dev);
1787         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1788         struct rndis_device *rndis_dev;
1789         int i;
1790
1791         if (!ndev)
1792                 return -ENODEV;
1793
1794         if (hfunc)
1795                 *hfunc = ETH_RSS_HASH_TOP;      /* Toeplitz */
1796
1797         rndis_dev = ndev->extension;
1798         if (indir) {
1799                 for (i = 0; i < ITAB_NUM; i++)
1800                         indir[i] = ndc->rx_table[i];
1801         }
1802
1803         if (key)
1804                 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1805
1806         return 0;
1807 }
1808
1809 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1810                            const u8 *key, const u8 hfunc)
1811 {
1812         struct net_device_context *ndc = netdev_priv(dev);
1813         struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
1814         struct rndis_device *rndis_dev;
1815         int i;
1816
1817         if (!ndev)
1818                 return -ENODEV;
1819
1820         if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1821                 return -EOPNOTSUPP;
1822
1823         rndis_dev = ndev->extension;
1824         if (indir) {
1825                 for (i = 0; i < ITAB_NUM; i++)
1826                         if (indir[i] >= ndev->num_chn)
1827                                 return -EINVAL;
1828
1829                 for (i = 0; i < ITAB_NUM; i++)
1830                         ndc->rx_table[i] = indir[i];
1831         }
1832
1833         if (!key) {
1834                 if (!indir)
1835                         return 0;
1836
1837                 key = rndis_dev->rss_key;
1838         }
1839
1840         return rndis_filter_set_rss_param(rndis_dev, key);
1841 }
1842
1843 /* Hyper-V RNDIS protocol does not have ring in the HW sense.
1844  * It does have pre-allocated receive area which is divided into sections.
1845  */
1846 static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1847                                    struct ethtool_ringparam *ring)
1848 {
1849         u32 max_buf_size;
1850
1851         ring->rx_pending = nvdev->recv_section_cnt;
1852         ring->tx_pending = nvdev->send_section_cnt;
1853
1854         if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1855                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1856         else
1857                 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1858
1859         ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1860         ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1861                 / nvdev->send_section_size;
1862 }
1863
1864 static void netvsc_get_ringparam(struct net_device *ndev,
1865                                  struct ethtool_ringparam *ring)
1866 {
1867         struct net_device_context *ndevctx = netdev_priv(ndev);
1868         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1869
1870         if (!nvdev)
1871                 return;
1872
1873         __netvsc_get_ringparam(nvdev, ring);
1874 }
1875
1876 static int netvsc_set_ringparam(struct net_device *ndev,
1877                                 struct ethtool_ringparam *ring)
1878 {
1879         struct net_device_context *ndevctx = netdev_priv(ndev);
1880         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1881         struct netvsc_device_info *device_info;
1882         struct ethtool_ringparam orig;
1883         u32 new_tx, new_rx;
1884         int ret = 0;
1885
1886         if (!nvdev || nvdev->destroy)
1887                 return -ENODEV;
1888
1889         memset(&orig, 0, sizeof(orig));
1890         __netvsc_get_ringparam(nvdev, &orig);
1891
1892         new_tx = clamp_t(u32, ring->tx_pending,
1893                          NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1894         new_rx = clamp_t(u32, ring->rx_pending,
1895                          NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1896
1897         if (new_tx == orig.tx_pending &&
1898             new_rx == orig.rx_pending)
1899                 return 0;        /* no change */
1900
1901         device_info = netvsc_devinfo_get(nvdev);
1902
1903         if (!device_info)
1904                 return -ENOMEM;
1905
1906         device_info->send_sections = new_tx;
1907         device_info->recv_sections = new_rx;
1908
1909         ret = netvsc_detach(ndev, nvdev);
1910         if (ret)
1911                 goto out;
1912
1913         ret = netvsc_attach(ndev, device_info);
1914         if (ret) {
1915                 device_info->send_sections = orig.tx_pending;
1916                 device_info->recv_sections = orig.rx_pending;
1917
1918                 if (netvsc_attach(ndev, device_info))
1919                         netdev_err(ndev, "restoring ringparam failed");
1920         }
1921
1922 out:
1923         netvsc_devinfo_put(device_info);
1924         return ret;
1925 }
1926
1927 static netdev_features_t netvsc_fix_features(struct net_device *ndev,
1928                                              netdev_features_t features)
1929 {
1930         struct net_device_context *ndevctx = netdev_priv(ndev);
1931         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1932
1933         if (!nvdev || nvdev->destroy)
1934                 return features;
1935
1936         if ((features & NETIF_F_LRO) && netvsc_xdp_get(nvdev)) {
1937                 features ^= NETIF_F_LRO;
1938                 netdev_info(ndev, "Skip LRO - unsupported with XDP\n");
1939         }
1940
1941         return features;
1942 }
1943
1944 static int netvsc_set_features(struct net_device *ndev,
1945                                netdev_features_t features)
1946 {
1947         netdev_features_t change = features ^ ndev->features;
1948         struct net_device_context *ndevctx = netdev_priv(ndev);
1949         struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1950         struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
1951         struct ndis_offload_params offloads;
1952         int ret = 0;
1953
1954         if (!nvdev || nvdev->destroy)
1955                 return -ENODEV;
1956
1957         if (!(change & NETIF_F_LRO))
1958                 goto syncvf;
1959
1960         memset(&offloads, 0, sizeof(struct ndis_offload_params));
1961
1962         if (features & NETIF_F_LRO) {
1963                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1964                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1965         } else {
1966                 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1967                 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1968         }
1969
1970         ret = rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1971
1972         if (ret) {
1973                 features ^= NETIF_F_LRO;
1974                 ndev->features = features;
1975         }
1976
1977 syncvf:
1978         if (!vf_netdev)
1979                 return ret;
1980
1981         vf_netdev->wanted_features = features;
1982         netdev_update_features(vf_netdev);
1983
1984         return ret;
1985 }
1986
1987 static int netvsc_get_regs_len(struct net_device *netdev)
1988 {
1989         return VRSS_SEND_TAB_SIZE * sizeof(u32);
1990 }
1991
1992 static void netvsc_get_regs(struct net_device *netdev,
1993                             struct ethtool_regs *regs, void *p)
1994 {
1995         struct net_device_context *ndc = netdev_priv(netdev);
1996         u32 *regs_buff = p;
1997
1998         /* increase the version, if buffer format is changed. */
1999         regs->version = 1;
2000
2001         memcpy(regs_buff, ndc->tx_table, VRSS_SEND_TAB_SIZE * sizeof(u32));
2002 }
2003
2004 static u32 netvsc_get_msglevel(struct net_device *ndev)
2005 {
2006         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2007
2008         return ndev_ctx->msg_enable;
2009 }
2010
2011 static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
2012 {
2013         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2014
2015         ndev_ctx->msg_enable = val;
2016 }
2017
2018 static const struct ethtool_ops ethtool_ops = {
2019         .get_drvinfo    = netvsc_get_drvinfo,
2020         .get_regs_len   = netvsc_get_regs_len,
2021         .get_regs       = netvsc_get_regs,
2022         .get_msglevel   = netvsc_get_msglevel,
2023         .set_msglevel   = netvsc_set_msglevel,
2024         .get_link       = ethtool_op_get_link,
2025         .get_ethtool_stats = netvsc_get_ethtool_stats,
2026         .get_sset_count = netvsc_get_sset_count,
2027         .get_strings    = netvsc_get_strings,
2028         .get_channels   = netvsc_get_channels,
2029         .set_channels   = netvsc_set_channels,
2030         .get_ts_info    = ethtool_op_get_ts_info,
2031         .get_rxnfc      = netvsc_get_rxnfc,
2032         .set_rxnfc      = netvsc_set_rxnfc,
2033         .get_rxfh_key_size = netvsc_get_rxfh_key_size,
2034         .get_rxfh_indir_size = netvsc_rss_indir_size,
2035         .get_rxfh       = netvsc_get_rxfh,
2036         .set_rxfh       = netvsc_set_rxfh,
2037         .get_link_ksettings = netvsc_get_link_ksettings,
2038         .set_link_ksettings = netvsc_set_link_ksettings,
2039         .get_ringparam  = netvsc_get_ringparam,
2040         .set_ringparam  = netvsc_set_ringparam,
2041 };
2042
2043 static const struct net_device_ops device_ops = {
2044         .ndo_open =                     netvsc_open,
2045         .ndo_stop =                     netvsc_close,
2046         .ndo_start_xmit =               netvsc_start_xmit,
2047         .ndo_change_rx_flags =          netvsc_change_rx_flags,
2048         .ndo_set_rx_mode =              netvsc_set_rx_mode,
2049         .ndo_fix_features =             netvsc_fix_features,
2050         .ndo_set_features =             netvsc_set_features,
2051         .ndo_change_mtu =               netvsc_change_mtu,
2052         .ndo_validate_addr =            eth_validate_addr,
2053         .ndo_set_mac_address =          netvsc_set_mac_addr,
2054         .ndo_select_queue =             netvsc_select_queue,
2055         .ndo_get_stats64 =              netvsc_get_stats64,
2056         .ndo_bpf =                      netvsc_bpf,
2057 };
2058
2059 /*
2060  * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
2061  * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
2062  * present send GARP packet to network peers with netif_notify_peers().
2063  */
2064 static void netvsc_link_change(struct work_struct *w)
2065 {
2066         struct net_device_context *ndev_ctx =
2067                 container_of(w, struct net_device_context, dwork.work);
2068         struct hv_device *device_obj = ndev_ctx->device_ctx;
2069         struct net_device *net = hv_get_drvdata(device_obj);
2070         unsigned long flags, next_reconfig, delay;
2071         struct netvsc_reconfig *event = NULL;
2072         struct netvsc_device *net_device;
2073         struct rndis_device *rdev;
2074         bool reschedule = false;
2075
2076         /* if changes are happening, comeback later */
2077         if (!rtnl_trylock()) {
2078                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2079                 return;
2080         }
2081
2082         net_device = rtnl_dereference(ndev_ctx->nvdev);
2083         if (!net_device)
2084                 goto out_unlock;
2085
2086         rdev = net_device->extension;
2087
2088         next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
2089         if (time_is_after_jiffies(next_reconfig)) {
2090                 /* link_watch only sends one notification with current state
2091                  * per second, avoid doing reconfig more frequently. Handle
2092                  * wrap around.
2093                  */
2094                 delay = next_reconfig - jiffies;
2095                 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
2096                 schedule_delayed_work(&ndev_ctx->dwork, delay);
2097                 goto out_unlock;
2098         }
2099         ndev_ctx->last_reconfig = jiffies;
2100
2101         spin_lock_irqsave(&ndev_ctx->lock, flags);
2102         if (!list_empty(&ndev_ctx->reconfig_events)) {
2103                 event = list_first_entry(&ndev_ctx->reconfig_events,
2104                                          struct netvsc_reconfig, list);
2105                 list_del(&event->list);
2106                 reschedule = !list_empty(&ndev_ctx->reconfig_events);
2107         }
2108         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2109
2110         if (!event)
2111                 goto out_unlock;
2112
2113         switch (event->event) {
2114                 /* Only the following events are possible due to the check in
2115                  * netvsc_linkstatus_callback()
2116                  */
2117         case RNDIS_STATUS_MEDIA_CONNECT:
2118                 if (rdev->link_state) {
2119                         rdev->link_state = false;
2120                         netif_carrier_on(net);
2121                         netvsc_tx_enable(net_device, net);
2122                 } else {
2123                         __netdev_notify_peers(net);
2124                 }
2125                 kfree(event);
2126                 break;
2127         case RNDIS_STATUS_MEDIA_DISCONNECT:
2128                 if (!rdev->link_state) {
2129                         rdev->link_state = true;
2130                         netif_carrier_off(net);
2131                         netvsc_tx_disable(net_device, net);
2132                 }
2133                 kfree(event);
2134                 break;
2135         case RNDIS_STATUS_NETWORK_CHANGE:
2136                 /* Only makes sense if carrier is present */
2137                 if (!rdev->link_state) {
2138                         rdev->link_state = true;
2139                         netif_carrier_off(net);
2140                         netvsc_tx_disable(net_device, net);
2141                         event->event = RNDIS_STATUS_MEDIA_CONNECT;
2142                         spin_lock_irqsave(&ndev_ctx->lock, flags);
2143                         list_add(&event->list, &ndev_ctx->reconfig_events);
2144                         spin_unlock_irqrestore(&ndev_ctx->lock, flags);
2145                         reschedule = true;
2146                 }
2147                 break;
2148         }
2149
2150         rtnl_unlock();
2151
2152         /* link_watch only sends one notification with current state per
2153          * second, handle next reconfig event in 2 seconds.
2154          */
2155         if (reschedule)
2156                 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
2157
2158         return;
2159
2160 out_unlock:
2161         rtnl_unlock();
2162 }
2163
2164 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
2165 {
2166         struct net_device_context *net_device_ctx;
2167         struct net_device *dev;
2168
2169         dev = netdev_master_upper_dev_get(vf_netdev);
2170         if (!dev || dev->netdev_ops != &device_ops)
2171                 return NULL;    /* not a netvsc device */
2172
2173         net_device_ctx = netdev_priv(dev);
2174         if (!rtnl_dereference(net_device_ctx->nvdev))
2175                 return NULL;    /* device is removed */
2176
2177         return dev;
2178 }
2179
2180 /* Called when VF is injecting data into network stack.
2181  * Change the associated network device from VF to netvsc.
2182  * note: already called with rcu_read_lock
2183  */
2184 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
2185 {
2186         struct sk_buff *skb = *pskb;
2187         struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
2188         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2189         struct netvsc_vf_pcpu_stats *pcpu_stats
2190                  = this_cpu_ptr(ndev_ctx->vf_stats);
2191
2192         skb = skb_share_check(skb, GFP_ATOMIC);
2193         if (unlikely(!skb))
2194                 return RX_HANDLER_CONSUMED;
2195
2196         *pskb = skb;
2197
2198         skb->dev = ndev;
2199
2200         u64_stats_update_begin(&pcpu_stats->syncp);
2201         pcpu_stats->rx_packets++;
2202         pcpu_stats->rx_bytes += skb->len;
2203         u64_stats_update_end(&pcpu_stats->syncp);
2204
2205         return RX_HANDLER_ANOTHER;
2206 }
2207
2208 static int netvsc_vf_join(struct net_device *vf_netdev,
2209                           struct net_device *ndev)
2210 {
2211         struct net_device_context *ndev_ctx = netdev_priv(ndev);
2212         int ret;
2213
2214         ret = netdev_rx_handler_register(vf_netdev,
2215                                          netvsc_vf_handle_frame, ndev);
2216         if (ret != 0) {
2217                 netdev_err(vf_netdev,
2218                            "can not register netvsc VF receive handler (err = %d)\n",
2219                            ret);
2220                 goto rx_handler_failed;
2221         }
2222
2223         ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2224                                            NULL, NULL, NULL);
2225         if (ret != 0) {
2226                 netdev_err(vf_netdev,
2227                            "can not set master device %s (err = %d)\n",
2228                            ndev->name, ret);
2229                 goto upper_link_failed;
2230         }
2231
2232         /* set slave flag before open to prevent IPv6 addrconf */
2233         vf_netdev->flags |= IFF_SLAVE;
2234
2235         schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2236
2237         call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2238
2239         netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2240         return 0;
2241
2242 upper_link_failed:
2243         netdev_rx_handler_unregister(vf_netdev);
2244 rx_handler_failed:
2245         return ret;
2246 }
2247
2248 static void __netvsc_vf_setup(struct net_device *ndev,
2249                               struct net_device *vf_netdev)
2250 {
2251         int ret;
2252
2253         /* Align MTU of VF with master */
2254         ret = dev_set_mtu(vf_netdev, ndev->mtu);
2255         if (ret)
2256                 netdev_warn(vf_netdev,
2257                             "unable to change mtu to %u\n", ndev->mtu);
2258
2259         /* set multicast etc flags on VF */
2260         dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
2261
2262         /* sync address list from ndev to VF */
2263         netif_addr_lock_bh(ndev);
2264         dev_uc_sync(vf_netdev, ndev);
2265         dev_mc_sync(vf_netdev, ndev);
2266         netif_addr_unlock_bh(ndev);
2267
2268         if (netif_running(ndev)) {
2269                 ret = dev_open(vf_netdev, NULL);
2270                 if (ret)
2271                         netdev_warn(vf_netdev,
2272                                     "unable to open: %d\n", ret);
2273         }
2274 }
2275
2276 /* Setup VF as slave of the synthetic device.
2277  * Runs in workqueue to avoid recursion in netlink callbacks.
2278  */
2279 static void netvsc_vf_setup(struct work_struct *w)
2280 {
2281         struct net_device_context *ndev_ctx
2282                 = container_of(w, struct net_device_context, vf_takeover.work);
2283         struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2284         struct net_device *vf_netdev;
2285
2286         if (!rtnl_trylock()) {
2287                 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
2288                 return;
2289         }
2290
2291         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2292         if (vf_netdev)
2293                 __netvsc_vf_setup(ndev, vf_netdev);
2294
2295         rtnl_unlock();
2296 }
2297
2298 /* Find netvsc by VF serial number.
2299  * The PCI hyperv controller records the serial number as the slot kobj name.
2300  */
2301 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2302 {
2303         struct device *parent = vf_netdev->dev.parent;
2304         struct net_device_context *ndev_ctx;
2305         struct pci_dev *pdev;
2306         u32 serial;
2307
2308         if (!parent || !dev_is_pci(parent))
2309                 return NULL; /* not a PCI device */
2310
2311         pdev = to_pci_dev(parent);
2312         if (!pdev->slot) {
2313                 netdev_notice(vf_netdev, "no PCI slot information\n");
2314                 return NULL;
2315         }
2316
2317         if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2318                 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2319                               pci_slot_name(pdev->slot));
2320                 return NULL;
2321         }
2322
2323         list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2324                 if (!ndev_ctx->vf_alloc)
2325                         continue;
2326
2327                 if (ndev_ctx->vf_serial == serial)
2328                         return hv_get_drvdata(ndev_ctx->device_ctx);
2329         }
2330
2331         netdev_notice(vf_netdev,
2332                       "no netdev found for vf serial:%u\n", serial);
2333         return NULL;
2334 }
2335
2336 static int netvsc_register_vf(struct net_device *vf_netdev)
2337 {
2338         struct net_device_context *net_device_ctx;
2339         struct netvsc_device *netvsc_dev;
2340         struct bpf_prog *prog;
2341         struct net_device *ndev;
2342         int ret;
2343
2344         if (vf_netdev->addr_len != ETH_ALEN)
2345                 return NOTIFY_DONE;
2346
2347         ndev = get_netvsc_byslot(vf_netdev);
2348         if (!ndev)
2349                 return NOTIFY_DONE;
2350
2351         net_device_ctx = netdev_priv(ndev);
2352         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2353         if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
2354                 return NOTIFY_DONE;
2355
2356         /* if synthetic interface is a different namespace,
2357          * then move the VF to that namespace; join will be
2358          * done again in that context.
2359          */
2360         if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2361                 ret = dev_change_net_namespace(vf_netdev,
2362                                                dev_net(ndev), "eth%d");
2363                 if (ret)
2364                         netdev_err(vf_netdev,
2365                                    "could not move to same namespace as %s: %d\n",
2366                                    ndev->name, ret);
2367                 else
2368                         netdev_info(vf_netdev,
2369                                     "VF moved to namespace with: %s\n",
2370                                     ndev->name);
2371                 return NOTIFY_DONE;
2372         }
2373
2374         netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
2375
2376         if (netvsc_vf_join(vf_netdev, ndev) != 0)
2377                 return NOTIFY_DONE;
2378
2379         dev_hold(vf_netdev);
2380         rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2381
2382         vf_netdev->wanted_features = ndev->features;
2383         netdev_update_features(vf_netdev);
2384
2385         prog = netvsc_xdp_get(netvsc_dev);
2386         netvsc_vf_setxdp(vf_netdev, prog);
2387
2388         return NOTIFY_OK;
2389 }
2390
2391 /* Change the data path when VF UP/DOWN/CHANGE are detected.
2392  *
2393  * Typically a UP or DOWN event is followed by a CHANGE event, so
2394  * net_device_ctx->data_path_is_vf is used to cache the current data path
2395  * to avoid the duplicate call of netvsc_switch_datapath() and the duplicate
2396  * message.
2397  *
2398  * During hibernation, if a VF NIC driver (e.g. mlx5) preserves the network
2399  * interface, there is only the CHANGE event and no UP or DOWN event.
2400  */
2401 static int netvsc_vf_changed(struct net_device *vf_netdev, unsigned long event)
2402 {
2403         struct net_device_context *net_device_ctx;
2404         struct netvsc_device *netvsc_dev;
2405         struct net_device *ndev;
2406         bool vf_is_up = false;
2407
2408         if (event != NETDEV_GOING_DOWN)
2409                 vf_is_up = netif_running(vf_netdev);
2410
2411         ndev = get_netvsc_byref(vf_netdev);
2412         if (!ndev)
2413                 return NOTIFY_DONE;
2414
2415         net_device_ctx = netdev_priv(ndev);
2416         netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
2417         if (!netvsc_dev)
2418                 return NOTIFY_DONE;
2419
2420         if (net_device_ctx->data_path_is_vf == vf_is_up)
2421                 return NOTIFY_OK;
2422
2423         netvsc_switch_datapath(ndev, vf_is_up);
2424         netdev_info(ndev, "Data path switched %s VF: %s\n",
2425                     vf_is_up ? "to" : "from", vf_netdev->name);
2426
2427         return NOTIFY_OK;
2428 }
2429
2430 static int netvsc_unregister_vf(struct net_device *vf_netdev)
2431 {
2432         struct net_device *ndev;
2433         struct net_device_context *net_device_ctx;
2434
2435         ndev = get_netvsc_byref(vf_netdev);
2436         if (!ndev)
2437                 return NOTIFY_DONE;
2438
2439         net_device_ctx = netdev_priv(ndev);
2440         cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
2441
2442         netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
2443
2444         netvsc_vf_setxdp(vf_netdev, NULL);
2445
2446         netdev_rx_handler_unregister(vf_netdev);
2447         netdev_upper_dev_unlink(vf_netdev, ndev);
2448         RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
2449         dev_put(vf_netdev);
2450
2451         return NOTIFY_OK;
2452 }
2453
2454 static int netvsc_probe(struct hv_device *dev,
2455                         const struct hv_vmbus_device_id *dev_id)
2456 {
2457         struct net_device *net = NULL;
2458         struct net_device_context *net_device_ctx;
2459         struct netvsc_device_info *device_info = NULL;
2460         struct netvsc_device *nvdev;
2461         int ret = -ENOMEM;
2462
2463         net = alloc_etherdev_mq(sizeof(struct net_device_context),
2464                                 VRSS_CHANNEL_MAX);
2465         if (!net)
2466                 goto no_net;
2467
2468         netif_carrier_off(net);
2469
2470         netvsc_init_settings(net);
2471
2472         net_device_ctx = netdev_priv(net);
2473         net_device_ctx->device_ctx = dev;
2474         net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2475         if (netif_msg_probe(net_device_ctx))
2476                 netdev_dbg(net, "netvsc msg_enable: %d\n",
2477                            net_device_ctx->msg_enable);
2478
2479         hv_set_drvdata(dev, net);
2480
2481         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
2482
2483         spin_lock_init(&net_device_ctx->lock);
2484         INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
2485         INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
2486
2487         net_device_ctx->vf_stats
2488                 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2489         if (!net_device_ctx->vf_stats)
2490                 goto no_stats;
2491
2492         net->netdev_ops = &device_ops;
2493         net->ethtool_ops = &ethtool_ops;
2494         SET_NETDEV_DEV(net, &dev->device);
2495
2496         /* We always need headroom for rndis header */
2497         net->needed_headroom = RNDIS_AND_PPI_SIZE;
2498
2499         /* Initialize the number of queues to be 1, we may change it if more
2500          * channels are offered later.
2501          */
2502         netif_set_real_num_tx_queues(net, 1);
2503         netif_set_real_num_rx_queues(net, 1);
2504
2505         /* Notify the netvsc driver of the new device */
2506         device_info = netvsc_devinfo_get(NULL);
2507
2508         if (!device_info) {
2509                 ret = -ENOMEM;
2510                 goto devinfo_failed;
2511         }
2512
2513         nvdev = rndis_filter_device_add(dev, device_info);
2514         if (IS_ERR(nvdev)) {
2515                 ret = PTR_ERR(nvdev);
2516                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
2517                 goto rndis_failed;
2518         }
2519
2520         memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
2521
2522         /* We must get rtnl lock before scheduling nvdev->subchan_work,
2523          * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2524          * all subchannels to show up, but that may not happen because
2525          * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2526          * -> ... -> device_add() -> ... -> __device_attach() can't get
2527          * the device lock, so all the subchannels can't be processed --
2528          * finally netvsc_subchan_work() hangs forever.
2529          */
2530         rtnl_lock();
2531
2532         if (nvdev->num_chn > 1)
2533                 schedule_work(&nvdev->subchan_work);
2534
2535         /* hw_features computed in rndis_netdev_set_hwcaps() */
2536         net->features = net->hw_features |
2537                 NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX |
2538                 NETIF_F_HW_VLAN_CTAG_RX;
2539         net->vlan_features = net->features;
2540
2541         netdev_lockdep_set_classes(net);
2542
2543         /* MTU range: 68 - 1500 or 65521 */
2544         net->min_mtu = NETVSC_MTU_MIN;
2545         if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2546                 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2547         else
2548                 net->max_mtu = ETH_DATA_LEN;
2549
2550         nvdev->tx_disable = false;
2551
2552         ret = register_netdevice(net);
2553         if (ret != 0) {
2554                 pr_err("Unable to register netdev.\n");
2555                 goto register_failed;
2556         }
2557
2558         list_add(&net_device_ctx->list, &netvsc_dev_list);
2559         rtnl_unlock();
2560
2561         netvsc_devinfo_put(device_info);
2562         return 0;
2563
2564 register_failed:
2565         rtnl_unlock();
2566         rndis_filter_device_remove(dev, nvdev);
2567 rndis_failed:
2568         netvsc_devinfo_put(device_info);
2569 devinfo_failed:
2570         free_percpu(net_device_ctx->vf_stats);
2571 no_stats:
2572         hv_set_drvdata(dev, NULL);
2573         free_netdev(net);
2574 no_net:
2575         return ret;
2576 }
2577
2578 static int netvsc_remove(struct hv_device *dev)
2579 {
2580         struct net_device_context *ndev_ctx;
2581         struct net_device *vf_netdev, *net;
2582         struct netvsc_device *nvdev;
2583
2584         net = hv_get_drvdata(dev);
2585         if (net == NULL) {
2586                 dev_err(&dev->device, "No net device to remove\n");
2587                 return 0;
2588         }
2589
2590         ndev_ctx = netdev_priv(net);
2591
2592         cancel_delayed_work_sync(&ndev_ctx->dwork);
2593
2594         rtnl_lock();
2595         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2596         if (nvdev) {
2597                 cancel_work_sync(&nvdev->subchan_work);
2598                 netvsc_xdp_set(net, NULL, NULL, nvdev);
2599         }
2600
2601         /*
2602          * Call to the vsc driver to let it know that the device is being
2603          * removed. Also blocks mtu and channel changes.
2604          */
2605         vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2606         if (vf_netdev)
2607                 netvsc_unregister_vf(vf_netdev);
2608
2609         if (nvdev)
2610                 rndis_filter_device_remove(dev, nvdev);
2611
2612         unregister_netdevice(net);
2613         list_del(&ndev_ctx->list);
2614
2615         rtnl_unlock();
2616
2617         hv_set_drvdata(dev, NULL);
2618
2619         free_percpu(ndev_ctx->vf_stats);
2620         free_netdev(net);
2621         return 0;
2622 }
2623
2624 static int netvsc_suspend(struct hv_device *dev)
2625 {
2626         struct net_device_context *ndev_ctx;
2627         struct netvsc_device *nvdev;
2628         struct net_device *net;
2629         int ret;
2630
2631         net = hv_get_drvdata(dev);
2632
2633         ndev_ctx = netdev_priv(net);
2634         cancel_delayed_work_sync(&ndev_ctx->dwork);
2635
2636         rtnl_lock();
2637
2638         nvdev = rtnl_dereference(ndev_ctx->nvdev);
2639         if (nvdev == NULL) {
2640                 ret = -ENODEV;
2641                 goto out;
2642         }
2643
2644         /* Save the current config info */
2645         ndev_ctx->saved_netvsc_dev_info = netvsc_devinfo_get(nvdev);
2646
2647         ret = netvsc_detach(net, nvdev);
2648 out:
2649         rtnl_unlock();
2650
2651         return ret;
2652 }
2653
2654 static int netvsc_resume(struct hv_device *dev)
2655 {
2656         struct net_device *net = hv_get_drvdata(dev);
2657         struct net_device_context *net_device_ctx;
2658         struct netvsc_device_info *device_info;
2659         int ret;
2660
2661         rtnl_lock();
2662
2663         net_device_ctx = netdev_priv(net);
2664
2665         /* Reset the data path to the netvsc NIC before re-opening the vmbus
2666          * channel. Later netvsc_netdev_event() will switch the data path to
2667          * the VF upon the UP or CHANGE event.
2668          */
2669         net_device_ctx->data_path_is_vf = false;
2670         device_info = net_device_ctx->saved_netvsc_dev_info;
2671
2672         ret = netvsc_attach(net, device_info);
2673
2674         netvsc_devinfo_put(device_info);
2675         net_device_ctx->saved_netvsc_dev_info = NULL;
2676
2677         rtnl_unlock();
2678
2679         return ret;
2680 }
2681 static const struct hv_vmbus_device_id id_table[] = {
2682         /* Network guid */
2683         { HV_NIC_GUID, },
2684         { },
2685 };
2686
2687 MODULE_DEVICE_TABLE(vmbus, id_table);
2688
2689 /* The one and only one */
2690 static struct  hv_driver netvsc_drv = {
2691         .name = KBUILD_MODNAME,
2692         .id_table = id_table,
2693         .probe = netvsc_probe,
2694         .remove = netvsc_remove,
2695         .suspend = netvsc_suspend,
2696         .resume = netvsc_resume,
2697         .driver = {
2698                 .probe_type = PROBE_FORCE_SYNCHRONOUS,
2699         },
2700 };
2701
2702 /*
2703  * On Hyper-V, every VF interface is matched with a corresponding
2704  * synthetic interface. The synthetic interface is presented first
2705  * to the guest. When the corresponding VF instance is registered,
2706  * we will take care of switching the data path.
2707  */
2708 static int netvsc_netdev_event(struct notifier_block *this,
2709                                unsigned long event, void *ptr)
2710 {
2711         struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2712
2713         /* Skip our own events */
2714         if (event_dev->netdev_ops == &device_ops)
2715                 return NOTIFY_DONE;
2716
2717         /* Avoid non-Ethernet type devices */
2718         if (event_dev->type != ARPHRD_ETHER)
2719                 return NOTIFY_DONE;
2720
2721         /* Avoid Vlan dev with same MAC registering as VF */
2722         if (is_vlan_dev(event_dev))
2723                 return NOTIFY_DONE;
2724
2725         /* Avoid Bonding master dev with same MAC registering as VF */
2726         if ((event_dev->priv_flags & IFF_BONDING) &&
2727             (event_dev->flags & IFF_MASTER))
2728                 return NOTIFY_DONE;
2729
2730         switch (event) {
2731         case NETDEV_REGISTER:
2732                 return netvsc_register_vf(event_dev);
2733         case NETDEV_UNREGISTER:
2734                 return netvsc_unregister_vf(event_dev);
2735         case NETDEV_UP:
2736         case NETDEV_DOWN:
2737         case NETDEV_CHANGE:
2738         case NETDEV_GOING_DOWN:
2739                 return netvsc_vf_changed(event_dev, event);
2740         default:
2741                 return NOTIFY_DONE;
2742         }
2743 }
2744
2745 static struct notifier_block netvsc_netdev_notifier = {
2746         .notifier_call = netvsc_netdev_event,
2747 };
2748
2749 static void __exit netvsc_drv_exit(void)
2750 {
2751         unregister_netdevice_notifier(&netvsc_netdev_notifier);
2752         vmbus_driver_unregister(&netvsc_drv);
2753 }
2754
2755 static int __init netvsc_drv_init(void)
2756 {
2757         int ret;
2758
2759         if (ring_size < RING_SIZE_MIN) {
2760                 ring_size = RING_SIZE_MIN;
2761                 pr_info("Increased ring_size to %u (min allowed)\n",
2762                         ring_size);
2763         }
2764         netvsc_ring_bytes = ring_size * PAGE_SIZE;
2765
2766         ret = vmbus_driver_register(&netvsc_drv);
2767         if (ret)
2768                 return ret;
2769
2770         register_netdevice_notifier(&netvsc_netdev_notifier);
2771         return 0;
2772 }
2773
2774 MODULE_LICENSE("GPL");
2775 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
2776
2777 module_init(netvsc_drv_init);
2778 module_exit(netvsc_drv_exit);